Whamcloud - gitweb
LU-4669 lmv: remove unused lmv_adjust_dirpages()
[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 #include <linux/slab.h>
39 #include <linux/module.h>
40 #include <linux/init.h>
41 #include <linux/slab.h>
42 #include <linux/pagemap.h>
43 #include <linux/mm.h>
44 #include <linux/math64.h>
45 #include <linux/seq_file.h>
46 #include <linux/namei.h>
47
48 #include <lustre/lustre_idl.h>
49 #include <obd_support.h>
50 #include <lustre_lib.h>
51 #include <lustre_net.h>
52 #include <obd_class.h>
53 #include <lustre_lmv.h>
54 #include <lprocfs_status.h>
55 #include <cl_object.h>
56 #include <lclient.h>
57 #include <lustre_lite.h>
58 #include <lustre_fid.h>
59 #include <lustre_ioctl.h>
60 #include "lmv_internal.h"
61
62 /* This hash is only for testing purpose */
63 static inline unsigned int
64 lmv_hash_all_chars(unsigned int count, const char *name, int namelen)
65 {
66         unsigned int c = 0;
67         const unsigned char *p = (const unsigned char *)name;
68
69         while (--namelen >= 0)
70                 c += p[namelen];
71
72         c = c % count;
73
74         return c;
75 }
76
77 static inline unsigned int
78 lmv_hash_fnv1a(unsigned int count, const char *name, int namelen)
79 {
80         __u64   hash;
81
82         hash = lustre_hash_fnv_1a_64(name, namelen);
83
84         hash = hash % count;
85
86         return hash;
87 }
88
89 int lmv_name_to_stripe_index(__u32 lmv_hash_type, unsigned int stripe_count,
90                              const char *name, int namelen)
91 {
92         int     idx;
93         __u32   hash_type = lmv_hash_type & LMV_HASH_TYPE_MASK;
94
95         LASSERT(namelen > 0);
96         if (stripe_count <= 1)
97                 return 0;
98
99         /* for migrating object, always start from 0 stripe */
100         if (lmv_hash_type & LMV_HASH_FLAG_MIGRATION)
101                 return 0;
102
103         switch (hash_type) {
104         case LMV_HASH_TYPE_ALL_CHARS:
105                 idx = lmv_hash_all_chars(stripe_count, name, namelen);
106                 break;
107         case LMV_HASH_TYPE_FNV_1A_64:
108                 idx = lmv_hash_fnv1a(stripe_count, name, namelen);
109                 break;
110         default:
111                 idx = -EBADFD;
112                 break;
113         }
114
115         CDEBUG(D_INFO, "name %.*s hash_type %d idx %d\n", namelen, name,
116                hash_type, idx);
117
118         return idx;
119 }
120
121 static void lmv_activate_target(struct lmv_obd *lmv,
122                                 struct lmv_tgt_desc *tgt,
123                                 int activate)
124 {
125         if (tgt->ltd_active == activate)
126                 return;
127
128         tgt->ltd_active = activate;
129         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
130 }
131
132 /**
133  * Error codes:
134  *
135  *  -EINVAL  : UUID can't be found in the LMV's target list
136  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
137  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
138  */
139 static int lmv_set_mdc_active(struct lmv_obd *lmv,
140                               const struct obd_uuid *uuid,
141                               int activate)
142 {
143         struct lmv_tgt_desc     *tgt = NULL;
144         struct obd_device       *obd;
145         __u32                    i;
146         int                      rc = 0;
147         ENTRY;
148
149         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
150                         lmv, uuid->uuid, activate);
151
152         spin_lock(&lmv->lmv_lock);
153         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
154                 tgt = lmv->tgts[i];
155                 if (tgt == NULL || tgt->ltd_exp == NULL)
156                         continue;
157
158                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
159                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
160
161                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
162                         break;
163         }
164
165         if (i == lmv->desc.ld_tgt_count)
166                 GOTO(out_lmv_lock, rc = -EINVAL);
167
168         obd = class_exp2obd(tgt->ltd_exp);
169         if (obd == NULL)
170                 GOTO(out_lmv_lock, rc = -ENOTCONN);
171
172         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
173                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
174                obd->obd_type->typ_name, i);
175         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
176
177         if (tgt->ltd_active == activate) {
178                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
179                        activate ? "" : "in");
180                 GOTO(out_lmv_lock, rc);
181         }
182
183         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
184                activate ? "" : "in");
185         lmv_activate_target(lmv, tgt, activate);
186         EXIT;
187
188  out_lmv_lock:
189         spin_unlock(&lmv->lmv_lock);
190         return rc;
191 }
192
193 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
194 {
195         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
196         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
197
198         return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
199 }
200
201 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
202                       enum obd_notify_event ev, void *data)
203 {
204         struct obd_connect_data *conn_data;
205         struct lmv_obd          *lmv = &obd->u.lmv;
206         struct obd_uuid         *uuid;
207         int                      rc = 0;
208         ENTRY;
209
210         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
211                 CERROR("unexpected notification of %s %s!\n",
212                        watched->obd_type->typ_name,
213                        watched->obd_name);
214                 RETURN(-EINVAL);
215         }
216
217         uuid = &watched->u.cli.cl_target_uuid;
218         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
219                 /*
220                  * Set MDC as active before notifying the observer, so the
221                  * observer can use the MDC normally.
222                  */
223                 rc = lmv_set_mdc_active(lmv, uuid,
224                                         ev == OBD_NOTIFY_ACTIVE);
225                 if (rc) {
226                         CERROR("%sactivation of %s failed: %d\n",
227                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
228                                uuid->uuid, rc);
229                         RETURN(rc);
230                 }
231         } else if (ev == OBD_NOTIFY_OCD) {
232                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
233                 /*
234                  * XXX: Make sure that ocd_connect_flags from all targets are
235                  * the same. Otherwise one of MDTs runs wrong version or
236                  * something like this.  --umka
237                  */
238                 obd->obd_self_export->exp_connect_data = *conn_data;
239         }
240 #if 0
241         else if (ev == OBD_NOTIFY_DISCON) {
242                 /*
243                  * For disconnect event, flush fld cache for failout MDS case.
244                  */
245                 fld_client_flush(&lmv->lmv_fld);
246         }
247 #endif
248         /*
249          * Pass the notification up the chain.
250          */
251         if (obd->obd_observer)
252                 rc = obd_notify(obd->obd_observer, watched, ev, data);
253
254         RETURN(rc);
255 }
256
257 /**
258  * This is fake connect function. Its purpose is to initialize lmv and say
259  * caller that everything is okay. Real connection will be performed later.
260  */
261 static int lmv_connect(const struct lu_env *env,
262                        struct obd_export **exp, struct obd_device *obd,
263                        struct obd_uuid *cluuid, struct obd_connect_data *data,
264                        void *localdata)
265 {
266         struct lmv_obd        *lmv = &obd->u.lmv;
267         struct lustre_handle  conn = { 0 };
268         int                    rc = 0;
269         ENTRY;
270
271         /*
272          * We don't want to actually do the underlying connections more than
273          * once, so keep track.
274          */
275         lmv->refcount++;
276         if (lmv->refcount > 1) {
277                 *exp = NULL;
278                 RETURN(0);
279         }
280
281         rc = class_connect(&conn, obd, cluuid);
282         if (rc) {
283                 CERROR("class_connection() returned %d\n", rc);
284                 RETURN(rc);
285         }
286
287         *exp = class_conn2export(&conn);
288         class_export_get(*exp);
289
290         lmv->exp = *exp;
291         lmv->connected = 0;
292         lmv->cluuid = *cluuid;
293
294         if (data)
295                 lmv->conn_data = *data;
296
297         if (lmv->targets_proc_entry == NULL) {
298                 lmv->targets_proc_entry = lprocfs_seq_register("target_obds",
299                                                         obd->obd_proc_entry,
300                                                         NULL, NULL);
301                 if (IS_ERR(lmv->targets_proc_entry)) {
302                         CERROR("%s: cannot register "
303                                "/proc/fs/lustre/%s/%s/target_obds\n",
304                                obd->obd_name, obd->obd_type->typ_name,
305                                obd->obd_name);
306                         lmv->targets_proc_entry = NULL;
307                 }
308         }
309
310         /*
311          * All real clients should perform actual connection right away, because
312          * it is possible, that LMV will not have opportunity to connect targets
313          * and MDC stuff will be called directly, for instance while reading
314          * ../mdc/../kbytesfree procfs file, etc.
315          */
316         if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
317                 rc = lmv_check_connect(obd);
318
319         if (rc && lmv->targets_proc_entry != NULL)
320                 lprocfs_remove(&lmv->targets_proc_entry);
321         RETURN(rc);
322 }
323
324 static void lmv_set_timeouts(struct obd_device *obd)
325 {
326         struct lmv_obd          *lmv;
327         __u32                    i;
328
329         lmv = &obd->u.lmv;
330         if (lmv->server_timeout == 0)
331                 return;
332
333         if (lmv->connected == 0)
334                 return;
335
336         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
337                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
338
339                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
340                         continue;
341
342                 obd_set_info_async(NULL, tgt->ltd_exp, sizeof(KEY_INTERMDS),
343                                    KEY_INTERMDS, 0, NULL, NULL);
344         }
345 }
346
347 static int lmv_init_ea_size(struct obd_export *exp, int easize,
348                             int def_easize, int cookiesize, int def_cookiesize)
349 {
350         struct obd_device       *obd = exp->exp_obd;
351         struct lmv_obd          *lmv = &obd->u.lmv;
352         __u32                    i;
353         int                      rc = 0;
354         int                      change = 0;
355         ENTRY;
356
357         if (lmv->max_easize < easize) {
358                 lmv->max_easize = easize;
359                 change = 1;
360         }
361         if (lmv->max_def_easize < def_easize) {
362                 lmv->max_def_easize = def_easize;
363                 change = 1;
364         }
365         if (lmv->max_cookiesize < cookiesize) {
366                 lmv->max_cookiesize = cookiesize;
367                 change = 1;
368         }
369         if (lmv->max_def_cookiesize < def_cookiesize) {
370                 lmv->max_def_cookiesize = def_cookiesize;
371                 change = 1;
372         }
373         if (change == 0)
374                 RETURN(0);
375
376         if (lmv->connected == 0)
377                 RETURN(0);
378
379         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
380                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
381
382                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
383                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
384                         continue;
385                 }
386
387                 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
388                                      cookiesize, def_cookiesize);
389                 if (rc) {
390                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
391                                " rc = %d\n", obd->obd_name, i, rc);
392                         break;
393                 }
394         }
395         RETURN(rc);
396 }
397
398 #define MAX_STRING_SIZE 128
399
400 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
401 {
402         struct lmv_obd          *lmv = &obd->u.lmv;
403         struct obd_uuid         *cluuid = &lmv->cluuid;
404         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
405         struct obd_device       *mdc_obd;
406         struct obd_export       *mdc_exp;
407         struct lu_fld_target     target;
408         int                      rc;
409         ENTRY;
410
411         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
412                                         &obd->obd_uuid);
413         if (!mdc_obd) {
414                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
415                 RETURN(-EINVAL);
416         }
417
418         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
419                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
420                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
421                 cluuid->uuid);
422
423         if (!mdc_obd->obd_set_up) {
424                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
425                 RETURN(-EINVAL);
426         }
427
428         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
429                          &lmv->conn_data, NULL);
430         if (rc) {
431                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
432                 RETURN(rc);
433         }
434
435         /*
436          * Init fid sequence client for this mdc and add new fld target.
437          */
438         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
439         if (rc)
440                 RETURN(rc);
441
442         target.ft_srv = NULL;
443         target.ft_exp = mdc_exp;
444         target.ft_idx = tgt->ltd_idx;
445
446         fld_client_add_target(&lmv->lmv_fld, &target);
447
448         rc = obd_register_observer(mdc_obd, obd);
449         if (rc) {
450                 obd_disconnect(mdc_exp);
451                 CERROR("target %s register_observer error %d\n",
452                        tgt->ltd_uuid.uuid, rc);
453                 RETURN(rc);
454         }
455
456         if (obd->obd_observer) {
457                 /*
458                  * Tell the observer about the new target.
459                  */
460                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
461                                 OBD_NOTIFY_ACTIVE,
462                                 (void *)(tgt - lmv->tgts[0]));
463                 if (rc) {
464                         obd_disconnect(mdc_exp);
465                         RETURN(rc);
466                 }
467         }
468
469         tgt->ltd_active = 1;
470         tgt->ltd_exp = mdc_exp;
471         lmv->desc.ld_active_tgt_count++;
472
473         md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize,
474                         lmv->max_cookiesize, lmv->max_def_cookiesize);
475
476         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
477                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
478                 atomic_read(&obd->obd_refcount));
479
480         if (lmv->targets_proc_entry != NULL) {
481                 struct proc_dir_entry *mdc_symlink;
482
483                 LASSERT(mdc_obd->obd_type != NULL);
484                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
485                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
486                                                   lmv->targets_proc_entry,
487                                                   "../../../%s/%s",
488                                                   mdc_obd->obd_type->typ_name,
489                                                   mdc_obd->obd_name);
490                 if (mdc_symlink == NULL) {
491                         CERROR("cannot register LMV target "
492                                "/proc/fs/lustre/%s/%s/target_obds/%s\n",
493                                obd->obd_type->typ_name, obd->obd_name,
494                                mdc_obd->obd_name);
495                 }
496         }
497         RETURN(0);
498 }
499
500 static void lmv_del_target(struct lmv_obd *lmv, int index)
501 {
502         if (lmv->tgts[index] == NULL)
503                 return;
504
505         OBD_FREE_PTR(lmv->tgts[index]);
506         lmv->tgts[index] = NULL;
507         return;
508 }
509
510 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
511                            __u32 index, int gen)
512 {
513         struct lmv_obd      *lmv = &obd->u.lmv;
514         struct lmv_tgt_desc *tgt;
515         int                  orig_tgt_count = 0;
516         int                  rc = 0;
517         ENTRY;
518
519         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
520
521         lmv_init_lock(lmv);
522
523         if (lmv->desc.ld_tgt_count == 0) {
524                 struct obd_device *mdc_obd;
525
526                 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
527                                                 &obd->obd_uuid);
528                 if (!mdc_obd) {
529                         lmv_init_unlock(lmv);
530                         CERROR("%s: Target %s not attached: rc = %d\n",
531                                obd->obd_name, uuidp->uuid, -EINVAL);
532                         RETURN(-EINVAL);
533                 }
534         }
535
536         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
537                 tgt = lmv->tgts[index];
538                 CERROR("%s: UUID %s already assigned at LOV target index %d:"
539                        " rc = %d\n", obd->obd_name,
540                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
541                 lmv_init_unlock(lmv);
542                 RETURN(-EEXIST);
543         }
544
545         if (index >= lmv->tgts_size) {
546                 /* We need to reallocate the lmv target array. */
547                 struct lmv_tgt_desc **newtgts, **old = NULL;
548                 __u32 newsize = 1;
549                 __u32 oldsize = 0;
550
551                 while (newsize < index + 1)
552                         newsize = newsize << 1;
553                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
554                 if (newtgts == NULL) {
555                         lmv_init_unlock(lmv);
556                         RETURN(-ENOMEM);
557                 }
558
559                 if (lmv->tgts_size) {
560                         memcpy(newtgts, lmv->tgts,
561                                sizeof(*newtgts) * lmv->tgts_size);
562                         old = lmv->tgts;
563                         oldsize = lmv->tgts_size;
564                 }
565
566                 lmv->tgts = newtgts;
567                 lmv->tgts_size = newsize;
568                 smp_rmb();
569                 if (old)
570                         OBD_FREE(old, sizeof(*old) * oldsize);
571
572                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
573                        lmv->tgts_size);
574         }
575
576         OBD_ALLOC_PTR(tgt);
577         if (!tgt) {
578                 lmv_init_unlock(lmv);
579                 RETURN(-ENOMEM);
580         }
581
582         mutex_init(&tgt->ltd_fid_mutex);
583         tgt->ltd_idx = index;
584         tgt->ltd_uuid = *uuidp;
585         tgt->ltd_active = 0;
586         lmv->tgts[index] = tgt;
587         if (index >= lmv->desc.ld_tgt_count) {
588                 orig_tgt_count = lmv->desc.ld_tgt_count;
589                 lmv->desc.ld_tgt_count = index + 1;
590         }
591
592         if (lmv->connected) {
593                 rc = lmv_connect_mdc(obd, tgt);
594                 if (rc != 0) {
595                         spin_lock(&lmv->lmv_lock);
596                         if (lmv->desc.ld_tgt_count == index + 1)
597                                 lmv->desc.ld_tgt_count = orig_tgt_count;
598                         memset(tgt, 0, sizeof(*tgt));
599                         spin_unlock(&lmv->lmv_lock);
600                 } else {
601                         int easize = sizeof(struct lmv_stripe_md) +
602                                 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
603                         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
604                 }
605         }
606
607         lmv_init_unlock(lmv);
608         RETURN(rc);
609 }
610
611 int lmv_check_connect(struct obd_device *obd)
612 {
613         struct lmv_obd          *lmv = &obd->u.lmv;
614         struct lmv_tgt_desc     *tgt;
615         __u32                    i;
616         int                      rc;
617         int                      easize;
618         ENTRY;
619
620         if (lmv->connected)
621                 RETURN(0);
622
623         lmv_init_lock(lmv);
624         if (lmv->connected) {
625                 lmv_init_unlock(lmv);
626                 RETURN(0);
627         }
628
629         if (lmv->desc.ld_tgt_count == 0) {
630                 lmv_init_unlock(lmv);
631                 CERROR("%s: no targets configured.\n", obd->obd_name);
632                 RETURN(-EINVAL);
633         }
634
635         LASSERT(lmv->tgts != NULL);
636
637         if (lmv->tgts[0] == NULL) {
638                 lmv_init_unlock(lmv);
639                 CERROR("%s: no target configured for index 0.\n",
640                        obd->obd_name);
641                 RETURN(-EINVAL);
642         }
643
644         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
645                lmv->cluuid.uuid, obd->obd_name);
646
647         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
648                 tgt = lmv->tgts[i];
649                 if (tgt == NULL)
650                         continue;
651                 rc = lmv_connect_mdc(obd, tgt);
652                 if (rc)
653                         GOTO(out_disc, rc);
654         }
655
656         lmv_set_timeouts(obd);
657         class_export_put(lmv->exp);
658         lmv->connected = 1;
659         easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
660         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
661         lmv_init_unlock(lmv);
662         RETURN(0);
663
664  out_disc:
665         while (i-- > 0) {
666                 int rc2;
667                 tgt = lmv->tgts[i];
668                 if (tgt == NULL)
669                         continue;
670                 tgt->ltd_active = 0;
671                 if (tgt->ltd_exp) {
672                         --lmv->desc.ld_active_tgt_count;
673                         rc2 = obd_disconnect(tgt->ltd_exp);
674                         if (rc2) {
675                                 CERROR("LMV target %s disconnect on "
676                                        "MDC idx %d: error %d\n",
677                                        tgt->ltd_uuid.uuid, i, rc2);
678                         }
679                 }
680         }
681         class_disconnect(lmv->exp);
682         lmv_init_unlock(lmv);
683         RETURN(rc);
684 }
685
686 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
687 {
688         struct lmv_obd         *lmv = &obd->u.lmv;
689         struct obd_device      *mdc_obd;
690         int                     rc;
691         ENTRY;
692
693         LASSERT(tgt != NULL);
694         LASSERT(obd != NULL);
695
696         mdc_obd = class_exp2obd(tgt->ltd_exp);
697
698         if (mdc_obd) {
699                 mdc_obd->obd_force = obd->obd_force;
700                 mdc_obd->obd_fail = obd->obd_fail;
701                 mdc_obd->obd_no_recov = obd->obd_no_recov;
702         }
703
704         if (lmv->targets_proc_entry != NULL)
705                 lprocfs_remove_proc_entry(mdc_obd->obd_name,
706                                           lmv->targets_proc_entry);
707
708         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
709         if (rc)
710                 CERROR("Can't finanize fids factory\n");
711
712         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
713                tgt->ltd_exp->exp_obd->obd_name,
714                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
715
716         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
717         rc = obd_disconnect(tgt->ltd_exp);
718         if (rc) {
719                 if (tgt->ltd_active) {
720                         CERROR("Target %s disconnect error %d\n",
721                                tgt->ltd_uuid.uuid, rc);
722                 }
723         }
724
725         lmv_activate_target(lmv, tgt, 0);
726         tgt->ltd_exp = NULL;
727         RETURN(0);
728 }
729
730 static int lmv_disconnect(struct obd_export *exp)
731 {
732         struct obd_device       *obd = class_exp2obd(exp);
733         struct lmv_obd          *lmv = &obd->u.lmv;
734         int                      rc;
735         __u32                    i;
736         ENTRY;
737
738         if (!lmv->tgts)
739                 goto out_local;
740
741         /*
742          * Only disconnect the underlying layers on the final disconnect.
743          */
744         lmv->refcount--;
745         if (lmv->refcount != 0)
746                 goto out_local;
747
748         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
749                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
750                         continue;
751
752                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
753         }
754
755         if (lmv->targets_proc_entry != NULL)
756                 lprocfs_remove(&lmv->targets_proc_entry);
757         else
758                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
759                        obd->obd_type->typ_name, obd->obd_name);
760
761 out_local:
762         /*
763          * This is the case when no real connection is established by
764          * lmv_check_connect().
765          */
766         if (!lmv->connected)
767                 class_export_put(exp);
768         rc = class_disconnect(exp);
769         if (lmv->refcount == 0)
770                 lmv->connected = 0;
771         RETURN(rc);
772 }
773
774 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
775 {
776         struct obd_device       *obddev = class_exp2obd(exp);
777         struct lmv_obd          *lmv = &obddev->u.lmv;
778         struct getinfo_fid2path *gf;
779         struct lmv_tgt_desc     *tgt;
780         struct getinfo_fid2path *remote_gf = NULL;
781         int                     remote_gf_size = 0;
782         int                     rc;
783
784         gf = (struct getinfo_fid2path *)karg;
785         tgt = lmv_find_target(lmv, &gf->gf_fid);
786         if (IS_ERR(tgt))
787                 RETURN(PTR_ERR(tgt));
788
789 repeat_fid2path:
790         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
791         if (rc != 0 && rc != -EREMOTE)
792                 GOTO(out_fid2path, rc);
793
794         /* If remote_gf != NULL, it means just building the
795          * path on the remote MDT, copy this path segement to gf */
796         if (remote_gf != NULL) {
797                 struct getinfo_fid2path *ori_gf;
798                 char *ptr;
799
800                 ori_gf = (struct getinfo_fid2path *)karg;
801                 if (strlen(ori_gf->gf_path) +
802                     strlen(gf->gf_path) > ori_gf->gf_pathlen)
803                         GOTO(out_fid2path, rc = -EOVERFLOW);
804
805                 ptr = ori_gf->gf_path;
806
807                 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
808                         strlen(ori_gf->gf_path));
809
810                 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
811                 ptr += strlen(gf->gf_path);
812                 *ptr = '/';
813         }
814
815         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
816                tgt->ltd_exp->exp_obd->obd_name,
817                gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
818                gf->gf_linkno);
819
820         if (rc == 0)
821                 GOTO(out_fid2path, rc);
822
823         /* sigh, has to go to another MDT to do path building further */
824         if (remote_gf == NULL) {
825                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
826                 OBD_ALLOC(remote_gf, remote_gf_size);
827                 if (remote_gf == NULL)
828                         GOTO(out_fid2path, rc = -ENOMEM);
829                 remote_gf->gf_pathlen = PATH_MAX;
830         }
831
832         if (!fid_is_sane(&gf->gf_fid)) {
833                 CERROR("%s: invalid FID "DFID": rc = %d\n",
834                        tgt->ltd_exp->exp_obd->obd_name,
835                        PFID(&gf->gf_fid), -EINVAL);
836                 GOTO(out_fid2path, rc = -EINVAL);
837         }
838
839         tgt = lmv_find_target(lmv, &gf->gf_fid);
840         if (IS_ERR(tgt))
841                 GOTO(out_fid2path, rc = -EINVAL);
842
843         remote_gf->gf_fid = gf->gf_fid;
844         remote_gf->gf_recno = -1;
845         remote_gf->gf_linkno = -1;
846         memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
847         gf = remote_gf;
848         goto repeat_fid2path;
849
850 out_fid2path:
851         if (remote_gf != NULL)
852                 OBD_FREE(remote_gf, remote_gf_size);
853         RETURN(rc);
854 }
855
856 static int lmv_hsm_req_count(struct lmv_obd *lmv,
857                              const struct hsm_user_request *hur,
858                              const struct lmv_tgt_desc *tgt_mds)
859 {
860         __u32                    i;
861         int                      nr = 0;
862         struct lmv_tgt_desc     *curr_tgt;
863
864         /* count how many requests must be sent to the given target */
865         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
866                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
867                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
868                         nr++;
869         }
870         return nr;
871 }
872
873 static void lmv_hsm_req_build(struct lmv_obd *lmv,
874                               struct hsm_user_request *hur_in,
875                               const struct lmv_tgt_desc *tgt_mds,
876                               struct hsm_user_request *hur_out)
877 {
878         __u32                    i, nr_out;
879         struct lmv_tgt_desc     *curr_tgt;
880
881         /* build the hsm_user_request for the given target */
882         hur_out->hur_request = hur_in->hur_request;
883         nr_out = 0;
884         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
885                 curr_tgt = lmv_find_target(lmv,
886                                            &hur_in->hur_user_item[i].hui_fid);
887                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
888                         hur_out->hur_user_item[nr_out] =
889                                                 hur_in->hur_user_item[i];
890                         nr_out++;
891                 }
892         }
893         hur_out->hur_request.hr_itemcount = nr_out;
894         memcpy(hur_data(hur_out), hur_data(hur_in),
895                hur_in->hur_request.hr_data_len);
896 }
897
898 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
899                                  struct lustre_kernelcomm *lk, void *uarg)
900 {
901         __u32                    i;
902         int                      rc;
903         struct kkuc_ct_data     *kcd = NULL;
904         ENTRY;
905
906         /* unregister request (call from llapi_hsm_copytool_fini) */
907         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
908                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
909
910                 if (tgt == NULL || tgt->ltd_exp == NULL)
911                         continue;
912                 /* best effort: try to clean as much as possible
913                  * (continue on error) */
914                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
915         }
916
917         /* Whatever the result, remove copytool from kuc groups.
918          * Unreached coordinators will get EPIPE on next requests
919          * and will unregister automatically.
920          */
921         rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
922         if (kcd != NULL)
923                 OBD_FREE_PTR(kcd);
924
925         RETURN(rc);
926 }
927
928 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
929                                struct lustre_kernelcomm *lk, void *uarg)
930 {
931         struct file             *filp;
932         __u32                    i, j;
933         int                      err, rc;
934         bool                     any_set = false;
935         struct kkuc_ct_data     *kcd;
936         ENTRY;
937
938         /* All or nothing: try to register to all MDS.
939          * In case of failure, unregister from previous MDS,
940          * except if it because of inactive target. */
941         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
942                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
943
944                 if (tgt == NULL || tgt->ltd_exp == NULL)
945                         continue;
946                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
947                 if (err) {
948                         if (tgt->ltd_active) {
949                                 /* permanent error */
950                                 CERROR("%s: iocontrol MDC %s on MDT"
951                                        " idx %d cmd %x: err = %d\n",
952                                        class_exp2obd(lmv->exp)->obd_name,
953                                        tgt->ltd_uuid.uuid, i, cmd, err);
954                                 rc = err;
955                                 lk->lk_flags |= LK_FLG_STOP;
956                                 /* unregister from previous MDS */
957                                 for (j = 0; j < i; j++) {
958                                         tgt = lmv->tgts[j];
959                                         if (tgt == NULL || tgt->ltd_exp == NULL)
960                                                 continue;
961                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
962                                                       lk, uarg);
963                                 }
964                                 RETURN(rc);
965                         }
966                         /* else: transient error.
967                          * kuc will register to the missing MDT
968                          * when it is back */
969                 } else {
970                         any_set = true;
971                 }
972         }
973
974         if (!any_set)
975                 /* no registration done: return error */
976                 RETURN(-ENOTCONN);
977
978         /* at least one registration done, with no failure */
979         filp = fget(lk->lk_wfd);
980         if (filp == NULL)
981                 RETURN(-EBADF);
982
983         OBD_ALLOC_PTR(kcd);
984         if (kcd == NULL) {
985                 fput(filp);
986                 RETURN(-ENOMEM);
987         }
988         kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
989         kcd->kcd_uuid = lmv->cluuid;
990         kcd->kcd_archive = lk->lk_data;
991
992         rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
993         if (rc != 0) {
994                 if (filp != NULL)
995                         fput(filp);
996                 OBD_FREE_PTR(kcd);
997         }
998
999         RETURN(rc);
1000 }
1001
1002
1003
1004
1005 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
1006                          int len, void *karg, void *uarg)
1007 {
1008         struct obd_device       *obddev = class_exp2obd(exp);
1009         struct lmv_obd          *lmv = &obddev->u.lmv;
1010         struct lmv_tgt_desc     *tgt = NULL;
1011         __u32                    i = 0;
1012         int                      rc = 0;
1013         int                      set = 0;
1014         __u32                    count = lmv->desc.ld_tgt_count;
1015         ENTRY;
1016
1017         if (count == 0)
1018                 RETURN(-ENOTTY);
1019
1020         switch (cmd) {
1021         case IOC_OBD_STATFS: {
1022                 struct obd_ioctl_data *data = karg;
1023                 struct obd_device *mdc_obd;
1024                 struct obd_statfs stat_buf = {0};
1025                 __u32 index;
1026
1027                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1028                 if ((index >= count))
1029                         RETURN(-ENODEV);
1030
1031                 tgt = lmv->tgts[index];
1032                 if (tgt == NULL || !tgt->ltd_active)
1033                         RETURN(-ENODATA);
1034
1035                 mdc_obd = class_exp2obd(tgt->ltd_exp);
1036                 if (!mdc_obd)
1037                         RETURN(-EINVAL);
1038
1039                 /* copy UUID */
1040                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
1041                                  min((int) data->ioc_plen2,
1042                                      (int) sizeof(struct obd_uuid))))
1043                         RETURN(-EFAULT);
1044
1045                 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
1046                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1047                                 0);
1048                 if (rc)
1049                         RETURN(rc);
1050                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1051                                  min((int) data->ioc_plen1,
1052                                      (int) sizeof(stat_buf))))
1053                         RETURN(-EFAULT);
1054                 break;
1055         }
1056         case OBD_IOC_QUOTACTL: {
1057                 struct if_quotactl *qctl = karg;
1058                 struct obd_quotactl *oqctl;
1059
1060                 if (qctl->qc_valid == QC_MDTIDX) {
1061                         if (count <= qctl->qc_idx)
1062                                 RETURN(-EINVAL);
1063
1064                         tgt = lmv->tgts[qctl->qc_idx];
1065                         if (tgt == NULL || tgt->ltd_exp == NULL)
1066                                 RETURN(-EINVAL);
1067                 } else if (qctl->qc_valid == QC_UUID) {
1068                         for (i = 0; i < count; i++) {
1069                                 tgt = lmv->tgts[i];
1070                                 if (tgt == NULL)
1071                                         continue;
1072                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
1073                                                      &qctl->obd_uuid))
1074                                         continue;
1075
1076                                 if (tgt->ltd_exp == NULL)
1077                                         RETURN(-EINVAL);
1078
1079                                 break;
1080                         }
1081                 } else {
1082                         RETURN(-EINVAL);
1083                 }
1084
1085                 if (i >= count)
1086                         RETURN(-EAGAIN);
1087
1088                 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1089                 OBD_ALLOC_PTR(oqctl);
1090                 if (!oqctl)
1091                         RETURN(-ENOMEM);
1092
1093                 QCTL_COPY(oqctl, qctl);
1094                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1095                 if (rc == 0) {
1096                         QCTL_COPY(qctl, oqctl);
1097                         qctl->qc_valid = QC_MDTIDX;
1098                         qctl->obd_uuid = tgt->ltd_uuid;
1099                 }
1100                 OBD_FREE_PTR(oqctl);
1101                 break;
1102         }
1103         case OBD_IOC_CHANGELOG_SEND:
1104         case OBD_IOC_CHANGELOG_CLEAR: {
1105                 struct ioc_changelog *icc = karg;
1106
1107                 if (icc->icc_mdtindex >= count)
1108                         RETURN(-ENODEV);
1109
1110                 tgt = lmv->tgts[icc->icc_mdtindex];
1111                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1112                         RETURN(-ENODEV);
1113                 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1114                 break;
1115         }
1116         case LL_IOC_GET_CONNECT_FLAGS: {
1117                 tgt = lmv->tgts[0];
1118                 if (tgt == NULL || tgt->ltd_exp == NULL)
1119                         RETURN(-ENODATA);
1120                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1121                 break;
1122         }
1123         case OBD_IOC_FID2PATH: {
1124                 rc = lmv_fid2path(exp, len, karg, uarg);
1125                 break;
1126         }
1127         case LL_IOC_HSM_STATE_GET:
1128         case LL_IOC_HSM_STATE_SET:
1129         case LL_IOC_HSM_ACTION: {
1130                 struct md_op_data       *op_data = karg;
1131
1132                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1133                 if (IS_ERR(tgt))
1134                         RETURN(PTR_ERR(tgt));
1135
1136                 if (tgt->ltd_exp == NULL)
1137                         RETURN(-EINVAL);
1138
1139                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1140                 break;
1141         }
1142         case LL_IOC_HSM_PROGRESS: {
1143                 const struct hsm_progress_kernel *hpk = karg;
1144
1145                 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1146                 if (IS_ERR(tgt))
1147                         RETURN(PTR_ERR(tgt));
1148                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1149                 break;
1150         }
1151         case LL_IOC_HSM_REQUEST: {
1152                 struct hsm_user_request *hur = karg;
1153                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1154
1155                 if (reqcount == 0)
1156                         RETURN(0);
1157
1158                 /* if the request is about a single fid
1159                  * or if there is a single MDS, no need to split
1160                  * the request. */
1161                 if (reqcount == 1 || count == 1) {
1162                         tgt = lmv_find_target(lmv,
1163                                               &hur->hur_user_item[0].hui_fid);
1164                         if (IS_ERR(tgt))
1165                                 RETURN(PTR_ERR(tgt));
1166                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1167                 } else {
1168                         /* split fid list to their respective MDS */
1169                         for (i = 0; i < count; i++) {
1170                                 unsigned int            nr, reqlen;
1171                                 int                     rc1;
1172                                 struct hsm_user_request *req;
1173
1174                                 tgt = lmv->tgts[i];
1175                                 if (tgt == NULL || tgt->ltd_exp == NULL)
1176                                         continue;
1177
1178                                 nr = lmv_hsm_req_count(lmv, hur, tgt);
1179                                 if (nr == 0) /* nothing for this MDS */
1180                                         continue;
1181
1182                                 /* build a request with fids for this MDS */
1183                                 reqlen = offsetof(typeof(*hur),
1184                                                   hur_user_item[nr])
1185                                                 + hur->hur_request.hr_data_len;
1186                                 OBD_ALLOC_LARGE(req, reqlen);
1187                                 if (req == NULL)
1188                                         RETURN(-ENOMEM);
1189
1190                                 lmv_hsm_req_build(lmv, hur, tgt, req);
1191
1192                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1193                                                     req, uarg);
1194                                 if (rc1 != 0 && rc == 0)
1195                                         rc = rc1;
1196                                 OBD_FREE_LARGE(req, reqlen);
1197                         }
1198                 }
1199                 break;
1200         }
1201         case LL_IOC_LOV_SWAP_LAYOUTS: {
1202                 struct md_op_data       *op_data = karg;
1203                 struct lmv_tgt_desc     *tgt1, *tgt2;
1204
1205                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1206                 if (IS_ERR(tgt1))
1207                         RETURN(PTR_ERR(tgt1));
1208
1209                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1210                 if (IS_ERR(tgt2))
1211                         RETURN(PTR_ERR(tgt2));
1212
1213                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1214                         RETURN(-EINVAL);
1215
1216                 /* only files on same MDT can have their layouts swapped */
1217                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1218                         RETURN(-EPERM);
1219
1220                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1221                 break;
1222         }
1223         case LL_IOC_HSM_CT_START: {
1224                 struct lustre_kernelcomm *lk = karg;
1225                 if (lk->lk_flags & LK_FLG_STOP)
1226                         rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1227                 else
1228                         rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1229                 break;
1230         }
1231         default:
1232                 for (i = 0; i < count; i++) {
1233                         struct obd_device *mdc_obd;
1234                         int err;
1235
1236                         tgt = lmv->tgts[i];
1237                         if (tgt == NULL || tgt->ltd_exp == NULL)
1238                                 continue;
1239                         /* ll_umount_begin() sets force flag but for lmv, not
1240                          * mdc. Let's pass it through */
1241                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1242                         mdc_obd->obd_force = obddev->obd_force;
1243                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1244                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1245                                 RETURN(err);
1246                         } else if (err) {
1247                                 if (tgt->ltd_active) {
1248                                         CERROR("error: iocontrol MDC %s on MDT"
1249                                                " idx %d cmd %x: err = %d\n",
1250                                                tgt->ltd_uuid.uuid, i, cmd, err);
1251                                         if (!rc)
1252                                                 rc = err;
1253                                 }
1254                         } else
1255                                 set = 1;
1256                 }
1257                 if (!set && !rc)
1258                         rc = -EIO;
1259         }
1260         RETURN(rc);
1261 }
1262
1263 #if 0
1264 static int lmv_all_chars_policy(int count, const char *name,
1265                                 int len)
1266 {
1267         unsigned int c = 0;
1268
1269         while (len > 0)
1270                 c += name[--len];
1271         c = c % count;
1272         return c;
1273 }
1274
1275 static int lmv_nid_policy(struct lmv_obd *lmv)
1276 {
1277         struct obd_import *imp;
1278         __u32              id;
1279
1280         /*
1281          * XXX: To get nid we assume that underlying obd device is mdc.
1282          */
1283         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1284         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1285         return id % lmv->desc.ld_tgt_count;
1286 }
1287
1288 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1289                           placement_policy_t placement)
1290 {
1291         switch (placement) {
1292         case PLACEMENT_CHAR_POLICY:
1293                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1294                                             op_data->op_name,
1295                                             op_data->op_namelen);
1296         case PLACEMENT_NID_POLICY:
1297                 return lmv_nid_policy(lmv);
1298
1299         default:
1300                 break;
1301         }
1302
1303         CERROR("Unsupported placement policy %x\n", placement);
1304         return -EINVAL;
1305 }
1306 #endif
1307
1308 /**
1309  * This is _inode_ placement policy function (not name).
1310  */
1311 static int lmv_placement_policy(struct obd_device *obd,
1312                                 struct md_op_data *op_data,
1313                                 mdsno_t *mds)
1314 {
1315         struct lmv_obd          *lmv = &obd->u.lmv;
1316         ENTRY;
1317
1318         LASSERT(mds != NULL);
1319
1320         if (lmv->desc.ld_tgt_count == 1) {
1321                 *mds = 0;
1322                 RETURN(0);
1323         }
1324
1325         /**
1326          * If stripe_offset is provided during setdirstripe
1327          * (setdirstripe -i xx), xx MDS will be choosen.
1328          */
1329         if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1330                 struct lmv_user_md *lum;
1331
1332                 lum = op_data->op_data;
1333
1334                 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1335                         *mds = le32_to_cpu(lum->lum_stripe_offset);
1336                 } else {
1337                         /* -1 means default, which will be in the same MDT with
1338                          * the stripe */
1339                         *mds = op_data->op_mds;
1340                         lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1341                 }
1342         } else {
1343                 /* Allocate new fid on target according to operation type and
1344                  * parent home mds. */
1345                 *mds = op_data->op_mds;
1346         }
1347
1348         RETURN(0);
1349 }
1350
1351 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid,
1352                     mdsno_t mds)
1353 {
1354         struct lmv_tgt_desc     *tgt;
1355         int                      rc;
1356         ENTRY;
1357
1358         tgt = lmv_get_target(lmv, mds, NULL);
1359         if (IS_ERR(tgt))
1360                 RETURN(PTR_ERR(tgt));
1361
1362         /*
1363          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1364          * on server that seq in new allocated fid is not yet known.
1365          */
1366         mutex_lock(&tgt->ltd_fid_mutex);
1367
1368         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1369                 GOTO(out, rc = -ENODEV);
1370
1371         /*
1372          * Asking underlying tgt layer to allocate new fid.
1373          */
1374         rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1375         if (rc > 0) {
1376                 LASSERT(fid_is_sane(fid));
1377                 rc = 0;
1378         }
1379
1380         EXIT;
1381 out:
1382         mutex_unlock(&tgt->ltd_fid_mutex);
1383         return rc;
1384 }
1385
1386 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1387                   struct lu_fid *fid, struct md_op_data *op_data)
1388 {
1389         struct obd_device     *obd = class_exp2obd(exp);
1390         struct lmv_obd        *lmv = &obd->u.lmv;
1391         mdsno_t                mds = 0;
1392         int                    rc;
1393         ENTRY;
1394
1395         LASSERT(op_data != NULL);
1396         LASSERT(fid != NULL);
1397
1398         rc = lmv_placement_policy(obd, op_data, &mds);
1399         if (rc) {
1400                 CERROR("Can't get target for allocating fid, "
1401                        "rc %d\n", rc);
1402                 RETURN(rc);
1403         }
1404
1405         rc = __lmv_fid_alloc(lmv, fid, mds);
1406         if (rc) {
1407                 CERROR("Can't alloc new fid, rc %d\n", rc);
1408                 RETURN(rc);
1409         }
1410
1411         RETURN(rc);
1412 }
1413
1414 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1415 {
1416         struct lmv_obd  *lmv = &obd->u.lmv;
1417         struct lmv_desc *desc;
1418         int             rc;
1419         ENTRY;
1420
1421         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1422                 CERROR("LMV setup requires a descriptor\n");
1423                 RETURN(-EINVAL);
1424         }
1425
1426         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1427         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1428                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1429                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1430                 RETURN(-EINVAL);
1431         }
1432
1433         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * 32);
1434         if (lmv->tgts == NULL)
1435                 RETURN(-ENOMEM);
1436         lmv->tgts_size = 32;
1437
1438         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1439         lmv->desc.ld_tgt_count = 0;
1440         lmv->desc.ld_active_tgt_count = 0;
1441         lmv->max_cookiesize = 0;
1442         lmv->max_def_easize = 0;
1443         lmv->max_easize = 0;
1444         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1445
1446         spin_lock_init(&lmv->lmv_lock);
1447         mutex_init(&lmv->init_mutex);
1448
1449 #ifdef LPROCFS
1450         obd->obd_vars = lprocfs_lmv_obd_vars;
1451         lprocfs_seq_obd_setup(obd);
1452         lprocfs_alloc_md_stats(obd, 0);
1453         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1454                                 0444, &lmv_proc_target_fops, obd);
1455         if (rc)
1456                 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1457                       obd->obd_name, rc);
1458 #endif
1459         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1460                              LUSTRE_CLI_FLD_HASH_DHT);
1461         if (rc) {
1462                 CERROR("Can't init FLD, err %d\n", rc);
1463                 GOTO(out, rc);
1464         }
1465
1466         RETURN(0);
1467
1468 out:
1469         return rc;
1470 }
1471
1472 static int lmv_cleanup(struct obd_device *obd)
1473 {
1474         struct lmv_obd   *lmv = &obd->u.lmv;
1475         ENTRY;
1476
1477         fld_client_fini(&lmv->lmv_fld);
1478         if (lmv->tgts != NULL) {
1479                 int i;
1480                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1481                         if (lmv->tgts[i] == NULL)
1482                                 continue;
1483                         lmv_del_target(lmv, i);
1484                 }
1485                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1486                 lmv->tgts_size = 0;
1487         }
1488         RETURN(0);
1489 }
1490
1491 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1492 {
1493         struct lustre_cfg       *lcfg = buf;
1494         struct obd_uuid         obd_uuid;
1495         int                     gen;
1496         __u32                   index;
1497         int                     rc;
1498         ENTRY;
1499
1500         switch (lcfg->lcfg_command) {
1501         case LCFG_ADD_MDC:
1502                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1503                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1504                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1505                         GOTO(out, rc = -EINVAL);
1506
1507                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1508
1509                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1510                         GOTO(out, rc = -EINVAL);
1511                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1512                         GOTO(out, rc = -EINVAL);
1513                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1514                 GOTO(out, rc);
1515         default:
1516                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1517                 GOTO(out, rc = -EINVAL);
1518         }
1519 out:
1520         RETURN(rc);
1521 }
1522
1523 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1524                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1525 {
1526         struct obd_device       *obd = class_exp2obd(exp);
1527         struct lmv_obd          *lmv = &obd->u.lmv;
1528         struct obd_statfs       *temp;
1529         int                      rc = 0;
1530         __u32                    i;
1531         ENTRY;
1532
1533         rc = lmv_check_connect(obd);
1534         if (rc)
1535                 RETURN(rc);
1536
1537         OBD_ALLOC(temp, sizeof(*temp));
1538         if (temp == NULL)
1539                 RETURN(-ENOMEM);
1540
1541         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1542                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1543                         continue;
1544
1545                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1546                                 max_age, flags);
1547                 if (rc) {
1548                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1549                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1550                                rc);
1551                         GOTO(out_free_temp, rc);
1552                 }
1553
1554                 if (i == 0) {
1555                         *osfs = *temp;
1556                         /* If the statfs is from mount, it will needs
1557                          * retrieve necessary information from MDT0.
1558                          * i.e. mount does not need the merged osfs
1559                          * from all of MDT.
1560                          * And also clients can be mounted as long as
1561                          * MDT0 is in service*/
1562                         if (flags & OBD_STATFS_FOR_MDT0)
1563                                 GOTO(out_free_temp, rc);
1564                 } else {
1565                         osfs->os_bavail += temp->os_bavail;
1566                         osfs->os_blocks += temp->os_blocks;
1567                         osfs->os_ffree += temp->os_ffree;
1568                         osfs->os_files += temp->os_files;
1569                 }
1570         }
1571
1572         EXIT;
1573 out_free_temp:
1574         OBD_FREE(temp, sizeof(*temp));
1575         return rc;
1576 }
1577
1578 static int lmv_getstatus(struct obd_export *exp,
1579                          struct lu_fid *fid,
1580                          struct obd_capa **pc)
1581 {
1582         struct obd_device    *obd = exp->exp_obd;
1583         struct lmv_obd       *lmv = &obd->u.lmv;
1584         int                   rc;
1585         ENTRY;
1586
1587         rc = lmv_check_connect(obd);
1588         if (rc)
1589                 RETURN(rc);
1590
1591         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1592         RETURN(rc);
1593 }
1594
1595 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1596                         struct obd_capa *oc, obd_valid valid, const char *name,
1597                         const char *input, int input_size, int output_size,
1598                         int flags, struct ptlrpc_request **request)
1599 {
1600         struct obd_device      *obd = exp->exp_obd;
1601         struct lmv_obd         *lmv = &obd->u.lmv;
1602         struct lmv_tgt_desc    *tgt;
1603         int                     rc;
1604         ENTRY;
1605
1606         rc = lmv_check_connect(obd);
1607         if (rc)
1608                 RETURN(rc);
1609
1610         tgt = lmv_find_target(lmv, fid);
1611         if (IS_ERR(tgt))
1612                 RETURN(PTR_ERR(tgt));
1613
1614         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1615                          input_size, output_size, flags, request);
1616
1617         RETURN(rc);
1618 }
1619
1620 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1621                         struct obd_capa *oc, obd_valid valid, const char *name,
1622                         const char *input, int input_size, int output_size,
1623                         int flags, __u32 suppgid,
1624                         struct ptlrpc_request **request)
1625 {
1626         struct obd_device      *obd = exp->exp_obd;
1627         struct lmv_obd         *lmv = &obd->u.lmv;
1628         struct lmv_tgt_desc    *tgt;
1629         int                     rc;
1630         ENTRY;
1631
1632         rc = lmv_check_connect(obd);
1633         if (rc)
1634                 RETURN(rc);
1635
1636         tgt = lmv_find_target(lmv, fid);
1637         if (IS_ERR(tgt))
1638                 RETURN(PTR_ERR(tgt));
1639
1640         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1641                          input_size, output_size, flags, suppgid,
1642                          request);
1643
1644         RETURN(rc);
1645 }
1646
1647 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1648                        struct ptlrpc_request **request)
1649 {
1650         struct obd_device       *obd = exp->exp_obd;
1651         struct lmv_obd          *lmv = &obd->u.lmv;
1652         struct lmv_tgt_desc     *tgt;
1653         int                      rc;
1654         ENTRY;
1655
1656         rc = lmv_check_connect(obd);
1657         if (rc)
1658                 RETURN(rc);
1659
1660         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1661         if (IS_ERR(tgt))
1662                 RETURN(PTR_ERR(tgt));
1663
1664         if (op_data->op_flags & MF_GET_MDT_IDX) {
1665                 op_data->op_mds = tgt->ltd_idx;
1666                 RETURN(0);
1667         }
1668
1669         rc = md_getattr(tgt->ltd_exp, op_data, request);
1670
1671         RETURN(rc);
1672 }
1673
1674 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1675 {
1676         struct obd_device   *obd = exp->exp_obd;
1677         struct lmv_obd      *lmv = &obd->u.lmv;
1678         __u32                i;
1679         int                  rc;
1680         ENTRY;
1681
1682         rc = lmv_check_connect(obd);
1683         if (rc)
1684                 RETURN(rc);
1685
1686         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1687
1688         /*
1689          * With DNE every object can have two locks in different namespaces:
1690          * lookup lock in space of MDT storing direntry and update/open lock in
1691          * space of MDT storing inode.
1692          */
1693         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1694                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1695                         continue;
1696                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1697         }
1698
1699         RETURN(0);
1700 }
1701
1702 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1703                            ldlm_iterator_t it, void *data)
1704 {
1705         struct obd_device       *obd = exp->exp_obd;
1706         struct lmv_obd          *lmv = &obd->u.lmv;
1707         int                     i;
1708         int                     tgt;
1709         int                     rc;
1710         ENTRY;
1711
1712         rc = lmv_check_connect(obd);
1713         if (rc)
1714                 RETURN(rc);
1715
1716         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1717
1718         /*
1719          * With DNE every object can have two locks in different namespaces:
1720          * lookup lock in space of MDT storing direntry and update/open lock in
1721          * space of MDT storing inode.  Try the MDT that the FID maps to first,
1722          * since this can be easily found, and only try others if that fails.
1723          */
1724         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1725              i < lmv->desc.ld_tgt_count;
1726              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1727                 if (tgt < 0) {
1728                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1729                                obd->obd_name, PFID(fid), tgt);
1730                         tgt = 0;
1731                 }
1732
1733                 if (lmv->tgts[tgt] == NULL ||
1734                     lmv->tgts[tgt]->ltd_exp == NULL)
1735                         continue;
1736
1737                 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1738                 if (rc)
1739                         RETURN(rc);
1740         }
1741
1742         RETURN(rc);
1743 }
1744
1745
1746 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1747                      struct md_open_data *mod, struct ptlrpc_request **request)
1748 {
1749         struct obd_device     *obd = exp->exp_obd;
1750         struct lmv_obd        *lmv = &obd->u.lmv;
1751         struct lmv_tgt_desc   *tgt;
1752         int                    rc;
1753         ENTRY;
1754
1755         rc = lmv_check_connect(obd);
1756         if (rc)
1757                 RETURN(rc);
1758
1759         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1760         if (IS_ERR(tgt))
1761                 RETURN(PTR_ERR(tgt));
1762
1763         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1764         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1765         RETURN(rc);
1766 }
1767
1768 /**
1769  * Choosing the MDT by name or FID in @op_data.
1770  * For non-striped directory, it will locate MDT by fid.
1771  * For striped-directory, it will locate MDT by name. And also
1772  * it will reset op_fid1 with the FID of the choosen stripe.
1773  **/
1774 struct lmv_tgt_desc *
1775 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1776                            const char *name, int namelen, struct lu_fid *fid,
1777                            mdsno_t *mds)
1778 {
1779         struct lmv_tgt_desc     *tgt;
1780         const struct lmv_oinfo  *oinfo;
1781
1782         oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1783         if (IS_ERR(oinfo))
1784                 RETURN(ERR_CAST(oinfo));
1785         *fid = oinfo->lmo_fid;
1786         *mds = oinfo->lmo_mds;
1787         tgt = lmv_get_target(lmv, *mds, NULL);
1788
1789         CDEBUG(D_INFO, "locate on mds %u "DFID"\n", *mds, PFID(fid));
1790         return tgt;
1791 }
1792
1793 /**
1794  * Locate mds by fid or name
1795  *
1796  * For striped directory (lsm != NULL), it will locate the stripe
1797  * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1798  * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1799  * walk through all of stripes to locate the entry.
1800  *
1801  * For normal direcotry, it will locate MDS by FID directly.
1802  * \param[in] lmv       LMV device
1803  * \param[in] op_data   client MD stack parameters, name, namelen
1804  *                      mds_num etc.
1805  * \param[in] fid       object FID used to locate MDS.
1806  *
1807  * retval               pointer to the lmv_tgt_desc if succeed.
1808  *                      ERR_PTR(errno) if failed.
1809  */
1810 struct lmv_tgt_desc
1811 *lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1812                 struct lu_fid *fid)
1813 {
1814         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1815         struct lmv_tgt_desc     *tgt;
1816
1817         if (lsm == NULL || op_data->op_namelen == 0) {
1818                 tgt = lmv_find_target(lmv, fid);
1819                 if (IS_ERR(tgt))
1820                         return tgt;
1821
1822                 op_data->op_mds = tgt->ltd_idx;
1823                 return tgt;
1824         }
1825
1826         return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1827                                           op_data->op_namelen, fid,
1828                                           &op_data->op_mds);
1829 }
1830
1831 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1832                const void *data, int datalen, int mode, __u32 uid,
1833                __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1834                struct ptlrpc_request **request)
1835 {
1836         struct obd_device       *obd = exp->exp_obd;
1837         struct lmv_obd          *lmv = &obd->u.lmv;
1838         struct lmv_tgt_desc     *tgt;
1839         int                      rc;
1840         ENTRY;
1841
1842         rc = lmv_check_connect(obd);
1843         if (rc)
1844                 RETURN(rc);
1845
1846         if (!lmv->desc.ld_active_tgt_count)
1847                 RETURN(-EIO);
1848
1849         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1850         if (IS_ERR(tgt))
1851                 RETURN(PTR_ERR(tgt));
1852
1853         CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1854                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1855                op_data->op_mds);
1856
1857         rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1858         if (rc)
1859                 RETURN(rc);
1860
1861         /* Send the create request to the MDT where the object
1862          * will be located */
1863         tgt = lmv_find_target(lmv, &op_data->op_fid2);
1864         if (IS_ERR(tgt))
1865                 RETURN(PTR_ERR(tgt));
1866
1867         op_data->op_mds = tgt->ltd_idx;
1868
1869         CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1870                PFID(&op_data->op_fid2), op_data->op_mds);
1871
1872         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1873         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1874                        cap_effective, rdev, request);
1875         if (rc == 0) {
1876                 if (*request == NULL)
1877                         RETURN(rc);
1878                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1879         }
1880         RETURN(rc);
1881 }
1882
1883 static int lmv_done_writing(struct obd_export *exp,
1884                             struct md_op_data *op_data,
1885                             struct md_open_data *mod)
1886 {
1887         struct obd_device     *obd = exp->exp_obd;
1888         struct lmv_obd        *lmv = &obd->u.lmv;
1889         struct lmv_tgt_desc   *tgt;
1890         int                    rc;
1891         ENTRY;
1892
1893         rc = lmv_check_connect(obd);
1894         if (rc)
1895                 RETURN(rc);
1896
1897         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1898         if (IS_ERR(tgt))
1899                 RETURN(PTR_ERR(tgt));
1900
1901         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1902         RETURN(rc);
1903 }
1904
1905 static int
1906 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1907             const union ldlm_policy_data *policy,
1908             struct lookup_intent *it, struct md_op_data *op_data,
1909             struct lustre_handle *lockh, __u64 extra_lock_flags)
1910 {
1911         struct obd_device        *obd = exp->exp_obd;
1912         struct lmv_obd           *lmv = &obd->u.lmv;
1913         struct lmv_tgt_desc      *tgt;
1914         int                       rc;
1915         ENTRY;
1916
1917         rc = lmv_check_connect(obd);
1918         if (rc)
1919                 RETURN(rc);
1920
1921         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1922                LL_IT2STR(it), PFID(&op_data->op_fid1));
1923
1924         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1925         if (IS_ERR(tgt))
1926                 RETURN(PTR_ERR(tgt));
1927
1928         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1929                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1930
1931         rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1932                         extra_lock_flags);
1933
1934         RETURN(rc);
1935 }
1936
1937 static int
1938 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1939                  struct ptlrpc_request **preq)
1940 {
1941         struct ptlrpc_request   *req = NULL;
1942         struct obd_device       *obd = exp->exp_obd;
1943         struct lmv_obd          *lmv = &obd->u.lmv;
1944         struct lmv_tgt_desc     *tgt;
1945         struct mdt_body         *body;
1946         int                      rc;
1947         ENTRY;
1948
1949         rc = lmv_check_connect(obd);
1950         if (rc)
1951                 RETURN(rc);
1952
1953         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1954         if (IS_ERR(tgt))
1955                 RETURN(PTR_ERR(tgt));
1956
1957         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1958                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1959                tgt->ltd_idx);
1960
1961         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1962         if (rc != 0)
1963                 RETURN(rc);
1964
1965         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1966         LASSERT(body != NULL);
1967
1968         if (body->mbo_valid & OBD_MD_MDS) {
1969                 struct lu_fid rid = body->mbo_fid1;
1970                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1971                        PFID(&rid));
1972
1973                 tgt = lmv_find_target(lmv, &rid);
1974                 if (IS_ERR(tgt)) {
1975                         ptlrpc_req_finished(*preq);
1976                         preq = NULL;
1977                         RETURN(PTR_ERR(tgt));
1978                 }
1979
1980                 op_data->op_fid1 = rid;
1981                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1982                 op_data->op_namelen = 0;
1983                 op_data->op_name = NULL;
1984                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1985                 ptlrpc_req_finished(*preq);
1986                 *preq = req;
1987         }
1988
1989         RETURN(rc);
1990 }
1991
1992 #define md_op_data_fid(op_data, fl)                     \
1993         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1994          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1995          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1996          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1997          NULL)
1998
1999 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
2000                             struct md_op_data *op_data,
2001                             int op_tgt, ldlm_mode_t mode, int bits, int flag)
2002 {
2003         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
2004         struct obd_device      *obd = exp->exp_obd;
2005         struct lmv_obd         *lmv = &obd->u.lmv;
2006         ldlm_policy_data_t      policy = {{ 0 }};
2007         int                     rc = 0;
2008         ENTRY;
2009
2010         if (!fid_is_sane(fid))
2011                 RETURN(0);
2012
2013         if (tgt == NULL) {
2014                 tgt = lmv_find_target(lmv, fid);
2015                 if (IS_ERR(tgt))
2016                         RETURN(PTR_ERR(tgt));
2017         }
2018
2019         if (tgt->ltd_idx != op_tgt) {
2020                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
2021                 policy.l_inodebits.bits = bits;
2022                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
2023                                       mode, LCF_ASYNC, NULL);
2024         } else {
2025                 CDEBUG(D_INODE,
2026                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
2027                        op_tgt, PFID(fid));
2028                 op_data->op_flags |= flag;
2029                 rc = 0;
2030         }
2031
2032         RETURN(rc);
2033 }
2034
2035 /*
2036  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2037  * op_data->op_fid2
2038  */
2039 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2040                     struct ptlrpc_request **request)
2041 {
2042         struct obd_device       *obd = exp->exp_obd;
2043         struct lmv_obd          *lmv = &obd->u.lmv;
2044         struct lmv_tgt_desc     *tgt;
2045         int                      rc;
2046         ENTRY;
2047
2048         rc = lmv_check_connect(obd);
2049         if (rc)
2050                 RETURN(rc);
2051
2052         LASSERT(op_data->op_namelen != 0);
2053
2054         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2055                PFID(&op_data->op_fid2), op_data->op_namelen,
2056                op_data->op_name, PFID(&op_data->op_fid1));
2057
2058         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2059         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2060         op_data->op_cap = cfs_curproc_cap_pack();
2061         if (op_data->op_mea2 != NULL) {
2062                 struct lmv_stripe_md    *lsm = op_data->op_mea2;
2063                 const struct lmv_oinfo  *oinfo;
2064
2065                 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2066                                                 op_data->op_namelen);
2067                 if (IS_ERR(oinfo))
2068                         RETURN(PTR_ERR(oinfo));
2069
2070                 op_data->op_fid2 = oinfo->lmo_fid;
2071         }
2072
2073         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2074         if (IS_ERR(tgt))
2075                 RETURN(PTR_ERR(tgt));
2076
2077         /*
2078          * Cancel UPDATE lock on child (fid1).
2079          */
2080         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2081         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2082                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2083         if (rc != 0)
2084                 RETURN(rc);
2085
2086         rc = md_link(tgt->ltd_exp, op_data, request);
2087
2088         RETURN(rc);
2089 }
2090
2091 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2092                       const char *old, int oldlen, const char *new, int newlen,
2093                       struct ptlrpc_request **request)
2094 {
2095         struct obd_device       *obd = exp->exp_obd;
2096         struct lmv_obd          *lmv = &obd->u.lmv;
2097         struct lmv_tgt_desc     *src_tgt;
2098         int                     rc;
2099         ENTRY;
2100
2101         LASSERT(oldlen != 0);
2102
2103         CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2104                oldlen, old, PFID(&op_data->op_fid1),
2105                op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2106                newlen, new, PFID(&op_data->op_fid2),
2107                op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2108
2109         rc = lmv_check_connect(obd);
2110         if (rc)
2111                 RETURN(rc);
2112
2113         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2114         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2115         op_data->op_cap = cfs_curproc_cap_pack();
2116         if (op_data->op_cli_flags & CLI_MIGRATE) {
2117                 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2118                          PFID(&op_data->op_fid3));
2119                 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2120                 if (rc)
2121                         RETURN(rc);
2122                 src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid3);
2123         } else {
2124                 if (op_data->op_mea1 != NULL) {
2125                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2126
2127                         src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2128                                                              oldlen,
2129                                                              &op_data->op_fid1,
2130                                                              &op_data->op_mds);
2131                         if (IS_ERR(src_tgt))
2132                                 RETURN(PTR_ERR(src_tgt));
2133                 } else {
2134                         src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2135                         if (IS_ERR(src_tgt))
2136                                 RETURN(PTR_ERR(src_tgt));
2137
2138                         op_data->op_mds = src_tgt->ltd_idx;
2139                 }
2140
2141                 if (op_data->op_mea2) {
2142                         struct lmv_stripe_md    *lsm = op_data->op_mea2;
2143                         const struct lmv_oinfo  *oinfo;
2144
2145                         oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2146                         if (IS_ERR(oinfo))
2147                                 RETURN(PTR_ERR(oinfo));
2148
2149                         op_data->op_fid2 = oinfo->lmo_fid;
2150                 }
2151         }
2152         if (IS_ERR(src_tgt))
2153                 RETURN(PTR_ERR(src_tgt));
2154
2155         /*
2156          * LOOKUP lock on src child (fid3) should also be cancelled for
2157          * src_tgt in mdc_rename.
2158          */
2159         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2160
2161         /*
2162          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2163          * own target.
2164          */
2165         rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2166                               LCK_EX, MDS_INODELOCK_UPDATE,
2167                               MF_MDC_CANCEL_FID2);
2168
2169         if (rc != 0)
2170                 RETURN(rc);
2171         /*
2172          * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2173          */
2174         if (fid_is_sane(&op_data->op_fid3)) {
2175                 struct lmv_tgt_desc *tgt;
2176
2177                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2178                 if (IS_ERR(tgt))
2179                         RETURN(PTR_ERR(tgt));
2180
2181                 /* Cancel LOOKUP lock on its parent */
2182                 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2183                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2184                                       MF_MDC_CANCEL_FID3);
2185                 if (rc != 0)
2186                         RETURN(rc);
2187
2188                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2189                                       LCK_EX, MDS_INODELOCK_FULL,
2190                                       MF_MDC_CANCEL_FID3);
2191                 if (rc != 0)
2192                         RETURN(rc);
2193         }
2194
2195         /*
2196          * Cancel all the locks on tgt child (fid4).
2197          */
2198         if (fid_is_sane(&op_data->op_fid4))
2199                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2200                                       LCK_EX, MDS_INODELOCK_FULL,
2201                                       MF_MDC_CANCEL_FID4);
2202
2203         CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2204                op_data->op_mds, PFID(&op_data->op_fid2));
2205
2206         rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2207                        request);
2208
2209         RETURN(rc);
2210 }
2211
2212 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2213                        void *ea, int ealen, void *ea2, int ea2len,
2214                        struct ptlrpc_request **request,
2215                        struct md_open_data **mod)
2216 {
2217         struct obd_device       *obd = exp->exp_obd;
2218         struct lmv_obd          *lmv = &obd->u.lmv;
2219         struct lmv_tgt_desc     *tgt;
2220         int                      rc = 0;
2221         ENTRY;
2222
2223         rc = lmv_check_connect(obd);
2224         if (rc)
2225                 RETURN(rc);
2226
2227         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2228                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2229
2230         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2231         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2232         if (IS_ERR(tgt))
2233                 RETURN(PTR_ERR(tgt));
2234
2235         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2236                         ea2len, request, mod);
2237
2238         RETURN(rc);
2239 }
2240
2241 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2242                      struct obd_capa *oc, struct ptlrpc_request **request)
2243 {
2244         struct obd_device       *obd = exp->exp_obd;
2245         struct lmv_obd          *lmv = &obd->u.lmv;
2246         struct lmv_tgt_desc     *tgt;
2247         int                      rc;
2248         ENTRY;
2249
2250         rc = lmv_check_connect(obd);
2251         if (rc != 0)
2252                 RETURN(rc);
2253
2254         tgt = lmv_find_target(lmv, fid);
2255         if (IS_ERR(tgt))
2256                 RETURN(PTR_ERR(tgt));
2257
2258         rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2259         RETURN(rc);
2260 }
2261
2262 /**
2263  * Get current minimum entry from striped directory
2264  *
2265  * This function will search the dir entry, whose hash value is the
2266  * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2267  * only being called for striped directory.
2268  *
2269  * \param[in] exp               export of LMV
2270  * \param[in] op_data           parameters transferred beween client MD stack
2271  *                              stripe_information will be included in this
2272  *                              parameter
2273  * \param[in] cb_op             ldlm callback being used in enqueue in
2274  *                              mdc_read_page
2275  * \param[in] hash_offset       the hash value, which is used to locate
2276  *                              minum(closet) dir entry
2277  * \param[in|out] stripe_offset the caller use this to indicate the stripe
2278  *                              index of last entry, so to avoid hash conflict
2279  *                              between stripes. It will also be used to
2280  *                              return the stripe index of current dir entry.
2281  * \param[in|out] entp          the minum entry and it also is being used
2282  *                              to input the last dir entry to resolve the
2283  *                              hash conflict
2284  *
2285  * \param[out] ppage            the page which holds the minum entry
2286  *
2287  * \retval                      = 0 get the entry successfully
2288  *                              negative errno (< 0) does not get the entry
2289  */
2290 static int lmv_get_min_striped_entry(struct obd_export *exp,
2291                                      struct md_op_data *op_data,
2292                                      struct md_callback *cb_op,
2293                                      __u64 hash_offset, int *stripe_offset,
2294                                      struct lu_dirent **entp,
2295                                      struct page **ppage)
2296 {
2297         struct obd_device       *obd = exp->exp_obd;
2298         struct lmv_obd          *lmv = &obd->u.lmv;
2299         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2300         struct lmv_tgt_desc     *tgt;
2301         int                     stripe_count;
2302         struct lu_dirent        *min_ent = NULL;
2303         struct page             *min_page = NULL;
2304         int                     min_idx = 0;
2305         int                     i;
2306         int                     rc = 0;
2307         ENTRY;
2308
2309         stripe_count = lsm->lsm_md_stripe_count;
2310         for (i = 0; i < stripe_count; i++) {
2311                 struct lu_dirent        *ent = NULL;
2312                 struct page             *page = NULL;
2313                 struct lu_dirpage       *dp;
2314                 __u64                   stripe_hash = hash_offset;
2315
2316                 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2317                 if (IS_ERR(tgt))
2318                         GOTO(out, rc = PTR_ERR(tgt));
2319
2320                 /* op_data will be shared by each stripe, so we need
2321                  * reset these value for each stripe */
2322                 op_data->op_stripe_offset = i;
2323                 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2324                 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2325                 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2326 next:
2327                 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2328                                   &page);
2329                 if (rc != 0)
2330                         GOTO(out, rc);
2331
2332                 dp = page_address(page);
2333                 for (ent = lu_dirent_start(dp); ent != NULL;
2334                      ent = lu_dirent_next(ent)) {
2335                         /* Skip dummy entry */
2336                         if (le16_to_cpu(ent->lde_namelen) == 0)
2337                                 continue;
2338
2339                         if (le64_to_cpu(ent->lde_hash) < hash_offset)
2340                                 continue;
2341
2342                         if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2343                             (*entp == ent || i < *stripe_offset))
2344                                 continue;
2345
2346                         /* skip . and .. for other stripes */
2347                         if (i != 0 &&
2348                             (strncmp(ent->lde_name, ".",
2349                                      le16_to_cpu(ent->lde_namelen)) == 0 ||
2350                              strncmp(ent->lde_name, "..",
2351                                      le16_to_cpu(ent->lde_namelen)) == 0))
2352                                 continue;
2353                         break;
2354                 }
2355
2356                 if (ent == NULL) {
2357                         stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2358
2359                         kunmap(page);
2360                         page_cache_release(page);
2361                         page = NULL;
2362
2363                         /* reach the end of current stripe, go to next stripe */
2364                         if (stripe_hash == MDS_DIR_END_OFF)
2365                                 continue;
2366                         else
2367                                 goto next;
2368                 }
2369
2370                 if (min_ent != NULL) {
2371                         if (le64_to_cpu(min_ent->lde_hash) >
2372                             le64_to_cpu(ent->lde_hash)) {
2373                                 min_ent = ent;
2374                                 kunmap(min_page);
2375                                 page_cache_release(min_page);
2376                                 min_idx = i;
2377                                 min_page = page;
2378                         } else {
2379                                 kunmap(page);
2380                                 page_cache_release(page);
2381                                 page = NULL;
2382                         }
2383                 } else {
2384                         min_ent = ent;
2385                         min_page = page;
2386                         min_idx = i;
2387                 }
2388         }
2389
2390 out:
2391         if (*ppage != NULL) {
2392                 kunmap(*ppage);
2393                 page_cache_release(*ppage);
2394         }
2395         *stripe_offset = min_idx;
2396         *entp = min_ent;
2397         *ppage = min_page;
2398         RETURN(rc);
2399 }
2400
2401 /**
2402  * Build dir entry page from a striped directory
2403  *
2404  * This function gets one entry by @offset from a striped directory. It will
2405  * read entries from all of stripes, and choose one closest to the required
2406  * offset(&offset). A few notes
2407  * 1. skip . and .. for non-zero stripes, because there can only have one .
2408  * and .. in a directory.
2409  * 2. op_data will be shared by all of stripes, instead of allocating new
2410  * one, so need to restore before reusing.
2411  * 3. release the entry page if that is not being chosen.
2412  *
2413  * \param[in] exp       obd export refer to LMV
2414  * \param[in] op_data   hold those MD parameters of read_entry
2415  * \param[in] cb_op     ldlm callback being used in enqueue in mdc_read_entry
2416  * \param[out] ldp      the entry being read
2417  * \param[out] ppage    the page holding the entry. Note: because the entry
2418  *                      will be accessed in upper layer, so we need hold the
2419  *                      page until the usages of entry is finished, see
2420  *                      ll_dir_entry_next.
2421  *
2422  * retval               =0 if get entry successfully
2423  *                      <0 cannot get entry
2424  */
2425 static int lmv_read_striped_page(struct obd_export *exp,
2426                                  struct md_op_data *op_data,
2427                                  struct md_callback *cb_op,
2428                                  __u64 offset, struct page **ppage)
2429 {
2430         struct obd_device       *obd = exp->exp_obd;
2431         struct lu_fid           master_fid = op_data->op_fid1;
2432         struct inode            *master_inode = op_data->op_data;
2433         __u64                   hash_offset = offset;
2434         struct lu_dirpage       *dp;
2435         struct page             *min_ent_page = NULL;
2436         struct page             *ent_page = NULL;
2437         struct lu_dirent        *ent;
2438         void                    *area;
2439         int                     ent_idx = 0;
2440         struct lu_dirent        *min_ent = NULL;
2441         struct lu_dirent        *last_ent;
2442         int                     left_bytes;
2443         int                     rc;
2444         ENTRY;
2445
2446         rc = lmv_check_connect(obd);
2447         if (rc)
2448                 RETURN(rc);
2449
2450         /* Allocate a page and read entries from all of stripes and fill
2451          * the page by hash order */
2452         ent_page = alloc_page(GFP_KERNEL);
2453         if (ent_page == NULL)
2454                 RETURN(-ENOMEM);
2455
2456         /* Initialize the entry page */
2457         dp = kmap(ent_page);
2458         memset(dp, 0, sizeof(*dp));
2459         dp->ldp_hash_start = cpu_to_le64(offset);
2460         dp->ldp_flags |= LDF_COLLIDE;
2461
2462         area = dp + 1;
2463         left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2464         ent = area;
2465         last_ent = ent;
2466         do {
2467                 __u16   ent_size;
2468
2469                 /* Find the minum entry from all sub-stripes */
2470                 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2471                                                &ent_idx, &min_ent,
2472                                                &min_ent_page);
2473                 if (rc != 0)
2474                         GOTO(out, rc);
2475
2476                 /* If it can not get minum entry, it means it already reaches
2477                  * the end of this directory */
2478                 if (min_ent == NULL) {
2479                         last_ent->lde_reclen = 0;
2480                         hash_offset = MDS_DIR_END_OFF;
2481                         GOTO(out, rc);
2482                 }
2483
2484                 ent_size = le16_to_cpu(min_ent->lde_reclen);
2485
2486                 /* the last entry lde_reclen is 0, but it might not
2487                  * the end of this entry of this temporay entry */
2488                 if (ent_size == 0)
2489                         ent_size = lu_dirent_calc_size(
2490                                         le16_to_cpu(min_ent->lde_namelen),
2491                                         le32_to_cpu(min_ent->lde_attrs));
2492                 if (ent_size > left_bytes) {
2493                         last_ent->lde_reclen = cpu_to_le16(0);
2494                         hash_offset = le64_to_cpu(min_ent->lde_hash);
2495                         GOTO(out, rc);
2496                 }
2497
2498                 memcpy(ent, min_ent, ent_size);
2499
2500                 /* Replace . with master FID and Replace .. with the parent FID
2501                  * of master object */
2502                 if (strncmp(ent->lde_name, ".",
2503                             le16_to_cpu(ent->lde_namelen)) == 0 &&
2504                     le16_to_cpu(ent->lde_namelen) == 1)
2505                         fid_cpu_to_le(&ent->lde_fid, &master_fid);
2506                 else if (strncmp(ent->lde_name, "..",
2507                                    le16_to_cpu(ent->lde_namelen)) == 0 &&
2508                            le16_to_cpu(ent->lde_namelen) == 2)
2509                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2510
2511                 left_bytes -= ent_size;
2512                 ent->lde_reclen = cpu_to_le16(ent_size);
2513                 last_ent = ent;
2514                 ent = (void *)ent + ent_size;
2515                 hash_offset = le64_to_cpu(min_ent->lde_hash);
2516                 if (hash_offset == MDS_DIR_END_OFF) {
2517                         last_ent->lde_reclen = 0;
2518                         break;
2519                 }
2520         } while (1);
2521 out:
2522         if (min_ent_page != NULL) {
2523                 kunmap(min_ent_page);
2524                 page_cache_release(min_ent_page);
2525         }
2526
2527         if (unlikely(rc != 0)) {
2528                 __free_page(ent_page);
2529                 ent_page = NULL;
2530         } else {
2531                 if (ent == area)
2532                         dp->ldp_flags |= LDF_EMPTY;
2533                 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2534                 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2535         }
2536
2537         /* We do not want to allocate md_op_data during each
2538          * dir entry reading, so op_data will be shared by every stripe,
2539          * then we need to restore it back to original value before
2540          * return to the upper layer */
2541         op_data->op_fid1 = master_fid;
2542         op_data->op_fid2 = master_fid;
2543         op_data->op_data = master_inode;
2544
2545         *ppage = ent_page;
2546
2547         RETURN(rc);
2548 }
2549
2550 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2551                   struct md_callback *cb_op, __u64 offset,
2552                   struct page **ppage)
2553 {
2554         struct obd_device       *obd = exp->exp_obd;
2555         struct lmv_obd          *lmv = &obd->u.lmv;
2556         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2557         struct lmv_tgt_desc     *tgt;
2558         int                     rc;
2559         ENTRY;
2560
2561         rc = lmv_check_connect(obd);
2562         if (rc != 0)
2563                 RETURN(rc);
2564
2565         if (unlikely(lsm != NULL)) {
2566                 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2567                 RETURN(rc);
2568         }
2569
2570         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2571         if (IS_ERR(tgt))
2572                 RETURN(PTR_ERR(tgt));
2573
2574         rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2575
2576         RETURN(rc);
2577 }
2578
2579 /**
2580  * Unlink a file/directory
2581  *
2582  * Unlink a file or directory under the parent dir. The unlink request
2583  * usually will be sent to the MDT where the child is located, but if
2584  * the client does not have the child FID then request will be sent to the
2585  * MDT where the parent is located.
2586  *
2587  * If the parent is a striped directory then it also needs to locate which
2588  * stripe the name of the child is located, and replace the parent FID
2589  * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2590  * it will walk through all of sub-stripes until the child is being
2591  * unlinked finally.
2592  *
2593  * \param[in] exp       export refer to LMV
2594  * \param[in] op_data   different parameters transferred beween client
2595  *                      MD stacks, name, namelen, FIDs etc.
2596  *                      op_fid1 is the parent FID, op_fid2 is the child
2597  *                      FID.
2598  * \param[out] request  point to the request of unlink.
2599  *
2600  * retval               0 if succeed
2601  *                      negative errno if failed.
2602  */
2603 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2604                       struct ptlrpc_request **request)
2605 {
2606         struct obd_device       *obd = exp->exp_obd;
2607         struct lmv_obd          *lmv = &obd->u.lmv;
2608         struct lmv_tgt_desc     *tgt = NULL;
2609         struct lmv_tgt_desc     *parent_tgt = NULL;
2610         struct mdt_body         *body;
2611         int                     rc;
2612         int                     stripe_index = 0;
2613         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2614         ENTRY;
2615
2616         rc = lmv_check_connect(obd);
2617         if (rc)
2618                 RETURN(rc);
2619 retry_unlink:
2620         /* For striped dir, we need to locate the parent as well */
2621         if (lsm != NULL) {
2622                 struct lmv_tgt_desc *tmp;
2623
2624                 LASSERT(op_data->op_name != NULL &&
2625                         op_data->op_namelen != 0);
2626
2627                 tmp = lmv_locate_target_for_name(lmv, lsm,
2628                                                  op_data->op_name,
2629                                                  op_data->op_namelen,
2630                                                  &op_data->op_fid1,
2631                                                  &op_data->op_mds);
2632
2633                 /* return -EBADFD means unknown hash type, might
2634                  * need try all sub-stripe here */
2635                 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2636                         RETURN(PTR_ERR(tmp));
2637
2638                 /* Note: both migrating dir and unknown hash dir need to
2639                  * try all of sub-stripes, so we need start search the
2640                  * name from stripe 0, but migrating dir is already handled
2641                  * inside lmv_locate_target_for_name(), so we only check
2642                  * unknown hash type directory here */
2643                 if (!lmv_is_known_hash_type(lsm)) {
2644                         struct lmv_oinfo *oinfo;
2645
2646                         oinfo = &lsm->lsm_md_oinfo[stripe_index];
2647
2648                         op_data->op_fid1 = oinfo->lmo_fid;
2649                         op_data->op_mds = oinfo->lmo_mds;
2650                 }
2651         }
2652
2653 try_next_stripe:
2654         /* Send unlink requests to the MDT where the child is located */
2655         if (likely(!fid_is_zero(&op_data->op_fid2)))
2656                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2657         else if (lsm != NULL)
2658                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2659         else
2660                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2661
2662         if (IS_ERR(tgt))
2663                 RETURN(PTR_ERR(tgt));
2664
2665         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2666         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2667         op_data->op_cap = cfs_curproc_cap_pack();
2668
2669         /*
2670          * If child's fid is given, cancel unused locks for it if it is from
2671          * another export than parent.
2672          *
2673          * LOOKUP lock for child (fid3) should also be cancelled on parent
2674          * tgt_tgt in mdc_unlink().
2675          */
2676         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2677
2678         /*
2679          * Cancel FULL locks on child (fid3).
2680          */
2681         parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2682         if (IS_ERR(parent_tgt))
2683                 RETURN(PTR_ERR(parent_tgt));
2684
2685         if (parent_tgt != tgt) {
2686                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2687                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2688                                       MF_MDC_CANCEL_FID3);
2689         }
2690
2691         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2692                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2693         if (rc != 0)
2694                 RETURN(rc);
2695
2696         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n",
2697                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2698
2699         rc = md_unlink(tgt->ltd_exp, op_data, request);
2700         if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2701                 RETURN(rc);
2702
2703         /* Try next stripe if it is needed. */
2704         if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2705                 struct lmv_oinfo *oinfo;
2706
2707                 stripe_index++;
2708                 if (stripe_index >= lsm->lsm_md_stripe_count)
2709                         RETURN(rc);
2710
2711                 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2712
2713                 op_data->op_fid1 = oinfo->lmo_fid;
2714                 op_data->op_mds = oinfo->lmo_mds;
2715
2716                 ptlrpc_req_finished(*request);
2717                 *request = NULL;
2718
2719                 goto try_next_stripe;
2720         }
2721
2722         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2723         if (body == NULL)
2724                 RETURN(-EPROTO);
2725
2726         /* Not cross-ref case, just get out of here. */
2727         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2728                 RETURN(0);
2729
2730         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2731                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2732
2733         /* This is a remote object, try remote MDT, Note: it may
2734          * try more than 1 time here, Considering following case
2735          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2736          * 1. Initially A does not know where remote1 is, it send
2737          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2738          *    resend unlink RPC to MDT1 (retry 1st time).
2739          *
2740          * 2. During the unlink RPC in flight,
2741          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2742          *    and create new remote1, but on MDT0
2743          *
2744          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2745          *    /mnt/lustre, then lookup get fid of remote1, and find
2746          *    it is remote dir again, and replay -EREMOTE again.
2747          *
2748          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2749          *
2750          * In theory, it might try unlimited time here, but it should
2751          * be very rare case.  */
2752         op_data->op_fid2 = body->mbo_fid1;
2753         ptlrpc_req_finished(*request);
2754         *request = NULL;
2755
2756         goto retry_unlink;
2757 }
2758
2759 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2760 {
2761         struct lmv_obd *lmv = &obd->u.lmv;
2762         int rc = 0;
2763
2764         switch (stage) {
2765         case OBD_CLEANUP_EARLY:
2766                 /* XXX: here should be calling obd_precleanup() down to
2767                  * stack. */
2768                 break;
2769         case OBD_CLEANUP_EXPORTS:
2770                 fld_client_proc_fini(&lmv->lmv_fld);
2771                 lprocfs_obd_cleanup(obd);
2772                 lprocfs_free_md_stats(obd);
2773                 break;
2774         default:
2775                 break;
2776         }
2777         RETURN(rc);
2778 }
2779
2780 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2781                         __u32 keylen, void *key, __u32 *vallen, void *val,
2782                         struct lov_stripe_md *lsm)
2783 {
2784         struct obd_device       *obd;
2785         struct lmv_obd          *lmv;
2786         int                      rc = 0;
2787         ENTRY;
2788
2789         obd = class_exp2obd(exp);
2790         if (obd == NULL) {
2791                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2792                        exp->exp_handle.h_cookie);
2793                 RETURN(-EINVAL);
2794         }
2795
2796         lmv = &obd->u.lmv;
2797         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2798                 int i;
2799
2800                 rc = lmv_check_connect(obd);
2801                 if (rc)
2802                         RETURN(rc);
2803
2804                 LASSERT(*vallen == sizeof(__u32));
2805                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2806                         struct lmv_tgt_desc *tgt = lmv->tgts[i];
2807                         /*
2808                          * All tgts should be connected when this gets called.
2809                          */
2810                         if (tgt == NULL || tgt->ltd_exp == NULL)
2811                                 continue;
2812
2813                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2814                                           vallen, val, NULL))
2815                                 RETURN(0);
2816                 }
2817                 RETURN(-EINVAL);
2818         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2819                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2820                    KEY_IS(KEY_MAX_COOKIESIZE) ||
2821                    KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2822                    KEY_IS(KEY_CONN_DATA)) {
2823                 rc = lmv_check_connect(obd);
2824                 if (rc)
2825                         RETURN(rc);
2826
2827                 /*
2828                  * Forwarding this request to first MDS, it should know LOV
2829                  * desc.
2830                  */
2831                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2832                                   vallen, val, NULL);
2833                 if (!rc && KEY_IS(KEY_CONN_DATA))
2834                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2835                 RETURN(rc);
2836         } else if (KEY_IS(KEY_TGT_COUNT)) {
2837                 *((int *)val) = lmv->desc.ld_tgt_count;
2838                 RETURN(0);
2839         }
2840
2841         CDEBUG(D_IOCTL, "Invalid key\n");
2842         RETURN(-EINVAL);
2843 }
2844
2845 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2846                        obd_count keylen, void *key, obd_count vallen,
2847                        void *val, struct ptlrpc_request_set *set)
2848 {
2849         struct lmv_tgt_desc    *tgt = NULL;
2850         struct obd_device      *obd;
2851         struct lmv_obd         *lmv;
2852         int rc = 0;
2853         ENTRY;
2854
2855         obd = class_exp2obd(exp);
2856         if (obd == NULL) {
2857                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2858                        exp->exp_handle.h_cookie);
2859                 RETURN(-EINVAL);
2860         }
2861         lmv = &obd->u.lmv;
2862
2863         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2864                 int i, err = 0;
2865
2866                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2867                         tgt = lmv->tgts[i];
2868
2869                         if (tgt == NULL || tgt->ltd_exp == NULL)
2870                                 continue;
2871
2872                         err = obd_set_info_async(env, tgt->ltd_exp,
2873                                                  keylen, key, vallen, val, set);
2874                         if (err && rc == 0)
2875                                 rc = err;
2876                 }
2877
2878                 RETURN(rc);
2879         }
2880
2881         RETURN(-EINVAL);
2882 }
2883
2884 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2885                           struct lmv_mds_md_v1 *lmm1)
2886 {
2887         int     cplen;
2888         int     i;
2889
2890         lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2891         lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2892         lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
2893         lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
2894         cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
2895                         sizeof(lmm1->lmv_pool_name));
2896         if (cplen >= sizeof(lmm1->lmv_pool_name))
2897                 return -E2BIG;
2898
2899         for (i = 0; i < lsm->lsm_md_stripe_count; i++)
2900                 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
2901                               &lsm->lsm_md_oinfo[i].lmo_fid);
2902         return 0;
2903 }
2904
2905 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
2906                 int stripe_count)
2907 {
2908         int     lmm_size = 0;
2909         bool    allocated = false;
2910         int     rc = 0;
2911         ENTRY;
2912
2913         LASSERT(lmmp != NULL);
2914         /* Free lmm */
2915         if (*lmmp != NULL && lsm == NULL) {
2916                 int stripe_count;
2917
2918                 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
2919                 lmm_size = lmv_mds_md_size(stripe_count,
2920                                            le32_to_cpu((*lmmp)->lmv_magic));
2921                 if (lmm_size == 0)
2922                         RETURN(-EINVAL);
2923                 OBD_FREE(*lmmp, lmm_size);
2924                 *lmmp = NULL;
2925                 RETURN(0);
2926         }
2927
2928         /* Alloc lmm */
2929         if (*lmmp == NULL && lsm == NULL) {
2930                 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
2931                 LASSERT(lmm_size > 0);
2932                 OBD_ALLOC(*lmmp, lmm_size);
2933                 if (*lmmp == NULL)
2934                         RETURN(-ENOMEM);
2935                 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
2936                 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
2937                 RETURN(lmm_size);
2938         }
2939
2940         /* pack lmm */
2941         LASSERT(lsm != NULL);
2942         lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
2943         if (*lmmp == NULL) {
2944                 OBD_ALLOC(*lmmp, lmm_size);
2945                 if (*lmmp == NULL)
2946                         RETURN(-ENOMEM);
2947                 allocated = true;
2948         }
2949
2950         switch (lsm->lsm_md_magic) {
2951         case LMV_MAGIC_V1:
2952                 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
2953                 break;
2954         default:
2955                 rc = -EINVAL;
2956                 break;
2957         }
2958
2959         if (rc != 0 && allocated) {
2960                 OBD_FREE(*lmmp, lmm_size);
2961                 *lmmp = NULL;
2962         }
2963
2964         RETURN(lmm_size);
2965 }
2966 EXPORT_SYMBOL(lmv_pack_md);
2967
2968 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2969                             const struct lmv_mds_md_v1 *lmm1)
2970 {
2971         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
2972         int             stripe_count;
2973         int             cplen;
2974         int             i;
2975         int             rc = 0;
2976         ENTRY;
2977
2978         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2979         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2980         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2981         if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
2982                 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
2983         else
2984                 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2985         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2986         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2987                         sizeof(lsm->lsm_md_pool_name));
2988
2989         if (cplen >= sizeof(lsm->lsm_md_pool_name))
2990                 RETURN(-E2BIG);
2991
2992         CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2993                "layout_version %d\n", lsm->lsm_md_stripe_count,
2994                lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2995                lsm->lsm_md_layout_version);
2996
2997         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2998         for (i = 0; i < le32_to_cpu(stripe_count); i++) {
2999                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3000                               &lmm1->lmv_stripe_fids[i]);
3001                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3002                                     &lsm->lsm_md_oinfo[i].lmo_mds);
3003                 if (rc != 0)
3004                         RETURN(rc);
3005                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3006                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3007         }
3008
3009         RETURN(rc);
3010 }
3011
3012 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3013                   const union lmv_mds_md *lmm, int stripe_count)
3014 {
3015         struct lmv_stripe_md     *lsm;
3016         int                      lsm_size;
3017         int                      rc;
3018         bool                     allocated = false;
3019         ENTRY;
3020
3021         LASSERT(lsmp != NULL);
3022
3023         lsm = *lsmp;
3024         /* Free memmd */
3025         if (lsm != NULL && lmm == NULL) {
3026                 int i;
3027                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3028                         /* For migrating inode, the master stripe and master
3029                          * object will be the same, so do not need iput, see
3030                          * ll_update_lsm_md */
3031                         if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
3032                               i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
3033                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
3034                 }
3035                 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3036                 OBD_FREE(lsm, lsm_size);
3037                 *lsmp = NULL;
3038                 RETURN(0);
3039         }
3040
3041         /* Alloc memmd */
3042         if (lsm == NULL && lmm == NULL) {
3043                 lsm_size = lmv_stripe_md_size(stripe_count);
3044                 OBD_ALLOC(lsm, lsm_size);
3045                 if (lsm == NULL)
3046                         RETURN(-ENOMEM);
3047                 lsm->lsm_md_stripe_count = stripe_count;
3048                 *lsmp = lsm;
3049                 RETURN(0);
3050         }
3051
3052         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3053                 RETURN(-EPERM);
3054
3055         /* Unpack memmd */
3056         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3057             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3058                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3059                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3060                        -EIO);
3061                 RETURN(-EIO);
3062         }
3063
3064         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3065                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3066         else
3067                 /**
3068                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3069                  * stripecount should be 0 then.
3070                  */
3071                 lsm_size = lmv_stripe_md_size(0);
3072
3073         lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3074         if (lsm == NULL) {
3075                 OBD_ALLOC(lsm, lsm_size);
3076                 if (lsm == NULL)
3077                         RETURN(-ENOMEM);
3078                 allocated = true;
3079                 *lsmp = lsm;
3080         }
3081
3082         switch (le32_to_cpu(lmm->lmv_magic)) {
3083         case LMV_MAGIC_V1:
3084                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3085                 break;
3086         default:
3087                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3088                        le32_to_cpu(lmm->lmv_magic));
3089                 rc = -EINVAL;
3090                 break;
3091         }
3092
3093         if (rc != 0 && allocated) {
3094                 OBD_FREE(lsm, lsm_size);
3095                 *lsmp = NULL;
3096                 lsm_size = rc;
3097         }
3098         RETURN(lsm_size);
3099 }
3100
3101 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
3102 {
3103         return lmv_unpack_md(NULL, lsmp, NULL, stripes);
3104 }
3105 EXPORT_SYMBOL(lmv_alloc_memmd);
3106
3107 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3108 {
3109         lmv_unpack_md(NULL, &lsm, NULL, 0);
3110 }
3111 EXPORT_SYMBOL(lmv_free_memmd);
3112
3113 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
3114                  struct lov_mds_md *lmm, int disk_len)
3115 {
3116         return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
3117                              (union lmv_mds_md *)lmm, disk_len);
3118 }
3119
3120 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
3121                struct lov_stripe_md *lsm)
3122 {
3123         struct obd_device               *obd = exp->exp_obd;
3124         struct lmv_obd                  *lmv_obd = &obd->u.lmv;
3125         const struct lmv_stripe_md      *lmv = (struct lmv_stripe_md *)lsm;
3126         int                             stripe_count;
3127
3128         if (lmmp == NULL) {
3129                 if (lsm != NULL)
3130                         stripe_count = lmv->lsm_md_stripe_count;
3131                 else
3132                         stripe_count = lmv_obd->desc.ld_tgt_count;
3133
3134                 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
3135         }
3136
3137         return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
3138 }
3139
3140 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3141                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
3142                              ldlm_cancel_flags_t flags, void *opaque)
3143 {
3144         struct obd_device       *obd = exp->exp_obd;
3145         struct lmv_obd          *lmv = &obd->u.lmv;
3146         int                      rc = 0;
3147         int                      err;
3148         __u32                    i;
3149         ENTRY;
3150
3151         LASSERT(fid != NULL);
3152
3153         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3154                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3155
3156                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3157                         continue;
3158
3159                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3160                                        opaque);
3161                 if (!rc)
3162                         rc = err;
3163         }
3164         RETURN(rc);
3165 }
3166
3167 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3168                       __u64 *bits)
3169 {
3170         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3171         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3172         int                      rc;
3173         ENTRY;
3174
3175         if (tgt == NULL || tgt->ltd_exp == NULL)
3176                 RETURN(-EINVAL);
3177         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3178         RETURN(rc);
3179 }
3180
3181 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3182                            const struct lu_fid *fid, ldlm_type_t type,
3183                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
3184                            struct lustre_handle *lockh)
3185 {
3186         struct obd_device       *obd = exp->exp_obd;
3187         struct lmv_obd          *lmv = &obd->u.lmv;
3188         ldlm_mode_t             rc;
3189         int                     tgt;
3190         int                     i;
3191         ENTRY;
3192
3193         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3194
3195         /*
3196          * With DNE every object can have two locks in different namespaces:
3197          * lookup lock in space of MDT storing direntry and update/open lock in
3198          * space of MDT storing inode.  Try the MDT that the FID maps to first,
3199          * since this can be easily found, and only try others if that fails.
3200          */
3201         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3202              i < lmv->desc.ld_tgt_count;
3203              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3204                 if (tgt < 0) {
3205                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3206                                obd->obd_name, PFID(fid), tgt);
3207                         tgt = 0;
3208                 }
3209
3210                 if (lmv->tgts[tgt] == NULL ||
3211                     lmv->tgts[tgt]->ltd_exp == NULL ||
3212                     lmv->tgts[tgt]->ltd_active == 0)
3213                         continue;
3214
3215                 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3216                           &n