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 DEBUG_SUBSYSTEM S_OST
36 #include <linux/module.h>
37 #include <linux/obd_ost.h>
38 #include <linux/lustre_net.h>
39 #include <linux/lustre_dlm.h>
40 #include <linux/lustre_export.h>
41 #include <linux/init.h>
42 #include <linux/lprocfs_status.h>
43 #include <linux/lustre_commit_confd.h>
44 #include <portals/list.h>
46 void oti_init(struct obd_trans_info *oti, struct ptlrpc_request *req)
50 memset(oti, 0, sizeof *oti);
52 if (req->rq_repmsg && req->rq_reqmsg != 0)
53 oti->oti_transno = req->rq_repmsg->transno;
56 void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req)
58 struct oti_req_ack_lock *ack_lock;
65 req->rq_repmsg->transno = oti->oti_transno;
67 /* XXX 4 == entries in oti_ack_locks??? */
68 for (ack_lock = oti->oti_ack_locks, i = 0; i < 4; i++, ack_lock++) {
71 memcpy(&req->rq_ack_locks[i].lock, &ack_lock->lock,
72 sizeof(req->rq_ack_locks[i].lock));
73 req->rq_ack_locks[i].mode = ack_lock->mode;
77 static int ost_destroy(struct ptlrpc_request *req, struct obd_trans_info *oti)
79 struct lustre_handle *conn = &req->rq_reqmsg->handle;
80 struct ost_body *body;
81 int rc, size = sizeof(*body);
84 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
88 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
92 if (body->oa.o_valid & OBD_MD_FLCOOKIE)
93 oti->oti_logcookies = obdo_logcookie(&body->oa);
94 req->rq_status = obd_destroy(conn, &body->oa, NULL, oti);
98 static int ost_getattr(struct ptlrpc_request *req)
100 struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
101 struct ost_body *body, *repbody;
102 int rc, size = sizeof(*body);
105 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
109 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
113 repbody = lustre_msg_buf (req->rq_repmsg, 0, sizeof(*repbody));
114 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
115 req->rq_status = obd_getattr(conn, &repbody->oa, NULL);
119 static int ost_statfs(struct ptlrpc_request *req)
121 struct obd_statfs *osfs;
122 int rc, size = sizeof(*osfs);
125 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
129 osfs = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*osfs));
131 req->rq_status = obd_statfs(req->rq_export->exp_obd, osfs, jiffies-HZ);
132 if (req->rq_status != 0)
133 CERROR("ost: statfs failed: rc %d\n", req->rq_status);
138 static int ost_syncfs(struct ptlrpc_request *req)
140 struct obd_statfs *osfs;
141 int rc, size = sizeof(*osfs);
144 rc = lustre_pack_msg(0, &size, NULL, &req->rq_replen, &req->rq_repmsg);
148 rc = obd_syncfs(req->rq_export);
150 CERROR("ost: syncfs failed: rc %d\n", rc);
158 static int ost_open(struct ptlrpc_request *req, struct obd_trans_info *oti)
160 struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
161 struct ost_body *body, *repbody;
162 int rc, size = sizeof(*repbody);
165 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
169 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
173 repbody = lustre_msg_buf (req->rq_repmsg, 0, sizeof(*repbody));
174 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
175 req->rq_status = obd_open(conn, &repbody->oa, NULL, oti, NULL);
179 static int ost_close(struct ptlrpc_request *req, struct obd_trans_info *oti)
181 struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
182 struct ost_body *body, *repbody;
183 int rc, size = sizeof(*repbody);
186 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
190 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
194 repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
195 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
196 req->rq_status = obd_close(conn, &repbody->oa, NULL, oti);
200 static int ost_create(struct ptlrpc_request *req, struct obd_trans_info *oti)
202 struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
203 struct ost_body *body, *repbody;
204 int rc, size = sizeof(*repbody);
207 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
211 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
215 repbody = lustre_msg_buf (req->rq_repmsg, 0, sizeof(*repbody));
216 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
217 oti->oti_logcookies = obdo_logcookie(&repbody->oa);
218 req->rq_status = obd_create(conn, &repbody->oa, NULL, oti);
219 //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
223 static int ost_punch(struct ptlrpc_request *req, struct obd_trans_info *oti)
225 struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
226 struct ost_body *body, *repbody;
227 int rc, size = sizeof(*repbody);
230 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
234 if ((body->oa.o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
235 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
238 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
242 repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
243 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
244 req->rq_status = obd_punch(conn, &repbody->oa, NULL, repbody->oa.o_size,
245 repbody->oa.o_blocks, oti);
249 static int ost_setattr(struct ptlrpc_request *req, struct obd_trans_info *oti)
251 struct lustre_handle *conn = &req->rq_reqmsg->handle;
252 struct ost_body *body, *repbody;
253 int rc, size = sizeof(*repbody);
256 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
260 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
264 repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
265 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
267 req->rq_status = obd_setattr(conn, &repbody->oa, NULL, oti);
271 static int ost_bulk_timeout(void *data)
274 /* We don't fail the connection here, because having the export
275 * killed makes the (vital) call to commitrw very sad.
280 static int get_per_page_niobufs(struct obd_ioobj *ioo, int nioo,
281 struct niobuf_remote *rnb, int nrnb,
282 struct niobuf_remote **pp_rnbp)
284 /* Copy a remote niobuf, splitting it into page-sized chunks
285 * and setting ioo[i].ioo_bufcnt accordingly */
286 struct niobuf_remote *pp_rnb;
293 /* first count and check the number of pages required */
294 for (i = 0; i < nioo; i++)
295 for (j = 0; j < ioo->ioo_bufcnt; j++, rnbidx++) {
296 obd_off offset = rnb[rnbidx].offset;
297 obd_off p0 = offset >> PAGE_SHIFT;
298 obd_off pn = (offset + rnb[rnbidx].len - 1)>>PAGE_SHIFT;
300 LASSERT(rnbidx < nrnb);
302 npages += (pn + 1 - p0);
304 if (rnb[rnbidx].len == 0) {
305 CERROR("zero len BRW: obj %d objid "LPX64
306 " buf %u\n", i, ioo[i].ioo_id, j);
310 rnb[rnbidx].offset <= rnb[rnbidx-1].offset) {
311 CERROR("unordered BRW: obj %d objid "LPX64
312 " buf %u offset "LPX64" <= "LPX64"\n",
313 i, ioo[i].ioo_id, j, rnb[rnbidx].offset,
319 LASSERT(rnbidx == nrnb);
321 if (npages == nrnb) { /* all niobufs are for single pages */
326 OBD_ALLOC(pp_rnb, sizeof(*pp_rnb) * npages);
330 /* now do the actual split */
332 for (i = 0; i < nioo; i++) {
335 for (j = 0; j < ioo[i].ioo_bufcnt; j++, rnbidx++) {
336 obd_off off = rnb[rnbidx].offset;
337 int nob = rnb[rnbidx].len;
339 LASSERT(rnbidx < nrnb);
341 obd_off poff = off & (PAGE_SIZE - 1);
342 int pnob = (poff + nob > PAGE_SIZE) ?
343 PAGE_SIZE - poff : nob;
345 LASSERT(page < npages);
346 pp_rnb[page].len = pnob;
347 pp_rnb[page].offset = off;
348 pp_rnb[page].flags = rnb->flags;
350 CDEBUG(D_PAGE, " obj %d id "LPX64
351 "page %d(%d) "LPX64" for %d\n",
352 i, ioo[i].ioo_id, obj_pages, page,
353 pp_rnb[page].offset, pp_rnb[page].len);
362 ioo[i].ioo_bufcnt = obj_pages;
364 LASSERT(page == npages);
370 static void free_per_page_niobufs (int npages, struct niobuf_remote *pp_rnb,
371 struct niobuf_remote *rnb)
373 if (pp_rnb == rnb) /* didn't allocate above */
376 OBD_FREE(pp_rnb, sizeof(*pp_rnb) * npages);
380 __u64 ost_checksum_bulk (struct ptlrpc_bulk_desc *desc)
383 struct ptlrpc_bulk_page *bp;
385 list_for_each_entry(bp, &desc->bd_page_list, bp_link) {
386 ost_checksum(&cksum, kmap(bp->bp_page) + bp->bp_pageoffset,
393 static int ost_brw_read(struct ptlrpc_request *req)
395 struct ptlrpc_bulk_desc *desc;
396 struct niobuf_remote *remote_nb;
397 struct niobuf_remote *pp_rnb;
398 struct niobuf_local *local_nb;
399 struct obd_ioobj *ioo;
400 struct ost_body *body, *repbody;
401 struct l_wait_info lwi;
402 struct obd_trans_info oti = { 0 };
403 int size[1] = { sizeof(*body) };
412 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
413 GOTO(out, rc = -EIO);
415 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
417 CERROR("Missing/short ost_body\n");
418 GOTO(out, rc = -EFAULT);
421 /* BUG 974: when we send back cache grants, don't clear this flag */
422 body->oa.o_valid &= ~OBD_MD_FLRDEV;
424 ioo = lustre_swab_reqbuf(req, 1, sizeof(*ioo), lustre_swab_obd_ioobj);
426 CERROR("Missing/short ioobj\n");
427 GOTO(out, rc = -EFAULT);
430 niocount = ioo->ioo_bufcnt;
431 remote_nb = lustre_swab_reqbuf(req, 2, niocount * sizeof(*remote_nb),
432 lustre_swab_niobuf_remote);
433 if (remote_nb == NULL) {
434 CERROR("Missing/short niobuf\n");
435 GOTO(out, rc = -EFAULT);
437 if (lustre_msg_swabbed(req->rq_reqmsg)) { /* swab remaining niobufs */
438 for (i = 1; i < niocount; i++)
439 lustre_swab_niobuf_remote (&remote_nb[i]);
442 size[0] = sizeof(*body);
443 rc = lustre_pack_msg(1, size, NULL, &req->rq_replen, &req->rq_repmsg);
447 /* FIXME all niobuf splitting should be done in obdfilter if needed */
448 /* CAVEAT EMPTOR this sets ioo->ioo_bufcnt to # pages */
449 npages = get_per_page_niobufs(ioo, 1, remote_nb, niocount, &pp_rnb);
451 GOTO(out, rc = npages);
453 OBD_ALLOC(local_nb, sizeof(*local_nb) * npages);
454 if (local_nb == NULL)
455 GOTO(out_pp_rnb, rc = -ENOMEM);
457 desc = ptlrpc_prep_bulk_exp(req, BULK_PUT_SOURCE, OST_BULK_PORTAL);
459 GOTO(out_local, rc = -ENOMEM);
461 rc = obd_preprw(OBD_BRW_READ, req->rq_export, &body->oa, 1,
462 ioo, npages, pp_rnb, local_nb, &oti);
467 for (i = 0; i < npages; i++) {
468 int page_rc = local_nb[i].rc;
470 if (page_rc < 0) { /* error */
475 LASSERT(page_rc <= pp_rnb[i].len);
477 if (page_rc != 0) { /* some data! */
478 LASSERT (local_nb[i].page != NULL);
479 rc = ptlrpc_prep_bulk_page(desc, local_nb[i].page,
480 pp_rnb[i].offset& ~PAGE_MASK,
486 if (page_rc != pp_rnb[i].len) { /* short read */
487 /* All subsequent pages should be 0 */
489 LASSERT(local_nb[i].rc == 0);
495 rc = ptlrpc_bulk_put(desc);
497 lwi = LWI_TIMEOUT(obd_timeout * HZ, ost_bulk_timeout,
499 rc = l_wait_event(desc->bd_waitq,
500 ptlrpc_bulk_complete(desc), &lwi);
502 LASSERT(rc == -ETIMEDOUT);
503 CERROR ("timeout waiting for bulk PUT\n");
504 ptlrpc_abort_bulk(desc);
507 CERROR("ptlrpc_bulk_put failed RC: %d\n", rc);
509 comms_error = rc != 0;
512 /* Must commit after prep above in all cases */
513 rc = obd_commitrw(OBD_BRW_READ, req->rq_export, &body->oa, 1,
514 ioo, npages, local_nb, &oti);
516 repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
517 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
521 repbody->oa.o_rdev = ost_checksum_bulk(desc);
522 repbody->oa.o_valid |= OBD_MD_FLCKSUM;
527 ptlrpc_free_bulk(desc);
529 OBD_FREE(local_nb, sizeof(*local_nb) * npages);
531 free_per_page_niobufs(npages, pp_rnb, remote_nb);
535 req->rq_status = nob;
537 } else if (!comms_error) {
538 /* only reply if comms OK */
542 if (req->rq_repmsg != NULL) {
543 /* reply out callback would free */
544 OBD_FREE(req->rq_repmsg, req->rq_replen);
546 CERROR("bulk IO comms error: evicting %s@%s nid "LPU64"\n",
547 req->rq_export->exp_client_uuid.uuid,
548 req->rq_connection->c_remote_uuid.uuid,
549 req->rq_connection->c_peer.peer_nid);
550 ptlrpc_fail_export(req->rq_export);
556 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
558 struct ptlrpc_bulk_desc *desc;
559 struct niobuf_remote *remote_nb;
560 struct niobuf_remote *pp_rnb;
561 struct niobuf_local *local_nb;
562 struct obd_ioobj *ioo;
563 struct ost_body *body, *repbody;
564 struct l_wait_info lwi;
566 int size[2] = { sizeof(*body) };
567 int objcount, niocount, npages;
569 int rc, rc2, swab, i, j;
572 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
573 GOTO(out, rc = -EIO);
575 /* pause before transaction has been started */
576 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK | OBD_FAIL_ONCE,
579 swab = lustre_msg_swabbed(req->rq_reqmsg);
580 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
582 CERROR("Missing/short ost_body\n");
583 GOTO(out, rc = -EFAULT);
586 /* BUG 974: when we send back cache grants, don't clear this flag */
587 body->oa.o_valid &= ~OBD_MD_FLRDEV;
589 LASSERT_REQSWAB(req, 1);
590 objcount = req->rq_reqmsg->buflens[1] / sizeof(*ioo);
592 CERROR("Missing/short ioobj\n");
593 GOTO(out, rc = -EFAULT);
595 ioo = lustre_msg_buf (req->rq_reqmsg, 1, objcount * sizeof(*ioo));
596 LASSERT (ioo != NULL);
597 for (niocount = i = 0; i < objcount; i++) {
599 lustre_swab_obd_ioobj (&ioo[i]);
600 if (ioo[i].ioo_bufcnt == 0) {
601 CERROR("ioo[%d] has zero bufcnt\n", i);
602 GOTO(out, rc = -EFAULT);
604 niocount += ioo[i].ioo_bufcnt;
607 remote_nb = lustre_swab_reqbuf(req, 2, niocount * sizeof(*remote_nb),
608 lustre_swab_niobuf_remote);
609 if (remote_nb == NULL) {
610 CERROR("Missing/short niobuf\n");
611 GOTO(out, rc = -EFAULT);
613 if (swab) { /* swab the remaining niobufs */
614 for (i = 1; i < niocount; i++)
615 lustre_swab_niobuf_remote (&remote_nb[i]);
618 size[1] = niocount * sizeof(*rcs);
619 rc = lustre_pack_msg(2, size, NULL, &req->rq_replen,
623 rcs = lustre_msg_buf(req->rq_repmsg, 1, niocount * sizeof(*rcs));
625 /* FIXME all niobuf splitting should be done in obdfilter if needed */
626 /* CAVEAT EMPTOR this sets ioo->ioo_bufcnt to # pages */
627 npages = get_per_page_niobufs(ioo, objcount,remote_nb,niocount,&pp_rnb);
629 GOTO(out, rc = npages);
631 OBD_ALLOC(local_nb, sizeof(*local_nb) * npages);
632 if (local_nb == NULL)
633 GOTO(out_pp_rnb, rc = -ENOMEM);
635 desc = ptlrpc_prep_bulk_exp(req, BULK_GET_SINK, OST_BULK_PORTAL);
637 GOTO(out_local, rc = -ENOMEM);
639 rc = obd_preprw(OBD_BRW_WRITE, req->rq_export, &body->oa, objcount,
640 ioo, npages, pp_rnb, local_nb, oti);
644 /* NB Having prepped, we must commit... */
646 for (i = 0; i < npages; i++) {
647 rc = ptlrpc_prep_bulk_page(desc, local_nb[i].page,
648 pp_rnb[i].offset & (PAGE_SIZE - 1),
655 rc = ptlrpc_bulk_get(desc);
657 lwi = LWI_TIMEOUT(obd_timeout * HZ, ost_bulk_timeout,
659 rc = l_wait_event(desc->bd_waitq,
660 ptlrpc_bulk_complete(desc), &lwi);
662 LASSERT(rc == -ETIMEDOUT);
663 CERROR("timeout waiting for bulk GET\n");
664 ptlrpc_abort_bulk(desc);
667 CERROR("ptlrpc_bulk_get failed RC: %d\n", rc);
669 comms_error = rc != 0;
672 repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
673 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
676 if (rc == 0 && (body->oa.o_valid & OBD_MD_FLCKSUM) != 0) {
677 static int cksum_counter;
678 __u64 client_cksum = body->oa.o_rdev;
679 __u64 cksum = ost_checksum_bulk(desc);
681 if (client_cksum != cksum) {
682 CERROR("Bad checksum: client "LPX64", server "LPX64
683 ", client NID "LPX64"\n", client_cksum, cksum,
684 req->rq_connection->c_peer.peer_nid);
686 repbody->oa.o_rdev = cksum;
689 if ((cksum_counter & (-cksum_counter)) == cksum_counter)
690 CERROR("Checksum %d from "LPX64": "LPX64" OK\n",
692 req->rq_connection->c_peer.peer_nid,
697 /* Must commit after prep above in all cases */
698 rc2 = obd_commitrw(OBD_BRW_WRITE, req->rq_export, &repbody->oa,
699 objcount, ioo, npages, local_nb, oti);
702 /* set per-requested niobuf return codes */
703 for (i = j = 0; i < niocount; i++) {
704 int nob = remote_nb[i].len;
709 if (local_nb[j].rc < 0)
710 rcs[i] = local_nb[j].rc;
711 nob -= pp_rnb[j].len;
716 LASSERT(j == npages);
722 ptlrpc_free_bulk(desc);
724 OBD_FREE(local_nb, sizeof(*local_nb) * npages);
726 free_per_page_niobufs(npages, pp_rnb, remote_nb);
729 oti_to_request(oti, req);
730 rc = ptlrpc_reply(req);
731 } else if (!comms_error) {
732 /* Only reply if there was no comms problem with bulk */
736 if (req->rq_repmsg != NULL) {
737 /* reply out callback would free */
738 OBD_FREE (req->rq_repmsg, req->rq_replen);
740 CERROR("bulk IO comms error: evicting %s@%s nid "LPU64"\n",
741 req->rq_export->exp_client_uuid.uuid,
742 req->rq_connection->c_remote_uuid.uuid,
743 req->rq_connection->c_peer.peer_nid);
744 ptlrpc_fail_export(req->rq_export);
749 static int ost_san_brw(struct ptlrpc_request *req, int cmd)
751 struct niobuf_remote *remote_nb, *res_nb;
752 struct obd_ioobj *ioo;
753 struct ost_body *body, *repbody;
754 int rc, i, j, objcount, niocount, size[2] = {sizeof(*body)};
759 /* XXX not set to use latest protocol */
761 swab = lustre_msg_swabbed(req->rq_reqmsg);
762 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
764 CERROR("Missing/short ost_body\n");
765 GOTO(out, rc = -EFAULT);
768 ioo = lustre_swab_reqbuf(req, 1, sizeof(*ioo), lustre_swab_obd_ioobj);
770 CERROR("Missing/short ioobj\n");
771 GOTO(out, rc = -EFAULT);
773 objcount = req->rq_reqmsg->buflens[1] / sizeof(*ioo);
774 niocount = ioo[0].ioo_bufcnt;
775 for (i = 1; i < objcount; i++) {
777 lustre_swab_obd_ioobj (&ioo[i]);
778 niocount += ioo[i].ioo_bufcnt;
781 remote_nb = lustre_swab_reqbuf(req, 2, niocount * sizeof(*remote_nb),
782 lustre_swab_niobuf_remote);
783 if (remote_nb == NULL) {
784 CERROR("Missing/short niobuf\n");
785 GOTO(out, rc = -EFAULT);
787 if (swab) { /* swab the remaining niobufs */
788 for (i = 1; i < niocount; i++)
789 lustre_swab_niobuf_remote (&remote_nb[i]);
792 for (i = n = 0; i < objcount; i++) {
793 for (j = 0; j < ioo[i].ioo_bufcnt; j++, n++) {
794 if (remote_nb[n].len == 0) {
795 CERROR("zero len BRW: objid "LPX64" buf %u\n",
797 GOTO(out, rc = -EINVAL);
799 if (j && remote_nb[n].offset <= remote_nb[n-1].offset) {
800 CERROR("unordered BRW: objid "LPX64
801 " buf %u offset "LPX64" <= "LPX64"\n",
802 ioo[i].ioo_id, j, remote_nb[n].offset,
803 remote_nb[n-1].offset);
804 GOTO(out, rc = -EINVAL);
809 size[1] = niocount * sizeof(*remote_nb);
810 rc = lustre_pack_msg(2, size, NULL, &req->rq_replen, &req->rq_repmsg);
814 req->rq_status = obd_san_preprw(cmd, req->rq_export, &body->oa,
815 objcount, ioo, niocount, remote_nb);
820 repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
821 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
823 res_nb = lustre_msg_buf(req->rq_repmsg, 1, size[1]);
824 memcpy(res_nb, remote_nb, size[1]);
828 OBD_FREE(req->rq_repmsg, req->rq_replen);
829 req->rq_repmsg = NULL;
838 static int ost_log_cancel(struct ptlrpc_request *req)
840 struct lustre_handle *conn;
841 struct llog_cookie *logcookies;
842 int num_cookies, rc = 0;
845 logcookies = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*logcookies));
846 if (logcookies == NULL) {
847 DEBUG_REQ(D_HA, req, "no cookies sent");
850 num_cookies = req->rq_reqmsg->buflens[0] / sizeof(*logcookies);
852 /* workaround until we don't need to send replies */
853 rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
856 req->rq_repmsg->status = 0;
859 conn = (struct lustre_handle *)&req->rq_reqmsg->handle;
860 rc = obd_log_cancel(conn, NULL, num_cookies, logcookies, 0);
865 static int ost_set_info(struct ptlrpc_request *req)
867 struct lustre_handle *conn;
872 key = lustre_msg_buf(req->rq_reqmsg, 0, 1);
874 DEBUG_REQ(D_HA, req, "no set_info key");
877 keylen = req->rq_reqmsg->buflens[0];
879 rc = lustre_pack_msg(0, NULL, NULL, &req->rq_replen, &req->rq_repmsg);
883 conn = (struct lustre_handle *)&req->rq_reqmsg->handle;
884 rc = obd_set_info(conn, keylen, key, 0, NULL);
885 req->rq_repmsg->status = 0;
889 static int filter_recovery_request(struct ptlrpc_request *req,
890 struct obd_device *obd, int *process)
892 switch (req->rq_reqmsg->opc) {
893 case OST_CONNECT: /* This will never get here, but for completeness. */
909 *process = target_queue_recovery_request(req, obd);
913 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
915 /* XXX what should we set rq_status to here? */
916 req->rq_status = -EAGAIN;
917 RETURN(ptlrpc_error(req));
923 static int ost_handle(struct ptlrpc_request *req)
925 struct obd_trans_info trans_info = { 0, };
926 struct obd_trans_info *oti = &trans_info;
927 int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
930 LASSERT(current->journal_info == NULL);
931 /* XXX identical to MDS */
932 if (req->rq_reqmsg->opc != OST_CONNECT) {
933 struct obd_device *obd;
934 int abort_recovery, recovering;
936 if (req->rq_export == NULL) {
937 CDEBUG(D_HA, "operation %d on unconnected OST\n",
938 req->rq_reqmsg->opc);
939 req->rq_status = -ENOTCONN;
940 GOTO(out, rc = -ENOTCONN);
943 obd = req->rq_export->exp_obd;
945 /* Check for aborted recovery. */
946 spin_lock_bh(&obd->obd_processing_task_lock);
947 abort_recovery = obd->obd_abort_recovery;
948 recovering = obd->obd_recovering;
949 spin_unlock_bh(&obd->obd_processing_task_lock);
950 if (abort_recovery) {
951 target_abort_recovery(obd);
952 } else if (recovering) {
953 rc = filter_recovery_request(req, obd, &should_process);
954 if (rc || !should_process)
959 if (strcmp(req->rq_obd->obd_type->typ_name, "ost") != 0)
960 GOTO(out, rc = -EINVAL);
964 switch (req->rq_reqmsg->opc) {
966 CDEBUG(D_INODE, "connect\n");
967 OBD_FAIL_RETURN(OBD_FAIL_OST_CONNECT_NET, 0);
968 rc = target_handle_connect(req, ost_handle);
971 CDEBUG(D_INODE, "disconnect\n");
972 OBD_FAIL_RETURN(OBD_FAIL_OST_DISCONNECT_NET, 0);
973 rc = target_handle_disconnect(req);
976 CDEBUG(D_INODE, "create\n");
977 OBD_FAIL_RETURN(OBD_FAIL_OST_CREATE_NET, 0);
978 rc = ost_create(req, oti);
981 CDEBUG(D_INODE, "destroy\n");
982 OBD_FAIL_RETURN(OBD_FAIL_OST_DESTROY_NET, 0);
983 rc = ost_destroy(req, oti);
986 CDEBUG(D_INODE, "getattr\n");
987 OBD_FAIL_RETURN(OBD_FAIL_OST_GETATTR_NET, 0);
988 rc = ost_getattr(req);
991 CDEBUG(D_INODE, "setattr\n");
992 OBD_FAIL_RETURN(OBD_FAIL_OST_SETATTR_NET, 0);
993 rc = ost_setattr(req, oti);
996 CDEBUG(D_INODE, "open\n");
997 OBD_FAIL_RETURN(OBD_FAIL_OST_OPEN_NET, 0);
998 rc = ost_open(req, oti);
1001 CDEBUG(D_INODE, "close\n");
1002 OBD_FAIL_RETURN(OBD_FAIL_OST_CLOSE_NET, 0);
1003 rc = ost_close(req, oti);
1006 CDEBUG(D_INODE, "write\n");
1007 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
1008 rc = ost_brw_write(req, oti);
1009 LASSERT(current->journal_info == NULL);
1010 /* ost_brw sends its own replies */
1013 CDEBUG(D_INODE, "read\n");
1014 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
1015 rc = ost_brw_read(req);
1016 LASSERT(current->journal_info == NULL);
1017 /* ost_brw sends its own replies */
1020 CDEBUG(D_INODE, "san read\n");
1021 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
1022 rc = ost_san_brw(req, OBD_BRW_READ);
1023 /* ost_san_brw sends its own replies */
1026 CDEBUG(D_INODE, "san write\n");
1027 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
1028 rc = ost_san_brw(req, OBD_BRW_WRITE);
1029 /* ost_san_brw sends its own replies */
1032 CDEBUG(D_INODE, "punch\n");
1033 OBD_FAIL_RETURN(OBD_FAIL_OST_PUNCH_NET, 0);
1034 rc = ost_punch(req, oti);
1037 CDEBUG(D_INODE, "statfs\n");
1038 OBD_FAIL_RETURN(OBD_FAIL_OST_STATFS_NET, 0);
1039 rc = ost_statfs(req);
1042 CDEBUG(D_INODE, "sync\n");
1043 OBD_FAIL_RETURN(OBD_FAIL_OST_SYNCFS_NET, 0);
1044 rc = ost_syncfs(req);
1047 DEBUG_REQ(D_INODE, req, "set_info");
1048 rc = ost_set_info(req);
1050 DEBUG_REQ(D_INODE, req, "ping");
1051 rc = target_handle_ping(req);
1053 case OBD_LOG_CANCEL:
1054 CDEBUG(D_INODE, "log cancel\n");
1055 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOG_CANCEL_NET, 0);
1056 rc = ost_log_cancel(req);
1059 CDEBUG(D_INODE, "enqueue\n");
1060 OBD_FAIL_RETURN(OBD_FAIL_LDLM_ENQUEUE, 0);
1061 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
1062 ldlm_server_blocking_ast);
1063 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
1066 CDEBUG(D_INODE, "convert\n");
1067 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CONVERT, 0);
1068 rc = ldlm_handle_convert(req);
1071 CDEBUG(D_INODE, "cancel\n");
1072 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CANCEL, 0);
1073 rc = ldlm_handle_cancel(req);
1075 case LDLM_BL_CALLBACK:
1076 case LDLM_CP_CALLBACK:
1077 CDEBUG(D_INODE, "callback\n");
1078 CERROR("callbacks should not happen on OST\n");
1081 CERROR("Unexpected opcode %d\n", req->rq_reqmsg->opc);
1082 req->rq_status = -ENOTSUPP;
1083 rc = ptlrpc_error(req);
1087 LASSERT(current->journal_info == NULL);
1090 /* If we're DISCONNECTing, the export_data is already freed */
1091 if (!rc && req->rq_reqmsg->opc != OST_DISCONNECT) {
1092 struct obd_device *obd = req->rq_export->exp_obd;
1093 if (!obd->obd_no_transno) {
1094 req->rq_repmsg->last_committed =
1095 obd->obd_last_committed;
1097 DEBUG_REQ(D_IOCTL, req,
1098 "not sending last_committed update");
1100 CDEBUG(D_INFO, "last_committed "LPU64", xid "LPX64"\n",
1101 obd->obd_last_committed, req->rq_xid);
1105 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LAST_REPLAY) {
1106 struct obd_device *obd = req->rq_export->exp_obd;
1108 if (obd && obd->obd_recovering) {
1109 DEBUG_REQ(D_HA, req, "LAST_REPLAY, queuing reply");
1110 return target_queue_final_reply(req, rc);
1112 /* Lost a race with recovery; let the error path DTRT. */
1113 rc = req->rq_status = -ENOTCONN;
1117 oti_to_request(oti, req);
1119 target_send_reply(req, rc, fail);
1123 static int ost_setup(struct obd_device *obddev, obd_count len, void *buf)
1125 struct ost_obd *ost = &obddev->u.ost;
1129 #ifdef ENABLE_ORPHANS
1130 err = llog_start_commit_thread();
1135 ost->ost_service = ptlrpc_init_svc(OST_NEVENTS, OST_NBUFS,
1136 OST_BUFSIZE, OST_MAXREQSIZE,
1137 OST_REQUEST_PORTAL, OSC_REPLY_PORTAL,
1138 ost_handle, "ost", obddev);
1139 if (!ost->ost_service) {
1140 CERROR("failed to start service\n");
1144 for (i = 0; i < OST_NUM_THREADS; i++) {
1146 sprintf(name, "ll_ost_%02d", i);
1147 err = ptlrpc_start_thread(obddev, ost->ost_service, name);
1149 CERROR("error starting thread #%d: rc %d\n", i, err);
1157 static int ost_cleanup(struct obd_device *obddev, int flags)
1159 struct ost_obd *ost = &obddev->u.ost;
1163 if (obddev->obd_recovering)
1164 target_cancel_recovery_timer(obddev);
1166 ptlrpc_stop_all_threads(ost->ost_service);
1167 ptlrpc_unregister_service(ost->ost_service);
1172 int ost_attach(struct obd_device *dev, obd_count len, void *data)
1174 struct lprocfs_static_vars lvars;
1176 lprocfs_init_vars(ost,&lvars);
1177 return lprocfs_obd_attach(dev, lvars.obd_vars);
1180 int ost_detach(struct obd_device *dev)
1182 return lprocfs_obd_detach(dev);
1185 /* I don't think this function is ever used, since nothing
1186 * connects directly to this module.
1188 static int ost_connect(struct lustre_handle *conn,
1189 struct obd_device *obd, struct obd_uuid *cluuid)
1191 struct obd_export *exp;
1195 if (!conn || !obd || !cluuid)
1198 rc = class_connect(conn, obd, cluuid);
1201 exp = class_conn2export(conn);
1203 class_export_put(exp);
1208 /* use obd ops to offer management infrastructure */
1209 static struct obd_ops ost_obd_ops = {
1210 o_owner: THIS_MODULE,
1211 o_attach: ost_attach,
1212 o_detach: ost_detach,
1214 o_cleanup: ost_cleanup,
1215 o_connect: ost_connect,
1218 static int __init ost_init(void)
1220 struct lprocfs_static_vars lvars;
1223 lprocfs_init_vars(ost,&lvars);
1224 RETURN(class_register_type(&ost_obd_ops, lvars.module_vars,
1228 static void /*__exit*/ ost_exit(void)
1230 class_unregister_type(LUSTRE_OST_NAME);
1233 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1234 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
1235 MODULE_LICENSE("GPL");
1237 module_init(ost_init);
1238 module_exit(ost_exit);