Whamcloud - gitweb
LU-7623 lmv: Mark lmv_hsm_ct_register/unregister uarg as __user
[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, 2015, 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/user_namespace.h>
42 #ifdef HAVE_UIDGID_HEADER
43 # include <linux/uidgid.h>
44 #endif
45 #include <linux/slab.h>
46 #include <linux/pagemap.h>
47 #include <linux/mm.h>
48 #include <linux/math64.h>
49 #include <linux/seq_file.h>
50 #include <linux/namei.h>
51
52 #include <lustre/lustre_idl.h>
53 #include <obd_support.h>
54 #include <lustre_lib.h>
55 #include <lustre_net.h>
56 #include <obd_class.h>
57 #include <lustre_lmv.h>
58 #include <lprocfs_status.h>
59 #include <cl_object.h>
60 #include <lustre_fid.h>
61 #include <lustre_ioctl.h>
62 #include <lustre_kernelcomm.h>
63 #include "lmv_internal.h"
64
65 static void lmv_activate_target(struct lmv_obd *lmv,
66                                 struct lmv_tgt_desc *tgt,
67                                 int activate)
68 {
69         if (tgt->ltd_active == activate)
70                 return;
71
72         tgt->ltd_active = activate;
73         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
74
75         tgt->ltd_exp->exp_obd->obd_inactive = !activate;
76 }
77
78 /**
79  * Error codes:
80  *
81  *  -EINVAL  : UUID can't be found in the LMV's target list
82  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
83  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
84  */
85 static int lmv_set_mdc_active(struct lmv_obd *lmv,
86                               const struct obd_uuid *uuid,
87                               int activate)
88 {
89         struct lmv_tgt_desc     *tgt = NULL;
90         struct obd_device       *obd;
91         __u32                    i;
92         int                      rc = 0;
93         ENTRY;
94
95         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
96                         lmv, uuid->uuid, activate);
97
98         spin_lock(&lmv->lmv_lock);
99         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
100                 tgt = lmv->tgts[i];
101                 if (tgt == NULL || tgt->ltd_exp == NULL)
102                         continue;
103
104                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
105                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
106
107                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
108                         break;
109         }
110
111         if (i == lmv->desc.ld_tgt_count)
112                 GOTO(out_lmv_lock, rc = -EINVAL);
113
114         obd = class_exp2obd(tgt->ltd_exp);
115         if (obd == NULL)
116                 GOTO(out_lmv_lock, rc = -ENOTCONN);
117
118         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
119                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
120                obd->obd_type->typ_name, i);
121         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
122
123         if (tgt->ltd_active == activate) {
124                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
125                        activate ? "" : "in");
126                 GOTO(out_lmv_lock, rc);
127         }
128
129         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
130                activate ? "" : "in");
131         lmv_activate_target(lmv, tgt, activate);
132         EXIT;
133
134  out_lmv_lock:
135         spin_unlock(&lmv->lmv_lock);
136         return rc;
137 }
138
139 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
140 {
141         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
142         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
143
144         return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
145 }
146
147 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
148                       enum obd_notify_event ev, void *data)
149 {
150         struct obd_connect_data *conn_data;
151         struct lmv_obd          *lmv = &obd->u.lmv;
152         struct obd_uuid         *uuid;
153         int                      rc = 0;
154         ENTRY;
155
156         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
157                 CERROR("unexpected notification of %s %s!\n",
158                        watched->obd_type->typ_name,
159                        watched->obd_name);
160                 RETURN(-EINVAL);
161         }
162
163         uuid = &watched->u.cli.cl_target_uuid;
164         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
165                 /*
166                  * Set MDC as active before notifying the observer, so the
167                  * observer can use the MDC normally.
168                  */
169                 rc = lmv_set_mdc_active(lmv, uuid,
170                                         ev == OBD_NOTIFY_ACTIVE);
171                 if (rc) {
172                         CERROR("%sactivation of %s failed: %d\n",
173                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
174                                uuid->uuid, rc);
175                         RETURN(rc);
176                 }
177         } else if (ev == OBD_NOTIFY_OCD) {
178                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
179                 /*
180                  * XXX: Make sure that ocd_connect_flags from all targets are
181                  * the same. Otherwise one of MDTs runs wrong version or
182                  * something like this.  --umka
183                  */
184                 obd->obd_self_export->exp_connect_data = *conn_data;
185         }
186 #if 0
187         else if (ev == OBD_NOTIFY_DISCON) {
188                 /*
189                  * For disconnect event, flush fld cache for failout MDS case.
190                  */
191                 fld_client_flush(&lmv->lmv_fld);
192         }
193 #endif
194         /*
195          * Pass the notification up the chain.
196          */
197         if (obd->obd_observer)
198                 rc = obd_notify(obd->obd_observer, watched, ev, data);
199
200         RETURN(rc);
201 }
202
203 /**
204  * This is fake connect function. Its purpose is to initialize lmv and say
205  * caller that everything is okay. Real connection will be performed later.
206  */
207 static int lmv_connect(const struct lu_env *env,
208                        struct obd_export **exp, struct obd_device *obd,
209                        struct obd_uuid *cluuid, struct obd_connect_data *data,
210                        void *localdata)
211 {
212         struct lmv_obd        *lmv = &obd->u.lmv;
213         struct lustre_handle  conn = { 0 };
214         int                    rc = 0;
215         ENTRY;
216
217         /*
218          * We don't want to actually do the underlying connections more than
219          * once, so keep track.
220          */
221         lmv->refcount++;
222         if (lmv->refcount > 1) {
223                 *exp = NULL;
224                 RETURN(0);
225         }
226
227         rc = class_connect(&conn, obd, cluuid);
228         if (rc) {
229                 CERROR("class_connection() returned %d\n", rc);
230                 RETURN(rc);
231         }
232
233         *exp = class_conn2export(&conn);
234         class_export_get(*exp);
235
236         lmv->exp = *exp;
237         lmv->connected = 0;
238         lmv->cluuid = *cluuid;
239
240         if (data)
241                 lmv->conn_data = *data;
242
243         if (lmv->targets_proc_entry == NULL) {
244                 lmv->targets_proc_entry = lprocfs_register("target_obds",
245                                                            obd->obd_proc_entry,
246                                                            NULL, NULL);
247                 if (IS_ERR(lmv->targets_proc_entry)) {
248                         CERROR("%s: cannot register "
249                                "/proc/fs/lustre/%s/%s/target_obds\n",
250                                obd->obd_name, obd->obd_type->typ_name,
251                                obd->obd_name);
252                         lmv->targets_proc_entry = NULL;
253                 }
254         }
255
256         /*
257          * All real clients should perform actual connection right away, because
258          * it is possible, that LMV will not have opportunity to connect targets
259          * and MDC stuff will be called directly, for instance while reading
260          * ../mdc/../kbytesfree procfs file, etc.
261          */
262         if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
263                 rc = lmv_check_connect(obd);
264
265         if (rc && lmv->targets_proc_entry != NULL)
266                 lprocfs_remove(&lmv->targets_proc_entry);
267         RETURN(rc);
268 }
269
270 static int lmv_init_ea_size(struct obd_export *exp, __u32 easize,
271                             __u32 def_easize)
272 {
273         struct obd_device       *obd = exp->exp_obd;
274         struct lmv_obd          *lmv = &obd->u.lmv;
275         __u32                    i;
276         int                      rc = 0;
277         int                      change = 0;
278         ENTRY;
279
280         if (lmv->max_easize < easize) {
281                 lmv->max_easize = easize;
282                 change = 1;
283         }
284         if (lmv->max_def_easize < def_easize) {
285                 lmv->max_def_easize = def_easize;
286                 change = 1;
287         }
288
289         if (change == 0)
290                 RETURN(0);
291
292         if (lmv->connected == 0)
293                 RETURN(0);
294
295         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
296                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
297
298                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
299                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
300                         continue;
301                 }
302
303                 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize);
304                 if (rc) {
305                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
306                                " rc = %d\n", obd->obd_name, i, rc);
307                         break;
308                 }
309         }
310         RETURN(rc);
311 }
312
313 #define MAX_STRING_SIZE 128
314
315 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
316 {
317         struct lmv_obd          *lmv = &obd->u.lmv;
318         struct obd_uuid         *cluuid = &lmv->cluuid;
319         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
320         struct obd_device       *mdc_obd;
321         struct obd_export       *mdc_exp;
322         struct lu_fld_target     target;
323         int                      rc;
324         ENTRY;
325
326         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
327                                         &obd->obd_uuid);
328         if (!mdc_obd) {
329                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
330                 RETURN(-EINVAL);
331         }
332
333         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
334                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
335                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
336                 cluuid->uuid);
337
338         if (!mdc_obd->obd_set_up) {
339                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
340                 RETURN(-EINVAL);
341         }
342
343         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
344                          &lmv->conn_data, NULL);
345         if (rc) {
346                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
347                 RETURN(rc);
348         }
349
350         /*
351          * Init fid sequence client for this mdc and add new fld target.
352          */
353         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
354         if (rc)
355                 RETURN(rc);
356
357         target.ft_srv = NULL;
358         target.ft_exp = mdc_exp;
359         target.ft_idx = tgt->ltd_idx;
360
361         fld_client_add_target(&lmv->lmv_fld, &target);
362
363         rc = obd_register_observer(mdc_obd, obd);
364         if (rc) {
365                 obd_disconnect(mdc_exp);
366                 CERROR("target %s register_observer error %d\n",
367                        tgt->ltd_uuid.uuid, rc);
368                 RETURN(rc);
369         }
370
371         if (obd->obd_observer) {
372                 /*
373                  * Tell the observer about the new target.
374                  */
375                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
376                                 OBD_NOTIFY_ACTIVE,
377                                 (void *)(tgt - lmv->tgts[0]));
378                 if (rc) {
379                         obd_disconnect(mdc_exp);
380                         RETURN(rc);
381                 }
382         }
383
384         tgt->ltd_active = 1;
385         tgt->ltd_exp = mdc_exp;
386         lmv->desc.ld_active_tgt_count++;
387
388         md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize);
389
390         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
391                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
392                 atomic_read(&obd->obd_refcount));
393
394         if (lmv->targets_proc_entry != NULL) {
395                 struct proc_dir_entry *mdc_symlink;
396
397                 LASSERT(mdc_obd->obd_type != NULL);
398                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
399                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
400                                                   lmv->targets_proc_entry,
401                                                   "../../../%s/%s",
402                                                   mdc_obd->obd_type->typ_name,
403                                                   mdc_obd->obd_name);
404                 if (mdc_symlink == NULL) {
405                         CERROR("cannot register LMV target "
406                                "/proc/fs/lustre/%s/%s/target_obds/%s\n",
407                                obd->obd_type->typ_name, obd->obd_name,
408                                mdc_obd->obd_name);
409                 }
410         }
411         RETURN(0);
412 }
413
414 static void lmv_del_target(struct lmv_obd *lmv, int index)
415 {
416         if (lmv->tgts[index] == NULL)
417                 return;
418
419         OBD_FREE_PTR(lmv->tgts[index]);
420         lmv->tgts[index] = NULL;
421         return;
422 }
423
424 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
425                            __u32 index, int gen)
426 {
427         struct obd_device *mdc_obd;
428         struct lmv_obd      *lmv = &obd->u.lmv;
429         struct lmv_tgt_desc *tgt;
430         int                  orig_tgt_count = 0;
431         int                  rc = 0;
432         ENTRY;
433
434         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
435         mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
436                                         &obd->obd_uuid);
437         if (!mdc_obd) {
438                 CERROR("%s: Target %s not attached: rc = %d\n",
439                        obd->obd_name, uuidp->uuid, -EINVAL);
440                 RETURN(-EINVAL);
441         }
442
443         mutex_lock(&lmv->lmv_init_mutex);
444         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
445                 tgt = lmv->tgts[index];
446                 CERROR("%s: UUID %s already assigned at LOV target index %d:"
447                        " rc = %d\n", obd->obd_name,
448                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
449                 mutex_unlock(&lmv->lmv_init_mutex);
450                 RETURN(-EEXIST);
451         }
452
453         if (index >= lmv->tgts_size) {
454                 /* We need to reallocate the lmv target array. */
455                 struct lmv_tgt_desc **newtgts, **old = NULL;
456                 __u32 newsize = 1;
457                 __u32 oldsize = 0;
458
459                 while (newsize < index + 1)
460                         newsize = newsize << 1;
461                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
462                 if (newtgts == NULL) {
463                         mutex_unlock(&lmv->lmv_init_mutex);
464                         RETURN(-ENOMEM);
465                 }
466
467                 if (lmv->tgts_size) {
468                         memcpy(newtgts, lmv->tgts,
469                                sizeof(*newtgts) * lmv->tgts_size);
470                         old = lmv->tgts;
471                         oldsize = lmv->tgts_size;
472                 }
473
474                 lmv->tgts = newtgts;
475                 lmv->tgts_size = newsize;
476                 smp_rmb();
477                 if (old)
478                         OBD_FREE(old, sizeof(*old) * oldsize);
479
480                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
481                        lmv->tgts_size);
482         }
483
484         OBD_ALLOC_PTR(tgt);
485         if (!tgt) {
486                 mutex_unlock(&lmv->lmv_init_mutex);
487                 RETURN(-ENOMEM);
488         }
489
490         mutex_init(&tgt->ltd_fid_mutex);
491         tgt->ltd_idx = index;
492         tgt->ltd_uuid = *uuidp;
493         tgt->ltd_active = 0;
494         lmv->tgts[index] = tgt;
495         if (index >= lmv->desc.ld_tgt_count) {
496                 orig_tgt_count = lmv->desc.ld_tgt_count;
497                 lmv->desc.ld_tgt_count = index + 1;
498         }
499
500         if (lmv->connected == 0) {
501                 /* lmv_check_connect() will connect this target. */
502                 mutex_unlock(&lmv->lmv_init_mutex);
503                 RETURN(0);
504         }
505
506         /* Otherwise let's connect it ourselves */
507         mutex_unlock(&lmv->lmv_init_mutex);
508         rc = lmv_connect_mdc(obd, tgt);
509         if (rc != 0) {
510                 spin_lock(&lmv->lmv_lock);
511                 if (lmv->desc.ld_tgt_count == index + 1)
512                         lmv->desc.ld_tgt_count = orig_tgt_count;
513                 memset(tgt, 0, sizeof(*tgt));
514                 spin_unlock(&lmv->lmv_lock);
515         } else {
516                 int easize = sizeof(struct lmv_stripe_md) +
517                         lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
518                 lmv_init_ea_size(obd->obd_self_export, easize, 0);
519         }
520
521         RETURN(rc);
522 }
523
524 int lmv_check_connect(struct obd_device *obd)
525 {
526         struct lmv_obd          *lmv = &obd->u.lmv;
527         struct lmv_tgt_desc     *tgt;
528         __u32                    i;
529         int                      rc;
530         int                      easize;
531         ENTRY;
532
533         if (lmv->connected)
534                 RETURN(0);
535
536         mutex_lock(&lmv->lmv_init_mutex);
537         if (lmv->connected) {
538                 mutex_unlock(&lmv->lmv_init_mutex);
539                 RETURN(0);
540         }
541
542         if (lmv->desc.ld_tgt_count == 0) {
543                 mutex_unlock(&lmv->lmv_init_mutex);
544                 CERROR("%s: no targets configured.\n", obd->obd_name);
545                 RETURN(-EINVAL);
546         }
547
548         LASSERT(lmv->tgts != NULL);
549
550         if (lmv->tgts[0] == NULL) {
551                 mutex_unlock(&lmv->lmv_init_mutex);
552                 CERROR("%s: no target configured for index 0.\n",
553                        obd->obd_name);
554                 RETURN(-EINVAL);
555         }
556
557         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
558                lmv->cluuid.uuid, obd->obd_name);
559
560         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
561                 tgt = lmv->tgts[i];
562                 if (tgt == NULL)
563                         continue;
564                 rc = lmv_connect_mdc(obd, tgt);
565                 if (rc)
566                         GOTO(out_disc, rc);
567         }
568
569         class_export_put(lmv->exp);
570         lmv->connected = 1;
571         easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
572         lmv_init_ea_size(obd->obd_self_export, easize, 0);
573         mutex_unlock(&lmv->lmv_init_mutex);
574         RETURN(0);
575
576  out_disc:
577         while (i-- > 0) {
578                 int rc2;
579                 tgt = lmv->tgts[i];
580                 if (tgt == NULL)
581                         continue;
582                 tgt->ltd_active = 0;
583                 if (tgt->ltd_exp) {
584                         --lmv->desc.ld_active_tgt_count;
585                         rc2 = obd_disconnect(tgt->ltd_exp);
586                         if (rc2) {
587                                 CERROR("LMV target %s disconnect on "
588                                        "MDC idx %d: error %d\n",
589                                        tgt->ltd_uuid.uuid, i, rc2);
590                         }
591                 }
592         }
593         class_disconnect(lmv->exp);
594         mutex_unlock(&lmv->lmv_init_mutex);
595         RETURN(rc);
596 }
597
598 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
599 {
600         struct lmv_obd         *lmv = &obd->u.lmv;
601         struct obd_device      *mdc_obd;
602         int                     rc;
603         ENTRY;
604
605         LASSERT(tgt != NULL);
606         LASSERT(obd != NULL);
607
608         mdc_obd = class_exp2obd(tgt->ltd_exp);
609
610         if (mdc_obd) {
611                 mdc_obd->obd_force = obd->obd_force;
612                 mdc_obd->obd_fail = obd->obd_fail;
613                 mdc_obd->obd_no_recov = obd->obd_no_recov;
614
615                 if (lmv->targets_proc_entry != NULL)
616                         lprocfs_remove_proc_entry(mdc_obd->obd_name,
617                                                   lmv->targets_proc_entry);
618         }
619
620         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
621         if (rc)
622                 CERROR("Can't finanize fids factory\n");
623
624         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
625                tgt->ltd_exp->exp_obd->obd_name,
626                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
627
628         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
629         rc = obd_disconnect(tgt->ltd_exp);
630         if (rc) {
631                 if (tgt->ltd_active) {
632                         CERROR("Target %s disconnect error %d\n",
633                                tgt->ltd_uuid.uuid, rc);
634                 }
635         }
636
637         lmv_activate_target(lmv, tgt, 0);
638         tgt->ltd_exp = NULL;
639         RETURN(0);
640 }
641
642 static int lmv_disconnect(struct obd_export *exp)
643 {
644         struct obd_device       *obd = class_exp2obd(exp);
645         struct lmv_obd          *lmv = &obd->u.lmv;
646         int                      rc;
647         __u32                    i;
648         ENTRY;
649
650         if (!lmv->tgts)
651                 goto out_local;
652
653         /*
654          * Only disconnect the underlying layers on the final disconnect.
655          */
656         lmv->refcount--;
657         if (lmv->refcount != 0)
658                 goto out_local;
659
660         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
661                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
662                         continue;
663
664                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
665         }
666
667         if (lmv->targets_proc_entry != NULL)
668                 lprocfs_remove(&lmv->targets_proc_entry);
669         else
670                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
671                        obd->obd_type->typ_name, obd->obd_name);
672
673 out_local:
674         /*
675          * This is the case when no real connection is established by
676          * lmv_check_connect().
677          */
678         if (!lmv->connected)
679                 class_export_put(exp);
680         rc = class_disconnect(exp);
681         if (lmv->refcount == 0)
682                 lmv->connected = 0;
683         RETURN(rc);
684 }
685
686 static int lmv_fid2path(struct obd_export *exp, int len, void *karg,
687                         void __user *uarg)
688 {
689         struct obd_device       *obddev = class_exp2obd(exp);
690         struct lmv_obd          *lmv = &obddev->u.lmv;
691         struct getinfo_fid2path *gf;
692         struct lmv_tgt_desc     *tgt;
693         struct getinfo_fid2path *remote_gf = NULL;
694         int                     remote_gf_size = 0;
695         int                     rc;
696
697         gf = (struct getinfo_fid2path *)karg;
698         tgt = lmv_find_target(lmv, &gf->gf_fid);
699         if (IS_ERR(tgt))
700                 RETURN(PTR_ERR(tgt));
701
702 repeat_fid2path:
703         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
704         if (rc != 0 && rc != -EREMOTE)
705                 GOTO(out_fid2path, rc);
706
707         /* If remote_gf != NULL, it means just building the
708          * path on the remote MDT, copy this path segement to gf */
709         if (remote_gf != NULL) {
710                 struct getinfo_fid2path *ori_gf;
711                 char *ptr;
712
713                 ori_gf = (struct getinfo_fid2path *)karg;
714                 if (strlen(ori_gf->gf_path) +
715                     strlen(gf->gf_path) > ori_gf->gf_pathlen)
716                         GOTO(out_fid2path, rc = -EOVERFLOW);
717
718                 ptr = ori_gf->gf_path;
719
720                 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
721                         strlen(ori_gf->gf_path));
722
723                 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
724                 ptr += strlen(gf->gf_path);
725                 *ptr = '/';
726         }
727
728         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
729                tgt->ltd_exp->exp_obd->obd_name,
730                gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
731                gf->gf_linkno);
732
733         if (rc == 0)
734                 GOTO(out_fid2path, rc);
735
736         /* sigh, has to go to another MDT to do path building further */
737         if (remote_gf == NULL) {
738                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
739                 OBD_ALLOC(remote_gf, remote_gf_size);
740                 if (remote_gf == NULL)
741                         GOTO(out_fid2path, rc = -ENOMEM);
742                 remote_gf->gf_pathlen = PATH_MAX;
743         }
744
745         if (!fid_is_sane(&gf->gf_fid)) {
746                 CERROR("%s: invalid FID "DFID": rc = %d\n",
747                        tgt->ltd_exp->exp_obd->obd_name,
748                        PFID(&gf->gf_fid), -EINVAL);
749                 GOTO(out_fid2path, rc = -EINVAL);
750         }
751
752         tgt = lmv_find_target(lmv, &gf->gf_fid);
753         if (IS_ERR(tgt))
754                 GOTO(out_fid2path, rc = -EINVAL);
755
756         remote_gf->gf_fid = gf->gf_fid;
757         remote_gf->gf_recno = -1;
758         remote_gf->gf_linkno = -1;
759         memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
760         gf = remote_gf;
761         goto repeat_fid2path;
762
763 out_fid2path:
764         if (remote_gf != NULL)
765                 OBD_FREE(remote_gf, remote_gf_size);
766         RETURN(rc);
767 }
768
769 static int lmv_hsm_req_count(struct lmv_obd *lmv,
770                              const struct hsm_user_request *hur,
771                              const struct lmv_tgt_desc *tgt_mds)
772 {
773         __u32                    i;
774         int                      nr = 0;
775         struct lmv_tgt_desc     *curr_tgt;
776
777         /* count how many requests must be sent to the given target */
778         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
779                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
780                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
781                         nr++;
782         }
783         return nr;
784 }
785
786 static void lmv_hsm_req_build(struct lmv_obd *lmv,
787                               struct hsm_user_request *hur_in,
788                               const struct lmv_tgt_desc *tgt_mds,
789                               struct hsm_user_request *hur_out)
790 {
791         __u32                    i, nr_out;
792         struct lmv_tgt_desc     *curr_tgt;
793
794         /* build the hsm_user_request for the given target */
795         hur_out->hur_request = hur_in->hur_request;
796         nr_out = 0;
797         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
798                 curr_tgt = lmv_find_target(lmv,
799                                            &hur_in->hur_user_item[i].hui_fid);
800                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
801                         hur_out->hur_user_item[nr_out] =
802                                                 hur_in->hur_user_item[i];
803                         nr_out++;
804                 }
805         }
806         hur_out->hur_request.hr_itemcount = nr_out;
807         memcpy(hur_data(hur_out), hur_data(hur_in),
808                hur_in->hur_request.hr_data_len);
809 }
810
811 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
812                                  struct lustre_kernelcomm *lk,
813                                  void __user *uarg)
814 {
815         __u32   i;
816         int     rc;
817         ENTRY;
818
819         /* unregister request (call from llapi_hsm_copytool_fini) */
820         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
821                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
822
823                 if (tgt == NULL || tgt->ltd_exp == NULL)
824                         continue;
825                 /* best effort: try to clean as much as possible
826                  * (continue on error) */
827                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
828         }
829
830         /* Whatever the result, remove copytool from kuc groups.
831          * Unreached coordinators will get EPIPE on next requests
832          * and will unregister automatically.
833          */
834         rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group);
835
836         RETURN(rc);
837 }
838
839 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
840                                struct lustre_kernelcomm *lk, __user void *uarg)
841 {
842         struct file             *filp;
843         __u32                    i, j;
844         int                      err, rc;
845         bool                     any_set = false;
846         struct kkuc_ct_data      kcd = { 0 };
847         ENTRY;
848
849         /* All or nothing: try to register to all MDS.
850          * In case of failure, unregister from previous MDS,
851          * except if it because of inactive target. */
852         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
853                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
854
855                 if (tgt == NULL || tgt->ltd_exp == NULL)
856                         continue;
857                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
858                 if (err) {
859                         if (tgt->ltd_active) {
860                                 /* permanent error */
861                                 CERROR("%s: iocontrol MDC %s on MDT"
862                                        " idx %d cmd %x: err = %d\n",
863                                        class_exp2obd(lmv->exp)->obd_name,
864                                        tgt->ltd_uuid.uuid, i, cmd, err);
865                                 rc = err;
866                                 lk->lk_flags |= LK_FLG_STOP;
867                                 /* unregister from previous MDS */
868                                 for (j = 0; j < i; j++) {
869                                         tgt = lmv->tgts[j];
870                                         if (tgt == NULL || tgt->ltd_exp == NULL)
871                                                 continue;
872                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
873                                                       lk, uarg);
874                                 }
875                                 RETURN(rc);
876                         }
877                         /* else: transient error.
878                          * kuc will register to the missing MDT
879                          * when it is back */
880                 } else {
881                         any_set = true;
882                 }
883         }
884
885         if (!any_set)
886                 /* no registration done: return error */
887                 RETURN(-ENOTCONN);
888
889         /* at least one registration done, with no failure */
890         filp = fget(lk->lk_wfd);
891         if (filp == NULL)
892                 RETURN(-EBADF);
893
894         kcd.kcd_magic = KKUC_CT_DATA_MAGIC;
895         kcd.kcd_uuid = lmv->cluuid;
896         kcd.kcd_archive = lk->lk_data;
897
898         rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group,
899                                    &kcd, sizeof(kcd));
900         if (rc != 0)
901                 fput(filp);
902
903         RETURN(rc);
904 }
905
906
907
908
909 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
910                          int len, void *karg, void __user *uarg)
911 {
912         struct obd_device       *obddev = class_exp2obd(exp);
913         struct lmv_obd          *lmv = &obddev->u.lmv;
914         struct lmv_tgt_desc     *tgt = NULL;
915         __u32                    i = 0;
916         int                      rc = 0;
917         int                      set = 0;
918         __u32                    count = lmv->desc.ld_tgt_count;
919         ENTRY;
920
921         if (count == 0)
922                 RETURN(-ENOTTY);
923
924         switch (cmd) {
925         case IOC_OBD_STATFS: {
926                 struct obd_ioctl_data *data = karg;
927                 struct obd_device *mdc_obd;
928                 struct obd_statfs stat_buf = {0};
929                 __u32 index;
930
931                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
932                 if ((index >= count))
933                         RETURN(-ENODEV);
934
935                 tgt = lmv->tgts[index];
936                 if (tgt == NULL || !tgt->ltd_active)
937                         RETURN(-ENODATA);
938
939                 mdc_obd = class_exp2obd(tgt->ltd_exp);
940                 if (!mdc_obd)
941                         RETURN(-EINVAL);
942
943                 /* copy UUID */
944                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
945                                  min((int) data->ioc_plen2,
946                                      (int) sizeof(struct obd_uuid))))
947                         RETURN(-EFAULT);
948
949                 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
950                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
951                                 0);
952                 if (rc)
953                         RETURN(rc);
954                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
955                                  min((int) data->ioc_plen1,
956                                      (int) sizeof(stat_buf))))
957                         RETURN(-EFAULT);
958                 break;
959         }
960         case OBD_IOC_QUOTACTL: {
961                 struct if_quotactl *qctl = karg;
962                 struct obd_quotactl *oqctl;
963
964                 if (qctl->qc_valid == QC_MDTIDX) {
965                         if (count <= qctl->qc_idx)
966                                 RETURN(-EINVAL);
967
968                         tgt = lmv->tgts[qctl->qc_idx];
969                         if (tgt == NULL || tgt->ltd_exp == NULL)
970                                 RETURN(-EINVAL);
971                 } else if (qctl->qc_valid == QC_UUID) {
972                         for (i = 0; i < count; i++) {
973                                 tgt = lmv->tgts[i];
974                                 if (tgt == NULL)
975                                         continue;
976                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
977                                                      &qctl->obd_uuid))
978                                         continue;
979
980                                 if (tgt->ltd_exp == NULL)
981                                         RETURN(-EINVAL);
982
983                                 break;
984                         }
985                 } else {
986                         RETURN(-EINVAL);
987                 }
988
989                 if (i >= count)
990                         RETURN(-EAGAIN);
991
992                 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
993                 OBD_ALLOC_PTR(oqctl);
994                 if (!oqctl)
995                         RETURN(-ENOMEM);
996
997                 QCTL_COPY(oqctl, qctl);
998                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
999                 if (rc == 0) {
1000                         QCTL_COPY(qctl, oqctl);
1001                         qctl->qc_valid = QC_MDTIDX;
1002                         qctl->obd_uuid = tgt->ltd_uuid;
1003                 }
1004                 OBD_FREE_PTR(oqctl);
1005                 break;
1006         }
1007         case OBD_IOC_CHANGELOG_SEND:
1008         case OBD_IOC_CHANGELOG_CLEAR: {
1009                 struct ioc_changelog *icc = karg;
1010
1011                 if (icc->icc_mdtindex >= count)
1012                         RETURN(-ENODEV);
1013
1014                 tgt = lmv->tgts[icc->icc_mdtindex];
1015                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1016                         RETURN(-ENODEV);
1017                 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1018                 break;
1019         }
1020         case LL_IOC_GET_CONNECT_FLAGS: {
1021                 tgt = lmv->tgts[0];
1022                 if (tgt == NULL || tgt->ltd_exp == NULL)
1023                         RETURN(-ENODATA);
1024                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1025                 break;
1026         }
1027         case LL_IOC_FID2MDTIDX: {
1028                 struct lu_fid *fid = karg;
1029                 int             mdt_index;
1030
1031                 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1032                 if (rc != 0)
1033                         RETURN(rc);
1034
1035                 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1036                  * point to user space memory for FID2MDTIDX. */
1037                 *(__u32 *)uarg = mdt_index;
1038                 break;
1039         }
1040         case OBD_IOC_FID2PATH: {
1041                 rc = lmv_fid2path(exp, len, karg, uarg);
1042                 break;
1043         }
1044         case LL_IOC_HSM_STATE_GET:
1045         case LL_IOC_HSM_STATE_SET:
1046         case LL_IOC_HSM_ACTION: {
1047                 struct md_op_data       *op_data = karg;
1048
1049                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1050                 if (IS_ERR(tgt))
1051                         RETURN(PTR_ERR(tgt));
1052
1053                 if (tgt->ltd_exp == NULL)
1054                         RETURN(-EINVAL);
1055
1056                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1057                 break;
1058         }
1059         case LL_IOC_HSM_PROGRESS: {
1060                 const struct hsm_progress_kernel *hpk = karg;
1061
1062                 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1063                 if (IS_ERR(tgt))
1064                         RETURN(PTR_ERR(tgt));
1065                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1066                 break;
1067         }
1068         case LL_IOC_HSM_REQUEST: {
1069                 struct hsm_user_request *hur = karg;
1070                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1071
1072                 if (reqcount == 0)
1073                         RETURN(0);
1074
1075                 /* if the request is about a single fid
1076                  * or if there is a single MDS, no need to split
1077                  * the request. */
1078                 if (reqcount == 1 || count == 1) {
1079                         tgt = lmv_find_target(lmv,
1080                                               &hur->hur_user_item[0].hui_fid);
1081                         if (IS_ERR(tgt))
1082                                 RETURN(PTR_ERR(tgt));
1083                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1084                 } else {
1085                         /* split fid list to their respective MDS */
1086                         for (i = 0; i < count; i++) {
1087                                 unsigned int            nr, reqlen;
1088                                 int                     rc1;
1089                                 struct hsm_user_request *req;
1090
1091                                 tgt = lmv->tgts[i];
1092                                 if (tgt == NULL || tgt->ltd_exp == NULL)
1093                                         continue;
1094
1095                                 nr = lmv_hsm_req_count(lmv, hur, tgt);
1096                                 if (nr == 0) /* nothing for this MDS */
1097                                         continue;
1098
1099                                 /* build a request with fids for this MDS */
1100                                 reqlen = offsetof(typeof(*hur),
1101                                                   hur_user_item[nr])
1102                                                 + hur->hur_request.hr_data_len;
1103                                 OBD_ALLOC_LARGE(req, reqlen);
1104                                 if (req == NULL)
1105                                         RETURN(-ENOMEM);
1106
1107                                 lmv_hsm_req_build(lmv, hur, tgt, req);
1108
1109                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1110                                                     req, uarg);
1111                                 if (rc1 != 0 && rc == 0)
1112                                         rc = rc1;
1113                                 OBD_FREE_LARGE(req, reqlen);
1114                         }
1115                 }
1116                 break;
1117         }
1118         case LL_IOC_LOV_SWAP_LAYOUTS: {
1119                 struct md_op_data       *op_data = karg;
1120                 struct lmv_tgt_desc     *tgt1, *tgt2;
1121
1122                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1123                 if (IS_ERR(tgt1))
1124                         RETURN(PTR_ERR(tgt1));
1125
1126                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1127                 if (IS_ERR(tgt2))
1128                         RETURN(PTR_ERR(tgt2));
1129
1130                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1131                         RETURN(-EINVAL);
1132
1133                 /* only files on same MDT can have their layouts swapped */
1134                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1135                         RETURN(-EPERM);
1136
1137                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1138                 break;
1139         }
1140         case LL_IOC_HSM_CT_START: {
1141                 struct lustre_kernelcomm *lk = karg;
1142                 if (lk->lk_flags & LK_FLG_STOP)
1143                         rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1144                 else
1145                         rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1146                 break;
1147         }
1148         default:
1149                 for (i = 0; i < count; i++) {
1150                         struct obd_device *mdc_obd;
1151                         int err;
1152
1153                         tgt = lmv->tgts[i];
1154                         if (tgt == NULL || tgt->ltd_exp == NULL)
1155                                 continue;
1156                         /* ll_umount_begin() sets force flag but for lmv, not
1157                          * mdc. Let's pass it through */
1158                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1159                         mdc_obd->obd_force = obddev->obd_force;
1160                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1161                         if (err) {
1162                                 if (tgt->ltd_active) {
1163                                         CERROR("error: iocontrol MDC %s on MDT"
1164                                                " idx %d cmd %x: err = %d\n",
1165                                                tgt->ltd_uuid.uuid, i, cmd, err);
1166                                         if (!rc)
1167                                                 rc = err;
1168                                 }
1169                         } else
1170                                 set = 1;
1171                 }
1172                 if (!set && !rc)
1173                         rc = -EIO;
1174         }
1175         RETURN(rc);
1176 }
1177
1178 #if 0
1179 static int lmv_all_chars_policy(int count, const char *name,
1180                                 int len)
1181 {
1182         unsigned int c = 0;
1183
1184         while (len > 0)
1185                 c += name[--len];
1186         c = c % count;
1187         return c;
1188 }
1189
1190 static int lmv_nid_policy(struct lmv_obd *lmv)
1191 {
1192         struct obd_import *imp;
1193         __u32              id;
1194
1195         /*
1196          * XXX: To get nid we assume that underlying obd device is mdc.
1197          */
1198         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1199         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1200         return id % lmv->desc.ld_tgt_count;
1201 }
1202
1203 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1204                           placement_policy_t placement)
1205 {
1206         switch (placement) {
1207         case PLACEMENT_CHAR_POLICY:
1208                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1209                                             op_data->op_name,
1210                                             op_data->op_namelen);
1211         case PLACEMENT_NID_POLICY:
1212                 return lmv_nid_policy(lmv);
1213
1214         default:
1215                 break;
1216         }
1217
1218         CERROR("Unsupported placement policy %x\n", placement);
1219         return -EINVAL;
1220 }
1221 #endif
1222
1223 /**
1224  * This is _inode_ placement policy function (not name).
1225  */
1226 static int lmv_placement_policy(struct obd_device *obd,
1227                                 struct md_op_data *op_data, u32 *mds)
1228 {
1229         struct lmv_obd          *lmv = &obd->u.lmv;
1230         ENTRY;
1231
1232         LASSERT(mds != NULL);
1233
1234         if (lmv->desc.ld_tgt_count == 1) {
1235                 *mds = 0;
1236                 RETURN(0);
1237         }
1238
1239         if (op_data->op_default_stripe_offset != -1) {
1240                 *mds = op_data->op_default_stripe_offset;
1241                 RETURN(0);
1242         }
1243
1244         /**
1245          * If stripe_offset is provided during setdirstripe
1246          * (setdirstripe -i xx), xx MDS will be choosen.
1247          */
1248         if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1249                 struct lmv_user_md *lum;
1250
1251                 lum = op_data->op_data;
1252
1253                 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1254                         *mds = le32_to_cpu(lum->lum_stripe_offset);
1255                 } else {
1256                         /* -1 means default, which will be in the same MDT with
1257                          * the stripe */
1258                         *mds = op_data->op_mds;
1259                         lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1260                 }
1261         } else {
1262                 /* Allocate new fid on target according to operation type and
1263                  * parent home mds. */
1264                 *mds = op_data->op_mds;
1265         }
1266
1267         RETURN(0);
1268 }
1269
1270 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1271 {
1272         struct lmv_tgt_desc     *tgt;
1273         int                      rc;
1274         ENTRY;
1275
1276         tgt = lmv_get_target(lmv, mds, NULL);
1277         if (IS_ERR(tgt))
1278                 RETURN(PTR_ERR(tgt));
1279
1280         /*
1281          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1282          * on server that seq in new allocated fid is not yet known.
1283          */
1284         mutex_lock(&tgt->ltd_fid_mutex);
1285
1286         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1287                 GOTO(out, rc = -ENODEV);
1288
1289         /*
1290          * Asking underlying tgt layer to allocate new fid.
1291          */
1292         rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1293         if (rc > 0) {
1294                 LASSERT(fid_is_sane(fid));
1295                 rc = 0;
1296         }
1297
1298         EXIT;
1299 out:
1300         mutex_unlock(&tgt->ltd_fid_mutex);
1301         return rc;
1302 }
1303
1304 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1305                   struct lu_fid *fid, struct md_op_data *op_data)
1306 {
1307         struct obd_device     *obd = class_exp2obd(exp);
1308         struct lmv_obd        *lmv = &obd->u.lmv;
1309         u32                    mds = 0;
1310         int                    rc;
1311         ENTRY;
1312
1313         LASSERT(op_data != NULL);
1314         LASSERT(fid != NULL);
1315
1316         rc = lmv_placement_policy(obd, op_data, &mds);
1317         if (rc) {
1318                 CERROR("Can't get target for allocating fid, "
1319                        "rc %d\n", rc);
1320                 RETURN(rc);
1321         }
1322
1323         rc = __lmv_fid_alloc(lmv, fid, mds);
1324         if (rc) {
1325                 CERROR("Can't alloc new fid, rc %d\n", rc);
1326                 RETURN(rc);
1327         }
1328
1329         RETURN(rc);
1330 }
1331
1332 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1333 {
1334         struct lmv_obd  *lmv = &obd->u.lmv;
1335         struct lmv_desc *desc;
1336         int             rc;
1337         ENTRY;
1338
1339         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1340                 CERROR("LMV setup requires a descriptor\n");
1341                 RETURN(-EINVAL);
1342         }
1343
1344         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1345         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1346                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1347                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1348                 RETURN(-EINVAL);
1349         }
1350
1351         lmv->tgts_size = 32U;
1352         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1353         if (lmv->tgts == NULL)
1354                 RETURN(-ENOMEM);
1355
1356         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1357         lmv->desc.ld_tgt_count = 0;
1358         lmv->desc.ld_active_tgt_count = 0;
1359         lmv->max_def_easize = 0;
1360         lmv->max_easize = 0;
1361         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1362
1363         spin_lock_init(&lmv->lmv_lock);
1364         mutex_init(&lmv->lmv_init_mutex);
1365
1366 #ifdef CONFIG_PROC_FS
1367         obd->obd_vars = lprocfs_lmv_obd_vars;
1368         lprocfs_obd_setup(obd);
1369         lprocfs_alloc_md_stats(obd, 0);
1370         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1371                                 0444, &lmv_proc_target_fops, obd);
1372         if (rc)
1373                 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1374                       obd->obd_name, rc);
1375 #endif
1376         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1377                              LUSTRE_CLI_FLD_HASH_DHT);
1378         if (rc) {
1379                 CERROR("Can't init FLD, err %d\n", rc);
1380                 GOTO(out, rc);
1381         }
1382
1383         RETURN(0);
1384
1385 out:
1386         return rc;
1387 }
1388
1389 static int lmv_cleanup(struct obd_device *obd)
1390 {
1391         struct lmv_obd   *lmv = &obd->u.lmv;
1392         ENTRY;
1393
1394         fld_client_fini(&lmv->lmv_fld);
1395         if (lmv->tgts != NULL) {
1396                 int i;
1397                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1398                         if (lmv->tgts[i] == NULL)
1399                                 continue;
1400                         lmv_del_target(lmv, i);
1401                 }
1402                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1403                 lmv->tgts_size = 0;
1404         }
1405         RETURN(0);
1406 }
1407
1408 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1409 {
1410         struct lustre_cfg       *lcfg = buf;
1411         struct obd_uuid         obd_uuid;
1412         int                     gen;
1413         __u32                   index;
1414         int                     rc;
1415         ENTRY;
1416
1417         switch (lcfg->lcfg_command) {
1418         case LCFG_ADD_MDC:
1419                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1420                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1421                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1422                         GOTO(out, rc = -EINVAL);
1423
1424                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1425
1426                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1427                         GOTO(out, rc = -EINVAL);
1428                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1429                         GOTO(out, rc = -EINVAL);
1430                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1431                 GOTO(out, rc);
1432         default:
1433                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1434                 GOTO(out, rc = -EINVAL);
1435         }
1436 out:
1437         RETURN(rc);
1438 }
1439
1440 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1441                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1442 {
1443         struct obd_device       *obd = class_exp2obd(exp);
1444         struct lmv_obd          *lmv = &obd->u.lmv;
1445         struct obd_statfs       *temp;
1446         int                      rc = 0;
1447         __u32                    i;
1448         ENTRY;
1449
1450         rc = lmv_check_connect(obd);
1451         if (rc)
1452                 RETURN(rc);
1453
1454         OBD_ALLOC(temp, sizeof(*temp));
1455         if (temp == NULL)
1456                 RETURN(-ENOMEM);
1457
1458         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1459                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1460                         continue;
1461
1462                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1463                                 max_age, flags);
1464                 if (rc) {
1465                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1466                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1467                                rc);
1468                         GOTO(out_free_temp, rc);
1469                 }
1470
1471                 if (i == 0) {
1472                         *osfs = *temp;
1473                         /* If the statfs is from mount, it will needs
1474                          * retrieve necessary information from MDT0.
1475                          * i.e. mount does not need the merged osfs
1476                          * from all of MDT.
1477                          * And also clients can be mounted as long as
1478                          * MDT0 is in service*/
1479                         if (flags & OBD_STATFS_FOR_MDT0)
1480                                 GOTO(out_free_temp, rc);
1481                 } else {
1482                         osfs->os_bavail += temp->os_bavail;
1483                         osfs->os_blocks += temp->os_blocks;
1484                         osfs->os_ffree += temp->os_ffree;
1485                         osfs->os_files += temp->os_files;
1486                 }
1487         }
1488
1489         EXIT;
1490 out_free_temp:
1491         OBD_FREE(temp, sizeof(*temp));
1492         return rc;
1493 }
1494
1495 static int lmv_getstatus(struct obd_export *exp, struct lu_fid *fid)
1496 {
1497         struct obd_device    *obd = exp->exp_obd;
1498         struct lmv_obd       *lmv = &obd->u.lmv;
1499         int                   rc;
1500         ENTRY;
1501
1502         rc = lmv_check_connect(obd);
1503         if (rc)
1504                 RETURN(rc);
1505
1506         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid);
1507         RETURN(rc);
1508 }
1509
1510 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1511                         u64 valid, const char *name,
1512                         const char *input, int input_size, int output_size,
1513                         int flags, struct ptlrpc_request **request)
1514 {
1515         struct obd_device      *obd = exp->exp_obd;
1516         struct lmv_obd         *lmv = &obd->u.lmv;
1517         struct lmv_tgt_desc    *tgt;
1518         int                     rc;
1519         ENTRY;
1520
1521         rc = lmv_check_connect(obd);
1522         if (rc)
1523                 RETURN(rc);
1524
1525         tgt = lmv_find_target(lmv, fid);
1526         if (IS_ERR(tgt))
1527                 RETURN(PTR_ERR(tgt));
1528
1529         rc = md_getxattr(tgt->ltd_exp, fid, valid, name, input,
1530                          input_size, output_size, flags, request);
1531
1532         RETURN(rc);
1533 }
1534
1535 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1536                         u64 valid, const char *name,
1537                         const char *input, int input_size, int output_size,
1538                         int flags, __u32 suppgid,
1539                         struct ptlrpc_request **request)
1540 {
1541         struct obd_device      *obd = exp->exp_obd;
1542         struct lmv_obd         *lmv = &obd->u.lmv;
1543         struct lmv_tgt_desc    *tgt;
1544         int                     rc;
1545         ENTRY;
1546
1547         rc = lmv_check_connect(obd);
1548         if (rc)
1549                 RETURN(rc);
1550
1551         tgt = lmv_find_target(lmv, fid);
1552         if (IS_ERR(tgt))
1553                 RETURN(PTR_ERR(tgt));
1554
1555         rc = md_setxattr(tgt->ltd_exp, fid, valid, name, input,
1556                          input_size, output_size, flags, suppgid,
1557                          request);
1558
1559         RETURN(rc);
1560 }
1561
1562 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1563                        struct ptlrpc_request **request)
1564 {
1565         struct obd_device       *obd = exp->exp_obd;
1566         struct lmv_obd          *lmv = &obd->u.lmv;
1567         struct lmv_tgt_desc     *tgt;
1568         int                      rc;
1569         ENTRY;
1570
1571         rc = lmv_check_connect(obd);
1572         if (rc)
1573                 RETURN(rc);
1574
1575         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1576         if (IS_ERR(tgt))
1577                 RETURN(PTR_ERR(tgt));
1578
1579         if (op_data->op_flags & MF_GET_MDT_IDX) {
1580                 op_data->op_mds = tgt->ltd_idx;
1581                 RETURN(0);
1582         }
1583
1584         rc = md_getattr(tgt->ltd_exp, op_data, request);
1585
1586         RETURN(rc);
1587 }
1588
1589 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1590 {
1591         struct obd_device   *obd = exp->exp_obd;
1592         struct lmv_obd      *lmv = &obd->u.lmv;
1593         __u32                i;
1594         int                  rc;
1595         ENTRY;
1596
1597         rc = lmv_check_connect(obd);
1598         if (rc)
1599                 RETURN(rc);
1600
1601         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1602
1603         /*
1604          * With DNE every object can have two locks in different namespaces:
1605          * lookup lock in space of MDT storing direntry and update/open lock in
1606          * space of MDT storing inode.
1607          */
1608         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1609                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1610                         continue;
1611                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1612         }
1613
1614         RETURN(0);
1615 }
1616
1617 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1618                            ldlm_iterator_t it, void *data)
1619 {
1620         struct obd_device       *obd = exp->exp_obd;
1621         struct lmv_obd          *lmv = &obd->u.lmv;
1622         int                     i;
1623         int                     tgt;
1624         int                     rc;
1625         ENTRY;
1626
1627         rc = lmv_check_connect(obd);
1628         if (rc)
1629                 RETURN(rc);
1630
1631         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1632
1633         /*
1634          * With DNE every object can have two locks in different namespaces:
1635          * lookup lock in space of MDT storing direntry and update/open lock in
1636          * space of MDT storing inode.  Try the MDT that the FID maps to first,
1637          * since this can be easily found, and only try others if that fails.
1638          */
1639         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1640              i < lmv->desc.ld_tgt_count;
1641              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1642                 if (tgt < 0) {
1643                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1644                                obd->obd_name, PFID(fid), tgt);
1645                         tgt = 0;
1646                 }
1647
1648                 if (lmv->tgts[tgt] == NULL ||
1649                     lmv->tgts[tgt]->ltd_exp == NULL)
1650                         continue;
1651
1652                 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1653                 if (rc)
1654                         RETURN(rc);
1655         }
1656
1657         RETURN(rc);
1658 }
1659
1660
1661 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1662                      struct md_open_data *mod, struct ptlrpc_request **request)
1663 {
1664         struct obd_device     *obd = exp->exp_obd;
1665         struct lmv_obd        *lmv = &obd->u.lmv;
1666         struct lmv_tgt_desc   *tgt;
1667         int                    rc;
1668         ENTRY;
1669
1670         rc = lmv_check_connect(obd);
1671         if (rc)
1672                 RETURN(rc);
1673
1674         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1675         if (IS_ERR(tgt))
1676                 RETURN(PTR_ERR(tgt));
1677
1678         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1679         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1680         RETURN(rc);
1681 }
1682
1683 /**
1684  * Choosing the MDT by name or FID in @op_data.
1685  * For non-striped directory, it will locate MDT by fid.
1686  * For striped-directory, it will locate MDT by name. And also
1687  * it will reset op_fid1 with the FID of the choosen stripe.
1688  **/
1689 struct lmv_tgt_desc *
1690 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1691                            const char *name, int namelen, struct lu_fid *fid,
1692                            u32 *mds)
1693 {
1694         struct lmv_tgt_desc     *tgt;
1695         const struct lmv_oinfo  *oinfo;
1696
1697         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1698                 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1699                         RETURN(ERR_PTR(-EBADF));
1700                 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1701         } else {
1702                 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1703                 if (IS_ERR(oinfo))
1704                         RETURN(ERR_CAST(oinfo));
1705         }
1706
1707         if (fid != NULL)
1708                 *fid = oinfo->lmo_fid;
1709         if (mds != NULL)
1710                 *mds = oinfo->lmo_mds;
1711
1712         tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1713
1714         CDEBUG(D_INFO, "locate on mds %u "DFID"\n", oinfo->lmo_mds,
1715                PFID(&oinfo->lmo_fid));
1716         return tgt;
1717 }
1718
1719 /**
1720  * Locate mds by fid or name
1721  *
1722  * For striped directory (lsm != NULL), it will locate the stripe
1723  * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1724  * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1725  * walk through all of stripes to locate the entry.
1726  *
1727  * For normal direcotry, it will locate MDS by FID directly.
1728  * \param[in] lmv       LMV device
1729  * \param[in] op_data   client MD stack parameters, name, namelen
1730  *                      mds_num etc.
1731  * \param[in] fid       object FID used to locate MDS.
1732  *
1733  * retval               pointer to the lmv_tgt_desc if succeed.
1734  *                      ERR_PTR(errno) if failed.
1735  */
1736 struct lmv_tgt_desc*
1737 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1738                struct lu_fid *fid)
1739 {
1740         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1741         struct lmv_tgt_desc     *tgt;
1742
1743         /* During creating VOLATILE file, it should honor the mdt
1744          * index if the file under striped dir is being restored, see
1745          * ct_restore(). */
1746         if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1747             (int)op_data->op_mds != -1) {
1748                 int i;
1749                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1750                 if (IS_ERR(tgt))
1751                         return tgt;
1752
1753                 if (lsm != NULL) {
1754                         /* refill the right parent fid */
1755                         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1756                                 struct lmv_oinfo *oinfo;
1757
1758                                 oinfo = &lsm->lsm_md_oinfo[i];
1759                                 if (oinfo->lmo_mds == op_data->op_mds) {
1760                                         *fid = oinfo->lmo_fid;
1761                                         break;
1762                                 }
1763                         }
1764
1765                         if (i == lsm->lsm_md_stripe_count)
1766                                 *fid = lsm->lsm_md_oinfo[0].lmo_fid;
1767                 }
1768
1769                 return tgt;
1770         }
1771
1772         if (lsm == NULL || op_data->op_namelen == 0) {
1773                 tgt = lmv_find_target(lmv, fid);
1774                 if (IS_ERR(tgt))
1775                         return tgt;
1776
1777                 op_data->op_mds = tgt->ltd_idx;
1778                 return tgt;
1779         }
1780
1781         return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1782                                           op_data->op_namelen, fid,
1783                                           &op_data->op_mds);
1784 }
1785
1786 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1787                 const void *data, size_t datalen, umode_t mode, uid_t uid,
1788                 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1789                 struct ptlrpc_request **request)
1790 {
1791         struct obd_device       *obd = exp->exp_obd;
1792         struct lmv_obd          *lmv = &obd->u.lmv;
1793         struct lmv_tgt_desc     *tgt;
1794         int                      rc;
1795         ENTRY;
1796
1797         rc = lmv_check_connect(obd);
1798         if (rc)
1799                 RETURN(rc);
1800
1801         if (!lmv->desc.ld_active_tgt_count)
1802                 RETURN(-EIO);
1803
1804         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1805         if (IS_ERR(tgt))
1806                 RETURN(PTR_ERR(tgt));
1807
1808         CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1809                 (int)op_data->op_namelen, op_data->op_name,
1810                 PFID(&op_data->op_fid1), op_data->op_mds);
1811
1812         rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1813         if (rc)
1814                 RETURN(rc);
1815         if (exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE) {
1816                 /* Send the create request to the MDT where the object
1817                  * will be located */
1818                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1819                 if (IS_ERR(tgt))
1820                         RETURN(PTR_ERR(tgt));
1821
1822                 op_data->op_mds = tgt->ltd_idx;
1823         } else {
1824                 CDEBUG(D_CONFIG, "Server doesn't support striped dirs\n");
1825         }
1826
1827         CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1828                PFID(&op_data->op_fid2), op_data->op_mds);
1829
1830         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1831         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1832                        cap_effective, rdev, request);
1833         if (rc == 0) {
1834                 if (*request == NULL)
1835                         RETURN(rc);
1836                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1837         }
1838         RETURN(rc);
1839 }
1840
1841 static int
1842 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1843             const union ldlm_policy_data *policy,
1844             struct lookup_intent *it, struct md_op_data *op_data,
1845             struct lustre_handle *lockh, __u64 extra_lock_flags)
1846 {
1847         struct obd_device        *obd = exp->exp_obd;
1848         struct lmv_obd           *lmv = &obd->u.lmv;
1849         struct lmv_tgt_desc      *tgt;
1850         int                       rc;
1851         ENTRY;
1852
1853         rc = lmv_check_connect(obd);
1854         if (rc)
1855                 RETURN(rc);
1856
1857         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1858                LL_IT2STR(it), PFID(&op_data->op_fid1));
1859
1860         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1861         if (IS_ERR(tgt))
1862                 RETURN(PTR_ERR(tgt));
1863
1864         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%u\n",
1865                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1866
1867         rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1868                         extra_lock_flags);
1869
1870         RETURN(rc);
1871 }
1872
1873 static int
1874 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1875                  struct ptlrpc_request **preq)
1876 {
1877         struct ptlrpc_request   *req = NULL;
1878         struct obd_device       *obd = exp->exp_obd;
1879         struct lmv_obd          *lmv = &obd->u.lmv;
1880         struct lmv_tgt_desc     *tgt;
1881         struct mdt_body         *body;
1882         int                      rc;
1883         ENTRY;
1884
1885         rc = lmv_check_connect(obd);
1886         if (rc)
1887                 RETURN(rc);
1888
1889         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1890         if (IS_ERR(tgt))
1891                 RETURN(PTR_ERR(tgt));
1892
1893         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1894                 (int)op_data->op_namelen, op_data->op_name,
1895                 PFID(&op_data->op_fid1), tgt->ltd_idx);
1896
1897         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1898         if (rc != 0)
1899                 RETURN(rc);
1900
1901         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1902         LASSERT(body != NULL);
1903
1904         if (body->mbo_valid & OBD_MD_MDS) {
1905                 struct lu_fid rid = body->mbo_fid1;
1906                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1907                        PFID(&rid));
1908
1909                 tgt = lmv_find_target(lmv, &rid);
1910                 if (IS_ERR(tgt)) {
1911                         ptlrpc_req_finished(*preq);
1912                         preq = NULL;
1913                         RETURN(PTR_ERR(tgt));
1914                 }
1915
1916                 op_data->op_fid1 = rid;
1917                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1918                 op_data->op_namelen = 0;
1919                 op_data->op_name = NULL;
1920                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1921                 ptlrpc_req_finished(*preq);
1922                 *preq = req;
1923         }
1924
1925         RETURN(rc);
1926 }
1927
1928 #define md_op_data_fid(op_data, fl)                     \
1929         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1930          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1931          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1932          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1933          NULL)
1934
1935 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1936                             struct md_op_data *op_data, __u32 op_tgt,
1937                             enum ldlm_mode mode, int bits, int flag)
1938 {
1939         struct lu_fid *fid = md_op_data_fid(op_data, flag);
1940         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
1941         union ldlm_policy_data policy = { { 0 } };
1942         int rc = 0;
1943         ENTRY;
1944
1945         if (!fid_is_sane(fid))
1946                 RETURN(0);
1947
1948         if (tgt == NULL) {
1949                 tgt = lmv_find_target(lmv, fid);
1950                 if (IS_ERR(tgt))
1951                         RETURN(PTR_ERR(tgt));
1952         }
1953
1954         if (tgt->ltd_idx != op_tgt) {
1955                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1956                 policy.l_inodebits.bits = bits;
1957                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1958                                       mode, LCF_ASYNC, NULL);
1959         } else {
1960                 CDEBUG(D_INODE,
1961                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1962                        op_tgt, PFID(fid));
1963                 op_data->op_flags |= flag;
1964                 rc = 0;
1965         }
1966
1967         RETURN(rc);
1968 }
1969
1970 /*
1971  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1972  * op_data->op_fid2
1973  */
1974 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1975                     struct ptlrpc_request **request)
1976 {
1977         struct obd_device       *obd = exp->exp_obd;
1978         struct lmv_obd          *lmv = &obd->u.lmv;
1979         struct lmv_tgt_desc     *tgt;
1980         int                      rc;
1981         ENTRY;
1982
1983         rc = lmv_check_connect(obd);
1984         if (rc)
1985                 RETURN(rc);
1986
1987         LASSERT(op_data->op_namelen != 0);
1988
1989         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
1990                PFID(&op_data->op_fid2), (int)op_data->op_namelen,
1991                op_data->op_name, PFID(&op_data->op_fid1));
1992
1993         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
1994         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
1995         op_data->op_cap = cfs_curproc_cap_pack();
1996         if (op_data->op_mea2 != NULL) {
1997                 struct lmv_stripe_md    *lsm = op_data->op_mea2;
1998                 const struct lmv_oinfo  *oinfo;
1999
2000                 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2001                                                 op_data->op_namelen);
2002                 if (IS_ERR(oinfo))
2003                         RETURN(PTR_ERR(oinfo));
2004
2005                 op_data->op_fid2 = oinfo->lmo_fid;
2006         }
2007
2008         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2009         if (IS_ERR(tgt))
2010                 RETURN(PTR_ERR(tgt));
2011
2012         /*
2013          * Cancel UPDATE lock on child (fid1).
2014          */
2015         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2016         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2017                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2018         if (rc != 0)
2019                 RETURN(rc);
2020
2021         rc = md_link(tgt->ltd_exp, op_data, request);
2022
2023         RETURN(rc);
2024 }
2025
2026 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2027                       const char *old, size_t oldlen,
2028                       const char *new, size_t newlen,
2029                       struct ptlrpc_request **request)
2030 {
2031         struct obd_device       *obd = exp->exp_obd;
2032         struct lmv_obd          *lmv = &obd->u.lmv;
2033         struct lmv_tgt_desc     *src_tgt;
2034         struct lmv_tgt_desc     *tgt_tgt;
2035         struct obd_export       *target_exp;
2036         struct mdt_body         *body;
2037         int                     rc;
2038         ENTRY;
2039
2040         LASSERT(oldlen != 0);
2041
2042         CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2043                (int)oldlen, old, PFID(&op_data->op_fid1),
2044                op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2045                (int)newlen, new, PFID(&op_data->op_fid2),
2046                op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2047
2048         rc = lmv_check_connect(obd);
2049         if (rc)
2050                 RETURN(rc);
2051
2052         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2053         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2054         op_data->op_cap = cfs_curproc_cap_pack();
2055         if (op_data->op_cli_flags & CLI_MIGRATE) {
2056                 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2057                          PFID(&op_data->op_fid3));
2058
2059                 if (op_data->op_mea1 != NULL) {
2060                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2061                         struct lmv_tgt_desc     *tmp;
2062
2063                         /* Fix the parent fid for striped dir */
2064                         tmp = lmv_locate_target_for_name(lmv, lsm, old,
2065                                                          oldlen,
2066                                                          &op_data->op_fid1,
2067                                                          NULL);
2068                         if (IS_ERR(tmp))
2069                                 RETURN(PTR_ERR(tmp));
2070                 }
2071
2072                 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2073                 if (rc != 0)
2074                         RETURN(rc);
2075
2076                 src_tgt = lmv_find_target(lmv, &op_data->op_fid3);
2077                 if (IS_ERR(src_tgt))
2078                         RETURN(PTR_ERR(src_tgt));
2079
2080                 target_exp = src_tgt->ltd_exp;
2081         } else {
2082                 if (op_data->op_mea1 != NULL) {
2083                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2084
2085                         src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2086                                                              oldlen,
2087                                                              &op_data->op_fid1,
2088                                                              &op_data->op_mds);
2089                 } else {
2090                         src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2091                 }
2092                 if (IS_ERR(src_tgt))
2093                         RETURN(PTR_ERR(src_tgt));
2094
2095
2096                 if (op_data->op_mea2 != NULL) {
2097                         struct lmv_stripe_md    *lsm = op_data->op_mea2;
2098
2099                         tgt_tgt = lmv_locate_target_for_name(lmv, lsm, new,
2100                                                              newlen,
2101                                                              &op_data->op_fid2,
2102                                                              &op_data->op_mds);
2103                 } else {
2104                         tgt_tgt = lmv_find_target(lmv, &op_data->op_fid2);
2105
2106                 }
2107                 if (IS_ERR(tgt_tgt))
2108                         RETURN(PTR_ERR(tgt_tgt));
2109
2110                 target_exp = tgt_tgt->ltd_exp;
2111         }
2112
2113         /*
2114          * LOOKUP lock on src child (fid3) should also be cancelled for
2115          * src_tgt in mdc_rename.
2116          */
2117         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2118
2119         /*
2120          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2121          * own target.
2122          */
2123         rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2124                               LCK_EX, MDS_INODELOCK_UPDATE,
2125                               MF_MDC_CANCEL_FID2);
2126
2127         if (rc != 0)
2128                 RETURN(rc);
2129         /*
2130          * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2131          */
2132         if (fid_is_sane(&op_data->op_fid3)) {
2133                 struct lmv_tgt_desc *tgt;
2134
2135                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2136                 if (IS_ERR(tgt))
2137                         RETURN(PTR_ERR(tgt));
2138
2139                 /* Cancel LOOKUP lock on its parent */
2140                 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2141                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2142                                       MF_MDC_CANCEL_FID3);
2143                 if (rc != 0)
2144                         RETURN(rc);
2145
2146                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2147                                       LCK_EX, MDS_INODELOCK_FULL,
2148                                       MF_MDC_CANCEL_FID3);
2149                 if (rc != 0)
2150                         RETURN(rc);
2151         }
2152
2153 retry_rename:
2154         /*
2155          * Cancel all the locks on tgt child (fid4).
2156          */
2157         if (fid_is_sane(&op_data->op_fid4)) {
2158                 struct lmv_tgt_desc *tgt;
2159
2160                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2161                                       LCK_EX, MDS_INODELOCK_FULL,
2162                                       MF_MDC_CANCEL_FID4);
2163                 if (rc != 0)
2164                         RETURN(rc);
2165
2166                 tgt = lmv_find_target(lmv, &op_data->op_fid4);
2167                 if (IS_ERR(tgt))
2168                         RETURN(PTR_ERR(tgt));
2169
2170                 /* Since the target child might be destroyed, and it might
2171                  * become orphan, and we can only check orphan on the local
2172                  * MDT right now, so we send rename request to the MDT where
2173                  * target child is located. If target child does not exist,
2174                  * then it will send the request to the target parent */
2175                 target_exp = tgt->ltd_exp;
2176         }
2177
2178         rc = md_rename(target_exp, op_data, old, oldlen, new, newlen,
2179                        request);
2180
2181         if (rc != 0 && rc != -EXDEV)
2182                 RETURN(rc);
2183
2184         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2185         if (body == NULL)
2186                 RETURN(-EPROTO);
2187
2188         /* Not cross-ref case, just get out of here. */
2189         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2190                 RETURN(rc);
2191
2192         CDEBUG(D_INODE, "%s: try rename to another MDT for "DFID"\n",
2193                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2194
2195         op_data->op_fid4 = body->mbo_fid1;
2196         ptlrpc_req_finished(*request);
2197         *request = NULL;
2198         goto retry_rename;
2199 }
2200
2201 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2202                        void *ea, size_t ealen, struct ptlrpc_request **request)
2203 {
2204         struct obd_device       *obd = exp->exp_obd;
2205         struct lmv_obd          *lmv = &obd->u.lmv;
2206         struct lmv_tgt_desc     *tgt;
2207         int                      rc = 0;
2208         ENTRY;
2209
2210         rc = lmv_check_connect(obd);
2211         if (rc)
2212                 RETURN(rc);
2213
2214         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2215                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2216
2217         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2218         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2219         if (IS_ERR(tgt))
2220                 RETURN(PTR_ERR(tgt));
2221
2222         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2223
2224         RETURN(rc);
2225 }
2226
2227 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2228                      struct ptlrpc_request **request)
2229 {
2230         struct obd_device       *obd = exp->exp_obd;
2231         struct lmv_obd          *lmv = &obd->u.lmv;
2232         struct lmv_tgt_desc     *tgt;
2233         int                      rc;
2234         ENTRY;
2235
2236         rc = lmv_check_connect(obd);
2237         if (rc != 0)
2238                 RETURN(rc);
2239
2240         tgt = lmv_find_target(lmv, fid);
2241         if (IS_ERR(tgt))
2242                 RETURN(PTR_ERR(tgt));
2243
2244         rc = md_fsync(tgt->ltd_exp, fid, request);
2245         RETURN(rc);
2246 }
2247
2248 /**
2249  * Get current minimum entry from striped directory
2250  *
2251  * This function will search the dir entry, whose hash value is the
2252  * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2253  * only being called for striped directory.
2254  *
2255  * \param[in] exp               export of LMV
2256  * \param[in] op_data           parameters transferred beween client MD stack
2257  *                              stripe_information will be included in this
2258  *                              parameter
2259  * \param[in] cb_op             ldlm callback being used in enqueue in
2260  *                              mdc_read_page
2261  * \param[in] hash_offset       the hash value, which is used to locate
2262  *                              minum(closet) dir entry
2263  * \param[in|out] stripe_offset the caller use this to indicate the stripe
2264  *                              index of last entry, so to avoid hash conflict
2265  *                              between stripes. It will also be used to
2266  *                              return the stripe index of current dir entry.
2267  * \param[in|out] entp          the minum entry and it also is being used
2268  *                              to input the last dir entry to resolve the
2269  *                              hash conflict
2270  *
2271  * \param[out] ppage            the page which holds the minum entry
2272  *
2273  * \retval                      = 0 get the entry successfully
2274  *                              negative errno (< 0) does not get the entry
2275  */
2276 static int lmv_get_min_striped_entry(struct obd_export *exp,
2277                                      struct md_op_data *op_data,
2278                                      struct md_callback *cb_op,
2279                                      __u64 hash_offset, int *stripe_offset,
2280                                      struct lu_dirent **entp,
2281                                      struct page **ppage)
2282 {
2283         struct obd_device       *obd = exp->exp_obd;
2284         struct lmv_obd          *lmv = &obd->u.lmv;
2285         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2286         struct lmv_tgt_desc     *tgt;
2287         int                     stripe_count;
2288         struct lu_dirent        *min_ent = NULL;
2289         struct page             *min_page = NULL;
2290         int                     min_idx = 0;
2291         int                     i;
2292         int                     rc = 0;
2293         ENTRY;
2294
2295         stripe_count = lsm->lsm_md_stripe_count;
2296         for (i = 0; i < stripe_count; i++) {
2297                 struct lu_dirent        *ent = NULL;
2298                 struct page             *page = NULL;
2299                 struct lu_dirpage       *dp;
2300                 __u64                   stripe_hash = hash_offset;
2301
2302                 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2303                 if (IS_ERR(tgt))
2304                         GOTO(out, rc = PTR_ERR(tgt));
2305
2306                 /* op_data will be shared by each stripe, so we need
2307                  * reset these value for each stripe */
2308                 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2309                 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2310                 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2311 next:
2312                 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2313                                   &page);
2314                 if (rc != 0)
2315                         GOTO(out, rc);
2316
2317                 dp = page_address(page);
2318                 for (ent = lu_dirent_start(dp); ent != NULL;
2319                      ent = lu_dirent_next(ent)) {
2320                         /* Skip dummy entry */
2321                         if (le16_to_cpu(ent->lde_namelen) == 0)
2322                                 continue;
2323
2324                         if (le64_to_cpu(ent->lde_hash) < hash_offset)
2325                                 continue;
2326
2327                         if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2328                             (*entp == ent || i < *stripe_offset))
2329                                 continue;
2330
2331                         /* skip . and .. for other stripes */
2332                         if (i != 0 &&
2333                             (strncmp(ent->lde_name, ".",
2334                                      le16_to_cpu(ent->lde_namelen)) == 0 ||
2335                              strncmp(ent->lde_name, "..",
2336                                      le16_to_cpu(ent->lde_namelen)) == 0))
2337                                 continue;
2338                         break;
2339                 }
2340
2341                 if (ent == NULL) {
2342                         stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2343
2344                         kunmap(page);
2345                         page_cache_release(page);
2346                         page = NULL;
2347
2348                         /* reach the end of current stripe, go to next stripe */
2349                         if (stripe_hash == MDS_DIR_END_OFF)
2350                                 continue;
2351                         else
2352                                 goto next;
2353                 }
2354
2355                 if (min_ent != NULL) {
2356                         if (le64_to_cpu(min_ent->lde_hash) >
2357                             le64_to_cpu(ent->lde_hash)) {
2358                                 min_ent = ent;
2359                                 kunmap(min_page);
2360                                 page_cache_release(min_page);
2361                                 min_idx = i;
2362                                 min_page = page;
2363                         } else {
2364                                 kunmap(page);
2365                                 page_cache_release(page);
2366                                 page = NULL;
2367                         }
2368                 } else {
2369                         min_ent = ent;
2370                         min_page = page;
2371                         min_idx = i;
2372                 }
2373         }
2374
2375 out:
2376         if (*ppage != NULL) {
2377                 kunmap(*ppage);
2378                 page_cache_release(*ppage);
2379         }
2380         *stripe_offset = min_idx;
2381         *entp = min_ent;
2382         *ppage = min_page;
2383         RETURN(rc);
2384 }
2385
2386 /**
2387  * Build dir entry page from a striped directory
2388  *
2389  * This function gets one entry by @offset from a striped directory. It will
2390  * read entries from all of stripes, and choose one closest to the required
2391  * offset(&offset). A few notes
2392  * 1. skip . and .. for non-zero stripes, because there can only have one .
2393  * and .. in a directory.
2394  * 2. op_data will be shared by all of stripes, instead of allocating new
2395  * one, so need to restore before reusing.
2396  * 3. release the entry page if that is not being chosen.
2397  *
2398  * \param[in] exp       obd export refer to LMV
2399  * \param[in] op_data   hold those MD parameters of read_entry
2400  * \param[in] cb_op     ldlm callback being used in enqueue in mdc_read_entry
2401  * \param[out] ldp      the entry being read
2402  * \param[out] ppage    the page holding the entry. Note: because the entry
2403  *                      will be accessed in upper layer, so we need hold the
2404  *                      page until the usages of entry is finished, see
2405  *                      ll_dir_entry_next.
2406  *
2407  * retval               =0 if get entry successfully
2408  *                      <0 cannot get entry
2409  */
2410 static int lmv_read_striped_page(struct obd_export *exp,
2411                                  struct md_op_data *op_data,
2412                                  struct md_callback *cb_op,
2413                                  __u64 offset, struct page **ppage)
2414 {
2415         struct obd_device       *obd = exp->exp_obd;
2416         struct lu_fid           master_fid = op_data->op_fid1;
2417         struct inode            *master_inode = op_data->op_data;
2418         __u64                   hash_offset = offset;
2419         struct lu_dirpage       *dp;
2420         struct page             *min_ent_page = NULL;
2421         struct page             *ent_page = NULL;
2422         struct lu_dirent        *ent;
2423         void                    *area;
2424         int                     ent_idx = 0;
2425         struct lu_dirent        *min_ent = NULL;
2426         struct lu_dirent        *last_ent;
2427         size_t                  left_bytes;
2428         int                     rc;
2429         ENTRY;
2430
2431         rc = lmv_check_connect(obd);
2432         if (rc)
2433                 RETURN(rc);
2434
2435         /* Allocate a page and read entries from all of stripes and fill
2436          * the page by hash order */
2437         ent_page = alloc_page(GFP_KERNEL);
2438         if (ent_page == NULL)
2439                 RETURN(-ENOMEM);
2440
2441         /* Initialize the entry page */
2442         dp = kmap(ent_page);
2443         memset(dp, 0, sizeof(*dp));
2444         dp->ldp_hash_start = cpu_to_le64(offset);
2445         dp->ldp_flags |= LDF_COLLIDE;
2446
2447         area = dp + 1;
2448         left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2449         ent = area;
2450         last_ent = ent;
2451         do {
2452                 __u16   ent_size;
2453
2454                 /* Find the minum entry from all sub-stripes */
2455                 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2456                                                &ent_idx, &min_ent,
2457                                                &min_ent_page);
2458                 if (rc != 0)
2459                         GOTO(out, rc);
2460
2461                 /* If it can not get minum entry, it means it already reaches
2462                  * the end of this directory */
2463                 if (min_ent == NULL) {
2464                         last_ent->lde_reclen = 0;
2465                         hash_offset = MDS_DIR_END_OFF;
2466                         GOTO(out, rc);
2467                 }
2468
2469                 ent_size = le16_to_cpu(min_ent->lde_reclen);
2470
2471                 /* the last entry lde_reclen is 0, but it might not
2472                  * the end of this entry of this temporay entry */
2473                 if (ent_size == 0)
2474                         ent_size = lu_dirent_calc_size(
2475                                         le16_to_cpu(min_ent->lde_namelen),
2476                                         le32_to_cpu(min_ent->lde_attrs));
2477                 if (ent_size > left_bytes) {
2478                         last_ent->lde_reclen = cpu_to_le16(0);
2479                         hash_offset = le64_to_cpu(min_ent->lde_hash);
2480                         GOTO(out, rc);
2481                 }
2482
2483                 memcpy(ent, min_ent, ent_size);
2484
2485                 /* Replace . with master FID and Replace .. with the parent FID
2486                  * of master object */
2487                 if (strncmp(ent->lde_name, ".",
2488                             le16_to_cpu(ent->lde_namelen)) == 0 &&
2489                     le16_to_cpu(ent->lde_namelen) == 1)
2490                         fid_cpu_to_le(&ent->lde_fid, &master_fid);
2491                 else if (strncmp(ent->lde_name, "..",
2492                                    le16_to_cpu(ent->lde_namelen)) == 0 &&
2493                            le16_to_cpu(ent->lde_namelen) == 2)
2494                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2495
2496                 left_bytes -= ent_size;
2497                 ent->lde_reclen = cpu_to_le16(ent_size);
2498                 last_ent = ent;
2499                 ent = (void *)ent + ent_size;
2500                 hash_offset = le64_to_cpu(min_ent->lde_hash);
2501                 if (hash_offset == MDS_DIR_END_OFF) {
2502                         last_ent->lde_reclen = 0;
2503                         break;
2504                 }
2505         } while (1);
2506 out:
2507         if (min_ent_page != NULL) {
2508                 kunmap(min_ent_page);
2509                 page_cache_release(min_ent_page);
2510         }
2511
2512         if (unlikely(rc != 0)) {
2513                 __free_page(ent_page);
2514                 ent_page = NULL;
2515         } else {
2516                 if (ent == area)
2517                         dp->ldp_flags |= LDF_EMPTY;
2518                 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2519                 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2520         }
2521
2522         /* We do not want to allocate md_op_data during each
2523          * dir entry reading, so op_data will be shared by every stripe,
2524          * then we need to restore it back to original value before
2525          * return to the upper layer */
2526         op_data->op_fid1 = master_fid;
2527         op_data->op_fid2 = master_fid;
2528         op_data->op_data = master_inode;
2529
2530         *ppage = ent_page;
2531
2532         RETURN(rc);
2533 }
2534
2535 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2536                   struct md_callback *cb_op, __u64 offset,
2537                   struct page **ppage)
2538 {
2539         struct obd_device       *obd = exp->exp_obd;
2540         struct lmv_obd          *lmv = &obd->u.lmv;
2541         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2542         struct lmv_tgt_desc     *tgt;
2543         int                     rc;
2544         ENTRY;
2545
2546         rc = lmv_check_connect(obd);
2547         if (rc != 0)
2548                 RETURN(rc);
2549
2550         if (unlikely(lsm != NULL)) {
2551                 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2552                 RETURN(rc);
2553         }
2554
2555         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2556         if (IS_ERR(tgt))
2557                 RETURN(PTR_ERR(tgt));
2558
2559         rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2560
2561         RETURN(rc);
2562 }
2563
2564 /**
2565  * Unlink a file/directory
2566  *
2567  * Unlink a file or directory under the parent dir. The unlink request
2568  * usually will be sent to the MDT where the child is located, but if
2569  * the client does not have the child FID then request will be sent to the
2570  * MDT where the parent is located.
2571  *
2572  * If the parent is a striped directory then it also needs to locate which
2573  * stripe the name of the child is located, and replace the parent FID
2574  * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2575  * it will walk through all of sub-stripes until the child is being
2576  * unlinked finally.
2577  *
2578  * \param[in] exp       export refer to LMV
2579  * \param[in] op_data   different parameters transferred beween client
2580  *                      MD stacks, name, namelen, FIDs etc.
2581  *                      op_fid1 is the parent FID, op_fid2 is the child
2582  *                      FID.
2583  * \param[out] request  point to the request of unlink.
2584  *
2585  * retval               0 if succeed
2586  *                      negative errno if failed.
2587  */
2588 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2589                       struct ptlrpc_request **request)
2590 {
2591         struct obd_device       *obd = exp->exp_obd;
2592         struct lmv_obd          *lmv = &obd->u.lmv;
2593         struct lmv_tgt_desc     *tgt = NULL;
2594         struct lmv_tgt_desc     *parent_tgt = NULL;
2595         struct mdt_body         *body;
2596         int                     rc;
2597         int                     stripe_index = 0;
2598         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2599         ENTRY;
2600
2601         rc = lmv_check_connect(obd);
2602         if (rc)
2603                 RETURN(rc);
2604 retry_unlink:
2605         /* For striped dir, we need to locate the parent as well */
2606         if (lsm != NULL) {
2607                 struct lmv_tgt_desc *tmp;
2608
2609                 LASSERT(op_data->op_name != NULL &&
2610                         op_data->op_namelen != 0);
2611
2612                 tmp = lmv_locate_target_for_name(lmv, lsm,
2613                                                  op_data->op_name,
2614                                                  op_data->op_namelen,
2615                                                  &op_data->op_fid1,
2616                                                  &op_data->op_mds);
2617
2618                 /* return -EBADFD means unknown hash type, might
2619                  * need try all sub-stripe here */
2620                 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2621                         RETURN(PTR_ERR(tmp));
2622
2623                 /* Note: both migrating dir and unknown hash dir need to
2624                  * try all of sub-stripes, so we need start search the
2625                  * name from stripe 0, but migrating dir is already handled
2626                  * inside lmv_locate_target_for_name(), so we only check
2627                  * unknown hash type directory here */
2628                 if (!lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
2629                         struct lmv_oinfo *oinfo;
2630
2631                         oinfo = &lsm->lsm_md_oinfo[stripe_index];
2632
2633                         op_data->op_fid1 = oinfo->lmo_fid;
2634                         op_data->op_mds = oinfo->lmo_mds;
2635                 }
2636         }
2637
2638 try_next_stripe:
2639         /* Send unlink requests to the MDT where the child is located */
2640         if (likely(!fid_is_zero(&op_data->op_fid2)))
2641                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2642         else if (lsm != NULL)
2643                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2644         else
2645                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2646
2647         if (IS_ERR(tgt))
2648                 RETURN(PTR_ERR(tgt));
2649
2650         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2651         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2652         op_data->op_cap = cfs_curproc_cap_pack();
2653
2654         /*
2655          * If child's fid is given, cancel unused locks for it if it is from
2656          * another export than parent.
2657          *
2658          * LOOKUP lock for child (fid3) should also be cancelled on parent
2659          * tgt_tgt in mdc_unlink().
2660          */
2661         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2662
2663         /*
2664          * Cancel FULL locks on child (fid3).
2665          */
2666         parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2667         if (IS_ERR(parent_tgt))
2668                 RETURN(PTR_ERR(parent_tgt));
2669
2670         if (parent_tgt != tgt) {
2671                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2672                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2673                                       MF_MDC_CANCEL_FID3);
2674         }
2675
2676         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2677                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2678         if (rc != 0)
2679                 RETURN(rc);
2680
2681         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2682                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2683
2684         rc = md_unlink(tgt->ltd_exp, op_data, request);
2685         if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2686                 RETURN(rc);
2687
2688         /* Try next stripe if it is needed. */
2689         if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2690                 struct lmv_oinfo *oinfo;
2691
2692                 stripe_index++;
2693                 if (stripe_index >= lsm->lsm_md_stripe_count)
2694                         RETURN(rc);
2695
2696                 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2697
2698                 op_data->op_fid1 = oinfo->lmo_fid;
2699                 op_data->op_mds = oinfo->lmo_mds;
2700
2701                 ptlrpc_req_finished(*request);
2702                 *request = NULL;
2703
2704                 goto try_next_stripe;
2705         }
2706
2707         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2708         if (body == NULL)
2709                 RETURN(-EPROTO);
2710
2711         /* Not cross-ref case, just get out of here. */
2712         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2713                 RETURN(rc);
2714
2715         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2716                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2717
2718         /* This is a remote object, try remote MDT, Note: it may
2719          * try more than 1 time here, Considering following case
2720          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2721          * 1. Initially A does not know where remote1 is, it send
2722          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2723          *    resend unlink RPC to MDT1 (retry 1st time).
2724          *
2725          * 2. During the unlink RPC in flight,
2726          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2727          *    and create new remote1, but on MDT0
2728          *
2729          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2730          *    /mnt/lustre, then lookup get fid of remote1, and find
2731          *    it is remote dir again, and replay -EREMOTE again.
2732          *
2733          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2734          *
2735          * In theory, it might try unlimited time here, but it should
2736          * be very rare case.  */
2737         op_data->op_fid2 = body->mbo_fid1;
2738         ptlrpc_req_finished(*request);
2739         *request = NULL;
2740
2741         goto retry_unlink;
2742 }
2743
2744 static int lmv_precleanup(struct obd_device *obd)
2745 {
2746         ENTRY;
2747         fld_client_proc_fini(&obd->u.lmv.lmv_fld);
2748         lprocfs_obd_cleanup(obd);
2749         lprocfs_free_md_stats(obd);
2750         RETURN(0);
2751 }
2752
2753 /**
2754  * Get by key a value associated with a LMV device.
2755  *
2756  * Dispatch request to lower-layer devices as needed.
2757  *
2758  * \param[in] env               execution environment for this thread
2759  * \param[in] exp               export for the LMV device
2760  * \param[in] keylen            length of key identifier
2761  * \param[in] key               identifier of key to get value for
2762  * \param[in] vallen            size of \a val
2763  * \param[out] val              pointer to storage location for value
2764  * \param[in] lsm               optional striping metadata of object
2765  *
2766  * \retval 0            on success
2767  * \retval negative     negated errno on failure
2768  */
2769 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2770                         __u32 keylen, void *key, __u32 *vallen, void *val)
2771 {
2772         struct obd_device       *obd;
2773         struct lmv_obd          *lmv;
2774         int                      rc = 0;
2775         ENTRY;
2776
2777         obd = class_exp2obd(exp);
2778         if (obd == NULL) {
2779                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2780                        exp->exp_handle.h_cookie);
2781                 RETURN(-EINVAL);
2782         }
2783
2784         lmv = &obd->u.lmv;
2785         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2786                 int i;
2787
2788                 rc = lmv_check_connect(obd);
2789                 if (rc)
2790                         RETURN(rc);
2791
2792                 LASSERT(*vallen == sizeof(__u32));
2793                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2794                         struct lmv_tgt_desc *tgt = lmv->tgts[i];
2795                         /*
2796                          * All tgts should be connected when this gets called.
2797                          */
2798                         if (tgt == NULL || tgt->ltd_exp == NULL)
2799                                 continue;
2800
2801                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2802                                           vallen, val))
2803                                 RETURN(0);
2804                 }
2805                 RETURN(-EINVAL);
2806         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2807                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2808                    KEY_IS(KEY_CONN_DATA)) {
2809                 rc = lmv_check_connect(obd);
2810                 if (rc)
2811                         RETURN(rc);
2812
2813                 /*
2814                  * Forwarding this request to first MDS, it should know LOV
2815                  * desc.
2816                  */
2817                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2818                                   vallen, val);
2819                 if (!rc && KEY_IS(KEY_CONN_DATA))
2820                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2821                 RETURN(rc);
2822         } else if (KEY_IS(KEY_TGT_COUNT)) {
2823                 *((int *)val) = lmv->desc.ld_tgt_count;
2824                 RETURN(0);
2825         }
2826
2827         CDEBUG(D_IOCTL, "Invalid key\n");
2828         RETURN(-EINVAL);
2829 }
2830
2831 /**
2832  * Asynchronously set by key a value associated with a LMV device.
2833  *
2834  * Dispatch request to lower-layer devices as needed.
2835  *
2836  * \param[in] env       execution environment for this thread
2837  * \param[in] exp       export for the LMV device
2838  * \param[in] keylen    length of key identifier
2839  * \param[in] key       identifier of key to store value for
2840  * \param[in] vallen    size of value to store
2841  * \param[in] val       pointer to data to be stored
2842  * \param[in] set       optional list of related ptlrpc requests
2843  *
2844  * \retval 0            on success
2845  * \retval negative     negated errno on failure
2846  */
2847 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2848                         __u32 keylen, void *key, __u32 vallen, void *val,
2849                         struct ptlrpc_request_set *set)
2850 {
2851         struct lmv_tgt_desc     *tgt = NULL;
2852         struct obd_device       *obd;
2853         struct lmv_obd          *lmv;
2854         int rc = 0;
2855         ENTRY;
2856
2857         obd = class_exp2obd(exp);
2858         if (obd == NULL) {
2859                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2860                        exp->exp_handle.h_cookie);
2861                 RETURN(-EINVAL);
2862         }
2863         lmv = &obd->u.lmv;
2864
2865         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
2866             KEY_IS(KEY_DEFAULT_EASIZE)) {
2867                 int i, err = 0;
2868
2869                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2870                         tgt = lmv->tgts[i];
2871
2872                         if (tgt == NULL || tgt->ltd_exp == NULL)
2873                                 continue;
2874
2875                         err = obd_set_info_async(env, tgt->ltd_exp,
2876                                                  keylen, key, vallen, val, set);
2877                         if (err && rc == 0)
2878                                 rc = err;
2879                 }
2880
2881                 RETURN(rc);
2882         }
2883
2884         RETURN(-EINVAL);
2885 }
2886
2887 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2888                             const struct lmv_mds_md_v1 *lmm1)
2889 {
2890         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
2891         int             stripe_count;
2892         int             cplen;
2893         int             i;
2894         int             rc = 0;
2895         ENTRY;
2896
2897         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2898         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2899         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2900         if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
2901                 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
2902         else
2903                 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2904         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2905         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2906                         sizeof(lsm->lsm_md_pool_name));
2907
2908         if (cplen >= sizeof(lsm->lsm_md_pool_name))
2909                 RETURN(-E2BIG);
2910
2911         CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2912                "layout_version %d\n", lsm->lsm_md_stripe_count,
2913                lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2914                lsm->lsm_md_layout_version);
2915
2916         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2917         for (i = 0; i < stripe_count; i++) {
2918                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2919                               &lmm1->lmv_stripe_fids[i]);
2920                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2921                                     &lsm->lsm_md_oinfo[i].lmo_mds);
2922                 if (rc != 0)
2923                         RETURN(rc);
2924                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
2925                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
2926         }
2927
2928         RETURN(rc);
2929 }
2930
2931 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
2932                         const union lmv_mds_md *lmm, size_t lmm_size)
2933 {
2934         struct lmv_stripe_md     *lsm;
2935         int                      lsm_size;
2936         int                      rc;
2937         bool                     allocated = false;
2938         ENTRY;
2939
2940         LASSERT(lsmp != NULL);
2941
2942         lsm = *lsmp;
2943         /* Free memmd */
2944         if (lsm != NULL && lmm == NULL) {
2945                 int i;
2946                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
2947                         /* For migrating inode, the master stripe and master
2948                          * object will be the same, so do not need iput, see
2949                          * ll_update_lsm_md */
2950                         if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
2951                               i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
2952                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
2953                 }
2954                 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
2955                 OBD_FREE(lsm, lsm_size);
2956                 *lsmp = NULL;
2957                 RETURN(0);
2958         }
2959
2960         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
2961                 RETURN(-EPERM);
2962
2963         /* Unpack memmd */
2964         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
2965             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
2966                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
2967                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
2968                        -EIO);
2969                 RETURN(-EIO);
2970         }
2971
2972         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
2973                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2974         else
2975                 /**
2976                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
2977                  * stripecount should be 0 then.
2978                  */
2979                 lsm_size = lmv_stripe_md_size(0);
2980
2981         lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2982         if (lsm == NULL) {
2983                 OBD_ALLOC(lsm, lsm_size);
2984                 if (lsm == NULL)
2985                         RETURN(-ENOMEM);
2986                 allocated = true;
2987                 *lsmp = lsm;
2988         }
2989
2990         switch (le32_to_cpu(lmm->lmv_magic)) {
2991         case LMV_MAGIC_V1:
2992                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
2993                 break;
2994         default:
2995                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
2996                        le32_to_cpu(lmm->lmv_magic));
2997                 rc = -EINVAL;
2998                 break;
2999         }
3000
3001         if (rc != 0 && allocated) {
3002                 OBD_FREE(lsm, lsm_size);
3003                 *lsmp = NULL;
3004                 lsm_size = rc;
3005         }
3006         RETURN(lsm_size);
3007 }
3008
3009 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3010 {
3011         lmv_unpackmd(NULL, &lsm, NULL, 0);
3012 }
3013 EXPORT_SYMBOL(lmv_free_memmd);
3014
3015 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3016                              union ldlm_policy_data *policy,
3017                              enum ldlm_mode mode, enum ldlm_cancel_flags flags,
3018                              void *opaque)
3019 {
3020         struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3021         int rc = 0;
3022         __u32 i;
3023         ENTRY;
3024
3025         LASSERT(fid != NULL);
3026
3027         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3028                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3029                 int err;
3030
3031                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3032                         continue;
3033
3034                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3035                                        opaque);
3036                 if (!rc)
3037                         rc = err;
3038         }
3039         RETURN(rc);
3040 }
3041
3042 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3043                       __u64 *bits)
3044 {
3045         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3046         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3047         int                      rc;
3048         ENTRY;
3049
3050         if (tgt == NULL || tgt->ltd_exp == NULL)
3051                 RETURN(-EINVAL);
3052         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3053         RETURN(rc);
3054 }
3055
3056 enum ldlm_mode lmv_lock_match(struct obd_export *exp, __u64 flags,
3057                               const struct lu_fid *fid, enum ldlm_type type,
3058                               union ldlm_policy_data *policy,
3059                               enum ldlm_mode mode, struct lustre_handle *lockh)
3060 {
3061         struct obd_device       *obd = exp->exp_obd;
3062         struct lmv_obd          *lmv = &obd->u.lmv;
3063         enum ldlm_mode          rc;
3064         int                     tgt;
3065         int                     i;
3066         ENTRY;
3067
3068         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3069
3070         /*
3071          * With DNE every object can have two locks in different namespaces:
3072          * lookup lock in space of MDT storing direntry and update/open lock in
3073          * space of MDT storing inode.  Try the MDT that the FID maps to first,
3074          * since this can be easily found, and only try others if that fails.
3075          */
3076         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3077              i < lmv->desc.ld_tgt_count;
3078              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3079                 if (tgt < 0) {
3080                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3081                                obd->obd_name, PFID(fid), tgt);
3082                         tgt = 0;
3083                 }
3084
3085                 if (lmv->tgts[tgt] == NULL ||
3086                     lmv->tgts[tgt]->ltd_exp == NULL ||
3087                     lmv->tgts[tgt]->ltd_active == 0)
3088                         continue;
3089
3090                 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3091                                    type, policy, mode, lockh);
3092                 if (rc)
3093                         RETURN(rc);
3094         }
3095
3096         RETURN(0);
3097 }
3098
3099 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3100                       struct obd_export *dt_exp, struct obd_export *md_exp,
3101                       struct lustre_md *md)
3102 {
3103         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3104         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3105
3106         if (tgt == NULL || tgt->ltd_exp == NULL)
3107                 RETURN(-EINVAL);
3108
3109         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3110 }
3111
3112 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3113 {
3114         struct obd_device       *obd = exp->exp_obd;
3115         struct lmv_obd          *lmv = &obd->u.lmv;
3116         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3117         ENTRY;
3118
3119         if (md->lmv != NULL) {
3120                 lmv_free_memmd(md->lmv);
3121                 md->lmv = NULL;
3122         }
3123         if (tgt == NULL || tgt->ltd_exp == NULL)
3124                 RETURN(-EINVAL);
3125         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3126 }
3127
3128 int lmv_set_open_replay_data(struct obd_export *exp,
3129                              struct obd_client_handle *och,
3130                              struct lookup_intent *it)
3131 {
3132         struct obd_device       *obd = exp->exp_obd;
3133         struct lmv_obd          *lmv = &obd->u.lmv;
3134         struct lmv_tgt_desc     *tgt;
3135         ENTRY;
3136
3137         tgt = lmv_find_target(lmv, &och->och_fid);
3138         if (IS_ERR(tgt))
3139                 RETURN(PTR_ERR(tgt));
3140
3141         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3142 }
3143
3144 int lmv_clear_open_replay_data(struct obd_export *exp,
3145                                struct obd_client_handle *och)
3146 {
3147         struct obd_device       *obd = exp->exp_obd;
3148         struct lmv_obd          *lmv = &obd->u.lmv;
3149         struct lmv_tgt_desc     *tgt;
3150         ENTRY;
3151
3152         tgt = lmv_find_target(lmv, &och->och_fid);
3153         if (IS_ERR(tgt))
3154                 RETURN(PTR_ERR(tgt));
3155
3156         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3157 }
3158
3159 static int lmv_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
3160                                u32 suppgid, struct ptlrpc_request **request)
3161 {
3162         struct obd_device       *obd = exp->exp_obd;
3163         struct lmv_obd          *lmv = &obd->u.lmv;
3164         struct lmv_tgt_desc     *tgt;
3165         int                      rc;
3166         ENTRY;
3167
3168         rc = lmv_check_connect(obd);
3169         if (rc)
3170                 RETURN(rc);
3171
3172         tgt = lmv_find_target(lmv, fid);
3173         if (IS_ERR(tgt))
3174                 RETURN(PTR_ERR(tgt));
3175
3176         rc = md_get_remote_perm(tgt->ltd_exp, fid, suppgid, request);
3177         RETURN(rc);
3178 }
3179
3180 int lmv_intent_getattr_async(struct obd_export *exp,
3181                              struct md_enqueue_info *minfo)
3182 {
3183         struct md_op_data       *op_data = &minfo->mi_data;
3184         struct obd_device       *obd = exp->exp_obd;
3185         struct lmv_obd          *lmv = &obd->u.lmv;
3186         struct lmv_tgt_desc     *ptgt = NULL;
3187         struct lmv_tgt_desc     *ctgt = NULL;
3188         int                      rc;
3189         ENTRY;
3190
3191         if (!fid_is_sane(&op_data->op_fid2))
3192                 RETURN(-EINVAL);
3193
3194         rc = lmv_check_connect(obd);
3195         if (rc)
3196                 RETURN(rc);
3197
3198         ptgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3199         if (IS_ERR(ptgt))
3200                 RETURN(PTR_ERR(ptgt));
3201
3202         ctgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
3203         if (IS_ERR(ctgt))
3204                 RETURN(PTR_ERR(ctgt));
3205
3206         /*
3207          * if child is on remote MDT, we need 2 async RPCs to fetch both LOOKUP
3208 &n