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, 2013, 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 <lustre_fid.h>
52 #include <lustre_fld.h>
53 #include <linux/init.h>
54 #include <lprocfs_status.h>
55 #include <libcfs/list.h>
56 #include <lustre_quota.h>
57 #include <lustre_fid.h>
58 #include "ost_internal.h"
59 #include <lustre_fid.h>
61 static int oss_num_threads;
62 CFS_MODULE_PARM(oss_num_threads, "i", int, 0444,
63 "number of OSS service threads to start");
65 static int ost_num_threads;
66 CFS_MODULE_PARM(ost_num_threads, "i", int, 0444,
67 "number of OST service threads to start (deprecated)");
69 static int oss_num_create_threads;
70 CFS_MODULE_PARM(oss_num_create_threads, "i", int, 0444,
71 "number of OSS create threads to start");
73 static char *oss_cpts;
74 CFS_MODULE_PARM(oss_cpts, "s", charp, 0444,
75 "CPU partitions OSS threads should run on");
77 static char *oss_io_cpts;
78 CFS_MODULE_PARM(oss_io_cpts, "s", charp, 0444,
79 "CPU partitions OSS IO threads should run on");
82 * this page is allocated statically when module is initializing
83 * it is used to simulate data corruptions, see ost_checksum_bulk()
84 * for details. as the original pages provided by the layers below
85 * can be remain in the internal cache, we do not want to modify
88 static struct page *ost_page_to_corrupt = NULL;
91 * Do not return server-side uid/gid to remote client
93 static void ost_drop_id(struct obd_export *exp, struct obdo *oa)
95 if (exp_connect_rmtclient(exp)) {
98 oa->o_valid &= ~(OBD_MD_FLUID | OBD_MD_FLGID);
103 * Validate oa from client.
104 * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
106 * a. objects in Single MDT FS seq = FID_SEQ_OST_MDT0, oi_id != 0
107 * b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
108 * pack ost_id. Because non-zero oi_seq will make it diffcult to tell
109 * whether this is oi_fid or real ostid. So it will check
110 * OBD_CONNECT_FID, then convert the ostid to FID for old client.
111 * c. Old FID-disable osc will send IDIF.
112 * d. new FID-enable osc/osp will send normal FID.
114 * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
115 * be used for LAST_ID file, and only being accessed inside OST now.
117 static int ost_validate_obdo(struct obd_export *exp, struct obdo *oa,
118 struct obd_ioobj *ioobj)
122 if (unlikely(!(exp_connect_flags(exp) & OBD_CONNECT_FID) &&
123 fid_seq_is_echo(oa->o_oi.oi.oi_seq) && oa != NULL)) {
124 /* Sigh 2.[123] client still sends echo req with oi_id = 0
125 * during create, and we will reset this to 1, since this
126 * oi_id is basically useless in the following create process,
127 * but oi_id == 0 will make it difficult to tell whether it is
128 * real FID or ost_id. */
129 oa->o_oi.oi_fid.f_oid = oa->o_oi.oi.oi_id ?: 1;
130 oa->o_oi.oi_fid.f_seq = FID_SEQ_ECHO;
131 oa->o_oi.oi_fid.f_ver = 0;
133 if (unlikely((oa == NULL) || ostid_id(&oa->o_oi) == 0))
134 GOTO(out, rc = -EPROTO);
136 /* Note: this check might be forced in 2.5 or 2.6, i.e.
137 * all of the requests are required to setup FLGROUP */
138 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
139 ostid_set_seq_mdt0(&oa->o_oi);
141 ostid_set_seq_mdt0(&ioobj->ioo_oid);
142 oa->o_valid |= OBD_MD_FLGROUP;
145 if (unlikely(!(fid_seq_is_idif(ostid_seq(&oa->o_oi)) ||
146 fid_seq_is_mdt0(ostid_seq(&oa->o_oi)) ||
147 fid_seq_is_norm(ostid_seq(&oa->o_oi)) ||
148 fid_seq_is_echo(ostid_seq(&oa->o_oi)))))
149 GOTO(out, rc = -EPROTO);
153 unsigned max_brw = ioobj_max_brw_get(ioobj);
155 if (unlikely((max_brw & (max_brw - 1)) != 0)) {
156 CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
157 ": rc = -EPROTO\n", exp->exp_obd->obd_name,
158 obd_export_nid2str(exp), max_brw,
160 GOTO(out, rc = -EPROTO);
162 ioobj->ioo_oid = oa->o_oi;
167 CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
168 exp->exp_obd->obd_name, obd_export_nid2str(exp),
169 oa ? ostid_seq(&oa->o_oi) : -1,
170 oa ? ostid_id(&oa->o_oi) : -1, rc);
174 void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req)
176 struct oti_req_ack_lock *ack_lock;
182 if (req->rq_repmsg) {
183 __u64 versions[PTLRPC_NUM_VERSIONS] = { 0 };
184 lustre_msg_set_transno(req->rq_repmsg, oti->oti_transno);
185 versions[0] = oti->oti_pre_version;
186 lustre_msg_set_versions(req->rq_repmsg, versions);
188 req->rq_transno = oti->oti_transno;
190 /* XXX 4 == entries in oti_ack_locks??? */
191 for (ack_lock = oti->oti_ack_locks, i = 0; i < 4; i++, ack_lock++) {
194 /* XXX not even calling target_send_reply in some cases... */
195 ptlrpc_save_lock (req, &ack_lock->lock, ack_lock->mode, 0);
199 static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req,
200 struct obd_trans_info *oti)
202 struct ost_body *body, *repbody;
203 struct lustre_capa *capa = NULL;
207 /* Get the request body */
208 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
212 if (ostid_id(&body->oa.o_oi) == 0)
215 rc = ost_validate_obdo(exp, &body->oa, NULL);
219 /* If there's a DLM request, cancel the locks mentioned in it*/
220 if (req_capsule_field_present(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT)) {
221 struct ldlm_request *dlm;
223 dlm = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
226 ldlm_request_cancel(req, dlm, 0);
229 /* If there's a capability, get it */
230 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
231 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
233 CERROR("Missing capability for OST DESTROY");
238 /* Prepare the reply */
239 rc = req_capsule_server_pack(&req->rq_pill);
243 /* Get the log cancellation cookie */
244 if (body->oa.o_valid & OBD_MD_FLCOOKIE)
245 oti->oti_logcookies = &body->oa.o_lcookie;
247 /* Finish the reply */
248 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
249 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
251 /* Do the destroy and set the reply status accordingly */
252 req->rq_status = obd_destroy(req->rq_svc_thread->t_env, exp,
253 &repbody->oa, NULL, oti, NULL, capa);
258 * Helper function for getting server side [start, start+count] DLM lock
259 * if asked by client.
261 static int ost_lock_get(struct obd_export *exp, struct obdo *oa,
262 __u64 start, __u64 count, struct lustre_handle *lh,
263 int mode, __u64 flags)
265 struct ldlm_res_id res_id;
266 ldlm_policy_data_t policy;
267 __u64 end = start + count;
271 LASSERT(!lustre_handle_is_used(lh));
272 /* o_id and o_gr are used for localizing resource, if client miss to set
273 * them, do not trigger ASSERTION. */
274 if (unlikely((oa->o_valid & (OBD_MD_FLID | OBD_MD_FLGROUP)) !=
275 (OBD_MD_FLID | OBD_MD_FLGROUP)))
278 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
279 !(oa->o_flags & OBD_FL_SRVLOCK))
282 if (mode == LCK_MINMODE)
285 ostid_build_res_name(&oa->o_oi, &res_id);
286 CDEBUG(D_INODE, "OST-side extent lock.\n");
288 policy.l_extent.start = start & CFS_PAGE_MASK;
290 /* If ->o_blocks is EOF it means "lock till the end of the
291 * file". Otherwise, it's size of a hole being punched (in bytes) */
292 if (count == OBD_OBJECT_EOF || end < start)
293 policy.l_extent.end = OBD_OBJECT_EOF;
295 policy.l_extent.end = end | ~CFS_PAGE_MASK;
297 RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
298 LDLM_EXTENT, &policy, mode, &flags,
299 ldlm_blocking_ast, ldlm_completion_ast,
300 ldlm_glimpse_ast, NULL, 0, LVB_T_NONE,
304 /* Helper function: release lock, if any. */
305 static void ost_lock_put(struct obd_export *exp,
306 struct lustre_handle *lh, int mode)
309 if (lustre_handle_is_used(lh))
310 ldlm_lock_decref(lh, mode);
314 static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
316 struct ost_body *body, *repbody;
317 struct obd_info *oinfo;
318 struct lustre_handle lh = { 0 };
319 struct lustre_capa *capa = NULL;
320 ldlm_mode_t lock_mode;
324 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
328 rc = ost_validate_obdo(exp, &body->oa, NULL);
332 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
333 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
335 CERROR("Missing capability for OST GETATTR");
340 rc = req_capsule_server_pack(&req->rq_pill);
344 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
345 repbody->oa = body->oa;
347 lock_mode = LCK_MINMODE;
348 if (body->oa.o_valid & OBD_MD_FLFLAGS &&
349 body->oa.o_flags & OBD_FL_SRVLOCK) {
351 if (body->oa.o_flags & OBD_FL_FLUSH)
354 rc = ost_lock_get(exp, &repbody->oa, 0, OBD_OBJECT_EOF, &lh,
359 OBD_ALLOC_PTR(oinfo);
361 GOTO(unlock, rc = -ENOMEM);
362 oinfo->oi_oa = &repbody->oa;
363 oinfo->oi_capa = capa;
365 req->rq_status = obd_getattr(req->rq_svc_thread->t_env, exp, oinfo);
369 ost_drop_id(exp, &repbody->oa);
371 if (!(repbody->oa.o_valid & OBD_MD_FLFLAGS)) {
372 repbody->oa.o_valid |= OBD_MD_FLFLAGS;
373 repbody->oa.o_flags = 0;
375 repbody->oa.o_flags |= OBD_FL_FLUSH;
378 ost_lock_put(exp, &lh, lock_mode);
382 static int ost_statfs(struct ptlrpc_request *req)
384 struct obd_statfs *osfs;
388 rc = req_capsule_server_pack(&req->rq_pill);
392 osfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
394 req->rq_status = obd_statfs(req->rq_svc_thread->t_env, req->rq_export,
396 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
398 if (req->rq_status != 0)
399 CERROR("ost: statfs failed: rc %d\n", req->rq_status);
401 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS))
402 req->rq_status = -EINPROGRESS;
407 static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
408 struct obd_trans_info *oti)
410 struct ost_body *body, *repbody;
414 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
418 rc = ost_validate_obdo(req->rq_export, &body->oa, NULL);
422 rc = req_capsule_server_pack(&req->rq_pill);
426 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
427 repbody->oa = body->oa;
428 oti->oti_logcookies = &body->oa.o_lcookie;
430 req->rq_status = obd_create(req->rq_svc_thread->t_env, exp,
431 &repbody->oa, NULL, oti);
432 //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
436 static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
437 struct obd_trans_info *oti)
439 struct ost_body *body, *repbody;
441 struct lustre_handle lh = {0,};
445 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
446 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
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_FLSIZE | OBD_MD_FLBLOCKS)) !=
457 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
460 rc = req_capsule_server_pack(&req->rq_pill);
464 /* standard truncate optimization: if file body is completely
465 * destroyed, don't send data back to the server. */
466 if (body->oa.o_size == 0)
467 flags |= LDLM_FL_AST_DISCARD_DATA;
469 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
470 repbody->oa = body->oa;
472 rc = ost_lock_get(exp, &repbody->oa, repbody->oa.o_size,
473 repbody->oa.o_blocks, &lh, LCK_PW, flags);
475 struct obd_info *oinfo;
476 struct lustre_capa *capa = NULL;
478 if (repbody->oa.o_valid & OBD_MD_FLFLAGS &&
479 repbody->oa.o_flags == OBD_FL_SRVLOCK)
481 * If OBD_FL_SRVLOCK is the only bit set in
482 * ->o_flags, clear OBD_MD_FLFLAGS to avoid falling
483 * through filter_setattr() to filter_iocontrol().
485 repbody->oa.o_valid &= ~OBD_MD_FLFLAGS;
487 if (repbody->oa.o_valid & OBD_MD_FLOSSCAPA) {
488 capa = req_capsule_client_get(&req->rq_pill,
491 CERROR("Missing capability for OST PUNCH");
492 GOTO(unlock, rc = -EFAULT);
496 OBD_ALLOC_PTR(oinfo);
498 GOTO(unlock, rc = -ENOMEM);
499 oinfo->oi_oa = &repbody->oa;
500 oinfo->oi_policy.l_extent.start = oinfo->oi_oa->o_size;
501 oinfo->oi_policy.l_extent.end = oinfo->oi_oa->o_blocks;
502 oinfo->oi_capa = capa;
503 oinfo->oi_flags = OBD_FL_PUNCH;
505 req->rq_status = obd_punch(req->rq_svc_thread->t_env, exp,
509 ost_lock_put(exp, &lh, LCK_PW);
512 ost_drop_id(exp, &repbody->oa);
516 static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req,
517 struct obd_trans_info *oti)
519 struct ost_body *body, *repbody;
520 struct obd_info *oinfo;
521 struct lustre_capa *capa = NULL;
525 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
529 rc = ost_validate_obdo(exp, &body->oa, NULL);
533 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
534 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
536 CERROR("Missing capability for OST SYNC");
541 rc = req_capsule_server_pack(&req->rq_pill);
545 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
546 repbody->oa = body->oa;
548 OBD_ALLOC_PTR(oinfo);
552 oinfo->oi_oa = &repbody->oa;
553 oinfo->oi_capa = capa;
554 oinfo->oi_jobid = oti->oti_jobid;
555 req->rq_status = obd_sync(req->rq_svc_thread->t_env, exp, oinfo,
556 repbody->oa.o_size, repbody->oa.o_blocks,
560 ost_drop_id(exp, &repbody->oa);
564 static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req,
565 struct obd_trans_info *oti)
567 struct ost_body *body, *repbody;
568 struct obd_info *oinfo;
569 struct lustre_capa *capa = NULL;
573 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
577 rc = ost_validate_obdo(req->rq_export, &body->oa, NULL);
581 rc = req_capsule_server_pack(&req->rq_pill);
585 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
586 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
588 CERROR("Missing capability for OST SETATTR");
593 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
594 repbody->oa = body->oa;
596 OBD_ALLOC_PTR(oinfo);
599 oinfo->oi_oa = &repbody->oa;
600 oinfo->oi_capa = capa;
602 req->rq_status = obd_setattr(req->rq_svc_thread->t_env, exp, oinfo,
607 ost_drop_id(exp, &repbody->oa);
611 static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc,
612 cksum_type_t cksum_type)
614 struct cfs_crypto_hash_desc *hdesc;
615 unsigned int bufsize;
617 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
620 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
622 CERROR("Unable to initialize checksum hash %s\n",
623 cfs_crypto_hash_name(cfs_alg));
624 return PTR_ERR(hdesc);
626 CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
627 for (i = 0; i < desc->bd_iov_count; i++) {
629 /* corrupt the data before we compute the checksum, to
630 * simulate a client->OST data error */
631 if (i == 0 && opc == OST_WRITE &&
632 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
633 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
634 int len = desc->bd_iov[i].kiov_len;
635 struct page *np = ost_page_to_corrupt;
636 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
639 char *ptr2 = kmap(np) + off;
641 memcpy(ptr2, ptr, len);
642 memcpy(ptr2, "bad3", min(4, len));
644 desc->bd_iov[i].kiov_page = np;
646 CERROR("can't alloc page for corruption\n");
649 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
650 desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
651 desc->bd_iov[i].kiov_len);
653 /* corrupt the data after we compute the checksum, to
654 * simulate an OST->client data error */
655 if (i == 0 && opc == OST_READ &&
656 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
657 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
658 int len = desc->bd_iov[i].kiov_len;
659 struct page *np = ost_page_to_corrupt;
660 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
663 char *ptr2 = kmap(np) + off;
665 memcpy(ptr2, ptr, len);
666 memcpy(ptr2, "bad4", min(4, len));
668 desc->bd_iov[i].kiov_page = np;
670 CERROR("can't alloc page for corruption\n");
676 err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
678 cfs_crypto_hash_final(hdesc, NULL, NULL);
683 static int ost_brw_lock_get(int mode, struct obd_export *exp,
684 struct obd_ioobj *obj, struct niobuf_remote *nb,
685 struct lustre_handle *lh)
688 int nrbufs = obj->ioo_bufcnt;
689 struct ldlm_res_id res_id;
690 ldlm_policy_data_t policy;
694 ostid_build_res_name(&obj->ioo_oid, &res_id);
695 LASSERT(mode == LCK_PR || mode == LCK_PW);
696 LASSERT(!lustre_handle_is_used(lh));
698 if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
701 for (i = 1; i < nrbufs; i ++)
702 if ((nb[0].flags & OBD_BRW_SRVLOCK) !=
703 (nb[i].flags & OBD_BRW_SRVLOCK))
706 policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
707 policy.l_extent.end = (nb[nrbufs - 1].offset +
708 nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
710 RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
711 LDLM_EXTENT, &policy, mode, &flags,
712 ldlm_blocking_ast, ldlm_completion_ast,
713 ldlm_glimpse_ast, NULL, 0, LVB_T_NONE,
717 static void ost_brw_lock_put(int mode,
718 struct obd_ioobj *obj, struct niobuf_remote *niob,
719 struct lustre_handle *lh)
722 LASSERT(mode == LCK_PR || mode == LCK_PW);
723 LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
724 lustre_handle_is_used(lh));
725 if (lustre_handle_is_used(lh))
726 ldlm_lock_decref(lh, mode);
730 /* Allocate thread local buffers if needed */
731 static struct ost_thread_local_cache *ost_tls_get(struct ptlrpc_request *r)
733 struct ost_thread_local_cache *tls =
734 (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
736 /* In normal mode of operation an I/O request is serviced only
737 * by ll_ost_io threads each of them has own tls buffers allocated by
738 * ost_io_thread_init().
739 * During recovery, an I/O request may be queued until any of the ost
740 * service threads process it. Not necessary it should be one of
741 * ll_ost_io threads. In that case we dynamically allocating tls
742 * buffers for the request service time. */
743 if (unlikely(tls == NULL)) {
744 LASSERT(r->rq_export->exp_in_recovery);
748 r->rq_svc_thread->t_data = tls;
754 /* Free thread local buffers if they were allocated only for servicing
755 * this one request */
756 static void ost_tls_put(struct ptlrpc_request *r)
758 struct ost_thread_local_cache *tls =
759 (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
761 if (unlikely(tls->temporary)) {
763 r->rq_svc_thread->t_data = NULL;
767 static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
769 struct ptlrpc_bulk_desc *desc = NULL;
770 struct obd_export *exp = req->rq_export;
771 struct niobuf_remote *remote_nb;
772 struct niobuf_local *local_nb;
773 struct obd_ioobj *ioo;
774 struct ost_body *body, *repbody;
775 struct lustre_capa *capa = NULL;
776 struct l_wait_info lwi;
777 struct lustre_handle lockh = { 0 };
778 int niocount, npages, nob = 0, rc, i;
780 struct ost_thread_local_cache *tls;
783 req->rq_bulk_read = 1;
785 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
786 GOTO(out, rc = -EIO);
788 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
790 /* Check if there is eviction in progress, and if so, wait for it to
792 if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
793 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
794 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
795 !cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress),
799 GOTO(out, rc = -ENOTCONN);
801 /* ost_body, ioobj & noibuf_remote are verified and swabbed in
802 * ost_rw_hpreq_check(). */
803 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
805 GOTO(out, rc = -EFAULT);
808 * A req_capsule_X_get_array(pill, field, ptr_to_element_count) function
809 * would be useful here and wherever we get &RMF_OBD_IOOBJ and
810 * &RMF_NIOBUF_REMOTE.
812 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
814 GOTO(out, rc = -EFAULT);
816 rc = ost_validate_obdo(exp, &body->oa, ioo);
820 niocount = ioo->ioo_bufcnt;
821 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
822 if (remote_nb == NULL)
823 GOTO(out, rc = -EFAULT);
825 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
826 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
828 CERROR("Missing capability for OST BRW READ");
829 GOTO(out, rc = -EFAULT);
833 rc = req_capsule_server_pack(&req->rq_pill);
837 tls = ost_tls_get(req);
839 GOTO(out_bulk, rc = -ENOMEM);
840 local_nb = tls->local;
842 rc = ost_brw_lock_get(LCK_PR, exp, ioo, remote_nb, &lockh);
847 * If getting the lock took more time than
848 * client was willing to wait, drop it. b=11330
850 if (cfs_time_current_sec() > req->rq_deadline ||
851 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
853 CERROR("Dropping timed-out read from %s because locking"
854 "object "DOSTID" took %ld seconds (limit was %ld).\n",
855 libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
856 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
857 req->rq_deadline - req->rq_arrival_time.tv_sec);
858 GOTO(out_lock, rc = -ETIMEDOUT);
861 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
862 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
864 npages = OST_THREAD_POOL_SIZE;
865 rc = obd_preprw(req->rq_svc_thread->t_env, OBD_BRW_READ, exp,
866 &repbody->oa, 1, ioo, remote_nb, &npages, local_nb,
871 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
872 BULK_PUT_SOURCE, OST_BULK_PORTAL);
874 GOTO(out_commitrw, rc = -ENOMEM);
877 for (i = 0; i < npages; i++) {
878 int page_rc = local_nb[i].rc;
880 if (page_rc < 0) { /* error */
886 if (page_rc != 0) { /* some data! */
887 LASSERT (local_nb[i].page != NULL);
888 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
889 local_nb[i].lnb_page_offset,
893 if (page_rc != local_nb[i].len) { /* short read */
894 /* All subsequent pages should be 0 */
896 LASSERT(local_nb[i].rc == 0);
901 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
902 cksum_type_t cksum_type =
903 cksum_type_unpack(repbody->oa.o_valid & OBD_MD_FLFLAGS ?
904 repbody->oa.o_flags : 0);
905 repbody->oa.o_flags = cksum_type_pack(cksum_type);
906 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
907 repbody->oa.o_cksum = ost_checksum_bulk(desc, OST_READ,cksum_type);
908 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
909 repbody->oa.o_cksum);
911 repbody->oa.o_valid = 0;
913 /* We're finishing using body->oa as an input variable */
915 /* Check if client was evicted while we were doing i/o before touching
918 if (likely(!CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2)))
919 rc = target_bulk_io(exp, desc, &lwi);
924 /* Must commit after prep above in all cases */
925 rc = obd_commitrw(req->rq_svc_thread->t_env, OBD_BRW_READ, exp,
926 &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
930 ost_drop_id(exp, &repbody->oa);
933 ost_brw_lock_put(LCK_PR, ioo, remote_nb, &lockh);
937 if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
938 ptlrpc_free_bulk_nopin(desc);
942 req->rq_status = nob;
943 ptlrpc_lprocfs_brw(req, nob);
944 target_committed_to_req(req);
946 } else if (!no_reply) {
947 /* Only reply if there was no comms problem with bulk */
948 target_committed_to_req(req);
952 /* reply out callback would free */
953 ptlrpc_req_drop_rs(req);
954 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
955 "client will retry: rc %d\n",
956 exp->exp_obd->obd_name,
957 obd_uuid2str(&exp->exp_client_uuid),
958 obd_export_nid2str(exp), rc);
960 /* send a bulk after reply to simulate a network delay or reordering
962 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
964 struct l_wait_info lwi1;
966 CDEBUG(D_INFO, "reorder BULK\n");
967 cfs_waitq_init(&waitq);
969 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
970 l_wait_event(waitq, 0, &lwi1);
971 rc = target_bulk_io(exp, desc, &lwi);
972 ptlrpc_free_bulk_nopin(desc);
978 static void ost_warn_on_cksum(struct ptlrpc_request *req,
979 struct ptlrpc_bulk_desc *desc,
980 struct niobuf_local *local_nb, int npages,
981 obd_count client_cksum, obd_count server_cksum,
984 struct obd_export *exp = req->rq_export;
985 struct ost_body *body;
989 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
990 LASSERT (body != NULL);
992 if (req->rq_peer.nid == desc->bd_sender) {
996 router = libcfs_nid2str(desc->bd_sender);
1000 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1001 client_cksum, server_cksum);
1005 LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1006 DFID" object "DOSTID" extent ["LPU64"-"LPU64
1007 "]: client csum %x, server csum %x\n",
1008 exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1010 body->oa.o_valid & OBD_MD_FLFID ?
1011 body->oa.o_parent_seq : (__u64)0,
1012 body->oa.o_valid & OBD_MD_FLFID ?
1013 body->oa.o_parent_oid : 0,
1014 body->oa.o_valid & OBD_MD_FLFID ?
1015 body->oa.o_parent_ver : 0,
1016 POSTID(&body->oa.o_oi),
1017 local_nb[0].lnb_file_offset,
1018 local_nb[npages-1].lnb_file_offset +
1019 local_nb[npages-1].len - 1,
1020 client_cksum, server_cksum);
1023 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
1025 struct ptlrpc_bulk_desc *desc = NULL;
1026 struct obd_export *exp = req->rq_export;
1027 struct niobuf_remote *remote_nb;
1028 struct niobuf_local *local_nb;
1029 struct obd_ioobj *ioo;
1030 struct ost_body *body, *repbody;
1031 struct l_wait_info lwi;
1032 struct lustre_handle lockh = {0};
1033 struct lustre_capa *capa = NULL;
1035 int objcount, niocount, npages;
1037 obd_count client_cksum = 0, server_cksum = 0;
1038 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
1039 int no_reply = 0, mmap = 0;
1040 __u32 o_uid = 0, o_gid = 0;
1041 struct ost_thread_local_cache *tls;
1044 req->rq_bulk_write = 1;
1046 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1047 GOTO(out, rc = -EIO);
1048 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1049 GOTO(out, rc = -EFAULT);
1051 /* pause before transaction has been started */
1052 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
1054 /* ost_body, ioobj & noibuf_remote are verified and swabbed in
1055 * ost_rw_hpreq_check(). */
1056 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1058 GOTO(out, rc = -EFAULT);
1060 objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1061 RCL_CLIENT) / sizeof(*ioo);
1062 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1064 GOTO(out, rc = -EFAULT);
1066 rc = ost_validate_obdo(exp, &body->oa, ioo);
1070 for (niocount = i = 0; i < objcount; i++)
1071 niocount += ioo[i].ioo_bufcnt;
1074 * It'd be nice to have a capsule function to indicate how many elements
1075 * there were in a buffer for an RMF that's declared to be an array.
1076 * It's easy enough to compute the number of elements here though.
1078 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1079 if (remote_nb == NULL || niocount != (req_capsule_get_size(&req->rq_pill,
1080 &RMF_NIOBUF_REMOTE, RCL_CLIENT) / sizeof(*remote_nb)))
1081 GOTO(out, rc = -EFAULT);
1083 if ((remote_nb[0].flags & OBD_BRW_MEMALLOC) &&
1084 (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
1085 memory_pressure_set();
1087 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
1088 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
1090 CERROR("Missing capability for OST BRW WRITE");
1091 GOTO(out, rc = -EFAULT);
1095 req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
1096 niocount * sizeof(*rcs));
1097 rc = req_capsule_server_pack(&req->rq_pill);
1100 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
1101 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
1103 tls = ost_tls_get(req);
1105 GOTO(out_bulk, rc = -ENOMEM);
1106 local_nb = tls->local;
1108 rc = ost_brw_lock_get(LCK_PW, exp, ioo, remote_nb, &lockh);
1113 * If getting the lock took more time than
1114 * client was willing to wait, drop it. b=11330
1116 if (cfs_time_current_sec() > req->rq_deadline ||
1117 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1119 CERROR("Dropping timed-out write from %s because locking "
1120 "object "DOSTID" took %ld seconds (limit was %ld).\n",
1121 libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1122 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1123 req->rq_deadline - req->rq_arrival_time.tv_sec);
1124 GOTO(out_lock, rc = -ETIMEDOUT);
1127 /* obd_preprw clobbers oa->valid, so save what we need */
1128 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1129 client_cksum = body->oa.o_cksum;
1130 if (body->oa.o_valid & OBD_MD_FLFLAGS)
1131 cksum_type = cksum_type_unpack(body->oa.o_flags);
1133 if (body->oa.o_valid & OBD_MD_FLFLAGS && body->oa.o_flags & OBD_FL_MMAP)
1136 /* Because we already sync grant info with client when reconnect,
1137 * grant info will be cleared for resent req, then fed_grant and
1138 * total_grant will not be modified in following preprw_write */
1139 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1140 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1141 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1144 if (exp_connect_rmtclient(exp)) {
1145 o_uid = body->oa.o_uid;
1146 o_gid = body->oa.o_gid;
1149 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1150 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
1152 npages = OST_THREAD_POOL_SIZE;
1153 rc = obd_preprw(req->rq_svc_thread->t_env, OBD_BRW_WRITE, exp,
1154 &repbody->oa, objcount, ioo, remote_nb, &npages,
1155 local_nb, oti, capa);
1159 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1160 BULK_GET_SINK, OST_BULK_PORTAL);
1162 GOTO(skip_transfer, rc = -ENOMEM);
1164 /* NB Having prepped, we must commit... */
1165 for (i = 0; i < npages; i++)
1166 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
1167 local_nb[i].lnb_page_offset,
1170 rc = sptlrpc_svc_prep_bulk(req, desc);
1174 rc = target_bulk_io(exp, desc, &lwi);
1178 if (client_cksum != 0 && rc == 0) {
1179 static int cksum_counter;
1180 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1181 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
1182 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
1183 server_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1184 repbody->oa.o_cksum = server_cksum;
1186 if (unlikely(client_cksum != server_cksum)) {
1187 ost_warn_on_cksum(req, desc, local_nb, npages,
1188 client_cksum, server_cksum, mmap);
1191 } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){
1192 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1193 cksum_counter, libcfs_id2str(req->rq_peer),
1198 /* Must commit after prep above in all cases */
1199 rc = obd_commitrw(req->rq_svc_thread->t_env, OBD_BRW_WRITE, exp,
1200 &repbody->oa, objcount, ioo, remote_nb, npages,
1202 if (rc == -ENOTCONN)
1203 /* quota acquire process has been given up because
1204 * either the client has been evicted or the client
1205 * has timed out the request already */
1208 if (exp_connect_rmtclient(exp)) {
1209 repbody->oa.o_uid = o_uid;
1210 repbody->oa.o_gid = o_gid;
1214 * Disable sending mtime back to the client. If the client locked the
1215 * whole object, then it has already updated the mtime on its side,
1216 * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
1218 repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
1223 /* set per-requested niobuf return codes */
1224 for (i = j = 0; i < niocount; i++) {
1225 int len = remote_nb[i].len;
1230 LASSERT(j < npages);
1231 if (local_nb[j].rc < 0)
1232 rcs[i] = local_nb[j].rc;
1233 len -= local_nb[j].len;
1238 LASSERT(j == npages);
1239 ptlrpc_lprocfs_brw(req, nob);
1243 ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh);
1248 ptlrpc_free_bulk_nopin(desc);
1251 oti_to_request(oti, req);
1252 target_committed_to_req(req);
1253 rc = ptlrpc_reply(req);
1254 } else if (!no_reply) {
1255 /* Only reply if there was no comms problem with bulk */
1256 target_committed_to_req(req);
1257 req->rq_status = rc;
1260 /* reply out callback would free */
1261 ptlrpc_req_drop_rs(req);
1262 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
1263 "client will retry: rc %d\n",
1264 exp->exp_obd->obd_name,
1265 obd_uuid2str(&exp->exp_client_uuid),
1266 obd_export_nid2str(exp), rc);
1268 memory_pressure_clr();
1273 * Implementation of OST_SET_INFO.
1275 * OST_SET_INFO is like ioctl(): heavily overloaded. Specifically, it takes a
1276 * "key" and a value RPC buffers as arguments, with the value's contents
1277 * interpreted according to the key.
1279 * Value types that need swabbing have swabbing done explicitly, either here or
1280 * in functions called from here. This should be corrected: all swabbing should
1281 * be done in the capsule abstraction, as that will then allow us to move
1282 * swabbing exclusively to the client without having to modify server code
1283 * outside the capsule abstraction's implementation itself. To correct this
1284 * will require minor changes to the capsule abstraction; see the comments for
1285 * req_capsule_extend() in layout.c.
1287 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
1289 struct ost_body *body = NULL, *repbody;
1290 char *key, *val = NULL;
1291 int keylen, vallen, rc = 0;
1292 int is_grant_shrink = 0;
1295 key = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
1297 DEBUG_REQ(D_HA, req, "no set_info key");
1300 keylen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_KEY,
1303 vallen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_VAL,
1306 if ((is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK)))
1307 /* In this case the value is actually an RMF_OST_BODY, so we
1308 * transmutate the type of this PTLRPC */
1309 req_capsule_extend(&req->rq_pill, &RQF_OST_SET_GRANT_INFO);
1311 rc = req_capsule_server_pack(&req->rq_pill);
1316 if (is_grant_shrink) {
1317 body = req_capsule_client_get(&req->rq_pill,
1322 repbody = req_capsule_server_get(&req->rq_pill,
1324 memcpy(repbody, body, sizeof(*body));
1325 val = (char*)repbody;
1327 val = req_capsule_client_get(&req->rq_pill,
1332 if (KEY_IS(KEY_EVICT_BY_NID)) {
1334 obd_export_evict_by_nid(exp->exp_obd, val);
1336 } else if (KEY_IS(KEY_MDS_CONN) && ptlrpc_req_need_swab(req)) {
1337 if (vallen < sizeof(__u32))
1339 __swab32s((__u32 *)val);
1342 /* OBD will also check if KEY_IS(KEY_GRANT_SHRINK), and will cast val to
1343 * a struct ost_body * value */
1344 rc = obd_set_info_async(req->rq_svc_thread->t_env, exp, keylen,
1345 key, vallen, val, NULL);
1347 lustre_msg_set_status(req->rq_repmsg, 0);
1351 struct locked_region {
1353 struct lustre_handle lh;
1356 static int lock_region(struct obd_export *exp, struct obdo *oa,
1357 unsigned long long begin, unsigned long long end,
1360 struct locked_region *region = NULL;
1363 LASSERT(begin <= end);
1364 OBD_ALLOC_PTR(region);
1368 rc = ost_lock_get(exp, oa, begin, end - begin, ®ion->lh, LCK_PR, 0);
1370 OBD_FREE_PTR(region);
1374 CDEBUG(D_OTHER, "ost lock [%llu,%llu], lh=%p\n",
1375 begin, end, ®ion->lh);
1376 cfs_list_add(®ion->list, locked);
1381 static int lock_zero_regions(struct obd_export *exp, struct obdo *oa,
1382 struct ll_user_fiemap *fiemap,
1385 __u64 begin = fiemap->fm_start;
1388 struct ll_fiemap_extent *fiemap_start = fiemap->fm_extents;
1391 CDEBUG(D_OTHER, "extents count %u\n", fiemap->fm_mapped_extents);
1392 for (i = 0; i < fiemap->fm_mapped_extents; i++) {
1393 if (fiemap_start[i].fe_logical > begin) {
1394 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1395 begin, fiemap_start[i].fe_logical);
1396 rc = lock_region(exp, oa, begin,
1397 fiemap_start[i].fe_logical, locked);
1402 begin = fiemap_start[i].fe_logical + fiemap_start[i].fe_length;
1405 if (begin < (fiemap->fm_start + fiemap->fm_length)) {
1406 CDEBUG(D_OTHER, "ost lock [%llu,%llu]\n",
1407 begin, fiemap->fm_start + fiemap->fm_length);
1408 rc = lock_region(exp, oa, begin,
1409 fiemap->fm_start + fiemap->fm_length, locked);
1415 static void unlock_zero_regions(struct obd_export *exp, cfs_list_t *locked)
1417 struct locked_region *entry, *temp;
1418 cfs_list_for_each_entry_safe(entry, temp, locked, list) {
1419 CDEBUG(D_OTHER, "ost unlock lh=%p\n", &entry->lh);
1420 ost_lock_put(exp, &entry->lh, LCK_PR);
1421 cfs_list_del(&entry->list);
1422 OBD_FREE_PTR(entry);
1426 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
1429 int keylen, replylen, rc = 0;
1430 struct req_capsule *pill = &req->rq_pill;
1431 cfs_list_t locked = CFS_LIST_HEAD_INIT(locked);
1432 struct ll_fiemap_info_key *fm_key = NULL;
1433 struct ll_user_fiemap *fiemap;
1436 /* this common part for get_info rpc */
1437 key = req_capsule_client_get(pill, &RMF_SETINFO_KEY);
1439 DEBUG_REQ(D_HA, req, "no get_info key");
1442 keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT);
1444 if (KEY_IS(KEY_FIEMAP)) {
1446 rc = ost_validate_obdo(exp, &fm_key->oa, NULL);
1451 rc = obd_get_info(req->rq_svc_thread->t_env, exp, keylen, key,
1452 &replylen, NULL, NULL);
1456 req_capsule_set_size(pill, &RMF_GENERIC_DATA,
1457 RCL_SERVER, replylen);
1459 rc = req_capsule_server_pack(pill);
1463 reply = req_capsule_server_get(pill, &RMF_GENERIC_DATA);
1467 if (KEY_IS(KEY_LAST_FID)) {
1471 req_capsule_extend(pill, &RQF_OST_GET_INFO_LAST_FID);
1472 val = req_capsule_client_get(pill, &RMF_SETINFO_VAL);
1473 vallen = req_capsule_get_size(pill, &RMF_SETINFO_VAL,
1475 if (val != NULL && vallen > 0 && replylen >= vallen) {
1476 memcpy(reply, val, vallen);
1478 CERROR("%s: invalid req val %p vallen %d replylen %d\n",
1479 exp->exp_obd->obd_name, val, vallen, replylen);
1484 /* call again to fill in the reply buffer */
1485 rc = obd_get_info(req->rq_svc_thread->t_env, exp, keylen, key,
1486 &replylen, reply, NULL);
1488 /* LU-3219: Lock the sparse areas to make sure dirty flushed back
1489 * from client, then call fiemap again. */
1490 if (KEY_IS(KEY_FIEMAP) && (fm_key->oa.o_valid & OBD_MD_FLFLAGS) &&
1491 (fm_key->oa.o_flags & OBD_FL_SRVLOCK)) {
1492 fiemap = (struct ll_user_fiemap *)reply;
1495 rc = lock_zero_regions(exp, &fm_key->oa, fiemap, &locked);
1496 if (rc == 0 && !cfs_list_empty(&locked))
1497 rc = obd_get_info(req->rq_svc_thread->t_env, exp,
1498 keylen, key, &replylen, reply, NULL);
1499 unlock_zero_regions(exp, &locked);
1504 lustre_msg_set_status(req->rq_repmsg, 0);
1509 static int ost_handle_quotactl(struct ptlrpc_request *req)
1511 struct obd_quotactl *oqctl, *repoqc;
1515 oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1517 GOTO(out, rc = -EPROTO);
1519 rc = req_capsule_server_pack(&req->rq_pill);
1523 repoqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1524 req->rq_status = obd_quotactl(req->rq_export, oqctl);
1531 static int ost_handle_quotacheck(struct ptlrpc_request *req)
1533 struct obd_quotactl *oqctl;
1537 oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1541 rc = req_capsule_server_pack(&req->rq_pill);
1545 /* deprecated, not used any more */
1546 req->rq_status = -EOPNOTSUPP;
1547 RETURN(-EOPNOTSUPP);
1550 static int ost_llog_handle_connect(struct obd_export *exp,
1551 struct ptlrpc_request *req)
1553 struct llogd_conn_body *body;
1557 body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_CONN_BODY);
1558 rc = obd_llog_connect(exp, body);
1562 #define ost_init_sec_none(reply, exp) \
1564 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT | \
1565 OBD_CONNECT_RMT_CLIENT_FORCE | \
1566 OBD_CONNECT_OSS_CAPA); \
1567 spin_lock(&exp->exp_lock); \
1568 *exp_connect_flags_ptr(exp) = reply->ocd_connect_flags; \
1569 spin_unlock(&exp->exp_lock); \
1572 static int ost_init_sec_level(struct ptlrpc_request *req)
1574 struct obd_export *exp = req->rq_export;
1575 struct req_capsule *pill = &req->rq_pill;
1576 struct obd_device *obd = exp->exp_obd;
1577 struct filter_obd *filter = &obd->u.filter;
1578 char *client = libcfs_nid2str(req->rq_peer.nid);
1579 struct obd_connect_data *data, *reply;
1583 data = req_capsule_client_get(pill, &RMF_CONNECT_DATA);
1584 reply = req_capsule_server_get(pill, &RMF_CONNECT_DATA);
1585 if (data == NULL || reply == NULL)
1588 /* connection from MDT is always trusted */
1589 if (req->rq_auth_usr_mdt) {
1590 ost_init_sec_none(reply, exp);
1594 /* no GSS support case */
1595 if (!req->rq_auth_gss) {
1596 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1597 CWARN("client %s -> target %s does not user GSS, "
1598 "can not run under security level %d.\n",
1599 client, obd->obd_name, filter->fo_sec_level);
1602 ost_init_sec_none(reply, exp);
1607 /* old version case */
1608 if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
1609 !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
1610 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1611 CWARN("client %s -> target %s uses old version, "
1612 "can not run under security level %d.\n",
1613 client, obd->obd_name, filter->fo_sec_level);
1616 CWARN("client %s -> target %s uses old version, "
1617 "run under security level %d.\n",
1618 client, obd->obd_name, filter->fo_sec_level);
1619 ost_init_sec_none(reply, exp);
1624 remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
1626 if (!req->rq_auth_remote)
1627 CDEBUG(D_SEC, "client (local realm) %s -> target %s "
1628 "asked to be remote.\n", client, obd->obd_name);
1629 } else if (req->rq_auth_remote) {
1631 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
1632 "as remote by default.\n", client, obd->obd_name);
1636 if (!filter->fo_fl_oss_capa) {
1637 CDEBUG(D_SEC, "client %s -> target %s is set as remote,"
1638 " but OSS capabilities are not enabled: %d.\n",
1639 client, obd->obd_name, filter->fo_fl_oss_capa);
1644 switch (filter->fo_sec_level) {
1645 case LUSTRE_SEC_NONE:
1647 ost_init_sec_none(reply, exp);
1650 CDEBUG(D_SEC, "client %s -> target %s is set as remote, "
1651 "can not run under security level %d.\n",
1652 client, obd->obd_name, filter->fo_sec_level);
1655 case LUSTRE_SEC_REMOTE:
1657 ost_init_sec_none(reply, exp);
1659 case LUSTRE_SEC_ALL:
1661 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
1662 OBD_CONNECT_RMT_CLIENT_FORCE);
1663 if (!filter->fo_fl_oss_capa)
1664 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
1666 spin_lock(&exp->exp_lock);
1667 *exp_connect_flags_ptr(exp) = reply->ocd_connect_flags;
1668 spin_unlock(&exp->exp_lock);
1680 * this should be done in filter_connect()/filter_reconnect(), but
1681 * we can't obtain information like NID, which stored in incoming
1682 * request, thus can't decide what flavor to use. so we do it here.
1684 * This hack should be removed after the OST stack be rewritten, just
1685 * like what we are doing in mdt_obd_connect()/mdt_obd_reconnect().
1687 static int ost_connect_check_sptlrpc(struct ptlrpc_request *req)
1689 struct obd_export *exp = req->rq_export;
1690 struct filter_obd *filter = &exp->exp_obd->u.filter;
1691 struct sptlrpc_flavor flvr;
1694 if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
1695 LUSTRE_ECHO_NAME) == 0)) {
1696 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
1700 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
1701 read_lock(&filter->fo_sptlrpc_lock);
1702 sptlrpc_target_choose_flavor(&filter->fo_sptlrpc_rset,
1706 read_unlock(&filter->fo_sptlrpc_lock);
1708 spin_lock(&exp->exp_lock);
1710 exp->exp_sp_peer = req->rq_sp_from;
1711 exp->exp_flvr = flvr;
1713 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
1714 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
1715 CERROR("unauthorized rpc flavor %x from %s, "
1716 "expect %x\n", req->rq_flvr.sf_rpc,
1717 libcfs_nid2str(req->rq_peer.nid),
1718 exp->exp_flvr.sf_rpc);
1722 spin_unlock(&exp->exp_lock);
1724 if (exp->exp_sp_peer != req->rq_sp_from) {
1725 CERROR("RPC source %s doesn't match %s\n",
1726 sptlrpc_part2name(req->rq_sp_from),
1727 sptlrpc_part2name(exp->exp_sp_peer));
1730 rc = sptlrpc_target_export_check(exp, req);
1737 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1739 int ost_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1740 void *data, int flag)
1743 __u32 sync_lock_cancel = 0;
1744 __u32 len = sizeof(sync_lock_cancel);
1749 rc = lu_env_init(&env, LCT_DT_THREAD);
1750 if (unlikely(rc != 0))
1753 rc = obd_get_info(&env, lock->l_export, sizeof(KEY_SYNC_LOCK_CANCEL),
1754 KEY_SYNC_LOCK_CANCEL, &len, &sync_lock_cancel, NULL);
1755 if (rc == 0 && flag == LDLM_CB_CANCELING &&
1756 (lock->l_granted_mode & (LCK_PW|LCK_GROUP)) &&
1757 (sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1758 (sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1759 lock->l_flags & LDLM_FL_CBPENDING))) {
1760 struct obd_info *oinfo;
1764 OBD_ALLOC_PTR(oinfo);
1766 GOTO(out_env, rc = -ENOMEM);
1769 OBD_FREE_PTR(oinfo);
1770 GOTO(out_env, rc = -ENOMEM);
1773 ostid_res_name_to_id(&oa->o_oi, &lock->l_resource->lr_name);
1774 oa->o_valid = OBD_MD_FLID|OBD_MD_FLGROUP;
1776 oinfo->oi_capa = BYPASS_CAPA;
1778 rc = obd_sync(&env, lock->l_export, oinfo,
1779 lock->l_policy_data.l_extent.start,
1780 lock->l_policy_data.l_extent.end, NULL);
1782 CERROR("Error %d syncing data on lock cancel\n", rc);
1785 OBD_FREE_PTR(oinfo);
1788 rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1794 static int ost_filter_recovery_request(struct ptlrpc_request *req,
1795 struct obd_device *obd, int *process)
1797 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1798 case OST_CONNECT: /* This will never get here, but for completeness. */
1799 case OST_DISCONNECT:
1810 case OBD_LOG_CANCEL:
1812 *process = target_queue_recovery_request(req, obd);
1816 DEBUG_REQ(D_WARNING, req, "not permitted during recovery");
1822 int ost_msg_check_version(struct lustre_msg *msg)
1826 switch(lustre_msg_get_opc(msg)) {
1828 case OST_DISCONNECT:
1831 case SEC_CTX_INIT_CONT:
1833 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
1835 CERROR("bad opc %u version %08x, expecting %08x\n",
1836 lustre_msg_get_opc(msg),
1837 lustre_msg_get_version(msg),
1838 LUSTRE_OBD_VERSION);
1851 case OST_QUOTACHECK:
1853 rc = lustre_msg_check_version(msg, LUSTRE_OST_VERSION);
1855 CERROR("bad opc %u version %08x, expecting %08x\n",
1856 lustre_msg_get_opc(msg),
1857 lustre_msg_get_version(msg),
1858 LUSTRE_OST_VERSION);
1863 case LDLM_BL_CALLBACK:
1864 case LDLM_CP_CALLBACK:
1865 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
1867 CERROR("bad opc %u version %08x, expecting %08x\n",
1868 lustre_msg_get_opc(msg),
1869 lustre_msg_get_version(msg),
1870 LUSTRE_DLM_VERSION);
1872 case LLOG_ORIGIN_CONNECT:
1873 case OBD_LOG_CANCEL:
1874 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
1876 CERROR("bad opc %u version %08x, expecting %08x\n",
1877 lustre_msg_get_opc(msg),
1878 lustre_msg_get_version(msg),
1879 LUSTRE_LOG_VERSION);
1881 case OST_QUOTA_ADJUST_QUNIT:
1883 CERROR("Quota adjust is deprecated as of 2.4.0\n");
1886 CERROR("Unexpected opcode %d\n", lustre_msg_get_opc(msg));
1892 struct ost_prolong_data {
1893 struct ptlrpc_request *opd_req;
1894 struct obd_export *opd_exp;
1895 struct obdo *opd_oa;
1896 struct ldlm_res_id opd_resid;
1897 struct ldlm_extent opd_extent;
1898 ldlm_mode_t opd_mode;
1899 unsigned int opd_locks;
1903 /* prolong locks for the current service time of the corresponding
1904 * portal (= OST_IO_PORTAL)
1906 static inline int prolong_timeout(struct ptlrpc_request *req)
1908 struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
1911 return obd_timeout / 2;
1913 return max(at_est2timeout(at_get(&svcpt->scp_at_estimate)),
1917 static void ost_prolong_lock_one(struct ost_prolong_data *opd,
1918 struct ldlm_lock *lock)
1920 LASSERT(lock->l_export == opd->opd_exp);
1922 if (lock->l_flags & LDLM_FL_DESTROYED) /* lock already cancelled */
1925 /* XXX: never try to grab resource lock here because we're inside
1926 * exp_bl_list_lock; in ldlm_lockd.c to handle waiting list we take
1927 * res lock and then exp_bl_list_lock. */
1929 if (!(lock->l_flags & LDLM_FL_AST_SENT))
1930 /* ignore locks not being cancelled */
1934 "refreshed for req x"LPU64" ext("LPU64"->"LPU64") to %ds.\n",
1935 opd->opd_req->rq_xid, opd->opd_extent.start,
1936 opd->opd_extent.end, opd->opd_timeout);
1938 /* OK. this is a possible lock the user holds doing I/O
1939 * let's refresh eviction timer for it */
1940 ldlm_refresh_waiting_lock(lock, opd->opd_timeout);
1944 static void ost_prolong_locks(struct ost_prolong_data *data)
1946 struct obd_export *exp = data->opd_exp;
1947 struct obdo *oa = data->opd_oa;
1948 struct ldlm_lock *lock;
1951 if (oa->o_valid & OBD_MD_FLHANDLE) {
1952 /* mostly a request should be covered by only one lock, try
1954 lock = ldlm_handle2lock(&oa->o_handle);
1956 /* Fast path to check if the lock covers the whole IO
1957 * region exclusively. */
1958 if (lock->l_granted_mode == LCK_PW &&
1959 ldlm_extent_contain(&lock->l_policy_data.l_extent,
1960 &data->opd_extent)) {
1962 ost_prolong_lock_one(data, lock);
1963 LDLM_LOCK_PUT(lock);
1966 LDLM_LOCK_PUT(lock);
1971 spin_lock_bh(&exp->exp_bl_list_lock);
1972 cfs_list_for_each_entry(lock, &exp->exp_bl_list, l_exp_list) {
1973 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1974 LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
1976 if (!ldlm_res_eq(&data->opd_resid, &lock->l_resource->lr_name))
1979 if (!ldlm_extent_overlap(&lock->l_policy_data.l_extent,
1983 ost_prolong_lock_one(data, lock);
1985 spin_unlock_bh(&exp->exp_bl_list_lock);
1991 * Returns 1 if the given PTLRPC matches the given LDLM locks, or 0 if it does
1994 static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req,
1995 struct ldlm_lock *lock)
1997 struct niobuf_remote *nb;
1998 struct obd_ioobj *ioo;
2000 struct ldlm_extent ext;
2003 opc = lustre_msg_get_opc(req->rq_reqmsg);
2004 LASSERT(opc == OST_READ || opc == OST_WRITE);
2006 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2007 LASSERT(ioo != NULL);
2009 nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2010 LASSERT(nb != NULL);
2012 ext.start = nb->offset;
2013 nb += ioo->ioo_bufcnt - 1;
2014 ext.end = nb->offset + nb->len - 1;
2016 LASSERT(lock->l_resource != NULL);
2017 if (!ostid_res_name_eq(&ioo->ioo_oid, &lock->l_resource->lr_name))
2021 if (opc == OST_READ)
2023 if (!(lock->l_granted_mode & mode))
2026 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
2030 * High-priority queue request check for whether the given PTLRPC request (\a
2031 * req) is blocking an LDLM lock cancel.
2033 * Returns 1 if the given given PTLRPC request (\a req) is blocking an LDLM lock
2034 * cancel, 0 if it is not, and -EFAULT if the request is malformed.
2036 * Only OST_READs, OST_WRITEs and OST_PUNCHes go on the h-p RPC queue. This
2037 * function looks only at OST_READs and OST_WRITEs.
2039 static int ost_rw_hpreq_check(struct ptlrpc_request *req)
2041 struct obd_device *obd = req->rq_export->exp_obd;
2042 struct ost_body *body;
2043 struct obd_ioobj *ioo;
2044 struct niobuf_remote *nb;
2045 struct ost_prolong_data opd = { 0 };
2050 * Use LASSERT to do sanity check because malformed RPCs should have
2051 * been filtered out in ost_hpreq_handler().
2053 opc = lustre_msg_get_opc(req->rq_reqmsg);
2054 LASSERT(opc == OST_READ || opc == OST_WRITE);
2056 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2057 LASSERT(body != NULL);
2059 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2060 LASSERT(ioo != NULL);
2062 nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2063 LASSERT(nb != NULL);
2064 LASSERT(!(nb->flags & OBD_BRW_SRVLOCK));
2066 ostid_build_res_name(&ioo->ioo_oid, &opd.opd_resid);
2070 if (opc == OST_READ)
2072 opd.opd_mode = mode;
2073 opd.opd_exp = req->rq_export;
2074 opd.opd_oa = &body->oa;
2075 opd.opd_extent.start = nb->offset;
2076 nb += ioo->ioo_bufcnt - 1;
2077 opd.opd_extent.end = nb->offset + nb->len - 1;
2078 opd.opd_timeout = prolong_timeout(req);
2080 DEBUG_REQ(D_RPCTRACE, req,
2081 "%s %s: refresh rw locks: " LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
2082 obd->obd_name, cfs_current()->comm,
2083 opd.opd_resid.name[0], opd.opd_resid.name[1],
2084 opd.opd_extent.start, opd.opd_extent.end);
2086 ost_prolong_locks(&opd);
2088 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2089 obd->obd_name, opd.opd_locks, req);
2091 RETURN(opd.opd_locks > 0);
2094 static void ost_rw_hpreq_fini(struct ptlrpc_request *req)
2096 (void)ost_rw_hpreq_check(req);
2100 * Like ost_rw_hpreq_lock_match(), but for OST_PUNCH RPCs.
2102 static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req,
2103 struct ldlm_lock *lock)
2105 struct ost_body *body;
2108 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2109 LASSERT(body != NULL);
2111 if (body->oa.o_valid & OBD_MD_FLHANDLE &&
2112 body->oa.o_handle.cookie == lock->l_handle.h_cookie)
2119 * Like ost_rw_hpreq_check(), but for OST_PUNCH RPCs.
2121 static int ost_punch_hpreq_check(struct ptlrpc_request *req)
2123 struct obd_device *obd = req->rq_export->exp_obd;
2124 struct ost_body *body;
2126 struct ost_prolong_data opd = { 0 };
2130 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2131 LASSERT(body != NULL);
2134 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS) ||
2135 !(oa->o_flags & OBD_FL_SRVLOCK));
2138 end = start + oa->o_blocks;
2141 opd.opd_mode = LCK_PW;
2142 opd.opd_exp = req->rq_export;
2144 opd.opd_extent.start = start;
2145 opd.opd_extent.end = end;
2146 if (oa->o_blocks == OBD_OBJECT_EOF)
2147 opd.opd_extent.end = OBD_OBJECT_EOF;
2148 opd.opd_timeout = prolong_timeout(req);
2150 ostid_build_res_name(&oa->o_oi, &opd.opd_resid);
2153 "%s: refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
2155 opd.opd_resid.name[0], opd.opd_resid.name[1],
2156 opd.opd_extent.start, opd.opd_extent.end);
2158 ost_prolong_locks(&opd);
2160 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
2161 obd->obd_name, opd.opd_locks, req);
2163 RETURN(opd.opd_locks > 0);
2166 static void ost_punch_hpreq_fini(struct ptlrpc_request *req)
2168 (void)ost_punch_hpreq_check(req);
2171 struct ptlrpc_hpreq_ops ost_hpreq_rw = {
2172 .hpreq_lock_match = ost_rw_hpreq_lock_match,
2173 .hpreq_check = ost_rw_hpreq_check,
2174 .hpreq_fini = ost_rw_hpreq_fini
2177 struct ptlrpc_hpreq_ops ost_hpreq_punch = {
2178 .hpreq_lock_match = ost_punch_hpreq_lock_match,
2179 .hpreq_check = ost_punch_hpreq_check,
2180 .hpreq_fini = ost_punch_hpreq_fini
2183 /** Assign high priority operations to the request if needed. */
2184 static int ost_io_hpreq_handler(struct ptlrpc_request *req)
2187 if (req->rq_export) {
2188 int opc = lustre_msg_get_opc(req->rq_reqmsg);
2189 struct ost_body *body;
2191 if (opc == OST_READ || opc == OST_WRITE) {
2192 struct niobuf_remote *nb;
2193 struct obd_ioobj *ioo;
2194 int objcount, niocount;
2198 /* RPCs on the H-P queue can be inspected before
2199 * ost_handler() initializes their pills, so we
2200 * initialize that here. Capsule initialization is
2201 * idempotent, as is setting the pill's format (provided
2202 * it doesn't change).
2204 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2205 if (opc == OST_READ)
2206 req_capsule_set(&req->rq_pill,
2209 req_capsule_set(&req->rq_pill,
2210 &RQF_OST_BRW_WRITE);
2212 body = req_capsule_client_get(&req->rq_pill,
2215 CERROR("Missing/short ost_body\n");
2219 objcount = req_capsule_get_size(&req->rq_pill,
2223 if (objcount == 0) {
2224 CERROR("Missing/short ioobj\n");
2228 CERROR("too many ioobjs (%d)\n", objcount);
2232 ioo = req_capsule_client_get(&req->rq_pill,
2235 CERROR("Missing/short ioobj\n");
2239 rc = ost_validate_obdo(req->rq_export, &body->oa, ioo);
2241 CERROR("invalid object ids\n");
2245 for (niocount = i = 0; i < objcount; i++) {
2246 if (ioo[i].ioo_bufcnt == 0) {
2247 CERROR("ioo[%d] has zero bufcnt\n", i);
2250 niocount += ioo[i].ioo_bufcnt;
2252 if (niocount > PTLRPC_MAX_BRW_PAGES) {
2253 DEBUG_REQ(D_RPCTRACE, req,
2254 "bulk has too many pages (%d)",
2259 nb = req_capsule_client_get(&req->rq_pill,
2260 &RMF_NIOBUF_REMOTE);
2262 CERROR("Missing/short niobuf\n");
2266 if (niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
2267 req->rq_ops = &ost_hpreq_rw;
2268 } else if (opc == OST_PUNCH) {
2269 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2270 req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
2272 body = req_capsule_client_get(&req->rq_pill,
2275 CERROR("Missing/short ost_body\n");
2279 if (!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
2280 !(body->oa.o_flags & OBD_FL_SRVLOCK))
2281 req->rq_ops = &ost_hpreq_punch;
2287 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
2288 int ost_handle(struct ptlrpc_request *req)
2290 struct obd_trans_info trans_info = { 0, };
2291 struct obd_trans_info *oti = &trans_info;
2292 int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
2293 struct obd_device *obd = NULL;
2294 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
2297 /* OST module is kept between remounts, but the last reference
2298 * to specific module (say, osd or ofd) kills all related keys
2299 * from the environment. so we have to refill it until the root
2300 * cause is fixed properly */
2301 lu_env_refill(req->rq_svc_thread->t_env);
2303 LASSERT(current->journal_info == NULL);
2305 /* primordial rpcs don't affect server recovery */
2308 case SEC_CTX_INIT_CONT:
2313 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2315 if (opc != OST_CONNECT) {
2316 if (!class_connected_export(req->rq_export)) {
2317 CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
2318 opc, libcfs_id2str(req->rq_peer));
2319 req->rq_status = -ENOTCONN;
2320 GOTO(out, rc = -ENOTCONN);
2323 obd = req->rq_export->exp_obd;
2325 /* Check for aborted recovery. */
2326 if (obd->obd_recovering) {
2327 rc = ost_filter_recovery_request(req, obd,
2329 if (rc || !should_process)
2331 else if (should_process < 0) {
2332 req->rq_status = should_process;
2333 rc = ptlrpc_error(req);
2341 rc = ost_msg_check_version(req->rq_reqmsg);
2345 if (req && req->rq_reqmsg && req->rq_export &&
2346 (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS))
2347 oti->oti_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
2351 CDEBUG(D_INODE, "connect\n");
2352 req_capsule_set(&req->rq_pill, &RQF_OST_CONNECT);
2353 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET))
2355 rc = target_handle_connect(req);
2356 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2))
2359 rc = ost_init_sec_level(req);
2361 rc = ost_connect_check_sptlrpc(req);
2365 case OST_DISCONNECT:
2366 CDEBUG(D_INODE, "disconnect\n");
2367 req_capsule_set(&req->rq_pill, &RQF_OST_DISCONNECT);
2368 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DISCONNECT_NET))
2370 rc = target_handle_disconnect(req);
2373 CDEBUG(D_INODE, "create\n");
2374 req_capsule_set(&req->rq_pill, &RQF_OST_CREATE);
2375 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET))
2377 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2378 GOTO(out, rc = -EROFS);
2379 rc = ost_create(req->rq_export, req, oti);
2382 CDEBUG(D_INODE, "destroy\n");
2383 req_capsule_set(&req->rq_pill, &RQF_OST_DESTROY);
2384 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DESTROY_NET))
2386 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2387 GOTO(out, rc = -EROFS);
2388 rc = ost_destroy(req->rq_export, req, oti);
2391 CDEBUG(D_INODE, "getattr\n");
2392 req_capsule_set(&req->rq_pill, &RQF_OST_GETATTR);
2393 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GETATTR_NET))
2395 rc = ost_getattr(req->rq_export, req);
2398 CDEBUG(D_INODE, "setattr\n");
2399 req_capsule_set(&req->rq_pill, &RQF_OST_SETATTR);
2400 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_NET))
2402 rc = ost_setattr(req->rq_export, req, oti);
2405 req_capsule_set(&req->rq_pill, &RQF_OST_BRW_WRITE);
2406 CDEBUG(D_INODE, "write\n");
2407 /* req->rq_request_portal would be nice, if it was set */
2408 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
2409 CERROR("%s: deny write request from %s to portal %u\n",
2410 req->rq_export->exp_obd->obd_name,
2411 obd_export_nid2str(req->rq_export),
2412 ptlrpc_req2svc(req)->srv_req_portal);
2413 GOTO(out, rc = -EPROTO);
2415 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2417 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2418 GOTO(out, rc = -ENOSPC);
2419 if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
2420 GOTO(out, rc = -EROFS);
2421 rc = ost_brw_write(req, oti);
2422 LASSERT(current->journal_info == NULL);
2423 /* ost_brw_write sends its own replies */
2426 req_capsule_set(&req->rq_pill, &RQF_OST_BRW_READ);
2427 CDEBUG(D_INODE, "read\n");
2428 /* req->rq_request_portal would be nice, if it was set */
2429 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
2430 CERROR("%s: deny read request from %s to portal %u\n",
2431 req->rq_export->exp_obd->obd_name,
2432 obd_export_nid2str(req->rq_export),
2433 ptlrpc_req2svc(req)->srv_req_portal);
2434 GOTO(out, rc = -EPROTO);
2436 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2438 rc = ost_brw_read(req, oti);
2439 LASSERT(current->journal_info == NULL);
2440 /* ost_brw_read sends its own replies */
2443 CDEBUG(D_INODE, "punch\n");
2444 req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
2445 if (OBD_FAIL_CHECK(OBD_FAIL_OST_PUNCH_NET))
2447 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2448 GOTO(out, rc = -EROFS);
2449 rc = ost_punch(req->rq_export, req, oti);
2452 CDEBUG(D_INODE, "statfs\n");
2453 req_capsule_set(&req->rq_pill, &RQF_OST_STATFS);
2454 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_NET))
2456 rc = ost_statfs(req);
2459 CDEBUG(D_INODE, "sync\n");
2460 req_capsule_set(&req->rq_pill, &RQF_OST_SYNC);
2461 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SYNC_NET))
2463 rc = ost_sync(req->rq_export, req, oti);
2466 DEBUG_REQ(D_INODE, req, "set_info");
2467 req_capsule_set(&req->rq_pill, &RQF_OBD_SET_INFO);
2468 rc = ost_set_info(req->rq_export, req);
2471 DEBUG_REQ(D_INODE, req, "get_info");
2472 req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC);
2473 rc = ost_get_info(req->rq_export, req);
2475 case OST_QUOTACHECK:
2476 CDEBUG(D_INODE, "quotacheck\n");
2477 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACHECK);
2478 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACHECK_NET))
2480 rc = ost_handle_quotacheck(req);
2483 CDEBUG(D_INODE, "quotactl\n");
2484 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACTL);
2485 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACTL_NET))
2487 rc = ost_handle_quotactl(req);
2490 DEBUG_REQ(D_INODE, req, "ping");
2491 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
2492 rc = target_handle_ping(req);
2494 /* FIXME - just reply status */
2495 case LLOG_ORIGIN_CONNECT:
2496 DEBUG_REQ(D_INODE, req, "log connect");
2497 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_CONNECT);
2498 rc = ost_llog_handle_connect(req->rq_export, req);
2499 req->rq_status = rc;
2500 rc = req_capsule_server_pack(&req->rq_pill);
2503 RETURN(ptlrpc_reply(req));
2505 CDEBUG(D_INODE, "enqueue\n");
2506 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
2507 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE_NET))
2509 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
2511 ldlm_server_glimpse_ast);
2512 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
2515 CDEBUG(D_INODE, "convert\n");
2516 req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT);
2517 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT_NET))
2519 rc = ldlm_handle_convert(req);
2522 CDEBUG(D_INODE, "cancel\n");
2523 req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL);
2524 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_NET))
2526 rc = ldlm_handle_cancel(req);
2528 case LDLM_BL_CALLBACK:
2529 case LDLM_CP_CALLBACK:
2530 CDEBUG(D_INODE, "callback\n");
2531 CERROR("callbacks should not happen on OST\n");
2534 CERROR("Unexpected opcode %d\n", opc);
2535 req->rq_status = -ENOTSUPP;
2536 rc = ptlrpc_error(req);
2540 LASSERT(current->journal_info == NULL);
2543 /* If we're DISCONNECTing, the export_data is already freed */
2544 if (!rc && opc != OST_DISCONNECT)
2545 target_committed_to_req(req);
2549 oti_to_request(oti, req);
2551 target_send_reply(req, rc, fail);
2554 EXPORT_SYMBOL(ost_handle);
2557 * free per-thread pool created by ost_io_thread_init().
2559 static void ost_io_thread_done(struct ptlrpc_thread *thread)
2561 struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local
2566 LASSERT(thread != NULL);
2569 * be prepared to handle partially-initialized pools (because this is
2570 * called from ost_io_thread_init() for cleanup.
2572 tls = thread->t_data;
2575 thread->t_data = NULL;
2581 * initialize per-thread page pool (bug 5137).
2583 static int ost_io_thread_init(struct ptlrpc_thread *thread)
2585 struct ost_thread_local_cache *tls;
2589 LASSERT(thread != NULL);
2590 LASSERT(thread->t_data == NULL);
2595 thread->t_data = tls;
2599 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
2601 static struct cfs_cpt_table *ost_io_cptable;
2603 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
2604 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
2606 static struct ptlrpc_service_conf svc_conf;
2607 struct ost_obd *ost = &obd->u.ost;
2608 struct lprocfs_static_vars lvars;
2613 rc = cfs_cleanup_group_info();
2617 lprocfs_ost_init_vars(&lvars);
2618 lprocfs_obd_setup(obd, lvars.obd_vars);
2620 mutex_init(&ost->ost_health_mutex);
2622 svc_conf = (typeof(svc_conf)) {
2623 .psc_name = LUSTRE_OSS_NAME,
2624 .psc_watchdog_factor = OSS_SERVICE_WATCHDOG_FACTOR,
2626 .bc_nbufs = OST_NBUFS,
2627 .bc_buf_size = OST_BUFSIZE,
2628 .bc_req_max_size = OST_MAXREQSIZE,
2629 .bc_rep_max_size = OST_MAXREPSIZE,
2630 .bc_req_portal = OST_REQUEST_PORTAL,
2631 .bc_rep_portal = OSC_REPLY_PORTAL,
2634 .tc_thr_name = "ll_ost",
2635 .tc_thr_factor = OSS_THR_FACTOR,
2636 .tc_nthrs_init = OSS_NTHRS_INIT,
2637 .tc_nthrs_base = OSS_NTHRS_BASE,
2638 .tc_nthrs_max = OSS_NTHRS_MAX,
2639 .tc_nthrs_user = oss_num_threads,
2640 .tc_cpu_affinity = 1,
2641 .tc_ctx_tags = LCT_DT_THREAD,
2644 .cc_pattern = oss_cpts,
2647 .so_req_handler = ost_handle,
2648 .so_req_printer = target_print_req,
2649 .so_hpreq_handler = ptlrpc_hpreq_handler,
2652 ost->ost_service = ptlrpc_register_service(&svc_conf,
2653 obd->obd_proc_entry);
2654 if (IS_ERR(ost->ost_service)) {
2655 rc = PTR_ERR(ost->ost_service);
2656 CERROR("failed to start service: %d\n", rc);
2657 GOTO(out_lprocfs, rc);
2660 memset(&svc_conf, 0, sizeof(svc_conf));
2661 svc_conf = (typeof(svc_conf)) {
2662 .psc_name = "ost_create",
2663 .psc_watchdog_factor = OSS_SERVICE_WATCHDOG_FACTOR,
2665 .bc_nbufs = OST_NBUFS,
2666 .bc_buf_size = OST_BUFSIZE,
2667 .bc_req_max_size = OST_MAXREQSIZE,
2668 .bc_rep_max_size = OST_MAXREPSIZE,
2669 .bc_req_portal = OST_CREATE_PORTAL,
2670 .bc_rep_portal = OSC_REPLY_PORTAL,
2673 .tc_thr_name = "ll_ost_create",
2674 .tc_thr_factor = OSS_CR_THR_FACTOR,
2675 .tc_nthrs_init = OSS_CR_NTHRS_INIT,
2676 .tc_nthrs_base = OSS_CR_NTHRS_BASE,
2677 .tc_nthrs_max = OSS_CR_NTHRS_MAX,
2678 .tc_nthrs_user = oss_num_create_threads,
2679 .tc_cpu_affinity = 1,
2680 .tc_ctx_tags = LCT_DT_THREAD,
2683 .cc_pattern = oss_cpts,
2686 .so_req_handler = ost_handle,
2687 .so_req_printer = target_print_req,
2690 ost->ost_create_service = ptlrpc_register_service(&svc_conf,
2691 obd->obd_proc_entry);
2692 if (IS_ERR(ost->ost_create_service)) {
2693 rc = PTR_ERR(ost->ost_create_service);
2694 CERROR("failed to start OST create service: %d\n", rc);
2695 GOTO(out_service, rc);
2698 mask = cfs_cpt_table->ctb_nodemask;
2699 /* event CPT feature is disabled in libcfs level by set partition
2700 * number to 1, we still want to set node affinity for io service */
2701 if (cfs_cpt_number(cfs_cpt_table) == 1 && nodes_weight(*mask) > 1) {
2705 ost_io_cptable = cfs_cpt_table_alloc(nodes_weight(*mask));
2706 for_each_node_mask(i, *mask) {
2707 if (ost_io_cptable == NULL) {
2708 CWARN("OSS failed to create CPT table\n");
2712 rc = cfs_cpt_set_node(ost_io_cptable, cpt++, i);
2714 CWARN("OSS Failed to set node %d for"
2715 "IO CPT table\n", i);
2716 cfs_cpt_table_free(ost_io_cptable);
2717 ost_io_cptable = NULL;
2723 memset(&svc_conf, 0, sizeof(svc_conf));
2724 svc_conf = (typeof(svc_conf)) {
2725 .psc_name = "ost_io",
2726 .psc_watchdog_factor = OSS_SERVICE_WATCHDOG_FACTOR,
2728 .bc_nbufs = OST_NBUFS,
2729 .bc_buf_size = OST_IO_BUFSIZE,
2730 .bc_req_max_size = OST_IO_MAXREQSIZE,
2731 .bc_rep_max_size = OST_IO_MAXREPSIZE,
2732 .bc_req_portal = OST_IO_PORTAL,
2733 .bc_rep_portal = OSC_REPLY_PORTAL,
2736 .tc_thr_name = "ll_ost_io",
2737 .tc_thr_factor = OSS_THR_FACTOR,
2738 .tc_nthrs_init = OSS_NTHRS_INIT,
2739 .tc_nthrs_base = OSS_NTHRS_BASE,
2740 .tc_nthrs_max = OSS_NTHRS_MAX,
2741 .tc_nthrs_user = oss_num_threads,
2742 .tc_cpu_affinity = 1,
2743 .tc_ctx_tags = LCT_DT_THREAD,
2746 .cc_cptable = ost_io_cptable,
2747 .cc_pattern = ost_io_cptable == NULL ?
2751 .so_thr_init = ost_io_thread_init,
2752 .so_thr_done = ost_io_thread_done,
2753 .so_req_handler = ost_handle,
2754 .so_hpreq_handler = ost_io_hpreq_handler,
2755 .so_req_printer = target_print_req,
2758 ost->ost_io_service = ptlrpc_register_service(&svc_conf,
2759 obd->obd_proc_entry);
2760 if (IS_ERR(ost->ost_io_service)) {
2761 rc = PTR_ERR(ost->ost_io_service);
2762 CERROR("failed to start OST I/O service: %d\n", rc);
2763 ost->ost_io_service = NULL;
2764 GOTO(out_create, rc);
2767 memset(&svc_conf, 0, sizeof(svc_conf));
2768 svc_conf = (typeof(svc_conf)) {
2769 .psc_name = "ost_seq",
2770 .psc_watchdog_factor = OSS_SERVICE_WATCHDOG_FACTOR,
2772 .bc_nbufs = OST_NBUFS,
2773 .bc_buf_size = OST_BUFSIZE,
2774 .bc_req_max_size = OST_MAXREQSIZE,
2775 .bc_rep_max_size = OST_MAXREPSIZE,
2776 .bc_req_portal = SEQ_DATA_PORTAL,
2777 .bc_rep_portal = OSC_REPLY_PORTAL,
2780 .tc_thr_name = "ll_ost_seq",
2781 .tc_thr_factor = OSS_CR_THR_FACTOR,
2782 .tc_nthrs_init = OSS_CR_NTHRS_INIT,
2783 .tc_nthrs_base = OSS_CR_NTHRS_BASE,
2784 .tc_nthrs_max = OSS_CR_NTHRS_MAX,
2785 .tc_nthrs_user = oss_num_create_threads,
2786 .tc_cpu_affinity = 1,
2787 .tc_ctx_tags = LCT_DT_THREAD,
2791 .cc_pattern = oss_cpts,
2794 .so_req_handler = tgt_request_handle,
2795 .so_req_printer = target_print_req,
2796 .so_hpreq_handler = NULL,
2799 ost->ost_seq_service = ptlrpc_register_service(&svc_conf,
2800 obd->obd_proc_entry);
2801 if (IS_ERR(ost->ost_seq_service)) {
2802 rc = PTR_ERR(ost->ost_seq_service);
2803 CERROR("failed to start OST seq service: %d\n", rc);
2804 ost->ost_seq_service = NULL;
2809 /* Object update service */
2810 memset(&svc_conf, 0, sizeof(svc_conf));
2811 svc_conf = (typeof(svc_conf)) {
2812 .psc_name = "ost_out",
2813 .psc_watchdog_factor = OSS_SERVICE_WATCHDOG_FACTOR,
2815 .bc_nbufs = OST_NBUFS,
2816 .bc_buf_size = OUT_BUFSIZE,
2817 .bc_req_max_size = OUT_MAXREQSIZE,
2818 .bc_rep_max_size = OUT_MAXREPSIZE,
2819 .bc_req_portal = OUT_PORTAL,
2820 .bc_rep_portal = OSC_REPLY_PORTAL,
2823 * We'd like to have a mechanism to set this on a per-device
2824 * basis, but alas...
2827 .tc_thr_name = "ll_ost_out",
2828 .tc_thr_factor = OSS_CR_THR_FACTOR,
2829 .tc_nthrs_init = OSS_CR_NTHRS_INIT,
2830 .tc_nthrs_base = OSS_CR_NTHRS_BASE,
2831 .tc_nthrs_max = OSS_CR_NTHRS_MAX,
2832 .tc_nthrs_user = oss_num_create_threads,
2833 .tc_cpu_affinity = 1,
2834 .tc_ctx_tags = LCT_DT_THREAD,
2837 .cc_pattern = oss_cpts,
2840 .so_req_handler = tgt_request_handle,
2841 .so_req_printer = target_print_req,
2842 .so_hpreq_handler = NULL,
2845 ost->ost_out_service = ptlrpc_register_service(&svc_conf,
2846 obd->obd_proc_entry);
2847 if (IS_ERR(ost->ost_out_service)) {
2848 rc = PTR_ERR(ost->ost_out_service);
2849 CERROR("failed to start out service: %d\n", rc);
2850 ost->ost_out_service = NULL;
2854 ping_evictor_start();
2858 ptlrpc_unregister_service(ost->ost_io_service);
2859 ost->ost_io_service = NULL;
2861 ptlrpc_unregister_service(ost->ost_create_service);
2862 ost->ost_create_service = NULL;
2864 ptlrpc_unregister_service(ost->ost_service);
2865 ost->ost_service = NULL;
2867 lprocfs_obd_cleanup(obd);
2871 static int ost_cleanup(struct obd_device *obd)
2873 struct ost_obd *ost = &obd->u.ost;
2877 ping_evictor_stop();
2879 /* there is no recovery for OST OBD, all recovery is controlled by
2881 LASSERT(obd->obd_recovering == 0);
2882 mutex_lock(&ost->ost_health_mutex);
2883 ptlrpc_unregister_service(ost->ost_service);
2884 ptlrpc_unregister_service(ost->ost_create_service);
2885 ptlrpc_unregister_service(ost->ost_io_service);
2886 ptlrpc_unregister_service(ost->ost_seq_service);
2888 ptlrpc_unregister_service(ost->ost_out_service);
2890 ost->ost_service = NULL;
2891 ost->ost_create_service = NULL;
2892 ost->ost_io_service = NULL;
2893 ost->ost_seq_service = NULL;
2894 ost->ost_out_service = NULL;
2896 mutex_unlock(&ost->ost_health_mutex);
2898 lprocfs_obd_cleanup(obd);
2900 if (ost_io_cptable != NULL) {
2901 cfs_cpt_table_free(ost_io_cptable);
2902 ost_io_cptable = NULL;
2908 static int ost_health_check(const struct lu_env *env, struct obd_device *obd)
2910 struct ost_obd *ost = &obd->u.ost;
2913 mutex_lock(&ost->ost_health_mutex);
2914 rc |= ptlrpc_service_health_check(ost->ost_service);
2915 rc |= ptlrpc_service_health_check(ost->ost_create_service);
2916 rc |= ptlrpc_service_health_check(ost->ost_io_service);
2917 mutex_unlock(&ost->ost_health_mutex);
2920 * health_check to return 0 on healthy
2921 * and 1 on unhealthy.
2929 struct ost_thread_local_cache *ost_tls(struct ptlrpc_request *r)
2931 return (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
2934 /* use obd ops to offer management infrastructure */
2935 static struct obd_ops ost_obd_ops = {
2936 .o_owner = THIS_MODULE,
2937 .o_setup = ost_setup,
2938 .o_cleanup = ost_cleanup,
2939 .o_health_check = ost_health_check,
2943 static int __init ost_init(void)
2945 struct lprocfs_static_vars lvars;
2949 ost_page_to_corrupt = alloc_page(GFP_IOFS);
2951 lprocfs_ost_init_vars(&lvars);
2952 rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars,
2953 LUSTRE_OSS_NAME, NULL);
2955 if (ost_num_threads != 0 && oss_num_threads == 0) {
2956 LCONSOLE_INFO("ost_num_threads module parameter is deprecated, "
2957 "use oss_num_threads instead or unset both for "
2958 "dynamic thread startup\n");
2959 oss_num_threads = ost_num_threads;
2965 static void /*__exit*/ ost_exit(void)
2967 if (ost_page_to_corrupt)
2968 page_cache_release(ost_page_to_corrupt);
2970 class_unregister_type(LUSTRE_OSS_NAME);
2973 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2974 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
2975 MODULE_LICENSE("GPL");
2977 module_init(ost_init);
2978 module_exit(ost_exit);