Whamcloud - gitweb
a83592f081c93d5f54704c5d5d4e68acaf04a27e
[fs/lustre-release.git] / lustre / ost / ost_handler.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  *
4  *  Copyright (C) 2001-2003 Cluster File Systems, Inc.
5  *   Author: Peter J. Braam <braam@clusterfs.com>
6  *   Author: Phil Schwan <phil@clusterfs.com>
7  *
8  *   This file is part of Lustre, http://www.lustre.org.
9  *
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.
13  *
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.
18  *
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.
22  *
23  *  Storage Target Handling functions
24  *  Lustre Object Server Module (OST)
25  *
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.)
31  */
32
33 #ifndef EXPORT_SYMTAB
34 # define EXPORT_SYMTAB
35 #endif
36 #define DEBUG_SUBSYSTEM S_OST
37
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>
47
48 void oti_init(struct obd_trans_info *oti, struct ptlrpc_request *req)
49 {
50         if (oti == NULL)
51                 return;
52         memset(oti, 0, sizeof *oti);
53
54         if (req->rq_repmsg && req->rq_reqmsg != 0)
55                 oti->oti_transno = req->rq_repmsg->transno;
56 }
57
58 void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req)
59 {
60         struct oti_req_ack_lock *ack_lock;
61         int i;
62
63         if (oti == NULL)
64                 return;
65
66         if (req->rq_repmsg)
67                 req->rq_repmsg->transno = oti->oti_transno;
68
69         /* XXX 4 == entries in oti_ack_locks??? */
70         for (ack_lock = oti->oti_ack_locks, i = 0; i < 4; i++, ack_lock++) {
71                 if (!ack_lock->mode)
72                         break;
73                 ldlm_put_lock_into_req(req, &ack_lock->lock, ack_lock->mode);
74         }
75 }
76
77 static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req, 
78                        struct obd_trans_info *oti)
79 {
80         struct ost_body *body, *repbody;
81         int rc, size = sizeof(*body);
82         ENTRY;
83
84         body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
85         if (body == NULL)
86                 RETURN(-EFAULT);
87
88         rc = lustre_pack_reply(req, 1, &size, NULL);
89         if (rc)
90                 RETURN(rc);
91
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);
97         RETURN(0);
98 }
99
100 static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
101 {
102         struct ost_body *body, *repbody;
103         int rc, size = sizeof(*body);
104         ENTRY;
105
106         body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
107         if (body == NULL)
108                 RETURN(-EFAULT);
109
110         rc = lustre_pack_reply(req, 1, &size, NULL);
111         if (rc)
112                 RETURN(rc);
113
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);
117         RETURN(0);
118 }
119
120 static int ost_statfs(struct ptlrpc_request *req)
121 {
122         struct obd_statfs *osfs;
123         int rc, size = sizeof(*osfs);
124         ENTRY;
125
126         rc = lustre_pack_reply(req, 1, &size, NULL);
127         if (rc)
128                 RETURN(rc);
129
130         osfs = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*osfs));
131
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);
135
136         RETURN(0);
137 }
138
139 static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
140                       struct obd_trans_info *oti)
141 {
142         struct ost_body *body, *repbody;
143         int rc, size = sizeof(*repbody);
144         ENTRY;
145
146         body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
147         if (body == NULL)
148                 RETURN(-EFAULT);
149
150         rc = lustre_pack_reply(req, 1, &size, NULL);
151         if (rc)
152                 RETURN(rc);
153
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);
159         RETURN(0);
160 }
161
162 static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req, 
163                      struct obd_trans_info *oti)
164 {
165         struct ost_body *body, *repbody;
166         int rc, size = sizeof(*repbody);
167         ENTRY;
168
169         body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
170         if (body == NULL)
171                 RETURN(-EFAULT);
172
173         if ((body->oa.o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
174             (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
175                 RETURN(-EINVAL);
176
177         rc = lustre_pack_reply(req, 1, &size, NULL);
178         if (rc)
179                 RETURN(rc);
180
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);
185         RETURN(0);
186 }
187
188 static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req)
189 {
190         struct ost_body *body, *repbody;
191         int rc, size = sizeof(*repbody);
192         ENTRY;
193
194         body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
195         if (body == NULL)
196                 RETURN(-EFAULT);
197
198         rc = lustre_pack_reply(req, 1, &size, NULL);
199         if (rc)
200                 RETURN(rc);
201
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);
206         RETURN(0);
207 }
208
209 static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req, 
210                        struct obd_trans_info *oti)
211 {
212         struct ost_body *body, *repbody;
213         int rc, size = sizeof(*repbody);
214         ENTRY;
215
216         body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
217         if (body == NULL)
218                 RETURN(-EFAULT);
219
220         rc = lustre_pack_reply(req, 1, &size, NULL);
221         if (rc)
222                 RETURN(rc);
223
224         repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
225         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
226
227         req->rq_status = obd_setattr(exp, &repbody->oa, NULL, oti);
228         RETURN(0);
229 }
230
231 static int ost_bulk_timeout(void *data)
232 {
233         ENTRY;
234         /* We don't fail the connection here, because having the export
235          * killed makes the (vital) call to commitrw very sad.
236          */
237         RETURN(1);
238 }
239
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)
243 {
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;
247         int   i;
248         int   j;
249         int   page;
250         int   rnbidx = 0;
251         int   npages = 0;
252
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;
259
260                         LASSERT(rnbidx < nrnb);
261
262                         npages += (pn + 1 - p0);
263
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);
267                                 return -EINVAL;
268                         }
269                         if (j > 0 &&
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,
274                                        rnb[rnbidx].offset);
275                                 return -EINVAL;
276                         }
277                 }
278
279         LASSERT(rnbidx == nrnb);
280
281         if (npages == nrnb) {       /* all niobufs are for single pages */
282                 *pp_rnbp = rnb;
283                 return npages;
284         }
285
286         OBD_ALLOC(pp_rnb, sizeof(*pp_rnb) * npages);
287         if (pp_rnb == NULL)
288                 return -ENOMEM;
289
290         /* now do the actual split */
291         page = rnbidx = 0;
292         for (i = 0; i < nioo; i++) {
293                 int  obj_pages = 0;
294
295                 for (j = 0; j < ioo[i].ioo_bufcnt; j++, rnbidx++) {
296                         obd_off off = rnb[rnbidx].offset;
297                         int     nob = rnb[rnbidx].len;
298
299                         LASSERT(rnbidx < nrnb);
300                         do {
301                                 obd_off  poff = off & (PAGE_SIZE - 1);
302                                 int      pnob = (poff + nob > PAGE_SIZE) ?
303                                                 PAGE_SIZE - poff : nob;
304
305                                 LASSERT(page < npages);
306                                 pp_rnb[page].len = pnob;
307                                 pp_rnb[page].offset = off;
308                                 pp_rnb[page].flags = rnb->flags;
309
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);
314                                 page++;
315                                 obj_pages++;
316
317                                 off += pnob;
318                                 nob -= pnob;
319                         } while (nob > 0);
320                         LASSERT(nob == 0);
321                 }
322                 ioo[i].ioo_bufcnt = obj_pages;
323         }
324         LASSERT(page == npages);
325
326         *pp_rnbp = pp_rnb;
327         return npages;
328 }
329
330 static void free_per_page_niobufs (int npages, struct niobuf_remote *pp_rnb,
331                                    struct niobuf_remote *rnb)
332 {
333         if (pp_rnb == rnb)                      /* didn't allocate above */
334                 return;
335
336         OBD_FREE(pp_rnb, sizeof(*pp_rnb) * npages);
337 }
338
339 #if CHECKSUM_BULK
340 obd_count ost_checksum_bulk(struct ptlrpc_bulk_desc *desc)
341 {
342         obd_count cksum = 0;
343         struct ptlrpc_bulk_page *bp;
344
345         list_for_each_entry(bp, &desc->bd_page_list, bp_link) {
346                 ost_checksum(&cksum, kmap(bp->bp_page) + bp->bp_pageoffset,
347                              bp->bp_buflen);
348                 kunmap(bp->bp_page);
349         }
350
351         return cksum;
352 }
353 #endif
354
355 static int ost_brw_read(struct ptlrpc_request *req)
356 {
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) };
367         int                      comms_error = 0;
368         int                      niocount;
369         int                      npages;
370         int                      nob = 0;
371         int                      rc;
372         int                      i;
373         ENTRY;
374
375         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
376                 GOTO(out, rc = -EIO);
377
378         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK | OBD_FAIL_ONCE,
379                          (obd_timeout + 1) / 4);
380
381         body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
382         if (body == NULL) {
383                 CERROR("Missing/short ost_body\n");
384                 GOTO(out, rc = -EFAULT);
385         }
386
387         /* BUG 974: when we send back cache grants, don't clear this flag */
388         body->oa.o_valid &= ~OBD_MD_FLRDEV;
389
390         ioo = lustre_swab_reqbuf(req, 1, sizeof(*ioo), lustre_swab_obd_ioobj);
391         if (ioo == NULL) {
392                 CERROR("Missing/short ioobj\n");
393                 GOTO(out, rc = -EFAULT);
394         }
395
396         niocount = ioo->ioo_bufcnt;
397         remote_nb = lustre_swab_reqbuf(req, 2, niocount * sizeof(*remote_nb),
398                                        lustre_swab_niobuf_remote);
399         if (remote_nb == NULL) {
400                 CERROR("Missing/short niobuf\n");
401                 GOTO(out, rc = -EFAULT);
402         }
403         if (lustre_msg_swabbed(req->rq_reqmsg)) { /* swab remaining niobufs */
404                 for (i = 1; i < niocount; i++)
405                         lustre_swab_niobuf_remote (&remote_nb[i]);
406         }
407
408         rc = lustre_pack_reply(req, 1, size, NULL);
409         if (rc)
410                 GOTO(out, rc);
411
412         /* FIXME all niobuf splitting should be done in obdfilter if needed */
413         /* CAVEAT EMPTOR this sets ioo->ioo_bufcnt to # pages */
414         npages = get_per_page_niobufs(ioo, 1, remote_nb, niocount, &pp_rnb);
415         if (npages < 0)
416                 GOTO(out, rc = npages);
417
418         OBD_ALLOC(local_nb, sizeof(*local_nb) * npages);
419         if (local_nb == NULL)
420                 GOTO(out_pp_rnb, rc = -ENOMEM);
421
422         desc = ptlrpc_prep_bulk_exp(req, BULK_PUT_SOURCE, OST_BULK_PORTAL);
423         if (desc == NULL)
424                 GOTO(out_local, rc = -ENOMEM);
425
426         rc = obd_preprw(OBD_BRW_READ, req->rq_export, &body->oa, 1,
427                         ioo, npages, pp_rnb, local_nb, &oti);
428         if (rc != 0)
429                 GOTO(out_bulk, rc);
430
431         nob = 0;
432         for (i = 0; i < npages; i++) {
433                 int page_rc = local_nb[i].rc;
434
435                 if (page_rc < 0) {              /* error */
436                         rc = page_rc;
437                         break;
438                 }
439
440                 LASSERT(page_rc <= pp_rnb[i].len);
441                 nob += page_rc;
442                 if (page_rc != 0) {             /* some data! */
443                         LASSERT (local_nb[i].page != NULL);
444                         rc = ptlrpc_prep_bulk_page(desc, local_nb[i].page,
445                                                    pp_rnb[i].offset& ~PAGE_MASK,
446                                                    page_rc);
447                         if (rc != 0)
448                                 break;
449                 }
450
451                 if (page_rc != pp_rnb[i].len) { /* short read */
452                         /* All subsequent pages should be 0 */
453                         while(++i < npages)
454                                 LASSERT(local_nb[i].rc == 0);
455                         break;
456                 }
457         }
458
459         if (rc == 0) {
460                 rc = ptlrpc_bulk_put(desc);
461                 if (rc == 0) {
462                         lwi = LWI_TIMEOUT(obd_timeout * HZ / 4,
463                                           ost_bulk_timeout, desc);
464                         rc = l_wait_event(desc->bd_waitq,
465                                           ptlrpc_bulk_complete(desc), &lwi);
466                         if (rc) {
467                                 LASSERT(rc == -ETIMEDOUT);
468                                 DEBUG_REQ(D_ERROR, req, "timeout on bulk PUT");
469                                 ptlrpc_abort_bulk(desc);
470                         }
471                 } else {
472                         DEBUG_REQ(D_ERROR, req, "bulk PUT failed: rc %d\n", rc);
473                 }
474                 comms_error = rc != 0;
475         }
476
477         /* Must commit after prep above in all cases */
478         rc = obd_commitrw(OBD_BRW_READ, req->rq_export, &body->oa, 1,
479                           ioo, npages, local_nb, &oti);
480
481         repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
482         memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
483
484 #if CHECKSUM_BULK
485         if (rc == 0) {
486                 repbody->oa.o_nlink = ost_checksum_bulk(desc);
487                 repbody->oa.o_valid |= OBD_MD_FLCKSUM;
488         }
489 #endif
490
491  out_bulk:
492         ptlrpc_free_bulk(desc);
493  out_local:
494         OBD_FREE(local_nb, sizeof(*local_nb) * npages);
495  out_pp_rnb:
496         free_per_page_niobufs(npages, pp_rnb, remote_nb);
497  out:
498         LASSERT(rc <= 0);
499         if (rc == 0) {
500                 req->rq_status = nob;
501                 ptlrpc_reply(req);
502         } else if (!comms_error) {
503                 /* only reply if comms OK */
504                 req->rq_status = rc;
505                 ptlrpc_error(req);
506         } else {
507                 if (req->rq_repmsg != NULL) {
508                         /* reply out callback would free */
509                         OBD_FREE(req->rq_repmsg, req->rq_replen);
510                 }
511                 if (req->rq_reqmsg->conn_cnt == req->rq_export->exp_conn_cnt) {
512                         CERROR("bulk IO comms error: "
513                                "evicting %s@%s nid "LPX64" (%s)\n",
514                                req->rq_export->exp_client_uuid.uuid,
515                                req->rq_connection->c_remote_uuid.uuid,
516                                req->rq_connection->c_peer.peer_nid,
517                                portals_nid2str(req->rq_connection->c_peer.peer_ni->pni_number,
518                                                req->rq_connection->c_peer.peer_nid,
519                                                str));
520                         ptlrpc_fail_export(req->rq_export);
521                 } else {
522                         CERROR("ignoring bulk IO comms error: "
523                                "client reconnected %s@%s nid "LPX64" (%s)\n",  
524                                req->rq_export->exp_client_uuid.uuid,
525                                req->rq_connection->c_remote_uuid.uuid,
526                                req->rq_connection->c_peer.peer_nid,
527                                portals_nid2str(req->rq_connection->c_peer.peer_ni->pni_number,
528                                                req->rq_connection->c_peer.peer_nid,
529                                                str));
530                 }
531         }
532
533         RETURN(rc);
534 }
535
536 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
537 {
538         struct ptlrpc_bulk_desc *desc;
539         struct niobuf_remote    *remote_nb;
540         struct niobuf_remote    *pp_rnb;
541         struct niobuf_local     *local_nb;
542         struct obd_ioobj        *ioo;
543         struct ost_body         *body, *repbody;
544         struct l_wait_info       lwi;
545         __u32                   *rcs;
546         int                      size[2] = { sizeof(*body) };
547         int                      objcount, niocount, npages;
548         int                      comms_error = 0;
549         int                      rc, rc2, swab, i, j;
550         char                    str[PTL_NALFMT_SIZE];
551         ENTRY;
552
553         if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
554                 GOTO(out, rc = -EIO);
555
556         /* pause before transaction has been started */
557         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK | OBD_FAIL_ONCE,
558                          (obd_timeout + 1) / 4);
559
560         swab = lustre_msg_swabbed(req->rq_reqmsg);
561         body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
562         if (body == NULL) {
563                 CERROR("Missing/short ost_body\n");
564                 GOTO(out, rc = -EFAULT);
565         }
566
567         /* BUG 974: when we send back cache grants, don't clear this flag */
568         body->oa.o_valid &= ~OBD_MD_FLRDEV;
569
570         LASSERT_REQSWAB(req, 1);
571         objcount = req->rq_reqmsg->buflens[1] / sizeof(*ioo);
572         if (objcount == 0) {
573                 CERROR("Missing/short ioobj\n");
574                 GOTO(out, rc = -EFAULT);
575         }
576         ioo = lustre_msg_buf (req->rq_reqmsg, 1, objcount * sizeof(*ioo));
577         LASSERT (ioo != NULL);
578         for (niocount = i = 0; i < objcount; i++) {
579                 if (swab)
580                         lustre_swab_obd_ioobj (&ioo[i]);
581                 if (ioo[i].ioo_bufcnt == 0) {
582                         CERROR("ioo[%d] has zero bufcnt\n", i);
583                         GOTO(out, rc = -EFAULT);
584                 }
585                 niocount += ioo[i].ioo_bufcnt;
586         }
587
588         remote_nb = lustre_swab_reqbuf(req, 2, niocount * sizeof(*remote_nb),
589                                        lustre_swab_niobuf_remote);
590         if (remote_nb == NULL) {
591                 CERROR("Missing/short niobuf\n");
592                 GOTO(out, rc = -EFAULT);
593         }
594         if (swab) {                             /* swab the remaining niobufs */
595                 for (i = 1; i < niocount; i++)
596                         lustre_swab_niobuf_remote (&remote_nb[i]);
597         }
598
599         size[1] = niocount * sizeof(*rcs);
600         rc = lustre_pack_reply(req, 2, size, NULL);
601         if (rc != 0)
602                 GOTO(out, rc);
603         rcs = lustre_msg_buf(req->rq_repmsg, 1, niocount * sizeof(*rcs));
604
605         /* FIXME all niobuf splitting should be done in obdfilter if needed */
606         /* CAVEAT EMPTOR this sets ioo->ioo_bufcnt to # pages */
607         npages = get_per_page_niobufs(ioo, objcount,remote_nb,niocount,&pp_rnb);
608         if (npages < 0)
609                 GOTO(out, rc = npages);
610
611         OBD_ALLOC(local_nb, sizeof(*local_nb) * npages);
612         if (local_nb == NULL)
613                 GOTO(out_pp_rnb, rc = -ENOMEM);
614
615         desc = ptlrpc_prep_bulk_exp(req, BULK_GET_SINK, OST_BULK_PORTAL);
616         if (desc == NULL)
617                 GOTO(out_local, rc = -ENOMEM);
618
619         rc = obd_preprw(OBD_BRW_WRITE, req->rq_export, &body->oa, objcount,
620                         ioo, npages, pp_rnb, local_nb, oti);
621         if (rc != 0)
622                 GOTO(out_bulk, rc);
623
624         /* NB Having prepped, we must commit... */
625
626         for (i = 0; i < npages; i++) {
627                 rc = ptlrpc_prep_bulk_page(desc, local_nb[i].page,
628                                            pp_rnb[i].offset & (PAGE_SIZE - 1),
629                                            pp_rnb[i].len);
630                 if (rc != 0)
631                         break;
632         }
633
634         if (rc == 0) {
635                 rc = ptlrpc_bulk_get(desc);
636                 if (rc == 0) {
637                         lwi = LWI_TIMEOUT(obd_timeout * HZ / 4,
638                                           ost_bulk_timeout, desc);
639                         rc = l_wait_event(desc->bd_waitq,
640                                           ptlrpc_bulk_complete(desc), &lwi);
641                         if (rc) {
642                                 LASSERT(rc == -ETIMEDOUT);
643                                 DEBUG_REQ(D_ERROR, req, "timeout on bulk GET");
644                                 ptlrpc_abort_bulk(desc);
645                         }
646                 } else {
647                         DEBUG_REQ(D_ERROR, req, "bulk GET failed: rc %d\n", rc);
648                 }
649                 comms_error = rc != 0;
650         }
651
652         repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
653         memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
654
655 #if CHECKSUM_BULK
656         if (rc == 0 && (body->oa.o_valid & OBD_MD_FLCKSUM) != 0) {
657                 static int cksum_counter;
658                 obd_count client_cksum = body->oa.o_nlink;
659                 obd_count cksum = ost_checksum_bulk(desc);
660
661                 portals_nid2str(req->rq_connection->c_peer.peer_ni->pni_number,
662                                 req->rq_connection->c_peer.peer_nid, str);
663                 if (client_cksum != cksum) {
664                         CERROR("Bad checksum: client %x, server %x, client NID "
665                                LPX64" (%s)\n", client_cksum, cksum,
666                                req->rq_connection->c_peer.peer_nid, str);
667                         cksum_counter = 1;
668                         repbody->oa.o_nlink = cksum;
669                 } else {
670                         cksum_counter++;
671                         if ((cksum_counter & (-cksum_counter)) == cksum_counter)
672                                 CWARN("Checksum %u from "LPX64": %x OK\n",
673                                       cksum_counter,
674                                       req->rq_connection->c_peer.peer_nid,
675                                       cksum);
676                 }
677         }
678 #endif
679         /* Must commit after prep above in all cases */
680         rc2 = obd_commitrw(OBD_BRW_WRITE, req->rq_export, &repbody->oa,
681                            objcount, ioo, npages, local_nb, oti);
682
683         if (rc == 0) {
684                 /* set per-requested niobuf return codes */
685                 for (i = j = 0; i < niocount; i++) {
686                         int nob = remote_nb[i].len;
687
688                         rcs[i] = 0;
689                         do {
690                                 LASSERT(j < npages);
691                                 if (local_nb[j].rc < 0)
692                                         rcs[i] = local_nb[j].rc;
693                                 nob -= pp_rnb[j].len;
694                                 j++;
695                         } while (nob > 0);
696                         LASSERT(nob == 0);
697                 }
698                 LASSERT(j == npages);
699         }
700         if (rc == 0)
701                 rc = rc2;
702
703  out_bulk:
704         ptlrpc_free_bulk(desc);
705  out_local:
706         OBD_FREE(local_nb, sizeof(*local_nb) * npages);
707  out_pp_rnb:
708         free_per_page_niobufs(npages, pp_rnb, remote_nb);
709  out:
710         if (rc == 0) {
711                 oti_to_request(oti, req);
712                 rc = ptlrpc_reply(req);
713         } else if (!comms_error) {
714                 /* Only reply if there was no comms problem with bulk */
715                 req->rq_status = rc;
716                 ptlrpc_error(req);
717         } else {
718                 if (req->rq_repmsg != NULL) {
719                         /* reply out callback would free */
720                         OBD_FREE (req->rq_repmsg, req->rq_replen);
721                 }
722                 if (req->rq_reqmsg->conn_cnt == req->rq_export->exp_conn_cnt) {
723                         CERROR("bulk IO comms error: "
724                                "evicting %s@%s nid "LPX64" (%s)\n",
725                                req->rq_export->exp_client_uuid.uuid,
726                                req->rq_connection->c_remote_uuid.uuid,
727                                req->rq_connection->c_peer.peer_nid,
728                                portals_nid2str(req->rq_connection->c_peer.peer_ni->pni_number,
729                                                req->rq_connection->c_peer.peer_nid,
730                                                str));
731                         ptlrpc_fail_export(req->rq_export);
732                 } else {
733                         CERROR("ignoring bulk IO comms error: "
734                                "client reconnected %s@%s nid "LPX64" (%s)\n",
735                                req->rq_export->exp_client_uuid.uuid,
736                                req->rq_connection->c_remote_uuid.uuid,
737                                req->rq_connection->c_peer.peer_nid,
738                                portals_nid2str(req->rq_connection->c_peer.peer_ni->pni_number,
739                                                req->rq_connection->c_peer.peer_nid,
740                                                str));
741                 }        
742         }
743         RETURN(rc);
744 }
745
746 static int ost_san_brw(struct ptlrpc_request *req, int cmd)
747 {
748         struct niobuf_remote *remote_nb, *res_nb, *pp_rnb;
749         struct obd_ioobj *ioo;
750         struct ost_body *body, *repbody;
751         int rc, i, objcount, niocount, size[2] = {sizeof(*body)}, npages;
752         int swab;
753         ENTRY;
754
755         /* XXX not set to use latest protocol */
756
757         swab = lustre_msg_swabbed(req->rq_reqmsg);
758         body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_ost_body);
759         if (body == NULL) {
760                 CERROR("Missing/short ost_body\n");
761                 GOTO(out, rc = -EFAULT);
762         }
763
764         ioo = lustre_swab_reqbuf(req, 1, sizeof(*ioo), lustre_swab_obd_ioobj);
765         if (ioo == NULL) {
766                 CERROR("Missing/short ioobj\n");
767                 GOTO(out, rc = -EFAULT);
768         }
769         objcount = req->rq_reqmsg->buflens[1] / sizeof(*ioo);
770         niocount = ioo[0].ioo_bufcnt;
771         for (i = 1; i < objcount; i++) {
772                 if (swab)
773                         lustre_swab_obd_ioobj (&ioo[i]);
774                 niocount += ioo[i].ioo_bufcnt;
775         }
776
777         remote_nb = lustre_swab_reqbuf(req, 2, niocount * sizeof(*remote_nb),
778                                        lustre_swab_niobuf_remote);
779         if (remote_nb == NULL) {
780                 CERROR("Missing/short niobuf\n");
781                 GOTO(out, rc = -EFAULT);
782         }
783         if (swab) {                             /* swab the remaining niobufs */
784                 for (i = 1; i < niocount; i++)
785                         lustre_swab_niobuf_remote (&remote_nb[i]);
786         }
787
788         /* CAVEAT EMPTOR this sets ioo->ioo_bufcnt to # pages */
789         npages = get_per_page_niobufs(ioo, objcount,remote_nb,niocount,&pp_rnb);
790         if (npages < 0)
791                 GOTO (out, rc = npages);
792  
793         size[1] = npages * sizeof(*pp_rnb);
794         rc = lustre_pack_reply(req, 2, size, NULL);
795         if (rc)
796                 GOTO(out_pp_rnb, rc);
797
798         req->rq_status = obd_san_preprw(cmd, req->rq_export, &body->oa,
799                                         objcount, ioo, npages, pp_rnb);
800
801         if (req->rq_status)
802                 GOTO(out_pp_rnb, rc = 0);
803
804         repbody = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*repbody));
805         memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
806
807         res_nb = lustre_msg_buf(req->rq_repmsg, 1, size[1]);
808         memcpy(res_nb, remote_nb, size[1]);
809         rc = 0;
810 out_pp_rnb:
811         free_per_page_niobufs(npages, pp_rnb, remote_nb);
812 out:
813         if (rc) {
814                 OBD_FREE(req->rq_repmsg, req->rq_replen);
815                 req->rq_repmsg = NULL;
816                 req->rq_status = rc;
817                 ptlrpc_error(req);
818         } else
819                 ptlrpc_reply(req);
820
821         return rc;
822 }
823
824
825 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
826 {
827         char *key;
828         int keylen, rc = 0;
829         ENTRY;
830
831         key = lustre_msg_buf(req->rq_reqmsg, 0, 1);
832         if (key == NULL) {
833                 DEBUG_REQ(D_HA, req, "no set_info key");
834                 RETURN(-EFAULT);
835         }
836         keylen = req->rq_reqmsg->buflens[0];
837
838         rc = lustre_pack_reply(req, 0, NULL, NULL);
839         if (rc)
840                 RETURN(rc);
841
842         rc = obd_set_info(exp, keylen, key, 0, NULL);
843         req->rq_repmsg->status = 0;
844         RETURN(rc);
845 }
846
847 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
848 {
849         char *key;
850         int keylen, rc = 0, size = sizeof(obd_id);
851         obd_id *reply;
852         ENTRY;
853
854         key = lustre_msg_buf(req->rq_reqmsg, 0, 1);
855         if (key == NULL) {
856                 DEBUG_REQ(D_HA, req, "no get_info key");
857                 RETURN(-EFAULT);
858         }
859         keylen = req->rq_reqmsg->buflens[0];
860
861         if (keylen < strlen("last_id") || memcmp(key, "last_id", 7) != 0)
862                 RETURN(-EPROTO);
863
864         rc = lustre_pack_reply(req, 1, &size, NULL);
865         if (rc)
866                 RETURN(rc);
867
868         reply = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*reply));
869         rc = obd_get_info(exp, keylen, key, &size, reply);
870         req->rq_repmsg->status = 0;
871         RETURN(rc);
872 }
873
874 static int ost_filter_recovery_request(struct ptlrpc_request *req,
875                                        struct obd_device *obd, int *process)
876 {
877         switch (req->rq_reqmsg->opc) {
878         case OST_CONNECT: /* This will never get here, but for completeness. */
879         case OST_DISCONNECT:
880                *process = 1;
881                RETURN(0);
882
883         case OBD_PING:
884         case OST_CREATE:
885         case OST_DESTROY:
886         case OST_PUNCH:
887         case OST_SETATTR:
888         case OST_SYNC:
889         case OST_WRITE:
890         case OBD_LOG_CANCEL:
891         case LDLM_ENQUEUE:
892                 *process = target_queue_recovery_request(req, obd);
893                 RETURN(0);
894
895         default:
896                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
897                 *process = 0;
898                 /* XXX what should we set rq_status to here? */
899                 req->rq_status = -EAGAIN;
900                 RETURN(ptlrpc_error(req));
901         }
902 }
903
904
905
906 static int ost_handle(struct ptlrpc_request *req)
907 {
908         struct obd_trans_info trans_info = { 0, };
909         struct obd_trans_info *oti = &trans_info;
910         int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
911         struct obd_export *exp = NULL;
912         ENTRY;
913
914         LASSERT(current->journal_info == NULL);
915         /* XXX identical to MDS */
916         if (req->rq_reqmsg->opc != OST_CONNECT) {
917                 struct obd_device *obd;
918                 int abort_recovery, recovering;
919
920                 exp = req->rq_export;
921
922                 if (exp == NULL) {
923                         CDEBUG(D_HA, "operation %d on unconnected OST\n",
924                                req->rq_reqmsg->opc);
925                         req->rq_status = -ENOTCONN;
926                         GOTO(out, rc = -ENOTCONN);
927                 }
928
929                 obd = exp->exp_obd;
930
931                 /* Check for aborted recovery. */
932                 spin_lock_bh(&obd->obd_processing_task_lock);
933                 abort_recovery = obd->obd_abort_recovery;
934                 recovering = obd->obd_recovering;
935                 spin_unlock_bh(&obd->obd_processing_task_lock);
936                 if (abort_recovery) {
937                         target_abort_recovery(obd);
938                 } else if (recovering) {
939                         rc = ost_filter_recovery_request(req, obd,
940                                                          &should_process);
941                         if (rc || !should_process)
942                                 RETURN(rc);
943                 }
944         }
945
946         oti_init(oti, req);
947
948         switch (req->rq_reqmsg->opc) {
949         case OST_CONNECT:
950                 CDEBUG(D_INODE, "connect\n");
951                 OBD_FAIL_RETURN(OBD_FAIL_OST_CONNECT_NET, 0);
952                 rc = target_handle_connect(req, ost_handle);
953                 break;
954         case OST_DISCONNECT:
955                 CDEBUG(D_INODE, "disconnect\n");
956                 OBD_FAIL_RETURN(OBD_FAIL_OST_DISCONNECT_NET, 0);
957                 rc = target_handle_disconnect(req);
958                 break;
959         case OST_CREATE:
960                 CDEBUG(D_INODE, "create\n");
961                 OBD_FAIL_RETURN(OBD_FAIL_OST_CREATE_NET, 0);
962                 rc = ost_create(exp, req, oti);
963                 break;
964         case OST_DESTROY:
965                 CDEBUG(D_INODE, "destroy\n");
966                 OBD_FAIL_RETURN(OBD_FAIL_OST_DESTROY_NET, 0);
967                 rc = ost_destroy(exp, req, oti);
968                 break;
969         case OST_GETATTR:
970                 CDEBUG(D_INODE, "getattr\n");
971                 OBD_FAIL_RETURN(OBD_FAIL_OST_GETATTR_NET, 0);
972                 rc = ost_getattr(exp, req);
973                 break;
974         case OST_SETATTR:
975                 CDEBUG(D_INODE, "setattr\n");
976                 OBD_FAIL_RETURN(OBD_FAIL_OST_SETATTR_NET, 0);
977                 rc = ost_setattr(exp, req, oti);
978                 break;
979         case OST_WRITE:
980                 CDEBUG(D_INODE, "write\n");
981                 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
982                 rc = ost_brw_write(req, oti);
983                 LASSERT(current->journal_info == NULL);
984                 /* ost_brw sends its own replies */
985                 RETURN(rc);
986         case OST_READ:
987                 CDEBUG(D_INODE, "read\n");
988                 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
989                 rc = ost_brw_read(req);
990                 LASSERT(current->journal_info == NULL);
991                 /* ost_brw sends its own replies */
992                 RETURN(rc);
993         case OST_SAN_READ:
994                 CDEBUG(D_INODE, "san read\n");
995                 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
996                 rc = ost_san_brw(req, OBD_BRW_READ);
997                 /* ost_san_brw sends its own replies */
998                 RETURN(rc);
999         case OST_SAN_WRITE:
1000                 CDEBUG(D_INODE, "san write\n");
1001                 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
1002                 rc = ost_san_brw(req, OBD_BRW_WRITE);
1003                 /* ost_san_brw sends its own replies */
1004                 RETURN(rc);
1005         case OST_PUNCH:
1006                 CDEBUG(D_INODE, "punch\n");
1007                 OBD_FAIL_RETURN(OBD_FAIL_OST_PUNCH_NET, 0);
1008                 rc = ost_punch(exp, req, oti);
1009                 break;
1010         case OST_STATFS:
1011                 CDEBUG(D_INODE, "statfs\n");
1012                 OBD_FAIL_RETURN(OBD_FAIL_OST_STATFS_NET, 0);
1013                 rc = ost_statfs(req);
1014                 break;
1015         case OST_SYNC:
1016                 CDEBUG(D_INODE, "sync\n");
1017                 OBD_FAIL_RETURN(OBD_FAIL_OST_SYNC_NET, 0);
1018                 rc = ost_sync(exp, req);
1019                 break;
1020         case OST_SET_INFO:
1021                 DEBUG_REQ(D_INODE, req, "set_info");
1022                 rc = ost_set_info(exp, req);
1023                 break;
1024         case OST_GET_INFO:
1025                 DEBUG_REQ(D_INODE, req, "get_info");
1026                 rc = ost_get_info(exp, req);
1027                 break;
1028         case OBD_PING:
1029                 DEBUG_REQ(D_INODE, req, "ping");
1030                 rc = target_handle_ping(req);
1031                 break;
1032 #ifdef ENABLE_ORPHANS
1033         /* FIXME - just reply status */
1034         case LLOG_ORIGIN_CONNECT:
1035                 DEBUG_REQ(D_INODE, req, "log connect\n");
1036                 rc = llog_handle_connect(req); 
1037                 req->rq_status = rc;
1038                 rc = lustre_pack_reply(req, 0, NULL, NULL);
1039                 if (rc)
1040                         RETURN(rc);
1041                 RETURN(ptlrpc_reply(req));
1042                 //break;
1043         case OBD_LOG_CANCEL:
1044                 CDEBUG(D_INODE, "log cancel\n");
1045                 OBD_FAIL_RETURN(OBD_FAIL_OBD_LOG_CANCEL_NET, 0);
1046                 rc = llog_origin_handle_cancel(req);
1047                 req->rq_status = rc;
1048                 rc = lustre_pack_reply(req, 0, NULL, NULL);
1049                 if (rc)
1050                         RETURN(rc);
1051                 RETURN(ptlrpc_reply(req));
1052                 //break;
1053 #endif
1054         case LDLM_ENQUEUE:
1055                 CDEBUG(D_INODE, "enqueue\n");
1056                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_ENQUEUE, 0);
1057                 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
1058                                          ldlm_server_blocking_ast);
1059                 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
1060                 break;
1061         case LDLM_CONVERT:
1062                 CDEBUG(D_INODE, "convert\n");
1063                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CONVERT, 0);
1064                 rc = ldlm_handle_convert(req);
1065                 break;
1066         case LDLM_CANCEL:
1067                 CDEBUG(D_INODE, "cancel\n");
1068                 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CANCEL, 0);
1069                 rc = ldlm_handle_cancel(req);
1070                 break;
1071         case LDLM_BL_CALLBACK:
1072         case LDLM_CP_CALLBACK:
1073                 CDEBUG(D_INODE, "callback\n");
1074                 CERROR("callbacks should not happen on OST\n");
1075                 /* fall through */
1076         default:
1077                 CERROR("Unexpected opcode %d\n", req->rq_reqmsg->opc);
1078                 req->rq_status = -ENOTSUPP;
1079                 rc = ptlrpc_error(req);
1080                 RETURN(rc);
1081         }
1082
1083         LASSERT(current->journal_info == NULL);
1084
1085         EXIT;
1086         /* If we're DISCONNECTing, the export_data is already freed */
1087         if (!rc && req->rq_reqmsg->opc != OST_DISCONNECT) {
1088                 struct obd_device *obd  = req->rq_export->exp_obd;
1089                 if (!obd->obd_no_transno) {
1090                         req->rq_repmsg->last_committed =
1091                                 obd->obd_last_committed;
1092                 } else {
1093                         DEBUG_REQ(D_IOCTL, req,
1094                                   "not sending last_committed update");
1095                 }
1096                 CDEBUG(D_INFO, "last_committed "LPU64", xid "LPX64"\n",
1097                        obd->obd_last_committed, req->rq_xid);
1098         }
1099
1100 out:
1101         if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LAST_REPLAY) {
1102                 struct obd_device *obd = req->rq_export->exp_obd;
1103
1104                 if (obd && obd->obd_recovering) {
1105                         DEBUG_REQ(D_HA, req, "LAST_REPLAY, queuing reply");
1106                         return target_queue_final_reply(req, rc);
1107                 }
1108                 /* Lost a race with recovery; let the error path DTRT. */
1109                 rc = req->rq_status = -ENOTCONN;
1110         }
1111
1112         if (!rc)
1113                 oti_to_request(oti, req);
1114
1115         target_send_reply(req, rc, fail);
1116         return 0;
1117 }
1118
1119 static int ost_setup(struct obd_device *obddev, obd_count len, void *buf)
1120 {
1121         struct ost_obd *ost = &obddev->u.ost;
1122         int rc;
1123         ENTRY;
1124
1125         /* Get rid of unneeded supplementary groups */
1126         current->ngroups = 0;
1127         memset(current->groups, 0, sizeof(current->groups));
1128
1129         rc = llog_start_commit_thread();
1130         if (rc < 0)
1131                 RETURN(rc);
1132
1133         ost->ost_service = ptlrpc_init_svc(OST_NEVENTS, OST_NBUFS,
1134                                            OST_BUFSIZE, OST_MAXREQSIZE,
1135                                            OST_REQUEST_PORTAL, OSC_REPLY_PORTAL,
1136                                            ost_handle, "ost", 
1137                                            obddev->obd_proc_entry);
1138         if (ost->ost_service == NULL) {
1139                 CERROR("failed to start service\n");
1140                 RETURN(-ENOMEM);
1141         }
1142         
1143         rc = ptlrpc_start_n_threads(obddev, ost->ost_service, OST_NUM_THREADS, 
1144                                  "ll_ost");
1145         if (rc)
1146                 GOTO(out, rc = -EINVAL);
1147
1148         ost->ost_create_service =
1149                 ptlrpc_init_svc(OST_NEVENTS, OST_NBUFS, OST_BUFSIZE,
1150                                 OST_MAXREQSIZE, OST_CREATE_PORTAL,
1151                                 OSC_REPLY_PORTAL, ost_handle, "ost_create",
1152                                 obddev->obd_proc_entry);
1153         if (ost->ost_create_service == NULL) {
1154                 CERROR("failed to start OST create service\n");
1155                 GOTO(out, rc = -ENOMEM);
1156         }
1157
1158         rc = ptlrpc_start_n_threads(obddev, ost->ost_create_service, 1,
1159                                     "ll_ost_create");
1160         if (rc) 
1161                 GOTO(out_create, rc = -EINVAL);
1162
1163         RETURN(0);
1164
1165 out_create:
1166         ptlrpc_unregister_service(ost->ost_create_service);
1167 out:
1168         ptlrpc_unregister_service(ost->ost_service);
1169         RETURN(rc);
1170 }
1171
1172 static int ost_cleanup(struct obd_device *obddev, int flags)
1173 {
1174         struct ost_obd *ost = &obddev->u.ost;
1175         int err = 0;
1176         ENTRY;
1177
1178         if (obddev->obd_recovering)
1179                 target_cancel_recovery_timer(obddev);
1180
1181         ptlrpc_stop_all_threads(ost->ost_service);
1182         ptlrpc_unregister_service(ost->ost_service);
1183
1184         ptlrpc_stop_all_threads(ost->ost_create_service);
1185         ptlrpc_unregister_service(ost->ost_create_service);
1186
1187         RETURN(err);
1188 }
1189
1190 int ost_attach(struct obd_device *dev, obd_count len, void *data)
1191 {
1192         struct lprocfs_static_vars lvars;
1193
1194         lprocfs_init_vars(ost,&lvars);
1195         return lprocfs_obd_attach(dev, lvars.obd_vars);
1196 }
1197
1198 int ost_detach(struct obd_device *dev)
1199 {
1200         return lprocfs_obd_detach(dev);
1201 }
1202
1203 /* use obd ops to offer management infrastructure */
1204 static struct obd_ops ost_obd_ops = {
1205         o_owner:        THIS_MODULE,
1206         o_attach:       ost_attach,
1207         o_detach:       ost_detach,
1208         o_setup:        ost_setup,
1209         o_cleanup:      ost_cleanup,
1210 };
1211
1212 static int __init ost_init(void)
1213 {
1214         struct lprocfs_static_vars lvars;
1215         ENTRY;
1216
1217         lprocfs_init_vars(ost,&lvars);
1218         RETURN(class_register_type(&ost_obd_ops, lvars.module_vars,
1219                                    LUSTRE_OST_NAME));
1220 }
1221
1222 static void /*__exit*/ ost_exit(void)
1223 {
1224         class_unregister_type(LUSTRE_OST_NAME);
1225 }
1226
1227 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1228 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
1229 MODULE_LICENSE("GPL");
1230
1231 module_init(ost_init);
1232 module_exit(ost_exit);