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, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/ost/ost_handler.c
38 * Author: Peter J. Braam <braam@clusterfs.com>
39 * Author: Phil Schwan <phil@clusterfs.com>
42 #define DEBUG_SUBSYSTEM S_OST
44 #include <linux/module.h>
45 #include <obd_cksum.h>
47 #include <lustre_net.h>
48 #include <lustre_dlm.h>
49 #include <lustre_export.h>
50 #include <lustre_debug.h>
51 #include <linux/init.h>
52 #include <lprocfs_status.h>
53 #include <libcfs/list.h>
54 #include "ost_internal.h"
56 static int oss_num_threads;
57 CFS_MODULE_PARM(oss_num_threads, "i", int, 0444,
58 "number of OSS service threads to start");
60 static int ost_num_threads;
61 CFS_MODULE_PARM(ost_num_threads, "i", int, 0444,
62 "number of OST service threads to start (deprecated)");
64 static int oss_num_create_threads;
65 CFS_MODULE_PARM(oss_num_create_threads, "i", int, 0444,
66 "number of OSS create threads to start");
68 static char *oss_cpts;
69 CFS_MODULE_PARM(oss_cpts, "s", charp, 0444,
70 "CPU partitions OSS threads should run on");
72 static char *oss_io_cpts;
73 CFS_MODULE_PARM(oss_io_cpts, "s", charp, 0444,
74 "CPU partitions OSS IO threads should run on");
77 * this page is allocated statically when module is initializing
78 * it is used to simulate data corruptions, see ost_checksum_bulk()
79 * for details. as the original pages provided by the layers below
80 * can be remain in the internal cache, we do not want to modify
83 static struct page *ost_page_to_corrupt = NULL;
86 * Do not return server-side uid/gid to remote client
88 static void ost_drop_id(struct obd_export *exp, struct obdo *oa)
90 if (exp_connect_rmtclient(exp)) {
93 oa->o_valid &= ~(OBD_MD_FLUID | OBD_MD_FLGID);
98 * Validate oa from client.
99 * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
101 * a. for single MDS seq = FID_SEQ_OST_MDT0,
102 * b. for CMD, seq = FID_SEQ_OST_MDT0, FID_SEQ_OST_MDT1 - FID_SEQ_OST_MAX
104 static int ost_validate_obdo(struct obd_export *exp, struct obdo *oa,
105 struct obd_ioobj *ioobj)
107 if (oa != NULL && !(oa->o_valid & OBD_MD_FLGROUP)) {
108 oa->o_seq = FID_SEQ_OST_MDT0;
110 ioobj->ioo_seq = FID_SEQ_OST_MDT0;
111 /* remove fid_seq_is_rsvd() after FID-on-OST allows SEQ > 9 */
112 } else if (oa == NULL || !(fid_seq_is_rsvd(oa->o_seq) ||
113 fid_seq_is_mdt0(oa->o_seq))) {
114 CERROR("%s: client %s sent invalid object "POSTID"\n",
115 exp->exp_obd->obd_name, obd_export_nid2str(exp),
116 oa ? oa->o_id : -1, oa ? oa->o_seq : -1);
119 obdo_from_ostid(oa, &oa->o_oi);
121 ioobj_from_obdo(ioobj, oa);
125 void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req)
127 struct oti_req_ack_lock *ack_lock;
133 if (req->rq_repmsg) {
134 __u64 versions[PTLRPC_NUM_VERSIONS] = { 0 };
135 lustre_msg_set_transno(req->rq_repmsg, oti->oti_transno);
136 versions[0] = oti->oti_pre_version;
137 lustre_msg_set_versions(req->rq_repmsg, versions);
139 req->rq_transno = oti->oti_transno;
141 /* XXX 4 == entries in oti_ack_locks??? */
142 for (ack_lock = oti->oti_ack_locks, i = 0; i < 4; i++, ack_lock++) {
145 /* XXX not even calling target_send_reply in some cases... */
146 ptlrpc_save_lock (req, &ack_lock->lock, ack_lock->mode, 0);
150 static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req,
151 struct obd_trans_info *oti)
153 struct ost_body *body, *repbody;
154 struct lustre_capa *capa = NULL;
158 /* Get the request body */
159 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
163 if (body->oa.o_id == 0)
166 rc = ost_validate_obdo(exp, &body->oa, NULL);
170 /* If there's a DLM request, cancel the locks mentioned in it*/
171 if (req_capsule_field_present(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT)) {
172 struct ldlm_request *dlm;
174 dlm = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
177 ldlm_request_cancel(req, dlm, 0);
180 /* If there's a capability, get it */
181 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
182 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
184 CERROR("Missing capability for OST DESTROY");
189 /* Prepare the reply */
190 rc = req_capsule_server_pack(&req->rq_pill);
194 /* Get the log cancellation cookie */
195 if (body->oa.o_valid & OBD_MD_FLCOOKIE)
196 oti->oti_logcookies = &body->oa.o_lcookie;
198 /* Finish the reply */
199 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
200 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
202 /* Do the destroy and set the reply status accordingly */
203 req->rq_status = obd_destroy(req->rq_svc_thread->t_env, exp,
204 &repbody->oa, NULL, oti, NULL, capa);
209 * Helper function for getting server side [start, start+count] DLM lock
210 * if asked by client.
212 static int ost_lock_get(struct obd_export *exp, struct obdo *oa,
213 __u64 start, __u64 count, struct lustre_handle *lh,
216 struct ldlm_res_id res_id;
217 ldlm_policy_data_t policy;
218 __u64 end = start + count;
222 LASSERT(!lustre_handle_is_used(lh));
223 /* o_id and o_gr are used for localizing resource, if client miss to set
224 * them, do not trigger ASSERTION. */
225 if (unlikely((oa->o_valid & (OBD_MD_FLID | OBD_MD_FLGROUP)) !=
226 (OBD_MD_FLID | OBD_MD_FLGROUP)))
229 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
230 !(oa->o_flags & OBD_FL_SRVLOCK))
233 osc_build_res_name(oa->o_id, oa->o_seq, &res_id);
234 CDEBUG(D_INODE, "OST-side extent lock.\n");
236 policy.l_extent.start = start & CFS_PAGE_MASK;
238 /* If ->o_blocks is EOF it means "lock till the end of the
239 * file". Otherwise, it's size of a hole being punched (in bytes) */
240 if (count == OBD_OBJECT_EOF || end < start)
241 policy.l_extent.end = OBD_OBJECT_EOF;
243 policy.l_extent.end = end | ~CFS_PAGE_MASK;
245 RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
246 LDLM_EXTENT, &policy, mode, &flags,
247 ldlm_blocking_ast, ldlm_completion_ast,
248 ldlm_glimpse_ast, NULL, 0, NULL, lh));
251 /* Helper function: release lock, if any. */
252 static void ost_lock_put(struct obd_export *exp,
253 struct lustre_handle *lh, int mode)
256 if (lustre_handle_is_used(lh))
257 ldlm_lock_decref(lh, mode);
261 static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
263 struct ost_body *body, *repbody;
264 struct obd_info *oinfo;
265 struct lustre_handle lh = { 0 };
266 struct lustre_capa *capa = NULL;
270 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
274 rc = ost_validate_obdo(exp, &body->oa, NULL);
278 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
279 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
281 CERROR("Missing capability for OST GETATTR");
286 rc = req_capsule_server_pack(&req->rq_pill);
290 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
291 repbody->oa = body->oa;
293 rc = ost_lock_get(exp, &repbody->oa, 0, OBD_OBJECT_EOF, &lh, LCK_PR, 0);
297 OBD_ALLOC_PTR(oinfo);
299 GOTO(unlock, rc = -ENOMEM);
300 oinfo->oi_oa = &repbody->oa;
301 oinfo->oi_capa = capa;
303 req->rq_status = obd_getattr(req->rq_svc_thread->t_env, exp, oinfo);
307 ost_drop_id(exp, &repbody->oa);
310 ost_lock_put(exp, &lh, LCK_PR);
314 static int ost_statfs(struct ptlrpc_request *req)
316 struct obd_statfs *osfs;
320 rc = req_capsule_server_pack(&req->rq_pill);
324 osfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
326 req->rq_status = obd_statfs(req->rq_svc_thread->t_env, req->rq_export,
328 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
330 if (req->rq_status != 0)
331 CERROR("ost: statfs failed: rc %d\n", req->rq_status);
333 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
334 req->rq_status = -EINPROGRESS;
339 static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
340 struct obd_trans_info *oti)
342 struct ost_body *body, *repbody;
346 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
350 rc = ost_validate_obdo(req->rq_export, &body->oa, NULL);
354 rc = req_capsule_server_pack(&req->rq_pill);
358 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
359 repbody->oa = body->oa;
360 oti->oti_logcookies = &body->oa.o_lcookie;
362 req->rq_status = obd_create(req->rq_svc_thread->t_env, exp,
363 &repbody->oa, NULL, oti);
364 //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
368 static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
369 struct obd_trans_info *oti)
371 struct ost_body *body, *repbody;
373 struct lustre_handle lh = {0,};
376 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
377 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
379 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
383 rc = ost_validate_obdo(exp, &body->oa, NULL);
387 if ((body->oa.o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
388 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
391 rc = req_capsule_server_pack(&req->rq_pill);
395 /* standard truncate optimization: if file body is completely
396 * destroyed, don't send data back to the server. */
397 if (body->oa.o_size == 0)
398 flags |= LDLM_AST_DISCARD_DATA;
400 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
401 repbody->oa = body->oa;
403 rc = ost_lock_get(exp, &repbody->oa, repbody->oa.o_size,
404 repbody->oa.o_blocks, &lh, LCK_PW, flags);
406 struct obd_info *oinfo;
407 struct lustre_capa *capa = NULL;
409 if (repbody->oa.o_valid & OBD_MD_FLFLAGS &&
410 repbody->oa.o_flags == OBD_FL_SRVLOCK)
412 * If OBD_FL_SRVLOCK is the only bit set in
413 * ->o_flags, clear OBD_MD_FLFLAGS to avoid falling
414 * through filter_setattr() to filter_iocontrol().
416 repbody->oa.o_valid &= ~OBD_MD_FLFLAGS;
418 if (repbody->oa.o_valid & OBD_MD_FLOSSCAPA) {
419 capa = req_capsule_client_get(&req->rq_pill,
422 CERROR("Missing capability for OST PUNCH");
423 GOTO(unlock, rc = -EFAULT);
427 OBD_ALLOC_PTR(oinfo);
429 GOTO(unlock, rc = -ENOMEM);
430 oinfo->oi_oa = &repbody->oa;
431 oinfo->oi_policy.l_extent.start = oinfo->oi_oa->o_size;
432 oinfo->oi_policy.l_extent.end = oinfo->oi_oa->o_blocks;
433 oinfo->oi_capa = capa;
434 oinfo->oi_flags = OBD_FL_PUNCH;
436 req->rq_status = obd_punch(req->rq_svc_thread->t_env, exp,
440 ost_lock_put(exp, &lh, LCK_PW);
443 ost_drop_id(exp, &repbody->oa);
447 static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req,
448 struct obd_trans_info *oti)
450 struct ost_body *body, *repbody;
451 struct obd_info *oinfo;
452 struct lustre_capa *capa = NULL;
456 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
460 rc = ost_validate_obdo(exp, &body->oa, NULL);
464 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
465 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
467 CERROR("Missing capability for OST SYNC");
472 rc = req_capsule_server_pack(&req->rq_pill);
476 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
477 repbody->oa = body->oa;
479 OBD_ALLOC_PTR(oinfo);
483 oinfo->oi_oa = &repbody->oa;
484 oinfo->oi_capa = capa;
485 oinfo->oi_jobid = oti->oti_jobid;
486 req->rq_status = obd_sync(req->rq_svc_thread->t_env, exp, oinfo,
487 repbody->oa.o_size, repbody->oa.o_blocks,
491 ost_drop_id(exp, &repbody->oa);
495 static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req,
496 struct obd_trans_info *oti)
498 struct ost_body *body, *repbody;
499 struct obd_info *oinfo;
500 struct lustre_capa *capa = NULL;
504 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
508 rc = ost_validate_obdo(req->rq_export, &body->oa, NULL);
512 rc = req_capsule_server_pack(&req->rq_pill);
516 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
517 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
519 CERROR("Missing capability for OST SETATTR");
524 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
525 repbody->oa = body->oa;
527 OBD_ALLOC_PTR(oinfo);
530 oinfo->oi_oa = &repbody->oa;
531 oinfo->oi_capa = capa;
533 req->rq_status = obd_setattr(req->rq_svc_thread->t_env, exp, oinfo,
538 ost_drop_id(exp, &repbody->oa);
542 static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc,
543 cksum_type_t cksum_type)
545 struct cfs_crypto_hash_desc *hdesc;
546 unsigned int bufsize;
548 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
551 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
553 CERROR("Unable to initialize checksum hash %s\n",
554 cfs_crypto_hash_name(cfs_alg));
555 return PTR_ERR(hdesc);
557 CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
558 for (i = 0; i < desc->bd_iov_count; i++) {
560 /* corrupt the data before we compute the checksum, to
561 * simulate a client->OST data error */
562 if (i == 0 && opc == OST_WRITE &&
563 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
564 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
565 int len = desc->bd_iov[i].kiov_len;
566 struct page *np = ost_page_to_corrupt;
567 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
570 char *ptr2 = kmap(np) + off;
572 memcpy(ptr2, ptr, len);
573 memcpy(ptr2, "bad3", min(4, len));
575 desc->bd_iov[i].kiov_page = np;
577 CERROR("can't alloc page for corruption\n");
580 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
581 desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
582 desc->bd_iov[i].kiov_len);
584 /* corrupt the data after we compute the checksum, to
585 * simulate an OST->client data error */
586 if (i == 0 && opc == OST_READ &&
587 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
588 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
589 int len = desc->bd_iov[i].kiov_len;
590 struct page *np = ost_page_to_corrupt;
591 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
594 char *ptr2 = kmap(np) + off;
596 memcpy(ptr2, ptr, len);
597 memcpy(ptr2, "bad4", min(4, len));
599 desc->bd_iov[i].kiov_page = np;
601 CERROR("can't alloc page for corruption\n");
607 err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
609 cfs_crypto_hash_final(hdesc, NULL, NULL);
614 static int ost_brw_lock_get(int mode, struct obd_export *exp,
615 struct obd_ioobj *obj, struct niobuf_remote *nb,
616 struct lustre_handle *lh)
619 int nrbufs = obj->ioo_bufcnt;
620 struct ldlm_res_id res_id;
621 ldlm_policy_data_t policy;
625 osc_build_res_name(obj->ioo_id, obj->ioo_seq, &res_id);
626 LASSERT(mode == LCK_PR || mode == LCK_PW);
627 LASSERT(!lustre_handle_is_used(lh));
629 if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
632 for (i = 1; i < nrbufs; i ++)
633 if ((nb[0].flags & OBD_BRW_SRVLOCK) !=
634 (nb[i].flags & OBD_BRW_SRVLOCK))
637 policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
638 policy.l_extent.end = (nb[nrbufs - 1].offset +
639 nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
641 RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
642 LDLM_EXTENT, &policy, mode, &flags,
643 ldlm_blocking_ast, ldlm_completion_ast,
644 ldlm_glimpse_ast, NULL, 0, NULL, lh));
647 static void ost_brw_lock_put(int mode,
648 struct obd_ioobj *obj, struct niobuf_remote *niob,
649 struct lustre_handle *lh)
652 LASSERT(mode == LCK_PR || mode == LCK_PW);
653 LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
654 lustre_handle_is_used(lh));
655 if (lustre_handle_is_used(lh))
656 ldlm_lock_decref(lh, mode);
660 /* Allocate thread local buffers if needed */
661 static struct ost_thread_local_cache *ost_tls_get(struct ptlrpc_request *r)
663 struct ost_thread_local_cache *tls =
664 (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
666 /* In normal mode of operation an I/O request is serviced only
667 * by ll_ost_io threads each of them has own tls buffers allocated by
669 * During recovery, an I/O request may be queued until any of the ost
670 * service threads process it. Not necessary it should be one of
671 * ll_ost_io threads. In that case we dynamically allocating tls
672 * buffers for the request service time. */
673 if (unlikely(tls == NULL)) {
674 LASSERT(r->rq_export->exp_in_recovery);
678 r->rq_svc_thread->t_data = tls;
684 /* Free thread local buffers if they were allocated only for servicing
685 * this one request */
686 static void ost_tls_put(struct ptlrpc_request *r)
688 struct ost_thread_local_cache *tls =
689 (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
691 if (unlikely(tls->temporary)) {
693 r->rq_svc_thread->t_data = NULL;
697 static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
699 struct ptlrpc_bulk_desc *desc = NULL;
700 struct obd_export *exp = req->rq_export;
701 struct niobuf_remote *remote_nb;
702 struct niobuf_local *local_nb;
703 struct obd_ioobj *ioo;
704 struct ost_body *body, *repbody;
705 struct lustre_capa *capa = NULL;
706 struct l_wait_info lwi;
707 struct lustre_handle lockh = { 0 };
708 int niocount, npages, nob = 0, rc, i;
710 struct ost_thread_local_cache *tls;
713 req->rq_bulk_read = 1;
715 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
716 GOTO(out, rc = -EIO);
718 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
720 /* Check if there is eviction in progress, and if so, wait for it to
722 if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
723 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
724 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
725 !cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress),
729 GOTO(out, rc = -ENOTCONN);
731 /* ost_body, ioobj & noibuf_remote are verified and swabbed in
732 * ost_rw_hpreq_check(). */
733 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
735 GOTO(out, rc = -EFAULT);
738 * A req_capsule_X_get_array(pill, field, ptr_to_element_count) function
739 * would be useful here and wherever we get &RMF_OBD_IOOBJ and
740 * &RMF_NIOBUF_REMOTE.
742 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
744 GOTO(out, rc = -EFAULT);
746 rc = ost_validate_obdo(exp, &body->oa, ioo);
750 niocount = ioo->ioo_bufcnt;
751 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
752 if (remote_nb == NULL)
753 GOTO(out, rc = -EFAULT);
755 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
756 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
758 CERROR("Missing capability for OST BRW READ");
759 GOTO(out, rc = -EFAULT);
763 rc = req_capsule_server_pack(&req->rq_pill);
767 tls = ost_tls_get(req);
769 GOTO(out_bulk, rc = -ENOMEM);
770 local_nb = tls->local;
772 rc = ost_brw_lock_get(LCK_PR, exp, ioo, remote_nb, &lockh);
777 * If getting the lock took more time than
778 * client was willing to wait, drop it. b=11330
780 if (cfs_time_current_sec() > req->rq_deadline ||
781 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
783 CERROR("Dropping timed-out read from %s because locking"
784 "object "LPX64" took %ld seconds (limit was %ld).\n",
785 libcfs_id2str(req->rq_peer), ioo->ioo_id,
786 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
787 req->rq_deadline - req->rq_arrival_time.tv_sec);
788 GOTO(out_lock, rc = -ETIMEDOUT);
791 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
792 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
794 npages = OST_THREAD_POOL_SIZE;
795 rc = obd_preprw(req->rq_svc_thread->t_env, OBD_BRW_READ, exp,
796 &repbody->oa, 1, ioo, remote_nb, &npages, local_nb,
801 desc = ptlrpc_prep_bulk_exp(req, npages,
802 BULK_PUT_SOURCE, OST_BULK_PORTAL);
804 GOTO(out_commitrw, rc = -ENOMEM);
807 for (i = 0; i < npages; i++) {
808 int page_rc = local_nb[i].rc;
810 if (page_rc < 0) { /* error */
816 if (page_rc != 0) { /* some data! */
817 LASSERT (local_nb[i].page != NULL);
818 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
819 local_nb[i].lnb_page_offset,
823 if (page_rc != local_nb[i].len) { /* short read */
824 /* All subsequent pages should be 0 */
826 LASSERT(local_nb[i].rc == 0);
831 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
832 cksum_type_t cksum_type =
833 cksum_type_unpack(repbody->oa.o_valid & OBD_MD_FLFLAGS ?
834 repbody->oa.o_flags : 0);
835 repbody->oa.o_flags = cksum_type_pack(cksum_type);
836 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
837 repbody->oa.o_cksum = ost_checksum_bulk(desc, OST_READ,cksum_type);
838 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
839 repbody->oa.o_cksum);
841 repbody->oa.o_valid = 0;
843 /* We're finishing using body->oa as an input variable */
845 /* Check if client was evicted while we were doing i/o before touching
848 if (likely(!CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2)))
849 rc = target_bulk_io(exp, desc, &lwi);
854 /* Must commit after prep above in all cases */
855 rc = obd_commitrw(req->rq_svc_thread->t_env, OBD_BRW_READ, exp,
856 &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
860 ost_drop_id(exp, &repbody->oa);
863 ost_brw_lock_put(LCK_PR, ioo, remote_nb, &lockh);
867 if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
868 ptlrpc_free_bulk_nopin(desc);
872 req->rq_status = nob;
873 ptlrpc_lprocfs_brw(req, nob);
874 target_committed_to_req(req);
876 } else if (!no_reply) {
877 /* Only reply if there was no comms problem with bulk */
878 target_committed_to_req(req);
882 /* reply out callback would free */
883 ptlrpc_req_drop_rs(req);
884 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
885 "client will retry: rc %d\n",
886 exp->exp_obd->obd_name,
887 obd_uuid2str(&exp->exp_client_uuid),
888 obd_export_nid2str(exp), rc);
890 /* send a bulk after reply to simulate a network delay or reordering
892 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
894 struct l_wait_info lwi1;
896 CDEBUG(D_INFO, "reorder BULK\n");
897 cfs_waitq_init(&waitq);
899 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
900 l_wait_event(waitq, 0, &lwi1);
901 rc = target_bulk_io(exp, desc, &lwi);
902 ptlrpc_free_bulk_nopin(desc);
908 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
910 struct ptlrpc_bulk_desc *desc = NULL;
911 struct obd_export *exp = req->rq_export;
912 struct niobuf_remote *remote_nb;
913 struct niobuf_local *local_nb;
914 struct obd_ioobj *ioo;
915 struct ost_body *body, *repbody;
916 struct l_wait_info lwi;
917 struct lustre_handle lockh = {0};
918 struct lustre_capa *capa = NULL;
920 int objcount, niocount, npages;
922 obd_count client_cksum = 0, server_cksum = 0;
923 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
924 int no_reply = 0, mmap = 0;
925 __u32 o_uid = 0, o_gid = 0;
926 struct ost_thread_local_cache *tls;
929 req->rq_bulk_write = 1;
931 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
932 GOTO(out, rc = -EIO);
933 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
934 GOTO(out, rc = -EFAULT);
936 /* pause before transaction has been started */
937 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
939 /* ost_body, ioobj & noibuf_remote are verified and swabbed in
940 * ost_rw_hpreq_check(). */
941 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
943 GOTO(out, rc = -EFAULT);
945 objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
946 RCL_CLIENT) / sizeof(*ioo);
947 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
949 GOTO(out, rc = -EFAULT);
951 rc = ost_validate_obdo(exp, &body->oa, ioo);
955 for (niocount = i = 0; i < objcount; i++)
956 niocount += ioo[i].ioo_bufcnt;
959 * It'd be nice to have a capsule function to indicate how many elements
960 * there were in a buffer for an RMF that's declared to be an array.
961 * It's easy enough to compute the number of elements here though.
963 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
964 if (remote_nb == NULL || niocount != (req_capsule_get_size(&req->rq_pill,
965 &RMF_NIOBUF_REMOTE, RCL_CLIENT) / sizeof(*remote_nb)))
966 GOTO(out, rc = -EFAULT);
968 if ((remote_nb[0].flags & OBD_BRW_MEMALLOC) &&
969 (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
970 cfs_memory_pressure_set();
972 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
973 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
975 CERROR("Missing capability for OST BRW WRITE");
976 GOTO(out, rc = -EFAULT);
980 req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
981 niocount * sizeof(*rcs));
982 rc = req_capsule_server_pack(&req->rq_pill);
985 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
986 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
988 tls = ost_tls_get(req);
990 GOTO(out_bulk, rc = -ENOMEM);
991 local_nb = tls->local;
993 rc = ost_brw_lock_get(LCK_PW, exp, ioo, remote_nb, &lockh);
998 * If getting the lock took more time than
999 * client was willing to wait, drop it. b=11330
1001 if (cfs_time_current_sec() > req->rq_deadline ||
1002 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1004 CERROR("Dropping timed-out write from %s because locking "
1005 "object "LPX64" took %ld seconds (limit was %ld).\n",
1006 libcfs_id2str(req->rq_peer), ioo->ioo_id,
1007 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1008 req->rq_deadline - req->rq_arrival_time.tv_sec);
1009 GOTO(out_lock, rc = -ETIMEDOUT);
1012 /* obd_preprw clobbers oa->valid, so save what we need */
1013 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1014 client_cksum = body->oa.o_cksum;
1015 if (body->oa.o_valid & OBD_MD_FLFLAGS)
1016 cksum_type = cksum_type_unpack(body->oa.o_flags);
1018 if (body->oa.o_valid & OBD_MD_FLFLAGS && body->oa.o_flags & OBD_FL_MMAP)
1021 /* Because we already sync grant info with client when reconnect,
1022 * grant info will be cleared for resent req, then fed_grant and
1023 * total_grant will not be modified in following preprw_write */
1024 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1025 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1026 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1029 if (exp_connect_rmtclient(exp)) {
1030 o_uid = body->oa.o_uid;
1031 o_gid = body->oa.o_gid;
1034 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1035 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
1037 npages = OST_THREAD_POOL_SIZE;
1038 rc = obd_preprw(req->rq_svc_thread->t_env, OBD_BRW_WRITE, exp,
1039 &repbody->oa, objcount, ioo, remote_nb, &npages,
1040 local_nb, oti, capa);
1044 desc = ptlrpc_prep_bulk_exp(req, npages,
1045 BULK_GET_SINK, OST_BULK_PORTAL);
1047 GOTO(skip_transfer, rc = -ENOMEM);
1049 /* NB Having prepped, we must commit... */
1051 for (i = 0; i < npages; i++)
1052 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
1053 local_nb[i].lnb_page_offset,
1056 rc = sptlrpc_svc_prep_bulk(req, desc);
1060 rc = target_bulk_io(exp, desc, &lwi);
1064 if (client_cksum != 0 && rc == 0) {
1065 static int cksum_counter;
1066 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1067 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
1068 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
1069 server_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1070 repbody->oa.o_cksum = server_cksum;
1072 if (unlikely(client_cksum != server_cksum)) {
1073 CDEBUG_LIMIT(mmap ? D_INFO : D_ERROR,
1074 "client csum %x, server csum %x\n",
1075 client_cksum, server_cksum);
1077 } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){
1078 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1079 cksum_counter, libcfs_id2str(req->rq_peer),
1084 /* Must commit after prep above in all cases */
1085 rc = obd_commitrw(req->rq_svc_thread->t_env, OBD_BRW_WRITE, exp,
1086 &repbody->oa, objcount, ioo, remote_nb, npages,
1088 if (rc == -ENOTCONN)
1089 /* quota acquire process has been given up because
1090 * either the client has been evicted or the client
1091 * has timed out the request already */
1094 if (exp_connect_rmtclient(exp)) {
1095 repbody->oa.o_uid = o_uid;
1096 repbody->oa.o_gid = o_gid;
1100 * Disable sending mtime back to the client. If the client locked the
1101 * whole object, then it has already updated the mtime on its side,
1102 * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
1104 repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
1106 if (unlikely(client_cksum != server_cksum && rc == 0 && !mmap)) {
1107 int new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1112 if (new_cksum == server_cksum)
1113 msg = "changed in transit before arrival at OST";
1114 else if (new_cksum == client_cksum)
1115 msg = "initial checksum before message complete";
1117 msg = "changed in transit AND after initial checksum";
1119 if (req->rq_peer.nid == desc->bd_sender) {
1123 router = libcfs_nid2str(desc->bd_sender);
1126 LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: %s from "
1127 "%s%s%s inode "DFID" object "
1128 LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
1129 exp->exp_obd->obd_name, msg,
1130 libcfs_id2str(req->rq_peer),
1132 body->oa.o_valid & OBD_MD_FLFID ?
1133 body->oa.o_parent_seq : (__u64)0,
1134 body->oa.o_valid & OBD_MD_FLFID ?
1135 body->oa.o_parent_oid : 0,
1136 body->oa.o_valid & OBD_MD_FLFID ?
1137 body->oa.o_parent_ver : 0,
1139 body->oa.o_valid & OBD_MD_FLGROUP ?
1140 body->oa.o_seq : (__u64)0,
1141 local_nb[0].lnb_file_offset,
1142 local_nb[npages-1].lnb_file_offset +
1143 local_nb[npages-1].len - 1 );
1144 CERROR("client csum %x, original server csum %x, "
1145 "server csum now %x\n",
1146 client_cksum, server_cksum, new_cksum);
1152 /* set per-requested niobuf return codes */
1153 for (i = j = 0; i < niocount; i++) {
1154 int len = remote_nb[i].len;
1159 LASSERT(j < npages);
1160 if (local_nb[j].rc < 0)
1161 rcs[i] = local_nb[j].rc;
1162 len -= local_nb[j].len;
1167 LASSERT(j == npages);
1168 ptlrpc_lprocfs_brw(req, nob);
1172 ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh);
1177 ptlrpc_free_bulk_nopin(desc);
1180 oti_to_request(oti, req);
1181 target_committed_to_req(req);
1182 rc = ptlrpc_reply(req);
1183 } else if (!no_reply) {
1184 /* Only reply if there was no comms problem with bulk */
1185 target_committed_to_req(req);
1186 req->rq_status = rc;
1189 /* reply out callback would free */
1190 ptlrpc_req_drop_rs(req);
1191 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
1192 "client will retry: rc %d\n",
1193 exp->exp_obd->obd_name,
1194 obd_uuid2str(&exp->exp_client_uuid),
1195 obd_export_nid2str(exp), rc);
1197 cfs_memory_pressure_clr();
1202 * Implementation of OST_SET_INFO.
1204 * OST_SET_INFO is like ioctl(): heavily overloaded. Specifically, it takes a
1205 * "key" and a value RPC buffers as arguments, with the value's contents
1206 * interpreted according to the key.
1208 * Value types that need swabbing have swabbing done explicitly, either here or
1209 * in functions called from here. This should be corrected: all swabbing should
1210 * be done in the capsule abstraction, as that will then allow us to move
1211 * swabbing exclusively to the client without having to modify server code
1212 * outside the capsule abstraction's implementation itself. To correct this
1213 * will require minor changes to the capsule abstraction; see the comments for
1214 * req_capsule_extend() in layout.c.
1216 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
1218 struct ost_body *body = NULL, *repbody;
1219 char *key, *val = NULL;
1220 int keylen, vallen, rc = 0;
1221 int is_grant_shrink = 0;
1224 key = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
1226 DEBUG_REQ(D_HA, req, "no set_info key");
1229 keylen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_KEY,
1232 vallen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_VAL,
1235 if ((is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK)))
1236 /* In this case the value is actually an RMF_OST_BODY, so we
1237 * transmutate the type of this PTLRPC */
1238 req_capsule_extend(&req->rq_pill, &RQF_OST_SET_GRANT_INFO);
1240 rc = req_capsule_server_pack(&req->rq_pill);
1245 if (is_grant_shrink) {
1246 body = req_capsule_client_get(&req->rq_pill,
1251 repbody = req_capsule_server_get(&req->rq_pill,
1253 memcpy(repbody, body, sizeof(*body));
1254 val = (char*)repbody;
1256 val = req_capsule_client_get(&req->rq_pill,
1261 if (KEY_IS(KEY_EVICT_BY_NID)) {
1263 obd_export_evict_by_nid(exp->exp_obd, val);
1265 } else if (KEY_IS(KEY_MDS_CONN) && ptlrpc_req_need_swab(req)) {
1266 if (vallen < sizeof(__u32))
1268 __swab32s((__u32 *)val);
1271 /* OBD will also check if KEY_IS(KEY_GRANT_SHRINK), and will cast val to
1272 * a struct ost_body * value */
1273 rc = obd_set_info_async(req->rq_svc_thread->t_env, exp, keylen,
1274 key, vallen, val, NULL);
1276 lustre_msg_set_status(req->rq_repmsg, 0);
1280 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
1283 int keylen, replylen, rc = 0;
1284 struct req_capsule *pill = &req->rq_pill;
1287 /* this common part for get_info rpc */
1288 key = req_capsule_client_get(pill, &RMF_SETINFO_KEY);
1290 DEBUG_REQ(D_HA, req, "no get_info key");
1293 keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT);
1295 if (KEY_IS(KEY_FIEMAP)) {
1296 struct ll_fiemap_info_key *fm_key = key;
1299 rc = ost_validate_obdo(exp, &fm_key->oa, NULL);
1304 rc = obd_get_info(req->rq_svc_thread->t_env, exp, keylen, key,
1305 &replylen, NULL, NULL);
1309 req_capsule_set_size(pill, &RMF_GENERIC_DATA,
1310 RCL_SERVER, replylen);
1312 rc = req_capsule_server_pack(pill);
1316 reply = req_capsule_server_get(pill, &RMF_GENERIC_DATA);
1320 /* call again to fill in the reply buffer */
1321 rc = obd_get_info(req->rq_svc_thread->t_env, exp, keylen, key,
1322 &replylen, reply, NULL);
1324 lustre_msg_set_status(req->rq_repmsg, 0);
1328 static int ost_handle_quotactl(struct ptlrpc_request *req)
1330 struct obd_quotactl *oqctl, *repoqc;
1334 oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1336 GOTO(out, rc = -EPROTO);
1338 rc = req_capsule_server_pack(&req->rq_pill);
1342 repoqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1343 req->rq_status = obd_quotactl(req->rq_export, oqctl);
1350 static int ost_handle_quotacheck(struct ptlrpc_request *req)
1352 struct obd_quotactl *oqctl;
1356 oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1360 rc = req_capsule_server_pack(&req->rq_pill);
1364 /* deprecated, not used any more */
1365 req->rq_status = -EOPNOTSUPP;
1366 RETURN(-EOPNOTSUPP);
1369 static int ost_llog_handle_connect(struct obd_export *exp,
1370 struct ptlrpc_request *req)
1372 struct llogd_conn_body *body;
1376 body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_CONN_BODY);
1377 rc = obd_llog_connect(exp, body);
1381 #define ost_init_sec_none(reply, exp) \
1383 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT | \
1384 OBD_CONNECT_RMT_CLIENT_FORCE | \
1385 OBD_CONNECT_OSS_CAPA); \
1386 cfs_spin_lock(&exp->exp_lock); \
1387 exp->exp_connect_flags = reply->ocd_connect_flags; \
1388 cfs_spin_unlock(&exp->exp_lock); \
1391 static int ost_init_sec_level(struct ptlrpc_request *req)
1393 struct obd_export *exp = req->rq_export;
1394 struct req_capsule *pill = &req->rq_pill;
1395 struct obd_device *obd = exp->exp_obd;
1396 struct filter_obd *filter = &obd->u.filter;
1397 char *client = libcfs_nid2str(req->rq_peer.nid);
1398 struct obd_connect_data *data, *reply;
1402 data = req_capsule_client_get(pill, &RMF_CONNECT_DATA);
1403 reply = req_capsule_server_get(pill, &RMF_CONNECT_DATA);
1404 if (data == NULL || reply == NULL)
1407 /* connection from MDT is always trusted */
1408 if (req->rq_auth_usr_mdt) {
1409 ost_init_sec_none(reply, exp);
1413 /* no GSS support case */
1414 if (!req->rq_auth_gss) {
1415 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1416 CWARN("client %s -> target %s does not user GSS, "
1417 "can not run under security level %d.\n",
1418 client, obd->obd_name, filter->fo_sec_level);
1421 ost_init_sec_none(reply, exp);
1426 /* old version case */
1427 if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
1428 !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
1429 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1430 CWARN("client %s -> target %s uses old version, "
1431 "can not run under security level %d.\n",
1432 client, obd->obd_name, filter->fo_sec_level);
1435 CWARN("client %s -> target %s uses old version, "
1436 "run under security level %d.\n",
1437 client, obd->obd_name, filter->fo_sec_level);
1438 ost_init_sec_none(reply, exp);
1443 remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
1445 if (!req->rq_auth_remote)
1446 CDEBUG(D_SEC, "client (local realm) %s -> target %s "
1447 "asked to be remote.\n", client, obd->obd_name);
1448 } else if (req->rq_auth_remote) {
1450 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
1451 "as remote by default.\n", client, obd->obd_name);
1455 if (!filter->fo_fl_oss_capa) {
1456 CDEBUG(D_SEC, "client %s -> target %s is set as remote,"
1457 " but OSS capabilities are not enabled: %d.\n",
1458 client, obd->obd_name, filter->fo_fl_oss_capa);
1463 switch (filter->fo_sec_level) {
1464 case LUSTRE_SEC_NONE:
1466 ost_init_sec_none(reply, exp);
1469 CDEBUG(D_SEC, "client %s -> target %s is set as remote, "
1470 "can not run under security level %d.\n",
1471 client, obd->obd_name, filter->fo_sec_level);
1474 case LUSTRE_SEC_REMOTE:
1476 ost_init_sec_none(reply, exp);
1478 case LUSTRE_SEC_ALL:
1480 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
1481 OBD_CONNECT_RMT_CLIENT_FORCE);
1482 if (!filter->fo_fl_oss_capa)
1483 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
1485 cfs_spin_lock(&exp->exp_lock);
1486 exp->exp_connect_flags = reply->ocd_connect_flags;
1487 cfs_spin_unlock(&exp->exp_lock);
1499 * this should be done in filter_connect()/filter_reconnect(), but
1500 * we can't obtain information like NID, which stored in incoming
1501 * request, thus can't decide what flavor to use. so we do it here.
1503 * This hack should be removed after the OST stack be rewritten, just
1504 * like what we are doing in mdt_obd_connect()/mdt_obd_reconnect().
1506 static int ost_connect_check_sptlrpc(struct ptlrpc_request *req)
1508 struct obd_export *exp = req->rq_export;
1509 struct filter_obd *filter = &exp->exp_obd->u.filter;
1510 struct sptlrpc_flavor flvr;
1513 if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
1514 LUSTRE_ECHO_NAME) == 0)) {
1515 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
1519 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
1520 cfs_read_lock(&filter->fo_sptlrpc_lock);
1521 sptlrpc_target_choose_flavor(&filter->fo_sptlrpc_rset,
1525 cfs_read_unlock(&filter->fo_sptlrpc_lock);
1527 cfs_spin_lock(&exp->exp_lock);
1529 exp->exp_sp_peer = req->rq_sp_from;
1530 exp->exp_flvr = flvr;
1532 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
1533 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
1534 CERROR("unauthorized rpc flavor %x from %s, "
1535 "expect %x\n", req->rq_flvr.sf_rpc,
1536 libcfs_nid2str(req->rq_peer.nid),
1537 exp->exp_flvr.sf_rpc);
1541 cfs_spin_unlock(&exp->exp_lock);
1543 if (exp->exp_sp_peer != req->rq_sp_from) {
1544 CERROR("RPC source %s doesn't match %s\n",
1545 sptlrpc_part2name(req->rq_sp_from),
1546 sptlrpc_part2name(exp->exp_sp_peer));
1549 rc = sptlrpc_target_export_check(exp, req);
1556 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1558 int ost_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1559 void *data, int flag)
1562 __u32 sync_lock_cancel = 0;
1563 __u32 len = sizeof(sync_lock_cancel);
1568 rc = lu_env_init(&env, LCT_DT_THREAD);
1569 if (unlikely(rc != 0))
1572 rc = obd_get_info(&env, lock->l_export, sizeof(KEY_SYNC_LOCK_CANCEL),
1573 KEY_SYNC_LOCK_CANCEL, &len, &sync_lock_cancel, NULL);
1574 if (rc == 0 && flag == LDLM_CB_CANCELING &&
1575 (lock->l_granted_mode & (LCK_PW|LCK_GROUP)) &&
1576 (sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1577 (sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1578 lock->l_flags & LDLM_FL_CBPENDING))) {
1579 struct obd_info *oinfo;
1583 OBD_ALLOC_PTR(oinfo);
1585 GOTO(out_env, rc = -ENOMEM);
1588 OBD_FREE_PTR(oinfo);
1589 GOTO(out_env, rc = -ENOMEM);
1591 oa->o_id = lock->l_resource->lr_name.name[0];
1592 oa->o_seq = lock->l_resource->lr_name.name[1];
1593 oa->o_valid = OBD_MD_FLID|OBD_MD_FLGROUP;
1595 oinfo->oi_capa = BYPASS_CAPA;
1597 rc = obd_sync(&env, lock->l_export, oinfo,
1598 lock->l_policy_data.l_extent.start,
1599 lock->l_policy_data.l_extent.end, NULL);
1601 CERROR("Error %d syncing data on lock cancel\n", rc);
1604 OBD_FREE_PTR(oinfo);
1607 rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1613 static int ost_filter_recovery_request(struct ptlrpc_request *req,
1614 struct obd_device *obd, int *process)
1616 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1617 case OST_CONNECT: /* This will never get here, but for completeness. */
1618 case OST_DISCONNECT:
1629 case OBD_LOG_CANCEL:
1631 *process = target_queue_recovery_request(req, obd);
1635 DEBUG_REQ(D_WARNING, req, "not permitted during recovery");
1641 int ost_msg_check_version(struct lustre_msg *msg)
1645 switch(lustre_msg_get_opc(msg)) {
1647 case OST_DISCONNECT:
1650 case SEC_CTX_INIT_CONT:
1652 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
1654 CERROR("bad opc %u version %08x, expecting %08x\n",
1655 lustre_msg_get_opc(msg),
1656 lustre_msg_get_version(msg),
1657 LUSTRE_OBD_VERSION);
1670 case OST_QUOTACHECK:
1672 rc = lustre_msg_check_version(msg, LUSTRE_OST_VERSION);
1674 CERROR("bad opc %u version %08x, expecting %08x\n",
1675 lustre_msg_get_opc(msg),
1676 lustre_msg_get_version(msg),
1677 LUSTRE_OST_VERSION);
1682 case LDLM_BL_CALLBACK:
1683 case LDLM_CP_CALLBACK:
1684 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
1686 CERROR("bad opc %u version %08x, expecting %08x\n",
1687 lustre_msg_get_opc(msg),
1688 lustre_msg_get_version(msg),
1689 LUSTRE_DLM_VERSION);
1691 case LLOG_ORIGIN_CONNECT:
1692 case OBD_LOG_CANCEL:
1693 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
1695 CERROR("bad opc %u version %08x, expecting %08x\n",
1696 lustre_msg_get_opc(msg),
1697 lustre_msg_get_version(msg),
1698 LUSTRE_LOG_VERSION);
1700 case OST_QUOTA_ADJUST_QUNIT:
1702 CERROR("Quota adjust is deprecated as of 2.4.0\n");
1705 CERROR("Unexpected opcode %d\n", lustre_msg_get_opc(msg));
1711 struct ost_prolong_data {
1712 struct ptlrpc_request *opd_req;
1713 struct obd_export *opd_exp;
1714 struct obdo *opd_oa;
1715 struct ldlm_res_id opd_resid;
1716 struct ldlm_extent opd_extent;
1717 ldlm_mode_t opd_mode;
1718 unsigned int opd_locks;
1722 /* prolong locks for the current service time of the corresponding
1723 * portal (= OST_IO_PORTAL)
1725 static inline int prolong_timeout(struct ptlrpc_request *req)
1727 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
1730 return obd_timeout / 2;
1732 return max(at_est2timeout(at_get(&svcpt->scp_at_estimate)),
1736 static void ost_prolong_lock_one(struct ost_prolong_data *opd,
1737 struct ldlm_lock *lock)
1739 LASSERT(lock->l_export == opd->opd_exp);
1741 if (lock->l_destroyed) /* lock already cancelled */
1744 /* XXX: never try to grab resource lock here because we're inside
1745 * exp_bl_list_lock; in ldlm_lockd.c to handle waiting list we take
1746 * res lock and then exp_bl_list_lock. */
1748 if (!(lock->l_flags & LDLM_FL_AST_SENT))
1749 /* ignore locks not being cancelled */
1753 "refreshed for req x"LPU64" ext("LPU64"->"LPU64") to %ds.\n",
1754 opd->opd_req->rq_xid, opd->opd_extent.start,
1755 opd->opd_extent.end, opd->opd_timeout);
1757 /* OK. this is a possible lock the user holds doing I/O
1758 * let's refresh eviction timer for it */
1759 ldlm_refresh_waiting_lock(lock, opd->opd_timeout);
1763 static void ost_prolong_locks(struct ost_prolong_data *data)
1765 struct obd_export *exp = data->opd_exp;
1766 struct obdo *oa = data->opd_oa;
1767 struct ldlm_lock *lock;
1770 if (oa->o_valid & OBD_MD_FLHANDLE) {
1771 /* mostly a request should be covered by only one lock, try
1773 lock = ldlm_handle2lock(&oa->o_handle);
1775 /* Fast path to check if the lock covers the whole IO
1776 * region exclusively. */
1777 if (lock->l_granted_mode == LCK_PW &&
1778 ldlm_extent_contain(&lock->l_policy_data.l_extent,
1779 &data->opd_extent)) {
1781 ost_prolong_lock_one(data, lock);
1782 LDLM_LOCK_PUT(lock);
1785 LDLM_LOCK_PUT(lock);
1790 cfs_spin_lock_bh(&exp->exp_bl_list_lock);
1791 cfs_list_for_each_entry(lock, &exp->exp_bl_list, l_exp_list) {
1792 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1793 LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
1795 if (!ldlm_res_eq(&data->opd_resid, &lock->l_resource->lr_name))
1798 if (!ldlm_extent_overlap(&lock->l_policy_data.l_extent,
1802 ost_prolong_lock_one(data, lock);
1804 cfs_spin_unlock_bh(&exp->exp_bl_list_lock);
1810 * Returns 1 if the given PTLRPC matches the given LDLM locks, or 0 if it does
1813 static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req,
1814 struct ldlm_lock *lock)
1816 struct niobuf_remote *nb;
1817 struct obd_ioobj *ioo;
1819 struct ldlm_extent ext;
1822 opc = lustre_msg_get_opc(req->rq_reqmsg);
1823 LASSERT(opc == OST_READ || opc == OST_WRITE);
1825 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1826 LASSERT(ioo != NULL);
1828 nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1829 LASSERT(nb != NULL);
1831 ext.start = nb->offset;
1832 nb += ioo->ioo_bufcnt - 1;
1833 ext.end = nb->offset + nb->len - 1;
1835 LASSERT(lock->l_resource != NULL);
1836 if (!osc_res_name_eq(ioo->ioo_id, ioo->ioo_seq,
1837 &lock->l_resource->lr_name))
1841 if (opc == OST_READ)
1843 if (!(lock->l_granted_mode & mode))
1846 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
1850 * High-priority queue request check for whether the given PTLRPC request (\a
1851 * req) is blocking an LDLM lock cancel.
1853 * Returns 1 if the given given PTLRPC request (\a req) is blocking an LDLM lock
1854 * cancel, 0 if it is not, and -EFAULT if the request is malformed.
1856 * Only OST_READs, OST_WRITEs and OST_PUNCHes go on the h-p RPC queue. This
1857 * function looks only at OST_READs and OST_WRITEs.
1859 static int ost_rw_hpreq_check(struct ptlrpc_request *req)
1861 struct obd_device *obd = req->rq_export->exp_obd;
1862 struct ost_body *body;
1863 struct obd_ioobj *ioo;
1864 struct niobuf_remote *nb;
1865 struct ost_prolong_data opd = { 0 };
1870 * Use LASSERT to do sanity check because malformed RPCs should have
1871 * been filtered out in ost_hpreq_handler().
1873 opc = lustre_msg_get_opc(req->rq_reqmsg);
1874 LASSERT(opc == OST_READ || opc == OST_WRITE);
1876 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1877 LASSERT(body != NULL);
1879 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1880 LASSERT(ioo != NULL);
1882 nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1883 LASSERT(nb != NULL);
1884 LASSERT(!(nb->flags & OBD_BRW_SRVLOCK));
1886 osc_build_res_name(ioo->ioo_id, ioo->ioo_seq, &opd.opd_resid);
1890 if (opc == OST_READ)
1892 opd.opd_mode = mode;
1893 opd.opd_exp = req->rq_export;
1894 opd.opd_oa = &body->oa;
1895 opd.opd_extent.start = nb->offset;
1896 nb += ioo->ioo_bufcnt - 1;
1897 opd.opd_extent.end = nb->offset + nb->len - 1;
1898 opd.opd_timeout = prolong_timeout(req);
1900 DEBUG_REQ(D_RPCTRACE, req,
1901 "%s %s: refresh rw locks: " LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1902 obd->obd_name, cfs_current()->comm,
1903 opd.opd_resid.name[0], opd.opd_resid.name[1],
1904 opd.opd_extent.start, opd.opd_extent.end);
1906 ost_prolong_locks(&opd);
1908 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1909 obd->obd_name, opd.opd_locks, req);
1911 RETURN(opd.opd_locks);
1914 static void ost_rw_hpreq_fini(struct ptlrpc_request *req)
1916 (void)ost_rw_hpreq_check(req);
1920 * Like ost_rw_hpreq_lock_match(), but for OST_PUNCH RPCs.
1922 static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req,
1923 struct ldlm_lock *lock)
1925 struct ost_body *body;
1928 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1929 LASSERT(body != NULL);
1931 if (body->oa.o_valid & OBD_MD_FLHANDLE &&
1932 body->oa.o_handle.cookie == lock->l_handle.h_cookie)
1939 * Like ost_rw_hpreq_check(), but for OST_PUNCH RPCs.
1941 static int ost_punch_hpreq_check(struct ptlrpc_request *req)
1943 struct obd_device *obd = req->rq_export->exp_obd;
1944 struct ost_body *body;
1946 struct ost_prolong_data opd = { 0 };
1950 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1951 LASSERT(body != NULL);
1954 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS) ||
1955 !(oa->o_flags & OBD_FL_SRVLOCK));
1958 end = start + oa->o_blocks;
1961 opd.opd_mode = LCK_PW;
1962 opd.opd_exp = req->rq_export;
1964 opd.opd_extent.start = start;
1965 opd.opd_extent.end = end;
1966 if (oa->o_blocks == OBD_OBJECT_EOF)
1967 opd.opd_extent.end = OBD_OBJECT_EOF;
1968 opd.opd_timeout = prolong_timeout(req);
1970 osc_build_res_name(oa->o_id, oa->o_seq, &opd.opd_resid);
1973 "%s: refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1975 opd.opd_resid.name[0], opd.opd_resid.name[1],
1976 opd.opd_extent.start, opd.opd_extent.end);
1978 ost_prolong_locks(&opd);
1980 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1981 obd->obd_name, opd.opd_locks, req);
1983 RETURN(opd.opd_locks > 0);
1986 static void ost_punch_hpreq_fini(struct ptlrpc_request *req)
1988 (void)ost_punch_hpreq_check(req);
1991 struct ptlrpc_hpreq_ops ost_hpreq_rw = {
1992 .hpreq_lock_match = ost_rw_hpreq_lock_match,
1993 .hpreq_check = ost_rw_hpreq_check,
1994 .hpreq_fini = ost_rw_hpreq_fini
1997 struct ptlrpc_hpreq_ops ost_hpreq_punch = {
1998 .hpreq_lock_match = ost_punch_hpreq_lock_match,
1999 .hpreq_check = ost_punch_hpreq_check,
2000 .hpreq_fini = ost_punch_hpreq_fini
2003 /** Assign high priority operations to the request if needed. */
2004 static int ost_hpreq_handler(struct ptlrpc_request *req)
2007 if (req->rq_export) {
2008 int opc = lustre_msg_get_opc(req->rq_reqmsg);
2009 struct ost_body *body;
2011 if (opc == OST_READ || opc == OST_WRITE) {
2012 struct niobuf_remote *nb;
2013 struct obd_ioobj *ioo;
2014 int objcount, niocount;
2018 /* RPCs on the H-P queue can be inspected before
2019 * ost_handler() initializes their pills, so we
2020 * initialize that here. Capsule initialization is
2021 * idempotent, as is setting the pill's format (provided
2022 * it doesn't change).
2024 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2025 if (opc == OST_READ)
2026 req_capsule_set(&req->rq_pill,
2029 req_capsule_set(&req->rq_pill,
2030 &RQF_OST_BRW_WRITE);
2032 body = req_capsule_client_get(&req->rq_pill,
2035 CERROR("Missing/short ost_body\n");
2039 objcount = req_capsule_get_size(&req->rq_pill,
2043 if (objcount == 0) {
2044 CERROR("Missing/short ioobj\n");
2048 CERROR("too many ioobjs (%d)\n", objcount);
2052 ioo = req_capsule_client_get(&req->rq_pill,
2055 CERROR("Missing/short ioobj\n");
2059 rc = ost_validate_obdo(req->rq_export, &body->oa, ioo);
2061 CERROR("invalid object ids\n");
2065 for (niocount = i = 0; i < objcount; i++) {
2066 if (ioo[i].ioo_bufcnt == 0) {
2067 CERROR("ioo[%d] has zero bufcnt\n", i);
2070 niocount += ioo[i].ioo_bufcnt;
2072 if (niocount > PTLRPC_MAX_BRW_PAGES) {
2073 DEBUG_REQ(D_RPCTRACE, req,
2074 "bulk has too many pages (%d)",
2079 nb = req_capsule_client_get(&req->rq_pill,
2080 &RMF_NIOBUF_REMOTE);
2082 CERROR("Missing/short niobuf\n");
2086 if (niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
2087 req->rq_ops = &ost_hpreq_rw;
2088 } else if (opc == OST_PUNCH) {
2089 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2090 req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
2092 body = req_capsule_client_get(&req->rq_pill,
2095 CERROR("Missing/short ost_body\n");
2099 if (!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
2100 !(body->oa.o_flags & OBD_FL_SRVLOCK))
2101 req->rq_ops = &ost_hpreq_punch;
2107 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
2108 int ost_handle(struct ptlrpc_request *req)
2110 struct obd_trans_info trans_info = { 0, };
2111 struct obd_trans_info *oti = &trans_info;
2112 int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
2113 struct obd_device *obd = NULL;
2116 /* OST module is kept between remounts, but the last reference
2117 * to specific module (say, osd or ofd) kills all related keys
2118 * from the environment. so we have to refill it until the root
2119 * cause is fixed properly */
2120 lu_env_refill(req->rq_svc_thread->t_env);
2122 LASSERT(current->journal_info == NULL);
2124 /* primordial rpcs don't affect server recovery */
2125 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2127 case SEC_CTX_INIT_CONT:
2132 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2134 if (lustre_msg_get_opc(req->rq_reqmsg) != OST_CONNECT) {
2135 if (!class_connected_export(req->rq_export)) {
2136 CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
2137 lustre_msg_get_opc(req->rq_reqmsg),
2138 libcfs_id2str(req->rq_peer));
2139 req->rq_status = -ENOTCONN;
2140 GOTO(out, rc = -ENOTCONN);
2143 obd = req->rq_export->exp_obd;
2145 /* Check for aborted recovery. */
2146 if (obd->obd_recovering) {
2147 rc = ost_filter_recovery_request(req, obd,
2149 if (rc || !should_process)
2151 else if (should_process < 0) {
2152 req->rq_status = should_process;
2153 rc = ptlrpc_error(req);
2161 rc = ost_msg_check_version(req->rq_reqmsg);
2165 if (req && req->rq_reqmsg && req->rq_export &&
2166 (req->rq_export->exp_connect_flags & OBD_CONNECT_JOBSTATS))
2167 oti->oti_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
2169 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2171 CDEBUG(D_INODE, "connect\n");
2172 req_capsule_set(&req->rq_pill, &RQF_OST_CONNECT);
2173 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET))
2175 rc = target_handle_connect(req);
2176 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2))
2179 rc = ost_init_sec_level(req);
2181 rc = ost_connect_check_sptlrpc(req);
2185 case OST_DISCONNECT:
2186 CDEBUG(D_INODE, "disconnect\n");
2187 req_capsule_set(&req->rq_pill, &RQF_OST_DISCONNECT);
2188 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DISCONNECT_NET))
2190 rc = target_handle_disconnect(req);
2193 CDEBUG(D_INODE, "create\n");
2194 req_capsule_set(&req->rq_pill, &RQF_OST_CREATE);
2195 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET))
2197 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2198 GOTO(out, rc = -EROFS);
2199 rc = ost_create(req->rq_export, req, oti);
2202 CDEBUG(D_INODE, "destroy\n");
2203 req_capsule_set(&req->rq_pill, &RQF_OST_DESTROY);
2204 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DESTROY_NET))
2206 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2207 GOTO(out, rc = -EROFS);
2208 rc = ost_destroy(req->rq_export, req, oti);
2211 CDEBUG(D_INODE, "getattr\n");
2212 req_capsule_set(&req->rq_pill, &RQF_OST_GETATTR);
2213 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GETATTR_NET))
2215 rc = ost_getattr(req->rq_export, req);
2218 CDEBUG(D_INODE, "setattr\n");
2219 req_capsule_set(&req->rq_pill, &RQF_OST_SETATTR);
2220 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_NET))
2222 rc = ost_setattr(req->rq_export, req, oti);
2225 req_capsule_set(&req->rq_pill, &RQF_OST_BRW_WRITE);
2226 CDEBUG(D_INODE, "write\n");
2227 /* req->rq_request_portal would be nice, if it was set */
2228 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
2229 CERROR("%s: deny write request from %s to portal %u\n",
2230 req->rq_export->exp_obd->obd_name,
2231 obd_export_nid2str(req->rq_export),
2232 ptlrpc_req2svc(req)->srv_req_portal);
2233 GOTO(out, rc = -EPROTO);
2235 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2237 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2238 GOTO(out, rc = -ENOSPC);
2239 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2240 GOTO(out, rc = -EROFS);
2241 rc = ost_brw_write(req, oti);
2242 LASSERT(current->journal_info == NULL);
2243 /* ost_brw_write sends its own replies */
2246 req_capsule_set(&req->rq_pill, &RQF_OST_BRW_READ);
2247 CDEBUG(D_INODE, "read\n");
2248 /* req->rq_request_portal would be nice, if it was set */
2249 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
2250 CERROR("%s: deny read request from %s to portal %u\n",
2251 req->rq_export->exp_obd->obd_name,
2252 obd_export_nid2str(req->rq_export),
2253 ptlrpc_req2svc(req)->srv_req_portal);
2254 GOTO(out, rc = -EPROTO);
2256 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2258 rc = ost_brw_read(req, oti);
2259 LASSERT(current->journal_info == NULL);
2260 /* ost_brw_read sends its own replies */
2263 CDEBUG(D_INODE, "punch\n");
2264 req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
2265 if (OBD_FAIL_CHECK(OBD_FAIL_OST_PUNCH_NET))
2267 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2268 GOTO(out, rc = -EROFS);
2269 rc = ost_punch(req->rq_export, req, oti);
2272 CDEBUG(D_INODE, "statfs\n");
2273 req_capsule_set(&req->rq_pill, &RQF_OST_STATFS);
2274 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_NET))
2276 rc = ost_statfs(req);
2279 CDEBUG(D_INODE, "sync\n");
2280 req_capsule_set(&req->rq_pill, &RQF_OST_SYNC);
2281 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SYNC_NET))
2283 rc = ost_sync(req->rq_export, req, oti);
2286 DEBUG_REQ(D_INODE, req, "set_info");
2287 req_capsule_set(&req->rq_pill, &RQF_OBD_SET_INFO);
2288 rc = ost_set_info(req->rq_export, req);
2291 DEBUG_REQ(D_INODE, req, "get_info");
2292 req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC);
2293 rc = ost_get_info(req->rq_export, req);
2295 case OST_QUOTACHECK:
2296 CDEBUG(D_INODE, "quotacheck\n");
2297 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACHECK);
2298 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACHECK_NET))
2300 rc = ost_handle_quotacheck(req);
2303 CDEBUG(D_INODE, "quotactl\n");
2304 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACTL);
2305 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACTL_NET))
2307 rc = ost_handle_quotactl(req);
2310 DEBUG_REQ(D_INODE, req, "ping");
2311 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
2312 rc = target_handle_ping(req);
2314 /* FIXME - just reply status */
2315 case LLOG_ORIGIN_CONNECT:
2316 DEBUG_REQ(D_INODE, req, "log connect");
2317 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_CONNECT);
2318 rc = ost_llog_handle_connect(req->rq_export, req);
2319 req->rq_status = rc;
2320 rc = req_capsule_server_pack(&req->rq_pill);
2323 RETURN(ptlrpc_reply(req));
2324 case OBD_LOG_CANCEL:
2325 CDEBUG(D_INODE, "log cancel\n");
2326 req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL);
2327 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
2329 rc = llog_origin_handle_cancel(req);
2330 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP))
2332 req->rq_status = rc;
2333 rc = req_capsule_server_pack(&req->rq_pill);
2336 RETURN(ptlrpc_reply(req));
2338 CDEBUG(D_INODE, "enqueue\n");
2339 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
2340 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE_NET))
2342 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
2344 ldlm_server_glimpse_ast);
2345 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
2348 CDEBUG(D_INODE, "convert\n");
2349 req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT);
2350 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT_NET))
2352 rc = ldlm_handle_convert(req);
2355 CDEBUG(D_INODE, "cancel\n");
2356 req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL);
2357 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_NET))
2359 rc = ldlm_handle_cancel(req);
2361 case LDLM_BL_CALLBACK:
2362 case LDLM_CP_CALLBACK:
2363 CDEBUG(D_INODE, "callback\n");
2364 CERROR("callbacks should not happen on OST\n");
2367 CERROR("Unexpected opcode %d\n",
2368 lustre_msg_get_opc(req->rq_reqmsg));
2369 req->rq_status = -ENOTSUPP;
2370 rc = ptlrpc_error(req);
2374 LASSERT(current->journal_info == NULL);
2377 /* If we're DISCONNECTing, the export_data is already freed */
2378 if (!rc && lustre_msg_get_opc(req->rq_reqmsg) != OST_DISCONNECT)
2379 target_committed_to_req(req);
2383 oti_to_request(oti, req);
2385 target_send_reply(req, rc, fail);
2388 EXPORT_SYMBOL(ost_handle);
2390 * free per-thread pool created by ost_thread_init().
2392 static void ost_thread_done(struct ptlrpc_thread *thread)
2394 struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local
2399 LASSERT(thread != NULL);
2402 * be prepared to handle partially-initialized pools (because this is
2403 * called from ost_thread_init() for cleanup.
2405 tls = thread->t_data;
2408 thread->t_data = NULL;
2414 * initialize per-thread page pool (bug 5137).
2416 static int ost_thread_init(struct ptlrpc_thread *thread)
2418 struct ost_thread_local_cache *tls;
2422 LASSERT(thread != NULL);
2423 LASSERT(thread->t_data == NULL);
2428 thread->t_data = tls;
2432 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
2434 static struct cfs_cpt_table *ost_io_cptable;
2436 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
2437 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
2439 static struct ptlrpc_service_conf svc_conf;
2440 struct ost_obd *ost = &obd->u.ost;
2441 struct lprocfs_static_vars lvars;
2446 rc = cfs_cleanup_group_info();
2450 lprocfs_ost_init_vars(&lvars);
2451 lprocfs_obd_setup(obd, lvars.obd_vars);
2453 cfs_mutex_init(&ost->ost_health_mutex);
2455 svc_conf = (typeof(svc_conf)) {
2456 .psc_name = LUSTRE_OSS_NAME,
2457 .psc_watchdog_factor = OSS_SERVICE_WATCHDOG_FACTOR,
2459 .bc_nbufs = OST_NBUFS,
2460 .bc_buf_size = OST_BUFSIZE,
2461 .bc_req_max_size = OST_MAXREQSIZE,
2462 .bc_rep_max_size = OST_MAXREPSIZE,
2463 .bc_req_portal = OST_REQUEST_PORTAL,
2464 .bc_rep_portal = OSC_REPLY_PORTAL,
2467 .tc_thr_name = "ll_ost",
2468 .tc_thr_factor = OSS_THR_FACTOR,
2469 .tc_nthrs_init = OSS_NTHRS_INIT,
2470 .tc_nthrs_base = OSS_NTHRS_BASE,
2471 .tc_nthrs_max = OSS_NTHRS_MAX,
2472 .tc_nthrs_user = oss_num_threads,
2473 .tc_cpu_affinity = 1,
2474 .tc_ctx_tags = LCT_DT_THREAD,
2477 .cc_pattern = oss_cpts,
2480 .so_req_handler = ost_handle,
2481 .so_req_printer = target_print_req,
2482 .so_hpreq_handler = ptlrpc_hpreq_handler,
2485 ost->ost_service = ptlrpc_register_service(&svc_conf,
2486 obd->obd_proc_entry);
2487 if (IS_ERR(ost->ost_service)) {
2488 rc = PTR_ERR(ost->ost_service);
2489 CERROR("failed to start service: %d\n", rc);
2490 GOTO(out_lprocfs, rc);
2493 memset(&svc_conf, 0, sizeof(svc_conf));
2494 svc_conf = (typeof(svc_conf)) {
2495 .psc_name = "ost_create",
2496 .psc_watchdog_factor = OSS_SERVICE_WATCHDOG_FACTOR,
2498 .bc_nbufs = OST_NBUFS,
2499 .bc_buf_size = OST_BUFSIZE,
2500 .bc_req_max_size = OST_MAXREQSIZE,
2501 .bc_rep_max_size = OST_MAXREPSIZE,
2502 .bc_req_portal = OST_CREATE_PORTAL,
2503 .bc_rep_portal = OSC_REPLY_PORTAL,
2506 .tc_thr_name = "ll_ost_create",
2507 .tc_thr_factor = OSS_CR_THR_FACTOR,
2508 .tc_nthrs_init = OSS_CR_NTHRS_INIT,
2509 .tc_nthrs_base = OSS_CR_NTHRS_BASE,
2510 .tc_nthrs_max = OSS_CR_NTHRS_MAX,
2511 .tc_nthrs_user = oss_num_create_threads,
2512 .tc_cpu_affinity = 1,
2513 .tc_ctx_tags = LCT_DT_THREAD,
2516 .cc_pattern = oss_cpts,
2519 .so_req_handler = ost_handle,
2520 .so_req_printer = target_print_req,
2521 .so_hpreq_handler = NULL,
2524 ost->ost_create_service = ptlrpc_register_service(&svc_conf,
2525 obd->obd_proc_entry);
2526 if (IS_ERR(ost->ost_create_service)) {
2527 rc = PTR_ERR(ost->ost_create_service);
2528 CERROR("failed to start OST create service: %d\n", rc);
2529 GOTO(out_service, rc);
2532 mask = cfs_cpt_table->ctb_nodemask;
2533 /* event CPT feature is disabled in libcfs level by set partition
2534 * number to 1, we still want to set node affinity for io service */
2535 if (cfs_cpt_number(cfs_cpt_table) == 1 && nodes_weight(*mask) > 1) {
2539 ost_io_cptable = cfs_cpt_table_alloc(nodes_weight(*mask));
2540 for_each_node_mask(i, *mask) {
2541 if (ost_io_cptable == NULL) {
2542 CWARN("OSS failed to create CPT table\n");
2546 rc = cfs_cpt_set_node(ost_io_cptable, cpt++, i);
2548 CWARN("OSS Failed to set node %d for"
2549 "IO CPT table\n", i);
2550 cfs_cpt_table_free(ost_io_cptable);
2551 ost_io_cptable = NULL;
2557 memset(&svc_conf, 0, sizeof(svc_conf));
2558 svc_conf = (typeof(svc_conf)) {
2559 .psc_name = "ost_io",
2560 .psc_watchdog_factor = OSS_SERVICE_WATCHDOG_FACTOR,
2562 .bc_nbufs = OST_NBUFS,
2563 .bc_buf_size = OST_BUFSIZE,
2564 .bc_req_max_size = OST_MAXREQSIZE,
2565 .bc_rep_max_size = OST_MAXREPSIZE,
2566 .bc_req_portal = OST_IO_PORTAL,
2567 .bc_rep_portal = OSC_REPLY_PORTAL,
2570 .tc_thr_name = "ll_ost_io",
2571 .tc_thr_factor = OSS_THR_FACTOR,
2572 .tc_nthrs_init = OSS_NTHRS_INIT,
2573 .tc_nthrs_base = OSS_NTHRS_BASE,
2574 .tc_nthrs_max = OSS_NTHRS_MAX,
2575 .tc_nthrs_user = oss_num_threads,
2576 .tc_cpu_affinity = 1,
2577 .tc_ctx_tags = LCT_DT_THREAD,
2580 .cc_cptable = ost_io_cptable,
2581 .cc_pattern = ost_io_cptable == NULL ?
2585 .so_thr_init = ost_thread_init,
2586 .so_thr_done = ost_thread_done,
2587 .so_req_handler = ost_handle,
2588 .so_hpreq_handler = ost_hpreq_handler,
2589 .so_req_printer = target_print_req,
2590 .so_hpreq_handler = NULL,
2593 ost->ost_io_service = ptlrpc_register_service(&svc_conf,
2594 obd->obd_proc_entry);
2595 if (IS_ERR(ost->ost_io_service)) {
2596 rc = PTR_ERR(ost->ost_io_service);
2597 CERROR("failed to start OST I/O service: %d\n", rc);
2598 ost->ost_io_service = NULL;
2599 GOTO(out_create, rc);
2602 ping_evictor_start();
2607 ptlrpc_unregister_service(ost->ost_create_service);
2608 ost->ost_create_service = NULL;
2610 ptlrpc_unregister_service(ost->ost_service);
2611 ost->ost_service = NULL;
2613 lprocfs_obd_cleanup(obd);
2617 static int ost_cleanup(struct obd_device *obd)
2619 struct ost_obd *ost = &obd->u.ost;
2623 ping_evictor_stop();
2625 /* there is no recovery for OST OBD, all recovery is controlled by
2627 LASSERT(obd->obd_recovering == 0);
2628 cfs_mutex_lock(&ost->ost_health_mutex);
2629 ptlrpc_unregister_service(ost->ost_service);
2630 ptlrpc_unregister_service(ost->ost_create_service);
2631 ptlrpc_unregister_service(ost->ost_io_service);
2632 ost->ost_service = NULL;
2633 ost->ost_create_service = NULL;
2634 ost->ost_io_service = NULL;
2636 cfs_mutex_unlock(&ost->ost_health_mutex);
2638 lprocfs_obd_cleanup(obd);
2640 if (ost_io_cptable != NULL) {
2641 cfs_cpt_table_free(ost_io_cptable);
2642 ost_io_cptable = NULL;
2648 static int ost_health_check(const struct lu_env *env, struct obd_device *obd)
2650 struct ost_obd *ost = &obd->u.ost;
2653 cfs_mutex_lock(&ost->ost_health_mutex);
2654 rc |= ptlrpc_service_health_check(ost->ost_service);
2655 rc |= ptlrpc_service_health_check(ost->ost_create_service);
2656 rc |= ptlrpc_service_health_check(ost->ost_io_service);
2657 cfs_mutex_unlock(&ost->ost_health_mutex);
2660 * health_check to return 0 on healthy
2661 * and 1 on unhealthy.
2669 struct ost_thread_local_cache *ost_tls(struct ptlrpc_request *r)
2671 return (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
2674 /* use obd ops to offer management infrastructure */
2675 static struct obd_ops ost_obd_ops = {
2676 .o_owner = THIS_MODULE,
2677 .o_setup = ost_setup,
2678 .o_cleanup = ost_cleanup,
2679 .o_health_check = ost_health_check,
2683 static int __init ost_init(void)
2685 struct lprocfs_static_vars lvars;
2689 ost_page_to_corrupt = cfs_alloc_page(CFS_ALLOC_STD);
2691 lprocfs_ost_init_vars(&lvars);
2692 rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars,
2693 LUSTRE_OSS_NAME, NULL);
2695 if (ost_num_threads != 0 && oss_num_threads == 0) {
2696 LCONSOLE_INFO("ost_num_threads module parameter is deprecated, "
2697 "use oss_num_threads instead or unset both for "
2698 "dynamic thread startup\n");
2699 oss_num_threads = ost_num_threads;
2705 static void /*__exit*/ ost_exit(void)
2707 if (ost_page_to_corrupt)
2708 page_cache_release(ost_page_to_corrupt);
2710 class_unregister_type(LUSTRE_OSS_NAME);
2713 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2714 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
2715 MODULE_LICENSE("GPL");
2717 module_init(ost_init);
2718 module_exit(ost_exit);