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 <lustre_quota.h>
55 #include "ost_internal.h"
57 static int oss_num_threads;
58 CFS_MODULE_PARM(oss_num_threads, "i", int, 0444,
59 "number of OSS service threads to start");
61 static int ost_num_threads;
62 CFS_MODULE_PARM(ost_num_threads, "i", int, 0444,
63 "number of OST service threads to start (deprecated)");
65 static int oss_num_create_threads;
66 CFS_MODULE_PARM(oss_num_create_threads, "i", int, 0444,
67 "number of OSS create threads to start");
69 static char *oss_cpts;
70 CFS_MODULE_PARM(oss_cpts, "s", charp, 0444,
71 "CPU partitions OSS threads should run on");
73 static char *oss_io_cpts;
74 CFS_MODULE_PARM(oss_io_cpts, "s", charp, 0444,
75 "CPU partitions OSS IO threads should run on");
78 * Do not return server-side uid/gid to remote client
80 static void ost_drop_id(struct obd_export *exp, struct obdo *oa)
82 if (exp_connect_rmtclient(exp)) {
85 oa->o_valid &= ~(OBD_MD_FLUID | OBD_MD_FLGID);
90 * Validate oa from client.
91 * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
93 * a. for single MDS seq = FID_SEQ_OST_MDT0,
94 * b. for CMD, seq = FID_SEQ_OST_MDT0, FID_SEQ_OST_MDT1 - FID_SEQ_OST_MAX
96 static int ost_validate_obdo(struct obd_export *exp, struct obdo *oa,
97 struct obd_ioobj *ioobj)
99 if (oa != NULL && !(oa->o_valid & OBD_MD_FLGROUP)) {
100 oa->o_seq = FID_SEQ_OST_MDT0;
102 ioobj->ioo_seq = FID_SEQ_OST_MDT0;
103 /* remove fid_seq_is_rsvd() after FID-on-OST allows SEQ > 9 */
104 } else if (oa == NULL || !(fid_seq_is_rsvd(oa->o_seq) ||
105 fid_seq_is_mdt0(oa->o_seq))) {
106 CERROR("%s: client %s sent invalid object "POSTID"\n",
107 exp->exp_obd->obd_name, obd_export_nid2str(exp),
108 oa ? oa->o_id : -1, oa ? oa->o_seq : -1);
111 obdo_from_ostid(oa, &oa->o_oi);
113 ioobj_from_obdo(ioobj, oa);
117 void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req)
119 struct oti_req_ack_lock *ack_lock;
125 if (req->rq_repmsg) {
126 __u64 versions[PTLRPC_NUM_VERSIONS] = { 0 };
127 lustre_msg_set_transno(req->rq_repmsg, oti->oti_transno);
128 versions[0] = oti->oti_pre_version;
129 lustre_msg_set_versions(req->rq_repmsg, versions);
131 req->rq_transno = oti->oti_transno;
133 /* XXX 4 == entries in oti_ack_locks??? */
134 for (ack_lock = oti->oti_ack_locks, i = 0; i < 4; i++, ack_lock++) {
137 /* XXX not even calling target_send_reply in some cases... */
138 ptlrpc_save_lock (req, &ack_lock->lock, ack_lock->mode, 0);
142 static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req,
143 struct obd_trans_info *oti)
145 struct ost_body *body, *repbody;
146 struct lustre_capa *capa = NULL;
150 /* Get the request body */
151 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
155 if (body->oa.o_id == 0)
158 rc = ost_validate_obdo(exp, &body->oa, NULL);
162 /* If there's a DLM request, cancel the locks mentioned in it*/
163 if (req_capsule_field_present(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT)) {
164 struct ldlm_request *dlm;
166 dlm = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
169 ldlm_request_cancel(req, dlm, 0);
172 /* If there's a capability, get it */
173 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
174 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
176 CERROR("Missing capability for OST DESTROY");
181 /* Prepare the reply */
182 rc = req_capsule_server_pack(&req->rq_pill);
186 /* Get the log cancellation cookie */
187 if (body->oa.o_valid & OBD_MD_FLCOOKIE)
188 oti->oti_logcookies = &body->oa.o_lcookie;
190 /* Finish the reply */
191 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
192 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
194 /* Do the destroy and set the reply status accordingly */
195 req->rq_status = obd_destroy(req->rq_svc_thread->t_env, exp,
196 &repbody->oa, NULL, oti, NULL, capa);
201 * Helper function for getting server side [start, start+count] DLM lock
202 * if asked by client.
204 static int ost_lock_get(struct obd_export *exp, struct obdo *oa,
205 __u64 start, __u64 count, struct lustre_handle *lh,
208 struct ldlm_res_id res_id;
209 ldlm_policy_data_t policy;
210 __u64 end = start + count;
214 LASSERT(!lustre_handle_is_used(lh));
215 /* o_id and o_gr are used for localizing resource, if client miss to set
216 * them, do not trigger ASSERTION. */
217 if (unlikely((oa->o_valid & (OBD_MD_FLID | OBD_MD_FLGROUP)) !=
218 (OBD_MD_FLID | OBD_MD_FLGROUP)))
221 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
222 !(oa->o_flags & OBD_FL_SRVLOCK))
225 osc_build_res_name(oa->o_id, oa->o_seq, &res_id);
226 CDEBUG(D_INODE, "OST-side extent lock.\n");
228 policy.l_extent.start = start & CFS_PAGE_MASK;
230 /* If ->o_blocks is EOF it means "lock till the end of the
231 * file". Otherwise, it's size of a hole being punched (in bytes) */
232 if (count == OBD_OBJECT_EOF || end < start)
233 policy.l_extent.end = OBD_OBJECT_EOF;
235 policy.l_extent.end = end | ~CFS_PAGE_MASK;
237 RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
238 LDLM_EXTENT, &policy, mode, &flags,
239 ldlm_blocking_ast, ldlm_completion_ast,
240 ldlm_glimpse_ast, NULL, 0, NULL, lh));
243 /* Helper function: release lock, if any. */
244 static void ost_lock_put(struct obd_export *exp,
245 struct lustre_handle *lh, int mode)
248 if (lustre_handle_is_used(lh))
249 ldlm_lock_decref(lh, mode);
253 static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
255 struct ost_body *body, *repbody;
256 struct obd_info *oinfo;
257 struct lustre_handle lh = { 0 };
258 struct lustre_capa *capa = NULL;
262 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
266 rc = ost_validate_obdo(exp, &body->oa, NULL);
270 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
271 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
273 CERROR("Missing capability for OST GETATTR");
278 rc = req_capsule_server_pack(&req->rq_pill);
282 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
283 repbody->oa = body->oa;
285 rc = ost_lock_get(exp, &repbody->oa, 0, OBD_OBJECT_EOF, &lh, LCK_PR, 0);
289 OBD_ALLOC_PTR(oinfo);
291 GOTO(unlock, rc = -ENOMEM);
292 oinfo->oi_oa = &repbody->oa;
293 oinfo->oi_capa = capa;
295 req->rq_status = obd_getattr(req->rq_svc_thread->t_env, exp, oinfo);
299 ost_drop_id(exp, &repbody->oa);
302 ost_lock_put(exp, &lh, LCK_PR);
306 static int ost_statfs(struct ptlrpc_request *req)
308 struct obd_statfs *osfs;
312 rc = req_capsule_server_pack(&req->rq_pill);
316 osfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
318 req->rq_status = obd_statfs(req->rq_svc_thread->t_env, req->rq_export,
320 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
322 if (req->rq_status != 0)
323 CERROR("ost: statfs failed: rc %d\n", req->rq_status);
325 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
326 req->rq_status = -EINPROGRESS;
331 static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
332 struct obd_trans_info *oti)
334 struct ost_body *body, *repbody;
338 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
342 rc = ost_validate_obdo(req->rq_export, &body->oa, NULL);
346 rc = req_capsule_server_pack(&req->rq_pill);
350 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
351 repbody->oa = body->oa;
352 oti->oti_logcookies = &body->oa.o_lcookie;
354 req->rq_status = obd_create(req->rq_svc_thread->t_env, exp,
355 &repbody->oa, NULL, oti);
356 //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
360 static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
361 struct obd_trans_info *oti)
363 struct ost_body *body, *repbody;
365 struct lustre_handle lh = {0,};
368 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
369 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
371 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
375 rc = ost_validate_obdo(exp, &body->oa, NULL);
379 if ((body->oa.o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
380 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
383 rc = req_capsule_server_pack(&req->rq_pill);
387 /* standard truncate optimization: if file body is completely
388 * destroyed, don't send data back to the server. */
389 if (body->oa.o_size == 0)
390 flags |= LDLM_AST_DISCARD_DATA;
392 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
393 repbody->oa = body->oa;
395 rc = ost_lock_get(exp, &repbody->oa, repbody->oa.o_size,
396 repbody->oa.o_blocks, &lh, LCK_PW, flags);
398 struct obd_info *oinfo;
399 struct lustre_capa *capa = NULL;
401 if (repbody->oa.o_valid & OBD_MD_FLFLAGS &&
402 repbody->oa.o_flags == OBD_FL_SRVLOCK)
404 * If OBD_FL_SRVLOCK is the only bit set in
405 * ->o_flags, clear OBD_MD_FLFLAGS to avoid falling
406 * through filter_setattr() to filter_iocontrol().
408 repbody->oa.o_valid &= ~OBD_MD_FLFLAGS;
410 if (repbody->oa.o_valid & OBD_MD_FLOSSCAPA) {
411 capa = req_capsule_client_get(&req->rq_pill,
414 CERROR("Missing capability for OST PUNCH");
415 GOTO(unlock, rc = -EFAULT);
419 OBD_ALLOC_PTR(oinfo);
421 GOTO(unlock, rc = -ENOMEM);
422 oinfo->oi_oa = &repbody->oa;
423 oinfo->oi_policy.l_extent.start = oinfo->oi_oa->o_size;
424 oinfo->oi_policy.l_extent.end = oinfo->oi_oa->o_blocks;
425 oinfo->oi_capa = capa;
426 oinfo->oi_flags = OBD_FL_PUNCH;
428 req->rq_status = obd_punch(req->rq_svc_thread->t_env, exp,
432 ost_lock_put(exp, &lh, LCK_PW);
435 ost_drop_id(exp, &repbody->oa);
439 static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req,
440 struct obd_trans_info *oti)
442 struct ost_body *body, *repbody;
443 struct obd_info *oinfo;
444 struct lustre_capa *capa = NULL;
448 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
452 rc = ost_validate_obdo(exp, &body->oa, NULL);
456 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
457 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
459 CERROR("Missing capability for OST SYNC");
464 rc = req_capsule_server_pack(&req->rq_pill);
468 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
469 repbody->oa = body->oa;
471 OBD_ALLOC_PTR(oinfo);
475 oinfo->oi_oa = &repbody->oa;
476 oinfo->oi_capa = capa;
477 oinfo->oi_jobid = oti->oti_jobid;
478 req->rq_status = obd_sync(req->rq_svc_thread->t_env, exp, oinfo,
479 repbody->oa.o_size, repbody->oa.o_blocks,
483 ost_drop_id(exp, &repbody->oa);
487 static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req,
488 struct obd_trans_info *oti)
490 struct ost_body *body, *repbody;
491 struct obd_info *oinfo;
492 struct lustre_capa *capa = NULL;
496 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
500 rc = ost_validate_obdo(req->rq_export, &body->oa, NULL);
504 rc = req_capsule_server_pack(&req->rq_pill);
508 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
509 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
511 CERROR("Missing capability for OST SETATTR");
516 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
517 repbody->oa = body->oa;
519 OBD_ALLOC_PTR(oinfo);
522 oinfo->oi_oa = &repbody->oa;
523 oinfo->oi_capa = capa;
525 req->rq_status = obd_setattr(req->rq_svc_thread->t_env, exp, oinfo,
530 ost_drop_id(exp, &repbody->oa);
534 static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc,
535 cksum_type_t cksum_type)
537 struct cfs_crypto_hash_desc *hdesc;
538 unsigned int bufsize;
540 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
543 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
545 CERROR("Unable to initialize checksum hash %s\n",
546 cfs_crypto_hash_name(cfs_alg));
547 return PTR_ERR(hdesc);
549 CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
550 for (i = 0; i < desc->bd_iov_count; i++) {
552 /* corrupt the data before we compute the checksum, to
553 * simulate a client->OST data error */
554 if (i == 0 && opc == OST_WRITE &&
555 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
556 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
557 int len = desc->bd_iov[i].kiov_len;
558 struct page *np = cfs_alloc_page(CFS_ALLOC_STD);
559 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
562 char *ptr2 = kmap(np) + off;
564 memcpy(ptr2, ptr, len);
565 memcpy(ptr2, "bad3", min(4, len));
567 cfs_page_unpin(desc->bd_iov[i].kiov_page);
568 desc->bd_iov[i].kiov_page = np;
570 CERROR("can't alloc page for corruption\n");
573 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
574 desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
575 desc->bd_iov[i].kiov_len);
577 /* corrupt the data after we compute the checksum, to
578 * simulate an OST->client data error */
579 if (i == 0 && opc == OST_READ &&
580 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
581 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
582 int len = desc->bd_iov[i].kiov_len;
583 struct page *np = cfs_alloc_page(CFS_ALLOC_STD);
584 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
587 char *ptr2 = kmap(np) + off;
589 memcpy(ptr2, ptr, len);
590 memcpy(ptr2, "bad4", min(4, len));
592 cfs_page_unpin(desc->bd_iov[i].kiov_page);
593 desc->bd_iov[i].kiov_page = np;
595 CERROR("can't alloc page for corruption\n");
601 err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
603 cfs_crypto_hash_final(hdesc, NULL, NULL);
608 static int ost_brw_lock_get(int mode, struct obd_export *exp,
609 struct obd_ioobj *obj, struct niobuf_remote *nb,
610 struct lustre_handle *lh)
613 int nrbufs = obj->ioo_bufcnt;
614 struct ldlm_res_id res_id;
615 ldlm_policy_data_t policy;
619 osc_build_res_name(obj->ioo_id, obj->ioo_seq, &res_id);
620 LASSERT(mode == LCK_PR || mode == LCK_PW);
621 LASSERT(!lustre_handle_is_used(lh));
623 if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
626 for (i = 1; i < nrbufs; i ++)
627 if ((nb[0].flags & OBD_BRW_SRVLOCK) !=
628 (nb[i].flags & OBD_BRW_SRVLOCK))
631 policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
632 policy.l_extent.end = (nb[nrbufs - 1].offset +
633 nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
635 RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
636 LDLM_EXTENT, &policy, mode, &flags,
637 ldlm_blocking_ast, ldlm_completion_ast,
638 ldlm_glimpse_ast, NULL, 0, NULL, lh));
641 static void ost_brw_lock_put(int mode,
642 struct obd_ioobj *obj, struct niobuf_remote *niob,
643 struct lustre_handle *lh)
646 LASSERT(mode == LCK_PR || mode == LCK_PW);
647 LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
648 lustre_handle_is_used(lh));
649 if (lustre_handle_is_used(lh))
650 ldlm_lock_decref(lh, mode);
654 /* Allocate thread local buffers if needed */
655 static struct ost_thread_local_cache *ost_tls_get(struct ptlrpc_request *r)
657 struct ost_thread_local_cache *tls =
658 (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
660 /* In normal mode of operation an I/O request is serviced only
661 * by ll_ost_io threads each of them has own tls buffers allocated by
663 * During recovery, an I/O request may be queued until any of the ost
664 * service threads process it. Not necessary it should be one of
665 * ll_ost_io threads. In that case we dynamically allocating tls
666 * buffers for the request service time. */
667 if (unlikely(tls == NULL)) {
668 LASSERT(r->rq_export->exp_in_recovery);
672 r->rq_svc_thread->t_data = tls;
678 /* Free thread local buffers if they were allocated only for servicing
679 * this one request */
680 static void ost_tls_put(struct ptlrpc_request *r)
682 struct ost_thread_local_cache *tls =
683 (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
685 if (unlikely(tls->temporary)) {
687 r->rq_svc_thread->t_data = NULL;
691 static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
693 struct ptlrpc_bulk_desc *desc = NULL;
694 struct obd_export *exp = req->rq_export;
695 struct niobuf_remote *remote_nb;
696 struct niobuf_local *local_nb;
697 struct obd_ioobj *ioo;
698 struct ost_body *body, *repbody;
699 struct lustre_capa *capa = NULL;
700 struct l_wait_info lwi;
701 struct lustre_handle lockh = { 0 };
702 int niocount, npages, nob = 0, rc, i;
704 struct ost_thread_local_cache *tls;
707 req->rq_bulk_read = 1;
709 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
710 GOTO(out, rc = -EIO);
712 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
714 /* Check if there is eviction in progress, and if so, wait for it to
716 if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
717 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
718 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
719 !cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress),
723 GOTO(out, rc = -ENOTCONN);
725 /* ost_body, ioobj & noibuf_remote are verified and swabbed in
726 * ost_rw_hpreq_check(). */
727 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
729 GOTO(out, rc = -EFAULT);
732 * A req_capsule_X_get_array(pill, field, ptr_to_element_count) function
733 * would be useful here and wherever we get &RMF_OBD_IOOBJ and
734 * &RMF_NIOBUF_REMOTE.
736 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
738 GOTO(out, rc = -EFAULT);
740 rc = ost_validate_obdo(exp, &body->oa, ioo);
744 niocount = ioo->ioo_bufcnt;
745 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
746 if (remote_nb == NULL)
747 GOTO(out, rc = -EFAULT);
749 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
750 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
752 CERROR("Missing capability for OST BRW READ");
753 GOTO(out, rc = -EFAULT);
757 rc = req_capsule_server_pack(&req->rq_pill);
761 tls = ost_tls_get(req);
763 GOTO(out_bulk, rc = -ENOMEM);
764 local_nb = tls->local;
766 rc = ost_brw_lock_get(LCK_PR, exp, ioo, remote_nb, &lockh);
771 * If getting the lock took more time than
772 * client was willing to wait, drop it. b=11330
774 if (cfs_time_current_sec() > req->rq_deadline ||
775 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
777 CERROR("Dropping timed-out read from %s because locking"
778 "object "LPX64" took %ld seconds (limit was %ld).\n",
779 libcfs_id2str(req->rq_peer), ioo->ioo_id,
780 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
781 req->rq_deadline - req->rq_arrival_time.tv_sec);
782 GOTO(out_lock, rc = -ETIMEDOUT);
785 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
786 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
788 npages = OST_THREAD_POOL_SIZE;
789 rc = obd_preprw(req->rq_svc_thread->t_env, OBD_BRW_READ, exp,
790 &repbody->oa, 1, ioo, remote_nb, &npages, local_nb,
795 desc = ptlrpc_prep_bulk_exp(req, npages,
796 BULK_PUT_SOURCE, OST_BULK_PORTAL);
798 GOTO(out_commitrw, rc = -ENOMEM);
801 for (i = 0; i < npages; i++) {
802 int page_rc = local_nb[i].rc;
804 if (page_rc < 0) { /* error */
810 if (page_rc != 0) { /* some data! */
811 LASSERT (local_nb[i].page != NULL);
812 ptlrpc_prep_bulk_page(desc, local_nb[i].page,
813 local_nb[i].lnb_page_offset,
817 if (page_rc != local_nb[i].len) { /* short read */
818 /* All subsequent pages should be 0 */
820 LASSERT(local_nb[i].rc == 0);
825 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
826 cksum_type_t cksum_type =
827 cksum_type_unpack(repbody->oa.o_valid & OBD_MD_FLFLAGS ?
828 repbody->oa.o_flags : 0);
829 repbody->oa.o_flags = cksum_type_pack(cksum_type);
830 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
831 repbody->oa.o_cksum = ost_checksum_bulk(desc, OST_READ,cksum_type);
832 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
833 repbody->oa.o_cksum);
835 repbody->oa.o_valid = 0;
837 /* We're finishing using body->oa as an input variable */
839 /* Check if client was evicted while we were doing i/o before touching
842 if (likely(!CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2)))
843 rc = target_bulk_io(exp, desc, &lwi);
848 /* Must commit after prep above in all cases */
849 rc = obd_commitrw(req->rq_svc_thread->t_env, OBD_BRW_READ, exp,
850 &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
854 ost_drop_id(exp, &repbody->oa);
857 ost_brw_lock_put(LCK_PR, ioo, remote_nb, &lockh);
861 if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
862 ptlrpc_free_bulk(desc);
866 req->rq_status = nob;
867 ptlrpc_lprocfs_brw(req, nob);
868 target_committed_to_req(req);
870 } else if (!no_reply) {
871 /* Only reply if there was no comms problem with bulk */
872 target_committed_to_req(req);
876 /* reply out callback would free */
877 ptlrpc_req_drop_rs(req);
878 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
879 "client will retry: rc %d\n",
880 exp->exp_obd->obd_name,
881 obd_uuid2str(&exp->exp_client_uuid),
882 obd_export_nid2str(exp), rc);
884 /* send a bulk after reply to simulate a network delay or reordering
886 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
888 struct l_wait_info lwi1;
890 CDEBUG(D_INFO, "reorder BULK\n");
891 cfs_waitq_init(&waitq);
893 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
894 l_wait_event(waitq, 0, &lwi1);
895 rc = target_bulk_io(exp, desc, &lwi);
896 ptlrpc_free_bulk(desc);
902 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
904 struct ptlrpc_bulk_desc *desc = NULL;
905 struct obd_export *exp = req->rq_export;
906 struct niobuf_remote *remote_nb;
907 struct niobuf_local *local_nb;
908 struct obd_ioobj *ioo;
909 struct ost_body *body, *repbody;
910 struct l_wait_info lwi;
911 struct lustre_handle lockh = {0};
912 struct lustre_capa *capa = NULL;
914 int objcount, niocount, npages;
916 obd_count client_cksum = 0, server_cksum = 0;
917 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
918 int no_reply = 0, mmap = 0;
919 __u32 o_uid = 0, o_gid = 0;
920 struct ost_thread_local_cache *tls;
923 req->rq_bulk_write = 1;
925 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
926 GOTO(out, rc = -EIO);
927 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
928 GOTO(out, rc = -EFAULT);
930 /* pause before transaction has been started */
931 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
933 /* ost_body, ioobj & noibuf_remote are verified and swabbed in
934 * ost_rw_hpreq_check(). */
935 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
937 GOTO(out, rc = -EFAULT);
939 objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
940 RCL_CLIENT) / sizeof(*ioo);
941 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
943 GOTO(out, rc = -EFAULT);
945 rc = ost_validate_obdo(exp, &body->oa, ioo);
949 for (niocount = i = 0; i < objcount; i++)
950 niocount += ioo[i].ioo_bufcnt;
953 * It'd be nice to have a capsule function to indicate how many elements
954 * there were in a buffer for an RMF that's declared to be an array.
955 * It's easy enough to compute the number of elements here though.
957 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
958 if (remote_nb == NULL || niocount != (req_capsule_get_size(&req->rq_pill,
959 &RMF_NIOBUF_REMOTE, RCL_CLIENT) / sizeof(*remote_nb)))
960 GOTO(out, rc = -EFAULT);
962 if ((remote_nb[0].flags & OBD_BRW_MEMALLOC) &&
963 (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
964 cfs_memory_pressure_set();
966 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
967 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
969 CERROR("Missing capability for OST BRW WRITE");
970 GOTO(out, rc = -EFAULT);
974 req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
975 niocount * sizeof(*rcs));
976 rc = req_capsule_server_pack(&req->rq_pill);
979 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
980 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
982 tls = ost_tls_get(req);
984 GOTO(out_bulk, rc = -ENOMEM);
985 local_nb = tls->local;
987 rc = ost_brw_lock_get(LCK_PW, exp, ioo, remote_nb, &lockh);
992 * If getting the lock took more time than
993 * client was willing to wait, drop it. b=11330
995 if (cfs_time_current_sec() > req->rq_deadline ||
996 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
998 CERROR("Dropping timed-out write from %s because locking "
999 "object "LPX64" took %ld seconds (limit was %ld).\n",
1000 libcfs_id2str(req->rq_peer), ioo->ioo_id,
1001 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1002 req->rq_deadline - req->rq_arrival_time.tv_sec);
1003 GOTO(out_lock, rc = -ETIMEDOUT);
1006 /* obd_preprw clobbers oa->valid, so save what we need */
1007 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1008 client_cksum = body->oa.o_cksum;
1009 if (body->oa.o_valid & OBD_MD_FLFLAGS)
1010 cksum_type = cksum_type_unpack(body->oa.o_flags);
1012 if (body->oa.o_valid & OBD_MD_FLFLAGS && body->oa.o_flags & OBD_FL_MMAP)
1015 /* Because we already sync grant info with client when reconnect,
1016 * grant info will be cleared for resent req, then fed_grant and
1017 * total_grant will not be modified in following preprw_write */
1018 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1019 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1020 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1023 if (exp_connect_rmtclient(exp)) {
1024 o_uid = body->oa.o_uid;
1025 o_gid = body->oa.o_gid;
1028 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1029 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
1031 npages = OST_THREAD_POOL_SIZE;
1032 rc = obd_preprw(req->rq_svc_thread->t_env, OBD_BRW_WRITE, exp,
1033 &repbody->oa, objcount, ioo, remote_nb, &npages,
1034 local_nb, oti, capa);
1038 desc = ptlrpc_prep_bulk_exp(req, npages,
1039 BULK_GET_SINK, OST_BULK_PORTAL);
1041 GOTO(skip_transfer, rc = -ENOMEM);
1043 /* NB Having prepped, we must commit... */
1045 for (i = 0; i < npages; i++)
1046 ptlrpc_prep_bulk_page(desc, local_nb[i].page,
1047 local_nb[i].lnb_page_offset,
1050 rc = sptlrpc_svc_prep_bulk(req, desc);
1054 rc = target_bulk_io(exp, desc, &lwi);
1058 if (client_cksum != 0 && rc == 0) {
1059 static int cksum_counter;
1060 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1061 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
1062 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
1063 server_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1064 repbody->oa.o_cksum = server_cksum;
1066 if (unlikely(client_cksum != server_cksum)) {
1067 CDEBUG_LIMIT(mmap ? D_INFO : D_ERROR,
1068 "client csum %x, server csum %x\n",
1069 client_cksum, server_cksum);
1071 } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){
1072 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1073 cksum_counter, libcfs_id2str(req->rq_peer),
1078 /* Must commit after prep above in all cases */
1079 rc = obd_commitrw(req->rq_svc_thread->t_env, OBD_BRW_WRITE, exp,
1080 &repbody->oa, objcount, ioo, remote_nb, npages,
1082 if (rc == -ENOTCONN)
1083 /* quota acquire process has been given up because
1084 * either the client has been evicted or the client
1085 * has timed out the request already */
1088 if (exp_connect_rmtclient(exp)) {
1089 repbody->oa.o_uid = o_uid;
1090 repbody->oa.o_gid = o_gid;
1094 * Disable sending mtime back to the client. If the client locked the
1095 * whole object, then it has already updated the mtime on its side,
1096 * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
1098 repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
1100 if (unlikely(client_cksum != server_cksum && rc == 0 && !mmap)) {
1101 int new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1106 if (new_cksum == server_cksum)
1107 msg = "changed in transit before arrival at OST";
1108 else if (new_cksum == client_cksum)
1109 msg = "initial checksum before message complete";
1111 msg = "changed in transit AND after initial checksum";
1113 if (req->rq_peer.nid == desc->bd_sender) {
1117 router = libcfs_nid2str(desc->bd_sender);
1120 LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: %s from "
1121 "%s%s%s inode "DFID" object "
1122 LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
1123 exp->exp_obd->obd_name, msg,
1124 libcfs_id2str(req->rq_peer),
1126 body->oa.o_valid & OBD_MD_FLFID ?
1127 body->oa.o_parent_seq : (__u64)0,
1128 body->oa.o_valid & OBD_MD_FLFID ?
1129 body->oa.o_parent_oid : 0,
1130 body->oa.o_valid & OBD_MD_FLFID ?
1131 body->oa.o_parent_ver : 0,
1133 body->oa.o_valid & OBD_MD_FLGROUP ?
1134 body->oa.o_seq : (__u64)0,
1135 local_nb[0].lnb_file_offset,
1136 local_nb[npages-1].lnb_file_offset +
1137 local_nb[npages-1].len - 1 );
1138 CERROR("client csum %x, original server csum %x, "
1139 "server csum now %x\n",
1140 client_cksum, server_cksum, new_cksum);
1146 /* set per-requested niobuf return codes */
1147 for (i = j = 0; i < niocount; i++) {
1148 int len = remote_nb[i].len;
1153 LASSERT(j < npages);
1154 if (local_nb[j].rc < 0)
1155 rcs[i] = local_nb[j].rc;
1156 len -= local_nb[j].len;
1161 LASSERT(j == npages);
1162 ptlrpc_lprocfs_brw(req, nob);
1166 ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh);
1171 ptlrpc_free_bulk(desc);
1174 oti_to_request(oti, req);
1175 target_committed_to_req(req);
1176 rc = ptlrpc_reply(req);
1177 } else if (!no_reply) {
1178 /* Only reply if there was no comms problem with bulk */
1179 target_committed_to_req(req);
1180 req->rq_status = rc;
1183 /* reply out callback would free */
1184 ptlrpc_req_drop_rs(req);
1185 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
1186 "client will retry: rc %d\n",
1187 exp->exp_obd->obd_name,
1188 obd_uuid2str(&exp->exp_client_uuid),
1189 obd_export_nid2str(exp), rc);
1191 cfs_memory_pressure_clr();
1196 * Implementation of OST_SET_INFO.
1198 * OST_SET_INFO is like ioctl(): heavily overloaded. Specifically, it takes a
1199 * "key" and a value RPC buffers as arguments, with the value's contents
1200 * interpreted according to the key.
1202 * Value types that need swabbing have swabbing done explicitly, either here or
1203 * in functions called from here. This should be corrected: all swabbing should
1204 * be done in the capsule abstraction, as that will then allow us to move
1205 * swabbing exclusively to the client without having to modify server code
1206 * outside the capsule abstraction's implementation itself. To correct this
1207 * will require minor changes to the capsule abstraction; see the comments for
1208 * req_capsule_extend() in layout.c.
1210 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
1212 struct ost_body *body = NULL, *repbody;
1213 char *key, *val = NULL;
1214 int keylen, vallen, rc = 0;
1215 int is_grant_shrink = 0;
1218 key = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
1220 DEBUG_REQ(D_HA, req, "no set_info key");
1223 keylen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_KEY,
1226 vallen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_VAL,
1229 if ((is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK)))
1230 /* In this case the value is actually an RMF_OST_BODY, so we
1231 * transmutate the type of this PTLRPC */
1232 req_capsule_extend(&req->rq_pill, &RQF_OST_SET_GRANT_INFO);
1234 rc = req_capsule_server_pack(&req->rq_pill);
1239 if (is_grant_shrink) {
1240 body = req_capsule_client_get(&req->rq_pill,
1245 repbody = req_capsule_server_get(&req->rq_pill,
1247 memcpy(repbody, body, sizeof(*body));
1248 val = (char*)repbody;
1250 val = req_capsule_client_get(&req->rq_pill,
1255 if (KEY_IS(KEY_EVICT_BY_NID)) {
1257 obd_export_evict_by_nid(exp->exp_obd, val);
1259 } else if (KEY_IS(KEY_MDS_CONN) && ptlrpc_req_need_swab(req)) {
1260 if (vallen < sizeof(__u32))
1262 __swab32s((__u32 *)val);
1265 /* OBD will also check if KEY_IS(KEY_GRANT_SHRINK), and will cast val to
1266 * a struct ost_body * value */
1267 rc = obd_set_info_async(req->rq_svc_thread->t_env, exp, keylen,
1268 key, vallen, val, NULL);
1270 lustre_msg_set_status(req->rq_repmsg, 0);
1274 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
1277 int keylen, replylen, rc = 0;
1278 struct req_capsule *pill = &req->rq_pill;
1281 /* this common part for get_info rpc */
1282 key = req_capsule_client_get(pill, &RMF_SETINFO_KEY);
1284 DEBUG_REQ(D_HA, req, "no get_info key");
1287 keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT);
1289 if (KEY_IS(KEY_FIEMAP)) {
1290 struct ll_fiemap_info_key *fm_key = key;
1293 rc = ost_validate_obdo(exp, &fm_key->oa, NULL);
1298 rc = obd_get_info(req->rq_svc_thread->t_env, exp, keylen, key,
1299 &replylen, NULL, NULL);
1303 req_capsule_set_size(pill, &RMF_GENERIC_DATA,
1304 RCL_SERVER, replylen);
1306 rc = req_capsule_server_pack(pill);
1310 reply = req_capsule_server_get(pill, &RMF_GENERIC_DATA);
1314 /* call again to fill in the reply buffer */
1315 rc = obd_get_info(req->rq_svc_thread->t_env, exp, keylen, key,
1316 &replylen, reply, NULL);
1318 lustre_msg_set_status(req->rq_repmsg, 0);
1322 static int ost_handle_quotactl(struct ptlrpc_request *req)
1324 struct obd_quotactl *oqctl, *repoqc;
1328 oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1330 GOTO(out, rc = -EPROTO);
1332 rc = req_capsule_server_pack(&req->rq_pill);
1336 repoqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1337 req->rq_status = obd_quotactl(req->rq_export, oqctl);
1344 static int ost_handle_quotacheck(struct ptlrpc_request *req)
1346 struct obd_quotactl *oqctl;
1350 oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1354 rc = req_capsule_server_pack(&req->rq_pill);
1358 /* deprecated, not used any more */
1359 req->rq_status = -EOPNOTSUPP;
1360 RETURN(-EOPNOTSUPP);
1363 static int ost_llog_handle_connect(struct obd_export *exp,
1364 struct ptlrpc_request *req)
1366 struct llogd_conn_body *body;
1370 body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_CONN_BODY);
1371 rc = obd_llog_connect(exp, body);
1375 #define ost_init_sec_none(reply, exp) \
1377 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT | \
1378 OBD_CONNECT_RMT_CLIENT_FORCE | \
1379 OBD_CONNECT_OSS_CAPA); \
1380 cfs_spin_lock(&exp->exp_lock); \
1381 exp->exp_connect_flags = reply->ocd_connect_flags; \
1382 cfs_spin_unlock(&exp->exp_lock); \
1385 static int ost_init_sec_level(struct ptlrpc_request *req)
1387 struct obd_export *exp = req->rq_export;
1388 struct req_capsule *pill = &req->rq_pill;
1389 struct obd_device *obd = exp->exp_obd;
1390 struct filter_obd *filter = &obd->u.filter;
1391 char *client = libcfs_nid2str(req->rq_peer.nid);
1392 struct obd_connect_data *data, *reply;
1396 data = req_capsule_client_get(pill, &RMF_CONNECT_DATA);
1397 reply = req_capsule_server_get(pill, &RMF_CONNECT_DATA);
1398 if (data == NULL || reply == NULL)
1401 /* connection from MDT is always trusted */
1402 if (req->rq_auth_usr_mdt) {
1403 ost_init_sec_none(reply, exp);
1407 /* no GSS support case */
1408 if (!req->rq_auth_gss) {
1409 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1410 CWARN("client %s -> target %s does not user GSS, "
1411 "can not run under security level %d.\n",
1412 client, obd->obd_name, filter->fo_sec_level);
1415 ost_init_sec_none(reply, exp);
1420 /* old version case */
1421 if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
1422 !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
1423 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1424 CWARN("client %s -> target %s uses old version, "
1425 "can not run under security level %d.\n",
1426 client, obd->obd_name, filter->fo_sec_level);
1429 CWARN("client %s -> target %s uses old version, "
1430 "run under security level %d.\n",
1431 client, obd->obd_name, filter->fo_sec_level);
1432 ost_init_sec_none(reply, exp);
1437 remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
1439 if (!req->rq_auth_remote)
1440 CDEBUG(D_SEC, "client (local realm) %s -> target %s "
1441 "asked to be remote.\n", client, obd->obd_name);
1442 } else if (req->rq_auth_remote) {
1444 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
1445 "as remote by default.\n", client, obd->obd_name);
1449 if (!filter->fo_fl_oss_capa) {
1450 CDEBUG(D_SEC, "client %s -> target %s is set as remote,"
1451 " but OSS capabilities are not enabled: %d.\n",
1452 client, obd->obd_name, filter->fo_fl_oss_capa);
1457 switch (filter->fo_sec_level) {
1458 case LUSTRE_SEC_NONE:
1460 ost_init_sec_none(reply, exp);
1463 CDEBUG(D_SEC, "client %s -> target %s is set as remote, "
1464 "can not run under security level %d.\n",
1465 client, obd->obd_name, filter->fo_sec_level);
1468 case LUSTRE_SEC_REMOTE:
1470 ost_init_sec_none(reply, exp);
1472 case LUSTRE_SEC_ALL:
1474 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
1475 OBD_CONNECT_RMT_CLIENT_FORCE);
1476 if (!filter->fo_fl_oss_capa)
1477 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
1479 cfs_spin_lock(&exp->exp_lock);
1480 exp->exp_connect_flags = reply->ocd_connect_flags;
1481 cfs_spin_unlock(&exp->exp_lock);
1493 * this should be done in filter_connect()/filter_reconnect(), but
1494 * we can't obtain information like NID, which stored in incoming
1495 * request, thus can't decide what flavor to use. so we do it here.
1497 * This hack should be removed after the OST stack be rewritten, just
1498 * like what we are doing in mdt_obd_connect()/mdt_obd_reconnect().
1500 static int ost_connect_check_sptlrpc(struct ptlrpc_request *req)
1502 struct obd_export *exp = req->rq_export;
1503 struct filter_obd *filter = &exp->exp_obd->u.filter;
1504 struct sptlrpc_flavor flvr;
1507 if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
1508 LUSTRE_ECHO_NAME) == 0)) {
1509 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
1513 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
1514 cfs_read_lock(&filter->fo_sptlrpc_lock);
1515 sptlrpc_target_choose_flavor(&filter->fo_sptlrpc_rset,
1519 cfs_read_unlock(&filter->fo_sptlrpc_lock);
1521 cfs_spin_lock(&exp->exp_lock);
1523 exp->exp_sp_peer = req->rq_sp_from;
1524 exp->exp_flvr = flvr;
1526 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
1527 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
1528 CERROR("unauthorized rpc flavor %x from %s, "
1529 "expect %x\n", req->rq_flvr.sf_rpc,
1530 libcfs_nid2str(req->rq_peer.nid),
1531 exp->exp_flvr.sf_rpc);
1535 cfs_spin_unlock(&exp->exp_lock);
1537 if (exp->exp_sp_peer != req->rq_sp_from) {
1538 CERROR("RPC source %s doesn't match %s\n",
1539 sptlrpc_part2name(req->rq_sp_from),
1540 sptlrpc_part2name(exp->exp_sp_peer));
1543 rc = sptlrpc_target_export_check(exp, req);
1550 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1552 int ost_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1553 void *data, int flag)
1556 __u32 sync_lock_cancel = 0;
1557 __u32 len = sizeof(sync_lock_cancel);
1562 rc = lu_env_init(&env, LCT_DT_THREAD);
1563 if (unlikely(rc != 0))
1566 rc = obd_get_info(&env, lock->l_export, sizeof(KEY_SYNC_LOCK_CANCEL),
1567 KEY_SYNC_LOCK_CANCEL, &len, &sync_lock_cancel, NULL);
1568 if (rc == 0 && flag == LDLM_CB_CANCELING &&
1569 (lock->l_granted_mode & (LCK_PW|LCK_GROUP)) &&
1570 (sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1571 (sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1572 lock->l_flags & LDLM_FL_CBPENDING))) {
1573 struct obd_info *oinfo;
1577 OBD_ALLOC_PTR(oinfo);
1579 GOTO(out_env, rc = -ENOMEM);
1582 OBD_FREE_PTR(oinfo);
1583 GOTO(out_env, rc = -ENOMEM);
1585 oa->o_id = lock->l_resource->lr_name.name[0];
1586 oa->o_seq = lock->l_resource->lr_name.name[1];
1587 oa->o_valid = OBD_MD_FLID|OBD_MD_FLGROUP;
1589 oinfo->oi_capa = BYPASS_CAPA;
1591 rc = obd_sync(&env, lock->l_export, oinfo,
1592 lock->l_policy_data.l_extent.start,
1593 lock->l_policy_data.l_extent.end, NULL);
1595 CERROR("Error %d syncing data on lock cancel\n", rc);
1598 OBD_FREE_PTR(oinfo);
1601 rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1607 static int ost_filter_recovery_request(struct ptlrpc_request *req,
1608 struct obd_device *obd, int *process)
1610 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1611 case OST_CONNECT: /* This will never get here, but for completeness. */
1612 case OST_DISCONNECT:
1623 case OBD_LOG_CANCEL:
1625 *process = target_queue_recovery_request(req, obd);
1629 DEBUG_REQ(D_WARNING, req, "not permitted during recovery");
1635 int ost_msg_check_version(struct lustre_msg *msg)
1639 switch(lustre_msg_get_opc(msg)) {
1641 case OST_DISCONNECT:
1644 case SEC_CTX_INIT_CONT:
1646 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
1648 CERROR("bad opc %u version %08x, expecting %08x\n",
1649 lustre_msg_get_opc(msg),
1650 lustre_msg_get_version(msg),
1651 LUSTRE_OBD_VERSION);
1664 case OST_QUOTACHECK:
1666 case OST_QUOTA_ADJUST_QUNIT:
1667 rc = lustre_msg_check_version(msg, LUSTRE_OST_VERSION);
1669 CERROR("bad opc %u version %08x, expecting %08x\n",
1670 lustre_msg_get_opc(msg),
1671 lustre_msg_get_version(msg),
1672 LUSTRE_OST_VERSION);
1677 case LDLM_BL_CALLBACK:
1678 case LDLM_CP_CALLBACK:
1679 rc = lustre_msg_check_version(msg, LUSTRE_DLM_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_DLM_VERSION);
1686 case LLOG_ORIGIN_CONNECT:
1687 case OBD_LOG_CANCEL:
1688 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
1690 CERROR("bad opc %u version %08x, expecting %08x\n",
1691 lustre_msg_get_opc(msg),
1692 lustre_msg_get_version(msg),
1693 LUSTRE_LOG_VERSION);
1696 CERROR("Unexpected opcode %d\n", lustre_msg_get_opc(msg));
1702 struct ost_prolong_data {
1703 struct ptlrpc_request *opd_req;
1704 struct obd_export *opd_exp;
1705 struct obdo *opd_oa;
1706 struct ldlm_res_id opd_resid;
1707 struct ldlm_extent opd_extent;
1708 ldlm_mode_t opd_mode;
1709 unsigned int opd_locks;
1713 /* prolong locks for the current service time of the corresponding
1714 * portal (= OST_IO_PORTAL)
1716 static inline int prolong_timeout(struct ptlrpc_request *req)
1718 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
1721 return obd_timeout / 2;
1723 return max(at_est2timeout(at_get(&svcpt->scp_at_estimate)),
1727 static void ost_prolong_lock_one(struct ost_prolong_data *opd,
1728 struct ldlm_lock *lock)
1730 LASSERT(lock->l_export == opd->opd_exp);
1732 if (lock->l_destroyed) /* lock already cancelled */
1735 /* XXX: never try to grab resource lock here because we're inside
1736 * exp_bl_list_lock; in ldlm_lockd.c to handle waiting list we take
1737 * res lock and then exp_bl_list_lock. */
1739 if (!(lock->l_flags & LDLM_FL_AST_SENT))
1740 /* ignore locks not being cancelled */
1744 "refreshed for req x"LPU64" ext("LPU64"->"LPU64") to %ds.\n",
1745 opd->opd_req->rq_xid, opd->opd_extent.start,
1746 opd->opd_extent.end, opd->opd_timeout);
1748 /* OK. this is a possible lock the user holds doing I/O
1749 * let's refresh eviction timer for it */
1750 ldlm_refresh_waiting_lock(lock, opd->opd_timeout);
1754 static void ost_prolong_locks(struct ost_prolong_data *data)
1756 struct obd_export *exp = data->opd_exp;
1757 struct obdo *oa = data->opd_oa;
1758 struct ldlm_lock *lock;
1761 if (oa->o_valid & OBD_MD_FLHANDLE) {
1762 /* mostly a request should be covered by only one lock, try
1764 lock = ldlm_handle2lock(&oa->o_handle);
1766 /* Fast path to check if the lock covers the whole IO
1767 * region exclusively. */
1768 if (lock->l_granted_mode == LCK_PW &&
1769 ldlm_extent_contain(&lock->l_policy_data.l_extent,
1770 &data->opd_extent)) {
1772 ost_prolong_lock_one(data, lock);
1773 LDLM_LOCK_PUT(lock);
1776 LDLM_LOCK_PUT(lock);
1781 cfs_spin_lock_bh(&exp->exp_bl_list_lock);
1782 cfs_list_for_each_entry(lock, &exp->exp_bl_list, l_exp_list) {
1783 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1784 LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
1786 if (!ldlm_res_eq(&data->opd_resid, &lock->l_resource->lr_name))
1789 if (!ldlm_extent_overlap(&lock->l_policy_data.l_extent,
1793 ost_prolong_lock_one(data, lock);
1795 cfs_spin_unlock_bh(&exp->exp_bl_list_lock);
1801 * Returns 1 if the given PTLRPC matches the given LDLM locks, or 0 if it does
1804 static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req,
1805 struct ldlm_lock *lock)
1807 struct niobuf_remote *nb;
1808 struct obd_ioobj *ioo;
1810 struct ldlm_extent ext;
1813 opc = lustre_msg_get_opc(req->rq_reqmsg);
1814 LASSERT(opc == OST_READ || opc == OST_WRITE);
1816 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1817 LASSERT(ioo != NULL);
1819 nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1820 LASSERT(nb != NULL);
1822 ext.start = nb->offset;
1823 nb += ioo->ioo_bufcnt - 1;
1824 ext.end = nb->offset + nb->len - 1;
1826 LASSERT(lock->l_resource != NULL);
1827 if (!osc_res_name_eq(ioo->ioo_id, ioo->ioo_seq,
1828 &lock->l_resource->lr_name))
1832 if (opc == OST_READ)
1834 if (!(lock->l_granted_mode & mode))
1837 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
1841 * High-priority queue request check for whether the given PTLRPC request (\a
1842 * req) is blocking an LDLM lock cancel.
1844 * Returns 1 if the given given PTLRPC request (\a req) is blocking an LDLM lock
1845 * cancel, 0 if it is not, and -EFAULT if the request is malformed.
1847 * Only OST_READs, OST_WRITEs and OST_PUNCHes go on the h-p RPC queue. This
1848 * function looks only at OST_READs and OST_WRITEs.
1850 static int ost_rw_hpreq_check(struct ptlrpc_request *req)
1852 struct obd_device *obd = req->rq_export->exp_obd;
1853 struct ost_body *body;
1854 struct obd_ioobj *ioo;
1855 struct niobuf_remote *nb;
1856 struct ost_prolong_data opd = { 0 };
1861 * Use LASSERT to do sanity check because malformed RPCs should have
1862 * been filtered out in ost_hpreq_handler().
1864 opc = lustre_msg_get_opc(req->rq_reqmsg);
1865 LASSERT(opc == OST_READ || opc == OST_WRITE);
1867 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1868 LASSERT(body != NULL);
1870 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1871 LASSERT(ioo != NULL);
1873 nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1874 LASSERT(nb != NULL);
1875 LASSERT(!(nb->flags & OBD_BRW_SRVLOCK));
1877 osc_build_res_name(ioo->ioo_id, ioo->ioo_seq, &opd.opd_resid);
1881 if (opc == OST_READ)
1883 opd.opd_mode = mode;
1884 opd.opd_exp = req->rq_export;
1885 opd.opd_oa = &body->oa;
1886 opd.opd_extent.start = nb->offset;
1887 nb += ioo->ioo_bufcnt - 1;
1888 opd.opd_extent.end = nb->offset + nb->len - 1;
1889 opd.opd_timeout = prolong_timeout(req);
1891 DEBUG_REQ(D_RPCTRACE, req,
1892 "%s %s: refresh rw locks: " LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1893 obd->obd_name, cfs_current()->comm,
1894 opd.opd_resid.name[0], opd.opd_resid.name[1],
1895 opd.opd_extent.start, opd.opd_extent.end);
1897 ost_prolong_locks(&opd);
1899 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1900 obd->obd_name, opd.opd_locks, req);
1902 RETURN(opd.opd_locks);
1905 static void ost_rw_hpreq_fini(struct ptlrpc_request *req)
1907 (void)ost_rw_hpreq_check(req);
1911 * Like ost_rw_hpreq_lock_match(), but for OST_PUNCH RPCs.
1913 static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req,
1914 struct ldlm_lock *lock)
1916 struct ost_body *body;
1919 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1920 LASSERT(body != NULL);
1922 if (body->oa.o_valid & OBD_MD_FLHANDLE &&
1923 body->oa.o_handle.cookie == lock->l_handle.h_cookie)
1930 * Like ost_rw_hpreq_check(), but for OST_PUNCH RPCs.
1932 static int ost_punch_hpreq_check(struct ptlrpc_request *req)
1934 struct obd_device *obd = req->rq_export->exp_obd;
1935 struct ost_body *body;
1937 struct ost_prolong_data opd = { 0 };
1941 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1942 LASSERT(body != NULL);
1945 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS) ||
1946 !(oa->o_flags & OBD_FL_SRVLOCK));
1949 end = start + oa->o_blocks;
1952 opd.opd_mode = LCK_PW;
1953 opd.opd_exp = req->rq_export;
1955 opd.opd_extent.start = start;
1956 opd.opd_extent.end = end;
1957 if (oa->o_blocks == OBD_OBJECT_EOF)
1958 opd.opd_extent.end = OBD_OBJECT_EOF;
1959 opd.opd_timeout = prolong_timeout(req);
1961 osc_build_res_name(oa->o_id, oa->o_seq, &opd.opd_resid);
1964 "%s: refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1966 opd.opd_resid.name[0], opd.opd_resid.name[1],
1967 opd.opd_extent.start, opd.opd_extent.end);
1969 ost_prolong_locks(&opd);
1971 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1972 obd->obd_name, opd.opd_locks, req);
1974 RETURN(opd.opd_locks > 0);
1977 static void ost_punch_hpreq_fini(struct ptlrpc_request *req)
1979 (void)ost_punch_hpreq_check(req);
1982 struct ptlrpc_hpreq_ops ost_hpreq_rw = {
1983 .hpreq_lock_match = ost_rw_hpreq_lock_match,
1984 .hpreq_check = ost_rw_hpreq_check,
1985 .hpreq_fini = ost_rw_hpreq_fini
1988 struct ptlrpc_hpreq_ops ost_hpreq_punch = {
1989 .hpreq_lock_match = ost_punch_hpreq_lock_match,
1990 .hpreq_check = ost_punch_hpreq_check,
1991 .hpreq_fini = ost_punch_hpreq_fini
1994 /** Assign high priority operations to the request if needed. */
1995 static int ost_hpreq_handler(struct ptlrpc_request *req)
1998 if (req->rq_export) {
1999 int opc = lustre_msg_get_opc(req->rq_reqmsg);
2000 struct ost_body *body;
2002 if (opc == OST_READ || opc == OST_WRITE) {
2003 struct niobuf_remote *nb;
2004 struct obd_ioobj *ioo;
2005 int objcount, niocount;
2009 /* RPCs on the H-P queue can be inspected before
2010 * ost_handler() initializes their pills, so we
2011 * initialize that here. Capsule initialization is
2012 * idempotent, as is setting the pill's format (provided
2013 * it doesn't change).
2015 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2016 if (opc == OST_READ)
2017 req_capsule_set(&req->rq_pill,
2020 req_capsule_set(&req->rq_pill,
2021 &RQF_OST_BRW_WRITE);
2023 body = req_capsule_client_get(&req->rq_pill,
2026 CERROR("Missing/short ost_body\n");
2030 objcount = req_capsule_get_size(&req->rq_pill,
2034 if (objcount == 0) {
2035 CERROR("Missing/short ioobj\n");
2039 CERROR("too many ioobjs (%d)\n", objcount);
2043 ioo = req_capsule_client_get(&req->rq_pill,
2046 CERROR("Missing/short ioobj\n");
2050 rc = ost_validate_obdo(req->rq_export, &body->oa, ioo);
2052 CERROR("invalid object ids\n");
2056 for (niocount = i = 0; i < objcount; i++) {
2057 if (ioo[i].ioo_bufcnt == 0) {
2058 CERROR("ioo[%d] has zero bufcnt\n", i);
2061 niocount += ioo[i].ioo_bufcnt;
2063 if (niocount > PTLRPC_MAX_BRW_PAGES) {
2064 DEBUG_REQ(D_RPCTRACE, req,
2065 "bulk has too many pages (%d)",
2070 nb = req_capsule_client_get(&req->rq_pill,
2071 &RMF_NIOBUF_REMOTE);
2073 CERROR("Missing/short niobuf\n");
2077 if (niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
2078 req->rq_ops = &ost_hpreq_rw;
2079 } else if (opc == OST_PUNCH) {
2080 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2081 req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
2083 body = req_capsule_client_get(&req->rq_pill,
2086 CERROR("Missing/short ost_body\n");
2090 if (!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
2091 !(body->oa.o_flags & OBD_FL_SRVLOCK))
2092 req->rq_ops = &ost_hpreq_punch;
2098 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
2099 int ost_handle(struct ptlrpc_request *req)
2101 struct obd_trans_info trans_info = { 0, };
2102 struct obd_trans_info *oti = &trans_info;
2103 int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
2104 struct obd_device *obd = NULL;
2107 /* OST module is kept between remounts, but the last reference
2108 * to specific module (say, osd or ofd) kills all related keys
2109 * from the environment. so we have to refill it until the root
2110 * cause is fixed properly */
2111 lu_env_refill(req->rq_svc_thread->t_env);
2113 LASSERT(current->journal_info == NULL);
2115 /* primordial rpcs don't affect server recovery */
2116 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2118 case SEC_CTX_INIT_CONT:
2123 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2125 if (lustre_msg_get_opc(req->rq_reqmsg) != OST_CONNECT) {
2126 if (!class_connected_export(req->rq_export)) {
2127 CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
2128 lustre_msg_get_opc(req->rq_reqmsg),
2129 libcfs_id2str(req->rq_peer));
2130 req->rq_status = -ENOTCONN;
2131 GOTO(out, rc = -ENOTCONN);
2134 obd = req->rq_export->exp_obd;
2136 /* Check for aborted recovery. */
2137 if (obd->obd_recovering) {
2138 rc = ost_filter_recovery_request(req, obd,
2140 if (rc || !should_process)
2142 else if (should_process < 0) {
2143 req->rq_status = should_process;
2144 rc = ptlrpc_error(req);
2152 rc = ost_msg_check_version(req->rq_reqmsg);
2156 if (req && req->rq_reqmsg && req->rq_export &&
2157 (req->rq_export->exp_connect_flags & OBD_CONNECT_JOBSTATS))
2158 oti->oti_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
2160 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2162 CDEBUG(D_INODE, "connect\n");
2163 req_capsule_set(&req->rq_pill, &RQF_OST_CONNECT);
2164 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET))
2166 rc = target_handle_connect(req);
2167 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2))
2170 rc = ost_init_sec_level(req);
2172 rc = ost_connect_check_sptlrpc(req);
2176 case OST_DISCONNECT:
2177 CDEBUG(D_INODE, "disconnect\n");
2178 req_capsule_set(&req->rq_pill, &RQF_OST_DISCONNECT);
2179 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DISCONNECT_NET))
2181 rc = target_handle_disconnect(req);
2184 CDEBUG(D_INODE, "create\n");
2185 req_capsule_set(&req->rq_pill, &RQF_OST_CREATE);
2186 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET))
2188 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2189 GOTO(out, rc = -EROFS);
2190 rc = ost_create(req->rq_export, req, oti);
2193 CDEBUG(D_INODE, "destroy\n");
2194 req_capsule_set(&req->rq_pill, &RQF_OST_DESTROY);
2195 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DESTROY_NET))
2197 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2198 GOTO(out, rc = -EROFS);
2199 rc = ost_destroy(req->rq_export, req, oti);
2202 CDEBUG(D_INODE, "getattr\n");
2203 req_capsule_set(&req->rq_pill, &RQF_OST_GETATTR);
2204 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GETATTR_NET))
2206 rc = ost_getattr(req->rq_export, req);
2209 CDEBUG(D_INODE, "setattr\n");
2210 req_capsule_set(&req->rq_pill, &RQF_OST_SETATTR);
2211 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_NET))
2213 rc = ost_setattr(req->rq_export, req, oti);
2216 req_capsule_set(&req->rq_pill, &RQF_OST_BRW_WRITE);
2217 CDEBUG(D_INODE, "write\n");
2218 /* req->rq_request_portal would be nice, if it was set */
2219 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
2220 CERROR("%s: deny write request from %s to portal %u\n",
2221 req->rq_export->exp_obd->obd_name,
2222 obd_export_nid2str(req->rq_export),
2223 ptlrpc_req2svc(req)->srv_req_portal);
2224 GOTO(out, rc = -EPROTO);
2226 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2228 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2229 GOTO(out, rc = -ENOSPC);
2230 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2231 GOTO(out, rc = -EROFS);
2232 rc = ost_brw_write(req, oti);
2233 LASSERT(current->journal_info == NULL);
2234 /* ost_brw_write sends its own replies */
2237 req_capsule_set(&req->rq_pill, &RQF_OST_BRW_READ);
2238 CDEBUG(D_INODE, "read\n");
2239 /* req->rq_request_portal would be nice, if it was set */
2240 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
2241 CERROR("%s: deny read request from %s to portal %u\n",
2242 req->rq_export->exp_obd->obd_name,
2243 obd_export_nid2str(req->rq_export),
2244 ptlrpc_req2svc(req)->srv_req_portal);
2245 GOTO(out, rc = -EPROTO);
2247 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2249 rc = ost_brw_read(req, oti);
2250 LASSERT(current->journal_info == NULL);
2251 /* ost_brw_read sends its own replies */
2254 CDEBUG(D_INODE, "punch\n");
2255 req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
2256 if (OBD_FAIL_CHECK(OBD_FAIL_OST_PUNCH_NET))
2258 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2259 GOTO(out, rc = -EROFS);
2260 rc = ost_punch(req->rq_export, req, oti);
2263 CDEBUG(D_INODE, "statfs\n");
2264 req_capsule_set(&req->rq_pill, &RQF_OST_STATFS);
2265 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_NET))
2267 rc = ost_statfs(req);
2270 CDEBUG(D_INODE, "sync\n");
2271 req_capsule_set(&req->rq_pill, &RQF_OST_SYNC);
2272 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SYNC_NET))
2274 rc = ost_sync(req->rq_export, req, oti);
2277 DEBUG_REQ(D_INODE, req, "set_info");
2278 req_capsule_set(&req->rq_pill, &RQF_OBD_SET_INFO);
2279 rc = ost_set_info(req->rq_export, req);
2282 DEBUG_REQ(D_INODE, req, "get_info");
2283 req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC);
2284 rc = ost_get_info(req->rq_export, req);
2286 case OST_QUOTACHECK:
2287 CDEBUG(D_INODE, "quotacheck\n");
2288 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACHECK);
2289 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACHECK_NET))
2291 rc = ost_handle_quotacheck(req);
2294 CDEBUG(D_INODE, "quotactl\n");
2295 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACTL);
2296 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACTL_NET))
2298 rc = ost_handle_quotactl(req);
2301 DEBUG_REQ(D_INODE, req, "ping");
2302 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
2303 rc = target_handle_ping(req);
2305 /* FIXME - just reply status */
2306 case LLOG_ORIGIN_CONNECT:
2307 DEBUG_REQ(D_INODE, req, "log connect");
2308 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_CONNECT);
2309 rc = ost_llog_handle_connect(req->rq_export, req);
2310 req->rq_status = rc;
2311 rc = req_capsule_server_pack(&req->rq_pill);
2314 RETURN(ptlrpc_reply(req));
2315 case OBD_LOG_CANCEL:
2316 CDEBUG(D_INODE, "log cancel\n");
2317 req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL);
2318 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
2320 rc = llog_origin_handle_cancel(req);
2321 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP))
2323 req->rq_status = rc;
2324 rc = req_capsule_server_pack(&req->rq_pill);
2327 RETURN(ptlrpc_reply(req));
2329 CDEBUG(D_INODE, "enqueue\n");
2330 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
2331 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE))
2333 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
2335 ldlm_server_glimpse_ast);
2336 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
2339 CDEBUG(D_INODE, "convert\n");
2340 req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT);
2341 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT))
2343 rc = ldlm_handle_convert(req);
2346 CDEBUG(D_INODE, "cancel\n");
2347 req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL);
2348 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL))
2350 rc = ldlm_handle_cancel(req);
2352 case LDLM_BL_CALLBACK:
2353 case LDLM_CP_CALLBACK:
2354 CDEBUG(D_INODE, "callback\n");
2355 CERROR("callbacks should not happen on OST\n");
2358 CERROR("Unexpected opcode %d\n",
2359 lustre_msg_get_opc(req->rq_reqmsg));
2360 req->rq_status = -ENOTSUPP;
2361 rc = ptlrpc_error(req);
2365 LASSERT(current->journal_info == NULL);
2368 /* If we're DISCONNECTing, the export_data is already freed */
2369 if (!rc && lustre_msg_get_opc(req->rq_reqmsg) != OST_DISCONNECT)
2370 target_committed_to_req(req);
2374 oti_to_request(oti, req);
2376 target_send_reply(req, rc, fail);
2379 EXPORT_SYMBOL(ost_handle);
2381 * free per-thread pool created by ost_thread_init().
2383 static void ost_thread_done(struct ptlrpc_thread *thread)
2385 struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local
2390 LASSERT(thread != NULL);
2393 * be prepared to handle partially-initialized pools (because this is
2394 * called from ost_thread_init() for cleanup.
2396 tls = thread->t_data;
2399 thread->t_data = NULL;
2405 * initialize per-thread page pool (bug 5137).
2407 static int ost_thread_init(struct ptlrpc_thread *thread)
2409 struct ost_thread_local_cache *tls;
2413 LASSERT(thread != NULL);
2414 LASSERT(thread->t_data == NULL);
2419 thread->t_data = tls;
2423 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
2425 static struct cfs_cpt_table *ost_io_cptable;
2427 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
2428 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
2430 static struct ptlrpc_service_conf svc_conf;
2431 struct ost_obd *ost = &obd->u.ost;
2432 struct lprocfs_static_vars lvars;
2437 rc = cfs_cleanup_group_info();
2441 lprocfs_ost_init_vars(&lvars);
2442 lprocfs_obd_setup(obd, lvars.obd_vars);
2444 cfs_mutex_init(&ost->ost_health_mutex);
2446 svc_conf = (typeof(svc_conf)) {
2447 .psc_name = LUSTRE_OSS_NAME,
2448 .psc_watchdog_factor = OSS_SERVICE_WATCHDOG_FACTOR,
2450 .bc_nbufs = OST_NBUFS,
2451 .bc_buf_size = OST_BUFSIZE,
2452 .bc_req_max_size = OST_MAXREQSIZE,
2453 .bc_rep_max_size = OST_MAXREPSIZE,
2454 .bc_req_portal = OST_REQUEST_PORTAL,
2455 .bc_rep_portal = OSC_REPLY_PORTAL,
2458 .tc_thr_name = "ll_ost",
2459 .tc_thr_factor = OSS_THR_FACTOR,
2460 .tc_nthrs_init = OSS_NTHRS_INIT,
2461 .tc_nthrs_base = OSS_NTHRS_BASE,
2462 .tc_nthrs_max = OSS_NTHRS_MAX,
2463 .tc_nthrs_user = oss_num_threads,
2464 .tc_cpu_affinity = 1,
2465 .tc_ctx_tags = LCT_DT_THREAD,
2468 .cc_pattern = oss_cpts,
2471 .so_req_handler = ost_handle,
2472 .so_req_printer = target_print_req,
2473 .so_hpreq_handler = ptlrpc_hpreq_handler,
2476 ost->ost_service = ptlrpc_register_service(&svc_conf,
2477 obd->obd_proc_entry);
2478 if (IS_ERR(ost->ost_service)) {
2479 rc = PTR_ERR(ost->ost_service);
2480 CERROR("failed to start service: %d\n", rc);
2481 GOTO(out_lprocfs, rc);
2484 memset(&svc_conf, 0, sizeof(svc_conf));
2485 svc_conf = (typeof(svc_conf)) {
2486 .psc_name = "ost_create",
2487 .psc_watchdog_factor = OSS_SERVICE_WATCHDOG_FACTOR,
2489 .bc_nbufs = OST_NBUFS,
2490 .bc_buf_size = OST_BUFSIZE,
2491 .bc_req_max_size = OST_MAXREQSIZE,
2492 .bc_rep_max_size = OST_MAXREPSIZE,
2493 .bc_req_portal = OST_CREATE_PORTAL,
2494 .bc_rep_portal = OSC_REPLY_PORTAL,
2497 .tc_thr_name = "ll_ost_create",
2498 .tc_thr_factor = OSS_CR_THR_FACTOR,
2499 .tc_nthrs_init = OSS_CR_NTHRS_INIT,
2500 .tc_nthrs_base = OSS_CR_NTHRS_BASE,
2501 .tc_nthrs_max = OSS_CR_NTHRS_MAX,
2502 .tc_nthrs_user = oss_num_create_threads,
2503 .tc_cpu_affinity = 1,
2504 .tc_ctx_tags = LCT_DT_THREAD,
2507 .cc_pattern = oss_cpts,
2510 .so_req_handler = ost_handle,
2511 .so_req_printer = target_print_req,
2512 .so_hpreq_handler = NULL,
2515 ost->ost_create_service = ptlrpc_register_service(&svc_conf,
2516 obd->obd_proc_entry);
2517 if (IS_ERR(ost->ost_create_service)) {
2518 rc = PTR_ERR(ost->ost_create_service);
2519 CERROR("failed to start OST create service: %d\n", rc);
2520 GOTO(out_service, rc);
2523 mask = cfs_cpt_table->ctb_nodemask;
2524 /* event CPT feature is disabled in libcfs level by set partition
2525 * number to 1, we still want to set node affinity for io service */
2526 if (cfs_cpt_number(cfs_cpt_table) == 1 && nodes_weight(*mask) > 1) {
2530 ost_io_cptable = cfs_cpt_table_alloc(nodes_weight(*mask));
2531 for_each_node_mask(i, *mask) {
2532 if (ost_io_cptable == NULL) {
2533 CWARN("OSS failed to create CPT table\n");
2537 rc = cfs_cpt_set_node(ost_io_cptable, cpt++, i);
2539 CWARN("OSS Failed to set node %d for"
2540 "IO CPT table\n", i);
2541 cfs_cpt_table_free(ost_io_cptable);
2542 ost_io_cptable = NULL;
2548 memset(&svc_conf, 0, sizeof(svc_conf));
2549 svc_conf = (typeof(svc_conf)) {
2550 .psc_name = "ost_io",
2551 .psc_watchdog_factor = OSS_SERVICE_WATCHDOG_FACTOR,
2553 .bc_nbufs = OST_NBUFS,
2554 .bc_buf_size = OST_BUFSIZE,
2555 .bc_req_max_size = OST_MAXREQSIZE,
2556 .bc_rep_max_size = OST_MAXREPSIZE,
2557 .bc_req_portal = OST_IO_PORTAL,
2558 .bc_rep_portal = OSC_REPLY_PORTAL,
2561 .tc_thr_name = "ll_ost_io",
2562 .tc_thr_factor = OSS_THR_FACTOR,
2563 .tc_nthrs_init = OSS_NTHRS_INIT,
2564 .tc_nthrs_base = OSS_NTHRS_BASE,
2565 .tc_nthrs_max = OSS_NTHRS_MAX,
2566 .tc_nthrs_user = oss_num_threads,
2567 .tc_cpu_affinity = 1,
2568 .tc_ctx_tags = LCT_DT_THREAD,
2571 .cc_cptable = ost_io_cptable,
2572 .cc_pattern = ost_io_cptable == NULL ?
2576 .so_thr_init = ost_thread_init,
2577 .so_thr_done = ost_thread_done,
2578 .so_req_handler = ost_handle,
2579 .so_hpreq_handler = ost_hpreq_handler,
2580 .so_req_printer = target_print_req,
2581 .so_hpreq_handler = NULL,
2584 ost->ost_io_service = ptlrpc_register_service(&svc_conf,
2585 obd->obd_proc_entry);
2586 if (IS_ERR(ost->ost_io_service)) {
2587 rc = PTR_ERR(ost->ost_io_service);
2588 CERROR("failed to start OST I/O service: %d\n", rc);
2589 ost->ost_io_service = NULL;
2590 GOTO(out_create, rc);
2593 ping_evictor_start();
2598 ptlrpc_unregister_service(ost->ost_create_service);
2599 ost->ost_create_service = NULL;
2601 ptlrpc_unregister_service(ost->ost_service);
2602 ost->ost_service = NULL;
2604 lprocfs_obd_cleanup(obd);
2608 static int ost_cleanup(struct obd_device *obd)
2610 struct ost_obd *ost = &obd->u.ost;
2614 ping_evictor_stop();
2616 /* there is no recovery for OST OBD, all recovery is controlled by
2618 LASSERT(obd->obd_recovering == 0);
2619 cfs_mutex_lock(&ost->ost_health_mutex);
2620 ptlrpc_unregister_service(ost->ost_service);
2621 ptlrpc_unregister_service(ost->ost_create_service);
2622 ptlrpc_unregister_service(ost->ost_io_service);
2623 ost->ost_service = NULL;
2624 ost->ost_create_service = NULL;
2625 ost->ost_io_service = NULL;
2627 cfs_mutex_unlock(&ost->ost_health_mutex);
2629 lprocfs_obd_cleanup(obd);
2631 if (ost_io_cptable != NULL) {
2632 cfs_cpt_table_free(ost_io_cptable);
2633 ost_io_cptable = NULL;
2639 static int ost_health_check(const struct lu_env *env, struct obd_device *obd)
2641 struct ost_obd *ost = &obd->u.ost;
2644 cfs_mutex_lock(&ost->ost_health_mutex);
2645 rc |= ptlrpc_service_health_check(ost->ost_service);
2646 rc |= ptlrpc_service_health_check(ost->ost_create_service);
2647 rc |= ptlrpc_service_health_check(ost->ost_io_service);
2648 cfs_mutex_unlock(&ost->ost_health_mutex);
2651 * health_check to return 0 on healthy
2652 * and 1 on unhealthy.
2660 struct ost_thread_local_cache *ost_tls(struct ptlrpc_request *r)
2662 return (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
2665 /* use obd ops to offer management infrastructure */
2666 static struct obd_ops ost_obd_ops = {
2667 .o_owner = THIS_MODULE,
2668 .o_setup = ost_setup,
2669 .o_cleanup = ost_cleanup,
2670 .o_health_check = ost_health_check,
2674 static int __init ost_init(void)
2676 struct lprocfs_static_vars lvars;
2680 lprocfs_ost_init_vars(&lvars);
2681 rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars,
2682 LUSTRE_OSS_NAME, NULL);
2684 if (ost_num_threads != 0 && oss_num_threads == 0) {
2685 LCONSOLE_INFO("ost_num_threads module parameter is deprecated, "
2686 "use oss_num_threads instead or unset both for "
2687 "dynamic thread startup\n");
2688 oss_num_threads = ost_num_threads;
2694 static void /*__exit*/ ost_exit(void)
2696 class_unregister_type(LUSTRE_OSS_NAME);
2699 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2700 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
2701 MODULE_LICENSE("GPL");
2703 module_init(ost_init);
2704 module_exit(ost_exit);