Whamcloud - gitweb
- if MDS aborts recovery it disconnects all the exports and lmv_disconnect()
[fs/lustre-release.git] / lustre / lmv / lmv_obd.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  *
4  * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
5  *
6  *   This file is part of Lustre, http://www.lustre.org.
7  *
8  *   Lustre is free software; you can redistribute it and/or
9  *   modify it under the terms of version 2 of the GNU General Public
10  *   License as published by the Free Software Foundation.
11  *
12  *   Lustre is distributed in the hope that it will be useful,
13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  *   GNU General Public License for more details.
16  *
17  *   You should have received a copy of the GNU General Public License
18  *   along with Lustre; if not, write to the Free Software
19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20  */
21
22 #ifndef EXPORT_SYMTAB
23 # define EXPORT_SYMTAB
24 #endif
25 #define DEBUG_SUBSYSTEM S_LMV
26 #ifdef __KERNEL__
27 #include <linux/slab.h>
28 #include <linux/module.h>
29 #include <linux/init.h>
30 #include <linux/slab.h>
31 #include <linux/pagemap.h>
32 #include <asm/div64.h>
33 #include <linux/seq_file.h>
34 #else
35 #include <liblustre.h>
36 #endif
37 #include <linux/ext2_fs.h>
38
39 #include <linux/obd_support.h>
40 #include <linux/lustre_lib.h>
41 #include <linux/lustre_net.h>
42 #include <linux/lustre_idl.h>
43 #include <linux/lustre_dlm.h>
44 #include <linux/lustre_mds.h>
45 #include <linux/obd_class.h>
46 #include <linux/obd_ost.h>
47 #include <linux/lprocfs_status.h>
48 #include <linux/lustre_fsfilt.h>
49 #include <linux/obd_lmv.h>
50 #include "lmv_internal.h"
51
52 /* Error codes:
53  *
54  *  -EINVAL  : UUID can't be found in the LMV's target list
55  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
56  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
57  */
58 static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid,
59                               int activate)
60 {
61         struct obd_device *obd;
62         struct lmv_tgt_desc *tgt;
63         int i, rc = 0;
64         ENTRY;
65
66         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
67                lmv, uuid->uuid, activate);
68
69         spin_lock(&lmv->lmv_lock);
70         for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) {
71                 if (tgt->ltd_exp == NULL)
72                         continue;
73
74                 CDEBUG(D_INFO, "lmv idx %d is %s conn "LPX64"\n",
75                        i, tgt->uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
76                 if (strncmp(uuid->uuid, tgt->uuid.uuid, sizeof uuid->uuid) == 0)
77                         break;
78         }
79
80         if (i == lmv->desc.ld_tgt_count)
81                 GOTO(out, rc = -EINVAL);
82
83         obd = class_exp2obd(tgt->ltd_exp);
84         if (obd == NULL)
85                 GOTO(out, rc = -ENOTCONN);
86
87         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
88                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
89                obd->obd_type->typ_name, i);
90         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
91
92         if (tgt->active == activate) {
93                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
94                        activate ? "" : "in");
95                 GOTO(out, rc);
96         }
97
98         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd, activate ? "" : "in");
99
100         tgt->active = activate;
101         if (activate)
102                 lmv->desc.ld_active_tgt_count++;
103         else
104                 lmv->desc.ld_active_tgt_count--;
105
106         EXIT;
107  out:
108         spin_unlock(&lmv->lmv_lock);
109         return rc;
110 }
111
112 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
113                       int active)
114 {
115         int rc;
116         struct obd_uuid *uuid;
117
118         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
119                 CERROR("unexpected notification of %s %s!\n",
120                        watched->obd_type->typ_name,
121                        watched->obd_name);
122                 return -EINVAL;
123         }
124         uuid = &watched->u.cli.cl_import->imp_target_uuid;
125
126         /* Set MDC as active before notifying the observer, so the
127          * observer can use the MDC normally.  
128          */
129         rc = lmv_set_mdc_active(&obd->u.lmv, uuid, active);
130         if (rc) {
131                 CERROR("%sactivation of %s failed: %d\n",
132                        active ? "" : "de", uuid->uuid, rc);
133                 RETURN(rc);
134         }
135
136         if (obd->obd_observer)
137                 /* Pass the notification up the chain. */
138                 rc = obd_notify(obd->obd_observer, watched, active);
139
140         RETURN(rc);
141 }
142
143 int lmv_attach(struct obd_device *dev, obd_count len, void *data)
144 {
145         struct lprocfs_static_vars lvars;
146         int rc;
147         ENTRY;
148
149         lprocfs_init_vars(lmv, &lvars);
150         rc = lprocfs_obd_attach(dev, lvars.obd_vars);
151         if (rc == 0) {
152 #ifdef __KERNEL__
153                 struct proc_dir_entry *entry;
154                 
155                 entry = create_proc_entry("target_obd", 0444, dev->obd_proc_entry);
156                 if (entry == NULL)
157                         RETURN(-ENOMEM);
158                 /* entry->proc_fops = &lmv_proc_target_fops; */
159                 entry->data = dev;
160 #endif
161        }
162         RETURN (rc);
163 }
164
165 int lmv_detach(struct obd_device *dev)
166 {
167         return lprocfs_obd_detach(dev);
168 }
169
170 /* This is fake connect function. Its purpose is to initialize lmv and 
171  * say caller that everything is okay. Real connection will be performed
172  * later. */
173 static int lmv_connect(struct lustre_handle *conn, struct obd_device *obd,
174                        struct obd_uuid *cluuid)
175 {
176         struct lmv_obd *lmv = &obd->u.lmv;
177         struct obd_export *exp;
178         int rc;
179         ENTRY;
180
181         rc = class_connect(conn, obd, cluuid);
182         if (rc) {
183                 CERROR("class_connection() returned %d\n", rc);
184                 RETURN(rc);
185         }
186
187         exp = class_conn2export(conn);
188         /* We don't want to actually do the underlying connections more than
189          * once, so keep track. */
190         lmv->refcount++;
191         if (lmv->refcount > 1) {
192                 class_export_put(exp);
193                 RETURN(0);
194         }
195
196         lmv->cluuid = *cluuid;
197         lmv->connected = 0;
198         lmv->exp = exp;
199
200         RETURN(0);
201 }
202
203 void lmv_set_timeouts(struct obd_device *obd)
204 {
205         struct lmv_tgt_desc *tgts;
206         struct lmv_obd *lmv;
207         int i;
208
209         lmv = &obd->u.lmv;
210         if (lmv->server_timeout == 0)
211                 return;
212
213         if (lmv->connected == 0)
214                 return;
215
216         for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
217                 if (tgts->ltd_exp == NULL)
218                         continue;
219                 obd_set_info(tgts->ltd_exp, strlen("inter_mds"),
220                              "inter_mds", 0, NULL);
221         }
222 }
223
224 /* Performs a check if passed obd is connected. If no - connect it. */
225 int lmv_check_connect(struct obd_device *obd) {
226         struct lmv_obd *lmv = &obd->u.lmv;
227         struct obd_uuid *cluuid;
228         struct lmv_tgt_desc *tgts;
229         struct obd_export *exp;
230         int rc, rc2, i;
231
232         if (lmv->connected)
233                 return 0;
234       
235         lmv->connected = 1;
236         cluuid = &lmv->cluuid;
237         exp = lmv->exp;
238         
239         CDEBUG(D_OTHER, "time to connect %s to %s\n",
240                cluuid->uuid, obd->obd_name);
241
242         for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
243                 struct obd_device *tgt_obd;
244                 struct obd_uuid lmv_osc_uuid = { "LMV_OSC_UUID" };
245                 struct lustre_handle conn = {0, };
246
247                 LASSERT(tgts != NULL);
248
249                 tgt_obd = class_find_client_obd(&tgts->uuid, LUSTRE_MDC_NAME, 
250                                                 &obd->obd_uuid);
251                 if (!tgt_obd) {
252                         CERROR("Target %s not attached\n", tgts->uuid.uuid);
253                         GOTO(out_disc, rc = -EINVAL);
254                 }
255
256                 /* for MDS: don't connect to yourself */
257                 if (obd_uuid_equals(&tgts->uuid, cluuid)) {
258                         CDEBUG(D_OTHER, "don't connect back to %s\n",
259                                cluuid->uuid);
260                         tgts->ltd_exp = NULL;
261                         continue;
262                 }
263
264                 CDEBUG(D_OTHER, "connect to %s(%s) - %s, %s FOR %s\n",
265                         tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
266                         tgts->uuid.uuid, obd->obd_uuid.uuid,
267                         cluuid->uuid);
268
269                 if (!tgt_obd->obd_set_up) {
270                         CERROR("Target %s not set up\n", tgts->uuid.uuid);
271                         GOTO(out_disc, rc = -EINVAL);
272                 }
273                 
274                 rc = obd_connect(&conn, tgt_obd, &lmv_osc_uuid);
275                 if (rc) {
276                         CERROR("Target %s connect error %d\n",
277                                 tgts->uuid.uuid, rc);
278                         GOTO(out_disc, rc);
279                 }
280                 tgts->ltd_exp = class_conn2export(&conn);
281
282                 obd_init_ea_size(tgts->ltd_exp, lmv->max_easize,
283                                  lmv->max_cookiesize);
284                 
285                 rc = obd_register_observer(tgt_obd, obd);
286                 if (rc) {
287                         CERROR("Target %s register_observer error %d\n",
288                                tgts->uuid.uuid, rc);
289                         obd_disconnect(tgts->ltd_exp, 0);
290                         GOTO(out_disc, rc);
291                 }
292
293                 lmv->desc.ld_active_tgt_count++;
294                 tgts->active = 1;
295                 
296                 CDEBUG(D_OTHER, "connected to %s(%s) successfully (%d)\n",
297                         tgt_obd->obd_name, tgt_obd->obd_uuid.uuid,
298                         atomic_read(&obd->obd_refcount));
299         }
300
301         lmv_set_timeouts(obd);
302
303         class_export_put(exp);
304         return 0;
305
306  out_disc:
307         while (i-- > 0) {
308                 struct obd_uuid uuid;
309                 --tgts;
310                 --lmv->desc.ld_active_tgt_count;
311                 tgts->active = 0;
312                 /* save for CERROR below; (we know it's terminated) */
313                 uuid = tgts->uuid;
314                 rc2 = obd_disconnect(tgts->ltd_exp, 0);
315                 if (rc2)
316                         CERROR("error: LMV target %s disconnect on MDT idx %d: "
317                                "error %d\n", uuid.uuid, i, rc2);
318         }
319         class_disconnect(exp, 0);
320         RETURN (rc);
321 }
322
323 static int lmv_disconnect(struct obd_export *exp, int flags)
324 {
325         struct obd_device *obd = class_exp2obd(exp);
326         struct lmv_obd *lmv = &obd->u.lmv;
327         int rc, i;
328         ENTRY;
329
330         if (!lmv->tgts)
331                 goto out_local;
332
333         /* Only disconnect the underlying layers on the final disconnect. */
334         lmv->refcount--;
335         if (lmv->refcount != 0)
336                 goto out_local;
337
338         lmv->connected = 0;
339
340         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
341                 if (lmv->tgts[i].ltd_exp == NULL)
342                         continue;
343
344                 if (obd->obd_no_recov) {
345                         /* Pass it on to our clients.
346                          * XXX This should be an argument to disconnect,
347                          * XXX not a back-door flag on the OBD.  Ah well.
348                          */
349                         struct obd_device *mdc_obd;
350                         mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp);
351                         if (mdc_obd)
352                                 mdc_obd->obd_no_recov = 1;
353                 }
354
355                 CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n",
356                         lmv->tgts[i].ltd_exp->exp_obd->obd_name,
357                         lmv->tgts[i].ltd_exp->exp_obd->obd_uuid.uuid);
358
359                 obd_register_observer(lmv->tgts[i].ltd_exp->exp_obd, NULL);
360
361                 rc = obd_disconnect(lmv->tgts[i].ltd_exp, flags);
362                 if (rc) {
363                         if (lmv->tgts[i].active) {
364                                 CERROR("Target %s disconnect error %d\n",
365                                        lmv->tgts[i].uuid.uuid, rc);
366                         }
367                         rc = 0;
368                 }
369                 if (lmv->tgts[i].active) {
370                         lmv->desc.ld_active_tgt_count--;
371                         lmv->tgts[i].active = 0;
372                 }
373                 lmv->tgts[i].ltd_exp = NULL;
374         }
375
376 out_local:
377         /* this is the case when no real connection is established by
378          * lmv_check_connect(). */
379         if (!lmv->connected)
380                 class_export_put(exp);
381         rc = class_disconnect(exp, 0);
382         RETURN(rc);
383 }
384
385 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
386                          int len, void *karg, void *uarg)
387 {
388         struct obd_device *obddev = class_exp2obd(exp);
389         struct lmv_obd *lmv = &obddev->u.lmv;
390         int i, rc = 0, set = 0;
391
392         ENTRY;
393
394         if (lmv->desc.ld_tgt_count == 0)
395                 RETURN(-ENOTTY);
396         
397         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
398                 int err;
399
400                 if (lmv->tgts[i].ltd_exp == NULL) {
401                         CWARN("%s: NULL export for %d\n", obddev->obd_name, i);
402                         continue;
403                 }
404
405                 err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, karg, uarg);
406                 if (err) {
407                         if (lmv->tgts[i].active) {
408                                 CERROR("error: iocontrol MDC %s on MDT"
409                                        "idx %d: err = %d\n",
410                                        lmv->tgts[i].uuid.uuid, i, err);
411                                 if (!rc)
412                                         rc = err;
413                         }
414                 } else
415                         set = 1;
416         }
417         if (!set && !rc)
418                 rc = -EIO;
419
420         RETURN(rc);
421 }
422
423 static int lmv_setup(struct obd_device *obd, obd_count len, void *buf)
424 {
425         int i, rc = 0;
426         struct lmv_desc *desc;
427         struct obd_uuid *uuids;
428         struct lmv_tgt_desc *tgts;
429         struct lustre_cfg *lcfg = buf;
430         struct lmv_obd *lmv = &obd->u.lmv;
431         ENTRY;
432
433         if (lcfg->lcfg_inllen1 < 1) {
434                 CERROR("LMV setup requires a descriptor\n");
435                 RETURN(-EINVAL);
436         }
437
438         if (lcfg->lcfg_inllen2 < 1) {
439                 CERROR("LMV setup requires an OST UUID list\n");
440                 RETURN(-EINVAL);
441         }
442
443         desc = (struct lmv_desc *)lcfg->lcfg_inlbuf1;
444         if (sizeof(*desc) > lcfg->lcfg_inllen1) {
445                 CERROR("descriptor size wrong: %d > %d\n",
446                        (int)sizeof(*desc), lcfg->lcfg_inllen1);
447                 RETURN(-EINVAL);
448         }
449
450         uuids = (struct obd_uuid *)lcfg->lcfg_inlbuf2;
451         if (sizeof(*uuids) * desc->ld_tgt_count != lcfg->lcfg_inllen2) {
452                 CERROR("UUID array size wrong: %u * %u != %u\n",
453                        sizeof(*uuids), desc->ld_tgt_count, lcfg->lcfg_inllen2);
454                 RETURN(-EINVAL);
455         }
456
457         lmv->bufsize = sizeof(struct lmv_tgt_desc) * desc->ld_tgt_count;
458         OBD_ALLOC(lmv->tgts, lmv->bufsize);
459         if (lmv->tgts == NULL) {
460                 CERROR("Out of memory\n");
461                 RETURN(-ENOMEM);
462         }
463
464         lmv->desc = *desc;
465         spin_lock_init(&lmv->lmv_lock);
466         
467         for (i = 0, tgts = lmv->tgts; i < desc->ld_tgt_count; i++, tgts++)
468                 tgts->uuid = uuids[i];
469         
470         lmv->max_cookiesize = 0;
471
472         lmv->max_easize = sizeof(struct ll_fid) *
473                 desc->ld_tgt_count + sizeof(struct mea);
474         
475         rc = lmv_setup_mgr(obd);
476         if (rc) {
477                 CERROR("Can't setup LMV object manager, "
478                        "error %d.\n", rc);
479                 OBD_FREE(lmv->tgts, lmv->bufsize);
480         }
481
482         RETURN(rc);
483 }
484
485 static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs,
486                       unsigned long max_age)
487 {
488         struct lmv_obd *lmv = &obd->u.lmv;
489         struct obd_statfs temp;
490         int rc = 0, i;
491         ENTRY;
492         
493         rc = lmv_check_connect(obd);
494         if (rc)
495                 RETURN(rc);
496                 
497         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
498                 if (lmv->tgts[i].ltd_exp == NULL) {
499                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
500                         continue;
501                 }
502
503                 rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, &temp, max_age);
504                 if (rc) {
505                         CERROR("can't stat MDS #%d (%s)\n", i,
506                                lmv->tgts[i].ltd_exp->exp_obd->obd_name);
507                         RETURN(rc);
508                 }
509                 if (i == 0) {
510                         memcpy(osfs, &temp, sizeof(temp));
511                 } else {
512                         osfs->os_bavail += temp.os_bavail;
513                         osfs->os_blocks += temp.os_blocks;
514                         osfs->os_ffree += temp.os_ffree;
515                         osfs->os_files += temp.os_files;
516                 }
517         }
518         RETURN(rc);
519 }
520
521 static int lmv_cleanup(struct obd_device *obd, int flags) 
522 {
523         struct lmv_obd *lmv = &obd->u.lmv;
524         ENTRY;
525         lmv_cleanup_mgr(obd);
526         OBD_FREE(lmv->tgts, lmv->bufsize);
527         RETURN(0);
528 }
529
530 static int lmv_getstatus(struct obd_export *exp, struct ll_fid *fid)
531 {
532         struct obd_device *obd = exp->exp_obd;
533         struct lmv_obd *lmv = &obd->u.lmv;
534         int rc;
535         ENTRY;
536         rc = lmv_check_connect(obd);
537         if (rc)
538                 RETURN(rc);
539         rc = md_getstatus(lmv->tgts[0].ltd_exp, fid);
540         fid->mds = 0;
541         RETURN(rc);
542 }
543
544 static int lmv_getattr(struct obd_export *exp, struct ll_fid *fid,
545                        unsigned long valid, unsigned int ea_size,
546                        struct ptlrpc_request **request)
547 {
548         struct obd_device *obd = exp->exp_obd;
549         struct lmv_obd *lmv = &obd->u.lmv;
550         int rc, i = fid->mds;
551         struct lmv_obj *obj;
552         ENTRY;
553
554         rc = lmv_check_connect(obd);
555         if (rc)
556                 RETURN(rc);
557
558         LASSERT(i < lmv->desc.ld_tgt_count);
559
560         rc = md_getattr(lmv->tgts[i].ltd_exp, fid, valid,
561                         ea_size, request);
562         if (rc)
563                 RETURN(rc);
564         
565         obj = lmv_grab_obj(obd, fid);
566         
567         CDEBUG(D_OTHER, "GETATTR for %lu/%lu/%lu %s\n",
568                (unsigned long)fid->mds, (unsigned long)fid->id,
569                (unsigned long)fid->generation, obj ? "(splitted)" : "");
570
571         /* if object is splitted, then we loop over all the slaves and gather
572          * size attribute. In ideal world we would have to gather also mds field
573          * from all slaves, as object is spread over the cluster and this is
574          * definitely interesting information and it is not good to loss it,
575          * but...*/
576         if (obj) {
577                 struct mds_body *body;
578
579                 if (*request == NULL) {
580                         lmv_put_obj(obj);
581                         RETURN(rc);
582                 }
583                         
584                 body = lustre_msg_buf((*request)->rq_repmsg, 0,
585                                       sizeof(*body));
586                 LASSERT(body != NULL);
587
588                 lmv_lock_obj(obj);
589         
590                 for (i = 0; i < obj->objcount; i++) {
591
592                         if (lmv->tgts[i].ltd_exp == NULL) {
593                                 CWARN("%s: NULL export for %d\n",
594                                       obd->obd_name, i);
595                                 continue;
596                         }
597
598                         /* skip master obj. */
599                         if (fid_equal(&obj->fid, &obj->objs[i].fid))
600                                 continue;
601                         
602                         body->size += obj->objs[i].size;
603                 }
604
605                 lmv_unlock_obj(obj);
606                 lmv_put_obj(obj);
607         }
608         
609         RETURN(rc);
610 }
611
612 static int lmv_change_cbdata(struct obd_export *exp, struct ll_fid *fid, 
613                              ldlm_iterator_t it, void *data)
614 {
615         struct obd_device *obd = exp->exp_obd;
616         struct lmv_obd *lmv = &obd->u.lmv;
617         int rc = 0;
618         ENTRY;
619         
620         rc = lmv_check_connect(obd);
621         if (rc)
622                 RETURN(rc);
623         
624         CDEBUG(D_OTHER, "CBDATA for %lu/%lu/%lu\n", (unsigned long)fid->mds,
625                (unsigned long)fid->id, (unsigned long)fid->generation);
626         
627         LASSERT(fid->mds < lmv->desc.ld_tgt_count);
628
629         rc = md_change_cbdata(lmv->tgts[fid->mds].ltd_exp,
630                               fid, it, data);
631         
632         RETURN(rc);
633 }
634
635 static int lmv_change_cbdata_name(struct obd_export *exp, struct ll_fid *pfid,
636                                   char *name, int len, struct ll_fid *cfid,
637                                   ldlm_iterator_t it, void *data)
638 {
639         struct obd_device *obd = exp->exp_obd;
640         struct lmv_obd *lmv = &obd->u.lmv;
641         struct lmv_obj *obj;
642         int rc = 0, mds;
643         ENTRY;
644
645         rc = lmv_check_connect(obd);
646         if (rc)
647                 RETURN(rc);
648
649         LASSERT(pfid->mds < lmv->desc.ld_tgt_count);
650         LASSERT(cfid->mds < lmv->desc.ld_tgt_count);
651         
652         CDEBUG(D_OTHER, "CBDATA for %lu/%lu/%lu:%*s -> %lu/%lu/%lu\n",
653                (unsigned long)pfid->mds, (unsigned long)pfid->id,
654                (unsigned long)pfid->generation, len, name,
655                (unsigned long)cfid->mds, (unsigned long)cfid->id,
656                (unsigned long)cfid->generation);
657
658         /* this is default mds for directory name belongs to. */
659         mds = pfid->mds;
660         obj = lmv_grab_obj(obd, pfid);
661         if (obj) {
662                 /* directory is splitted. look for right mds for this name. */
663                 mds = raw_name2idx(obj->objcount, name, len);
664                 mds = obj->objs[mds].fid.mds;
665                 lmv_put_obj(obj);
666         }
667         rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, cfid, it, data);
668         RETURN(rc);
669 }
670
671 static int lmv_valid_attrs(struct obd_export *exp, struct ll_fid *fid) 
672 {
673         struct obd_device *obd = exp->exp_obd;
674         struct lmv_obd *lmv = &obd->u.lmv;
675         int rc = 0;
676         ENTRY;
677         rc = lmv_check_connect(obd);
678         if (rc)
679                 RETURN(rc);
680         CDEBUG(D_OTHER, "validate %lu/%lu/%lu\n", (unsigned long) fid->mds,
681                (unsigned long) fid->id, (unsigned long) fid->generation);
682         LASSERT(fid->mds < lmv->desc.ld_tgt_count);
683         rc = md_valid_attrs(lmv->tgts[fid->mds].ltd_exp, fid);
684         RETURN(rc);
685 }
686
687 int lmv_close(struct obd_export *exp, struct obdo *obdo,
688                   struct obd_client_handle *och,
689                   struct ptlrpc_request **request)
690 {
691         struct obd_device *obd = exp->exp_obd;
692         struct lmv_obd *lmv = &obd->u.lmv;
693         int rc, i = obdo->o_mds;
694         ENTRY;
695         rc = lmv_check_connect(obd);
696         if (rc)
697                 RETURN(rc);
698         LASSERT(i < lmv->desc.ld_tgt_count);
699         CDEBUG(D_OTHER, "CLOSE %lu/%lu/%lu\n", (unsigned long) obdo->o_mds,
700                (unsigned long) obdo->o_id, (unsigned long) obdo->o_generation);
701         rc = md_close(lmv->tgts[i].ltd_exp, obdo, och, request);
702         RETURN(rc);
703 }
704
705 int lmv_get_mea_and_update_object(struct obd_export *exp, struct ll_fid *fid)
706 {
707         struct obd_device *obd = exp->exp_obd;
708         struct lmv_obd *lmv = &obd->u.lmv;
709         struct ptlrpc_request *req = NULL;
710         struct lmv_obj *obj;
711         struct lustre_md md;
712         unsigned long valid;
713         int mealen, rc;
714
715         md.mea = NULL;
716         mealen = MEA_SIZE_LMV(lmv);
717         
718         valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
719
720         /* time to update mea of parent fid */
721         rc = md_getattr(lmv->tgts[fid->mds].ltd_exp, fid,
722                         valid, mealen, &req);
723         if (rc) {
724                 CERROR("md_getattr() failed, error %d\n", rc);
725                 GOTO(cleanup, rc);
726         }
727
728         rc = mdc_req2lustre_md(exp, req, 0, NULL, &md);
729         if (rc) {
730                 CERROR("mdc_req2lustre_md() failed, error %d\n", rc);
731                 GOTO(cleanup, rc);
732         }
733
734         if (md.mea == NULL)
735                 GOTO(cleanup, rc = -ENODATA);
736
737         obj = lmv_create_obj(exp, fid, md.mea);
738         if (IS_ERR(obj))
739                 rc = PTR_ERR(obj);
740         
741         lmv_put_obj(obj);
742         obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea);
743
744 cleanup:
745         if (req)
746                 ptlrpc_req_finished(req);
747         RETURN(rc);
748 }
749
750 int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data,
751                const void *data, int datalen, int mode, __u32 uid,
752                __u32 gid, __u64 rdev, struct ptlrpc_request **request)
753 {
754         struct obd_device *obd = exp->exp_obd;
755         struct lmv_obd *lmv = &obd->u.lmv;
756         struct mds_body *body;
757         struct lmv_obj *obj;
758         int rc, mds, loop = 0;
759         ENTRY;
760
761         rc = lmv_check_connect(obd);
762         if (rc)
763                 RETURN(rc);
764
765         if (!lmv->desc.ld_active_tgt_count)
766                 RETURN(-EIO);
767 repeat:
768         LASSERT(++loop <= 2);
769         obj = lmv_grab_obj(obd, &op_data->fid1);
770         if (obj) {
771                 mds = raw_name2idx(obj->objcount, op_data->name,
772                                    op_data->namelen);
773                 op_data->fid1 = obj->objs[mds].fid;
774                 lmv_put_obj(obj);
775         }
776
777         CDEBUG(D_OTHER, "CREATE '%*s' on %lu/%lu/%lu\n", op_data->namelen,
778                op_data->name, (unsigned long)op_data->fid1.mds,
779                (unsigned long)op_data->fid1.id,
780                (unsigned long)op_data->fid1.generation);
781         
782         rc = md_create(lmv->tgts[op_data->fid1.mds].ltd_exp, op_data, data,
783                        datalen, mode, uid, gid, rdev, request);
784         if (rc == 0) {
785                 if (*request == NULL)
786                         RETURN(rc);
787
788                 body = lustre_msg_buf((*request)->rq_repmsg, 0,
789                                       sizeof(*body));
790                 LASSERT(body != NULL);
791                 
792                 CDEBUG(D_OTHER, "created. id = %lu, generation = %lu, "
793                        "mds = %d\n", (unsigned long)body->fid1.id,
794                        (unsigned long)body->fid1.generation, op_data->fid1.mds);
795                 
796                 LASSERT(body->valid & OBD_MD_MDS ||
797                         body->mds == op_data->fid1.mds);
798         } else if (rc == -ERESTART) {
799                 /* directory got splitted. time to update local object and
800                  * repeat the request with proper MDS */
801                 rc = lmv_get_mea_and_update_object(exp, &op_data->fid1);
802                 if (rc == 0) {
803                         ptlrpc_req_finished(*request);
804                         goto repeat;
805                 }
806         }
807         RETURN(rc);
808 }
809
810 int lmv_done_writing(struct obd_export *exp, struct obdo *obdo)
811 {
812         struct obd_device *obd = exp->exp_obd;
813         struct lmv_obd *lmv = &obd->u.lmv;
814         int rc;
815         ENTRY;
816         rc = lmv_check_connect(obd);
817         if (rc)
818                 RETURN(rc);
819
820         /* FIXME: choose right MDC here */
821         CWARN("this method isn't implemented yet\n");
822         rc = md_done_writing(lmv->tgts[0].ltd_exp, obdo);
823         RETURN(rc);
824 }
825
826 int lmv_enqueue_slaves(struct obd_export *exp, int locktype,
827                        struct lookup_intent *it, int lockmode,
828                        struct mdc_op_data *data, struct lustre_handle *lockh,
829                        void *lmm, int lmmsize, ldlm_completion_callback cb_completion,
830                        ldlm_blocking_callback cb_blocking, void *cb_data)
831 {
832         struct obd_device *obd = exp->exp_obd;
833         struct lmv_obd *lmv = &obd->u.lmv;
834         struct mea *mea = data->mea1;
835         struct mdc_op_data data2;
836         int i, rc, mds;
837         ENTRY;
838
839         LASSERT(mea != NULL);
840         for (i = 0; i < mea->mea_count; i++) {
841                 memset(&data2, 0, sizeof(data2));
842                 data2.fid1 = mea->mea_fids[i];
843                 mds = data2.fid1.mds;
844                 
845                 if (lmv->tgts[mds].ltd_exp == NULL)
846                         continue;
847
848                 rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it, lockmode,
849                                 &data2, lockh + i, lmm, lmmsize, cb_completion,
850                                 cb_blocking, cb_data);
851                 
852                 CDEBUG(D_OTHER, "take lock on slave %lu/%lu/%lu -> %d/%d\n",
853                        (unsigned long)mea->mea_fids[i].mds,
854                        (unsigned long)mea->mea_fids[i].id,
855                        (unsigned long)mea->mea_fids[i].generation,
856                        rc, it->d.lustre.it_status);
857                 if (rc)
858                         GOTO(cleanup, rc);
859                 if (it->d.lustre.it_data) {
860                         struct ptlrpc_request *req;
861                         req = (struct ptlrpc_request *) it->d.lustre.it_data;
862                         ptlrpc_req_finished(req);
863                 }
864                 
865                 if (it->d.lustre.it_status)
866                         GOTO(cleanup, rc = it->d.lustre.it_status);
867         }
868         RETURN(0);
869         
870 cleanup:
871         /* drop all taken locks */
872         while (--i >= 0) {
873                 if (lockh[i].cookie)
874                         ldlm_lock_decref(lockh + i, lockmode);
875                 lockh[i].cookie = 0;
876         }
877         RETURN(rc);
878 }
879
880 int lmv_enqueue(struct obd_export *exp, int lock_type,
881                 struct lookup_intent *it, int lock_mode,
882                 struct mdc_op_data *data, struct lustre_handle *lockh,
883                 void *lmm, int lmmsize, ldlm_completion_callback cb_completion,
884                 ldlm_blocking_callback cb_blocking, void *cb_data)
885 {
886         struct obd_device *obd = exp->exp_obd;
887         struct lmv_obd *lmv = &obd->u.lmv;
888         struct lmv_obj *obj;
889         int rc, mds;
890         ENTRY;
891
892         rc = lmv_check_connect(obd);
893         if (rc)
894                 RETURN(rc);
895
896         if (it->it_op == IT_UNLINK) {
897                 rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode,
898                                         data, lockh, lmm, lmmsize,
899                                         cb_completion, cb_blocking, cb_data);
900                 RETURN(rc);
901         }
902
903         if (data->namelen) {
904                 obj = lmv_grab_obj(obd, &data->fid1);
905                 if (obj) {
906                         /* directory is splitted. look for right mds for this
907                          * name */
908                         mds = raw_name2idx(obj->objcount, (char *)data->name,
909                                            data->namelen);
910                         data->fid1 = obj->objs[mds].fid;
911                         lmv_put_obj(obj);
912                 }
913         }
914         CDEBUG(D_OTHER, "ENQUEUE '%s' on %lu/%lu\n", LL_IT2STR(it),
915                (unsigned long)data->fid1.id, (unsigned long)data->fid1.generation);
916         
917         rc = md_enqueue(lmv->tgts[data->fid1.mds].ltd_exp, lock_type, it,
918                         lock_mode, data, lockh, lmm, lmmsize, cb_completion,
919                         cb_blocking, cb_data);
920
921         RETURN(rc);
922 }
923
924 int lmv_getattr_name(struct obd_export *exp, struct ll_fid *fid,
925                      char *filename, int namelen, unsigned long valid,
926                      unsigned int ea_size, struct ptlrpc_request **request)
927 {
928         struct obd_device *obd = exp->exp_obd;
929         struct lmv_obd *lmv = &obd->u.lmv;
930         struct ll_fid rfid = *fid;
931         int rc, mds = fid->mds, loop = 0;
932         struct mds_body *body;
933         struct lmv_obj *obj;
934         ENTRY;
935         rc = lmv_check_connect(obd);
936         if (rc)
937                 RETURN(rc);
938 repeat:
939         LASSERT(++loop <= 2);
940         obj = lmv_grab_obj(obd, fid);
941         if (obj) {
942                 /* directory is splitted. look for right mds for this name */
943                 mds = raw_name2idx(obj->objcount, filename, namelen - 1);
944                 rfid = obj->objs[mds].fid;
945                 lmv_put_obj(obj);
946         }
947         
948         CDEBUG(D_OTHER, "getattr_name for %*s on %lu/%lu/%lu -> %lu/%lu/%lu\n",
949                namelen, filename, (unsigned long)fid->mds,
950                (unsigned long)fid->id, (unsigned long)fid->generation,
951                (unsigned long)rfid.mds, (unsigned long)rfid.id,
952                (unsigned long)rfid.generation);
953
954         rc = md_getattr_name(lmv->tgts[rfid.mds].ltd_exp, &rfid, filename,
955                              namelen, valid, ea_size, request);
956         if (rc == 0) {
957                 /* this could be cross-node reference. in this case all we have
958                  * right now is mds/ino/generation triple. we'd like to find
959                  * other attributes */
960                 body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body));
961                 LASSERT(body != NULL);
962                 if (body->valid & OBD_MD_MDS) {
963                         struct ptlrpc_request *req = NULL;
964                         rfid = body->fid1;
965                         CDEBUG(D_OTHER, "request attrs for %lu/%lu/%lu\n",
966                                (unsigned long) rfid.mds,
967                                (unsigned long) rfid.id,
968                                (unsigned long) rfid.generation);
969                         rc = md_getattr_name(lmv->tgts[rfid.mds].ltd_exp, &rfid,
970                                              NULL, 1, valid, ea_size, &req);
971                         ptlrpc_req_finished(*request);
972                         *request = req;
973                 }
974         } else if (rc == -ERESTART) {
975                 /* directory got splitted. time to update local object and
976                  * repeat the request with proper MDS */
977                 rc = lmv_get_mea_and_update_object(exp, &rfid);
978                 if (rc == 0) {
979                         ptlrpc_req_finished(*request);
980                         goto repeat;
981                 }
982         }
983         RETURN(rc);
984 }
985
986
987 /*
988  * llite passes fid of an target inode in data->fid1 and fid of directory in
989  * data->fid2
990  */
991 int lmv_link(struct obd_export *exp, struct mdc_op_data *data,
992              struct ptlrpc_request **request)
993 {
994         struct obd_device *obd = exp->exp_obd;
995         struct lmv_obd *lmv = &obd->u.lmv;
996         struct lmv_obj *obj;
997         int rc;
998         ENTRY;
999         
1000         rc = lmv_check_connect(obd);
1001         if (rc)
1002                 RETURN(rc);
1003
1004         if (data->namelen != 0) {
1005                 /* usual link request */
1006                 obj = lmv_grab_obj(obd, &data->fid1);
1007                 if (obj) {
1008                         rc = raw_name2idx(obj->objcount, data->name,
1009                                           data->namelen);
1010                         data->fid1 = obj->objs[rc].fid;
1011                         lmv_put_obj(obj);
1012                 }
1013                 
1014                 CDEBUG(D_OTHER,"link %lu/%lu/%lu:%*s to %lu/%lu/%lu mds %lu\n",
1015                        (unsigned long)data->fid2.mds,
1016                        (unsigned long)data->fid2.id,
1017                        (unsigned long)data->fid2.generation,
1018                        data->namelen, data->name,
1019                        (unsigned long)data->fid1.mds,
1020                        (unsigned long)data->fid1.id,
1021                        (unsigned long)data->fid1.generation,
1022                        (unsigned long)data->fid1.mds);
1023         } else {
1024                 /* request from MDS to acquire i_links for inode by fid1 */
1025                 CDEBUG(D_OTHER, "inc i_nlinks for %lu/%lu/%lu\n",
1026                        (unsigned long)data->fid1.mds,
1027                        (unsigned long)data->fid1.id,
1028                        (unsigned long)data->fid1.generation);
1029         }
1030                         
1031         rc = md_link(lmv->tgts[data->fid1.mds].ltd_exp, data, request);
1032         RETURN(rc);
1033 }
1034
1035 int lmv_rename(struct obd_export *exp, struct mdc_op_data *data,
1036                const char *old, int oldlen, const char *new, int newlen,
1037                struct ptlrpc_request **request)
1038 {
1039         struct obd_device *obd = exp->exp_obd;
1040         struct lmv_obd *lmv = &obd->u.lmv;
1041         struct lmv_obj *obj;
1042         int rc, mds;
1043         ENTRY;
1044
1045         CDEBUG(D_OTHER, "rename %*s in %lu/%lu/%lu to %*s in %lu/%lu/%lu\n",
1046                oldlen, old, (unsigned long)data->fid1.mds,
1047                (unsigned long)data->fid1.id,
1048                (unsigned long)data->fid1.generation,
1049                newlen, new, (unsigned long) data->fid2.mds,
1050                (unsigned long) data->fid2.id,
1051                (unsigned long) data->fid2.generation);
1052         
1053         if (!fid_equal(&data->fid1, &data->fid2))
1054                 CWARN("cross-node rename %lu/%lu/%lu:%*s to %lu/%lu/%lu:%*s\n",
1055                       (unsigned long)data->fid1.mds,
1056                       (unsigned long)data->fid1.id,
1057                       (unsigned long)data->fid1.generation, oldlen, old,
1058                       (unsigned long)data->fid2.mds,
1059                       (unsigned long)data->fid2.id,
1060                       (unsigned long)data->fid2.generation, newlen, new);
1061
1062         rc = lmv_check_connect(obd);
1063         if (rc)
1064                 RETURN(rc);
1065
1066         if (oldlen == 0) {
1067                 /* MDS with old dir entry is asking another MDS to create name
1068                  * there */
1069                 CDEBUG(D_OTHER,
1070                        "create %*s(%d/%d) in %lu/%lu/%lu pointing to %lu/%lu/%lu\n",
1071                        newlen, new, oldlen, newlen,
1072                        (unsigned long)data->fid2.mds,
1073                        (unsigned long)data->fid2.id,
1074                        (unsigned long)data->fid2.generation,
1075                        (unsigned long)data->fid1.mds,
1076                        (unsigned long)data->fid1.id,
1077                        (unsigned long)data->fid1.generation);
1078                 mds = data->fid2.mds;
1079                 goto request;
1080         }
1081
1082         obj = lmv_grab_obj(obd, &data->fid1);
1083         if (obj) {
1084                 /* directory is already splitted, so we have to forward request
1085                  * to the right MDS */
1086                 mds = raw_name2idx(obj->objcount, (char *)old, oldlen);
1087                 data->fid1 = obj->objs[mds].fid;
1088                 CDEBUG(D_OTHER, "forward to MDS #%u (%lu/%lu/%lu)\n", mds,
1089                        (unsigned long)obj->objs[mds].fid.mds,
1090                        (unsigned long)obj->objs[mds].fid.id,
1091                        (unsigned long)obj->objs[mds].fid.generation);
1092                 lmv_put_obj(obj);
1093         }
1094
1095         obj = lmv_grab_obj(obd, &data->fid2);
1096         if (obj) {
1097                 /* directory is already splitted, so we have to forward request
1098                  * to the right MDS */
1099                 mds = raw_name2idx(obj->objcount, (char *)new, newlen);
1100                 data->fid2 = obj->objs[mds].fid;
1101                 CDEBUG(D_OTHER, "forward to MDS #%u (%lu/%lu/%lu)\n", mds,
1102                        (unsigned long)obj->objs[mds].fid.mds,
1103                        (unsigned long)obj->objs[mds].fid.id,
1104                        (unsigned long)obj->objs[mds].fid.generation);
1105                 lmv_put_obj(obj);
1106         }
1107         
1108         mds = data->fid1.mds;
1109
1110 request:
1111         rc = md_rename(lmv->tgts[mds].ltd_exp, data, old, oldlen,
1112                        new, newlen, request); 
1113         RETURN(rc);
1114 }
1115
1116 int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data,
1117                 struct iattr *iattr, void *ea, int ealen, void *ea2,
1118                 int ea2len, struct ptlrpc_request **request)
1119 {
1120         struct obd_device *obd = exp->exp_obd;
1121         struct lmv_obd *lmv = &obd->u.lmv;
1122         struct ptlrpc_request *req;
1123         struct mds_body *body;
1124         struct lmv_obj *obj;
1125         int rc = 0, i;
1126         ENTRY;
1127
1128         rc = lmv_check_connect(obd);
1129         if (rc)
1130                 RETURN(rc);
1131
1132         obj = lmv_grab_obj(obd, &data->fid1);
1133         
1134         CDEBUG(D_OTHER, "SETATTR for %lu/%lu/%lu, valid 0x%x%s\n",
1135                (unsigned long)data->fid1.mds, (unsigned long)data->fid1.id,
1136                (unsigned long)data->fid1.generation, iattr->ia_valid,
1137                obj ? ", splitted" : "");
1138         
1139         if (obj) {
1140                 for (i = 0; i < obj->objcount; i++) {
1141                         data->fid1 = obj->objs[i].fid;
1142                         
1143                         rc = md_setattr(lmv->tgts[data->fid1.mds].ltd_exp, data,
1144                                         iattr, ea, ealen, ea2, ea2len, &req);
1145
1146                         if (fid_equal(&obj->fid, &obj->objs[i].fid)) {
1147                                 /* this is master object and this request should
1148                                  * be returned back to llite */
1149                                 *request = req;
1150                         } else {
1151                                 ptlrpc_req_finished(req);
1152                         }
1153
1154                         if (rc)
1155                                 break;
1156                 }
1157                 lmv_put_obj(obj);
1158         } else {
1159                 LASSERT(data->fid1.mds < lmv->desc.ld_tgt_count);
1160                 rc = md_setattr(lmv->tgts[data->fid1.mds].ltd_exp, data,
1161                                 iattr, ea, ealen, ea2, ea2len, request); 
1162                 if (rc == 0) {
1163                         body = lustre_msg_buf((*request)->rq_repmsg, 0,
1164                                               sizeof(*body));
1165                         LASSERT(body != NULL);
1166                         LASSERT(body->mds == data->fid1.mds);
1167                 }
1168         }
1169         RETURN(rc);
1170 }
1171
1172 int lmv_sync(struct obd_export *exp, struct ll_fid *fid,
1173              struct ptlrpc_request **request)
1174 {
1175         struct obd_device *obd = exp->exp_obd;
1176         struct lmv_obd *lmv = &obd->u.lmv;
1177         int rc;
1178         ENTRY;
1179
1180         rc = lmv_check_connect(obd);
1181         if (rc)
1182                 RETURN(rc);
1183
1184         rc = md_sync(lmv->tgts[fid->mds].ltd_exp, fid, request); 
1185         RETURN(rc);
1186 }
1187
1188 int lmv_dirobj_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1189                             void *data, int flag)
1190 {
1191         struct lustre_handle lockh;
1192         struct lmv_obj *obj;
1193         int rc;
1194         ENTRY;
1195
1196         switch (flag) {
1197         case LDLM_CB_BLOCKING:
1198                 ldlm_lock2handle(lock, &lockh);
1199                 rc = ldlm_cli_cancel(&lockh);
1200                 if (rc < 0) {
1201                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
1202                         RETURN(rc);
1203                 }
1204                 break;
1205         case LDLM_CB_CANCELING:
1206                 /* time to drop cached attrs for dirobj */
1207                 obj = lock->l_ast_data;
1208                 if (obj) {
1209                         CDEBUG(D_OTHER, "cancel %s on %lu/%lu, master %lu/%lu/%lu\n",
1210                                lock->l_resource->lr_name.name[3] == 1 ? "LOOKUP" : "UPDATE",
1211                                (unsigned long)lock->l_resource->lr_name.name[0],
1212                                (unsigned long)lock->l_resource->lr_name.name[1],
1213                                (unsigned long)obj->fid.mds, (unsigned long)obj->fid.id,
1214                                (unsigned long)obj->fid.generation);
1215                         lmv_put_obj(obj);
1216                 }
1217                 break;
1218         default:
1219                 LBUG();
1220         }
1221         RETURN(0);
1222 }
1223
1224 void lmv_remove_dots(struct page *page)
1225 {
1226         char *kaddr = page_address(page);
1227         unsigned limit = PAGE_CACHE_SIZE;
1228         unsigned offs, rec_len;
1229         struct ext2_dir_entry_2 *p;
1230
1231         for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
1232                 p = (struct ext2_dir_entry_2 *)(kaddr + offs);
1233                 rec_len = le16_to_cpu(p->rec_len);
1234
1235                 if ((p->name_len == 1 && p->name[0] == '.') ||
1236                     (p->name_len == 2 && p->name[0] == '.' && p->name[1] == '.'))
1237                         p->inode = 0;
1238         }
1239 }
1240
1241 int lmv_readpage(struct obd_export *exp, struct ll_fid *mdc_fid,
1242                  __u64 offset, struct page *page,
1243                  struct ptlrpc_request **request)
1244 {
1245         struct obd_device *obd = exp->exp_obd;
1246         struct lmv_obd *lmv = &obd->u.lmv;
1247         struct ll_fid rfid = *mdc_fid;
1248         struct lmv_obj *obj;
1249         int rc, i;
1250         ENTRY;
1251
1252         rc = lmv_check_connect(obd);
1253         if (rc)
1254                 RETURN(rc);
1255
1256         LASSERT(mdc_fid->mds < lmv->desc.ld_tgt_count);
1257         CDEBUG(D_OTHER, "READPAGE at %llu from %lu/%lu/%lu\n",
1258                offset, (unsigned long) rfid.mds,
1259                (unsigned long) rfid.id,
1260                (unsigned long) rfid.generation);
1261
1262         obj = lmv_grab_obj(obd, mdc_fid);
1263         if (obj) {
1264                 lmv_lock_obj(obj);
1265
1266                 /* find dirobj containing page with requested offset. */
1267                 for (i = 0; i < obj->objcount; i++) {
1268                         if (offset < obj->objs[i].size)
1269                                 break;
1270                         offset -= obj->objs[i].size;
1271                 }
1272                 rfid = obj->objs[i].fid;
1273                 
1274                 lmv_unlock_obj(obj);
1275                 lmv_put_obj(obj);
1276                 
1277                 CDEBUG(D_OTHER, "forward to %lu/%lu/%lu with offset %lu\n",
1278                        (unsigned long)rfid.mds, (unsigned long)rfid.id,
1279                        (unsigned long)rfid.generation, (unsigned long)offset);
1280         }
1281         rc = md_readpage(lmv->tgts[rfid.mds].ltd_exp, &rfid, offset,
1282                          page, request);
1283         
1284         if (rc == 0 && !fid_equal(&rfid, mdc_fid))
1285                 /* this page isn't from master object. To avoid "." and ".." 
1286                  * duplication in directory, we have to remove them from all
1287                  * slave objects */
1288                 lmv_remove_dots(page);
1289         
1290         RETURN(rc);
1291 }
1292
1293 int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data,
1294                       struct ptlrpc_request **req)
1295 {
1296         struct obd_device *obd = exp->exp_obd;
1297         struct lmv_obd *lmv = &obd->u.lmv;
1298         struct mea *mea = data->mea1;
1299         struct mdc_op_data data2;
1300         int i, rc = 0, mds;
1301         ENTRY;
1302
1303         LASSERT(mea != NULL);
1304         for (i = 0; i < mea->mea_count; i++) {
1305                 memset(&data2, 0, sizeof(data2));
1306                 data2.fid1 = mea->mea_fids[i];
1307                 data2.create_mode = MDS_MODE_DONT_LOCK | S_IFDIR;
1308                 mds = data2.fid1.mds;
1309
1310                 if (lmv->tgts[mds].ltd_exp == NULL)
1311                         continue;
1312
1313                 rc = md_unlink(lmv->tgts[mds].ltd_exp, &data2, req);
1314                 CDEBUG(D_OTHER, "unlink slave %lu/%lu/%lu -> %d\n",
1315                        (unsigned long) mea->mea_fids[i].mds,
1316                        (unsigned long) mea->mea_fids[i].id,
1317                        (unsigned long) mea->mea_fids[i].generation, rc);
1318                 if (*req) {
1319                         ptlrpc_req_finished(*req);
1320                         *req = NULL;
1321                 }
1322                 if (rc)
1323                         break;
1324         }
1325         RETURN(rc);
1326 }
1327
1328 int lmv_delete_object(struct obd_export *exp, struct ll_fid *fid)
1329 {
1330         ENTRY;
1331
1332         if (!lmv_delete_obj(exp, fid)) {
1333                 CDEBUG(D_OTHER, "Object %lu/%lu/%lu is not found.\n",
1334                        (unsigned long)fid->mds, (unsigned long)fid->id,
1335                        (unsigned long)fid->generation);
1336         }
1337         
1338         RETURN(0);
1339 }
1340
1341 int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data,
1342                struct ptlrpc_request **request)
1343 {
1344         struct obd_device *obd = exp->exp_obd;
1345         struct lmv_obd *lmv = &obd->u.lmv;
1346         int rc, i = 0;
1347         ENTRY;
1348         
1349         rc = lmv_check_connect(obd);
1350         if (rc)
1351                 RETURN(rc);
1352
1353         if (data->namelen == 0 && data->mea1 != NULL) {
1354                 /* mds asks to remove slave objects */
1355                 rc = lmv_unlink_slaves(exp, data, request);
1356                 RETURN(rc);
1357         } else if (data->namelen != 0) {
1358                 struct lmv_obj *obj;
1359                 
1360                 obj = lmv_grab_obj(obd, &data->fid1);
1361                 if (obj) {
1362                         i = raw_name2idx(obj->objcount, data->name,
1363                                          data->namelen);
1364                         data->fid1 = obj->objs[i].fid;
1365                         lmv_put_obj(obj);
1366                 }
1367                 CDEBUG(D_OTHER, "unlink '%*s' in %lu/%lu/%lu -> %u\n",
1368                        data->namelen, data->name,
1369                        (unsigned long) data->fid1.mds,
1370                        (unsigned long) data->fid1.id,
1371                        (unsigned long) data->fid1.generation, i);
1372         } else {
1373                 CDEBUG(D_OTHER, "drop i_nlink on %lu/%lu/%lu\n",
1374                        (unsigned long) data->fid1.mds,
1375                        (unsigned long) data->fid1.id,
1376                        (unsigned long) data->fid1.generation);
1377         }
1378         rc = md_unlink(lmv->tgts[data->fid1.mds].ltd_exp, data, request); 
1379         RETURN(rc);
1380 }
1381
1382 struct obd_device *lmv_get_real_obd(struct obd_export *exp,
1383                                     char *name, int len)
1384 {
1385         struct obd_device *obd = exp->exp_obd;
1386         struct lmv_obd *lmv = &obd->u.lmv;
1387         int rc;
1388         ENTRY;
1389
1390         rc = lmv_check_connect(obd);
1391         if (rc)
1392                 RETURN(ERR_PTR(rc));
1393 #warning "we need well-desgined readdir() implementation to remove this mess"
1394         obd = lmv->tgts[0].ltd_exp->exp_obd;
1395         EXIT;
1396         return obd;
1397 }
1398
1399 int lmv_init_ea_size(struct obd_export *exp, int easize, int cookiesize)
1400 {
1401         struct obd_device *obd = exp->exp_obd;
1402         struct lmv_obd *lmv = &obd->u.lmv;
1403         int i, rc = 0, change = 0;
1404         ENTRY;
1405
1406         if (lmv->max_easize < easize) {
1407                 lmv->max_easize = easize;
1408                 change = 1;
1409         }
1410         if (lmv->max_cookiesize < cookiesize) {
1411                 lmv->max_cookiesize = cookiesize;
1412                 change = 1;
1413         }
1414         if (change == 0)
1415                 RETURN(0);
1416         
1417         if (lmv->connected == 0)
1418                 RETURN(0);
1419
1420         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1421                 if (lmv->tgts[i].ltd_exp == NULL) {
1422                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
1423                         continue;
1424                 }
1425
1426                 rc = obd_init_ea_size(lmv->tgts[i].ltd_exp, easize, cookiesize);
1427                 if (rc) {
1428                         CERROR("obd_init_ea_size() failed on MDT target %d, "
1429                                "error %d.\n", i, rc);
1430                         break;
1431                 }
1432         }
1433         RETURN(rc);
1434 }
1435
1436 int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa,
1437                           struct lov_stripe_md **ea, struct obd_trans_info *oti)
1438 {
1439         struct obd_device *obd = exp->exp_obd;
1440         struct lmv_obd *lmv = &obd->u.lmv;
1441         struct lov_stripe_md obj_md;
1442         struct lov_stripe_md *obj_mdp = &obj_md;
1443         int rc = 0;
1444         ENTRY;
1445
1446         rc = lmv_check_connect(obd);
1447         if (rc)
1448                 RETURN(rc);
1449
1450         LASSERT(ea == NULL);
1451         LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1452
1453         rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp, oa, &obj_mdp, oti);
1454
1455         RETURN(rc);
1456 }
1457
1458 /*
1459  * to be called from MDS only
1460  */
1461 int lmv_obd_create(struct obd_export *exp, struct obdo *oa,
1462                    struct lov_stripe_md **ea, struct obd_trans_info *oti)
1463 {
1464         struct obd_device *obd = exp->exp_obd;
1465         struct lmv_obd *lmv = &obd->u.lmv;
1466         int i, c, rc = 0;
1467         struct mea *mea;
1468         struct ll_fid mfid;
1469         int lcount;
1470         ENTRY;
1471
1472         rc = lmv_check_connect(obd);
1473         if (rc)
1474                 RETURN(rc);
1475
1476         LASSERT(oa != NULL);
1477         
1478         if (ea == NULL) {
1479                 rc = lmv_obd_create_single(exp, oa, NULL, oti);
1480                 RETURN(rc);
1481         }
1482
1483         if (*ea == NULL) {
1484                 rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea);
1485                 if (rc < 0) {
1486                         CERROR("obd_alloc_diskmd() failed, error %d\n",
1487                                rc);
1488                         RETURN(rc);
1489                 }
1490                 
1491                 if (*ea == NULL)
1492                         RETURN(-EINVAL);
1493         }
1494
1495         rc = 0;
1496         mfid.id = oa->o_id;
1497         mfid.generation = oa->o_generation;
1498         
1499         mea = (struct mea *)*ea;
1500         if (!mea->mea_count || mea->mea_count > lmv->desc.ld_tgt_count)
1501                 mea->mea_count = lmv->desc.ld_tgt_count;
1502
1503         mea->mea_master = -1;
1504         lcount = lmv->desc.ld_tgt_count;
1505         for (i = 0, c = 0; c < mea->mea_count && i < lcount; i++) {
1506                 struct lov_stripe_md obj_md;
1507                 struct lov_stripe_md *obj_mdp = &obj_md;
1508                
1509                 if (lmv->tgts[i].ltd_exp == NULL) {
1510                         /* this is master MDS */
1511                         mea->mea_fids[c].id = mfid.id;
1512                         mea->mea_fids[c].generation = mfid.generation;
1513                         mea->mea_fids[c].mds = i;
1514                         mea->mea_master = i;
1515                         c++;
1516                         continue;
1517                 }
1518
1519                 /* "master" MDS should always be part of stripped dir, so scan
1520                    for it. */
1521                 if (mea->mea_master == -1 && c == mea->mea_count - 1)
1522                         continue;
1523
1524                 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE
1525                         | OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID;
1526
1527                 rc = obd_create(lmv->tgts[c].ltd_exp, oa, &obj_mdp, oti);
1528                 if (rc) {
1529                         CERROR("obd_create() failed on MDT target %d, "
1530                                "error %d\n", c, rc);
1531                         RETURN(rc);
1532                 }
1533
1534                 mea->mea_fids[c].id = oa->o_id;
1535                 mea->mea_fids[c].generation = oa->o_generation;
1536                 mea->mea_fids[c].mds = i;
1537                 c++;
1538                 CDEBUG(D_OTHER, "dirobj at mds %d: "LPU64"/%u\n",
1539                        i, oa->o_id, oa->o_generation);
1540         }
1541         LASSERT(c == mea->mea_count);
1542         CDEBUG(D_OTHER, "%d dirobjects created\n", (int) mea->mea_count);
1543
1544         RETURN(rc);
1545 }
1546
1547 static int lmv_get_info(struct obd_export *exp, __u32 keylen,
1548                         void *key, __u32 *vallen, void *val)
1549 {
1550         struct obd_device *obd;
1551         struct lmv_obd *lmv;
1552         ENTRY;
1553
1554         obd = class_exp2obd(exp);
1555         if (obd == NULL) {
1556                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1557                        exp->exp_handle.h_cookie);
1558                 RETURN(-EINVAL);
1559         }
1560
1561         lmv = &obd->u.lmv;
1562         if (keylen == 6 && memcmp(key, "mdsize", 6) == 0) {
1563                 __u32 *mdsize = val;
1564                 *vallen = sizeof(__u32);
1565                 *mdsize = sizeof(struct ll_fid) * lmv->desc.ld_tgt_count
1566                                 + sizeof(struct mea);
1567                 RETURN(0);
1568         } else if (keylen == 6 && memcmp(key, "mdsnum", 6) == 0) {
1569                 struct obd_uuid *cluuid = &lmv->cluuid;
1570                 struct lmv_tgt_desc *tgts;
1571                 __u32 *mdsnum = val;
1572                 int i;
1573
1574                 for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) {
1575                         if (obd_uuid_equals(&tgts->uuid, cluuid)) {
1576                                 *vallen = sizeof(__u32);
1577                                 *mdsnum = i;
1578                                 RETURN(0);
1579                         }
1580                 }
1581                 LASSERT(0);
1582         }
1583
1584         CDEBUG(D_IOCTL, "invalid key\n");
1585         RETURN(-EINVAL);
1586 }
1587
1588 int lmv_set_info(struct obd_export *exp, obd_count keylen,
1589                  void *key, obd_count vallen, void *val)
1590 {
1591         struct obd_device *obd;
1592         struct lmv_obd *lmv;
1593         ENTRY;
1594
1595         obd = class_exp2obd(exp);
1596         if (obd == NULL) {
1597                 CDEBUG(D_IOCTL, "invalid client cookie "LPX64"\n",
1598                        exp->exp_handle.h_cookie);
1599                 RETURN(-EINVAL);
1600         }
1601         lmv = &obd->u.lmv;
1602
1603         if (keylen >= strlen("client") && strcmp(key, "client") == 0) {
1604                 struct lmv_tgt_desc *tgts;
1605                 int i, rc;
1606
1607                 rc = lmv_check_connect(obd);
1608                 if (rc)
1609                         RETURN(rc);
1610
1611                 for (i = 0, tgts = lmv->tgts; 
1612                         i < lmv->desc.ld_tgt_count; i++, tgts++) {
1613                         rc = obd_set_info(tgts->ltd_exp, keylen, key, vallen, val);
1614                         if (rc)
1615                                 RETURN(rc);
1616                 }
1617                 RETURN(0);
1618         } else if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) {
1619                 lmv->server_timeout = 1;
1620                 lmv_set_timeouts(obd);
1621                 RETURN(0);
1622         }
1623         
1624         RETURN(-EINVAL);
1625 }
1626
1627 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
1628                struct lov_stripe_md *lsm)
1629 {
1630         struct obd_device *obd = class_exp2obd(exp);
1631         struct lmv_obd *lmv = &obd->u.lmv;
1632         int mea_size;
1633         ENTRY;
1634
1635         mea_size = sizeof(struct ll_fid) * 
1636                 lmv->desc.ld_tgt_count + sizeof(struct mea);
1637         if (!lmmp)
1638                 RETURN(mea_size);
1639
1640         if (*lmmp && !lsm) {
1641                 OBD_FREE(*lmmp, mea_size);
1642                 *lmmp = NULL;
1643                 RETURN(0);
1644         }
1645
1646         if (*lmmp == NULL) {
1647                 OBD_ALLOC(*lmmp, mea_size);
1648                 if (*lmmp == NULL)
1649                         RETURN(-ENOMEM);
1650         }
1651
1652         if (!lsm)
1653                 RETURN(mea_size);
1654
1655 #warning "MEA packing/convertation must be here! -bzzz"
1656         memcpy(*lmmp, lsm, mea_size);
1657         RETURN(mea_size);
1658 }
1659
1660 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **mem_tgt,
1661                  struct lov_mds_md *disk_src, int mdsize)
1662 {
1663         struct obd_device *obd = class_exp2obd(exp);
1664         struct lmv_obd *lmv = &obd->u.lmv;
1665         struct mea **tmea = (struct mea **) mem_tgt;
1666         struct mea *mea = (void *) disk_src;
1667         int mea_size;
1668         ENTRY;
1669
1670         mea_size = sizeof(struct ll_fid) * 
1671                 lmv->desc.ld_tgt_count + sizeof(struct mea);
1672         if (mem_tgt == NULL)
1673                 return mea_size;
1674
1675         if (*mem_tgt != NULL && disk_src == NULL) {
1676                 OBD_FREE(*tmea, mea_size);
1677                 RETURN(0);
1678         }
1679
1680         LASSERT(mea_size == mdsize);
1681
1682         OBD_ALLOC(*tmea, mea_size);
1683         if (*tmea == NULL)
1684                 RETURN(-ENOMEM);
1685
1686         if (!disk_src)
1687                 RETURN(mea_size);
1688
1689 #warning "MEA unpacking/convertation must be here! -bzzz"
1690         memcpy(*tmea, mea, mdsize);
1691         RETURN(mea_size);
1692 }
1693
1694 int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa,
1695             struct lov_stripe_md *ea, obd_count oa_bufs,
1696             struct brw_page *pgarr, struct obd_trans_info *oti)
1697 {
1698         struct obd_device *obd = exp->exp_obd;
1699         struct lmv_obd *lmv = &obd->u.lmv;
1700         struct mea *mea = (struct mea *) ea;
1701         int err;
1702       
1703         LASSERT(oa != NULL);
1704         LASSERT(ea != NULL);
1705         LASSERT(pgarr != NULL);
1706         LASSERT(oa->o_mds < lmv->desc.ld_tgt_count);
1707
1708         oa->o_gr = mea->mea_fids[oa->o_mds].generation;
1709         oa->o_id = mea->mea_fids[oa->o_mds].id;
1710         oa->o_valid =  OBD_MD_FLID | OBD_MD_FLGROUP;
1711         err = obd_brw(rw, lmv->tgts[oa->o_mds].ltd_exp, oa,
1712                       NULL, oa_bufs, pgarr, oti);
1713         RETURN(err);
1714 }
1715
1716 struct obd_ops lmv_obd_ops = {
1717         .o_owner                = THIS_MODULE,
1718         .o_attach               = lmv_attach,
1719         .o_detach               = lmv_detach,
1720         .o_setup                = lmv_setup,
1721         .o_cleanup              = lmv_cleanup,
1722         .o_connect              = lmv_connect,
1723         .o_disconnect           = lmv_disconnect,
1724         .o_statfs               = lmv_statfs,
1725         .o_get_info             = lmv_get_info,
1726         .o_set_info             = lmv_set_info,
1727         .o_create               = lmv_obd_create,
1728         .o_packmd               = lmv_packmd,
1729         .o_unpackmd             = lmv_unpackmd,
1730         .o_brw                  = lmv_brw,
1731         .o_init_ea_size         = lmv_init_ea_size,
1732         .o_notify               = lmv_notify,
1733         .o_iocontrol            = lmv_iocontrol,
1734 };
1735
1736 struct md_ops lmv_md_ops = {
1737         .m_getstatus            = lmv_getstatus,
1738         .m_getattr              = lmv_getattr,
1739         .m_change_cbdata        = lmv_change_cbdata,
1740         .m_change_cbdata_name   = lmv_change_cbdata_name,
1741         .m_close                = lmv_close,
1742         .m_create               = lmv_create,
1743         .m_done_writing         = lmv_done_writing,
1744         .m_enqueue              = lmv_enqueue,
1745         .m_getattr_name         = lmv_getattr_name,
1746         .m_intent_lock          = lmv_intent_lock,
1747         .m_link                 = lmv_link,
1748         .m_rename               = lmv_rename,
1749         .m_setattr              = lmv_setattr,
1750         .m_sync                 = lmv_sync,
1751         .m_readpage             = lmv_readpage,
1752         .m_unlink               = lmv_unlink,
1753         .m_get_real_obd         = lmv_get_real_obd,
1754         .m_valid_attrs          = lmv_valid_attrs,
1755         .m_delete_object        = lmv_delete_object,
1756 };
1757
1758 int __init lmv_init(void)
1759 {
1760         struct lprocfs_static_vars lvars;
1761         int rc;
1762
1763         lprocfs_init_vars(lmv, &lvars);
1764         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
1765                                  lvars.module_vars, OBD_LMV_DEVICENAME);
1766         RETURN(rc);
1767 }
1768
1769 #ifdef __KERNEL__
1770 static void lmv_exit(void)
1771 {
1772         class_unregister_type(OBD_LMV_DEVICENAME);
1773 }
1774
1775 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1776 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
1777 MODULE_LICENSE("GPL");
1778
1779 module_init(lmv_init);
1780 module_exit(lmv_exit);
1781 #endif