1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2001-2003 Cluster File Systems, Inc.
5 * Author: Peter J. Braam <braam@clusterfs.com>
6 * Author: Phil Schwan <phil@clusterfs.com>
8 * This file is part of Lustre, http://www.lustre.org.
10 * Lustre is free software; you can redistribute it and/or
11 * modify it under the terms of version 2 of the GNU General Public
12 * License as published by the Free Software Foundation.
14 * Lustre is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with Lustre; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 * Storage Target Handling functions
24 * Lustre Object Server Module (OST)
26 * This server is single threaded at present (but can easily be multi
27 * threaded). For testing and management it is treated as an
28 * obd_device, although it does not export a full OBD method table
29 * (the requests are coming in over the wire, so object target
30 * modules do not have a full method table.)
34 # define EXPORT_SYMTAB
36 #define DEBUG_SUBSYSTEM S_OST
38 #include <linux/module.h>
39 #include <linux/obd_ost.h>
40 #include <linux/lustre_net.h>
41 #include <linux/lustre_dlm.h>
42 #include <linux/lustre_export.h>
43 #include <linux/init.h>
44 #include <linux/lprocfs_status.h>
45 #include <linux/lustre_commit_confd.h>
46 #include <portals/list.h>
48 void oti_init(struct obd_trans_info *oti, struct ptlrpc_request *req)
52 memset(oti, 0, sizeof *oti);
54 if (req->rq_repmsg && req->rq_reqmsg != 0)
55 oti->oti_transno = req->rq_repmsg->transno;
58 void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req)
60 struct oti_req_ack_lock *ack_lock;
67 req->rq_repmsg->transno = oti->oti_transno;
69 /* XXX 4 == entries in oti_ack_locks??? */
70 for (ack_lock = oti->oti_ack_locks, i = 0; i < 4; i++, ack_lock++) {
73 ldlm_put_lock_into_req(req, &ack_lock->lock, ack_lock->mode);
77 static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req,
78 struct obd_trans_info *oti)
80 struct ost_body *body, *repbody;
81 int rc, size = sizeof(*body);
84 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
88 rc = lustre_pack_reply(req, 1, &size, NULL);
92 if (body->oa.o_valid & OBD_MD_FLCOOKIE)
93 oti->oti_logcookies = obdo_logcookie(&body->oa);
94 repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
95 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
96 req->rq_status = obd_destroy(exp, &body->oa, NULL, oti);
100 static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
102 struct ost_body *body, *repbody;
103 int rc, size = sizeof(*body);
106 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
110 rc = lustre_pack_reply(req, 1, &size, NULL);
114 repbody = lustre_msg_buf (req->rq_repmsg, 0, sizeof(*repbody));
115 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
116 req->rq_status = obd_getattr(exp, &repbody->oa, NULL);
120 static int ost_statfs(struct ptlrpc_request *req)
122 struct obd_statfs *osfs;
123 int rc, size = sizeof(*osfs);
126 rc = lustre_pack_reply(req, 1, &size, NULL);
130 osfs = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*osfs));
132 req->rq_status = obd_statfs(req->rq_export->exp_obd, osfs, jiffies-HZ);
133 if (req->rq_status != 0)
134 CERROR("ost: statfs failed: rc %d\n", req->rq_status);
139 static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
140 struct obd_trans_info *oti)
142 struct ost_body *body, *repbody;
143 int rc, size = sizeof(*repbody);
146 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
150 rc = lustre_pack_reply(req, 1, &size, NULL);
154 repbody = lustre_msg_buf (req->rq_repmsg, 0, sizeof(*repbody));
155 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
156 oti->oti_logcookies = obdo_logcookie(&repbody->oa);
157 req->rq_status = obd_create(exp, &repbody->oa, NULL, oti);
158 //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
162 static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
163 struct obd_trans_info *oti)
165 struct ost_body *body, *repbody;
166 int rc, size = sizeof(*repbody);
169 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
173 if ((body->oa.o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
174 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
177 rc = lustre_pack_reply(req, 1, &size, NULL);
181 repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
182 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
183 req->rq_status = obd_punch(exp, &repbody->oa, NULL, repbody->oa.o_size,
184 repbody->oa.o_blocks, oti);
188 static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req)
190 struct ost_body *body, *repbody;
191 int rc, size = sizeof(*repbody);
194 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
198 rc = lustre_pack_reply(req, 1, &size, NULL);
202 repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
203 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
204 req->rq_status = obd_sync(exp, &repbody->oa, NULL, repbody->oa.o_size,
205 repbody->oa.o_blocks);
209 static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req,
210 struct obd_trans_info *oti)
212 struct ost_body *body, *repbody;
213 int rc, size = sizeof(*repbody);
216 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
220 rc = lustre_pack_reply(req, 1, &size, NULL);
224 repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
225 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
227 req->rq_status = obd_setattr(exp, &repbody->oa, NULL, oti);
231 static int ost_bulk_timeout(void *data)
234 /* We don't fail the connection here, because having the export
235 * killed makes the (vital) call to commitrw very sad.
240 static int get_per_page_niobufs(struct obd_ioobj *ioo, int nioo,
241 struct niobuf_remote *rnb, int nrnb,
242 struct niobuf_remote **pp_rnbp)
244 /* Copy a remote niobuf, splitting it into page-sized chunks
245 * and setting ioo[i].ioo_bufcnt accordingly */
246 struct niobuf_remote *pp_rnb;
253 /* first count and check the number of pages required */
254 for (i = 0; i < nioo; i++)
255 for (j = 0; j < ioo->ioo_bufcnt; j++, rnbidx++) {
256 obd_off offset = rnb[rnbidx].offset;
257 obd_off p0 = offset >> PAGE_SHIFT;
258 obd_off pn = (offset + rnb[rnbidx].len - 1)>>PAGE_SHIFT;
260 LASSERT(rnbidx < nrnb);
262 npages += (pn + 1 - p0);
264 if (rnb[rnbidx].len == 0) {
265 CERROR("zero len BRW: obj %d objid "LPX64
266 " buf %u\n", i, ioo[i].ioo_id, j);
270 rnb[rnbidx].offset <= rnb[rnbidx-1].offset) {
271 CERROR("unordered BRW: obj %d objid "LPX64
272 " buf %u offset "LPX64" <= "LPX64"\n",
273 i, ioo[i].ioo_id, j, rnb[rnbidx].offset,
279 LASSERT(rnbidx == nrnb);
281 if (npages == nrnb) { /* all niobufs are for single pages */
286 OBD_ALLOC(pp_rnb, sizeof(*pp_rnb) * npages);
290 /* now do the actual split */
292 for (i = 0; i < nioo; i++) {
295 for (j = 0; j < ioo[i].ioo_bufcnt; j++, rnbidx++) {
296 obd_off off = rnb[rnbidx].offset;
297 int nob = rnb[rnbidx].len;
299 LASSERT(rnbidx < nrnb);
301 obd_off poff = off & (PAGE_SIZE - 1);
302 int pnob = (poff + nob > PAGE_SIZE) ?
303 PAGE_SIZE - poff : nob;
305 LASSERT(page < npages);
306 pp_rnb[page].len = pnob;
307 pp_rnb[page].offset = off;
308 pp_rnb[page].flags = rnb->flags;
310 CDEBUG(D_PAGE, " obj %d id "LPX64
311 "page %d(%d) "LPX64" for %d\n",
312 i, ioo[i].ioo_id, obj_pages, page,
313 pp_rnb[page].offset, pp_rnb[page].len);
322 ioo[i].ioo_bufcnt = obj_pages;
324 LASSERT(page == npages);
330 static void free_per_page_niobufs (int npages, struct niobuf_remote *pp_rnb,
331 struct niobuf_remote *rnb)
333 if (pp_rnb == rnb) /* didn't allocate above */
336 OBD_FREE(pp_rnb, sizeof(*pp_rnb) * npages);
340 obd_count ost_checksum_bulk(struct ptlrpc_bulk_desc *desc)
343 struct ptlrpc_bulk_page *bp;
345 list_for_each_entry(bp, &desc->bd_page_list, bp_link) {
346 ost_checksum(&cksum, kmap(bp->bp_page) + bp->bp_pageoffset,
355 static int ost_brw_read(struct ptlrpc_request *req)
357 struct ptlrpc_bulk_desc *desc;
358 struct niobuf_remote *remote_nb;
359 struct niobuf_remote *pp_rnb;
360 struct niobuf_local *local_nb;
361 struct obd_ioobj *ioo;
362 struct ost_body *body, *repbody;
363 struct l_wait_info lwi;
364 struct obd_trans_info oti = { 0 };
365 char str[PTL_NALFMT_SIZE];
366 int size[1] = { sizeof(*body) };
375 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
376 GOTO(out, rc = -EIO);
378 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK | OBD_FAIL_ONCE,
379 (obd_timeout + 1) / 4);
381 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
383 CERROR("Missing/short ost_body\n");
384 GOTO(out, rc = -EFAULT);
387 ioo = lustre_swab_reqbuf(req, 1, sizeof(*ioo), lustre_swab_obd_ioobj);
389 CERROR("Missing/short ioobj\n");
390 GOTO(out, rc = -EFAULT);
393 niocount = ioo->ioo_bufcnt;
394 remote_nb = lustre_swab_reqbuf(req, 2, niocount * sizeof(*remote_nb),
395 lustre_swab_niobuf_remote);
396 if (remote_nb == NULL) {
397 CERROR("Missing/short niobuf\n");
398 GOTO(out, rc = -EFAULT);
400 if (lustre_msg_swabbed(req->rq_reqmsg)) { /* swab remaining niobufs */
401 for (i = 1; i < niocount; i++)
402 lustre_swab_niobuf_remote (&remote_nb[i]);
405 rc = lustre_pack_reply(req, 1, size, NULL);
409 /* FIXME all niobuf splitting should be done in obdfilter if needed */
410 /* CAVEAT EMPTOR this sets ioo->ioo_bufcnt to # pages */
411 npages = get_per_page_niobufs(ioo, 1, remote_nb, niocount, &pp_rnb);
413 GOTO(out, rc = npages);
415 OBD_ALLOC(local_nb, sizeof(*local_nb) * npages);
416 if (local_nb == NULL)
417 GOTO(out_pp_rnb, rc = -ENOMEM);
419 desc = ptlrpc_prep_bulk_exp(req, BULK_PUT_SOURCE, OST_BULK_PORTAL);
421 GOTO(out_local, rc = -ENOMEM);
423 rc = obd_preprw(OBD_BRW_READ, req->rq_export, &body->oa, 1,
424 ioo, npages, pp_rnb, local_nb, &oti);
429 for (i = 0; i < npages; i++) {
430 int page_rc = local_nb[i].rc;
432 if (page_rc < 0) { /* error */
437 LASSERT(page_rc <= pp_rnb[i].len);
439 if (page_rc != 0) { /* some data! */
440 LASSERT (local_nb[i].page != NULL);
441 rc = ptlrpc_prep_bulk_page(desc, local_nb[i].page,
442 pp_rnb[i].offset& ~PAGE_MASK,
448 if (page_rc != pp_rnb[i].len) { /* short read */
449 /* All subsequent pages should be 0 */
451 LASSERT(local_nb[i].rc == 0);
457 rc = ptlrpc_bulk_put(desc);
459 lwi = LWI_TIMEOUT(obd_timeout * HZ / 4,
460 ost_bulk_timeout, desc);
461 rc = l_wait_event(desc->bd_waitq,
462 ptlrpc_bulk_complete(desc), &lwi);
464 LASSERT(rc == -ETIMEDOUT);
465 DEBUG_REQ(D_ERROR, req, "timeout on bulk PUT");
466 ptlrpc_abort_bulk(desc);
469 DEBUG_REQ(D_ERROR, req, "bulk PUT failed: rc %d\n", rc);
471 comms_error = rc != 0;
474 /* Must commit after prep above in all cases */
475 rc = obd_commitrw(OBD_BRW_READ, req->rq_export, &body->oa, 1,
476 ioo, npages, local_nb, &oti);
479 repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
480 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
483 repbody->oa.o_nlink = ost_checksum_bulk(desc);
484 repbody->oa.o_valid |= OBD_MD_FLCKSUM;
489 ptlrpc_free_bulk(desc);
491 OBD_FREE(local_nb, sizeof(*local_nb) * npages);
493 free_per_page_niobufs(npages, pp_rnb, remote_nb);
497 req->rq_status = nob;
499 } else if (!comms_error) {
500 /* only reply if comms OK */
504 if (req->rq_repmsg != NULL) {
505 /* reply out callback would free */
506 OBD_FREE(req->rq_repmsg, req->rq_replen);
508 if (req->rq_reqmsg->conn_cnt == req->rq_export->exp_conn_cnt) {
509 CERROR("bulk IO comms error: "
510 "evicting %s@%s nid "LPX64" (%s)\n",
511 req->rq_export->exp_client_uuid.uuid,
512 req->rq_connection->c_remote_uuid.uuid,
513 req->rq_connection->c_peer.peer_nid,
514 portals_nid2str(req->rq_connection->c_peer.peer_ni->pni_number,
515 req->rq_connection->c_peer.peer_nid,
517 ptlrpc_fail_export(req->rq_export);
519 CERROR("ignoring bulk IO comms error: "
520 "client reconnected %s@%s nid "LPX64" (%s)\n",
521 req->rq_export->exp_client_uuid.uuid,
522 req->rq_connection->c_remote_uuid.uuid,
523 req->rq_connection->c_peer.peer_nid,
524 portals_nid2str(req->rq_connection->c_peer.peer_ni->pni_number,
525 req->rq_connection->c_peer.peer_nid,
533 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
535 struct ptlrpc_bulk_desc *desc;
536 struct niobuf_remote *remote_nb;
537 struct niobuf_remote *pp_rnb;
538 struct niobuf_local *local_nb;
539 struct obd_ioobj *ioo;
540 struct ost_body *body, *repbody;
541 struct l_wait_info lwi;
543 int size[2] = { sizeof(*body) };
544 int objcount, niocount, npages;
546 int rc, rc2, swab, i, j;
547 char str[PTL_NALFMT_SIZE];
550 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
551 GOTO(out, rc = -EIO);
553 /* pause before transaction has been started */
554 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK | OBD_FAIL_ONCE,
555 (obd_timeout + 1) / 4);
557 swab = lustre_msg_swabbed(req->rq_reqmsg);
558 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
560 CERROR("Missing/short ost_body\n");
561 GOTO(out, rc = -EFAULT);
564 LASSERT_REQSWAB(req, 1);
565 objcount = req->rq_reqmsg->buflens[1] / sizeof(*ioo);
567 CERROR("Missing/short ioobj\n");
568 GOTO(out, rc = -EFAULT);
570 ioo = lustre_msg_buf (req->rq_reqmsg, 1, objcount * sizeof(*ioo));
571 LASSERT (ioo != NULL);
572 for (niocount = i = 0; i < objcount; i++) {
574 lustre_swab_obd_ioobj (&ioo[i]);
575 if (ioo[i].ioo_bufcnt == 0) {
576 CERROR("ioo[%d] has zero bufcnt\n", i);
577 GOTO(out, rc = -EFAULT);
579 niocount += ioo[i].ioo_bufcnt;
582 remote_nb = lustre_swab_reqbuf(req, 2, niocount * sizeof(*remote_nb),
583 lustre_swab_niobuf_remote);
584 if (remote_nb == NULL) {
585 CERROR("Missing/short niobuf\n");
586 GOTO(out, rc = -EFAULT);
588 if (swab) { /* swab the remaining niobufs */
589 for (i = 1; i < niocount; i++)
590 lustre_swab_niobuf_remote (&remote_nb[i]);
593 size[1] = niocount * sizeof(*rcs);
594 rc = lustre_pack_reply(req, 2, size, NULL);
597 rcs = lustre_msg_buf(req->rq_repmsg, 1, niocount * sizeof(*rcs));
599 /* FIXME all niobuf splitting should be done in obdfilter if needed */
600 /* CAVEAT EMPTOR this sets ioo->ioo_bufcnt to # pages */
601 npages = get_per_page_niobufs(ioo, objcount,remote_nb,niocount,&pp_rnb);
603 GOTO(out, rc = npages);
605 OBD_ALLOC(local_nb, sizeof(*local_nb) * npages);
606 if (local_nb == NULL)
607 GOTO(out_pp_rnb, rc = -ENOMEM);
609 desc = ptlrpc_prep_bulk_exp(req, BULK_GET_SINK, OST_BULK_PORTAL);
611 GOTO(out_local, rc = -ENOMEM);
613 rc = obd_preprw(OBD_BRW_WRITE, req->rq_export, &body->oa, objcount,
614 ioo, npages, pp_rnb, local_nb, oti);
618 /* NB Having prepped, we must commit... */
620 for (i = 0; i < npages; i++) {
621 rc = ptlrpc_prep_bulk_page(desc, local_nb[i].page,
622 pp_rnb[i].offset & (PAGE_SIZE - 1),
629 rc = ptlrpc_bulk_get(desc);
631 lwi = LWI_TIMEOUT(obd_timeout * HZ / 4,
632 ost_bulk_timeout, desc);
633 rc = l_wait_event(desc->bd_waitq,
634 ptlrpc_bulk_complete(desc), &lwi);
636 LASSERT(rc == -ETIMEDOUT);
637 DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
638 ptlrpc_abort_bulk(desc);
641 DEBUG_REQ(D_ERROR, req, "bulk GET failed: rc %d\n", rc);
643 comms_error = rc != 0;
646 repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
647 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
650 if (rc == 0 && (body->oa.o_valid & OBD_MD_FLCKSUM) != 0) {
651 static int cksum_counter;
652 obd_count client_cksum = body->oa.o_nlink;
653 obd_count cksum = ost_checksum_bulk(desc);
655 portals_nid2str(req->rq_connection->c_peer.peer_ni->pni_number,
656 req->rq_connection->c_peer.peer_nid, str);
657 if (client_cksum != cksum) {
658 CERROR("Bad checksum: client %x, server %x, client NID "
659 LPX64" (%s)\n", client_cksum, cksum,
660 req->rq_connection->c_peer.peer_nid, str);
662 repbody->oa.o_nlink = cksum;
665 if ((cksum_counter & (-cksum_counter)) == cksum_counter)
666 CWARN("Checksum %u from "LPX64": %x OK\n",
668 req->rq_connection->c_peer.peer_nid,
673 /* Must commit after prep above in all cases */
674 rc2 = obd_commitrw(OBD_BRW_WRITE, req->rq_export, &repbody->oa,
675 objcount, ioo, npages, local_nb, oti);
678 /* set per-requested niobuf return codes */
679 for (i = j = 0; i < niocount; i++) {
680 int nob = remote_nb[i].len;
685 if (local_nb[j].rc < 0)
686 rcs[i] = local_nb[j].rc;
687 nob -= pp_rnb[j].len;
692 LASSERT(j == npages);
698 ptlrpc_free_bulk(desc);
700 OBD_FREE(local_nb, sizeof(*local_nb) * npages);
702 free_per_page_niobufs(npages, pp_rnb, remote_nb);
705 oti_to_request(oti, req);
706 rc = ptlrpc_reply(req);
707 } else if (!comms_error) {
708 /* Only reply if there was no comms problem with bulk */
712 if (req->rq_repmsg != NULL) {
713 /* reply out callback would free */
714 OBD_FREE (req->rq_repmsg, req->rq_replen);
716 if (req->rq_reqmsg->conn_cnt == req->rq_export->exp_conn_cnt) {
717 CERROR("bulk IO comms error: "
718 "evicting %s@%s nid "LPX64" (%s)\n",
719 req->rq_export->exp_client_uuid.uuid,
720 req->rq_connection->c_remote_uuid.uuid,
721 req->rq_connection->c_peer.peer_nid,
722 portals_nid2str(req->rq_connection->c_peer.peer_ni->pni_number,
723 req->rq_connection->c_peer.peer_nid,
725 ptlrpc_fail_export(req->rq_export);
727 CERROR("ignoring bulk IO comms error: "
728 "client reconnected %s@%s nid "LPX64" (%s)\n",
729 req->rq_export->exp_client_uuid.uuid,
730 req->rq_connection->c_remote_uuid.uuid,
731 req->rq_connection->c_peer.peer_nid,
732 portals_nid2str(req->rq_connection->c_peer.peer_ni->pni_number,
733 req->rq_connection->c_peer.peer_nid,
740 static int ost_san_brw(struct ptlrpc_request *req, int cmd)
742 struct niobuf_remote *remote_nb, *res_nb, *pp_rnb;
743 struct obd_ioobj *ioo;
744 struct ost_body *body, *repbody;
745 int rc, i, objcount, niocount, size[2] = {sizeof(*body)}, npages;
749 /* XXX not set to use latest protocol */
751 swab = lustre_msg_swabbed(req->rq_reqmsg);
752 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
754 CERROR("Missing/short ost_body\n");
755 GOTO(out, rc = -EFAULT);
758 ioo = lustre_swab_reqbuf(req, 1, sizeof(*ioo), lustre_swab_obd_ioobj);
760 CERROR("Missing/short ioobj\n");
761 GOTO(out, rc = -EFAULT);
763 objcount = req->rq_reqmsg->buflens[1] / sizeof(*ioo);
764 niocount = ioo[0].ioo_bufcnt;
765 for (i = 1; i < objcount; i++) {
767 lustre_swab_obd_ioobj (&ioo[i]);
768 niocount += ioo[i].ioo_bufcnt;
771 remote_nb = lustre_swab_reqbuf(req, 2, niocount * sizeof(*remote_nb),
772 lustre_swab_niobuf_remote);
773 if (remote_nb == NULL) {
774 CERROR("Missing/short niobuf\n");
775 GOTO(out, rc = -EFAULT);
777 if (swab) { /* swab the remaining niobufs */
778 for (i = 1; i < niocount; i++)
779 lustre_swab_niobuf_remote (&remote_nb[i]);
782 /* CAVEAT EMPTOR this sets ioo->ioo_bufcnt to # pages */
783 npages = get_per_page_niobufs(ioo, objcount,remote_nb,niocount,&pp_rnb);
785 GOTO (out, rc = npages);
787 size[1] = npages * sizeof(*pp_rnb);
788 rc = lustre_pack_reply(req, 2, size, NULL);
790 GOTO(out_pp_rnb, rc);
792 req->rq_status = obd_san_preprw(cmd, req->rq_export, &body->oa,
793 objcount, ioo, npages, pp_rnb);
796 GOTO(out_pp_rnb, rc = 0);
798 repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
799 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
801 res_nb = lustre_msg_buf(req->rq_repmsg, 1, size[1]);
802 memcpy(res_nb, remote_nb, size[1]);
805 free_per_page_niobufs(npages, pp_rnb, remote_nb);
808 OBD_FREE(req->rq_repmsg, req->rq_replen);
809 req->rq_repmsg = NULL;
819 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
825 key = lustre_msg_buf(req->rq_reqmsg, 0, 1);
827 DEBUG_REQ(D_HA, req, "no set_info key");
830 keylen = req->rq_reqmsg->buflens[0];
832 rc = lustre_pack_reply(req, 0, NULL, NULL);
836 rc = obd_set_info(exp, keylen, key, 0, NULL);
837 req->rq_repmsg->status = 0;
841 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
844 int keylen, rc = 0, size = sizeof(obd_id);
848 key = lustre_msg_buf(req->rq_reqmsg, 0, 1);
850 DEBUG_REQ(D_HA, req, "no get_info key");
853 keylen = req->rq_reqmsg->buflens[0];
855 if (keylen < strlen("last_id") || memcmp(key, "last_id", 7) != 0)
858 rc = lustre_pack_reply(req, 1, &size, NULL);
862 reply = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*reply));
863 rc = obd_get_info(exp, keylen, key, &size, reply);
864 req->rq_repmsg->status = 0;
868 static int ost_filter_recovery_request(struct ptlrpc_request *req,
869 struct obd_device *obd, int *process)
871 switch (req->rq_reqmsg->opc) {
872 case OST_CONNECT: /* This will never get here, but for completeness. */
886 *process = target_queue_recovery_request(req, obd);
890 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
892 /* XXX what should we set rq_status to here? */
893 req->rq_status = -EAGAIN;
894 RETURN(ptlrpc_error(req));
900 static int ost_handle(struct ptlrpc_request *req)
902 struct obd_trans_info trans_info = { 0, };
903 struct obd_trans_info *oti = &trans_info;
904 int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
905 struct obd_export *exp = NULL;
908 LASSERT(current->journal_info == NULL);
909 /* XXX identical to MDS */
910 if (req->rq_reqmsg->opc != OST_CONNECT) {
911 struct obd_device *obd;
912 int abort_recovery, recovering;
914 exp = req->rq_export;
917 CDEBUG(D_HA, "operation %d on unconnected OST\n",
918 req->rq_reqmsg->opc);
919 req->rq_status = -ENOTCONN;
920 GOTO(out, rc = -ENOTCONN);
925 /* Check for aborted recovery. */
926 spin_lock_bh(&obd->obd_processing_task_lock);
927 abort_recovery = obd->obd_abort_recovery;
928 recovering = obd->obd_recovering;
929 spin_unlock_bh(&obd->obd_processing_task_lock);
930 if (abort_recovery) {
931 target_abort_recovery(obd);
932 } else if (recovering) {
933 rc = ost_filter_recovery_request(req, obd,
935 if (rc || !should_process)
942 switch (req->rq_reqmsg->opc) {
944 CDEBUG(D_INODE, "connect\n");
945 OBD_FAIL_RETURN(OBD_FAIL_OST_CONNECT_NET, 0);
946 rc = target_handle_connect(req, ost_handle);
949 CDEBUG(D_INODE, "disconnect\n");
950 OBD_FAIL_RETURN(OBD_FAIL_OST_DISCONNECT_NET, 0);
951 rc = target_handle_disconnect(req);
954 CDEBUG(D_INODE, "create\n");
955 OBD_FAIL_RETURN(OBD_FAIL_OST_CREATE_NET, 0);
956 rc = ost_create(exp, req, oti);
959 CDEBUG(D_INODE, "destroy\n");
960 OBD_FAIL_RETURN(OBD_FAIL_OST_DESTROY_NET, 0);
961 rc = ost_destroy(exp, req, oti);
964 CDEBUG(D_INODE, "getattr\n");
965 OBD_FAIL_RETURN(OBD_FAIL_OST_GETATTR_NET, 0);
966 rc = ost_getattr(exp, req);
969 CDEBUG(D_INODE, "setattr\n");
970 OBD_FAIL_RETURN(OBD_FAIL_OST_SETATTR_NET, 0);
971 rc = ost_setattr(exp, req, oti);
974 CDEBUG(D_INODE, "write\n");
975 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
976 rc = ost_brw_write(req, oti);
977 LASSERT(current->journal_info == NULL);
978 /* ost_brw sends its own replies */
981 CDEBUG(D_INODE, "read\n");
982 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
983 rc = ost_brw_read(req);
984 LASSERT(current->journal_info == NULL);
985 /* ost_brw sends its own replies */
988 CDEBUG(D_INODE, "san read\n");
989 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
990 rc = ost_san_brw(req, OBD_BRW_READ);
991 /* ost_san_brw sends its own replies */
994 CDEBUG(D_INODE, "san write\n");
995 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
996 rc = ost_san_brw(req, OBD_BRW_WRITE);
997 /* ost_san_brw sends its own replies */
1000 CDEBUG(D_INODE, "punch\n");
1001 OBD_FAIL_RETURN(OBD_FAIL_OST_PUNCH_NET, 0);
1002 rc = ost_punch(exp, req, oti);
1005 CDEBUG(D_INODE, "statfs\n");
1006 OBD_FAIL_RETURN(OBD_FAIL_OST_STATFS_NET, 0);
1007 rc = ost_statfs(req);
1010 CDEBUG(D_INODE, "sync\n");
1011 OBD_FAIL_RETURN(OBD_FAIL_OST_SYNC_NET, 0);
1012 rc = ost_sync(exp, req);
1015 DEBUG_REQ(D_INODE, req, "set_info");
1016 rc = ost_set_info(exp, req);
1019 DEBUG_REQ(D_INODE, req, "get_info");
1020 rc = ost_get_info(exp, req);
1023 DEBUG_REQ(D_INODE, req, "ping");
1024 rc = target_handle_ping(req);
1026 #ifdef ENABLE_ORPHANS
1027 /* FIXME - just reply status */
1028 case LLOG_ORIGIN_CONNECT:
1029 DEBUG_REQ(D_INODE, req, "log connect\n");
1030 rc = llog_handle_connect(req);
1031 req->rq_status = rc;
1032 rc = lustre_pack_reply(req, 0, NULL, NULL);
1035 RETURN(ptlrpc_reply(req));
1037 case OBD_LOG_CANCEL:
1038 CDEBUG(D_INODE, "log cancel\n");
1039 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOG_CANCEL_NET, 0);
1040 rc = llog_origin_handle_cancel(req);
1041 req->rq_status = rc;
1042 rc = lustre_pack_reply(req, 0, NULL, NULL);
1045 RETURN(ptlrpc_reply(req));
1049 CDEBUG(D_INODE, "enqueue\n");
1050 OBD_FAIL_RETURN(OBD_FAIL_LDLM_ENQUEUE, 0);
1051 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
1052 ldlm_server_blocking_ast);
1053 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
1056 CDEBUG(D_INODE, "convert\n");
1057 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CONVERT, 0);
1058 rc = ldlm_handle_convert(req);
1061 CDEBUG(D_INODE, "cancel\n");
1062 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CANCEL, 0);
1063 rc = ldlm_handle_cancel(req);
1065 case LDLM_BL_CALLBACK:
1066 case LDLM_CP_CALLBACK:
1067 CDEBUG(D_INODE, "callback\n");
1068 CERROR("callbacks should not happen on OST\n");
1071 CERROR("Unexpected opcode %d\n", req->rq_reqmsg->opc);
1072 req->rq_status = -ENOTSUPP;
1073 rc = ptlrpc_error(req);
1077 LASSERT(current->journal_info == NULL);
1080 /* If we're DISCONNECTing, the export_data is already freed */
1081 if (!rc && req->rq_reqmsg->opc != OST_DISCONNECT) {
1082 struct obd_device *obd = req->rq_export->exp_obd;
1083 if (!obd->obd_no_transno) {
1084 req->rq_repmsg->last_committed =
1085 obd->obd_last_committed;
1087 DEBUG_REQ(D_IOCTL, req,
1088 "not sending last_committed update");
1090 CDEBUG(D_INFO, "last_committed "LPU64", xid "LPX64"\n",
1091 obd->obd_last_committed, req->rq_xid);
1095 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LAST_REPLAY) {
1096 struct obd_device *obd = req->rq_export->exp_obd;
1098 if (obd && obd->obd_recovering) {
1099 DEBUG_REQ(D_HA, req, "LAST_REPLAY, queuing reply");
1100 return target_queue_final_reply(req, rc);
1102 /* Lost a race with recovery; let the error path DTRT. */
1103 rc = req->rq_status = -ENOTCONN;
1107 oti_to_request(oti, req);
1109 target_send_reply(req, rc, fail);
1113 static int ost_setup(struct obd_device *obddev, obd_count len, void *buf)
1115 struct ost_obd *ost = &obddev->u.ost;
1119 /* Get rid of unneeded supplementary groups */
1120 current->ngroups = 0;
1121 memset(current->groups, 0, sizeof(current->groups));
1123 rc = llog_start_commit_thread();
1127 ost->ost_service = ptlrpc_init_svc(OST_NEVENTS, OST_NBUFS,
1128 OST_BUFSIZE, OST_MAXREQSIZE,
1129 OST_REQUEST_PORTAL, OSC_REPLY_PORTAL,
1131 obddev->obd_proc_entry);
1132 if (ost->ost_service == NULL) {
1133 CERROR("failed to start service\n");
1137 rc = ptlrpc_start_n_threads(obddev, ost->ost_service, OST_NUM_THREADS,
1140 GOTO(out, rc = -EINVAL);
1142 ost->ost_create_service =
1143 ptlrpc_init_svc(OST_NEVENTS, OST_NBUFS, OST_BUFSIZE,
1144 OST_MAXREQSIZE, OST_CREATE_PORTAL,
1145 OSC_REPLY_PORTAL, ost_handle, "ost_create",
1146 obddev->obd_proc_entry);
1147 if (ost->ost_create_service == NULL) {
1148 CERROR("failed to start OST create service\n");
1149 GOTO(out, rc = -ENOMEM);
1152 rc = ptlrpc_start_n_threads(obddev, ost->ost_create_service, 1,
1155 GOTO(out_create, rc = -EINVAL);
1160 ptlrpc_unregister_service(ost->ost_create_service);
1162 ptlrpc_unregister_service(ost->ost_service);
1166 static int ost_cleanup(struct obd_device *obddev, int flags)
1168 struct ost_obd *ost = &obddev->u.ost;
1172 if (obddev->obd_recovering)
1173 target_cancel_recovery_timer(obddev);
1175 ptlrpc_stop_all_threads(ost->ost_service);
1176 ptlrpc_unregister_service(ost->ost_service);
1178 ptlrpc_stop_all_threads(ost->ost_create_service);
1179 ptlrpc_unregister_service(ost->ost_create_service);
1184 int ost_attach(struct obd_device *dev, obd_count len, void *data)
1186 struct lprocfs_static_vars lvars;
1188 lprocfs_init_vars(ost,&lvars);
1189 return lprocfs_obd_attach(dev, lvars.obd_vars);
1192 int ost_detach(struct obd_device *dev)
1194 return lprocfs_obd_detach(dev);
1197 /* use obd ops to offer management infrastructure */
1198 static struct obd_ops ost_obd_ops = {
1199 o_owner: THIS_MODULE,
1200 o_attach: ost_attach,
1201 o_detach: ost_detach,
1203 o_cleanup: ost_cleanup,
1206 static int __init ost_init(void)
1208 struct lprocfs_static_vars lvars;
1211 lprocfs_init_vars(ost,&lvars);
1212 RETURN(class_register_type(&ost_obd_ops, lvars.module_vars,
1216 static void /*__exit*/ ost_exit(void)
1218 class_unregister_type(LUSTRE_OST_NAME);
1221 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1222 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
1223 MODULE_LICENSE("GPL");
1225 module_init(ost_init);
1226 module_exit(ost_exit);