Whamcloud - gitweb
LU-3185 build: fix autogen with modern autoconf tools
[fs/lustre-release.git] / lustre / lmv / lmv_obd.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2013, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  */
36
37 #define DEBUG_SUBSYSTEM S_LMV
38 #ifdef __KERNEL__
39 #include <linux/slab.h>
40 #include <linux/module.h>
41 #include <linux/init.h>
42 #include <linux/slab.h>
43 #include <linux/pagemap.h>
44 #include <linux/mm.h>
45 #include <asm/div64.h>
46 #include <linux/seq_file.h>
47 #include <linux/namei.h>
48 #else
49 #include <liblustre.h>
50 #endif
51
52 #include <lustre/lustre_idl.h>
53 #include <obd_support.h>
54 #include <lustre_lib.h>
55 #include <lustre_net.h>
56 #include <obd_class.h>
57 #include <lprocfs_status.h>
58 #include <lustre_lite.h>
59 #include <lustre_fid.h>
60 #include "lmv_internal.h"
61
62 static void lmv_activate_target(struct lmv_obd *lmv,
63                                 struct lmv_tgt_desc *tgt,
64                                 int activate)
65 {
66         if (tgt->ltd_active == activate)
67                 return;
68
69         tgt->ltd_active = activate;
70         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
71 }
72
73 /**
74  * Error codes:
75  *
76  *  -EINVAL  : UUID can't be found in the LMV's target list
77  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
78  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
79  */
80 static int lmv_set_mdc_active(struct lmv_obd *lmv,
81                               const struct obd_uuid *uuid,
82                               int activate)
83 {
84         struct lmv_tgt_desc     *tgt = NULL;
85         struct obd_device       *obd;
86         __u32                    i;
87         int                      rc = 0;
88         ENTRY;
89
90         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
91                         lmv, uuid->uuid, activate);
92
93         spin_lock(&lmv->lmv_lock);
94         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
95                 tgt = lmv->tgts[i];
96                 if (tgt == NULL || tgt->ltd_exp == NULL)
97                         continue;
98
99                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
100                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
101
102                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
103                         break;
104         }
105
106         if (i == lmv->desc.ld_tgt_count)
107                 GOTO(out_lmv_lock, rc = -EINVAL);
108
109         obd = class_exp2obd(tgt->ltd_exp);
110         if (obd == NULL)
111                 GOTO(out_lmv_lock, rc = -ENOTCONN);
112
113         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
114                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
115                obd->obd_type->typ_name, i);
116         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
117
118         if (tgt->ltd_active == activate) {
119                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
120                        activate ? "" : "in");
121                 GOTO(out_lmv_lock, rc);
122         }
123
124         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
125                activate ? "" : "in");
126         lmv_activate_target(lmv, tgt, activate);
127         EXIT;
128
129  out_lmv_lock:
130         spin_unlock(&lmv->lmv_lock);
131         return rc;
132 }
133
134 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
135 {
136         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
137
138         return obd_get_uuid(lmv->tgts[0]->ltd_exp);
139 }
140
141 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
142                       enum obd_notify_event ev, void *data)
143 {
144         struct obd_connect_data *conn_data;
145         struct lmv_obd          *lmv = &obd->u.lmv;
146         struct obd_uuid         *uuid;
147         int                      rc = 0;
148         ENTRY;
149
150         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
151                 CERROR("unexpected notification of %s %s!\n",
152                        watched->obd_type->typ_name,
153                        watched->obd_name);
154                 RETURN(-EINVAL);
155         }
156
157         uuid = &watched->u.cli.cl_target_uuid;
158         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
159                 /*
160                  * Set MDC as active before notifying the observer, so the
161                  * observer can use the MDC normally.
162                  */
163                 rc = lmv_set_mdc_active(lmv, uuid,
164                                         ev == OBD_NOTIFY_ACTIVE);
165                 if (rc) {
166                         CERROR("%sactivation of %s failed: %d\n",
167                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
168                                uuid->uuid, rc);
169                         RETURN(rc);
170                 }
171         } else if (ev == OBD_NOTIFY_OCD) {
172                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
173                 /*
174                  * XXX: Make sure that ocd_connect_flags from all targets are
175                  * the same. Otherwise one of MDTs runs wrong version or
176                  * something like this.  --umka
177                  */
178                 obd->obd_self_export->exp_connect_data = *conn_data;
179         }
180 #if 0
181         else if (ev == OBD_NOTIFY_DISCON) {
182                 /*
183                  * For disconnect event, flush fld cache for failout MDS case.
184                  */
185                 fld_client_flush(&lmv->lmv_fld);
186         }
187 #endif
188         /*
189          * Pass the notification up the chain.
190          */
191         if (obd->obd_observer)
192                 rc = obd_notify(obd->obd_observer, watched, ev, data);
193
194         RETURN(rc);
195 }
196
197 /**
198  * This is fake connect function. Its purpose is to initialize lmv and say
199  * caller that everything is okay. Real connection will be performed later.
200  */
201 static int lmv_connect(const struct lu_env *env,
202                        struct obd_export **exp, struct obd_device *obd,
203                        struct obd_uuid *cluuid, struct obd_connect_data *data,
204                        void *localdata)
205 {
206 #ifdef __KERNEL__
207         struct proc_dir_entry *lmv_proc_dir;
208 #endif
209         struct lmv_obd        *lmv = &obd->u.lmv;
210         struct lustre_handle  conn = { 0 };
211         int                    rc = 0;
212         ENTRY;
213
214         /*
215          * We don't want to actually do the underlying connections more than
216          * once, so keep track.
217          */
218         lmv->refcount++;
219         if (lmv->refcount > 1) {
220                 *exp = NULL;
221                 RETURN(0);
222         }
223
224         rc = class_connect(&conn, obd, cluuid);
225         if (rc) {
226                 CERROR("class_connection() returned %d\n", rc);
227                 RETURN(rc);
228         }
229
230         *exp = class_conn2export(&conn);
231         class_export_get(*exp);
232
233         lmv->exp = *exp;
234         lmv->connected = 0;
235         lmv->cluuid = *cluuid;
236
237         if (data)
238                 lmv->conn_data = *data;
239
240 #ifdef __KERNEL__
241         lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry,
242                                         NULL, NULL);
243         if (IS_ERR(lmv_proc_dir)) {
244                 CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.",
245                        obd->obd_type->typ_name, obd->obd_name);
246                 lmv_proc_dir = NULL;
247         }
248 #endif
249
250         /*
251          * All real clients should perform actual connection right away, because
252          * it is possible, that LMV will not have opportunity to connect targets
253          * and MDC stuff will be called directly, for instance while reading
254          * ../mdc/../kbytesfree procfs file, etc.
255          */
256         if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
257                 rc = lmv_check_connect(obd);
258
259 #ifdef __KERNEL__
260         if (rc) {
261                 if (lmv_proc_dir)
262                         lprocfs_remove(&lmv_proc_dir);
263         }
264 #endif
265
266         RETURN(rc);
267 }
268
269 static void lmv_set_timeouts(struct obd_device *obd)
270 {
271         struct lmv_tgt_desc     *tgt;
272         struct lmv_obd          *lmv;
273         __u32                    i;
274
275         lmv = &obd->u.lmv;
276         if (lmv->server_timeout == 0)
277                 return;
278
279         if (lmv->connected == 0)
280                 return;
281
282         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
283                 tgt = lmv->tgts[i];
284                 if (tgt == NULL || tgt->ltd_exp == NULL || tgt->ltd_active == 0)
285                         continue;
286
287                 obd_set_info_async(NULL, tgt->ltd_exp, sizeof(KEY_INTERMDS),
288                                    KEY_INTERMDS, 0, NULL, NULL);
289         }
290 }
291
292 static int lmv_init_ea_size(struct obd_export *exp, int easize,
293                             int def_easize, int cookiesize)
294 {
295         struct obd_device       *obd = exp->exp_obd;
296         struct lmv_obd          *lmv = &obd->u.lmv;
297         __u32                    i;
298         int                      rc = 0;
299         int                      change = 0;
300         ENTRY;
301
302         if (lmv->max_easize < easize) {
303                 lmv->max_easize = easize;
304                 change = 1;
305         }
306         if (lmv->max_def_easize < def_easize) {
307                 lmv->max_def_easize = def_easize;
308                 change = 1;
309         }
310         if (lmv->max_cookiesize < cookiesize) {
311                 lmv->max_cookiesize = cookiesize;
312                 change = 1;
313         }
314         if (change == 0)
315                 RETURN(0);
316
317         if (lmv->connected == 0)
318                 RETURN(0);
319
320         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
321                 if (lmv->tgts[i] == NULL ||
322                     lmv->tgts[i]->ltd_exp == NULL ||
323                     lmv->tgts[i]->ltd_active == 0) {
324                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
325                         continue;
326                 }
327
328                 rc = md_init_ea_size(lmv->tgts[i]->ltd_exp, easize, def_easize,
329                                      cookiesize);
330                 if (rc) {
331                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
332                                " rc = %d.\n", obd->obd_name, i, rc);
333                         break;
334                 }
335         }
336         RETURN(rc);
337 }
338
339 #define MAX_STRING_SIZE 128
340
341 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
342 {
343 #ifdef __KERNEL__
344         struct proc_dir_entry   *lmv_proc_dir;
345 #endif
346         struct lmv_obd          *lmv = &obd->u.lmv;
347         struct obd_uuid         *cluuid = &lmv->cluuid;
348         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
349         struct obd_device       *mdc_obd;
350         struct obd_export       *mdc_exp;
351         struct lu_fld_target     target;
352         int                      rc;
353         ENTRY;
354
355         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
356                                         &obd->obd_uuid);
357         if (!mdc_obd) {
358                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
359                 RETURN(-EINVAL);
360         }
361
362         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
363                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
364                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
365                 cluuid->uuid);
366
367         if (!mdc_obd->obd_set_up) {
368                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
369                 RETURN(-EINVAL);
370         }
371
372         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
373                          &lmv->conn_data, NULL);
374         if (rc) {
375                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
376                 RETURN(rc);
377         }
378
379         /*
380          * Init fid sequence client for this mdc and add new fld target.
381          */
382         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
383         if (rc)
384                 RETURN(rc);
385
386         target.ft_srv = NULL;
387         target.ft_exp = mdc_exp;
388         target.ft_idx = tgt->ltd_idx;
389
390         fld_client_add_target(&lmv->lmv_fld, &target);
391
392         rc = obd_register_observer(mdc_obd, obd);
393         if (rc) {
394                 obd_disconnect(mdc_exp);
395                 CERROR("target %s register_observer error %d\n",
396                        tgt->ltd_uuid.uuid, rc);
397                 RETURN(rc);
398         }
399
400         if (obd->obd_observer) {
401                 /*
402                  * Tell the observer about the new target.
403                  */
404                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
405                                 OBD_NOTIFY_ACTIVE,
406                                 (void *)(tgt - lmv->tgts[0]));
407                 if (rc) {
408                         obd_disconnect(mdc_exp);
409                         RETURN(rc);
410                 }
411         }
412
413         tgt->ltd_active = 1;
414         tgt->ltd_exp = mdc_exp;
415         lmv->desc.ld_active_tgt_count++;
416
417         md_init_ea_size(tgt->ltd_exp, lmv->max_easize,
418                         lmv->max_def_easize, lmv->max_cookiesize);
419
420         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
421                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
422                 cfs_atomic_read(&obd->obd_refcount));
423
424 #ifdef __KERNEL__
425         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
426         if (lmv_proc_dir) {
427                 struct proc_dir_entry *mdc_symlink;
428
429                 LASSERT(mdc_obd->obd_type != NULL);
430                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
431                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
432                                                   lmv_proc_dir,
433                                                   "../../../%s/%s",
434                                                   mdc_obd->obd_type->typ_name,
435                                                   mdc_obd->obd_name);
436                 if (mdc_symlink == NULL) {
437                         CERROR("Could not register LMV target "
438                                "/proc/fs/lustre/%s/%s/target_obds/%s.",
439                                obd->obd_type->typ_name, obd->obd_name,
440                                mdc_obd->obd_name);
441                         lprocfs_remove(&lmv_proc_dir);
442                         lmv_proc_dir = NULL;
443                 }
444         }
445 #endif
446         RETURN(0);
447 }
448
449 static void lmv_del_target(struct lmv_obd *lmv, int index)
450 {
451         if (lmv->tgts[index] == NULL)
452                 return;
453
454         OBD_FREE_PTR(lmv->tgts[index]);
455         lmv->tgts[index] = NULL;
456         return;
457 }
458
459 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
460                            __u32 index, int gen)
461 {
462         struct lmv_obd      *lmv = &obd->u.lmv;
463         struct lmv_tgt_desc *tgt;
464         int                  rc = 0;
465         ENTRY;
466
467         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
468
469         lmv_init_lock(lmv);
470
471         if (lmv->desc.ld_tgt_count == 0) {
472                 struct obd_device *mdc_obd;
473
474                 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
475                                                 &obd->obd_uuid);
476                 if (!mdc_obd) {
477                         lmv_init_unlock(lmv);
478                         CERROR("%s: Target %s not attached: rc = %d\n",
479                                obd->obd_name, uuidp->uuid, -EINVAL);
480                         RETURN(-EINVAL);
481                 }
482         }
483
484         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
485                 tgt = lmv->tgts[index];
486                 CERROR("%s: UUID %s already assigned at LOV target index %d:"
487                        " rc = %d\n", obd->obd_name,
488                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
489                 lmv_init_unlock(lmv);
490                 RETURN(-EEXIST);
491         }
492
493         if (index >= lmv->tgts_size) {
494                 /* We need to reallocate the lmv target array. */
495                 struct lmv_tgt_desc **newtgts, **old = NULL;
496                 __u32 newsize = 1;
497                 __u32 oldsize = 0;
498
499                 while (newsize < index + 1)
500                         newsize = newsize << 1;
501                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
502                 if (newtgts == NULL) {
503                         lmv_init_unlock(lmv);
504                         RETURN(-ENOMEM);
505                 }
506
507                 if (lmv->tgts_size) {
508                         memcpy(newtgts, lmv->tgts,
509                                sizeof(*newtgts) * lmv->tgts_size);
510                         old = lmv->tgts;
511                         oldsize = lmv->tgts_size;
512                 }
513
514                 lmv->tgts = newtgts;
515                 lmv->tgts_size = newsize;
516                 smp_rmb();
517                 if (old)
518                         OBD_FREE(old, sizeof(*old) * oldsize);
519
520                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
521                        lmv->tgts_size);
522         }
523
524         OBD_ALLOC_PTR(tgt);
525         if (!tgt) {
526                 lmv_init_unlock(lmv);
527                 RETURN(-ENOMEM);
528         }
529
530         mutex_init(&tgt->ltd_fid_mutex);
531         tgt->ltd_idx = index;
532         tgt->ltd_uuid = *uuidp;
533         tgt->ltd_active = 0;
534         lmv->tgts[index] = tgt;
535         if (index >= lmv->desc.ld_tgt_count)
536                 lmv->desc.ld_tgt_count = index + 1;
537
538         if (lmv->connected) {
539                 rc = lmv_connect_mdc(obd, tgt);
540                 if (rc) {
541                         spin_lock(&lmv->lmv_lock);
542                         lmv->desc.ld_tgt_count--;
543                         memset(tgt, 0, sizeof(*tgt));
544                         spin_unlock(&lmv->lmv_lock);
545                 } else {
546                         int easize = sizeof(struct lmv_stripe_md) +
547                                      lmv->desc.ld_tgt_count *
548                                      sizeof(struct lu_fid);
549                         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
550                 }
551         }
552
553         lmv_init_unlock(lmv);
554         RETURN(rc);
555 }
556
557 int lmv_check_connect(struct obd_device *obd)
558 {
559         struct lmv_obd          *lmv = &obd->u.lmv;
560         struct lmv_tgt_desc     *tgt;
561         __u32                    i;
562         int                      rc;
563         int                      easize;
564         ENTRY;
565
566         if (lmv->connected)
567                 RETURN(0);
568
569         lmv_init_lock(lmv);
570         if (lmv->connected) {
571                 lmv_init_unlock(lmv);
572                 RETURN(0);
573         }
574
575         if (lmv->desc.ld_tgt_count == 0) {
576                 lmv_init_unlock(lmv);
577                 CERROR("%s: no targets configured.\n", obd->obd_name);
578                 RETURN(-EINVAL);
579         }
580
581         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
582                lmv->cluuid.uuid, obd->obd_name);
583
584         LASSERT(lmv->tgts != NULL);
585
586         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
587                 tgt = lmv->tgts[i];
588                 if (tgt == NULL)
589                         continue;
590                 rc = lmv_connect_mdc(obd, tgt);
591                 if (rc)
592                         GOTO(out_disc, rc);
593         }
594
595         lmv_set_timeouts(obd);
596         class_export_put(lmv->exp);
597         lmv->connected = 1;
598         easize = lmv_get_easize(lmv);
599         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0);
600         lmv_init_unlock(lmv);
601         RETURN(0);
602
603  out_disc:
604         while (i-- > 0) {
605                 int rc2;
606                 tgt = lmv->tgts[i];
607                 if (tgt == NULL)
608                         continue;
609                 tgt->ltd_active = 0;
610                 if (tgt->ltd_exp) {
611                         --lmv->desc.ld_active_tgt_count;
612                         rc2 = obd_disconnect(tgt->ltd_exp);
613                         if (rc2) {
614                                 CERROR("LMV target %s disconnect on "
615                                        "MDC idx %d: error %d\n",
616                                        tgt->ltd_uuid.uuid, i, rc2);
617                         }
618                 }
619         }
620         class_disconnect(lmv->exp);
621         lmv_init_unlock(lmv);
622         RETURN(rc);
623 }
624
625 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
626 {
627 #ifdef __KERNEL__
628         struct proc_dir_entry  *lmv_proc_dir;
629 #endif
630         struct lmv_obd         *lmv = &obd->u.lmv;
631         struct obd_device      *mdc_obd;
632         int                     rc;
633         ENTRY;
634
635         LASSERT(tgt != NULL);
636         LASSERT(obd != NULL);
637
638         mdc_obd = class_exp2obd(tgt->ltd_exp);
639
640         if (mdc_obd) {
641                 mdc_obd->obd_force = obd->obd_force;
642                 mdc_obd->obd_fail = obd->obd_fail;
643                 mdc_obd->obd_no_recov = obd->obd_no_recov;
644         }
645
646 #ifdef __KERNEL__
647         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
648         if (lmv_proc_dir) {
649                 struct proc_dir_entry *mdc_symlink;
650
651                 mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name);
652                 if (mdc_symlink) {
653                         lprocfs_remove(&mdc_symlink);
654                 } else {
655                         CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n",
656                                obd->obd_type->typ_name, obd->obd_name,
657                                mdc_obd->obd_name);
658                 }
659         }
660 #endif
661         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
662         if (rc)
663                 CERROR("Can't finanize fids factory\n");
664
665         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
666                tgt->ltd_exp->exp_obd->obd_name,
667                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
668
669         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
670         rc = obd_disconnect(tgt->ltd_exp);
671         if (rc) {
672                 if (tgt->ltd_active) {
673                         CERROR("Target %s disconnect error %d\n",
674                                tgt->ltd_uuid.uuid, rc);
675                 }
676         }
677
678         lmv_activate_target(lmv, tgt, 0);
679         tgt->ltd_exp = NULL;
680         RETURN(0);
681 }
682
683 static int lmv_disconnect(struct obd_export *exp)
684 {
685         struct obd_device       *obd = class_exp2obd(exp);
686 #ifdef __KERNEL__
687         struct proc_dir_entry   *lmv_proc_dir;
688 #endif
689         struct lmv_obd          *lmv = &obd->u.lmv;
690         int                      rc;
691         __u32                    i;
692         ENTRY;
693
694         if (!lmv->tgts)
695                 goto out_local;
696
697         /*
698          * Only disconnect the underlying layers on the final disconnect.
699          */
700         lmv->refcount--;
701         if (lmv->refcount != 0)
702                 goto out_local;
703
704         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
705                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
706                         continue;
707
708                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
709         }
710
711 #ifdef __KERNEL__
712         lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds");
713         if (lmv_proc_dir) {
714                 lprocfs_remove(&lmv_proc_dir);
715         } else {
716                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
717                        obd->obd_type->typ_name, obd->obd_name);
718         }
719 #endif
720
721 out_local:
722         /*
723          * This is the case when no real connection is established by
724          * lmv_check_connect().
725          */
726         if (!lmv->connected)
727                 class_export_put(exp);
728         rc = class_disconnect(exp);
729         if (lmv->refcount == 0)
730                 lmv->connected = 0;
731         RETURN(rc);
732 }
733
734 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
735 {
736         struct obd_device       *obddev = class_exp2obd(exp);
737         struct lmv_obd          *lmv = &obddev->u.lmv;
738         struct getinfo_fid2path *gf;
739         struct lmv_tgt_desc     *tgt;
740         struct getinfo_fid2path *remote_gf = NULL;
741         int                     remote_gf_size = 0;
742         int                     rc;
743
744         gf = (struct getinfo_fid2path *)karg;
745         tgt = lmv_find_target(lmv, &gf->gf_fid);
746         if (IS_ERR(tgt))
747                 RETURN(PTR_ERR(tgt));
748
749 repeat_fid2path:
750         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
751         if (rc != 0 && rc != -EREMOTE)
752                 GOTO(out_fid2path, rc);
753
754         /* If remote_gf != NULL, it means just building the
755          * path on the remote MDT, copy this path segement to gf */
756         if (remote_gf != NULL) {
757                 struct getinfo_fid2path *ori_gf;
758                 char *ptr;
759
760                 ori_gf = (struct getinfo_fid2path *)karg;
761                 if (strlen(ori_gf->gf_path) +
762                     strlen(gf->gf_path) > ori_gf->gf_pathlen)
763                         GOTO(out_fid2path, rc = -EOVERFLOW);
764
765                 ptr = ori_gf->gf_path;
766
767                 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
768                         strlen(ori_gf->gf_path));
769
770                 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
771                 ptr += strlen(gf->gf_path);
772                 *ptr = '/';
773         }
774
775         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
776                tgt->ltd_exp->exp_obd->obd_name,
777                gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
778                gf->gf_linkno);
779
780         if (rc == 0)
781                 GOTO(out_fid2path, rc);
782
783         /* sigh, has to go to another MDT to do path building further */
784         if (remote_gf == NULL) {
785                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
786                 OBD_ALLOC(remote_gf, remote_gf_size);
787                 if (remote_gf == NULL)
788                         GOTO(out_fid2path, rc = -ENOMEM);
789                 remote_gf->gf_pathlen = PATH_MAX;
790         }
791
792         if (!fid_is_sane(&gf->gf_fid)) {
793                 CERROR("%s: invalid FID "DFID": rc = %d\n",
794                        tgt->ltd_exp->exp_obd->obd_name,
795                        PFID(&gf->gf_fid), -EINVAL);
796                 GOTO(out_fid2path, rc = -EINVAL);
797         }
798
799         tgt = lmv_find_target(lmv, &gf->gf_fid);
800         if (IS_ERR(tgt))
801                 GOTO(out_fid2path, rc = -EINVAL);
802
803         remote_gf->gf_fid = gf->gf_fid;
804         remote_gf->gf_recno = -1;
805         remote_gf->gf_linkno = -1;
806         memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
807         gf = remote_gf;
808         goto repeat_fid2path;
809
810 out_fid2path:
811         if (remote_gf != NULL)
812                 OBD_FREE(remote_gf, remote_gf_size);
813         RETURN(rc);
814 }
815
816 static int lmv_hsm_req_count(struct lmv_obd *lmv,
817                              const struct hsm_user_request *hur,
818                              const struct lmv_tgt_desc *tgt_mds)
819 {
820         __u32                    i;
821         int                      nr = 0;
822         struct lmv_tgt_desc     *curr_tgt;
823
824         /* count how many requests must be sent to the given target */
825         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
826                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
827                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
828                         nr++;
829         }
830         return nr;
831 }
832
833 static void lmv_hsm_req_build(struct lmv_obd *lmv,
834                               struct hsm_user_request *hur_in,
835                               const struct lmv_tgt_desc *tgt_mds,
836                               struct hsm_user_request *hur_out)
837 {
838         __u32                    i, nr_out;
839         struct lmv_tgt_desc     *curr_tgt;
840
841         /* build the hsm_user_request for the given target */
842         hur_out->hur_request = hur_in->hur_request;
843         nr_out = 0;
844         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
845                 curr_tgt = lmv_find_target(lmv,
846                                            &hur_in->hur_user_item[i].hui_fid);
847                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
848                         hur_out->hur_user_item[nr_out] =
849                                                 hur_in->hur_user_item[i];
850                         nr_out++;
851                 }
852         }
853         hur_out->hur_request.hr_itemcount = nr_out;
854         memcpy(hur_data(hur_out), hur_data(hur_in),
855                hur_in->hur_request.hr_data_len);
856 }
857
858 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
859                                  struct lustre_kernelcomm *lk, void *uarg)
860 {
861         __u32   i;
862         int     rc;
863         ENTRY;
864
865         /* unregister request (call from llapi_hsm_copytool_fini) */
866         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
867                 /* best effort: try to clean as much as possible
868                  * (continue on error) */
869                 obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp, len, lk, uarg);
870         }
871
872         /* Whatever the result, remove copytool from kuc groups.
873          * Unreached coordinators will get EPIPE on next requests
874          * and will unregister automatically.
875          */
876         rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group);
877         RETURN(rc);
878 }
879
880 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
881                                struct lustre_kernelcomm *lk, void *uarg)
882 {
883         struct file     *filp;
884         __u32            i, j;
885         int              err, rc;
886         bool             any_set = false;
887         ENTRY;
888
889         /* All or nothing: try to register to all MDS.
890          * In case of failure, unregister from previous MDS,
891          * except if it because of inactive target. */
892         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
893                 err = obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp,
894                                    len, lk, uarg);
895                 if (err) {
896                         if (lmv->tgts[i]->ltd_active) {
897                                 /* permanent error */
898                                 CERROR("%s: iocontrol MDC %s on MDT"
899                                        " idx %d cmd %x: err = %d\n",
900                                        class_exp2obd(lmv->exp)->obd_name,
901                                        lmv->tgts[i]->ltd_uuid.uuid,
902                                        i, cmd, err);
903                                 rc = err;
904                                 lk->lk_flags |= LK_FLG_STOP;
905                                 /* unregister from previous MDS */
906                                 for (j = 0; j < i; j++)
907                                         obd_iocontrol(cmd,
908                                                       lmv->tgts[j]->ltd_exp,
909                                                       len, lk, uarg);
910                                 RETURN(rc);
911                         }
912                         /* else: transient error.
913                          * kuc will register to the missing MDT
914                          * when it is back */
915                 } else {
916                         any_set = true;
917                 }
918         }
919
920         if (!any_set)
921                 /* no registration done: return error */
922                 RETURN(-ENOTCONN);
923
924         /* at least one registration done, with no failure */
925         filp = fget(lk->lk_wfd);
926         if (filp == NULL) {
927                 RETURN(-EBADF);
928         }
929         rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, lk->lk_data);
930         if (rc != 0 && filp != NULL)
931                 fput(filp);
932         RETURN(rc);
933 }
934
935
936
937
938 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
939                          int len, void *karg, void *uarg)
940 {
941         struct obd_device       *obddev = class_exp2obd(exp);
942         struct lmv_obd          *lmv = &obddev->u.lmv;
943         __u32                    i = 0;
944         int                      rc = 0;
945         int                      set = 0;
946         __u32                    count = lmv->desc.ld_tgt_count;
947         ENTRY;
948
949         if (count == 0)
950                 RETURN(-ENOTTY);
951
952         switch (cmd) {
953         case IOC_OBD_STATFS: {
954                 struct obd_ioctl_data *data = karg;
955                 struct obd_device *mdc_obd;
956                 struct obd_statfs stat_buf = {0};
957                 __u32 index;
958
959                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
960                 if ((index >= count))
961                         RETURN(-ENODEV);
962
963                 if (lmv->tgts[index] == NULL ||
964                     lmv->tgts[index]->ltd_active == 0)
965                         RETURN(-ENODATA);
966
967                 mdc_obd = class_exp2obd(lmv->tgts[index]->ltd_exp);
968                 if (!mdc_obd)
969                         RETURN(-EINVAL);
970
971                 /* copy UUID */
972                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
973                                  min((int) data->ioc_plen2,
974                                      (int) sizeof(struct obd_uuid))))
975                         RETURN(-EFAULT);
976
977                 rc = obd_statfs(NULL, lmv->tgts[index]->ltd_exp, &stat_buf,
978                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
979                                 0);
980                 if (rc)
981                         RETURN(rc);
982                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
983                                  min((int) data->ioc_plen1,
984                                      (int) sizeof(stat_buf))))
985                         RETURN(-EFAULT);
986                 break;
987         }
988         case OBD_IOC_QUOTACTL: {
989                 struct if_quotactl *qctl = karg;
990                 struct lmv_tgt_desc *tgt = NULL;
991                 struct obd_quotactl *oqctl;
992
993                 if (qctl->qc_valid == QC_MDTIDX) {
994                         if (count <= qctl->qc_idx)
995                                 RETURN(-EINVAL);
996
997                         tgt = lmv->tgts[qctl->qc_idx];
998                         if (tgt == NULL || tgt->ltd_exp == NULL)
999                                 RETURN(-EINVAL);
1000                 } else if (qctl->qc_valid == QC_UUID) {
1001                         for (i = 0; i < count; i++) {
1002                                 tgt = lmv->tgts[i];
1003                                 if (tgt == NULL)
1004                                         continue;
1005                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
1006                                                      &qctl->obd_uuid))
1007                                         continue;
1008
1009                                 if (tgt->ltd_exp == NULL)
1010                                         RETURN(-EINVAL);
1011
1012                                 break;
1013                         }
1014                 } else {
1015                         RETURN(-EINVAL);
1016                 }
1017
1018                 if (i >= count)
1019                         RETURN(-EAGAIN);
1020
1021                 LASSERT(tgt && tgt->ltd_exp);
1022                 OBD_ALLOC_PTR(oqctl);
1023                 if (!oqctl)
1024                         RETURN(-ENOMEM);
1025
1026                 QCTL_COPY(oqctl, qctl);
1027                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1028                 if (rc == 0) {
1029                         QCTL_COPY(qctl, oqctl);
1030                         qctl->qc_valid = QC_MDTIDX;
1031                         qctl->obd_uuid = tgt->ltd_uuid;
1032                 }
1033                 OBD_FREE_PTR(oqctl);
1034                 break;
1035         }
1036         case OBD_IOC_CHANGELOG_SEND:
1037         case OBD_IOC_CHANGELOG_CLEAR: {
1038                 struct ioc_changelog *icc = karg;
1039
1040                 if (icc->icc_mdtindex >= count)
1041                         RETURN(-ENODEV);
1042
1043                 if (lmv->tgts[icc->icc_mdtindex] == NULL ||
1044                     lmv->tgts[icc->icc_mdtindex]->ltd_exp == NULL ||
1045                     lmv->tgts[icc->icc_mdtindex]->ltd_active == 0)
1046                         RETURN(-ENODEV);
1047                 rc = obd_iocontrol(cmd, lmv->tgts[icc->icc_mdtindex]->ltd_exp,
1048                                    sizeof(*icc), icc, NULL);
1049                 break;
1050         }
1051         case LL_IOC_GET_CONNECT_FLAGS: {
1052                 if (lmv->tgts[0] == NULL)
1053                         RETURN(-ENODATA);
1054                 rc = obd_iocontrol(cmd, lmv->tgts[0]->ltd_exp, len, karg, uarg);
1055                 break;
1056         }
1057         case OBD_IOC_FID2PATH: {
1058                 rc = lmv_fid2path(exp, len, karg, uarg);
1059                 break;
1060         }
1061         case LL_IOC_HSM_STATE_GET:
1062         case LL_IOC_HSM_STATE_SET:
1063         case LL_IOC_HSM_ACTION: {
1064                 struct md_op_data       *op_data = karg;
1065                 struct lmv_tgt_desc     *tgt;
1066
1067                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1068                 if (IS_ERR(tgt))
1069                         RETURN(PTR_ERR(tgt));
1070
1071                 if (tgt->ltd_exp == NULL)
1072                         RETURN(-EINVAL);
1073
1074                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1075                 break;
1076         }
1077         case LL_IOC_HSM_PROGRESS: {
1078                 const struct hsm_progress_kernel *hpk = karg;
1079                 struct lmv_tgt_desc     *tgt;
1080
1081                 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1082                 if (IS_ERR(tgt))
1083                         RETURN(PTR_ERR(tgt));
1084                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1085                 break;
1086         }
1087         case LL_IOC_HSM_REQUEST: {
1088                 struct hsm_user_request *hur = karg;
1089                 struct lmv_tgt_desc     *tgt;
1090                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1091
1092                 if (reqcount == 0)
1093                         RETURN(0);
1094
1095                 /* if the request is about a single fid
1096                  * or if there is a single MDS, no need to split
1097                  * the request. */
1098                 if (reqcount == 1 || count == 1) {
1099                         tgt = lmv_find_target(lmv,
1100                                               &hur->hur_user_item[0].hui_fid);
1101                         if (IS_ERR(tgt))
1102                                 RETURN(PTR_ERR(tgt));
1103                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1104                 } else {
1105                         /* split fid list to their respective MDS */
1106                         for (i = 0; i < count; i++) {
1107                                 unsigned int            nr, reqlen;
1108                                 int                     rc1;
1109                                 struct hsm_user_request *req;
1110
1111                                 nr = lmv_hsm_req_count(lmv, hur, lmv->tgts[i]);
1112                                 if (nr == 0) /* nothing for this MDS */
1113                                         continue;
1114
1115                                 /* build a request with fids for this MDS */
1116                                 reqlen = offsetof(typeof(*hur),
1117                                                   hur_user_item[nr])
1118                                                 + hur->hur_request.hr_data_len;
1119                                 OBD_ALLOC_LARGE(req, reqlen);
1120                                 if (req == NULL)
1121                                         RETURN(-ENOMEM);
1122
1123                                 lmv_hsm_req_build(lmv, hur, lmv->tgts[i], req);
1124
1125                                 rc1 = obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp,
1126                                                     reqlen, req, uarg);
1127                                 if (rc1 != 0 && rc == 0)
1128                                         rc = rc1;
1129                                 OBD_FREE_LARGE(req, reqlen);
1130                         }
1131                 }
1132                 break;
1133         }
1134         case LL_IOC_LOV_SWAP_LAYOUTS: {
1135                 struct md_op_data       *op_data = karg;
1136                 struct lmv_tgt_desc     *tgt1, *tgt2;
1137
1138                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1139                 if (IS_ERR(tgt1))
1140                         RETURN(PTR_ERR(tgt1));
1141
1142                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1143                 if (IS_ERR(tgt2))
1144                         RETURN(PTR_ERR(tgt2));
1145
1146                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1147                         RETURN(-EINVAL);
1148
1149                 /* only files on same MDT can have their layouts swapped */
1150                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1151                         RETURN(-EPERM);
1152
1153                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1154                 break;
1155         }
1156         case LL_IOC_HSM_CT_START: {
1157                 struct lustre_kernelcomm *lk = karg;
1158                 if (lk->lk_flags & LK_FLG_STOP)
1159                         rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1160                 else
1161                         rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1162                 break;
1163         }
1164         default:
1165                 for (i = 0; i < count; i++) {
1166                         struct obd_device *mdc_obd;
1167                         int err;
1168
1169                         if (lmv->tgts[i] == NULL ||
1170                             lmv->tgts[i]->ltd_exp == NULL)
1171                                 continue;
1172                         /* ll_umount_begin() sets force flag but for lmv, not
1173                          * mdc. Let's pass it through */
1174                         mdc_obd = class_exp2obd(lmv->tgts[i]->ltd_exp);
1175                         mdc_obd->obd_force = obddev->obd_force;
1176                         err = obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp, len,
1177                                             karg, uarg);
1178                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1179                                 RETURN(err);
1180                         } else if (err) {
1181                                 if (lmv->tgts[i]->ltd_active) {
1182                                         CERROR("error: iocontrol MDC %s on MDT"
1183                                                " idx %d cmd %x: err = %d\n",
1184                                                lmv->tgts[i]->ltd_uuid.uuid,
1185                                                i, cmd, err);
1186                                         if (!rc)
1187                                                 rc = err;
1188                                 }
1189                         } else
1190                                 set = 1;
1191                 }
1192                 if (!set && !rc)
1193                         rc = -EIO;
1194         }
1195         RETURN(rc);
1196 }
1197
1198 #if 0
1199 static int lmv_all_chars_policy(int count, const char *name,
1200                                 int len)
1201 {
1202         unsigned int c = 0;
1203
1204         while (len > 0)
1205                 c += name[--len];
1206         c = c % count;
1207         return c;
1208 }
1209
1210 static int lmv_nid_policy(struct lmv_obd *lmv)
1211 {
1212         struct obd_import *imp;
1213         __u32              id;
1214
1215         /*
1216          * XXX: To get nid we assume that underlying obd device is mdc.
1217          */
1218         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1219         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1220         return id % lmv->desc.ld_tgt_count;
1221 }
1222
1223 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1224                           placement_policy_t placement)
1225 {
1226         switch (placement) {
1227         case PLACEMENT_CHAR_POLICY:
1228                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1229                                             op_data->op_name,
1230                                             op_data->op_namelen);
1231         case PLACEMENT_NID_POLICY:
1232                 return lmv_nid_policy(lmv);
1233
1234         default:
1235                 break;
1236         }
1237
1238         CERROR("Unsupported placement policy %x\n", placement);
1239         return -EINVAL;
1240 }
1241 #endif
1242
1243 /**
1244  * This is _inode_ placement policy function (not name).
1245  */
1246 static int lmv_placement_policy(struct obd_device *obd,
1247                                 struct md_op_data *op_data,
1248                                 mdsno_t *mds)
1249 {
1250         struct lmv_obd          *lmv = &obd->u.lmv;
1251         ENTRY;
1252
1253         LASSERT(mds != NULL);
1254
1255         if (lmv->desc.ld_tgt_count == 1) {
1256                 *mds = 0;
1257                 RETURN(0);
1258         }
1259
1260         /**
1261          * If stripe_offset is provided during setdirstripe
1262          * (setdirstripe -i xx), xx MDS will be choosen.
1263          */
1264         if (op_data->op_cli_flags & CLI_SET_MEA) {
1265                 struct lmv_user_md *lum;
1266
1267                 lum = (struct lmv_user_md *)op_data->op_data;
1268                 if (lum->lum_type == LMV_STRIPE_TYPE &&
1269                     lum->lum_stripe_offset != -1) {
1270                         if (lum->lum_stripe_offset >= lmv->desc.ld_tgt_count) {
1271                                 CERROR("%s: Stripe_offset %d > MDT count %d:"
1272                                        " rc = %d\n", obd->obd_name,
1273                                        lum->lum_stripe_offset,
1274                                        lmv->desc.ld_tgt_count, -ERANGE);
1275                                 RETURN(-ERANGE);
1276                         }
1277                         *mds = lum->lum_stripe_offset;
1278                         RETURN(0);
1279                 }
1280         }
1281
1282         /* Allocate new fid on target according to operation type and parent
1283          * home mds. */
1284         *mds = op_data->op_mds;
1285         RETURN(0);
1286 }
1287
1288 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
1289                     mdsno_t mds)
1290 {
1291         struct lmv_tgt_desc     *tgt;
1292         int                      rc;
1293         ENTRY;
1294
1295         tgt = lmv_get_target(lmv, mds);
1296         if (IS_ERR(tgt))
1297                 RETURN(PTR_ERR(tgt));
1298
1299         /*
1300          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1301          * on server that seq in new allocated fid is not yet known.
1302          */
1303         mutex_lock(&tgt->ltd_fid_mutex);
1304
1305         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1306                 GOTO(out, rc = -ENODEV);
1307
1308         /*
1309          * Asking underlaying tgt layer to allocate new fid.
1310          */
1311         rc = obd_fid_alloc(tgt->ltd_exp, fid, NULL);
1312         if (rc > 0) {
1313                 LASSERT(fid_is_sane(fid));
1314                 rc = 0;
1315         }
1316
1317         EXIT;
1318 out:
1319         mutex_unlock(&tgt->ltd_fid_mutex);
1320         return rc;
1321 }
1322
1323 int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1324                   struct md_op_data *op_data)
1325 {
1326         struct obd_device     *obd = class_exp2obd(exp);
1327         struct lmv_obd        *lmv = &obd->u.lmv;
1328         mdsno_t                mds = 0;
1329         int                    rc;
1330         ENTRY;
1331
1332         LASSERT(op_data != NULL);
1333         LASSERT(fid != NULL);
1334
1335         rc = lmv_placement_policy(obd, op_data, &mds);
1336         if (rc) {
1337                 CERROR("Can't get target for allocating fid, "
1338                        "rc %d\n", rc);
1339                 RETURN(rc);
1340         }
1341
1342         rc = __lmv_fid_alloc(lmv, fid, mds);
1343         if (rc) {
1344                 CERROR("Can't alloc new fid, rc %d\n", rc);
1345                 RETURN(rc);
1346         }
1347
1348         RETURN(rc);
1349 }
1350
1351 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1352 {
1353         struct lmv_obd             *lmv = &obd->u.lmv;
1354         struct lprocfs_static_vars  lvars;
1355         struct lmv_desc            *desc;
1356         int                         rc;
1357         ENTRY;
1358
1359         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1360                 CERROR("LMV setup requires a descriptor\n");
1361                 RETURN(-EINVAL);
1362         }
1363
1364         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1365         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1366                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1367                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1368                 RETURN(-EINVAL);
1369         }
1370
1371         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * 32);
1372         if (lmv->tgts == NULL)
1373                 RETURN(-ENOMEM);
1374         lmv->tgts_size = 32;
1375
1376         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1377         lmv->desc.ld_tgt_count = 0;
1378         lmv->desc.ld_active_tgt_count = 0;
1379         lmv->max_cookiesize = 0;
1380         lmv->max_def_easize = 0;
1381         lmv->max_easize = 0;
1382         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1383
1384         spin_lock_init(&lmv->lmv_lock);
1385         mutex_init(&lmv->init_mutex);
1386
1387         lprocfs_lmv_init_vars(&lvars);
1388
1389         lprocfs_obd_setup(obd, lvars.obd_vars);
1390         lprocfs_alloc_md_stats(obd, 0);
1391 #ifdef LPROCFS
1392         {
1393                 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1394                                         0444, &lmv_proc_target_fops, obd);
1395                 if (rc)
1396                         CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1397                               obd->obd_name, rc);
1398         }
1399 #endif
1400         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1401                              LUSTRE_CLI_FLD_HASH_DHT);
1402         if (rc) {
1403                 CERROR("Can't init FLD, err %d\n", rc);
1404                 GOTO(out, rc);
1405         }
1406
1407         RETURN(0);
1408
1409 out:
1410         return rc;
1411 }
1412
1413 static int lmv_cleanup(struct obd_device *obd)
1414 {
1415         struct lmv_obd   *lmv = &obd->u.lmv;
1416         ENTRY;
1417
1418         fld_client_fini(&lmv->lmv_fld);
1419         if (lmv->tgts != NULL) {
1420                 int i;
1421                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1422                         if (lmv->tgts[i] == NULL)
1423                                 continue;
1424                         lmv_del_target(lmv, i);
1425                 }
1426                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1427                 lmv->tgts_size = 0;
1428         }
1429         RETURN(0);
1430 }
1431
1432 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1433 {
1434         struct lustre_cfg       *lcfg = buf;
1435         struct obd_uuid         obd_uuid;
1436         int                     gen;
1437         __u32                   index;
1438         int                     rc;
1439         ENTRY;
1440
1441         switch (lcfg->lcfg_command) {
1442         case LCFG_ADD_MDC:
1443                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1444                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1445                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1446                         GOTO(out, rc = -EINVAL);
1447
1448                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1449
1450                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
1451                         GOTO(out, rc = -EINVAL);
1452                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1453                         GOTO(out, rc = -EINVAL);
1454                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1455                 GOTO(out, rc);
1456         default:
1457                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1458                 GOTO(out, rc = -EINVAL);
1459         }
1460 out:
1461         RETURN(rc);
1462 }
1463
1464 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1465                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1466 {
1467         struct obd_device       *obd = class_exp2obd(exp);
1468         struct lmv_obd          *lmv = &obd->u.lmv;
1469         struct obd_statfs       *temp;
1470         int                      rc = 0;
1471         __u32                    i;
1472         ENTRY;
1473
1474         rc = lmv_check_connect(obd);
1475         if (rc)
1476                 RETURN(rc);
1477
1478         OBD_ALLOC(temp, sizeof(*temp));
1479         if (temp == NULL)
1480                 RETURN(-ENOMEM);
1481
1482         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1483                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1484                         continue;
1485
1486                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1487                                 max_age, flags);
1488                 if (rc) {
1489                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1490                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1491                                rc);
1492                         GOTO(out_free_temp, rc);
1493                 }
1494
1495                 if (i == 0) {
1496                         *osfs = *temp;
1497                         /* If the statfs is from mount, it will needs
1498                          * retrieve necessary information from MDT0.
1499                          * i.e. mount does not need the merged osfs
1500                          * from all of MDT.
1501                          * And also clients can be mounted as long as
1502                          * MDT0 is in service*/
1503                         if (flags & OBD_STATFS_FOR_MDT0)
1504                                 GOTO(out_free_temp, rc);
1505                 } else {
1506                         osfs->os_bavail += temp->os_bavail;
1507                         osfs->os_blocks += temp->os_blocks;
1508                         osfs->os_ffree += temp->os_ffree;
1509                         osfs->os_files += temp->os_files;
1510                 }
1511         }
1512
1513         EXIT;
1514 out_free_temp:
1515         OBD_FREE(temp, sizeof(*temp));
1516         return rc;
1517 }
1518
1519 static int lmv_getstatus(struct obd_export *exp,
1520                          struct lu_fid *fid,
1521                          struct obd_capa **pc)
1522 {
1523         struct obd_device    *obd = exp->exp_obd;
1524         struct lmv_obd       *lmv = &obd->u.lmv;
1525         int                   rc;
1526         ENTRY;
1527
1528         rc = lmv_check_connect(obd);
1529         if (rc)
1530                 RETURN(rc);
1531
1532         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1533         RETURN(rc);
1534 }
1535
1536 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1537                         struct obd_capa *oc, obd_valid valid, const char *name,
1538                         const char *input, int input_size, int output_size,
1539                         int flags, struct ptlrpc_request **request)
1540 {
1541         struct obd_device      *obd = exp->exp_obd;
1542         struct lmv_obd         *lmv = &obd->u.lmv;
1543         struct lmv_tgt_desc    *tgt;
1544         int                     rc;
1545         ENTRY;
1546
1547         rc = lmv_check_connect(obd);
1548         if (rc)
1549                 RETURN(rc);
1550
1551         tgt = lmv_find_target(lmv, fid);
1552         if (IS_ERR(tgt))
1553                 RETURN(PTR_ERR(tgt));
1554
1555         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1556                          input_size, output_size, flags, request);
1557
1558         RETURN(rc);
1559 }
1560
1561 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1562                         struct obd_capa *oc, obd_valid valid, const char *name,
1563                         const char *input, int input_size, int output_size,
1564                         int flags, __u32 suppgid,
1565                         struct ptlrpc_request **request)
1566 {
1567         struct obd_device      *obd = exp->exp_obd;
1568         struct lmv_obd         *lmv = &obd->u.lmv;
1569         struct lmv_tgt_desc    *tgt;
1570         int                     rc;
1571         ENTRY;
1572
1573         rc = lmv_check_connect(obd);
1574         if (rc)
1575                 RETURN(rc);
1576
1577         tgt = lmv_find_target(lmv, fid);
1578         if (IS_ERR(tgt))
1579                 RETURN(PTR_ERR(tgt));
1580
1581         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1582                          input_size, output_size, flags, suppgid,
1583                          request);
1584
1585         RETURN(rc);
1586 }
1587
1588 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1589                        struct ptlrpc_request **request)
1590 {
1591         struct obd_device       *obd = exp->exp_obd;
1592         struct lmv_obd          *lmv = &obd->u.lmv;
1593         struct lmv_tgt_desc     *tgt;
1594         int                      rc;
1595         ENTRY;
1596
1597         rc = lmv_check_connect(obd);
1598         if (rc)
1599                 RETURN(rc);
1600
1601         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1602         if (IS_ERR(tgt))
1603                 RETURN(PTR_ERR(tgt));
1604
1605         if (op_data->op_flags & MF_GET_MDT_IDX) {
1606                 op_data->op_mds = tgt->ltd_idx;
1607                 RETURN(0);
1608         }
1609
1610         rc = md_getattr(tgt->ltd_exp, op_data, request);
1611
1612         RETURN(rc);
1613 }
1614
1615 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1616 {
1617         struct obd_device   *obd = exp->exp_obd;
1618         struct lmv_obd      *lmv = &obd->u.lmv;
1619         __u32                i;
1620         int                  rc;
1621         ENTRY;
1622
1623         rc = lmv_check_connect(obd);
1624         if (rc)
1625                 RETURN(rc);
1626
1627         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1628
1629         /*
1630          * With DNE every object can have two locks in different namespaces:
1631          * lookup lock in space of MDT storing direntry and update/open lock in
1632          * space of MDT storing inode.
1633          */
1634         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1635                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1636                         continue;
1637                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1638         }
1639
1640         RETURN(0);
1641 }
1642
1643 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1644                            ldlm_iterator_t it, void *data)
1645 {
1646         struct obd_device   *obd = exp->exp_obd;
1647         struct lmv_obd      *lmv = &obd->u.lmv;
1648         __u32                i;
1649         int                  rc;
1650         ENTRY;
1651
1652         rc = lmv_check_connect(obd);
1653         if (rc)
1654                 RETURN(rc);
1655
1656         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1657
1658         /*
1659          * With DNE every object can have two locks in different namespaces:
1660          * lookup lock in space of MDT storing direntry and update/open lock in
1661          * space of MDT storing inode.
1662          */
1663         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1664                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1665                         continue;
1666                 rc = md_find_cbdata(lmv->tgts[i]->ltd_exp, fid, it, data);
1667                 if (rc)
1668                         RETURN(rc);
1669         }
1670
1671         RETURN(rc);
1672 }
1673
1674
1675 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1676                      struct md_open_data *mod, struct ptlrpc_request **request)
1677 {
1678         struct obd_device     *obd = exp->exp_obd;
1679         struct lmv_obd        *lmv = &obd->u.lmv;
1680         struct lmv_tgt_desc   *tgt;
1681         int                    rc;
1682         ENTRY;
1683
1684         rc = lmv_check_connect(obd);
1685         if (rc)
1686                 RETURN(rc);
1687
1688         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1689         if (IS_ERR(tgt))
1690                 RETURN(PTR_ERR(tgt));
1691
1692         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1693         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1694         RETURN(rc);
1695 }
1696
1697 struct lmv_tgt_desc
1698 *lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1699                 struct lu_fid *fid)
1700 {
1701         struct lmv_tgt_desc *tgt;
1702
1703         tgt = lmv_find_target(lmv, fid);
1704         if (IS_ERR(tgt))
1705                 return tgt;
1706
1707         op_data->op_mds = tgt->ltd_idx;
1708
1709         return tgt;
1710 }
1711
1712 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1713                const void *data, int datalen, int mode, __u32 uid,
1714                __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1715                struct ptlrpc_request **request)
1716 {
1717         struct obd_device       *obd = exp->exp_obd;
1718         struct lmv_obd          *lmv = &obd->u.lmv;
1719         struct lmv_tgt_desc     *tgt;
1720         int                      rc;
1721         ENTRY;
1722
1723         rc = lmv_check_connect(obd);
1724         if (rc)
1725                 RETURN(rc);
1726
1727         if (!lmv->desc.ld_active_tgt_count)
1728                 RETURN(-EIO);
1729
1730         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1731         if (IS_ERR(tgt))
1732                 RETURN(PTR_ERR(tgt));
1733
1734         rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data);
1735         if (rc)
1736                 RETURN(rc);
1737
1738         CDEBUG(D_INODE, "CREATE '%*s' on "DFID" -> mds #%x\n",
1739                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1740                op_data->op_mds);
1741
1742         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1743         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1744                        cap_effective, rdev, request);
1745
1746         if (rc == 0) {
1747                 if (*request == NULL)
1748                         RETURN(rc);
1749                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1750         }
1751         RETURN(rc);
1752 }
1753
1754 static int lmv_done_writing(struct obd_export *exp,
1755                             struct md_op_data *op_data,
1756                             struct md_open_data *mod)
1757 {
1758         struct obd_device     *obd = exp->exp_obd;
1759         struct lmv_obd        *lmv = &obd->u.lmv;
1760         struct lmv_tgt_desc   *tgt;
1761         int                    rc;
1762         ENTRY;
1763
1764         rc = lmv_check_connect(obd);
1765         if (rc)
1766                 RETURN(rc);
1767
1768         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1769         if (IS_ERR(tgt))
1770                 RETURN(PTR_ERR(tgt));
1771
1772         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1773         RETURN(rc);
1774 }
1775
1776 static int
1777 lmv_enqueue_remote(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1778                    struct lookup_intent *it, struct md_op_data *op_data,
1779                    struct lustre_handle *lockh, void *lmm, int lmmsize,
1780                    int extra_lock_flags)
1781 {
1782         struct ptlrpc_request      *req = it->d.lustre.it_data;
1783         struct obd_device          *obd = exp->exp_obd;
1784         struct lmv_obd             *lmv = &obd->u.lmv;
1785         struct lustre_handle        plock;
1786         struct lmv_tgt_desc        *tgt;
1787         struct md_op_data          *rdata;
1788         struct lu_fid               fid1;
1789         struct mdt_body            *body;
1790         int                         rc = 0;
1791         int                         pmode;
1792         ENTRY;
1793
1794         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1795         LASSERT(body != NULL);
1796
1797         if (!(body->valid & OBD_MD_MDS))
1798                 RETURN(0);
1799
1800         CDEBUG(D_INODE, "REMOTE_ENQUEUE '%s' on "DFID" -> "DFID"\n",
1801                LL_IT2STR(it), PFID(&op_data->op_fid1), PFID(&body->fid1));
1802
1803         /*
1804          * We got LOOKUP lock, but we really need attrs.
1805          */
1806         pmode = it->d.lustre.it_lock_mode;
1807         LASSERT(pmode != 0);
1808         memcpy(&plock, lockh, sizeof(plock));
1809         it->d.lustre.it_lock_mode = 0;
1810         it->d.lustre.it_data = NULL;
1811         fid1 = body->fid1;
1812
1813         it->d.lustre.it_disposition &= ~DISP_ENQ_COMPLETE;
1814         ptlrpc_req_finished(req);
1815
1816         tgt = lmv_find_target(lmv, &fid1);
1817         if (IS_ERR(tgt))
1818                 GOTO(out, rc = PTR_ERR(tgt));
1819
1820         OBD_ALLOC_PTR(rdata);
1821         if (rdata == NULL)
1822                 GOTO(out, rc = -ENOMEM);
1823
1824         rdata->op_fid1 = fid1;
1825         rdata->op_bias = MDS_CROSS_REF;
1826
1827         rc = md_enqueue(tgt->ltd_exp, einfo, it, rdata, lockh,
1828                         lmm, lmmsize, NULL, extra_lock_flags);
1829         OBD_FREE_PTR(rdata);
1830         EXIT;
1831 out:
1832         ldlm_lock_decref(&plock, pmode);
1833         return rc;
1834 }
1835
1836 static int
1837 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1838             struct lookup_intent *it, struct md_op_data *op_data,
1839             struct lustre_handle *lockh, void *lmm, int lmmsize,
1840             struct ptlrpc_request **req, __u64 extra_lock_flags)
1841 {
1842         struct obd_device        *obd = exp->exp_obd;
1843         struct lmv_obd           *lmv = &obd->u.lmv;
1844         struct lmv_tgt_desc      *tgt;
1845         int                       rc;
1846         ENTRY;
1847
1848         rc = lmv_check_connect(obd);
1849         if (rc)
1850                 RETURN(rc);
1851
1852         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1853                LL_IT2STR(it), PFID(&op_data->op_fid1));
1854
1855         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1856         if (IS_ERR(tgt))
1857                 RETURN(PTR_ERR(tgt));
1858
1859         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1860                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1861
1862         rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data, lockh,
1863                         lmm, lmmsize, req, extra_lock_flags);
1864
1865         if (rc == 0 && it && it->it_op == IT_OPEN) {
1866                 rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh,
1867                                         lmm, lmmsize, extra_lock_flags);
1868         }
1869         RETURN(rc);
1870 }
1871
1872 static int
1873 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1874                  struct ptlrpc_request **request)
1875 {
1876         struct ptlrpc_request   *req = NULL;
1877         struct obd_device       *obd = exp->exp_obd;
1878         struct lmv_obd          *lmv = &obd->u.lmv;
1879         struct lmv_tgt_desc     *tgt;
1880         struct mdt_body         *body;
1881         int                      rc;
1882         ENTRY;
1883
1884         rc = lmv_check_connect(obd);
1885         if (rc)
1886                 RETURN(rc);
1887
1888         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1889         if (IS_ERR(tgt))
1890                 RETURN(PTR_ERR(tgt));
1891
1892         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1893                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1894                tgt->ltd_idx);
1895
1896         rc = md_getattr_name(tgt->ltd_exp, op_data, request);
1897         if (rc != 0)
1898                 RETURN(rc);
1899
1900         body = req_capsule_server_get(&(*request)->rq_pill,
1901                                       &RMF_MDT_BODY);
1902         LASSERT(body != NULL);
1903
1904         if (body->valid & OBD_MD_MDS) {
1905                 struct lu_fid rid = body->fid1;
1906                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1907                        PFID(&rid));
1908
1909                 tgt = lmv_find_target(lmv, &rid);
1910                 if (IS_ERR(tgt)) {
1911                         ptlrpc_req_finished(*request);
1912                         RETURN(PTR_ERR(tgt));
1913                 }
1914
1915                 op_data->op_fid1 = rid;
1916                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1917                 op_data->op_namelen = 0;
1918                 op_data->op_name = NULL;
1919                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1920                 ptlrpc_req_finished(*request);
1921                 *request = req;
1922         }
1923
1924         RETURN(rc);
1925 }
1926
1927 #define md_op_data_fid(op_data, fl)                     \
1928         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1929          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1930          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1931          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1932          NULL)
1933
1934 static int lmv_early_cancel(struct obd_export *exp, struct md_op_data *op_data,
1935                             int op_tgt, ldlm_mode_t mode, int bits, int flag)
1936 {
1937         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
1938         struct obd_device      *obd = exp->exp_obd;
1939         struct lmv_obd         *lmv = &obd->u.lmv;
1940         struct lmv_tgt_desc    *tgt;
1941         ldlm_policy_data_t      policy = {{0}};
1942         int                     rc = 0;
1943         ENTRY;
1944
1945         if (!fid_is_sane(fid))
1946                 RETURN(0);
1947
1948         tgt = lmv_find_target(lmv, fid);
1949         if (IS_ERR(tgt))
1950                 RETURN(PTR_ERR(tgt));
1951
1952         if (tgt->ltd_idx != op_tgt) {
1953                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1954                 policy.l_inodebits.bits = bits;
1955                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1956                                       mode, LCF_ASYNC, NULL);
1957         } else {
1958                 CDEBUG(D_INODE,
1959                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1960                        op_tgt, PFID(fid));
1961                 op_data->op_flags |= flag;
1962                 rc = 0;
1963         }
1964
1965         RETURN(rc);
1966 }
1967
1968 /*
1969  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1970  * op_data->op_fid2
1971  */
1972 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1973                     struct ptlrpc_request **request)
1974 {
1975         struct obd_device       *obd = exp->exp_obd;
1976         struct lmv_obd          *lmv = &obd->u.lmv;
1977         struct lmv_tgt_desc     *tgt;
1978         int                      rc;
1979         ENTRY;
1980
1981         rc = lmv_check_connect(obd);
1982         if (rc)
1983                 RETURN(rc);
1984
1985         LASSERT(op_data->op_namelen != 0);
1986
1987         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
1988                PFID(&op_data->op_fid2), op_data->op_namelen,
1989                op_data->op_name, PFID(&op_data->op_fid1));
1990
1991         op_data->op_fsuid = current_fsuid();
1992         op_data->op_fsgid = current_fsgid();
1993         op_data->op_cap = cfs_curproc_cap_pack();
1994         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
1995         if (IS_ERR(tgt))
1996                 RETURN(PTR_ERR(tgt));
1997
1998         /*
1999          * Cancel UPDATE lock on child (fid1).
2000          */
2001         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2002         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
2003                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2004         if (rc != 0)
2005                 RETURN(rc);
2006
2007         rc = md_link(tgt->ltd_exp, op_data, request);
2008
2009         RETURN(rc);
2010 }
2011
2012 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2013                       const char *old, int oldlen, const char *new, int newlen,
2014                       struct ptlrpc_request **request)
2015 {
2016         struct obd_device       *obd = exp->exp_obd;
2017         struct lmv_obd          *lmv = &obd->u.lmv;
2018         struct lmv_tgt_desc     *src_tgt;
2019         struct lmv_tgt_desc     *tgt_tgt;
2020         int                     rc;
2021         ENTRY;
2022
2023         LASSERT(oldlen != 0);
2024
2025         CDEBUG(D_INODE, "RENAME %*s in "DFID" to %*s in "DFID"\n",
2026                oldlen, old, PFID(&op_data->op_fid1),
2027                newlen, new, PFID(&op_data->op_fid2));
2028
2029         rc = lmv_check_connect(obd);
2030         if (rc)
2031                 RETURN(rc);
2032
2033         op_data->op_fsuid = current_fsuid();
2034         op_data->op_fsgid = current_fsgid();
2035         op_data->op_cap = cfs_curproc_cap_pack();
2036         src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2037         if (IS_ERR(src_tgt))
2038                 RETURN(PTR_ERR(src_tgt));
2039
2040         tgt_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2041         if (IS_ERR(tgt_tgt))
2042                 RETURN(PTR_ERR(tgt_tgt));
2043         /*
2044          * LOOKUP lock on src child (fid3) should also be cancelled for
2045          * src_tgt in mdc_rename.
2046          */
2047         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2048
2049         /*
2050          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2051          * own target.
2052          */
2053         rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2054                               LCK_EX, MDS_INODELOCK_UPDATE,
2055                               MF_MDC_CANCEL_FID2);
2056
2057         /*
2058          * Cancel LOOKUP locks on tgt child (fid4) for parent tgt_tgt.
2059          */
2060         if (rc == 0) {
2061                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2062                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2063                                       MF_MDC_CANCEL_FID4);
2064         }
2065
2066         /*
2067          * Cancel all the locks on tgt child (fid4).
2068          */
2069         if (rc == 0)
2070                 rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx,
2071                                       LCK_EX, MDS_INODELOCK_FULL,
2072                                       MF_MDC_CANCEL_FID4);
2073
2074         if (rc == 0)
2075                 rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen,
2076                                new, newlen, request);
2077         RETURN(rc);
2078 }
2079
2080 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2081                        void *ea, int ealen, void *ea2, int ea2len,
2082                        struct ptlrpc_request **request,
2083                        struct md_open_data **mod)
2084 {
2085         struct obd_device       *obd = exp->exp_obd;
2086         struct lmv_obd          *lmv = &obd->u.lmv;
2087         struct lmv_tgt_desc     *tgt;
2088         int                      rc = 0;
2089         ENTRY;
2090
2091         rc = lmv_check_connect(obd);
2092         if (rc)
2093                 RETURN(rc);
2094
2095         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2096                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2097
2098         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2099         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2100         if (IS_ERR(tgt))
2101                 RETURN(PTR_ERR(tgt));
2102
2103         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2104                         ea2len, request, mod);
2105
2106         RETURN(rc);
2107 }
2108
2109 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2110                      struct obd_capa *oc, struct ptlrpc_request **request)
2111 {
2112         struct obd_device       *obd = exp->exp_obd;
2113         struct lmv_obd          *lmv = &obd->u.lmv;
2114         struct lmv_tgt_desc     *tgt;
2115         int                      rc;
2116         ENTRY;
2117
2118         rc = lmv_check_connect(obd);
2119         if (rc != 0)
2120                 RETURN(rc);
2121
2122         tgt = lmv_find_target(lmv, fid);
2123         if (IS_ERR(tgt))
2124                 RETURN(PTR_ERR(tgt));
2125
2126         rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2127         RETURN(rc);
2128 }
2129
2130 /*
2131  * Adjust a set of pages, each page containing an array of lu_dirpages,
2132  * so that each page can be used as a single logical lu_dirpage.
2133  *
2134  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
2135  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
2136  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
2137  * value is used as a cookie to request the next lu_dirpage in a
2138  * directory listing that spans multiple pages (two in this example):
2139  *   ________
2140  *  |        |
2141  * .|--------v-------   -----.
2142  * |s|e|f|p|ent|ent| ... |ent|
2143  * '--|--------------   -----'   Each CFS_PAGE contains a single
2144  *    '------.                   lu_dirpage.
2145  * .---------v-------   -----.
2146  * |s|e|f|p|ent| 0 | ... | 0 |
2147  * '-----------------   -----'
2148  *
2149  * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
2150  * larger than LU_PAGE_SIZE, a single host page may contain multiple
2151  * lu_dirpages. After reading the lu_dirpages from the MDS, the
2152  * ldp_hash_end of the first lu_dirpage refers to the one immediately
2153  * after it in the same CFS_PAGE (arrows simplified for brevity, but
2154  * in general e0==s1, e1==s2, etc.):
2155  *
2156  * .--------------------   -----.
2157  * |s0|e0|f0|p|ent|ent| ... |ent|
2158  * |---v----------------   -----|
2159  * |s1|e1|f1|p|ent|ent| ... |ent|
2160  * |---v----------------   -----|  Here, each CFS_PAGE contains
2161  *             ...                 multiple lu_dirpages.
2162  * |---v----------------   -----|
2163  * |s'|e'|f'|p|ent|ent| ... |ent|
2164  * '---|----------------   -----'
2165  *     v
2166  * .----------------------------.
2167  * |        next CFS_PAGE       |
2168  *
2169  * This structure is transformed into a single logical lu_dirpage as follows:
2170  *
2171  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
2172  *   labeled 'next CFS_PAGE'.
2173  *
2174  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
2175  *   a hash collision with the next page exists.
2176  *
2177  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
2178  *   to the first entry of the next lu_dirpage.
2179  */
2180 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
2181 static void lmv_adjust_dirpages(struct page **pages, int ncfspgs, int nlupgs)
2182 {
2183         int i;
2184
2185         for (i = 0; i < ncfspgs; i++) {
2186                 struct lu_dirpage       *dp = kmap(pages[i]);
2187                 struct lu_dirpage       *first = dp;
2188                 struct lu_dirent        *end_dirent = NULL;
2189                 struct lu_dirent        *ent;
2190                 __u64                   hash_end = dp->ldp_hash_end;
2191                 __u32                   flags = dp->ldp_flags;
2192
2193                 while (--nlupgs > 0) {
2194                         ent = lu_dirent_start(dp);
2195                         for (end_dirent = ent; ent != NULL;
2196                              end_dirent = ent, ent = lu_dirent_next(ent));
2197
2198                         /* Advance dp to next lu_dirpage. */
2199                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
2200
2201                         /* Check if we've reached the end of the CFS_PAGE. */
2202                         if (!((unsigned long)dp & ~CFS_PAGE_MASK))
2203                                 break;
2204
2205                         /* Save the hash and flags of this lu_dirpage. */
2206                         hash_end = dp->ldp_hash_end;
2207                         flags = dp->ldp_flags;
2208
2209                         /* Check if lu_dirpage contains no entries. */
2210                         if (!end_dirent)
2211                                 break;
2212
2213                         /* Enlarge the end entry lde_reclen from 0 to
2214                          * first entry of next lu_dirpage. */
2215                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
2216                         end_dirent->lde_reclen =
2217                                 cpu_to_le16((char *)(dp->ldp_entries) -
2218                                             (char *)end_dirent);
2219                 }
2220
2221                 first->ldp_hash_end = hash_end;
2222                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
2223                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
2224
2225                 kunmap(pages[i]);
2226         }
2227         LASSERTF(nlupgs == 0, "left = %d", nlupgs);
2228 }
2229 #else
2230 #define lmv_adjust_dirpages(pages, ncfspgs, nlupgs) do {} while (0)
2231 #endif  /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
2232
2233 static int lmv_readpage(struct obd_export *exp, struct md_op_data *op_data,
2234                         struct page **pages, struct ptlrpc_request **request)
2235 {
2236         struct obd_device       *obd = exp->exp_obd;
2237         struct lmv_obd          *lmv = &obd->u.lmv;
2238         __u64                   offset = op_data->op_offset;
2239         int                     rc;
2240         int                     ncfspgs; /* pages read in PAGE_CACHE_SIZE */
2241         int                     nlupgs; /* pages read in LU_PAGE_SIZE */
2242         struct lmv_tgt_desc     *tgt;
2243         ENTRY;
2244
2245         rc = lmv_check_connect(obd);
2246         if (rc)
2247                 RETURN(rc);
2248
2249         CDEBUG(D_INODE, "READPAGE at "LPX64" from "DFID"\n",
2250                offset, PFID(&op_data->op_fid1));
2251
2252         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2253         if (IS_ERR(tgt))
2254                 RETURN(PTR_ERR(tgt));
2255
2256         rc = md_readpage(tgt->ltd_exp, op_data, pages, request);
2257         if (rc != 0)
2258                 RETURN(rc);
2259
2260         ncfspgs = ((*request)->rq_bulk->bd_nob_transferred +
2261                    PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
2262         nlupgs = (*request)->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
2263         LASSERT(!((*request)->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
2264         LASSERT(ncfspgs > 0 && ncfspgs <= op_data->op_npages);
2265
2266         CDEBUG(D_INODE, "read %d(%d)/%d pages\n", ncfspgs, nlupgs,
2267                op_data->op_npages);
2268
2269         lmv_adjust_dirpages(pages, ncfspgs, nlupgs);
2270
2271         RETURN(rc);
2272 }
2273
2274 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2275                       struct ptlrpc_request **request)
2276 {
2277         struct obd_device       *obd = exp->exp_obd;
2278         struct lmv_obd          *lmv = &obd->u.lmv;
2279         struct lmv_tgt_desc     *tgt = NULL;
2280         struct mdt_body         *body;
2281         int                     rc;
2282         ENTRY;
2283
2284         rc = lmv_check_connect(obd);
2285         if (rc)
2286                 RETURN(rc);
2287 retry:
2288         /* Send unlink requests to the MDT where the child is located */
2289         if (likely(!fid_is_zero(&op_data->op_fid2)))
2290                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2291         else
2292                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2293         if (IS_ERR(tgt))
2294                 RETURN(PTR_ERR(tgt));
2295
2296         op_data->op_fsuid = current_fsuid();
2297         op_data->op_fsgid = current_fsgid();
2298         op_data->op_cap = cfs_curproc_cap_pack();
2299
2300         /*
2301          * If child's fid is given, cancel unused locks for it if it is from
2302          * another export than parent.
2303          *
2304          * LOOKUP lock for child (fid3) should also be cancelled on parent
2305          * tgt_tgt in mdc_unlink().
2306          */
2307         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2308
2309         /*
2310          * Cancel FULL locks on child (fid3).
2311          */
2312         rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX,
2313                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2314
2315         if (rc != 0)
2316                 RETURN(rc);
2317
2318         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n",
2319                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2320
2321         rc = md_unlink(tgt->ltd_exp, op_data, request);
2322         if (rc != 0 && rc != -EREMOTE)
2323                 RETURN(rc);
2324
2325         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2326         if (body == NULL)
2327                 RETURN(-EPROTO);
2328
2329         /* Not cross-ref case, just get out of here. */
2330         if (likely(!(body->valid & OBD_MD_MDS)))
2331                 RETURN(0);
2332
2333         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2334                exp->exp_obd->obd_name, PFID(&body->fid1));
2335
2336         /* This is a remote object, try remote MDT, Note: it may
2337          * try more than 1 time here, Considering following case
2338          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2339          * 1. Initially A does not know where remote1 is, it send
2340          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2341          *    resend unlink RPC to MDT1 (retry 1st time).
2342          *
2343          * 2. During the unlink RPC in flight,
2344          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2345          *    and create new remote1, but on MDT0
2346          *
2347          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2348          *    /mnt/lustre, then lookup get fid of remote1, and find
2349          *    it is remote dir again, and replay -EREMOTE again.
2350          *
2351          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2352          *
2353          * In theory, it might try unlimited time here, but it should
2354          * be very rare case.  */
2355         op_data->op_fid2 = body->fid1;
2356         ptlrpc_req_finished(*request);
2357         *request = NULL;
2358
2359         goto retry;
2360 }
2361
2362 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2363 {
2364         struct lmv_obd *lmv = &obd->u.lmv;
2365         int rc = 0;
2366
2367         switch (stage) {
2368         case OBD_CLEANUP_EARLY:
2369                 /* XXX: here should be calling obd_precleanup() down to
2370                  * stack. */
2371                 break;
2372         case OBD_CLEANUP_EXPORTS:
2373                 fld_client_proc_fini(&lmv->lmv_fld);
2374                 lprocfs_obd_cleanup(obd);
2375                 lprocfs_free_md_stats(obd);
2376                 break;
2377         default:
2378                 break;
2379         }
2380         RETURN(rc);
2381 }
2382
2383 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2384                         __u32 keylen, void *key, __u32 *vallen, void *val,
2385                         struct lov_stripe_md *lsm)
2386 {
2387         struct obd_device       *obd;
2388         struct lmv_obd          *lmv;
2389         int                      rc = 0;
2390         ENTRY;
2391
2392         obd = class_exp2obd(exp);
2393         if (obd == NULL) {
2394                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2395                        exp->exp_handle.h_cookie);
2396                 RETURN(-EINVAL);
2397         }
2398
2399         lmv = &obd->u.lmv;
2400         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2401                 struct lmv_tgt_desc *tgt;
2402                 int i;
2403
2404                 rc = lmv_check_connect(obd);
2405                 if (rc)
2406                         RETURN(rc);
2407
2408                 LASSERT(*vallen == sizeof(__u32));
2409                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2410                         tgt = lmv->tgts[i];
2411                         /*
2412                          * All tgts should be connected when this gets called.
2413                          */
2414                         if (tgt == NULL || tgt->ltd_exp == NULL)
2415                                 continue;
2416
2417                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2418                                           vallen, val, NULL))
2419                                 RETURN(0);
2420                 }
2421                 RETURN(-EINVAL);
2422         } else if (KEY_IS(KEY_MAX_EASIZE) || KEY_IS(KEY_CONN_DATA)) {
2423                 rc = lmv_check_connect(obd);
2424                 if (rc)
2425                         RETURN(rc);
2426
2427                 /*
2428                  * Forwarding this request to first MDS, it should know LOV
2429                  * desc.
2430                  */
2431                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2432                                   vallen, val, NULL);
2433                 if (!rc && KEY_IS(KEY_CONN_DATA))
2434                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2435                 RETURN(rc);
2436         } else if (KEY_IS(KEY_TGT_COUNT)) {
2437                 *((int *)val) = lmv->desc.ld_tgt_count;
2438                 RETURN(0);
2439         }
2440
2441         CDEBUG(D_IOCTL, "Invalid key\n");
2442         RETURN(-EINVAL);
2443 }
2444
2445 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2446                        obd_count keylen, void *key, obd_count vallen,
2447                        void *val, struct ptlrpc_request_set *set)
2448 {
2449         struct lmv_tgt_desc    *tgt;
2450         struct obd_device      *obd;
2451         struct lmv_obd         *lmv;
2452         int rc = 0;
2453         ENTRY;
2454
2455         obd = class_exp2obd(exp);
2456         if (obd == NULL) {
2457                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2458                        exp->exp_handle.h_cookie);
2459                 RETURN(-EINVAL);
2460         }
2461         lmv = &obd->u.lmv;
2462
2463         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2464                 int i, err = 0;
2465
2466                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2467                         tgt = lmv->tgts[i];
2468
2469                         if (tgt == NULL || tgt->ltd_exp == NULL)
2470                                 continue;
2471
2472                         err = obd_set_info_async(env, tgt->ltd_exp,
2473                                                  keylen, key, vallen, val, set);
2474                         if (err && rc == 0)
2475                                 rc = err;
2476                 }
2477
2478                 RETURN(rc);
2479         }
2480
2481         RETURN(-EINVAL);
2482 }
2483
2484 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
2485                struct lov_stripe_md *lsm)
2486 {
2487         struct obd_device         *obd = class_exp2obd(exp);
2488         struct lmv_obd            *lmv = &obd->u.lmv;
2489         struct lmv_stripe_md      *meap;
2490         struct lmv_stripe_md      *lsmp;
2491         int                        mea_size;
2492         __u32                      i;
2493         ENTRY;
2494
2495         mea_size = lmv_get_easize(lmv);
2496         if (!lmmp)
2497                 RETURN(mea_size);
2498
2499         if (*lmmp && !lsm) {
2500                 OBD_FREE_LARGE(*lmmp, mea_size);
2501                 *lmmp = NULL;
2502                 RETURN(0);
2503         }
2504
2505         if (*lmmp == NULL) {
2506                 OBD_ALLOC_LARGE(*lmmp, mea_size);
2507                 if (*lmmp == NULL)
2508                         RETURN(-ENOMEM);
2509         }
2510
2511         if (!lsm)
2512                 RETURN(mea_size);
2513
2514         lsmp = (struct lmv_stripe_md *)lsm;
2515         meap = (struct lmv_stripe_md *)*lmmp;
2516
2517         if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR &&
2518             lsmp->mea_magic != MEA_MAGIC_ALL_CHARS)
2519                 RETURN(-EINVAL);
2520
2521         meap->mea_magic = cpu_to_le32(lsmp->mea_magic);
2522         meap->mea_count = cpu_to_le32(lsmp->mea_count);
2523         meap->mea_master = cpu_to_le32(lsmp->mea_master);
2524
2525         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2526                 meap->mea_ids[i] = lsmp->mea_ids[i];
2527                 fid_cpu_to_le(&meap->mea_ids[i], &lsmp->mea_ids[i]);
2528         }
2529
2530         RETURN(mea_size);
2531 }
2532
2533 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
2534                  struct lov_mds_md *lmm, int lmm_size)
2535 {
2536         struct obd_device          *obd = class_exp2obd(exp);
2537         struct lmv_stripe_md      **tmea = (struct lmv_stripe_md **)lsmp;
2538         struct lmv_stripe_md       *mea = (struct lmv_stripe_md *)lmm;
2539         struct lmv_obd             *lmv = &obd->u.lmv;
2540         int                         mea_size;
2541         __u32                       i;
2542         __u32                       magic;
2543         ENTRY;
2544
2545         mea_size = lmv_get_easize(lmv);
2546         if (lsmp == NULL)
2547                 return mea_size;
2548
2549         if (*lsmp != NULL && lmm == NULL) {
2550                 OBD_FREE_LARGE(*tmea, mea_size);
2551                 *lsmp = NULL;
2552                 RETURN(0);
2553         }
2554
2555         LASSERT(mea_size == lmm_size);
2556
2557         OBD_ALLOC_LARGE(*tmea, mea_size);
2558         if (*tmea == NULL)
2559                 RETURN(-ENOMEM);
2560
2561         if (!lmm)
2562                 RETURN(mea_size);
2563
2564         if (mea->mea_magic == MEA_MAGIC_LAST_CHAR ||
2565             mea->mea_magic == MEA_MAGIC_ALL_CHARS ||
2566             mea->mea_magic == MEA_MAGIC_HASH_SEGMENT)
2567         {
2568                 magic = le32_to_cpu(mea->mea_magic);
2569         } else {
2570                 /*
2571                  * Old mea is not handled here.
2572                  */
2573                 CERROR("Old not supportable EA is found\n");
2574                 LBUG();
2575         }
2576
2577         (*tmea)->mea_magic = magic;
2578         (*tmea)->mea_count = le32_to_cpu(mea->mea_count);
2579         (*tmea)->mea_master = le32_to_cpu(mea->mea_master);
2580
2581         for (i = 0; i < (*tmea)->mea_count; i++) {
2582                 (*tmea)->mea_ids[i] = mea->mea_ids[i];
2583                 fid_le_to_cpu(&(*tmea)->mea_ids[i], &(*tmea)->mea_ids[i]);
2584         }
2585         RETURN(mea_size);
2586 }
2587
2588 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2589                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
2590                              ldlm_cancel_flags_t flags, void *opaque)
2591 {
2592         struct obd_device       *obd = exp->exp_obd;
2593         struct lmv_obd          *lmv = &obd->u.lmv;
2594         int                      rc = 0;
2595         int                      err;
2596         __u32                    i;
2597         ENTRY;
2598
2599         LASSERT(fid != NULL);
2600
2601         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2602                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL ||
2603                     lmv->tgts[i]->ltd_active == 0)
2604                         continue;
2605
2606                 err = md_cancel_unused(lmv->tgts[i]->ltd_exp, fid,
2607                                        policy, mode, flags, opaque);
2608                 if (!rc)
2609                         rc = err;
2610         }
2611         RETURN(rc);
2612 }
2613
2614 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
2615                       __u64 *bits)
2616 {
2617         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
2618         int                      rc;
2619         ENTRY;
2620
2621         rc =  md_set_lock_data(lmv->tgts[0]->ltd_exp, lockh, data, bits);
2622         RETURN(rc);
2623 }
2624
2625 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
2626                            const struct lu_fid *fid, ldlm_type_t type,
2627                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
2628                            struct lustre_handle *lockh)
2629 {
2630         struct obd_device       *obd = exp->exp_obd;
2631         struct lmv_obd          *lmv = &obd->u.lmv;
2632         ldlm_mode_t              rc;
2633         __u32                    i;
2634         ENTRY;
2635
2636         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
2637
2638         /*
2639          * With CMD every object can have two locks in different namespaces:
2640          * lookup lock in space of mds storing direntry and update/open lock in
2641          * space of mds storing inode. Thus we check all targets, not only that
2642          * one fid was created in.
2643          */
2644         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2645                 if (lmv->tgts[i] == NULL ||
2646                     lmv->tgts[i]->ltd_exp == NULL ||
2647                     lmv->tgts[i]->ltd_active == 0)
2648                         continue;
2649
2650                 rc = md_lock_match(lmv->tgts[i]->ltd_exp, flags, fid,
2651                                    type, policy, mode, lockh);
2652                 if (rc)
2653                         RETURN(rc);
2654         }
2655
2656         RETURN(0);
2657 }
2658
2659 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
2660                       struct obd_export *dt_exp, struct obd_export *md_exp,
2661                       struct lustre_md *md)
2662 {
2663         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
2664
2665         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
2666 }
2667
2668 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
2669 {
2670         struct obd_device       *obd = exp->exp_obd;
2671         struct lmv_obd          *lmv = &obd->u.lmv;
2672         ENTRY;
2673
2674         if (md->mea)
2675                 obd_free_memmd(exp, (void *)&md->mea);
2676         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
2677 }
2678
2679 int lmv_set_open_replay_data(struct obd_export *exp,
2680                              struct obd_client_handle *och,
2681                              struct ptlrpc_request *open_req)
2682 {
2683         struct obd_device       *obd = exp->exp_obd;
2684         struct lmv_obd          *lmv = &obd->u.lmv;
2685         struct lmv_tgt_desc     *tgt;
2686         ENTRY;
2687
2688         tgt = lmv_find_target(lmv, &och->och_fid);
2689         if (IS_ERR(tgt))
2690                 RETURN(PTR_ERR(tgt));
2691
2692         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, open_req));
2693 }
2694
2695 int lmv_clear_open_replay_data(struct obd_export *exp,
2696                                struct obd_client_handle *och)
2697 {
2698         struct obd_device       *obd = exp->exp_obd;
2699         struct lmv_obd          *lmv = &obd->u.lmv;
2700         struct lmv_tgt_desc     *tgt;
2701         ENTRY;
2702
2703         tgt = lmv_find_target(lmv, &och->och_fid);
2704         if (IS_ERR(tgt))
2705                 RETURN(PTR_ERR(tgt));
2706
2707         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
2708 }
2709
2710 static int lmv_get_remote_perm(struct obd_export *exp,
2711                                const struct lu_fid *fid,
2712                                struct obd_capa *oc, __u32 suppgid,
2713                                struct ptlrpc_request **request)
2714 {
2715         struct obd_device       *obd = exp->exp_obd;
2716         struct lmv_obd          *lmv = &obd->u.lmv;
2717         struct lmv_tgt_desc     *tgt;
2718         int                      rc;
2719         ENTRY;
2720
2721         rc = lmv_check_connect(obd);
2722         if (rc)
2723                 RETURN(rc);
2724
2725         tgt = lmv_find_target(lmv, fid);
2726         if (IS_ERR(tgt))
2727                 RETURN(PTR_ERR(tgt));
2728
2729         rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
2730         RETURN(rc);
2731 }
2732
2733 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2734                           renew_capa_cb_t cb)
2735 {
2736         struct obd_device       *obd = exp->exp_obd;
2737         struct lmv_obd          *lmv = &obd->u.lmv;
2738         struct lmv_tgt_desc     *tgt;
2739         int                      rc;
2740         ENTRY;
2741
2742         rc = lmv_check_connect(obd);
2743         if (rc)
2744                 RETURN(rc);
2745
2746         tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
2747         if (IS_ERR(tgt))
2748                 RETURN(PTR_ERR(tgt));
2749
2750         rc = md_renew_capa(tgt->ltd_exp, oc, cb);
2751         RETURN(rc);
2752 }
2753
2754 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
2755                     const struct req_msg_field *field, struct obd_capa **oc)
2756 {
2757         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2758
2759         return md_unpack_capa(lmv->tgts[0]->ltd_exp, req, field, oc);
2760 }
2761
2762 int lmv_intent_getattr_async(struct obd_export *exp,
2763                              struct md_enqueue_info *minfo,
2764                              struct ldlm_enqueue_info *einfo)
2765 {
2766         struct md_op_data       *op_data = &minfo->mi_data;
2767         struct obd_device       *obd = exp->exp_obd;
2768         struct lmv_obd          *lmv = &obd->u.lmv;
2769         struct lmv_tgt_desc     *tgt = NULL;
2770         int                      rc;
2771         ENTRY;
2772
2773         rc = lmv_check_connect(obd);
2774         if (rc)
2775                 RETURN(rc);
2776
2777         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2778         if (IS_ERR(tgt))
2779                 RETURN(PTR_ERR(tgt));
2780
2781         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
2782         RETURN(rc);
2783 }
2784
2785 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
2786                         struct lu_fid *fid, __u64 *bits)
2787 {
2788         struct obd_device       *obd = exp->exp_obd;
2789         struct lmv_obd          *lmv = &obd->u.lmv;
2790         struct lmv_tgt_desc     *tgt;
2791         int                      rc;
2792         ENTRY;
2793
2794         rc = lmv_check_connect(obd);
2795         if (rc)
2796                 RETURN(rc);
2797
2798         tgt = lmv_find_target(lmv, fid);
2799         if (IS_ERR(tgt))
2800                 RETURN(PTR_ERR(tgt));
2801
2802         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
2803         RETURN(rc);
2804 }
2805
2806 /**
2807  * For lmv, only need to send request to master MDT, and the master MDT will
2808  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
2809  * we directly fetch data from the slave MDTs.
2810  */
2811 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
2812                  struct obd_quotactl *oqctl)
2813 {
2814         struct obd_device   *obd = class_exp2obd(exp);
2815         struct lmv_obd      *lmv = &obd->u.lmv;
2816         struct lmv_tgt_desc *tgt = lmv->tgts[0];
2817         int                  rc = 0;
2818         __u32                i;
2819         __u64                curspace, curinodes;
2820         ENTRY;
2821
2822         if (!lmv->desc.ld_tgt_count || !tgt->ltd_active) {
2823                 CERROR("master lmv inactive\n");
2824                 RETURN(-EIO);
2825         }
2826
2827         if (oqctl->qc_cmd != Q_GETOQUOTA) {
2828                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
2829                 RETURN(rc);
2830         }
2831
2832         curspace = curinodes = 0;
2833         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2834                 int err;
2835                 tgt = lmv->tgts[i];
2836
2837                 if (tgt == NULL || tgt->ltd_exp == NULL || tgt->ltd_active == 0)
2838                         continue;
2839                 if (!tgt->ltd_active) {
2840                         CDEBUG(D_HA, "mdt %d is inactive.\n", i);
2841                         continue;
2842                 }
2843
2844                 err = obd_quotactl(tgt->ltd_exp, oqctl);
2845                 if (err) {
2846                         CERROR("getquota on mdt %d failed. %d\n", i, err);
2847                         if (!rc)
2848                                 rc = err;
2849                 } else {
2850                         curspace += oqctl->qc_dqblk.dqb_curspace;
2851                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
2852                 }
2853         }
2854         oqctl->qc_dqblk.dqb_curspace = curspace;
2855         oqctl->qc_dqblk.dqb_curinodes = curinodes;
2856
2857         RETURN(rc);
2858 }
2859
2860 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
2861                    struct obd_quotactl *oqctl)
2862 {
2863         struct obd_device       *obd = class_exp2obd(exp);
2864         struct lmv_obd          *lmv = &obd->u.lmv;
2865         struct lmv_tgt_desc     *tgt;
2866         __u32                    i;
2867         int                      rc = 0;
2868         ENTRY;
2869
2870         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2871                 int err;
2872                 tgt = lmv->tgts[i];
2873                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
2874                         CERROR("lmv idx %d inactive\n", i);
2875                         RETURN(-EIO);
2876                 }
2877
2878                 err = obd_quotacheck(tgt->ltd_exp, oqctl);
2879                 if (err && !rc)
2880                         rc = err;
2881         }
2882
2883         RETURN(rc);
2884 }
2885
2886 struct obd_ops lmv_obd_ops = {
2887         .o_owner                = THIS_MODULE,
2888         .o_setup                = lmv_setup,
2889         .o_cleanup              = lmv_cleanup,
2890         .o_precleanup           = lmv_precleanup,
2891         .o_process_config       = lmv_process_config,
2892         .o_connect              = lmv_connect,
2893         .o_disconnect           = lmv_disconnect,
2894         .o_statfs               = lmv_statfs,
2895         .o_get_info             = lmv_get_info,
2896         .o_set_info_async       = lmv_set_info_async,
2897         .o_packmd               = lmv_packmd,
2898         .o_unpackmd             = lmv_unpackmd,
2899         .o_notify               = lmv_notify,
2900         .o_get_uuid             = lmv_get_uuid,
2901         .o_iocontrol            = lmv_iocontrol,
2902         .o_quotacheck           = lmv_quotacheck,
2903         .o_quotactl             = lmv_quotactl
2904 };
2905
2906 struct md_ops lmv_md_ops = {
2907         .m_getstatus            = lmv_getstatus,
2908         .m_null_inode           = lmv_null_inode,
2909         .m_find_cbdata          = lmv_find_cbdata,
2910         .m_close                = lmv_close,
2911         .m_create               = lmv_create,
2912         .m_done_writing         = lmv_done_writing,
2913         .m_enqueue              = lmv_enqueue,
2914         .m_getattr              = lmv_getattr,
2915         .m_getxattr             = lmv_getxattr,
2916         .m_getattr_name         = lmv_getattr_name,
2917         .m_intent_lock          = lmv_intent_lock,
2918         .m_link                 = lmv_link,
2919         .m_rename               = lmv_rename,
2920         .m_setattr              = lmv_setattr,
2921         .m_setxattr             = lmv_setxattr,
2922         .m_fsync                = lmv_fsync,
2923         .m_readpage             = lmv_readpage,
2924         .m_unlink               = lmv_unlink,
2925         .m_init_ea_size         = lmv_init_ea_size,
2926         .m_cancel_unused        = lmv_cancel_unused,
2927         .m_set_lock_data        = lmv_set_lock_data,
2928         .m_lock_match           = lmv_lock_match,
2929         .m_get_lustre_md        = lmv_get_lustre_md,
2930         .m_free_lustre_md       = lmv_free_lustre_md,
2931         .m_set_open_replay_data = lmv_set_open_replay_data,
2932         .m_clear_open_replay_data = lmv_clear_open_replay_data,
2933         .m_renew_capa           = lmv_renew_capa,
2934         .m_unpack_capa          = lmv_unpack_capa,
2935         .m_get_remote_perm      = lmv_get_remote_perm,
2936         .m_intent_getattr_async = lmv_intent_getattr_async,
2937         .m_revalidate_lock      = lmv_revalidate_lock
2938 };
2939
2940 int __init lmv_init(void)
2941 {
2942         struct lprocfs_static_vars lvars;
2943         int                        rc;
2944
2945         lprocfs_lmv_init_vars(&lvars);
2946
2947         rc = class_register_type(&lmv_obd_ops, &lmv_md_ops,
2948                                  lvars.module_vars, LUSTRE_LMV_NAME, NULL);
2949         return rc;
2950 }
2951
2952 #ifdef __KERNEL__
2953 static void lmv_exit(void)
2954 {
2955         class_unregister_type(LUSTRE_LMV_NAME);
2956 }
2957
2958 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2959 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
2960 MODULE_LICENSE("GPL");
2961
2962 module_init(lmv_init);
2963 module_exit(lmv_exit);
2964 #endif