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, Intel Corporation.
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,
214 int mode, __u64 flags)
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, LVB_T_NONE,
252 /* Helper function: release lock, if any. */
253 static void ost_lock_put(struct obd_export *exp,
254 struct lustre_handle *lh, int mode)
257 if (lustre_handle_is_used(lh))
258 ldlm_lock_decref(lh, mode);
262 static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
264 struct ost_body *body, *repbody;
265 struct obd_info *oinfo;
266 struct lustre_handle lh = { 0 };
267 struct lustre_capa *capa = NULL;
271 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
275 rc = ost_validate_obdo(exp, &body->oa, NULL);
279 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
280 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
282 CERROR("Missing capability for OST GETATTR");
287 rc = req_capsule_server_pack(&req->rq_pill);
291 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
292 repbody->oa = body->oa;
294 rc = ost_lock_get(exp, &repbody->oa, 0, OBD_OBJECT_EOF, &lh, LCK_PR, 0);
298 OBD_ALLOC_PTR(oinfo);
300 GOTO(unlock, rc = -ENOMEM);
301 oinfo->oi_oa = &repbody->oa;
302 oinfo->oi_capa = capa;
304 req->rq_status = obd_getattr(req->rq_svc_thread->t_env, exp, oinfo);
308 ost_drop_id(exp, &repbody->oa);
311 ost_lock_put(exp, &lh, LCK_PR);
315 static int ost_statfs(struct ptlrpc_request *req)
317 struct obd_statfs *osfs;
321 rc = req_capsule_server_pack(&req->rq_pill);
325 osfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
327 req->rq_status = obd_statfs(req->rq_svc_thread->t_env, req->rq_export,
329 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
331 if (req->rq_status != 0)
332 CERROR("ost: statfs failed: rc %d\n", req->rq_status);
334 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
335 req->rq_status = -EINPROGRESS;
340 static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
341 struct obd_trans_info *oti)
343 struct ost_body *body, *repbody;
347 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
351 rc = ost_validate_obdo(req->rq_export, &body->oa, NULL);
355 rc = req_capsule_server_pack(&req->rq_pill);
359 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
360 repbody->oa = body->oa;
361 oti->oti_logcookies = &body->oa.o_lcookie;
363 req->rq_status = obd_create(req->rq_svc_thread->t_env, exp,
364 &repbody->oa, NULL, oti);
365 //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
369 static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
370 struct obd_trans_info *oti)
372 struct ost_body *body, *repbody;
374 struct lustre_handle lh = {0,};
378 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
379 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
381 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
385 rc = ost_validate_obdo(exp, &body->oa, NULL);
389 if ((body->oa.o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
390 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
393 rc = req_capsule_server_pack(&req->rq_pill);
397 /* standard truncate optimization: if file body is completely
398 * destroyed, don't send data back to the server. */
399 if (body->oa.o_size == 0)
400 flags |= LDLM_AST_DISCARD_DATA;
402 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
403 repbody->oa = body->oa;
405 rc = ost_lock_get(exp, &repbody->oa, repbody->oa.o_size,
406 repbody->oa.o_blocks, &lh, LCK_PW, flags);
408 struct obd_info *oinfo;
409 struct lustre_capa *capa = NULL;
411 if (repbody->oa.o_valid & OBD_MD_FLFLAGS &&
412 repbody->oa.o_flags == OBD_FL_SRVLOCK)
414 * If OBD_FL_SRVLOCK is the only bit set in
415 * ->o_flags, clear OBD_MD_FLFLAGS to avoid falling
416 * through filter_setattr() to filter_iocontrol().
418 repbody->oa.o_valid &= ~OBD_MD_FLFLAGS;
420 if (repbody->oa.o_valid & OBD_MD_FLOSSCAPA) {
421 capa = req_capsule_client_get(&req->rq_pill,
424 CERROR("Missing capability for OST PUNCH");
425 GOTO(unlock, rc = -EFAULT);
429 OBD_ALLOC_PTR(oinfo);
431 GOTO(unlock, rc = -ENOMEM);
432 oinfo->oi_oa = &repbody->oa;
433 oinfo->oi_policy.l_extent.start = oinfo->oi_oa->o_size;
434 oinfo->oi_policy.l_extent.end = oinfo->oi_oa->o_blocks;
435 oinfo->oi_capa = capa;
436 oinfo->oi_flags = OBD_FL_PUNCH;
438 req->rq_status = obd_punch(req->rq_svc_thread->t_env, exp,
442 ost_lock_put(exp, &lh, LCK_PW);
445 ost_drop_id(exp, &repbody->oa);
449 static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req,
450 struct obd_trans_info *oti)
452 struct ost_body *body, *repbody;
453 struct obd_info *oinfo;
454 struct lustre_capa *capa = NULL;
458 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
462 rc = ost_validate_obdo(exp, &body->oa, NULL);
466 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
467 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
469 CERROR("Missing capability for OST SYNC");
474 rc = req_capsule_server_pack(&req->rq_pill);
478 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
479 repbody->oa = body->oa;
481 OBD_ALLOC_PTR(oinfo);
485 oinfo->oi_oa = &repbody->oa;
486 oinfo->oi_capa = capa;
487 oinfo->oi_jobid = oti->oti_jobid;
488 req->rq_status = obd_sync(req->rq_svc_thread->t_env, exp, oinfo,
489 repbody->oa.o_size, repbody->oa.o_blocks,
493 ost_drop_id(exp, &repbody->oa);
497 static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req,
498 struct obd_trans_info *oti)
500 struct ost_body *body, *repbody;
501 struct obd_info *oinfo;
502 struct lustre_capa *capa = NULL;
506 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
510 rc = ost_validate_obdo(req->rq_export, &body->oa, NULL);
514 rc = req_capsule_server_pack(&req->rq_pill);
518 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
519 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
521 CERROR("Missing capability for OST SETATTR");
526 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
527 repbody->oa = body->oa;
529 OBD_ALLOC_PTR(oinfo);
532 oinfo->oi_oa = &repbody->oa;
533 oinfo->oi_capa = capa;
535 req->rq_status = obd_setattr(req->rq_svc_thread->t_env, exp, oinfo,
540 ost_drop_id(exp, &repbody->oa);
544 static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc,
545 cksum_type_t cksum_type)
547 struct cfs_crypto_hash_desc *hdesc;
548 unsigned int bufsize;
550 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
553 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
555 CERROR("Unable to initialize checksum hash %s\n",
556 cfs_crypto_hash_name(cfs_alg));
557 return PTR_ERR(hdesc);
559 CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
560 for (i = 0; i < desc->bd_iov_count; i++) {
562 /* corrupt the data before we compute the checksum, to
563 * simulate a client->OST data error */
564 if (i == 0 && opc == OST_WRITE &&
565 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
566 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
567 int len = desc->bd_iov[i].kiov_len;
568 struct page *np = ost_page_to_corrupt;
569 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
572 char *ptr2 = kmap(np) + off;
574 memcpy(ptr2, ptr, len);
575 memcpy(ptr2, "bad3", min(4, len));
577 desc->bd_iov[i].kiov_page = np;
579 CERROR("can't alloc page for corruption\n");
582 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
583 desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
584 desc->bd_iov[i].kiov_len);
586 /* corrupt the data after we compute the checksum, to
587 * simulate an OST->client data error */
588 if (i == 0 && opc == OST_READ &&
589 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
590 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
591 int len = desc->bd_iov[i].kiov_len;
592 struct page *np = ost_page_to_corrupt;
593 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
596 char *ptr2 = kmap(np) + off;
598 memcpy(ptr2, ptr, len);
599 memcpy(ptr2, "bad4", min(4, len));
601 desc->bd_iov[i].kiov_page = np;
603 CERROR("can't alloc page for corruption\n");
609 err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
611 cfs_crypto_hash_final(hdesc, NULL, NULL);
616 static int ost_brw_lock_get(int mode, struct obd_export *exp,
617 struct obd_ioobj *obj, struct niobuf_remote *nb,
618 struct lustre_handle *lh)
621 int nrbufs = obj->ioo_bufcnt;
622 struct ldlm_res_id res_id;
623 ldlm_policy_data_t policy;
627 osc_build_res_name(obj->ioo_id, obj->ioo_seq, &res_id);
628 LASSERT(mode == LCK_PR || mode == LCK_PW);
629 LASSERT(!lustre_handle_is_used(lh));
631 if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
634 for (i = 1; i < nrbufs; i ++)
635 if ((nb[0].flags & OBD_BRW_SRVLOCK) !=
636 (nb[i].flags & OBD_BRW_SRVLOCK))
639 policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
640 policy.l_extent.end = (nb[nrbufs - 1].offset +
641 nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
643 RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
644 LDLM_EXTENT, &policy, mode, &flags,
645 ldlm_blocking_ast, ldlm_completion_ast,
646 ldlm_glimpse_ast, NULL, 0, LVB_T_NONE,
650 static void ost_brw_lock_put(int mode,
651 struct obd_ioobj *obj, struct niobuf_remote *niob,
652 struct lustre_handle *lh)
655 LASSERT(mode == LCK_PR || mode == LCK_PW);
656 LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
657 lustre_handle_is_used(lh));
658 if (lustre_handle_is_used(lh))
659 ldlm_lock_decref(lh, mode);
663 /* Allocate thread local buffers if needed */
664 static struct ost_thread_local_cache *ost_tls_get(struct ptlrpc_request *r)
666 struct ost_thread_local_cache *tls =
667 (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
669 /* In normal mode of operation an I/O request is serviced only
670 * by ll_ost_io threads each of them has own tls buffers allocated by
671 * ost_io_thread_init().
672 * During recovery, an I/O request may be queued until any of the ost
673 * service threads process it. Not necessary it should be one of
674 * ll_ost_io threads. In that case we dynamically allocating tls
675 * buffers for the request service time. */
676 if (unlikely(tls == NULL)) {
677 LASSERT(r->rq_export->exp_in_recovery);
681 r->rq_svc_thread->t_data = tls;
687 /* Free thread local buffers if they were allocated only for servicing
688 * this one request */
689 static void ost_tls_put(struct ptlrpc_request *r)
691 struct ost_thread_local_cache *tls =
692 (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
694 if (unlikely(tls->temporary)) {
696 r->rq_svc_thread->t_data = NULL;
700 static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
702 struct ptlrpc_bulk_desc *desc = NULL;
703 struct obd_export *exp = req->rq_export;
704 struct niobuf_remote *remote_nb;
705 struct niobuf_local *local_nb;
706 struct obd_ioobj *ioo;
707 struct ost_body *body, *repbody;
708 struct lustre_capa *capa = NULL;
709 struct l_wait_info lwi;
710 struct lustre_handle lockh = { 0 };
711 int niocount, npages, nob = 0, rc, i;
713 struct ost_thread_local_cache *tls;
716 req->rq_bulk_read = 1;
718 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
719 GOTO(out, rc = -EIO);
721 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
723 /* Check if there is eviction in progress, and if so, wait for it to
725 if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
726 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
727 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
728 !cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress),
732 GOTO(out, rc = -ENOTCONN);
734 /* ost_body, ioobj & noibuf_remote are verified and swabbed in
735 * ost_rw_hpreq_check(). */
736 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
738 GOTO(out, rc = -EFAULT);
741 * A req_capsule_X_get_array(pill, field, ptr_to_element_count) function
742 * would be useful here and wherever we get &RMF_OBD_IOOBJ and
743 * &RMF_NIOBUF_REMOTE.
745 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
747 GOTO(out, rc = -EFAULT);
749 rc = ost_validate_obdo(exp, &body->oa, ioo);
753 niocount = ioo->ioo_bufcnt;
754 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
755 if (remote_nb == NULL)
756 GOTO(out, rc = -EFAULT);
758 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
759 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
761 CERROR("Missing capability for OST BRW READ");
762 GOTO(out, rc = -EFAULT);
766 rc = req_capsule_server_pack(&req->rq_pill);
770 tls = ost_tls_get(req);
772 GOTO(out_bulk, rc = -ENOMEM);
773 local_nb = tls->local;
775 rc = ost_brw_lock_get(LCK_PR, exp, ioo, remote_nb, &lockh);
780 * If getting the lock took more time than
781 * client was willing to wait, drop it. b=11330
783 if (cfs_time_current_sec() > req->rq_deadline ||
784 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
786 CERROR("Dropping timed-out read from %s because locking"
787 "object "LPX64" took %ld seconds (limit was %ld).\n",
788 libcfs_id2str(req->rq_peer), ioo->ioo_id,
789 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
790 req->rq_deadline - req->rq_arrival_time.tv_sec);
791 GOTO(out_lock, rc = -ETIMEDOUT);
794 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
795 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
797 npages = OST_THREAD_POOL_SIZE;
798 rc = obd_preprw(req->rq_svc_thread->t_env, OBD_BRW_READ, exp,
799 &repbody->oa, 1, ioo, remote_nb, &npages, local_nb,
804 desc = ptlrpc_prep_bulk_exp(req, npages,
805 BULK_PUT_SOURCE, OST_BULK_PORTAL);
807 GOTO(out_commitrw, rc = -ENOMEM);
810 for (i = 0; i < npages; i++) {
811 int page_rc = local_nb[i].rc;
813 if (page_rc < 0) { /* error */
819 if (page_rc != 0) { /* some data! */
820 LASSERT (local_nb[i].page != NULL);
821 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
822 local_nb[i].lnb_page_offset,
826 if (page_rc != local_nb[i].len) { /* short read */
827 /* All subsequent pages should be 0 */
829 LASSERT(local_nb[i].rc == 0);
834 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
835 cksum_type_t cksum_type =
836 cksum_type_unpack(repbody->oa.o_valid & OBD_MD_FLFLAGS ?
837 repbody->oa.o_flags : 0);
838 repbody->oa.o_flags = cksum_type_pack(cksum_type);
839 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
840 repbody->oa.o_cksum = ost_checksum_bulk(desc, OST_READ,cksum_type);
841 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
842 repbody->oa.o_cksum);
844 repbody->oa.o_valid = 0;
846 /* We're finishing using body->oa as an input variable */
848 /* Check if client was evicted while we were doing i/o before touching
851 if (likely(!CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2)))
852 rc = target_bulk_io(exp, desc, &lwi);
857 /* Must commit after prep above in all cases */
858 rc = obd_commitrw(req->rq_svc_thread->t_env, OBD_BRW_READ, exp,
859 &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
863 ost_drop_id(exp, &repbody->oa);
866 ost_brw_lock_put(LCK_PR, ioo, remote_nb, &lockh);
870 if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
871 ptlrpc_free_bulk_nopin(desc);
875 req->rq_status = nob;
876 ptlrpc_lprocfs_brw(req, nob);
877 target_committed_to_req(req);
879 } else if (!no_reply) {
880 /* Only reply if there was no comms problem with bulk */
881 target_committed_to_req(req);
885 /* reply out callback would free */
886 ptlrpc_req_drop_rs(req);
887 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
888 "client will retry: rc %d\n",
889 exp->exp_obd->obd_name,
890 obd_uuid2str(&exp->exp_client_uuid),
891 obd_export_nid2str(exp), rc);
893 /* send a bulk after reply to simulate a network delay or reordering
895 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
897 struct l_wait_info lwi1;
899 CDEBUG(D_INFO, "reorder BULK\n");
900 cfs_waitq_init(&waitq);
902 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
903 l_wait_event(waitq, 0, &lwi1);
904 rc = target_bulk_io(exp, desc, &lwi);
905 ptlrpc_free_bulk_nopin(desc);
911 static void ost_warn_on_cksum(struct ptlrpc_request *req,
912 struct ptlrpc_bulk_desc *desc,
913 struct niobuf_local *local_nb, int npages,
914 obd_count client_cksum, obd_count server_cksum,
917 struct obd_export *exp = req->rq_export;
918 struct ost_body *body;
922 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
923 LASSERT (body != NULL);
925 if (req->rq_peer.nid == desc->bd_sender) {
929 router = libcfs_nid2str(desc->bd_sender);
933 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
934 client_cksum, server_cksum);
938 LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
939 DFID" object "LPU64"/"LPU64" extent ["LPU64"-"LPU64
940 "]: client csum %x, server csum %x\n",
941 exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
943 body->oa.o_valid & OBD_MD_FLFID ?
944 body->oa.o_parent_seq : (__u64)0,
945 body->oa.o_valid & OBD_MD_FLFID ?
946 body->oa.o_parent_oid : 0,
947 body->oa.o_valid & OBD_MD_FLFID ?
948 body->oa.o_parent_ver : 0,
950 body->oa.o_valid & OBD_MD_FLGROUP ?
951 body->oa.o_seq : (__u64)0,
952 local_nb[0].lnb_file_offset,
953 local_nb[npages-1].lnb_file_offset +
954 local_nb[npages-1].len - 1,
955 client_cksum, server_cksum);
958 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
960 struct ptlrpc_bulk_desc *desc = NULL;
961 struct obd_export *exp = req->rq_export;
962 struct niobuf_remote *remote_nb;
963 struct niobuf_local *local_nb;
964 struct obd_ioobj *ioo;
965 struct ost_body *body, *repbody;
966 struct l_wait_info lwi;
967 struct lustre_handle lockh = {0};
968 struct lustre_capa *capa = NULL;
970 int objcount, niocount, npages;
972 obd_count client_cksum = 0, server_cksum = 0;
973 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
974 int no_reply = 0, mmap = 0;
975 __u32 o_uid = 0, o_gid = 0;
976 struct ost_thread_local_cache *tls;
979 req->rq_bulk_write = 1;
981 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
982 GOTO(out, rc = -EIO);
983 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
984 GOTO(out, rc = -EFAULT);
986 /* pause before transaction has been started */
987 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
989 /* ost_body, ioobj & noibuf_remote are verified and swabbed in
990 * ost_rw_hpreq_check(). */
991 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
993 GOTO(out, rc = -EFAULT);
995 objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
996 RCL_CLIENT) / sizeof(*ioo);
997 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
999 GOTO(out, rc = -EFAULT);
1001 rc = ost_validate_obdo(exp, &body->oa, ioo);
1005 for (niocount = i = 0; i < objcount; i++)
1006 niocount += ioo[i].ioo_bufcnt;
1009 * It'd be nice to have a capsule function to indicate how many elements
1010 * there were in a buffer for an RMF that's declared to be an array.
1011 * It's easy enough to compute the number of elements here though.
1013 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1014 if (remote_nb == NULL || niocount != (req_capsule_get_size(&req->rq_pill,
1015 &RMF_NIOBUF_REMOTE, RCL_CLIENT) / sizeof(*remote_nb)))
1016 GOTO(out, rc = -EFAULT);
1018 if ((remote_nb[0].flags & OBD_BRW_MEMALLOC) &&
1019 (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
1020 cfs_memory_pressure_set();
1022 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
1023 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
1025 CERROR("Missing capability for OST BRW WRITE");
1026 GOTO(out, rc = -EFAULT);
1030 req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
1031 niocount * sizeof(*rcs));
1032 rc = req_capsule_server_pack(&req->rq_pill);
1035 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
1036 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
1038 tls = ost_tls_get(req);
1040 GOTO(out_bulk, rc = -ENOMEM);
1041 local_nb = tls->local;
1043 rc = ost_brw_lock_get(LCK_PW, exp, ioo, remote_nb, &lockh);
1048 * If getting the lock took more time than
1049 * client was willing to wait, drop it. b=11330
1051 if (cfs_time_current_sec() > req->rq_deadline ||
1052 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1054 CERROR("Dropping timed-out write from %s because locking "
1055 "object "LPX64" took %ld seconds (limit was %ld).\n",
1056 libcfs_id2str(req->rq_peer), ioo->ioo_id,
1057 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1058 req->rq_deadline - req->rq_arrival_time.tv_sec);
1059 GOTO(out_lock, rc = -ETIMEDOUT);
1062 /* obd_preprw clobbers oa->valid, so save what we need */
1063 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1064 client_cksum = body->oa.o_cksum;
1065 if (body->oa.o_valid & OBD_MD_FLFLAGS)
1066 cksum_type = cksum_type_unpack(body->oa.o_flags);
1068 if (body->oa.o_valid & OBD_MD_FLFLAGS && body->oa.o_flags & OBD_FL_MMAP)
1071 /* Because we already sync grant info with client when reconnect,
1072 * grant info will be cleared for resent req, then fed_grant and
1073 * total_grant will not be modified in following preprw_write */
1074 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1075 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1076 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1079 if (exp_connect_rmtclient(exp)) {
1080 o_uid = body->oa.o_uid;
1081 o_gid = body->oa.o_gid;
1084 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1085 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
1087 npages = OST_THREAD_POOL_SIZE;
1088 rc = obd_preprw(req->rq_svc_thread->t_env, OBD_BRW_WRITE, exp,
1089 &repbody->oa, objcount, ioo, remote_nb, &npages,
1090 local_nb, oti, capa);
1094 desc = ptlrpc_prep_bulk_exp(req, npages,
1095 BULK_GET_SINK, OST_BULK_PORTAL);
1097 GOTO(skip_transfer, rc = -ENOMEM);
1099 /* NB Having prepped, we must commit... */
1101 for (i = 0; i < npages; i++)
1102 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
1103 local_nb[i].lnb_page_offset,
1106 rc = sptlrpc_svc_prep_bulk(req, desc);
1110 rc = target_bulk_io(exp, desc, &lwi);
1114 if (client_cksum != 0 && rc == 0) {
1115 static int cksum_counter;
1116 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1117 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
1118 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
1119 server_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1120 repbody->oa.o_cksum = server_cksum;
1122 if (unlikely(client_cksum != server_cksum)) {
1123 ost_warn_on_cksum(req, desc, local_nb, npages,
1124 client_cksum, server_cksum, mmap);
1127 } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){
1128 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1129 cksum_counter, libcfs_id2str(req->rq_peer),
1134 /* Must commit after prep above in all cases */
1135 rc = obd_commitrw(req->rq_svc_thread->t_env, OBD_BRW_WRITE, exp,
1136 &repbody->oa, objcount, ioo, remote_nb, npages,
1138 if (rc == -ENOTCONN)
1139 /* quota acquire process has been given up because
1140 * either the client has been evicted or the client
1141 * has timed out the request already */
1144 if (exp_connect_rmtclient(exp)) {
1145 repbody->oa.o_uid = o_uid;
1146 repbody->oa.o_gid = o_gid;
1150 * Disable sending mtime back to the client. If the client locked the
1151 * whole object, then it has already updated the mtime on its side,
1152 * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
1154 repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
1159 /* set per-requested niobuf return codes */
1160 for (i = j = 0; i < niocount; i++) {
1161 int len = remote_nb[i].len;
1166 LASSERT(j < npages);
1167 if (local_nb[j].rc < 0)
1168 rcs[i] = local_nb[j].rc;
1169 len -= local_nb[j].len;
1174 LASSERT(j == npages);
1175 ptlrpc_lprocfs_brw(req, nob);
1179 ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh);
1184 ptlrpc_free_bulk_nopin(desc);
1187 oti_to_request(oti, req);
1188 target_committed_to_req(req);
1189 rc = ptlrpc_reply(req);
1190 } else if (!no_reply) {
1191 /* Only reply if there was no comms problem with bulk */
1192 target_committed_to_req(req);
1193 req->rq_status = rc;
1196 /* reply out callback would free */
1197 ptlrpc_req_drop_rs(req);
1198 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
1199 "client will retry: rc %d\n",
1200 exp->exp_obd->obd_name,
1201 obd_uuid2str(&exp->exp_client_uuid),
1202 obd_export_nid2str(exp), rc);
1204 cfs_memory_pressure_clr();
1209 * Implementation of OST_SET_INFO.
1211 * OST_SET_INFO is like ioctl(): heavily overloaded. Specifically, it takes a
1212 * "key" and a value RPC buffers as arguments, with the value's contents
1213 * interpreted according to the key.
1215 * Value types that need swabbing have swabbing done explicitly, either here or
1216 * in functions called from here. This should be corrected: all swabbing should
1217 * be done in the capsule abstraction, as that will then allow us to move
1218 * swabbing exclusively to the client without having to modify server code
1219 * outside the capsule abstraction's implementation itself. To correct this
1220 * will require minor changes to the capsule abstraction; see the comments for
1221 * req_capsule_extend() in layout.c.
1223 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
1225 struct ost_body *body = NULL, *repbody;
1226 char *key, *val = NULL;
1227 int keylen, vallen, rc = 0;
1228 int is_grant_shrink = 0;
1231 key = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
1233 DEBUG_REQ(D_HA, req, "no set_info key");
1236 keylen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_KEY,
1239 vallen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_VAL,
1242 if ((is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK)))
1243 /* In this case the value is actually an RMF_OST_BODY, so we
1244 * transmutate the type of this PTLRPC */
1245 req_capsule_extend(&req->rq_pill, &RQF_OST_SET_GRANT_INFO);
1247 rc = req_capsule_server_pack(&req->rq_pill);
1252 if (is_grant_shrink) {
1253 body = req_capsule_client_get(&req->rq_pill,
1258 repbody = req_capsule_server_get(&req->rq_pill,
1260 memcpy(repbody, body, sizeof(*body));
1261 val = (char*)repbody;
1263 val = req_capsule_client_get(&req->rq_pill,
1268 if (KEY_IS(KEY_EVICT_BY_NID)) {
1270 obd_export_evict_by_nid(exp->exp_obd, val);
1272 } else if (KEY_IS(KEY_MDS_CONN) && ptlrpc_req_need_swab(req)) {
1273 if (vallen < sizeof(__u32))
1275 __swab32s((__u32 *)val);
1278 /* OBD will also check if KEY_IS(KEY_GRANT_SHRINK), and will cast val to
1279 * a struct ost_body * value */
1280 rc = obd_set_info_async(req->rq_svc_thread->t_env, exp, keylen,
1281 key, vallen, val, NULL);
1283 lustre_msg_set_status(req->rq_repmsg, 0);
1287 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
1290 int keylen, replylen, rc = 0;
1291 struct req_capsule *pill = &req->rq_pill;
1294 /* this common part for get_info rpc */
1295 key = req_capsule_client_get(pill, &RMF_SETINFO_KEY);
1297 DEBUG_REQ(D_HA, req, "no get_info key");
1300 keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT);
1302 if (KEY_IS(KEY_FIEMAP)) {
1303 struct ll_fiemap_info_key *fm_key = key;
1306 rc = ost_validate_obdo(exp, &fm_key->oa, NULL);
1311 rc = obd_get_info(req->rq_svc_thread->t_env, exp, keylen, key,
1312 &replylen, NULL, NULL);
1316 req_capsule_set_size(pill, &RMF_GENERIC_DATA,
1317 RCL_SERVER, replylen);
1319 rc = req_capsule_server_pack(pill);
1323 reply = req_capsule_server_get(pill, &RMF_GENERIC_DATA);
1327 /* call again to fill in the reply buffer */
1328 rc = obd_get_info(req->rq_svc_thread->t_env, exp, keylen, key,
1329 &replylen, reply, NULL);
1331 lustre_msg_set_status(req->rq_repmsg, 0);
1335 static int ost_handle_quotactl(struct ptlrpc_request *req)
1337 struct obd_quotactl *oqctl, *repoqc;
1341 oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1343 GOTO(out, rc = -EPROTO);
1345 rc = req_capsule_server_pack(&req->rq_pill);
1349 repoqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1350 req->rq_status = obd_quotactl(req->rq_export, oqctl);
1357 static int ost_handle_quotacheck(struct ptlrpc_request *req)
1359 struct obd_quotactl *oqctl;
1363 oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1367 rc = req_capsule_server_pack(&req->rq_pill);
1371 /* deprecated, not used any more */
1372 req->rq_status = -EOPNOTSUPP;
1373 RETURN(-EOPNOTSUPP);
1376 static int ost_llog_handle_connect(struct obd_export *exp,
1377 struct ptlrpc_request *req)
1379 struct llogd_conn_body *body;
1383 body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_CONN_BODY);
1384 rc = obd_llog_connect(exp, body);
1388 #define ost_init_sec_none(reply, exp) \
1390 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT | \
1391 OBD_CONNECT_RMT_CLIENT_FORCE | \
1392 OBD_CONNECT_OSS_CAPA); \
1393 spin_lock(&exp->exp_lock); \
1394 exp->exp_connect_flags = reply->ocd_connect_flags; \
1395 spin_unlock(&exp->exp_lock); \
1398 static int ost_init_sec_level(struct ptlrpc_request *req)
1400 struct obd_export *exp = req->rq_export;
1401 struct req_capsule *pill = &req->rq_pill;
1402 struct obd_device *obd = exp->exp_obd;
1403 struct filter_obd *filter = &obd->u.filter;
1404 char *client = libcfs_nid2str(req->rq_peer.nid);
1405 struct obd_connect_data *data, *reply;
1409 data = req_capsule_client_get(pill, &RMF_CONNECT_DATA);
1410 reply = req_capsule_server_get(pill, &RMF_CONNECT_DATA);
1411 if (data == NULL || reply == NULL)
1414 /* connection from MDT is always trusted */
1415 if (req->rq_auth_usr_mdt) {
1416 ost_init_sec_none(reply, exp);
1420 /* no GSS support case */
1421 if (!req->rq_auth_gss) {
1422 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1423 CWARN("client %s -> target %s does not user GSS, "
1424 "can not run under security level %d.\n",
1425 client, obd->obd_name, filter->fo_sec_level);
1428 ost_init_sec_none(reply, exp);
1433 /* old version case */
1434 if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
1435 !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
1436 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1437 CWARN("client %s -> target %s uses old version, "
1438 "can not run under security level %d.\n",
1439 client, obd->obd_name, filter->fo_sec_level);
1442 CWARN("client %s -> target %s uses old version, "
1443 "run under security level %d.\n",
1444 client, obd->obd_name, filter->fo_sec_level);
1445 ost_init_sec_none(reply, exp);
1450 remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
1452 if (!req->rq_auth_remote)
1453 CDEBUG(D_SEC, "client (local realm) %s -> target %s "
1454 "asked to be remote.\n", client, obd->obd_name);
1455 } else if (req->rq_auth_remote) {
1457 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
1458 "as remote by default.\n", client, obd->obd_name);
1462 if (!filter->fo_fl_oss_capa) {
1463 CDEBUG(D_SEC, "client %s -> target %s is set as remote,"
1464 " but OSS capabilities are not enabled: %d.\n",
1465 client, obd->obd_name, filter->fo_fl_oss_capa);
1470 switch (filter->fo_sec_level) {
1471 case LUSTRE_SEC_NONE:
1473 ost_init_sec_none(reply, exp);
1476 CDEBUG(D_SEC, "client %s -> target %s is set as remote, "
1477 "can not run under security level %d.\n",
1478 client, obd->obd_name, filter->fo_sec_level);
1481 case LUSTRE_SEC_REMOTE:
1483 ost_init_sec_none(reply, exp);
1485 case LUSTRE_SEC_ALL:
1487 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
1488 OBD_CONNECT_RMT_CLIENT_FORCE);
1489 if (!filter->fo_fl_oss_capa)
1490 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
1492 spin_lock(&exp->exp_lock);
1493 exp->exp_connect_flags = reply->ocd_connect_flags;
1494 spin_unlock(&exp->exp_lock);
1506 * this should be done in filter_connect()/filter_reconnect(), but
1507 * we can't obtain information like NID, which stored in incoming
1508 * request, thus can't decide what flavor to use. so we do it here.
1510 * This hack should be removed after the OST stack be rewritten, just
1511 * like what we are doing in mdt_obd_connect()/mdt_obd_reconnect().
1513 static int ost_connect_check_sptlrpc(struct ptlrpc_request *req)
1515 struct obd_export *exp = req->rq_export;
1516 struct filter_obd *filter = &exp->exp_obd->u.filter;
1517 struct sptlrpc_flavor flvr;
1520 if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
1521 LUSTRE_ECHO_NAME) == 0)) {
1522 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
1526 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
1527 read_lock(&filter->fo_sptlrpc_lock);
1528 sptlrpc_target_choose_flavor(&filter->fo_sptlrpc_rset,
1532 read_unlock(&filter->fo_sptlrpc_lock);
1534 spin_lock(&exp->exp_lock);
1536 exp->exp_sp_peer = req->rq_sp_from;
1537 exp->exp_flvr = flvr;
1539 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
1540 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
1541 CERROR("unauthorized rpc flavor %x from %s, "
1542 "expect %x\n", req->rq_flvr.sf_rpc,
1543 libcfs_nid2str(req->rq_peer.nid),
1544 exp->exp_flvr.sf_rpc);
1548 spin_unlock(&exp->exp_lock);
1550 if (exp->exp_sp_peer != req->rq_sp_from) {
1551 CERROR("RPC source %s doesn't match %s\n",
1552 sptlrpc_part2name(req->rq_sp_from),
1553 sptlrpc_part2name(exp->exp_sp_peer));
1556 rc = sptlrpc_target_export_check(exp, req);
1563 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1565 int ost_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1566 void *data, int flag)
1569 __u32 sync_lock_cancel = 0;
1570 __u32 len = sizeof(sync_lock_cancel);
1575 rc = lu_env_init(&env, LCT_DT_THREAD);
1576 if (unlikely(rc != 0))
1579 rc = obd_get_info(&env, lock->l_export, sizeof(KEY_SYNC_LOCK_CANCEL),
1580 KEY_SYNC_LOCK_CANCEL, &len, &sync_lock_cancel, NULL);
1581 if (rc == 0 && flag == LDLM_CB_CANCELING &&
1582 (lock->l_granted_mode & (LCK_PW|LCK_GROUP)) &&
1583 (sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1584 (sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1585 lock->l_flags & LDLM_FL_CBPENDING))) {
1586 struct obd_info *oinfo;
1590 OBD_ALLOC_PTR(oinfo);
1592 GOTO(out_env, rc = -ENOMEM);
1595 OBD_FREE_PTR(oinfo);
1596 GOTO(out_env, rc = -ENOMEM);
1598 oa->o_id = lock->l_resource->lr_name.name[0];
1599 oa->o_seq = lock->l_resource->lr_name.name[1];
1600 oa->o_valid = OBD_MD_FLID|OBD_MD_FLGROUP;
1602 oinfo->oi_capa = BYPASS_CAPA;
1604 rc = obd_sync(&env, lock->l_export, oinfo,
1605 lock->l_policy_data.l_extent.start,
1606 lock->l_policy_data.l_extent.end, NULL);
1608 CERROR("Error %d syncing data on lock cancel\n", rc);
1611 OBD_FREE_PTR(oinfo);
1614 rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1620 static int ost_filter_recovery_request(struct ptlrpc_request *req,
1621 struct obd_device *obd, int *process)
1623 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1624 case OST_CONNECT: /* This will never get here, but for completeness. */
1625 case OST_DISCONNECT:
1636 case OBD_LOG_CANCEL:
1638 *process = target_queue_recovery_request(req, obd);
1642 DEBUG_REQ(D_WARNING, req, "not permitted during recovery");
1648 int ost_msg_check_version(struct lustre_msg *msg)
1652 switch(lustre_msg_get_opc(msg)) {
1654 case OST_DISCONNECT:
1657 case SEC_CTX_INIT_CONT:
1659 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
1661 CERROR("bad opc %u version %08x, expecting %08x\n",
1662 lustre_msg_get_opc(msg),
1663 lustre_msg_get_version(msg),
1664 LUSTRE_OBD_VERSION);
1677 case OST_QUOTACHECK:
1679 rc = lustre_msg_check_version(msg, LUSTRE_OST_VERSION);
1681 CERROR("bad opc %u version %08x, expecting %08x\n",
1682 lustre_msg_get_opc(msg),
1683 lustre_msg_get_version(msg),
1684 LUSTRE_OST_VERSION);
1689 case LDLM_BL_CALLBACK:
1690 case LDLM_CP_CALLBACK:
1691 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
1693 CERROR("bad opc %u version %08x, expecting %08x\n",
1694 lustre_msg_get_opc(msg),
1695 lustre_msg_get_version(msg),
1696 LUSTRE_DLM_VERSION);
1698 case LLOG_ORIGIN_CONNECT:
1699 case OBD_LOG_CANCEL:
1700 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
1702 CERROR("bad opc %u version %08x, expecting %08x\n",
1703 lustre_msg_get_opc(msg),
1704 lustre_msg_get_version(msg),
1705 LUSTRE_LOG_VERSION);
1707 case OST_QUOTA_ADJUST_QUNIT:
1709 CERROR("Quota adjust is deprecated as of 2.4.0\n");
1712 CERROR("Unexpected opcode %d\n", lustre_msg_get_opc(msg));
1718 struct ost_prolong_data {
1719 struct ptlrpc_request *opd_req;
1720 struct obd_export *opd_exp;
1721 struct obdo *opd_oa;
1722 struct ldlm_res_id opd_resid;
1723 struct ldlm_extent opd_extent;
1724 ldlm_mode_t opd_mode;
1725 unsigned int opd_locks;
1729 /* prolong locks for the current service time of the corresponding
1730 * portal (= OST_IO_PORTAL)
1732 static inline int prolong_timeout(struct ptlrpc_request *req)
1734 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
1737 return obd_timeout / 2;
1739 return max(at_est2timeout(at_get(&svcpt->scp_at_estimate)),
1743 static void ost_prolong_lock_one(struct ost_prolong_data *opd,
1744 struct ldlm_lock *lock)
1746 LASSERT(lock->l_export == opd->opd_exp);
1748 if (lock->l_destroyed) /* lock already cancelled */
1751 /* XXX: never try to grab resource lock here because we're inside
1752 * exp_bl_list_lock; in ldlm_lockd.c to handle waiting list we take
1753 * res lock and then exp_bl_list_lock. */
1755 if (!(lock->l_flags & LDLM_FL_AST_SENT))
1756 /* ignore locks not being cancelled */
1760 "refreshed for req x"LPU64" ext("LPU64"->"LPU64") to %ds.\n",
1761 opd->opd_req->rq_xid, opd->opd_extent.start,
1762 opd->opd_extent.end, opd->opd_timeout);
1764 /* OK. this is a possible lock the user holds doing I/O
1765 * let's refresh eviction timer for it */
1766 ldlm_refresh_waiting_lock(lock, opd->opd_timeout);
1770 static void ost_prolong_locks(struct ost_prolong_data *data)
1772 struct obd_export *exp = data->opd_exp;
1773 struct obdo *oa = data->opd_oa;
1774 struct ldlm_lock *lock;
1777 if (oa->o_valid & OBD_MD_FLHANDLE) {
1778 /* mostly a request should be covered by only one lock, try
1780 lock = ldlm_handle2lock(&oa->o_handle);
1782 /* Fast path to check if the lock covers the whole IO
1783 * region exclusively. */
1784 if (lock->l_granted_mode == LCK_PW &&
1785 ldlm_extent_contain(&lock->l_policy_data.l_extent,
1786 &data->opd_extent)) {
1788 ost_prolong_lock_one(data, lock);
1789 LDLM_LOCK_PUT(lock);
1792 LDLM_LOCK_PUT(lock);
1797 spin_lock_bh(&exp->exp_bl_list_lock);
1798 cfs_list_for_each_entry(lock, &exp->exp_bl_list, l_exp_list) {
1799 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1800 LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
1802 if (!ldlm_res_eq(&data->opd_resid, &lock->l_resource->lr_name))
1805 if (!ldlm_extent_overlap(&lock->l_policy_data.l_extent,
1809 ost_prolong_lock_one(data, lock);
1811 spin_unlock_bh(&exp->exp_bl_list_lock);
1817 * Returns 1 if the given PTLRPC matches the given LDLM locks, or 0 if it does
1820 static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req,
1821 struct ldlm_lock *lock)
1823 struct niobuf_remote *nb;
1824 struct obd_ioobj *ioo;
1826 struct ldlm_extent ext;
1829 opc = lustre_msg_get_opc(req->rq_reqmsg);
1830 LASSERT(opc == OST_READ || opc == OST_WRITE);
1832 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1833 LASSERT(ioo != NULL);
1835 nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1836 LASSERT(nb != NULL);
1838 ext.start = nb->offset;
1839 nb += ioo->ioo_bufcnt - 1;
1840 ext.end = nb->offset + nb->len - 1;
1842 LASSERT(lock->l_resource != NULL);
1843 if (!osc_res_name_eq(ioo->ioo_id, ioo->ioo_seq,
1844 &lock->l_resource->lr_name))
1848 if (opc == OST_READ)
1850 if (!(lock->l_granted_mode & mode))
1853 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
1857 * High-priority queue request check for whether the given PTLRPC request (\a
1858 * req) is blocking an LDLM lock cancel.
1860 * Returns 1 if the given given PTLRPC request (\a req) is blocking an LDLM lock
1861 * cancel, 0 if it is not, and -EFAULT if the request is malformed.
1863 * Only OST_READs, OST_WRITEs and OST_PUNCHes go on the h-p RPC queue. This
1864 * function looks only at OST_READs and OST_WRITEs.
1866 static int ost_rw_hpreq_check(struct ptlrpc_request *req)
1868 struct obd_device *obd = req->rq_export->exp_obd;
1869 struct ost_body *body;
1870 struct obd_ioobj *ioo;
1871 struct niobuf_remote *nb;
1872 struct ost_prolong_data opd = { 0 };
1877 * Use LASSERT to do sanity check because malformed RPCs should have
1878 * been filtered out in ost_hpreq_handler().
1880 opc = lustre_msg_get_opc(req->rq_reqmsg);
1881 LASSERT(opc == OST_READ || opc == OST_WRITE);
1883 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1884 LASSERT(body != NULL);
1886 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1887 LASSERT(ioo != NULL);
1889 nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1890 LASSERT(nb != NULL);
1891 LASSERT(!(nb->flags & OBD_BRW_SRVLOCK));
1893 osc_build_res_name(ioo->ioo_id, ioo->ioo_seq, &opd.opd_resid);
1897 if (opc == OST_READ)
1899 opd.opd_mode = mode;
1900 opd.opd_exp = req->rq_export;
1901 opd.opd_oa = &body->oa;
1902 opd.opd_extent.start = nb->offset;
1903 nb += ioo->ioo_bufcnt - 1;
1904 opd.opd_extent.end = nb->offset + nb->len - 1;
1905 opd.opd_timeout = prolong_timeout(req);
1907 DEBUG_REQ(D_RPCTRACE, req,
1908 "%s %s: refresh rw locks: " LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1909 obd->obd_name, cfs_current()->comm,
1910 opd.opd_resid.name[0], opd.opd_resid.name[1],
1911 opd.opd_extent.start, opd.opd_extent.end);
1913 ost_prolong_locks(&opd);
1915 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1916 obd->obd_name, opd.opd_locks, req);
1918 RETURN(opd.opd_locks);
1921 static void ost_rw_hpreq_fini(struct ptlrpc_request *req)
1923 (void)ost_rw_hpreq_check(req);
1927 * Like ost_rw_hpreq_lock_match(), but for OST_PUNCH RPCs.
1929 static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req,
1930 struct ldlm_lock *lock)
1932 struct ost_body *body;
1935 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1936 LASSERT(body != NULL);
1938 if (body->oa.o_valid & OBD_MD_FLHANDLE &&
1939 body->oa.o_handle.cookie == lock->l_handle.h_cookie)
1946 * Like ost_rw_hpreq_check(), but for OST_PUNCH RPCs.
1948 static int ost_punch_hpreq_check(struct ptlrpc_request *req)
1950 struct obd_device *obd = req->rq_export->exp_obd;
1951 struct ost_body *body;
1953 struct ost_prolong_data opd = { 0 };
1957 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1958 LASSERT(body != NULL);
1961 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS) ||
1962 !(oa->o_flags & OBD_FL_SRVLOCK));
1965 end = start + oa->o_blocks;
1968 opd.opd_mode = LCK_PW;
1969 opd.opd_exp = req->rq_export;
1971 opd.opd_extent.start = start;
1972 opd.opd_extent.end = end;
1973 if (oa->o_blocks == OBD_OBJECT_EOF)
1974 opd.opd_extent.end = OBD_OBJECT_EOF;
1975 opd.opd_timeout = prolong_timeout(req);
1977 osc_build_res_name(oa->o_id, oa->o_seq, &opd.opd_resid);
1980 "%s: refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1982 opd.opd_resid.name[0], opd.opd_resid.name[1],
1983 opd.opd_extent.start, opd.opd_extent.end);
1985 ost_prolong_locks(&opd);
1987 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1988 obd->obd_name, opd.opd_locks, req);
1990 RETURN(opd.opd_locks > 0);
1993 static void ost_punch_hpreq_fini(struct ptlrpc_request *req)
1995 (void)ost_punch_hpreq_check(req);
1998 struct ptlrpc_hpreq_ops ost_hpreq_rw = {
1999 .hpreq_lock_match = ost_rw_hpreq_lock_match,
2000 .hpreq_check = ost_rw_hpreq_check,
2001 .hpreq_fini = ost_rw_hpreq_fini
2004 struct ptlrpc_hpreq_ops ost_hpreq_punch = {
2005 .hpreq_lock_match = ost_punch_hpreq_lock_match,
2006 .hpreq_check = ost_punch_hpreq_check,
2007 .hpreq_fini = ost_punch_hpreq_fini
2010 /** Assign high priority operations to the request if needed. */
2011 static int ost_io_hpreq_handler(struct ptlrpc_request *req)
2014 if (req->rq_export) {
2015 int opc = lustre_msg_get_opc(req->rq_reqmsg);
2016 struct ost_body *body;
2018 if (opc == OST_READ || opc == OST_WRITE) {
2019 struct niobuf_remote *nb;
2020 struct obd_ioobj *ioo;
2021 int objcount, niocount;
2025 /* RPCs on the H-P queue can be inspected before
2026 * ost_handler() initializes their pills, so we
2027 * initialize that here. Capsule initialization is
2028 * idempotent, as is setting the pill's format (provided
2029 * it doesn't change).
2031 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2032 if (opc == OST_READ)
2033 req_capsule_set(&req->rq_pill,
2036 req_capsule_set(&req->rq_pill,
2037 &RQF_OST_BRW_WRITE);
2039 body = req_capsule_client_get(&req->rq_pill,
2042 CERROR("Missing/short ost_body\n");
2046 objcount = req_capsule_get_size(&req->rq_pill,
2050 if (objcount == 0) {
2051 CERROR("Missing/short ioobj\n");
2055 CERROR("too many ioobjs (%d)\n", objcount);
2059 ioo = req_capsule_client_get(&req->rq_pill,
2062 CERROR("Missing/short ioobj\n");
2066 rc = ost_validate_obdo(req->rq_export, &body->oa, ioo);
2068 CERROR("invalid object ids\n");
2072 for (niocount = i = 0; i < objcount; i++) {
2073 if (ioo[i].ioo_bufcnt == 0) {
2074 CERROR("ioo[%d] has zero bufcnt\n", i);
2077 niocount += ioo[i].ioo_bufcnt;
2079 if (niocount > PTLRPC_MAX_BRW_PAGES) {
2080 DEBUG_REQ(D_RPCTRACE, req,
2081 "bulk has too many pages (%d)",
2086 nb = req_capsule_client_get(&req->rq_pill,
2087 &RMF_NIOBUF_REMOTE);
2089 CERROR("Missing/short niobuf\n");
2093 if (niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
2094 req->rq_ops = &ost_hpreq_rw;
2095 } else if (opc == OST_PUNCH) {
2096 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2097 req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
2099 body = req_capsule_client_get(&req->rq_pill,
2102 CERROR("Missing/short ost_body\n");
2106 if (!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
2107 !(body->oa.o_flags & OBD_FL_SRVLOCK))
2108 req->rq_ops = &ost_hpreq_punch;
2114 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
2115 int ost_handle(struct ptlrpc_request *req)
2117 struct obd_trans_info trans_info = { 0, };
2118 struct obd_trans_info *oti = &trans_info;
2119 int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
2120 struct obd_device *obd = NULL;
2123 /* OST module is kept between remounts, but the last reference
2124 * to specific module (say, osd or ofd) kills all related keys
2125 * from the environment. so we have to refill it until the root
2126 * cause is fixed properly */
2127 lu_env_refill(req->rq_svc_thread->t_env);
2129 LASSERT(current->journal_info == NULL);
2131 /* primordial rpcs don't affect server recovery */
2132 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2134 case SEC_CTX_INIT_CONT:
2139 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2141 if (lustre_msg_get_opc(req->rq_reqmsg) != OST_CONNECT) {
2142 if (!class_connected_export(req->rq_export)) {
2143 CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
2144 lustre_msg_get_opc(req->rq_reqmsg),
2145 libcfs_id2str(req->rq_peer));
2146 req->rq_status = -ENOTCONN;
2147 GOTO(out, rc = -ENOTCONN);
2150 obd = req->rq_export->exp_obd;
2152 /* Check for aborted recovery. */
2153 if (obd->obd_recovering) {
2154 rc = ost_filter_recovery_request(req, obd,
2156 if (rc || !should_process)
2158 else if (should_process < 0) {
2159 req->rq_status = should_process;
2160 rc = ptlrpc_error(req);
2168 rc = ost_msg_check_version(req->rq_reqmsg);
2172 if (req && req->rq_reqmsg && req->rq_export &&
2173 (req->rq_export->exp_connect_flags & OBD_CONNECT_JOBSTATS))
2174 oti->oti_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
2176 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2178 CDEBUG(D_INODE, "connect\n");
2179 req_capsule_set(&req->rq_pill, &RQF_OST_CONNECT);
2180 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET))
2182 rc = target_handle_connect(req);
2183 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2))
2186 rc = ost_init_sec_level(req);
2188 rc = ost_connect_check_sptlrpc(req);
2192 case OST_DISCONNECT:
2193 CDEBUG(D_INODE, "disconnect\n");
2194 req_capsule_set(&req->rq_pill, &RQF_OST_DISCONNECT);
2195 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DISCONNECT_NET))
2197 rc = target_handle_disconnect(req);
2200 CDEBUG(D_INODE, "create\n");
2201 req_capsule_set(&req->rq_pill, &RQF_OST_CREATE);
2202 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET))
2204 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2205 GOTO(out, rc = -EROFS);
2206 rc = ost_create(req->rq_export, req, oti);
2209 CDEBUG(D_INODE, "destroy\n");
2210 req_capsule_set(&req->rq_pill, &RQF_OST_DESTROY);
2211 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DESTROY_NET))
2213 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2214 GOTO(out, rc = -EROFS);
2215 rc = ost_destroy(req->rq_export, req, oti);
2218 CDEBUG(D_INODE, "getattr\n");
2219 req_capsule_set(&req->rq_pill, &RQF_OST_GETATTR);
2220 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GETATTR_NET))
2222 rc = ost_getattr(req->rq_export, req);
2225 CDEBUG(D_INODE, "setattr\n");
2226 req_capsule_set(&req->rq_pill, &RQF_OST_SETATTR);
2227 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_NET))
2229 rc = ost_setattr(req->rq_export, req, oti);
2232 req_capsule_set(&req->rq_pill, &RQF_OST_BRW_WRITE);
2233 CDEBUG(D_INODE, "write\n");
2234 /* req->rq_request_portal would be nice, if it was set */
2235 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
2236 CERROR("%s: deny write request from %s to portal %u\n",
2237 req->rq_export->exp_obd->obd_name,
2238 obd_export_nid2str(req->rq_export),
2239 ptlrpc_req2svc(req)->srv_req_portal);
2240 GOTO(out, rc = -EPROTO);
2242 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2244 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2245 GOTO(out, rc = -ENOSPC);
2246 if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
2247 GOTO(out, rc = -EROFS);
2248 rc = ost_brw_write(req, oti);
2249 LASSERT(current->journal_info == NULL);
2250 /* ost_brw_write sends its own replies */
2253 req_capsule_set(&req->rq_pill, &RQF_OST_BRW_READ);
2254 CDEBUG(D_INODE, "read\n");
2255 /* req->rq_request_portal would be nice, if it was set */
2256 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
2257 CERROR("%s: deny read request from %s to portal %u\n",
2258 req->rq_export->exp_obd->obd_name,
2259 obd_export_nid2str(req->rq_export),
2260 ptlrpc_req2svc(req)->srv_req_portal);
2261 GOTO(out, rc = -EPROTO);
2263 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2265 rc = ost_brw_read(req, oti);
2266 LASSERT(current->journal_info == NULL);
2267 /* ost_brw_read sends its own replies */
2270 CDEBUG(D_INODE, "punch\n");
2271 req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
2272 if (OBD_FAIL_CHECK(OBD_FAIL_OST_PUNCH_NET))
2274 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2275 GOTO(out, rc = -EROFS);
2276 rc = ost_punch(req->rq_export, req, oti);
2279 CDEBUG(D_INODE, "statfs\n");
2280 req_capsule_set(&req->rq_pill, &RQF_OST_STATFS);
2281 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_NET))
2283 rc = ost_statfs(req);
2286 CDEBUG(D_INODE, "sync\n");
2287 req_capsule_set(&req->rq_pill, &RQF_OST_SYNC);
2288 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SYNC_NET))
2290 rc = ost_sync(req->rq_export, req, oti);
2293 DEBUG_REQ(D_INODE, req, "set_info");
2294 req_capsule_set(&req->rq_pill, &RQF_OBD_SET_INFO);
2295 rc = ost_set_info(req->rq_export, req);
2298 DEBUG_REQ(D_INODE, req, "get_info");
2299 req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC);
2300 rc = ost_get_info(req->rq_export, req);
2302 case OST_QUOTACHECK:
2303 CDEBUG(D_INODE, "quotacheck\n");
2304 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACHECK);
2305 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACHECK_NET))
2307 rc = ost_handle_quotacheck(req);
2310 CDEBUG(D_INODE, "quotactl\n");
2311 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACTL);
2312 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACTL_NET))
2314 rc = ost_handle_quotactl(req);
2317 DEBUG_REQ(D_INODE, req, "ping");
2318 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
2319 rc = target_handle_ping(req);
2321 /* FIXME - just reply status */
2322 case LLOG_ORIGIN_CONNECT:
2323 DEBUG_REQ(D_INODE, req, "log connect");
2324 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_CONNECT);
2325 rc = ost_llog_handle_connect(req->rq_export, req);
2326 req->rq_status = rc;
2327 rc = req_capsule_server_pack(&req->rq_pill);
2330 RETURN(ptlrpc_reply(req));
2331 case OBD_LOG_CANCEL:
2332 CDEBUG(D_INODE, "log cancel\n");
2333 req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL);
2334 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
2336 rc = llog_origin_handle_cancel(req);
2337 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP))
2339 req->rq_status = rc;
2340 rc = req_capsule_server_pack(&req->rq_pill);
2343 RETURN(ptlrpc_reply(req));
2345 CDEBUG(D_INODE, "enqueue\n");
2346 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
2347 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE_NET))
2349 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
2351 ldlm_server_glimpse_ast);
2352 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
2355 CDEBUG(D_INODE, "convert\n");
2356 req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT);
2357 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT_NET))
2359 rc = ldlm_handle_convert(req);
2362 CDEBUG(D_INODE, "cancel\n");
2363 req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL);
2364 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_NET))
2366 rc = ldlm_handle_cancel(req);
2368 case LDLM_BL_CALLBACK:
2369 case LDLM_CP_CALLBACK:
2370 CDEBUG(D_INODE, "callback\n");
2371 CERROR("callbacks should not happen on OST\n");
2374 CERROR("Unexpected opcode %d\n",
2375 lustre_msg_get_opc(req->rq_reqmsg));
2376 req->rq_status = -ENOTSUPP;
2377 rc = ptlrpc_error(req);
2381 LASSERT(current->journal_info == NULL);
2384 /* If we're DISCONNECTing, the export_data is already freed */
2385 if (!rc && lustre_msg_get_opc(req->rq_reqmsg) != OST_DISCONNECT)
2386 target_committed_to_req(req);
2390 oti_to_request(oti, req);
2392 target_send_reply(req, rc, fail);
2395 EXPORT_SYMBOL(ost_handle);
2398 * free per-thread pool created by ost_io_thread_init().
2400 static void ost_io_thread_done(struct ptlrpc_thread *thread)
2402 struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local
2407 LASSERT(thread != NULL);
2410 * be prepared to handle partially-initialized pools (because this is
2411 * called from ost_io_thread_init() for cleanup.
2413 tls = thread->t_data;
2416 thread->t_data = NULL;
2422 * initialize per-thread page pool (bug 5137).
2424 static int ost_io_thread_init(struct ptlrpc_thread *thread)
2426 struct ost_thread_local_cache *tls;
2430 LASSERT(thread != NULL);
2431 LASSERT(thread->t_data == NULL);
2436 thread->t_data = tls;
2440 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
2442 static struct cfs_cpt_table *ost_io_cptable;
2444 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
2445 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
2447 static struct ptlrpc_service_conf svc_conf;
2448 struct ost_obd *ost = &obd->u.ost;
2449 struct lprocfs_static_vars lvars;
2454 rc = cfs_cleanup_group_info();
2458 lprocfs_ost_init_vars(&lvars);
2459 lprocfs_obd_setup(obd, lvars.obd_vars);
2461 mutex_init(&ost->ost_health_mutex);
2463 svc_conf = (typeof(svc_conf)) {
2464 .psc_name = LUSTRE_OSS_NAME,
2465 .psc_watchdog_factor = OSS_SERVICE_WATCHDOG_FACTOR,
2467 .bc_nbufs = OST_NBUFS,
2468 .bc_buf_size = OST_BUFSIZE,
2469 .bc_req_max_size = OST_MAXREQSIZE,
2470 .bc_rep_max_size = OST_MAXREPSIZE,
2471 .bc_req_portal = OST_REQUEST_PORTAL,
2472 .bc_rep_portal = OSC_REPLY_PORTAL,
2475 .tc_thr_name = "ll_ost",
2476 .tc_thr_factor = OSS_THR_FACTOR,
2477 .tc_nthrs_init = OSS_NTHRS_INIT,
2478 .tc_nthrs_base = OSS_NTHRS_BASE,
2479 .tc_nthrs_max = OSS_NTHRS_MAX,
2480 .tc_nthrs_user = oss_num_threads,
2481 .tc_cpu_affinity = 1,
2482 .tc_ctx_tags = LCT_DT_THREAD,
2485 .cc_pattern = oss_cpts,
2488 .so_req_handler = ost_handle,
2489 .so_req_printer = target_print_req,
2490 .so_hpreq_handler = ptlrpc_hpreq_handler,
2493 ost->ost_service = ptlrpc_register_service(&svc_conf,
2494 obd->obd_proc_entry);
2495 if (IS_ERR(ost->ost_service)) {
2496 rc = PTR_ERR(ost->ost_service);
2497 CERROR("failed to start service: %d\n", rc);
2498 GOTO(out_lprocfs, rc);
2501 memset(&svc_conf, 0, sizeof(svc_conf));
2502 svc_conf = (typeof(svc_conf)) {
2503 .psc_name = "ost_create",
2504 .psc_watchdog_factor = OSS_SERVICE_WATCHDOG_FACTOR,
2506 .bc_nbufs = OST_NBUFS,
2507 .bc_buf_size = OST_BUFSIZE,
2508 .bc_req_max_size = OST_MAXREQSIZE,
2509 .bc_rep_max_size = OST_MAXREPSIZE,
2510 .bc_req_portal = OST_CREATE_PORTAL,
2511 .bc_rep_portal = OSC_REPLY_PORTAL,
2514 .tc_thr_name = "ll_ost_create",
2515 .tc_thr_factor = OSS_CR_THR_FACTOR,
2516 .tc_nthrs_init = OSS_CR_NTHRS_INIT,
2517 .tc_nthrs_base = OSS_CR_NTHRS_BASE,
2518 .tc_nthrs_max = OSS_CR_NTHRS_MAX,
2519 .tc_nthrs_user = oss_num_create_threads,
2520 .tc_cpu_affinity = 1,
2521 .tc_ctx_tags = LCT_DT_THREAD,
2524 .cc_pattern = oss_cpts,
2527 .so_req_handler = ost_handle,
2528 .so_req_printer = target_print_req,
2531 ost->ost_create_service = ptlrpc_register_service(&svc_conf,
2532 obd->obd_proc_entry);
2533 if (IS_ERR(ost->ost_create_service)) {
2534 rc = PTR_ERR(ost->ost_create_service);
2535 CERROR("failed to start OST create service: %d\n", rc);
2536 GOTO(out_service, rc);
2539 mask = cfs_cpt_table->ctb_nodemask;
2540 /* event CPT feature is disabled in libcfs level by set partition
2541 * number to 1, we still want to set node affinity for io service */
2542 if (cfs_cpt_number(cfs_cpt_table) == 1 && nodes_weight(*mask) > 1) {
2546 ost_io_cptable = cfs_cpt_table_alloc(nodes_weight(*mask));
2547 for_each_node_mask(i, *mask) {
2548 if (ost_io_cptable == NULL) {
2549 CWARN("OSS failed to create CPT table\n");
2553 rc = cfs_cpt_set_node(ost_io_cptable, cpt++, i);
2555 CWARN("OSS Failed to set node %d for"
2556 "IO CPT table\n", i);
2557 cfs_cpt_table_free(ost_io_cptable);
2558 ost_io_cptable = NULL;
2564 memset(&svc_conf, 0, sizeof(svc_conf));
2565 svc_conf = (typeof(svc_conf)) {
2566 .psc_name = "ost_io",
2567 .psc_watchdog_factor = OSS_SERVICE_WATCHDOG_FACTOR,
2569 .bc_nbufs = OST_NBUFS,
2570 .bc_buf_size = OST_BUFSIZE,
2571 .bc_req_max_size = OST_MAXREQSIZE,
2572 .bc_rep_max_size = OST_MAXREPSIZE,
2573 .bc_req_portal = OST_IO_PORTAL,
2574 .bc_rep_portal = OSC_REPLY_PORTAL,
2577 .tc_thr_name = "ll_ost_io",
2578 .tc_thr_factor = OSS_THR_FACTOR,
2579 .tc_nthrs_init = OSS_NTHRS_INIT,
2580 .tc_nthrs_base = OSS_NTHRS_BASE,
2581 .tc_nthrs_max = OSS_NTHRS_MAX,
2582 .tc_nthrs_user = oss_num_threads,
2583 .tc_cpu_affinity = 1,
2584 .tc_ctx_tags = LCT_DT_THREAD,
2587 .cc_cptable = ost_io_cptable,
2588 .cc_pattern = ost_io_cptable == NULL ?
2592 .so_thr_init = ost_io_thread_init,
2593 .so_thr_done = ost_io_thread_done,
2594 .so_req_handler = ost_handle,
2595 .so_hpreq_handler = ost_io_hpreq_handler,
2596 .so_req_printer = target_print_req,
2599 ost->ost_io_service = ptlrpc_register_service(&svc_conf,
2600 obd->obd_proc_entry);
2601 if (IS_ERR(ost->ost_io_service)) {
2602 rc = PTR_ERR(ost->ost_io_service);
2603 CERROR("failed to start OST I/O service: %d\n", rc);
2604 ost->ost_io_service = NULL;
2605 GOTO(out_create, rc);
2608 ping_evictor_start();
2613 ptlrpc_unregister_service(ost->ost_create_service);
2614 ost->ost_create_service = NULL;
2616 ptlrpc_unregister_service(ost->ost_service);
2617 ost->ost_service = NULL;
2619 lprocfs_obd_cleanup(obd);
2623 static int ost_cleanup(struct obd_device *obd)
2625 struct ost_obd *ost = &obd->u.ost;
2629 ping_evictor_stop();
2631 /* there is no recovery for OST OBD, all recovery is controlled by
2633 LASSERT(obd->obd_recovering == 0);
2634 mutex_lock(&ost->ost_health_mutex);
2635 ptlrpc_unregister_service(ost->ost_service);
2636 ptlrpc_unregister_service(ost->ost_create_service);
2637 ptlrpc_unregister_service(ost->ost_io_service);
2638 ost->ost_service = NULL;
2639 ost->ost_create_service = NULL;
2640 ost->ost_io_service = NULL;
2642 mutex_unlock(&ost->ost_health_mutex);
2644 lprocfs_obd_cleanup(obd);
2646 if (ost_io_cptable != NULL) {
2647 cfs_cpt_table_free(ost_io_cptable);
2648 ost_io_cptable = NULL;
2654 static int ost_health_check(const struct lu_env *env, struct obd_device *obd)
2656 struct ost_obd *ost = &obd->u.ost;
2659 mutex_lock(&ost->ost_health_mutex);
2660 rc |= ptlrpc_service_health_check(ost->ost_service);
2661 rc |= ptlrpc_service_health_check(ost->ost_create_service);
2662 rc |= ptlrpc_service_health_check(ost->ost_io_service);
2663 mutex_unlock(&ost->ost_health_mutex);
2666 * health_check to return 0 on healthy
2667 * and 1 on unhealthy.
2675 struct ost_thread_local_cache *ost_tls(struct ptlrpc_request *r)
2677 return (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
2680 /* use obd ops to offer management infrastructure */
2681 static struct obd_ops ost_obd_ops = {
2682 .o_owner = THIS_MODULE,
2683 .o_setup = ost_setup,
2684 .o_cleanup = ost_cleanup,
2685 .o_health_check = ost_health_check,
2689 static int __init ost_init(void)
2691 struct lprocfs_static_vars lvars;
2695 ost_page_to_corrupt = cfs_alloc_page(CFS_ALLOC_STD);
2697 lprocfs_ost_init_vars(&lvars);
2698 rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars,
2699 LUSTRE_OSS_NAME, NULL);
2701 if (ost_num_threads != 0 && oss_num_threads == 0) {
2702 LCONSOLE_INFO("ost_num_threads module parameter is deprecated, "
2703 "use oss_num_threads instead or unset both for "
2704 "dynamic thread startup\n");
2705 oss_num_threads = ost_num_threads;
2711 static void /*__exit*/ ost_exit(void)
2713 if (ost_page_to_corrupt)
2714 page_cache_release(ost_page_to_corrupt);
2716 class_unregister_type(LUSTRE_OSS_NAME);
2719 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2720 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
2721 MODULE_LICENSE("GPL");
2723 module_init(ost_init);
2724 module_exit(ost_exit);