1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
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>
43 # define EXPORT_SYMTAB
45 #define DEBUG_SUBSYSTEM S_OST
47 #include <linux/module.h>
48 #include <obd_cksum.h>
50 #include <lustre_net.h>
51 #include <lustre_dlm.h>
52 #include <lustre_export.h>
53 #include <lustre_debug.h>
54 #include <linux/init.h>
55 #include <lprocfs_status.h>
56 #include <libcfs/list.h>
57 #include <lustre_quota.h>
58 #include "ost_internal.h"
60 static int oss_num_threads;
61 CFS_MODULE_PARM(oss_num_threads, "i", int, 0444,
62 "number of OSS service threads to start");
64 static int ost_num_threads;
65 CFS_MODULE_PARM(ost_num_threads, "i", int, 0444,
66 "number of OST service threads to start (deprecated)");
68 static int oss_num_create_threads;
69 CFS_MODULE_PARM(oss_num_create_threads, "i", int, 0444,
70 "number of OSS create threads to start");
73 * Do not return server-side uid/gid to remote client
75 static void ost_drop_id(struct obd_export *exp, struct obdo *oa)
77 if (exp_connect_rmtclient(exp)) {
80 oa->o_valid &= ~(OBD_MD_FLUID | OBD_MD_FLGID);
84 void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req)
86 struct oti_req_ack_lock *ack_lock;
93 lustre_msg_set_transno(req->rq_repmsg, oti->oti_transno);
94 req->rq_transno = oti->oti_transno;
96 /* XXX 4 == entries in oti_ack_locks??? */
97 for (ack_lock = oti->oti_ack_locks, i = 0; i < 4; i++, ack_lock++) {
100 /* XXX not even calling target_send_reply in some cases... */
101 ptlrpc_save_lock (req, &ack_lock->lock, ack_lock->mode, 0);
105 static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req,
106 struct obd_trans_info *oti)
108 struct ost_body *body, *repbody;
109 __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
110 struct lustre_capa *capa = NULL;
114 body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
115 lustre_swab_ost_body);
119 if (body->oa.o_id == 0)
122 if (lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1)) {
123 struct ldlm_request *dlm;
124 dlm = lustre_swab_reqbuf(req, REQ_REC_OFF + 1, sizeof(*dlm),
125 lustre_swab_ldlm_request);
128 ldlm_request_cancel(req, dlm, 0);
131 if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
132 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 2);
134 rc = lustre_pack_reply(req, 2, size, NULL);
138 if (body->oa.o_valid & OBD_MD_FLCOOKIE)
139 oti->oti_logcookies = &body->oa.o_lcookie;
140 repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
142 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
143 req->rq_status = obd_destroy(exp, &body->oa, NULL, oti, NULL, capa);
147 static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
149 struct ost_body *body, *repbody;
150 struct obd_info oinfo = { { { 0 } } };
151 __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
155 body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
156 lustre_swab_ost_body);
160 rc = lustre_pack_reply(req, 2, size, NULL);
164 repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
166 repbody->oa = body->oa;
168 oinfo.oi_oa = &repbody->oa;
169 if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
170 oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg,
172 req->rq_status = obd_getattr(exp, &oinfo);
173 ost_drop_id(exp, &repbody->oa);
177 static int ost_statfs(struct ptlrpc_request *req)
179 struct obd_statfs *osfs;
180 __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*osfs) };
184 rc = lustre_pack_reply(req, 2, size, NULL);
188 osfs = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*osfs));
190 req->rq_status = obd_statfs(req->rq_export->exp_obd, osfs,
191 cfs_time_current_64() - HZ, 0);
192 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
193 osfs->os_bfree = osfs->os_bavail = 64;
194 if (req->rq_status != 0)
195 CERROR("ost: statfs failed: rc %d\n", req->rq_status);
200 static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
201 struct obd_trans_info *oti)
203 struct ost_body *body, *repbody;
204 __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
208 body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
209 lustre_swab_ost_body);
213 rc = lustre_pack_reply(req, 2, size, NULL);
217 repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
219 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
220 oti->oti_logcookies = &repbody->oa.o_lcookie;
221 req->rq_status = obd_create(exp, &repbody->oa, NULL, oti);
222 //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
227 * Helper function for ost_punch(): if asked by client, acquire [size, EOF]
228 * lock on the file being truncated.
230 static int ost_punch_lock_get(struct obd_export *exp, struct obdo *oa,
231 struct lustre_handle *lh)
234 struct ldlm_res_id res_id;
235 ldlm_policy_data_t policy;
241 osc_build_res_name(oa->o_id, oa->o_gr, &res_id);
242 LASSERT(!lustre_handle_is_used(lh));
244 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
245 !(oa->o_flags & OBD_FL_TRUNCLOCK))
248 CDEBUG(D_INODE, "OST-side truncate lock.\n");
251 finis = start + oa->o_blocks;
254 * standard truncate optimization: if file body is completely
255 * destroyed, don't send data back to the server.
257 flags = (start == 0) ? LDLM_AST_DISCARD_DATA : 0;
259 policy.l_extent.start = start & CFS_PAGE_MASK;
262 * If ->o_blocks is EOF it means "lock till the end of the
263 * file". Otherwise, it's size of a hole being punched (in bytes)
265 if (oa->o_blocks == OBD_OBJECT_EOF || finis < start)
266 policy.l_extent.end = OBD_OBJECT_EOF;
268 policy.l_extent.end = finis | ~CFS_PAGE_MASK;
270 RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
271 LDLM_EXTENT, &policy, LCK_PW, &flags,
272 ldlm_blocking_ast, ldlm_completion_ast,
273 ldlm_glimpse_ast, NULL, 0, NULL,
278 * Helper function for ost_punch(): release lock acquired by
279 * ost_punch_lock_get(), if any.
281 static void ost_punch_lock_put(struct obd_export *exp, struct obdo *oa,
282 struct lustre_handle *lh)
285 if (lustre_handle_is_used(lh))
286 ldlm_lock_decref(lh, LCK_PW);
290 static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
291 struct obd_trans_info *oti)
293 struct obd_info oinfo = { { { 0 } } };
294 struct ost_body *body, *repbody;
295 __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
297 struct lustre_handle lh = {0,};
300 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
301 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
303 /* ost_body is varified and swabbed in ost_hpreq_handler() */
304 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
305 LASSERT(body != NULL);
307 oinfo.oi_oa = &body->oa;
308 oinfo.oi_policy.l_extent.start = oinfo.oi_oa->o_size;
309 oinfo.oi_policy.l_extent.end = oinfo.oi_oa->o_blocks;
311 if ((oinfo.oi_oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
312 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
315 rc = lustre_pack_reply(req, 2, size, NULL);
319 repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
321 rc = ost_punch_lock_get(exp, oinfo.oi_oa, &lh);
323 if (oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS &&
324 oinfo.oi_oa->o_flags == OBD_FL_TRUNCLOCK)
326 * If OBD_FL_TRUNCLOCK is the only bit set in
327 * ->o_flags, clear OBD_MD_FLFLAGS to avoid falling
328 * through filter_setattr() to filter_iocontrol().
330 oinfo.oi_oa->o_valid &= ~OBD_MD_FLFLAGS;
332 if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
333 oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg,
335 req->rq_status = obd_punch(exp, &oinfo, oti, NULL);
336 ost_punch_lock_put(exp, oinfo.oi_oa, &lh);
338 repbody->oa = *oinfo.oi_oa;
339 ost_drop_id(exp, &repbody->oa);
343 static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req)
345 struct ost_body *body, *repbody;
346 struct lustre_capa *capa = NULL;
347 __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
351 body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
352 lustre_swab_ost_body);
356 if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
357 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 1);
359 rc = lustre_pack_reply(req, 2, size, NULL);
363 repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
365 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
366 req->rq_status = obd_sync(exp, &repbody->oa, NULL, repbody->oa.o_size,
367 repbody->oa.o_blocks, capa);
368 ost_drop_id(exp, &repbody->oa);
372 static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req,
373 struct obd_trans_info *oti)
375 struct ost_body *body, *repbody;
376 __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) };
378 struct obd_info oinfo = { { { 0 } } };
381 body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
382 lustre_swab_ost_body);
386 rc = lustre_pack_reply(req, 2, size, NULL);
390 repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
392 repbody->oa = body->oa;
394 oinfo.oi_oa = &repbody->oa;
395 if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA)
396 oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg,
398 req->rq_status = obd_setattr(exp, &oinfo, oti);
399 ost_drop_id(exp, &repbody->oa);
403 static int ost_bulk_timeout(void *data)
406 /* We don't fail the connection here, because having the export
407 * killed makes the (vital) call to commitrw very sad.
412 static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc,
413 cksum_type_t cksum_type)
418 cksum = init_checksum(cksum_type);
419 for (i = 0; i < desc->bd_iov_count; i++) {
420 struct page *page = desc->bd_iov[i].kiov_page;
421 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
422 char *ptr = kmap(page) + off;
423 int len = desc->bd_iov[i].kiov_len;
425 /* corrupt the data before we compute the checksum, to
426 * simulate a client->OST data error */
427 if (i == 0 && opc == OST_WRITE &&
428 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE))
429 memcpy(ptr, "bad3", min(4, len));
430 cksum = compute_checksum(cksum, ptr, len, cksum_type);
431 /* corrupt the data after we compute the checksum, to
432 * simulate an OST->client data error */
433 if (i == 0 && opc == OST_READ &&
434 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
435 memcpy(ptr, "bad4", min(4, len));
436 /* nobody should use corrupted page again */
437 ClearPageUptodate(page);
445 static int ost_brw_lock_get(int mode, struct obd_export *exp,
446 struct obd_ioobj *obj, struct niobuf_remote *nb,
447 struct lustre_handle *lh)
450 int nrbufs = obj->ioo_bufcnt;
451 struct ldlm_res_id res_id;
452 ldlm_policy_data_t policy;
456 osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id);
457 LASSERT(mode == LCK_PR || mode == LCK_PW);
458 LASSERT(!lustre_handle_is_used(lh));
460 if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
463 /* EXPENSIVE ASSERTION */
464 for (i = 1; i < nrbufs; i ++)
465 LASSERT((nb[0].flags & OBD_BRW_SRVLOCK) ==
466 (nb[i].flags & OBD_BRW_SRVLOCK));
468 policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
469 policy.l_extent.end = (nb[nrbufs - 1].offset +
470 nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
472 RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
473 LDLM_EXTENT, &policy, mode, &flags,
474 ldlm_blocking_ast, ldlm_completion_ast,
475 ldlm_glimpse_ast, NULL, 0, NULL,
479 static void ost_brw_lock_put(int mode,
480 struct obd_ioobj *obj, struct niobuf_remote *niob,
481 struct lustre_handle *lh)
484 LASSERT(mode == LCK_PR || mode == LCK_PW);
485 LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
486 lustre_handle_is_used(lh));
487 if (lustre_handle_is_used(lh))
488 ldlm_lock_decref(lh, mode);
492 struct ost_prolong_data {
493 struct obd_export *opd_exp;
494 ldlm_policy_data_t opd_policy;
496 ldlm_mode_t opd_mode;
501 static int ost_prolong_locks_iter(struct ldlm_lock *lock, void *data)
503 struct ost_prolong_data *opd = data;
505 LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
507 if (lock->l_req_mode != lock->l_granted_mode) {
508 /* scan granted locks only */
509 return LDLM_ITER_STOP;
512 if (lock->l_export != opd->opd_exp) {
513 /* prolong locks only for given client */
514 return LDLM_ITER_CONTINUE;
517 if (!(lock->l_granted_mode & opd->opd_mode)) {
518 /* we aren't interesting in all type of locks */
519 return LDLM_ITER_CONTINUE;
522 if (lock->l_policy_data.l_extent.end < opd->opd_policy.l_extent.start ||
523 lock->l_policy_data.l_extent.start > opd->opd_policy.l_extent.end) {
524 /* the request doesn't cross the lock, skip it */
525 return LDLM_ITER_CONTINUE;
528 /* Fill the obdo with the matched lock handle.
529 * XXX: it is possible in some cases the IO RPC is covered by several
530 * locks, even for the write case, so it may need to be a lock list. */
531 if (opd->opd_oa && !(opd->opd_oa->o_valid & OBD_MD_FLHANDLE)) {
532 opd->opd_oa->o_handle.cookie = lock->l_handle.h_cookie;
533 opd->opd_oa->o_valid |= OBD_MD_FLHANDLE;
536 if (!(lock->l_flags & LDLM_FL_AST_SENT)) {
537 /* ignore locks not being cancelled */
538 return LDLM_ITER_CONTINUE;
541 CDEBUG(D_DLMTRACE,"refresh lock: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
542 lock->l_resource->lr_name.name[0],
543 lock->l_resource->lr_name.name[1],
544 opd->opd_policy.l_extent.start, opd->opd_policy.l_extent.end);
545 /* OK. this is a possible lock the user holds doing I/O
546 * let's refresh eviction timer for it */
547 ldlm_refresh_waiting_lock(lock, opd->opd_timeout);
548 opd->opd_lock_match = 1;
550 return LDLM_ITER_CONTINUE;
553 static int ost_rw_prolong_locks(struct ptlrpc_request *req, struct obd_ioobj *obj,
554 struct niobuf_remote *nb, struct obdo *oa,
557 struct ldlm_res_id res_id;
558 int nrbufs = obj->ioo_bufcnt;
559 struct ost_prolong_data opd = { 0 };
562 osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id);
565 opd.opd_exp = req->rq_export;
566 opd.opd_policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
567 opd.opd_policy.l_extent.end = (nb[nrbufs - 1].offset +
568 nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
570 /* prolong locks for the current service time of the corresponding
571 * portal (= OST_IO_PORTAL) */
572 opd.opd_timeout = AT_OFF ? obd_timeout / 2:
573 max(at_est2timeout(at_get(&req->rq_rqbd->
574 rqbd_service->srv_at_estimate)), ldlm_timeout);
576 CDEBUG(D_INFO,"refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
577 res_id.name[0], res_id.name[1], opd.opd_policy.l_extent.start,
578 opd.opd_policy.l_extent.end);
580 if (oa->o_valid & OBD_MD_FLHANDLE) {
581 struct ldlm_lock *lock;
583 lock = ldlm_handle2lock(&oa->o_handle);
585 ost_prolong_locks_iter(lock, &opd);
586 if (opd.opd_lock_match) {
591 /* Check if the lock covers the whole IO region,
592 * otherwise iterate through the resource. */
593 if (lock->l_policy_data.l_extent.end >=
594 opd.opd_policy.l_extent.end &&
595 lock->l_policy_data.l_extent.start <=
596 opd.opd_policy.l_extent.start) {
605 ldlm_resource_iterate(req->rq_export->exp_obd->obd_namespace, &res_id,
606 ost_prolong_locks_iter, &opd);
607 RETURN(opd.opd_lock_match);
610 static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
612 struct ptlrpc_bulk_desc *desc = NULL;
613 struct obd_export *exp = req->rq_export;
614 struct niobuf_remote *remote_nb;
615 struct niobuf_local *local_nb;
616 struct obd_ioobj *ioo;
617 struct ost_body *body, *repbody;
618 struct lustre_capa *capa = NULL;
619 struct l_wait_info lwi;
620 struct lustre_handle lockh = { 0 };
621 __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
622 int niocount, npages, nob = 0, rc, i;
626 req->rq_bulk_read = 1;
628 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
629 GOTO(out, rc = -EIO);
631 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
633 /* Check if there is eviction in progress, and if so, wait for it to
635 if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
636 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
637 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
638 !atomic_read(&exp->exp_obd->obd_evict_inprogress),
642 GOTO(out, rc = -ENOTCONN);
644 /* ost_body, ioobj & noibuf_remote are verified and swabbed in
645 * ost_rw_hpreq_check(). */
646 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
647 LASSERT(body != NULL);
649 ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1, sizeof(*ioo));
650 LASSERT(ioo != NULL);
652 niocount = ioo->ioo_bufcnt;
653 remote_nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
654 niocount * sizeof(*remote_nb));
655 LASSERT(remote_nb != NULL);
657 if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
658 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 3);
660 rc = lustre_pack_reply(req, 2, size, NULL);
665 * Per-thread array of struct niobuf_{local,remote}'s was allocated by
668 local_nb = ost_tls(req)->local;
670 rc = ost_brw_lock_get(LCK_PR, exp, ioo, remote_nb, &lockh);
675 * If getting the lock took more time than
676 * client was willing to wait, drop it. b=11330
678 if (cfs_time_current_sec() > req->rq_deadline ||
679 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
681 CERROR("Dropping timed-out read from %s because locking"
682 "object "LPX64" took %ld seconds (limit was %ld).\n",
683 libcfs_id2str(req->rq_peer), ioo->ioo_id,
684 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
685 req->rq_deadline - req->rq_arrival_time.tv_sec);
686 GOTO(out_lock, rc = -ETIMEDOUT);
689 npages = OST_THREAD_POOL_SIZE;
690 rc = obd_preprw(OBD_BRW_READ, exp, &body->oa, 1, ioo,
691 remote_nb, &npages, local_nb, oti, capa);
695 desc = ptlrpc_prep_bulk_exp(req, npages,
696 BULK_PUT_SOURCE, OST_BULK_PORTAL);
697 if (desc == NULL) /* XXX: check all cleanup stuff */
698 GOTO(out, rc = -ENOMEM);
700 ost_rw_prolong_locks(req, ioo, remote_nb, &body->oa, LCK_PW | LCK_PR);
703 for (i = 0; i < npages; i++) {
704 int page_rc = local_nb[i].rc;
706 if (page_rc < 0) { /* error */
712 if (page_rc != 0) { /* some data! */
713 LASSERT (local_nb[i].page != NULL);
714 ptlrpc_prep_bulk_page(desc, local_nb[i].page,
715 local_nb[i].offset & ~CFS_PAGE_MASK,
719 if (page_rc != local_nb[i].len) { /* short read */
720 /* All subsequent pages should be 0 */
722 LASSERT(local_nb[i].rc == 0);
727 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
728 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
730 if (body->oa.o_valid & OBD_MD_FLFLAGS)
731 cksum_type = cksum_type_unpack(body->oa.o_flags);
732 body->oa.o_flags = cksum_type_pack(cksum_type);
733 body->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
734 body->oa.o_cksum = ost_checksum_bulk(desc, OST_READ, cksum_type);
735 CDEBUG(D_PAGE,"checksum at read origin: %x\n",body->oa.o_cksum);
737 body->oa.o_valid = 0;
739 /* We're finishing using body->oa as an input variable */
741 /* Check if client was evicted while we were doing i/o before touching
744 /* Check if there is eviction in progress, and if so, wait for
746 if (unlikely(atomic_read(&exp->exp_obd->
747 obd_evict_inprogress))) {
748 lwi = LWI_INTR(NULL, NULL);
749 rc = l_wait_event(exp->exp_obd->
750 obd_evict_inprogress_waitq,
751 !atomic_read(&exp->exp_obd->
752 obd_evict_inprogress),
758 rc = sptlrpc_svc_wrap_bulk(req, desc);
760 rc = ptlrpc_start_bulk_transfer(desc);
764 time_t start = cfs_time_current_sec();
766 long timeoutl = req->rq_deadline -
767 cfs_time_current_sec();
768 cfs_duration_t timeout = (timeoutl <= 0 || rc) ?
769 CFS_TICK : cfs_time_seconds(timeoutl);
770 lwi = LWI_TIMEOUT_INTERVAL(timeout,
774 rc = l_wait_event(desc->bd_waitq,
775 !ptlrpc_server_bulk_active(desc) ||
776 exp->exp_failed, &lwi);
777 LASSERT(rc == 0 || rc == -ETIMEDOUT);
778 /* Wait again if we changed deadline */
779 } while ((rc == -ETIMEDOUT) &&
780 (req->rq_deadline > cfs_time_current_sec()));
782 if (rc == -ETIMEDOUT) {
783 DEBUG_REQ(D_ERROR, req,
784 "timeout on bulk PUT after %ld%+lds",
785 req->rq_deadline - start,
786 cfs_time_current_sec() -
788 ptlrpc_abort_bulk(desc);
789 } else if (exp->exp_failed) {
790 DEBUG_REQ(D_ERROR, req, "Eviction on bulk PUT");
792 ptlrpc_abort_bulk(desc);
793 } else if (!desc->bd_success ||
794 desc->bd_nob_transferred != desc->bd_nob) {
795 DEBUG_REQ(D_ERROR, req, "%s bulk PUT %d(%d)",
797 "truncated" : "network error on",
798 desc->bd_nob_transferred,
800 /* XXX should this be a different errno? */
804 DEBUG_REQ(D_ERROR, req, "bulk PUT failed: rc %d", rc);
809 /* Must commit after prep above in all cases */
810 rc = obd_commitrw(OBD_BRW_READ, exp, &body->oa, 1, ioo,
811 remote_nb, npages, local_nb, oti, rc);
814 repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
816 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
817 ost_drop_id(exp, &repbody->oa);
821 ost_brw_lock_put(LCK_PR, ioo, remote_nb, &lockh);
824 ptlrpc_free_bulk(desc);
828 req->rq_status = nob;
829 ptlrpc_lprocfs_brw(req, nob);
830 target_committed_to_req(req);
832 } else if (!no_reply) {
833 /* Only reply if there was no comms problem with bulk */
834 target_committed_to_req(req);
838 /* reply out callback would free */
839 ptlrpc_req_drop_rs(req);
840 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
841 "client will retry\n",
842 exp->exp_obd->obd_name,
843 exp->exp_client_uuid.uuid,
844 exp->exp_connection->c_remote_uuid.uuid,
845 libcfs_id2str(req->rq_peer));
851 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
853 struct ptlrpc_bulk_desc *desc = NULL;
854 struct obd_export *exp = req->rq_export;
855 struct niobuf_remote *remote_nb;
856 struct niobuf_local *local_nb;
857 struct obd_ioobj *ioo;
858 struct ost_body *body, *repbody;
859 struct l_wait_info lwi;
860 struct lustre_handle lockh = {0};
861 struct lustre_capa *capa = NULL;
863 __u32 size[3] = { sizeof(struct ptlrpc_body), sizeof(*body) };
864 int objcount, niocount, npages;
866 obd_count client_cksum = 0, server_cksum = 0;
867 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
869 __u32 o_uid = 0, o_gid = 0;
872 req->rq_bulk_write = 1;
874 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
875 GOTO(out, rc = -EIO);
876 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
877 GOTO(out, rc = -EFAULT);
879 /* pause before transaction has been started */
880 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
882 /* Check if there is eviction in progress, and if so, wait for it to
884 if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
885 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
886 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
887 !atomic_read(&exp->exp_obd->obd_evict_inprogress),
891 GOTO(out, rc = -ENOTCONN);
893 /* ost_body, ioobj & noibuf_remote are verified and swabbed in
894 * ost_rw_hpreq_check(). */
895 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
896 LASSERT(body != NULL);
898 objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
900 ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,
901 objcount * sizeof(*ioo));
902 LASSERT(ioo != NULL);
903 for (niocount = i = 0; i < objcount; i++)
904 niocount += ioo[i].ioo_bufcnt;
906 remote_nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
907 niocount * sizeof(*remote_nb));
908 LASSERT(remote_nb != NULL);
910 if (body->oa.o_valid & OBD_MD_FLOSSCAPA)
911 capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 3);
913 size[REPLY_REC_OFF + 1] = niocount * sizeof(*rcs);
914 rc = lustre_pack_reply(req, 3, size, NULL);
917 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, obd_fail_val);
918 rcs = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF + 1,
919 niocount * sizeof(*rcs));
922 * Per-thread array of struct niobuf_{local,remote}'s was allocated by
925 local_nb = ost_tls(req)->local;
927 rc = ost_brw_lock_get(LCK_PW, exp, ioo, remote_nb, &lockh);
932 * If getting the lock took more time than
933 * client was willing to wait, drop it. b=11330
935 if (cfs_time_current_sec() > req->rq_deadline ||
936 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
938 CERROR("Dropping timed-out write from %s because locking "
939 "object "LPX64" took %ld seconds (limit was %ld).\n",
940 libcfs_id2str(req->rq_peer), ioo->ioo_id,
941 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
942 req->rq_deadline - req->rq_arrival_time.tv_sec);
943 GOTO(out_lock, rc = -ETIMEDOUT);
946 ost_rw_prolong_locks(req, ioo, remote_nb,&body->oa, LCK_PW);
948 /* obd_preprw clobbers oa->valid, so save what we need */
949 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
950 client_cksum = body->oa.o_cksum;
951 if (body->oa.o_valid & OBD_MD_FLFLAGS)
952 cksum_type = cksum_type_unpack(body->oa.o_flags);
955 /* Because we already sync grant info with client when reconnect,
956 * grant info will be cleared for resent req, then fed_grant and
957 * total_grant will not be modified in following preprw_write */
958 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
959 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
960 body->oa.o_valid &= ~OBD_MD_FLGRANT;
963 if (exp_connect_rmtclient(exp)) {
964 o_uid = body->oa.o_uid;
965 o_gid = body->oa.o_gid;
967 npages = OST_THREAD_POOL_SIZE;
968 rc = obd_preprw(OBD_BRW_WRITE, exp, &body->oa, objcount,
969 ioo, remote_nb, &npages, local_nb, oti, capa);
973 desc = ptlrpc_prep_bulk_exp(req, npages,
974 BULK_GET_SINK, OST_BULK_PORTAL);
976 GOTO(out, rc = -ENOMEM);
978 /* NB Having prepped, we must commit... */
980 for (i = 0; i < npages; i++)
981 ptlrpc_prep_bulk_page(desc, local_nb[i].page,
982 local_nb[i].offset & ~CFS_PAGE_MASK,
985 rc = sptlrpc_svc_prep_bulk(req, desc);
989 /* Check if client was evicted while we were doing i/o before touching
991 if (desc->bd_export->exp_failed)
994 rc = ptlrpc_start_bulk_transfer(desc);
996 time_t start = cfs_time_current_sec();
998 long timeoutl = req->rq_deadline -
999 cfs_time_current_sec();
1000 cfs_duration_t timeout = (timeoutl <= 0 || rc) ?
1001 CFS_TICK : cfs_time_seconds(timeoutl);
1002 lwi = LWI_TIMEOUT_INTERVAL(timeout, cfs_time_seconds(1),
1003 ost_bulk_timeout, desc);
1004 rc = l_wait_event(desc->bd_waitq,
1005 !ptlrpc_server_bulk_active(desc) ||
1006 desc->bd_export->exp_failed, &lwi);
1007 LASSERT(rc == 0 || rc == -ETIMEDOUT);
1008 /* Wait again if we changed deadline */
1009 } while ((rc == -ETIMEDOUT) &&
1010 (req->rq_deadline > cfs_time_current_sec()));
1012 if (rc == -ETIMEDOUT) {
1013 DEBUG_REQ(D_ERROR, req,
1014 "timeout on bulk GET after %ld%+lds",
1015 req->rq_deadline - start,
1016 cfs_time_current_sec() -
1018 ptlrpc_abort_bulk(desc);
1019 } else if (desc->bd_export->exp_failed) {
1020 DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET");
1022 ptlrpc_abort_bulk(desc);
1023 } else if (!desc->bd_success) {
1024 DEBUG_REQ(D_ERROR, req, "network error on bulk GET");
1025 /* XXX should this be a different errno? */
1028 rc = sptlrpc_svc_unwrap_bulk(req, desc);
1031 DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc);
1035 repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1037 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
1039 if (unlikely(client_cksum != 0 && rc == 0)) {
1040 static int cksum_counter;
1041 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1042 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
1043 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
1044 server_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1045 repbody->oa.o_cksum = server_cksum;
1047 if (unlikely(client_cksum != server_cksum)) {
1048 CERROR("client csum %x, server csum %x\n",
1049 client_cksum, server_cksum);
1051 } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){
1052 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1053 cksum_counter, libcfs_id2str(req->rq_peer),
1058 /* Must commit after prep above in all cases */
1059 rc = obd_commitrw(OBD_BRW_WRITE, exp, &repbody->oa, objcount, ioo,
1060 remote_nb, npages, local_nb, oti, rc);
1061 if (exp_connect_rmtclient(exp)) {
1062 repbody->oa.o_uid = o_uid;
1063 repbody->oa.o_gid = o_gid;
1066 if (unlikely(client_cksum != server_cksum && rc == 0)) {
1067 int new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1072 if (new_cksum == server_cksum)
1073 msg = "changed in transit before arrival at OST";
1074 else if (new_cksum == client_cksum)
1075 msg = "initial checksum before message complete";
1077 msg = "changed in transit AND after initial checksum";
1079 if (req->rq_peer.nid == desc->bd_sender) {
1083 router = libcfs_nid2str(desc->bd_sender);
1086 LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: %s from "
1087 "%s%s%s inum "LPU64"/"LPU64" object "
1088 LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
1089 exp->exp_obd->obd_name, msg,
1090 libcfs_id2str(req->rq_peer),
1092 body->oa.o_valid & OBD_MD_FLFID ?
1093 body->oa.o_fid : (__u64)0,
1094 body->oa.o_valid & OBD_MD_FLFID ?
1095 body->oa.o_generation :(__u64)0,
1097 body->oa.o_valid & OBD_MD_FLGROUP ?
1098 body->oa.o_gr : (__u64)0,
1100 local_nb[npages-1].offset +
1101 local_nb[npages-1].len - 1 );
1102 CERROR("client csum %x, original server csum %x, "
1103 "server csum now %x\n",
1104 client_cksum, server_cksum, new_cksum);
1110 /* set per-requested niobuf return codes */
1111 for (i = j = 0; i < niocount; i++) {
1112 int len = remote_nb[i].len;
1117 LASSERT(j < npages);
1118 if (local_nb[j].rc < 0)
1119 rcs[i] = local_nb[j].rc;
1120 len -= local_nb[j].len;
1125 LASSERT(j == npages);
1126 ptlrpc_lprocfs_brw(req, nob);
1130 ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh);
1133 ptlrpc_free_bulk(desc);
1136 oti_to_request(oti, req);
1137 target_committed_to_req(req);
1138 rc = ptlrpc_reply(req);
1139 } else if (!no_reply) {
1140 /* Only reply if there was no comms problem with bulk */
1141 target_committed_to_req(req);
1142 req->rq_status = rc;
1145 /* reply out callback would free */
1146 ptlrpc_req_drop_rs(req);
1147 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
1148 "client will retry\n",
1149 exp->exp_obd->obd_name,
1150 exp->exp_client_uuid.uuid,
1151 exp->exp_connection->c_remote_uuid.uuid,
1152 libcfs_id2str(req->rq_peer));
1157 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
1159 struct ost_body *body = NULL, *repbody;
1160 __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) };
1161 char *key, *val = NULL;
1162 int keylen, vallen, rc = 0;
1165 key = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, 1);
1167 DEBUG_REQ(D_HA, req, "no set_info key");
1170 keylen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF);
1172 if (KEY_IS(KEY_GRANT_SHRINK)) {
1173 rc = lustre_pack_reply(req, 2, size, NULL);
1177 rc = lustre_pack_reply(req, 1, NULL, NULL);
1182 vallen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1);
1184 if (KEY_IS(KEY_GRANT_SHRINK)) {
1185 body = lustre_swab_reqbuf(req, REQ_REC_OFF + 1,
1187 lustre_swab_ost_body);
1191 repbody = lustre_msg_buf(req->rq_repmsg,
1194 memcpy(repbody, body, sizeof(*body));
1195 val = (char*)repbody;
1197 val = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,0);
1200 if (KEY_IS(KEY_EVICT_BY_NID)) {
1202 obd_export_evict_by_nid(exp->exp_obd, val);
1207 rc = obd_set_info_async(exp, keylen, key, vallen, val, NULL);
1209 lustre_msg_set_status(req->rq_repmsg, 0);
1213 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
1216 int keylen, replylen, rc = 0;
1217 struct req_capsule *pill = &req->rq_pill;
1220 req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC);
1222 /* this common part for get_info rpc */
1223 key = req_capsule_client_get(pill, &RMF_SETINFO_KEY);
1225 DEBUG_REQ(D_HA, req, "no get_info key");
1228 keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT);
1230 rc = obd_get_info(exp, keylen, key, &replylen, NULL, NULL);
1234 req_capsule_set_size(pill, &RMF_GENERIC_DATA,
1235 RCL_SERVER, replylen);
1237 rc = req_capsule_server_pack(pill);
1241 reply = req_capsule_server_get(pill, &RMF_GENERIC_DATA);
1245 /* call again to fill in the reply buffer */
1246 rc = obd_get_info(exp, keylen, key, &replylen, reply, NULL);
1248 lustre_msg_set_status(req->rq_repmsg, 0);
1252 #ifdef HAVE_QUOTA_SUPPORT
1253 static int ost_handle_quotactl(struct ptlrpc_request *req)
1255 struct obd_quotactl *oqctl, *repoqc;
1259 oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1261 GOTO(out, rc = -EPROTO);
1263 rc = req_capsule_server_pack(&req->rq_pill);
1267 repoqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1268 req->rq_status = obd_quotactl(req->rq_export, oqctl);
1275 static int ost_handle_quotacheck(struct ptlrpc_request *req)
1277 struct obd_quotactl *oqctl;
1281 oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1285 rc = req_capsule_server_pack(&req->rq_pill);
1289 req->rq_status = obd_quotacheck(req->rq_export, oqctl);
1293 static int ost_handle_quota_adjust_qunit(struct ptlrpc_request *req)
1295 struct quota_adjust_qunit *oqaq, *repoqa;
1296 struct lustre_quota_ctxt *qctxt;
1300 qctxt = &req->rq_export->exp_obd->u.obt.obt_qctxt;
1301 oqaq = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
1303 GOTO(out, rc = -EPROTO);
1305 rc = req_capsule_server_pack(&req->rq_pill);
1309 repoqa = req_capsule_server_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
1310 req->rq_status = obd_quota_adjust_qunit(req->rq_export, oqaq, qctxt);
1318 static int ost_llog_handle_connect(struct obd_export *exp,
1319 struct ptlrpc_request *req)
1321 struct llogd_conn_body *body;
1325 body = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*body));
1326 rc = obd_llog_connect(exp, body);
1330 #define ost_init_sec_none(reply, exp) \
1332 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT | \
1333 OBD_CONNECT_RMT_CLIENT_FORCE | \
1334 OBD_CONNECT_OSS_CAPA); \
1335 spin_lock(&exp->exp_lock); \
1336 exp->exp_connect_flags = reply->ocd_connect_flags; \
1337 spin_unlock(&exp->exp_lock); \
1340 static int ost_init_sec_level(struct ptlrpc_request *req)
1342 struct obd_export *exp = req->rq_export;
1343 struct req_capsule *pill = &req->rq_pill;
1344 struct obd_device *obd = exp->exp_obd;
1345 struct filter_obd *filter = &obd->u.filter;
1346 char *client = libcfs_nid2str(req->rq_peer.nid);
1347 struct obd_connect_data *data, *reply;
1351 data = req_capsule_client_get(pill, &RMF_CONNECT_DATA);
1352 reply = req_capsule_server_get(pill, &RMF_CONNECT_DATA);
1353 if (data == NULL || reply == NULL)
1356 /* connection from MDT is always trusted */
1357 if (req->rq_auth_usr_mdt) {
1358 ost_init_sec_none(reply, exp);
1362 /* no GSS support case */
1363 if (!req->rq_auth_gss) {
1364 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1365 CWARN("client %s -> target %s does not user GSS, "
1366 "can not run under security level %d.\n",
1367 client, obd->obd_name, filter->fo_sec_level);
1370 ost_init_sec_none(reply, exp);
1375 /* old version case */
1376 if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
1377 !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
1378 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1379 CWARN("client %s -> target %s uses old version, "
1380 "can not run under security level %d.\n",
1381 client, obd->obd_name, filter->fo_sec_level);
1384 CWARN("client %s -> target %s uses old version, "
1385 "run under security level %d.\n",
1386 client, obd->obd_name, filter->fo_sec_level);
1387 ost_init_sec_none(reply, exp);
1392 remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
1394 if (!req->rq_auth_remote)
1395 CDEBUG(D_SEC, "client (local realm) %s -> target %s "
1396 "asked to be remote.\n", client, obd->obd_name);
1397 } else if (req->rq_auth_remote) {
1399 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
1400 "as remote by default.\n", client, obd->obd_name);
1404 if (!filter->fo_fl_oss_capa) {
1405 CDEBUG(D_SEC, "client %s -> target %s is set as remote,"
1406 " but OSS capabilities are not enabled: %d.\n",
1407 client, obd->obd_name, filter->fo_fl_oss_capa);
1412 switch (filter->fo_sec_level) {
1413 case LUSTRE_SEC_NONE:
1415 ost_init_sec_none(reply, exp);
1418 CDEBUG(D_SEC, "client %s -> target %s is set as remote, "
1419 "can not run under security level %d.\n",
1420 client, obd->obd_name, filter->fo_sec_level);
1423 case LUSTRE_SEC_REMOTE:
1425 ost_init_sec_none(reply, exp);
1427 case LUSTRE_SEC_ALL:
1429 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
1430 OBD_CONNECT_RMT_CLIENT_FORCE);
1431 if (!filter->fo_fl_oss_capa)
1432 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
1434 spin_lock(&exp->exp_lock);
1435 exp->exp_connect_flags = reply->ocd_connect_flags;
1436 spin_unlock(&exp->exp_lock);
1448 * this should be done in filter_connect()/filter_reconnect(), but
1449 * we can't obtain information like NID, which stored in incoming
1450 * request, thus can't decide what flavor to use. so we do it here.
1452 * This hack should be removed after the OST stack be rewritten, just
1453 * like what we are doing in mdt_obd_connect()/mdt_obd_reconnect().
1455 static int ost_connect_check_sptlrpc(struct ptlrpc_request *req)
1457 struct obd_export *exp = req->rq_export;
1458 struct filter_obd *filter = &exp->exp_obd->u.filter;
1459 struct sptlrpc_flavor flvr;
1462 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
1463 read_lock(&filter->fo_sptlrpc_lock);
1464 sptlrpc_target_choose_flavor(&filter->fo_sptlrpc_rset,
1468 read_unlock(&filter->fo_sptlrpc_lock);
1470 spin_lock(&exp->exp_lock);
1472 exp->exp_sp_peer = req->rq_sp_from;
1473 exp->exp_flvr = flvr;
1475 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
1476 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
1477 CERROR("unauthorized rpc flavor %x from %s, "
1478 "expect %x\n", req->rq_flvr.sf_rpc,
1479 libcfs_nid2str(req->rq_peer.nid),
1480 exp->exp_flvr.sf_rpc);
1484 spin_unlock(&exp->exp_lock);
1486 if (exp->exp_sp_peer != req->rq_sp_from) {
1487 CERROR("RPC source %s doesn't match %s\n",
1488 sptlrpc_part2name(req->rq_sp_from),
1489 sptlrpc_part2name(exp->exp_sp_peer));
1492 rc = sptlrpc_target_export_check(exp, req);
1499 static int ost_filter_recovery_request(struct ptlrpc_request *req,
1500 struct obd_device *obd, int *process)
1502 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1503 case OST_CONNECT: /* This will never get here, but for completeness. */
1504 case OST_DISCONNECT:
1515 case OBD_LOG_CANCEL:
1517 *process = target_queue_recovery_request(req, obd);
1521 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
1527 int ost_msg_check_version(struct lustre_msg *msg)
1531 switch(lustre_msg_get_opc(msg)) {
1533 case OST_DISCONNECT:
1536 case SEC_CTX_INIT_CONT:
1538 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
1540 CERROR("bad opc %u version %08x, expecting %08x\n",
1541 lustre_msg_get_opc(msg),
1542 lustre_msg_get_version(msg),
1543 LUSTRE_OBD_VERSION);
1556 #ifdef HAVE_QUOTA_SUPPORT
1557 case OST_QUOTACHECK:
1559 case OST_QUOTA_ADJUST_QUNIT:
1561 rc = lustre_msg_check_version(msg, LUSTRE_OST_VERSION);
1563 CERROR("bad opc %u version %08x, expecting %08x\n",
1564 lustre_msg_get_opc(msg),
1565 lustre_msg_get_version(msg),
1566 LUSTRE_OST_VERSION);
1571 case LDLM_BL_CALLBACK:
1572 case LDLM_CP_CALLBACK:
1573 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
1575 CERROR("bad opc %u version %08x, expecting %08x\n",
1576 lustre_msg_get_opc(msg),
1577 lustre_msg_get_version(msg),
1578 LUSTRE_DLM_VERSION);
1580 case LLOG_ORIGIN_CONNECT:
1581 case OBD_LOG_CANCEL:
1582 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
1584 CERROR("bad opc %u version %08x, expecting %08x\n",
1585 lustre_msg_get_opc(msg),
1586 lustre_msg_get_version(msg),
1587 LUSTRE_LOG_VERSION);
1590 CERROR("Unexpected opcode %d\n", lustre_msg_get_opc(msg));
1596 static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req,
1597 struct ldlm_lock *lock)
1599 struct niobuf_remote *nb;
1600 struct obd_ioobj *ioo;
1601 struct ost_body *body;
1602 int objcount, niocount;
1607 opc = lustre_msg_get_opc(req->rq_reqmsg);
1608 LASSERT(opc == OST_READ || opc == OST_WRITE);
1610 /* As the request may be covered by several locks, do not look at
1611 * o_handle, look at the RPC IO region. */
1612 body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
1614 objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
1616 ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,
1617 objcount * sizeof(*ioo));
1618 LASSERT(ioo != NULL);
1619 for (niocount = i = 0; i < objcount; i++)
1620 niocount += ioo[i].ioo_bufcnt;
1622 nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
1623 niocount * sizeof(*nb));
1624 LASSERT(nb != NULL);
1627 if (opc == OST_READ)
1630 start = nb[0].offset & CFS_PAGE_MASK;
1631 end = (nb[ioo->ioo_bufcnt - 1].offset +
1632 nb[ioo->ioo_bufcnt - 1].len - 1) | ~CFS_PAGE_MASK;
1634 LASSERT(lock->l_resource != NULL);
1635 if (!osc_res_name_eq(ioo->ioo_id, ioo->ioo_gr,
1636 &lock->l_resource->lr_name))
1639 if (!(lock->l_granted_mode & mode))
1642 if (lock->l_policy_data.l_extent.end < start ||
1643 lock->l_policy_data.l_extent.start > end)
1650 * Swab buffers needed to call ost_rw_prolong_locks() and call it.
1651 * Return the value from ost_rw_prolong_locks() which is non-zero if
1652 * there is a cancelled lock which is waiting for this IO request.
1654 static int ost_rw_hpreq_check(struct ptlrpc_request *req)
1656 struct niobuf_remote *nb;
1657 struct obd_ioobj *ioo;
1658 struct ost_body *body;
1659 int objcount, niocount;
1663 opc = lustre_msg_get_opc(req->rq_reqmsg);
1664 LASSERT(opc == OST_READ || opc == OST_WRITE);
1666 body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body));
1667 LASSERT(body != NULL);
1669 objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) /
1671 ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1,
1672 objcount * sizeof(*ioo));
1673 LASSERT(ioo != NULL);
1675 for (niocount = i = 0; i < objcount; i++)
1676 niocount += ioo[i].ioo_bufcnt;
1677 nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2,
1678 niocount * sizeof(*nb));
1679 LASSERT(nb != NULL);
1680 LASSERT(niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK));
1683 if (opc == OST_READ)
1685 RETURN(ost_rw_prolong_locks(req, ioo, nb, &body->oa, mode));
1688 static int ost_punch_prolong_locks(struct ptlrpc_request *req, struct obdo *oa)
1690 struct ldlm_res_id res_id = { .name = { oa->o_id } };
1691 struct ost_prolong_data opd = { 0 };
1696 end = start + oa->o_blocks;
1698 opd.opd_mode = LCK_PW;
1699 opd.opd_exp = req->rq_export;
1700 opd.opd_policy.l_extent.start = start & CFS_PAGE_MASK;
1701 if (oa->o_blocks == OBD_OBJECT_EOF || end < start)
1702 opd.opd_policy.l_extent.end = OBD_OBJECT_EOF;
1704 opd.opd_policy.l_extent.end = end | ~CFS_PAGE_MASK;
1706 /* prolong locks for the current service time of the corresponding
1707 * portal (= OST_IO_PORTAL) */
1708 opd.opd_timeout = AT_OFF ? obd_timeout / 2:
1709 max(at_est2timeout(at_get(&req->rq_rqbd->
1710 rqbd_service->srv_at_estimate)), ldlm_timeout);
1712 CDEBUG(D_DLMTRACE,"refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1713 res_id.name[0], res_id.name[1], opd.opd_policy.l_extent.start,
1714 opd.opd_policy.l_extent.end);
1717 ldlm_resource_iterate(req->rq_export->exp_obd->obd_namespace, &res_id,
1718 ost_prolong_locks_iter, &opd);
1719 RETURN(opd.opd_lock_match);
1722 static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req,
1723 struct ldlm_lock *lock)
1725 struct ost_body *body;
1728 body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body),
1730 LASSERT(body != NULL);
1732 if (body->oa.o_valid & OBD_MD_FLHANDLE &&
1733 body->oa.o_handle.cookie == lock->l_handle.h_cookie)
1738 static int ost_punch_hpreq_check(struct ptlrpc_request *req)
1740 struct ost_body *body = lustre_msg_buf(req->rq_reqmsg,
1741 REQ_REC_OFF, sizeof(*body));
1742 LASSERT(body != NULL);
1743 LASSERT(!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
1744 !(body->oa.o_flags & OBD_FL_TRUNCLOCK));
1746 RETURN(ost_punch_prolong_locks(req, &body->oa));
1749 struct ptlrpc_hpreq_ops ost_hpreq_rw = {
1750 .hpreq_lock_match = ost_rw_hpreq_lock_match,
1751 .hpreq_check = ost_rw_hpreq_check,
1754 struct ptlrpc_hpreq_ops ost_hpreq_punch = {
1755 .hpreq_lock_match = ost_punch_hpreq_lock_match,
1756 .hpreq_check = ost_punch_hpreq_check,
1759 /** Assign high priority operations to the request if needed. */
1760 static int ost_hpreq_handler(struct ptlrpc_request *req)
1763 if (req->rq_export) {
1764 int opc = lustre_msg_get_opc(req->rq_reqmsg);
1765 struct ost_body *body;
1767 if (opc == OST_READ || opc == OST_WRITE) {
1768 struct niobuf_remote *nb;
1769 struct obd_ioobj *ioo;
1770 int objcount, niocount;
1773 body = lustre_swab_reqbuf(req, REQ_REC_OFF,
1777 CERROR("Missing/short ost_body\n");
1780 objcount = lustre_msg_buflen(req->rq_reqmsg,
1783 if (objcount == 0) {
1784 CERROR("Missing/short ioobj\n");
1788 CERROR("too many ioobjs (%d)\n", objcount);
1792 swab = !lustre_req_swabbed(req, REQ_REC_OFF + 1) &&
1793 lustre_msg_swabbed(req->rq_reqmsg);
1794 ioo = lustre_swab_reqbuf(req, REQ_REC_OFF + 1,
1795 objcount * sizeof(*ioo),
1796 lustre_swab_obd_ioobj);
1798 CERROR("Missing/short ioobj\n");
1801 for (niocount = i = 0; i < objcount; i++) {
1803 lustre_swab_obd_ioobj(&ioo[i]);
1804 if (ioo[i].ioo_bufcnt == 0) {
1805 CERROR("ioo[%d] has zero bufcnt\n", i);
1808 niocount += ioo[i].ioo_bufcnt;
1810 if (niocount > PTLRPC_MAX_BRW_PAGES) {
1811 DEBUG_REQ(D_ERROR, req, "bulk has too many "
1812 "pages (%d)", niocount);
1816 swab = !lustre_req_swabbed(req, REQ_REC_OFF + 2) &&
1817 lustre_msg_swabbed(req->rq_reqmsg);
1818 nb = lustre_swab_reqbuf(req, REQ_REC_OFF + 2,
1819 niocount * sizeof(*nb),
1820 lustre_swab_niobuf_remote);
1822 CERROR("Missing/short niobuf\n");
1827 /* swab remaining niobufs */
1828 for (i = 1; i < niocount; i++)
1829 lustre_swab_niobuf_remote(&nb[i]);
1832 if (niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
1833 req->rq_ops = &ost_hpreq_rw;
1834 } else if (opc == OST_PUNCH) {
1835 body = lustre_swab_reqbuf(req, REQ_REC_OFF,
1839 CERROR("Missing/short ost_body\n");
1843 if (!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
1844 !(body->oa.o_flags & OBD_FL_TRUNCLOCK))
1845 req->rq_ops = &ost_hpreq_punch;
1851 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
1852 int ost_handle(struct ptlrpc_request *req)
1854 struct obd_trans_info trans_info = { 0, };
1855 struct obd_trans_info *oti = &trans_info;
1856 int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
1857 struct obd_device *obd = NULL;
1860 LASSERT(current->journal_info == NULL);
1862 /* primordial rpcs don't affect server recovery */
1863 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1865 case SEC_CTX_INIT_CONT:
1870 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
1872 /* XXX identical to MDS */
1873 if (lustre_msg_get_opc(req->rq_reqmsg) != OST_CONNECT) {
1876 if (req->rq_export == NULL) {
1877 CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
1878 lustre_msg_get_opc(req->rq_reqmsg),
1879 libcfs_id2str(req->rq_peer));
1880 req->rq_status = -ENOTCONN;
1881 GOTO(out, rc = -ENOTCONN);
1884 obd = req->rq_export->exp_obd;
1886 /* Check for aborted recovery. */
1887 spin_lock_bh(&obd->obd_processing_task_lock);
1888 recovering = obd->obd_recovering;
1889 spin_unlock_bh(&obd->obd_processing_task_lock);
1891 rc = ost_filter_recovery_request(req, obd,
1893 if (rc || !should_process)
1895 else if (should_process < 0) {
1896 req->rq_status = should_process;
1897 rc = ptlrpc_error(req);
1905 rc = ost_msg_check_version(req->rq_reqmsg);
1909 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1911 CDEBUG(D_INODE, "connect\n");
1912 req_capsule_set(&req->rq_pill, &RQF_OST_CONNECT);
1913 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET))
1915 rc = target_handle_connect(req);
1916 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2))
1919 rc = ost_init_sec_level(req);
1921 rc = ost_connect_check_sptlrpc(req);
1925 case OST_DISCONNECT:
1926 CDEBUG(D_INODE, "disconnect\n");
1927 req_capsule_set(&req->rq_pill, &RQF_OST_DISCONNECT);
1928 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DISCONNECT_NET))
1930 rc = target_handle_disconnect(req);
1933 CDEBUG(D_INODE, "create\n");
1934 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET))
1936 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1937 GOTO(out, rc = -ENOSPC);
1938 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1939 GOTO(out, rc = -EROFS);
1940 rc = ost_create(req->rq_export, req, oti);
1943 CDEBUG(D_INODE, "destroy\n");
1944 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DESTROY_NET))
1946 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1947 GOTO(out, rc = -EROFS);
1948 rc = ost_destroy(req->rq_export, req, oti);
1951 CDEBUG(D_INODE, "getattr\n");
1952 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GETATTR_NET))
1954 rc = ost_getattr(req->rq_export, req);
1957 CDEBUG(D_INODE, "setattr\n");
1958 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_NET))
1960 rc = ost_setattr(req->rq_export, req, oti);
1963 CDEBUG(D_INODE, "write\n");
1964 /* req->rq_request_portal would be nice, if it was set */
1965 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
1966 CERROR("%s: deny write request from %s to portal %u\n",
1967 req->rq_export->exp_obd->obd_name,
1968 obd_export_nid2str(req->rq_export),
1969 req->rq_rqbd->rqbd_service->srv_req_portal);
1970 GOTO(out, rc = -EPROTO);
1972 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
1974 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1975 GOTO(out, rc = -ENOSPC);
1976 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
1977 GOTO(out, rc = -EROFS);
1978 rc = ost_brw_write(req, oti);
1979 LASSERT(current->journal_info == NULL);
1980 /* ost_brw_write sends its own replies */
1983 CDEBUG(D_INODE, "read\n");
1984 /* req->rq_request_portal would be nice, if it was set */
1985 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
1986 CERROR("%s: deny read request from %s to portal %u\n",
1987 req->rq_export->exp_obd->obd_name,
1988 obd_export_nid2str(req->rq_export),
1989 req->rq_rqbd->rqbd_service->srv_req_portal);
1990 GOTO(out, rc = -EPROTO);
1992 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
1994 rc = ost_brw_read(req, oti);
1995 LASSERT(current->journal_info == NULL);
1996 /* ost_brw_read sends its own replies */
1999 CDEBUG(D_INODE, "punch\n");
2000 if (OBD_FAIL_CHECK(OBD_FAIL_OST_PUNCH_NET))
2002 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2003 GOTO(out, rc = -EROFS);
2004 rc = ost_punch(req->rq_export, req, oti);
2007 CDEBUG(D_INODE, "statfs\n");
2008 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_NET))
2010 rc = ost_statfs(req);
2013 CDEBUG(D_INODE, "sync\n");
2014 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SYNC_NET))
2016 rc = ost_sync(req->rq_export, req);
2019 DEBUG_REQ(D_INODE, req, "set_info");
2020 req_capsule_set(&req->rq_pill, &RQF_OST_SET_INFO);
2021 rc = ost_set_info(req->rq_export, req);
2024 DEBUG_REQ(D_INODE, req, "get_info");
2025 rc = ost_get_info(req->rq_export, req);
2027 #ifdef HAVE_QUOTA_SUPPORT
2028 case OST_QUOTACHECK:
2029 CDEBUG(D_INODE, "quotacheck\n");
2030 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACHECK);
2031 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACHECK_NET))
2033 rc = ost_handle_quotacheck(req);
2036 CDEBUG(D_INODE, "quotactl\n");
2037 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACTL);
2038 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACTL_NET))
2040 rc = ost_handle_quotactl(req);
2042 case OST_QUOTA_ADJUST_QUNIT:
2043 CDEBUG(D_INODE, "quota_adjust_qunit\n");
2044 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTA_ADJUST_QUNIT);
2045 rc = ost_handle_quota_adjust_qunit(req);
2049 DEBUG_REQ(D_INODE, req, "ping");
2050 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
2051 rc = target_handle_ping(req);
2053 /* FIXME - just reply status */
2054 case LLOG_ORIGIN_CONNECT:
2055 DEBUG_REQ(D_INODE, req, "log connect");
2056 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_CONNECT);
2057 rc = ost_llog_handle_connect(req->rq_export, req);
2058 req->rq_status = rc;
2059 rc = req_capsule_server_pack(&req->rq_pill);
2062 RETURN(ptlrpc_reply(req));
2063 case OBD_LOG_CANCEL:
2064 CDEBUG(D_INODE, "log cancel\n");
2065 req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL);
2066 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
2068 rc = llog_origin_handle_cancel(req);
2069 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP))
2071 req->rq_status = rc;
2072 rc = req_capsule_server_pack(&req->rq_pill);
2075 RETURN(ptlrpc_reply(req));
2077 CDEBUG(D_INODE, "enqueue\n");
2078 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
2079 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE))
2081 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
2082 ldlm_server_blocking_ast,
2083 ldlm_server_glimpse_ast);
2084 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
2087 CDEBUG(D_INODE, "convert\n");
2088 req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT);
2089 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT))
2091 rc = ldlm_handle_convert(req);
2094 CDEBUG(D_INODE, "cancel\n");
2095 req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL);
2096 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL))
2098 rc = ldlm_handle_cancel(req);
2100 case LDLM_BL_CALLBACK:
2101 case LDLM_CP_CALLBACK:
2102 CDEBUG(D_INODE, "callback\n");
2103 CERROR("callbacks should not happen on OST\n");
2106 CERROR("Unexpected opcode %d\n",
2107 lustre_msg_get_opc(req->rq_reqmsg));
2108 req->rq_status = -ENOTSUPP;
2109 rc = ptlrpc_error(req);
2113 LASSERT(current->journal_info == NULL);
2116 /* If we're DISCONNECTing, the export_data is already freed */
2117 if (!rc && lustre_msg_get_opc(req->rq_reqmsg) != OST_DISCONNECT)
2118 target_committed_to_req(req);
2122 oti_to_request(oti, req);
2124 target_send_reply(req, rc, fail);
2127 EXPORT_SYMBOL(ost_handle);
2129 * free per-thread pool created by ost_thread_init().
2131 static void ost_thread_done(struct ptlrpc_thread *thread)
2133 struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local
2138 LASSERT(thread != NULL);
2141 * be prepared to handle partially-initialized pools (because this is
2142 * called from ost_thread_init() for cleanup.
2144 tls = thread->t_data;
2147 thread->t_data = NULL;
2153 * initialize per-thread page pool (bug 5137).
2155 static int ost_thread_init(struct ptlrpc_thread *thread)
2157 struct ost_thread_local_cache *tls;
2161 LASSERT(thread != NULL);
2162 LASSERT(thread->t_data == NULL);
2163 LASSERTF(thread->t_id <= OSS_THREADS_MAX, "%u\n", thread->t_id);
2168 thread->t_data = tls;
2172 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
2174 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
2175 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
2177 struct ost_obd *ost = &obd->u.ost;
2178 struct lprocfs_static_vars lvars;
2179 int oss_min_threads;
2180 int oss_max_threads;
2181 int oss_min_create_threads;
2182 int oss_max_create_threads;
2186 rc = cleanup_group_info();
2190 lprocfs_ost_init_vars(&lvars);
2191 lprocfs_obd_setup(obd, lvars.obd_vars);
2193 sema_init(&ost->ost_health_sem, 1);
2195 if (oss_num_threads) {
2196 /* If oss_num_threads is set, it is the min and the max. */
2197 if (oss_num_threads > OSS_THREADS_MAX)
2198 oss_num_threads = OSS_THREADS_MAX;
2199 if (oss_num_threads < OSS_THREADS_MIN)
2200 oss_num_threads = OSS_THREADS_MIN;
2201 oss_max_threads = oss_min_threads = oss_num_threads;
2203 /* Base min threads on memory and cpus */
2204 oss_min_threads = num_possible_cpus() * num_physpages >>
2205 (27 - CFS_PAGE_SHIFT);
2206 if (oss_min_threads < OSS_THREADS_MIN)
2207 oss_min_threads = OSS_THREADS_MIN;
2208 /* Insure a 4x range for dynamic threads */
2209 if (oss_min_threads > OSS_THREADS_MAX / 4)
2210 oss_min_threads = OSS_THREADS_MAX / 4;
2211 oss_max_threads = min(OSS_THREADS_MAX, oss_min_threads * 4 + 1);
2215 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2216 OST_MAXREPSIZE, OST_REQUEST_PORTAL,
2217 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2218 ost_handle, LUSTRE_OSS_NAME,
2219 obd->obd_proc_entry, target_print_req,
2220 oss_min_threads, oss_max_threads,
2221 "ll_ost", LCT_DT_THREAD, NULL);
2222 if (ost->ost_service == NULL) {
2223 CERROR("failed to start service\n");
2224 GOTO(out_lprocfs, rc = -ENOMEM);
2227 rc = ptlrpc_start_threads(obd, ost->ost_service);
2229 GOTO(out_service, rc = -EINVAL);
2231 if (oss_num_create_threads) {
2232 if (oss_num_create_threads > OSS_MAX_CREATE_THREADS)
2233 oss_num_create_threads = OSS_MAX_CREATE_THREADS;
2234 if (oss_num_create_threads < OSS_MIN_CREATE_THREADS)
2235 oss_num_create_threads = OSS_MIN_CREATE_THREADS;
2236 oss_min_create_threads = oss_max_create_threads =
2237 oss_num_create_threads;
2239 oss_min_create_threads = OSS_MIN_CREATE_THREADS;
2240 oss_max_create_threads = OSS_MAX_CREATE_THREADS;
2243 ost->ost_create_service =
2244 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2245 OST_MAXREPSIZE, OST_CREATE_PORTAL,
2246 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2247 ost_handle, "ost_create",
2248 obd->obd_proc_entry, target_print_req,
2249 oss_min_create_threads, oss_max_create_threads,
2250 "ll_ost_creat", LCT_DT_THREAD, NULL);
2251 if (ost->ost_create_service == NULL) {
2252 CERROR("failed to start OST create service\n");
2253 GOTO(out_service, rc = -ENOMEM);
2256 rc = ptlrpc_start_threads(obd, ost->ost_create_service);
2258 GOTO(out_create, rc = -EINVAL);
2260 ost->ost_io_service =
2261 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2262 OST_MAXREPSIZE, OST_IO_PORTAL,
2263 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2264 ost_handle, "ost_io",
2265 obd->obd_proc_entry, target_print_req,
2266 oss_min_threads, oss_max_threads,
2267 "ll_ost_io", LCT_DT_THREAD, ost_hpreq_handler);
2268 if (ost->ost_io_service == NULL) {
2269 CERROR("failed to start OST I/O service\n");
2270 GOTO(out_create, rc = -ENOMEM);
2273 ost->ost_io_service->srv_init = ost_thread_init;
2274 ost->ost_io_service->srv_done = ost_thread_done;
2275 ost->ost_io_service->srv_cpu_affinity = 1;
2276 rc = ptlrpc_start_threads(obd, ost->ost_io_service);
2278 GOTO(out_io, rc = -EINVAL);
2280 ping_evictor_start();
2285 ptlrpc_unregister_service(ost->ost_io_service);
2286 ost->ost_io_service = NULL;
2288 ptlrpc_unregister_service(ost->ost_create_service);
2289 ost->ost_create_service = NULL;
2291 ptlrpc_unregister_service(ost->ost_service);
2292 ost->ost_service = NULL;
2294 lprocfs_obd_cleanup(obd);
2298 static int ost_cleanup(struct obd_device *obd)
2300 struct ost_obd *ost = &obd->u.ost;
2304 ping_evictor_stop();
2306 spin_lock_bh(&obd->obd_processing_task_lock);
2307 if (obd->obd_recovering) {
2308 target_cancel_recovery_timer(obd);
2309 obd->obd_recovering = 0;
2311 spin_unlock_bh(&obd->obd_processing_task_lock);
2313 down(&ost->ost_health_sem);
2314 ptlrpc_unregister_service(ost->ost_service);
2315 ptlrpc_unregister_service(ost->ost_create_service);
2316 ptlrpc_unregister_service(ost->ost_io_service);
2317 ost->ost_service = NULL;
2318 ost->ost_create_service = NULL;
2319 up(&ost->ost_health_sem);
2321 lprocfs_obd_cleanup(obd);
2326 static int ost_health_check(struct obd_device *obd)
2328 struct ost_obd *ost = &obd->u.ost;
2331 down(&ost->ost_health_sem);
2332 rc |= ptlrpc_service_health_check(ost->ost_service);
2333 rc |= ptlrpc_service_health_check(ost->ost_create_service);
2334 rc |= ptlrpc_service_health_check(ost->ost_io_service);
2335 up(&ost->ost_health_sem);
2338 * health_check to return 0 on healthy
2339 * and 1 on unhealthy.
2347 struct ost_thread_local_cache *ost_tls(struct ptlrpc_request *r)
2349 return (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
2352 /* use obd ops to offer management infrastructure */
2353 static struct obd_ops ost_obd_ops = {
2354 .o_owner = THIS_MODULE,
2355 .o_setup = ost_setup,
2356 .o_cleanup = ost_cleanup,
2357 .o_health_check = ost_health_check,
2361 static int __init ost_init(void)
2363 struct lprocfs_static_vars lvars;
2367 lprocfs_ost_init_vars(&lvars);
2368 rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars,
2369 LUSTRE_OSS_NAME, NULL);
2371 if (ost_num_threads != 0 && oss_num_threads == 0) {
2372 LCONSOLE_INFO("ost_num_threads module parameter is deprecated, "
2373 "use oss_num_threads instead or unset both for "
2374 "dynamic thread startup\n");
2375 oss_num_threads = ost_num_threads;
2381 static void /*__exit*/ ost_exit(void)
2383 class_unregister_type(LUSTRE_OSS_NAME);
2386 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2387 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
2388 MODULE_LICENSE("GPL");
2390 module_init(ost_init);
2391 module_exit(ost_exit);