Whamcloud - gitweb
LU-5577 mdc: fix comparison between signed and unsigned
[fs/lustre-release.git] / lustre / lmv / lmv_obd.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2013, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  */
36
37 #define DEBUG_SUBSYSTEM S_LMV
38 #include <linux/slab.h>
39 #include <linux/module.h>
40 #include <linux/init.h>
41 #include <linux/slab.h>
42 #include <linux/pagemap.h>
43 #include <linux/mm.h>
44 #include <linux/math64.h>
45 #include <linux/seq_file.h>
46 #include <linux/namei.h>
47
48 #include <lustre/lustre_idl.h>
49 #include <obd_support.h>
50 #include <lustre_lib.h>
51 #include <lustre_net.h>
52 #include <obd_class.h>
53 #include <lustre_lmv.h>
54 #include <lprocfs_status.h>
55 #include <cl_object.h>
56 #include <lclient.h>
57 #include <lustre_fid.h>
58 #include <lustre_ioctl.h>
59 #include "lmv_internal.h"
60
61 static void lmv_activate_target(struct lmv_obd *lmv,
62                                 struct lmv_tgt_desc *tgt,
63                                 int activate)
64 {
65         if (tgt->ltd_active == activate)
66                 return;
67
68         tgt->ltd_active = activate;
69         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
70 }
71
72 /**
73  * Error codes:
74  *
75  *  -EINVAL  : UUID can't be found in the LMV's target list
76  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
77  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
78  */
79 static int lmv_set_mdc_active(struct lmv_obd *lmv,
80                               const struct obd_uuid *uuid,
81                               int activate)
82 {
83         struct lmv_tgt_desc     *tgt = NULL;
84         struct obd_device       *obd;
85         __u32                    i;
86         int                      rc = 0;
87         ENTRY;
88
89         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
90                         lmv, uuid->uuid, activate);
91
92         spin_lock(&lmv->lmv_lock);
93         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
94                 tgt = lmv->tgts[i];
95                 if (tgt == NULL || tgt->ltd_exp == NULL)
96                         continue;
97
98                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
99                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
100
101                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
102                         break;
103         }
104
105         if (i == lmv->desc.ld_tgt_count)
106                 GOTO(out_lmv_lock, rc = -EINVAL);
107
108         obd = class_exp2obd(tgt->ltd_exp);
109         if (obd == NULL)
110                 GOTO(out_lmv_lock, rc = -ENOTCONN);
111
112         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
113                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
114                obd->obd_type->typ_name, i);
115         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
116
117         if (tgt->ltd_active == activate) {
118                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
119                        activate ? "" : "in");
120                 GOTO(out_lmv_lock, rc);
121         }
122
123         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
124                activate ? "" : "in");
125         lmv_activate_target(lmv, tgt, activate);
126         EXIT;
127
128  out_lmv_lock:
129         spin_unlock(&lmv->lmv_lock);
130         return rc;
131 }
132
133 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
134 {
135         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
136         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
137
138         return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
139 }
140
141 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
142                       enum obd_notify_event ev, void *data)
143 {
144         struct obd_connect_data *conn_data;
145         struct lmv_obd          *lmv = &obd->u.lmv;
146         struct obd_uuid         *uuid;
147         int                      rc = 0;
148         ENTRY;
149
150         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
151                 CERROR("unexpected notification of %s %s!\n",
152                        watched->obd_type->typ_name,
153                        watched->obd_name);
154                 RETURN(-EINVAL);
155         }
156
157         uuid = &watched->u.cli.cl_target_uuid;
158         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
159                 /*
160                  * Set MDC as active before notifying the observer, so the
161                  * observer can use the MDC normally.
162                  */
163                 rc = lmv_set_mdc_active(lmv, uuid,
164                                         ev == OBD_NOTIFY_ACTIVE);
165                 if (rc) {
166                         CERROR("%sactivation of %s failed: %d\n",
167                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
168                                uuid->uuid, rc);
169                         RETURN(rc);
170                 }
171         } else if (ev == OBD_NOTIFY_OCD) {
172                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
173                 /*
174                  * XXX: Make sure that ocd_connect_flags from all targets are
175                  * the same. Otherwise one of MDTs runs wrong version or
176                  * something like this.  --umka
177                  */
178                 obd->obd_self_export->exp_connect_data = *conn_data;
179         }
180 #if 0
181         else if (ev == OBD_NOTIFY_DISCON) {
182                 /*
183                  * For disconnect event, flush fld cache for failout MDS case.
184                  */
185                 fld_client_flush(&lmv->lmv_fld);
186         }
187 #endif
188         /*
189          * Pass the notification up the chain.
190          */
191         if (obd->obd_observer)
192                 rc = obd_notify(obd->obd_observer, watched, ev, data);
193
194         RETURN(rc);
195 }
196
197 /**
198  * This is fake connect function. Its purpose is to initialize lmv and say
199  * caller that everything is okay. Real connection will be performed later.
200  */
201 static int lmv_connect(const struct lu_env *env,
202                        struct obd_export **exp, struct obd_device *obd,
203                        struct obd_uuid *cluuid, struct obd_connect_data *data,
204                        void *localdata)
205 {
206         struct lmv_obd        *lmv = &obd->u.lmv;
207         struct lustre_handle  conn = { 0 };
208         int                    rc = 0;
209         ENTRY;
210
211         /*
212          * We don't want to actually do the underlying connections more than
213          * once, so keep track.
214          */
215         lmv->refcount++;
216         if (lmv->refcount > 1) {
217                 *exp = NULL;
218                 RETURN(0);
219         }
220
221         rc = class_connect(&conn, obd, cluuid);
222         if (rc) {
223                 CERROR("class_connection() returned %d\n", rc);
224                 RETURN(rc);
225         }
226
227         *exp = class_conn2export(&conn);
228         class_export_get(*exp);
229
230         lmv->exp = *exp;
231         lmv->connected = 0;
232         lmv->cluuid = *cluuid;
233
234         if (data)
235                 lmv->conn_data = *data;
236
237         if (lmv->targets_proc_entry == NULL) {
238                 lmv->targets_proc_entry = lprocfs_seq_register("target_obds",
239                                                         obd->obd_proc_entry,
240                                                         NULL, NULL);
241                 if (IS_ERR(lmv->targets_proc_entry)) {
242                         CERROR("%s: cannot register "
243                                "/proc/fs/lustre/%s/%s/target_obds\n",
244                                obd->obd_name, obd->obd_type->typ_name,
245                                obd->obd_name);
246                         lmv->targets_proc_entry = NULL;
247                 }
248         }
249
250         /*
251          * All real clients should perform actual connection right away, because
252          * it is possible, that LMV will not have opportunity to connect targets
253          * and MDC stuff will be called directly, for instance while reading
254          * ../mdc/../kbytesfree procfs file, etc.
255          */
256         if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
257                 rc = lmv_check_connect(obd);
258
259         if (rc && lmv->targets_proc_entry != NULL)
260                 lprocfs_remove(&lmv->targets_proc_entry);
261         RETURN(rc);
262 }
263
264 static int lmv_init_ea_size(struct obd_export *exp,
265                             __u32 easize, __u32 def_easize,
266                             __u32 cookiesize, __u32 def_cookiesize)
267 {
268         struct obd_device       *obd = exp->exp_obd;
269         struct lmv_obd          *lmv = &obd->u.lmv;
270         __u32                    i;
271         int                      rc = 0;
272         int                      change = 0;
273         ENTRY;
274
275         if (lmv->max_easize < easize) {
276                 lmv->max_easize = easize;
277                 change = 1;
278         }
279         if (lmv->max_def_easize < def_easize) {
280                 lmv->max_def_easize = def_easize;
281                 change = 1;
282         }
283         if (lmv->max_cookiesize < cookiesize) {
284                 lmv->max_cookiesize = cookiesize;
285                 change = 1;
286         }
287         if (lmv->max_def_cookiesize < def_cookiesize) {
288                 lmv->max_def_cookiesize = def_cookiesize;
289                 change = 1;
290         }
291         if (change == 0)
292                 RETURN(0);
293
294         if (lmv->connected == 0)
295                 RETURN(0);
296
297         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
298                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
299
300                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
301                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
302                         continue;
303                 }
304
305                 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
306                                      cookiesize, def_cookiesize);
307                 if (rc) {
308                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
309                                " rc = %d\n", obd->obd_name, i, rc);
310                         break;
311                 }
312         }
313         RETURN(rc);
314 }
315
316 #define MAX_STRING_SIZE 128
317
318 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
319 {
320         struct lmv_obd          *lmv = &obd->u.lmv;
321         struct obd_uuid         *cluuid = &lmv->cluuid;
322         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
323         struct obd_device       *mdc_obd;
324         struct obd_export       *mdc_exp;
325         struct lu_fld_target     target;
326         int                      rc;
327         ENTRY;
328
329         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
330                                         &obd->obd_uuid);
331         if (!mdc_obd) {
332                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
333                 RETURN(-EINVAL);
334         }
335
336         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
337                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
338                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
339                 cluuid->uuid);
340
341         if (!mdc_obd->obd_set_up) {
342                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
343                 RETURN(-EINVAL);
344         }
345
346         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
347                          &lmv->conn_data, NULL);
348         if (rc) {
349                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
350                 RETURN(rc);
351         }
352
353         /*
354          * Init fid sequence client for this mdc and add new fld target.
355          */
356         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
357         if (rc)
358                 RETURN(rc);
359
360         target.ft_srv = NULL;
361         target.ft_exp = mdc_exp;
362         target.ft_idx = tgt->ltd_idx;
363
364         fld_client_add_target(&lmv->lmv_fld, &target);
365
366         rc = obd_register_observer(mdc_obd, obd);
367         if (rc) {
368                 obd_disconnect(mdc_exp);
369                 CERROR("target %s register_observer error %d\n",
370                        tgt->ltd_uuid.uuid, rc);
371                 RETURN(rc);
372         }
373
374         if (obd->obd_observer) {
375                 /*
376                  * Tell the observer about the new target.
377                  */
378                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
379                                 OBD_NOTIFY_ACTIVE,
380                                 (void *)(tgt - lmv->tgts[0]));
381                 if (rc) {
382                         obd_disconnect(mdc_exp);
383                         RETURN(rc);
384                 }
385         }
386
387         tgt->ltd_active = 1;
388         tgt->ltd_exp = mdc_exp;
389         lmv->desc.ld_active_tgt_count++;
390
391         md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize,
392                         lmv->max_cookiesize, lmv->max_def_cookiesize);
393
394         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
395                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
396                 atomic_read(&obd->obd_refcount));
397
398         if (lmv->targets_proc_entry != NULL) {
399                 struct proc_dir_entry *mdc_symlink;
400
401                 LASSERT(mdc_obd->obd_type != NULL);
402                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
403                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
404                                                   lmv->targets_proc_entry,
405                                                   "../../../%s/%s",
406                                                   mdc_obd->obd_type->typ_name,
407                                                   mdc_obd->obd_name);
408                 if (mdc_symlink == NULL) {
409                         CERROR("cannot register LMV target "
410                                "/proc/fs/lustre/%s/%s/target_obds/%s\n",
411                                obd->obd_type->typ_name, obd->obd_name,
412                                mdc_obd->obd_name);
413                 }
414         }
415         RETURN(0);
416 }
417
418 static void lmv_del_target(struct lmv_obd *lmv, int index)
419 {
420         if (lmv->tgts[index] == NULL)
421                 return;
422
423         OBD_FREE_PTR(lmv->tgts[index]);
424         lmv->tgts[index] = NULL;
425         return;
426 }
427
428 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
429                            __u32 index, int gen)
430 {
431         struct lmv_obd      *lmv = &obd->u.lmv;
432         struct lmv_tgt_desc *tgt;
433         int                  orig_tgt_count = 0;
434         int                  rc = 0;
435         ENTRY;
436
437         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
438
439         lmv_init_lock(lmv);
440
441         if (lmv->desc.ld_tgt_count == 0) {
442                 struct obd_device *mdc_obd;
443
444                 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
445                                                 &obd->obd_uuid);
446                 if (!mdc_obd) {
447                         lmv_init_unlock(lmv);
448                         CERROR("%s: Target %s not attached: rc = %d\n",
449                                obd->obd_name, uuidp->uuid, -EINVAL);
450                         RETURN(-EINVAL);
451                 }
452         }
453
454         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
455                 tgt = lmv->tgts[index];
456                 CERROR("%s: UUID %s already assigned at LOV target index %d:"
457                        " rc = %d\n", obd->obd_name,
458                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
459                 lmv_init_unlock(lmv);
460                 RETURN(-EEXIST);
461         }
462
463         if (index >= lmv->tgts_size) {
464                 /* We need to reallocate the lmv target array. */
465                 struct lmv_tgt_desc **newtgts, **old = NULL;
466                 __u32 newsize = 1;
467                 __u32 oldsize = 0;
468
469                 while (newsize < index + 1)
470                         newsize = newsize << 1;
471                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
472                 if (newtgts == NULL) {
473                         lmv_init_unlock(lmv);
474                         RETURN(-ENOMEM);
475                 }
476
477                 if (lmv->tgts_size) {
478                         memcpy(newtgts, lmv->tgts,
479                                sizeof(*newtgts) * lmv->tgts_size);
480                         old = lmv->tgts;
481                         oldsize = lmv->tgts_size;
482                 }
483
484                 lmv->tgts = newtgts;
485                 lmv->tgts_size = newsize;
486                 smp_rmb();
487                 if (old)
488                         OBD_FREE(old, sizeof(*old) * oldsize);
489
490                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
491                        lmv->tgts_size);
492         }
493
494         OBD_ALLOC_PTR(tgt);
495         if (!tgt) {
496                 lmv_init_unlock(lmv);
497                 RETURN(-ENOMEM);
498         }
499
500         mutex_init(&tgt->ltd_fid_mutex);
501         tgt->ltd_idx = index;
502         tgt->ltd_uuid = *uuidp;
503         tgt->ltd_active = 0;
504         lmv->tgts[index] = tgt;
505         if (index >= lmv->desc.ld_tgt_count) {
506                 orig_tgt_count = lmv->desc.ld_tgt_count;
507                 lmv->desc.ld_tgt_count = index + 1;
508         }
509
510         if (lmv->connected) {
511                 rc = lmv_connect_mdc(obd, tgt);
512                 if (rc != 0) {
513                         spin_lock(&lmv->lmv_lock);
514                         if (lmv->desc.ld_tgt_count == index + 1)
515                                 lmv->desc.ld_tgt_count = orig_tgt_count;
516                         memset(tgt, 0, sizeof(*tgt));
517                         spin_unlock(&lmv->lmv_lock);
518                 } else {
519                         int easize = sizeof(struct lmv_stripe_md) +
520                                 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
521                         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
522                 }
523         }
524
525         lmv_init_unlock(lmv);
526         RETURN(rc);
527 }
528
529 int lmv_check_connect(struct obd_device *obd)
530 {
531         struct lmv_obd          *lmv = &obd->u.lmv;
532         struct lmv_tgt_desc     *tgt;
533         __u32                    i;
534         int                      rc;
535         int                      easize;
536         ENTRY;
537
538         if (lmv->connected)
539                 RETURN(0);
540
541         lmv_init_lock(lmv);
542         if (lmv->connected) {
543                 lmv_init_unlock(lmv);
544                 RETURN(0);
545         }
546
547         if (lmv->desc.ld_tgt_count == 0) {
548                 lmv_init_unlock(lmv);
549                 CERROR("%s: no targets configured.\n", obd->obd_name);
550                 RETURN(-EINVAL);
551         }
552
553         LASSERT(lmv->tgts != NULL);
554
555         if (lmv->tgts[0] == NULL) {
556                 lmv_init_unlock(lmv);
557                 CERROR("%s: no target configured for index 0.\n",
558                        obd->obd_name);
559                 RETURN(-EINVAL);
560         }
561
562         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
563                lmv->cluuid.uuid, obd->obd_name);
564
565         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
566                 tgt = lmv->tgts[i];
567                 if (tgt == NULL)
568                         continue;
569                 rc = lmv_connect_mdc(obd, tgt);
570                 if (rc)
571                         GOTO(out_disc, rc);
572         }
573
574         class_export_put(lmv->exp);
575         lmv->connected = 1;
576         easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
577         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
578         lmv_init_unlock(lmv);
579         RETURN(0);
580
581  out_disc:
582         while (i-- > 0) {
583                 int rc2;
584                 tgt = lmv->tgts[i];
585                 if (tgt == NULL)
586                         continue;
587                 tgt->ltd_active = 0;
588                 if (tgt->ltd_exp) {
589                         --lmv->desc.ld_active_tgt_count;
590                         rc2 = obd_disconnect(tgt->ltd_exp);
591                         if (rc2) {
592                                 CERROR("LMV target %s disconnect on "
593                                        "MDC idx %d: error %d\n",
594                                        tgt->ltd_uuid.uuid, i, rc2);
595                         }
596                 }
597         }
598         class_disconnect(lmv->exp);
599         lmv_init_unlock(lmv);
600         RETURN(rc);
601 }
602
603 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
604 {
605         struct lmv_obd         *lmv = &obd->u.lmv;
606         struct obd_device      *mdc_obd;
607         int                     rc;
608         ENTRY;
609
610         LASSERT(tgt != NULL);
611         LASSERT(obd != NULL);
612
613         mdc_obd = class_exp2obd(tgt->ltd_exp);
614
615         if (mdc_obd) {
616                 mdc_obd->obd_force = obd->obd_force;
617                 mdc_obd->obd_fail = obd->obd_fail;
618                 mdc_obd->obd_no_recov = obd->obd_no_recov;
619         }
620
621         if (lmv->targets_proc_entry != NULL)
622                 lprocfs_remove_proc_entry(mdc_obd->obd_name,
623                                           lmv->targets_proc_entry);
624
625         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
626         if (rc)
627                 CERROR("Can't finanize fids factory\n");
628
629         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
630                tgt->ltd_exp->exp_obd->obd_name,
631                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
632
633         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
634         rc = obd_disconnect(tgt->ltd_exp);
635         if (rc) {
636                 if (tgt->ltd_active) {
637                         CERROR("Target %s disconnect error %d\n",
638                                tgt->ltd_uuid.uuid, rc);
639                 }
640         }
641
642         lmv_activate_target(lmv, tgt, 0);
643         tgt->ltd_exp = NULL;
644         RETURN(0);
645 }
646
647 static int lmv_disconnect(struct obd_export *exp)
648 {
649         struct obd_device       *obd = class_exp2obd(exp);
650         struct lmv_obd          *lmv = &obd->u.lmv;
651         int                      rc;
652         __u32                    i;
653         ENTRY;
654
655         if (!lmv->tgts)
656                 goto out_local;
657
658         /*
659          * Only disconnect the underlying layers on the final disconnect.
660          */
661         lmv->refcount--;
662         if (lmv->refcount != 0)
663                 goto out_local;
664
665         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
666                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
667                         continue;
668
669                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
670         }
671
672         if (lmv->targets_proc_entry != NULL)
673                 lprocfs_remove(&lmv->targets_proc_entry);
674         else
675                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
676                        obd->obd_type->typ_name, obd->obd_name);
677
678 out_local:
679         /*
680          * This is the case when no real connection is established by
681          * lmv_check_connect().
682          */
683         if (!lmv->connected)
684                 class_export_put(exp);
685         rc = class_disconnect(exp);
686         if (lmv->refcount == 0)
687                 lmv->connected = 0;
688         RETURN(rc);
689 }
690
691 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
692 {
693         struct obd_device       *obddev = class_exp2obd(exp);
694         struct lmv_obd          *lmv = &obddev->u.lmv;
695         struct getinfo_fid2path *gf;
696         struct lmv_tgt_desc     *tgt;
697         struct getinfo_fid2path *remote_gf = NULL;
698         int                     remote_gf_size = 0;
699         int                     rc;
700
701         gf = (struct getinfo_fid2path *)karg;
702         tgt = lmv_find_target(lmv, &gf->gf_fid);
703         if (IS_ERR(tgt))
704                 RETURN(PTR_ERR(tgt));
705
706 repeat_fid2path:
707         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
708         if (rc != 0 && rc != -EREMOTE)
709                 GOTO(out_fid2path, rc);
710
711         /* If remote_gf != NULL, it means just building the
712          * path on the remote MDT, copy this path segement to gf */
713         if (remote_gf != NULL) {
714                 struct getinfo_fid2path *ori_gf;
715                 char *ptr;
716
717                 ori_gf = (struct getinfo_fid2path *)karg;
718                 if (strlen(ori_gf->gf_path) +
719                     strlen(gf->gf_path) > ori_gf->gf_pathlen)
720                         GOTO(out_fid2path, rc = -EOVERFLOW);
721
722                 ptr = ori_gf->gf_path;
723
724                 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
725                         strlen(ori_gf->gf_path));
726
727                 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
728                 ptr += strlen(gf->gf_path);
729                 *ptr = '/';
730         }
731
732         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
733                tgt->ltd_exp->exp_obd->obd_name,
734                gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
735                gf->gf_linkno);
736
737         if (rc == 0)
738                 GOTO(out_fid2path, rc);
739
740         /* sigh, has to go to another MDT to do path building further */
741         if (remote_gf == NULL) {
742                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
743                 OBD_ALLOC(remote_gf, remote_gf_size);
744                 if (remote_gf == NULL)
745                         GOTO(out_fid2path, rc = -ENOMEM);
746                 remote_gf->gf_pathlen = PATH_MAX;
747         }
748
749         if (!fid_is_sane(&gf->gf_fid)) {
750                 CERROR("%s: invalid FID "DFID": rc = %d\n",
751                        tgt->ltd_exp->exp_obd->obd_name,
752                        PFID(&gf->gf_fid), -EINVAL);
753                 GOTO(out_fid2path, rc = -EINVAL);
754         }
755
756         tgt = lmv_find_target(lmv, &gf->gf_fid);
757         if (IS_ERR(tgt))
758                 GOTO(out_fid2path, rc = -EINVAL);
759
760         remote_gf->gf_fid = gf->gf_fid;
761         remote_gf->gf_recno = -1;
762         remote_gf->gf_linkno = -1;
763         memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
764         gf = remote_gf;
765         goto repeat_fid2path;
766
767 out_fid2path:
768         if (remote_gf != NULL)
769                 OBD_FREE(remote_gf, remote_gf_size);
770         RETURN(rc);
771 }
772
773 static int lmv_hsm_req_count(struct lmv_obd *lmv,
774                              const struct hsm_user_request *hur,
775                              const struct lmv_tgt_desc *tgt_mds)
776 {
777         __u32                    i;
778         int                      nr = 0;
779         struct lmv_tgt_desc     *curr_tgt;
780
781         /* count how many requests must be sent to the given target */
782         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
783                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
784                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
785                         nr++;
786         }
787         return nr;
788 }
789
790 static void lmv_hsm_req_build(struct lmv_obd *lmv,
791                               struct hsm_user_request *hur_in,
792                               const struct lmv_tgt_desc *tgt_mds,
793                               struct hsm_user_request *hur_out)
794 {
795         __u32                    i, nr_out;
796         struct lmv_tgt_desc     *curr_tgt;
797
798         /* build the hsm_user_request for the given target */
799         hur_out->hur_request = hur_in->hur_request;
800         nr_out = 0;
801         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
802                 curr_tgt = lmv_find_target(lmv,
803                                            &hur_in->hur_user_item[i].hui_fid);
804                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
805                         hur_out->hur_user_item[nr_out] =
806                                                 hur_in->hur_user_item[i];
807                         nr_out++;
808                 }
809         }
810         hur_out->hur_request.hr_itemcount = nr_out;
811         memcpy(hur_data(hur_out), hur_data(hur_in),
812                hur_in->hur_request.hr_data_len);
813 }
814
815 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
816                                  struct lustre_kernelcomm *lk, void *uarg)
817 {
818         __u32                    i;
819         int                      rc;
820         struct kkuc_ct_data     *kcd = NULL;
821         ENTRY;
822
823         /* unregister request (call from llapi_hsm_copytool_fini) */
824         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
825                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
826
827                 if (tgt == NULL || tgt->ltd_exp == NULL)
828                         continue;
829                 /* best effort: try to clean as much as possible
830                  * (continue on error) */
831                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
832         }
833
834         /* Whatever the result, remove copytool from kuc groups.
835          * Unreached coordinators will get EPIPE on next requests
836          * and will unregister automatically.
837          */
838         rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
839         if (kcd != NULL)
840                 OBD_FREE_PTR(kcd);
841
842         RETURN(rc);
843 }
844
845 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
846                                struct lustre_kernelcomm *lk, void *uarg)
847 {
848         struct file             *filp;
849         __u32                    i, j;
850         int                      err, rc;
851         bool                     any_set = false;
852         struct kkuc_ct_data     *kcd;
853         ENTRY;
854
855         /* All or nothing: try to register to all MDS.
856          * In case of failure, unregister from previous MDS,
857          * except if it because of inactive target. */
858         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
859                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
860
861                 if (tgt == NULL || tgt->ltd_exp == NULL)
862                         continue;
863                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
864                 if (err) {
865                         if (tgt->ltd_active) {
866                                 /* permanent error */
867                                 CERROR("%s: iocontrol MDC %s on MDT"
868                                        " idx %d cmd %x: err = %d\n",
869                                        class_exp2obd(lmv->exp)->obd_name,
870                                        tgt->ltd_uuid.uuid, i, cmd, err);
871                                 rc = err;
872                                 lk->lk_flags |= LK_FLG_STOP;
873                                 /* unregister from previous MDS */
874                                 for (j = 0; j < i; j++) {
875                                         tgt = lmv->tgts[j];
876                                         if (tgt == NULL || tgt->ltd_exp == NULL)
877                                                 continue;
878                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
879                                                       lk, uarg);
880                                 }
881                                 RETURN(rc);
882                         }
883                         /* else: transient error.
884                          * kuc will register to the missing MDT
885                          * when it is back */
886                 } else {
887                         any_set = true;
888                 }
889         }
890
891         if (!any_set)
892                 /* no registration done: return error */
893                 RETURN(-ENOTCONN);
894
895         /* at least one registration done, with no failure */
896         filp = fget(lk->lk_wfd);
897         if (filp == NULL)
898                 RETURN(-EBADF);
899
900         OBD_ALLOC_PTR(kcd);
901         if (kcd == NULL) {
902                 fput(filp);
903                 RETURN(-ENOMEM);
904         }
905         kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
906         kcd->kcd_uuid = lmv->cluuid;
907         kcd->kcd_archive = lk->lk_data;
908
909         rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
910         if (rc != 0) {
911                 if (filp != NULL)
912                         fput(filp);
913                 OBD_FREE_PTR(kcd);
914         }
915
916         RETURN(rc);
917 }
918
919
920
921
922 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
923                          int len, void *karg, void *uarg)
924 {
925         struct obd_device       *obddev = class_exp2obd(exp);
926         struct lmv_obd          *lmv = &obddev->u.lmv;
927         struct lmv_tgt_desc     *tgt = NULL;
928         __u32                    i = 0;
929         int                      rc = 0;
930         int                      set = 0;
931         __u32                    count = lmv->desc.ld_tgt_count;
932         ENTRY;
933
934         if (count == 0)
935                 RETURN(-ENOTTY);
936
937         switch (cmd) {
938         case IOC_OBD_STATFS: {
939                 struct obd_ioctl_data *data = karg;
940                 struct obd_device *mdc_obd;
941                 struct obd_statfs stat_buf = {0};
942                 __u32 index;
943
944                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
945                 if ((index >= count))
946                         RETURN(-ENODEV);
947
948                 tgt = lmv->tgts[index];
949                 if (tgt == NULL || !tgt->ltd_active)
950                         RETURN(-ENODATA);
951
952                 mdc_obd = class_exp2obd(tgt->ltd_exp);
953                 if (!mdc_obd)
954                         RETURN(-EINVAL);
955
956                 /* copy UUID */
957                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
958                                  min((int) data->ioc_plen2,
959                                      (int) sizeof(struct obd_uuid))))
960                         RETURN(-EFAULT);
961
962                 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
963                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
964                                 0);
965                 if (rc)
966                         RETURN(rc);
967                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
968                                  min((int) data->ioc_plen1,
969                                      (int) sizeof(stat_buf))))
970                         RETURN(-EFAULT);
971                 break;
972         }
973         case OBD_IOC_QUOTACTL: {
974                 struct if_quotactl *qctl = karg;
975                 struct obd_quotactl *oqctl;
976
977                 if (qctl->qc_valid == QC_MDTIDX) {
978                         if (count <= qctl->qc_idx)
979                                 RETURN(-EINVAL);
980
981                         tgt = lmv->tgts[qctl->qc_idx];
982                         if (tgt == NULL || tgt->ltd_exp == NULL)
983                                 RETURN(-EINVAL);
984                 } else if (qctl->qc_valid == QC_UUID) {
985                         for (i = 0; i < count; i++) {
986                                 tgt = lmv->tgts[i];
987                                 if (tgt == NULL)
988                                         continue;
989                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
990                                                      &qctl->obd_uuid))
991                                         continue;
992
993                                 if (tgt->ltd_exp == NULL)
994                                         RETURN(-EINVAL);
995
996                                 break;
997                         }
998                 } else {
999                         RETURN(-EINVAL);
1000                 }
1001
1002                 if (i >= count)
1003                         RETURN(-EAGAIN);
1004
1005                 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1006                 OBD_ALLOC_PTR(oqctl);
1007                 if (!oqctl)
1008                         RETURN(-ENOMEM);
1009
1010                 QCTL_COPY(oqctl, qctl);
1011                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1012                 if (rc == 0) {
1013                         QCTL_COPY(qctl, oqctl);
1014                         qctl->qc_valid = QC_MDTIDX;
1015                         qctl->obd_uuid = tgt->ltd_uuid;
1016                 }
1017                 OBD_FREE_PTR(oqctl);
1018                 break;
1019         }
1020         case OBD_IOC_CHANGELOG_SEND:
1021         case OBD_IOC_CHANGELOG_CLEAR: {
1022                 struct ioc_changelog *icc = karg;
1023
1024                 if (icc->icc_mdtindex >= count)
1025                         RETURN(-ENODEV);
1026
1027                 tgt = lmv->tgts[icc->icc_mdtindex];
1028                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1029                         RETURN(-ENODEV);
1030                 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1031                 break;
1032         }
1033         case LL_IOC_GET_CONNECT_FLAGS: {
1034                 tgt = lmv->tgts[0];
1035                 if (tgt == NULL || tgt->ltd_exp == NULL)
1036                         RETURN(-ENODATA);
1037                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1038                 break;
1039         }
1040         case LL_IOC_FID2MDTIDX: {
1041                 struct lu_fid *fid = karg;
1042                 int             mdt_index;
1043
1044                 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1045                 if (rc != 0)
1046                         RETURN(rc);
1047
1048                 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1049                  * point to user space memory for FID2MDTIDX. */
1050                 *(__u32 *)uarg = mdt_index;
1051                 break;
1052         }
1053         case OBD_IOC_FID2PATH: {
1054                 rc = lmv_fid2path(exp, len, karg, uarg);
1055                 break;
1056         }
1057         case LL_IOC_HSM_STATE_GET:
1058         case LL_IOC_HSM_STATE_SET:
1059         case LL_IOC_HSM_ACTION: {
1060                 struct md_op_data       *op_data = karg;
1061
1062                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1063                 if (IS_ERR(tgt))
1064                         RETURN(PTR_ERR(tgt));
1065
1066                 if (tgt->ltd_exp == NULL)
1067                         RETURN(-EINVAL);
1068
1069                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1070                 break;
1071         }
1072         case LL_IOC_HSM_PROGRESS: {
1073                 const struct hsm_progress_kernel *hpk = karg;
1074
1075                 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1076                 if (IS_ERR(tgt))
1077                         RETURN(PTR_ERR(tgt));
1078                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1079                 break;
1080         }
1081         case LL_IOC_HSM_REQUEST: {
1082                 struct hsm_user_request *hur = karg;
1083                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1084
1085                 if (reqcount == 0)
1086                         RETURN(0);
1087
1088                 /* if the request is about a single fid
1089                  * or if there is a single MDS, no need to split
1090                  * the request. */
1091                 if (reqcount == 1 || count == 1) {
1092                         tgt = lmv_find_target(lmv,
1093                                               &hur->hur_user_item[0].hui_fid);
1094                         if (IS_ERR(tgt))
1095                                 RETURN(PTR_ERR(tgt));
1096                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1097                 } else {
1098                         /* split fid list to their respective MDS */
1099                         for (i = 0; i < count; i++) {
1100                                 unsigned int            nr, reqlen;
1101                                 int                     rc1;
1102                                 struct hsm_user_request *req;
1103
1104                                 tgt = lmv->tgts[i];
1105                                 if (tgt == NULL || tgt->ltd_exp == NULL)
1106                                         continue;
1107
1108                                 nr = lmv_hsm_req_count(lmv, hur, tgt);
1109                                 if (nr == 0) /* nothing for this MDS */
1110                                         continue;
1111
1112                                 /* build a request with fids for this MDS */
1113                                 reqlen = offsetof(typeof(*hur),
1114                                                   hur_user_item[nr])
1115                                                 + hur->hur_request.hr_data_len;
1116                                 OBD_ALLOC_LARGE(req, reqlen);
1117                                 if (req == NULL)
1118                                         RETURN(-ENOMEM);
1119
1120                                 lmv_hsm_req_build(lmv, hur, tgt, req);
1121
1122                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1123                                                     req, uarg);
1124                                 if (rc1 != 0 && rc == 0)
1125                                         rc = rc1;
1126                                 OBD_FREE_LARGE(req, reqlen);
1127                         }
1128                 }
1129                 break;
1130         }
1131         case LL_IOC_LOV_SWAP_LAYOUTS: {
1132                 struct md_op_data       *op_data = karg;
1133                 struct lmv_tgt_desc     *tgt1, *tgt2;
1134
1135                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1136                 if (IS_ERR(tgt1))
1137                         RETURN(PTR_ERR(tgt1));
1138
1139                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1140                 if (IS_ERR(tgt2))
1141                         RETURN(PTR_ERR(tgt2));
1142
1143                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1144                         RETURN(-EINVAL);
1145
1146                 /* only files on same MDT can have their layouts swapped */
1147                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1148                         RETURN(-EPERM);
1149
1150                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1151                 break;
1152         }
1153         case LL_IOC_HSM_CT_START: {
1154                 struct lustre_kernelcomm *lk = karg;
1155                 if (lk->lk_flags & LK_FLG_STOP)
1156                         rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1157                 else
1158                         rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1159                 break;
1160         }
1161         default:
1162                 for (i = 0; i < count; i++) {
1163                         struct obd_device *mdc_obd;
1164                         int err;
1165
1166                         tgt = lmv->tgts[i];
1167                         if (tgt == NULL || tgt->ltd_exp == NULL)
1168                                 continue;
1169                         /* ll_umount_begin() sets force flag but for lmv, not
1170                          * mdc. Let's pass it through */
1171                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1172                         mdc_obd->obd_force = obddev->obd_force;
1173                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1174                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1175                                 RETURN(err);
1176                         } else if (err) {
1177                                 if (tgt->ltd_active) {
1178                                         CERROR("error: iocontrol MDC %s on MDT"
1179                                                " idx %d cmd %x: err = %d\n",
1180                                                tgt->ltd_uuid.uuid, i, cmd, err);
1181                                         if (!rc)
1182                                                 rc = err;
1183                                 }
1184                         } else
1185                                 set = 1;
1186                 }
1187                 if (!set && !rc)
1188                         rc = -EIO;
1189         }
1190         RETURN(rc);
1191 }
1192
1193 #if 0
1194 static int lmv_all_chars_policy(int count, const char *name,
1195                                 int len)
1196 {
1197         unsigned int c = 0;
1198
1199         while (len > 0)
1200                 c += name[--len];
1201         c = c % count;
1202         return c;
1203 }
1204
1205 static int lmv_nid_policy(struct lmv_obd *lmv)
1206 {
1207         struct obd_import *imp;
1208         __u32              id;
1209
1210         /*
1211          * XXX: To get nid we assume that underlying obd device is mdc.
1212          */
1213         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1214         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1215         return id % lmv->desc.ld_tgt_count;
1216 }
1217
1218 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1219                           placement_policy_t placement)
1220 {
1221         switch (placement) {
1222         case PLACEMENT_CHAR_POLICY:
1223                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1224                                             op_data->op_name,
1225                                             op_data->op_namelen);
1226         case PLACEMENT_NID_POLICY:
1227                 return lmv_nid_policy(lmv);
1228
1229         default:
1230                 break;
1231         }
1232
1233         CERROR("Unsupported placement policy %x\n", placement);
1234         return -EINVAL;
1235 }
1236 #endif
1237
1238 /**
1239  * This is _inode_ placement policy function (not name).
1240  */
1241 static int lmv_placement_policy(struct obd_device *obd,
1242                                 struct md_op_data *op_data, u32 *mds)
1243 {
1244         struct lmv_obd          *lmv = &obd->u.lmv;
1245         ENTRY;
1246
1247         LASSERT(mds != NULL);
1248
1249         if (lmv->desc.ld_tgt_count == 1) {
1250                 *mds = 0;
1251                 RETURN(0);
1252         }
1253
1254         /**
1255          * If stripe_offset is provided during setdirstripe
1256          * (setdirstripe -i xx), xx MDS will be choosen.
1257          */
1258         if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1259                 struct lmv_user_md *lum;
1260
1261                 lum = op_data->op_data;
1262
1263                 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1264                         *mds = le32_to_cpu(lum->lum_stripe_offset);
1265                 } else {
1266                         /* -1 means default, which will be in the same MDT with
1267                          * the stripe */
1268                         *mds = op_data->op_mds;
1269                         lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1270                 }
1271         } else {
1272                 /* Allocate new fid on target according to operation type and
1273                  * parent home mds. */
1274                 *mds = op_data->op_mds;
1275         }
1276
1277         RETURN(0);
1278 }
1279
1280 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1281 {
1282         struct lmv_tgt_desc     *tgt;
1283         int                      rc;
1284         ENTRY;
1285
1286         tgt = lmv_get_target(lmv, mds, NULL);
1287         if (IS_ERR(tgt))
1288                 RETURN(PTR_ERR(tgt));
1289
1290         /*
1291          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1292          * on server that seq in new allocated fid is not yet known.
1293          */
1294         mutex_lock(&tgt->ltd_fid_mutex);
1295
1296         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1297                 GOTO(out, rc = -ENODEV);
1298
1299         /*
1300          * Asking underlying tgt layer to allocate new fid.
1301          */
1302         rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1303         if (rc > 0) {
1304                 LASSERT(fid_is_sane(fid));
1305                 rc = 0;
1306         }
1307
1308         EXIT;
1309 out:
1310         mutex_unlock(&tgt->ltd_fid_mutex);
1311         return rc;
1312 }
1313
1314 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1315                   struct lu_fid *fid, struct md_op_data *op_data)
1316 {
1317         struct obd_device     *obd = class_exp2obd(exp);
1318         struct lmv_obd        *lmv = &obd->u.lmv;
1319         u32                    mds = 0;
1320         int                    rc;
1321         ENTRY;
1322
1323         LASSERT(op_data != NULL);
1324         LASSERT(fid != NULL);
1325
1326         rc = lmv_placement_policy(obd, op_data, &mds);
1327         if (rc) {
1328                 CERROR("Can't get target for allocating fid, "
1329                        "rc %d\n", rc);
1330                 RETURN(rc);
1331         }
1332
1333         rc = __lmv_fid_alloc(lmv, fid, mds);
1334         if (rc) {
1335                 CERROR("Can't alloc new fid, rc %d\n", rc);
1336                 RETURN(rc);
1337         }
1338
1339         RETURN(rc);
1340 }
1341
1342 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1343 {
1344         struct lmv_obd  *lmv = &obd->u.lmv;
1345         struct lmv_desc *desc;
1346         int             rc;
1347         ENTRY;
1348
1349         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1350                 CERROR("LMV setup requires a descriptor\n");
1351                 RETURN(-EINVAL);
1352         }
1353
1354         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1355         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1356                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1357                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1358                 RETURN(-EINVAL);
1359         }
1360
1361         lmv->tgts_size = 32U;
1362         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1363         if (lmv->tgts == NULL)
1364                 RETURN(-ENOMEM);
1365
1366         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1367         lmv->desc.ld_tgt_count = 0;
1368         lmv->desc.ld_active_tgt_count = 0;
1369         lmv->max_cookiesize = 0;
1370         lmv->max_def_easize = 0;
1371         lmv->max_easize = 0;
1372         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1373
1374         spin_lock_init(&lmv->lmv_lock);
1375         mutex_init(&lmv->init_mutex);
1376
1377 #ifdef LPROCFS
1378         obd->obd_vars = lprocfs_lmv_obd_vars;
1379         lprocfs_obd_setup(obd);
1380         lprocfs_alloc_md_stats(obd, 0);
1381         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1382                                 0444, &lmv_proc_target_fops, obd);
1383         if (rc)
1384                 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1385                       obd->obd_name, rc);
1386 #endif
1387         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1388                              LUSTRE_CLI_FLD_HASH_DHT);
1389         if (rc) {
1390                 CERROR("Can't init FLD, err %d\n", rc);
1391                 GOTO(out, rc);
1392         }
1393
1394         RETURN(0);
1395
1396 out:
1397         return rc;
1398 }
1399
1400 static int lmv_cleanup(struct obd_device *obd)
1401 {
1402         struct lmv_obd   *lmv = &obd->u.lmv;
1403         ENTRY;
1404
1405         fld_client_fini(&lmv->lmv_fld);
1406         if (lmv->tgts != NULL) {
1407                 int i;
1408                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1409                         if (lmv->tgts[i] == NULL)
1410                                 continue;
1411                         lmv_del_target(lmv, i);
1412                 }
1413                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1414                 lmv->tgts_size = 0;
1415         }
1416         RETURN(0);
1417 }
1418
1419 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1420 {
1421         struct lustre_cfg       *lcfg = buf;
1422         struct obd_uuid         obd_uuid;
1423         int                     gen;
1424         __u32                   index;
1425         int                     rc;
1426         ENTRY;
1427
1428         switch (lcfg->lcfg_command) {
1429         case LCFG_ADD_MDC:
1430                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1431                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1432                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1433                         GOTO(out, rc = -EINVAL);
1434
1435                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1436
1437                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1438                         GOTO(out, rc = -EINVAL);
1439                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1440                         GOTO(out, rc = -EINVAL);
1441                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1442                 GOTO(out, rc);
1443         default:
1444                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1445                 GOTO(out, rc = -EINVAL);
1446         }
1447 out:
1448         RETURN(rc);
1449 }
1450
1451 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1452                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1453 {
1454         struct obd_device       *obd = class_exp2obd(exp);
1455         struct lmv_obd          *lmv = &obd->u.lmv;
1456         struct obd_statfs       *temp;
1457         int                      rc = 0;
1458         __u32                    i;
1459         ENTRY;
1460
1461         rc = lmv_check_connect(obd);
1462         if (rc)
1463                 RETURN(rc);
1464
1465         OBD_ALLOC(temp, sizeof(*temp));
1466         if (temp == NULL)
1467                 RETURN(-ENOMEM);
1468
1469         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1470                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1471                         continue;
1472
1473                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1474                                 max_age, flags);
1475                 if (rc) {
1476                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1477                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1478                                rc);
1479                         GOTO(out_free_temp, rc);
1480                 }
1481
1482                 if (i == 0) {
1483                         *osfs = *temp;
1484                         /* If the statfs is from mount, it will needs
1485                          * retrieve necessary information from MDT0.
1486                          * i.e. mount does not need the merged osfs
1487                          * from all of MDT.
1488                          * And also clients can be mounted as long as
1489                          * MDT0 is in service*/
1490                         if (flags & OBD_STATFS_FOR_MDT0)
1491                                 GOTO(out_free_temp, rc);
1492                 } else {
1493                         osfs->os_bavail += temp->os_bavail;
1494                         osfs->os_blocks += temp->os_blocks;
1495                         osfs->os_ffree += temp->os_ffree;
1496                         osfs->os_files += temp->os_files;
1497                 }
1498         }
1499
1500         EXIT;
1501 out_free_temp:
1502         OBD_FREE(temp, sizeof(*temp));
1503         return rc;
1504 }
1505
1506 static int lmv_getstatus(struct obd_export *exp,
1507                          struct lu_fid *fid,
1508                          struct obd_capa **pc)
1509 {
1510         struct obd_device    *obd = exp->exp_obd;
1511         struct lmv_obd       *lmv = &obd->u.lmv;
1512         int                   rc;
1513         ENTRY;
1514
1515         rc = lmv_check_connect(obd);
1516         if (rc)
1517                 RETURN(rc);
1518
1519         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1520         RETURN(rc);
1521 }
1522
1523 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1524                         struct obd_capa *oc, obd_valid valid, const char *name,
1525                         const char *input, int input_size, int output_size,
1526                         int flags, struct ptlrpc_request **request)
1527 {
1528         struct obd_device      *obd = exp->exp_obd;
1529         struct lmv_obd         *lmv = &obd->u.lmv;
1530         struct lmv_tgt_desc    *tgt;
1531         int                     rc;
1532         ENTRY;
1533
1534         rc = lmv_check_connect(obd);
1535         if (rc)
1536                 RETURN(rc);
1537
1538         tgt = lmv_find_target(lmv, fid);
1539         if (IS_ERR(tgt))
1540                 RETURN(PTR_ERR(tgt));
1541
1542         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1543                          input_size, output_size, flags, request);
1544
1545         RETURN(rc);
1546 }
1547
1548 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1549                         struct obd_capa *oc, obd_valid valid, const char *name,
1550                         const char *input, int input_size, int output_size,
1551                         int flags, __u32 suppgid,
1552                         struct ptlrpc_request **request)
1553 {
1554         struct obd_device      *obd = exp->exp_obd;
1555         struct lmv_obd         *lmv = &obd->u.lmv;
1556         struct lmv_tgt_desc    *tgt;
1557         int                     rc;
1558         ENTRY;
1559
1560         rc = lmv_check_connect(obd);
1561         if (rc)
1562                 RETURN(rc);
1563
1564         tgt = lmv_find_target(lmv, fid);
1565         if (IS_ERR(tgt))
1566                 RETURN(PTR_ERR(tgt));
1567
1568         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1569                          input_size, output_size, flags, suppgid,
1570                          request);
1571
1572         RETURN(rc);
1573 }
1574
1575 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1576                        struct ptlrpc_request **request)
1577 {
1578         struct obd_device       *obd = exp->exp_obd;
1579         struct lmv_obd          *lmv = &obd->u.lmv;
1580         struct lmv_tgt_desc     *tgt;
1581         int                      rc;
1582         ENTRY;
1583
1584         rc = lmv_check_connect(obd);
1585         if (rc)
1586                 RETURN(rc);
1587
1588         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1589         if (IS_ERR(tgt))
1590                 RETURN(PTR_ERR(tgt));
1591
1592         if (op_data->op_flags & MF_GET_MDT_IDX) {
1593                 op_data->op_mds = tgt->ltd_idx;
1594                 RETURN(0);
1595         }
1596
1597         rc = md_getattr(tgt->ltd_exp, op_data, request);
1598
1599         RETURN(rc);
1600 }
1601
1602 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1603 {
1604         struct obd_device   *obd = exp->exp_obd;
1605         struct lmv_obd      *lmv = &obd->u.lmv;
1606         __u32                i;
1607         int                  rc;
1608         ENTRY;
1609
1610         rc = lmv_check_connect(obd);
1611         if (rc)
1612                 RETURN(rc);
1613
1614         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1615
1616         /*
1617          * With DNE every object can have two locks in different namespaces:
1618          * lookup lock in space of MDT storing direntry and update/open lock in
1619          * space of MDT storing inode.
1620          */
1621         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1622                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1623                         continue;
1624                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1625         }
1626
1627         RETURN(0);
1628 }
1629
1630 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1631                            ldlm_iterator_t it, void *data)
1632 {
1633         struct obd_device       *obd = exp->exp_obd;
1634         struct lmv_obd          *lmv = &obd->u.lmv;
1635         int                     i;
1636         int                     tgt;
1637         int                     rc;
1638         ENTRY;
1639
1640         rc = lmv_check_connect(obd);
1641         if (rc)
1642                 RETURN(rc);
1643
1644         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1645
1646         /*
1647          * With DNE every object can have two locks in different namespaces:
1648          * lookup lock in space of MDT storing direntry and update/open lock in
1649          * space of MDT storing inode.  Try the MDT that the FID maps to first,
1650          * since this can be easily found, and only try others if that fails.
1651          */
1652         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1653              i < lmv->desc.ld_tgt_count;
1654              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1655                 if (tgt < 0) {
1656                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1657                                obd->obd_name, PFID(fid), tgt);
1658                         tgt = 0;
1659                 }
1660
1661                 if (lmv->tgts[tgt] == NULL ||
1662                     lmv->tgts[tgt]->ltd_exp == NULL)
1663                         continue;
1664
1665                 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1666                 if (rc)
1667                         RETURN(rc);
1668         }
1669
1670         RETURN(rc);
1671 }
1672
1673
1674 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1675                      struct md_open_data *mod, struct ptlrpc_request **request)
1676 {
1677         struct obd_device     *obd = exp->exp_obd;
1678         struct lmv_obd        *lmv = &obd->u.lmv;
1679         struct lmv_tgt_desc   *tgt;
1680         int                    rc;
1681         ENTRY;
1682
1683         rc = lmv_check_connect(obd);
1684         if (rc)
1685                 RETURN(rc);
1686
1687         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1688         if (IS_ERR(tgt))
1689                 RETURN(PTR_ERR(tgt));
1690
1691         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1692         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1693         RETURN(rc);
1694 }
1695
1696 /**
1697  * Choosing the MDT by name or FID in @op_data.
1698  * For non-striped directory, it will locate MDT by fid.
1699  * For striped-directory, it will locate MDT by name. And also
1700  * it will reset op_fid1 with the FID of the choosen stripe.
1701  **/
1702 struct lmv_tgt_desc *
1703 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1704                            const char *name, int namelen, struct lu_fid *fid,
1705                            u32 *mds)
1706 {
1707         struct lmv_tgt_desc     *tgt;
1708         const struct lmv_oinfo  *oinfo;
1709
1710         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1711                 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1712         } else {
1713                 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1714                 if (IS_ERR(oinfo))
1715                         RETURN(ERR_CAST(oinfo));
1716         }
1717
1718         *fid = oinfo->lmo_fid;
1719         *mds = oinfo->lmo_mds;
1720         tgt = lmv_get_target(lmv, *mds, NULL);
1721
1722         CDEBUG(D_INFO, "locate on mds %u "DFID"\n", *mds, PFID(fid));
1723         return tgt;
1724 }
1725
1726 /**
1727  * Locate mds by fid or name
1728  *
1729  * For striped directory (lsm != NULL), it will locate the stripe
1730  * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1731  * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1732  * walk through all of stripes to locate the entry.
1733  *
1734  * For normal direcotry, it will locate MDS by FID directly.
1735  * \param[in] lmv       LMV device
1736  * \param[in] op_data   client MD stack parameters, name, namelen
1737  *                      mds_num etc.
1738  * \param[in] fid       object FID used to locate MDS.
1739  *
1740  * retval               pointer to the lmv_tgt_desc if succeed.
1741  *                      ERR_PTR(errno) if failed.
1742  */
1743 struct lmv_tgt_desc*
1744 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1745                struct lu_fid *fid)
1746 {
1747         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1748         struct lmv_tgt_desc     *tgt;
1749
1750         /* During creating VOLATILE file, it should honor the mdt
1751          * index if the file under striped dir is being restored, see
1752          * ct_restore(). */
1753         if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1754             (int)op_data->op_mds != -1 && lsm != NULL) {
1755                 int i;
1756                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1757                 if (IS_ERR(tgt))
1758                         return tgt;
1759
1760                 /* refill the right parent fid */
1761                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1762                         struct lmv_oinfo *oinfo;
1763
1764                         oinfo = &lsm->lsm_md_oinfo[i];
1765                         if (oinfo->lmo_mds == op_data->op_mds) {
1766                                 *fid = oinfo->lmo_fid;
1767                                 break;
1768                         }
1769                 }
1770
1771                 /* Hmm, can not find the stripe by mdt_index(op_mds) */
1772                 if (i == lsm->lsm_md_stripe_count)
1773                         tgt = ERR_PTR(-EINVAL);
1774
1775                 return tgt;
1776         }
1777
1778         if (lsm == NULL || op_data->op_namelen == 0) {
1779                 tgt = lmv_find_target(lmv, fid);
1780                 if (IS_ERR(tgt))
1781                         return tgt;
1782
1783                 op_data->op_mds = tgt->ltd_idx;
1784                 return tgt;
1785         }
1786
1787         return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1788                                           op_data->op_namelen, fid,
1789                                           &op_data->op_mds);
1790 }
1791
1792 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1793                 const void *data, size_t datalen, umode_t mode, uid_t uid,
1794                 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1795                 struct ptlrpc_request **request)
1796 {
1797         struct obd_device       *obd = exp->exp_obd;
1798         struct lmv_obd          *lmv = &obd->u.lmv;
1799         struct lmv_tgt_desc     *tgt;
1800         int                      rc;
1801         ENTRY;
1802
1803         rc = lmv_check_connect(obd);
1804         if (rc)
1805                 RETURN(rc);
1806
1807         if (!lmv->desc.ld_active_tgt_count)
1808                 RETURN(-EIO);
1809
1810         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1811         if (IS_ERR(tgt))
1812                 RETURN(PTR_ERR(tgt));
1813
1814         CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1815                 (int)op_data->op_namelen, op_data->op_name,
1816                 PFID(&op_data->op_fid1), op_data->op_mds);
1817
1818         rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1819         if (rc)
1820                 RETURN(rc);
1821
1822         /* Send the create request to the MDT where the object
1823          * will be located */
1824         tgt = lmv_find_target(lmv, &op_data->op_fid2);
1825         if (IS_ERR(tgt))
1826                 RETURN(PTR_ERR(tgt));
1827
1828         op_data->op_mds = tgt->ltd_idx;
1829
1830         CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1831                PFID(&op_data->op_fid2), op_data->op_mds);
1832
1833         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1834         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1835                        cap_effective, rdev, request);
1836         if (rc == 0) {
1837                 if (*request == NULL)
1838                         RETURN(rc);
1839                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1840         }
1841         RETURN(rc);
1842 }
1843
1844 static int lmv_done_writing(struct obd_export *exp,
1845                             struct md_op_data *op_data,
1846                             struct md_open_data *mod)
1847 {
1848         struct obd_device     *obd = exp->exp_obd;
1849         struct lmv_obd        *lmv = &obd->u.lmv;
1850         struct lmv_tgt_desc   *tgt;
1851         int                    rc;
1852         ENTRY;
1853
1854         rc = lmv_check_connect(obd);
1855         if (rc)
1856                 RETURN(rc);
1857
1858         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1859         if (IS_ERR(tgt))
1860                 RETURN(PTR_ERR(tgt));
1861
1862         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1863         RETURN(rc);
1864 }
1865
1866 static int
1867 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1868             const union ldlm_policy_data *policy,
1869             struct lookup_intent *it, struct md_op_data *op_data,
1870             struct lustre_handle *lockh, __u64 extra_lock_flags)
1871 {
1872         struct obd_device        *obd = exp->exp_obd;
1873         struct lmv_obd           *lmv = &obd->u.lmv;
1874         struct lmv_tgt_desc      *tgt;
1875         int                       rc;
1876         ENTRY;
1877
1878         rc = lmv_check_connect(obd);
1879         if (rc)
1880                 RETURN(rc);
1881
1882         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1883                LL_IT2STR(it), PFID(&op_data->op_fid1));
1884
1885         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1886         if (IS_ERR(tgt))
1887                 RETURN(PTR_ERR(tgt));
1888
1889         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%u\n",
1890                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1891
1892         rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1893                         extra_lock_flags);
1894
1895         RETURN(rc);
1896 }
1897
1898 static int
1899 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1900                  struct ptlrpc_request **preq)
1901 {
1902         struct ptlrpc_request   *req = NULL;
1903         struct obd_device       *obd = exp->exp_obd;
1904         struct lmv_obd          *lmv = &obd->u.lmv;
1905         struct lmv_tgt_desc     *tgt;
1906         struct mdt_body         *body;
1907         int                      rc;
1908         ENTRY;
1909
1910         rc = lmv_check_connect(obd);
1911         if (rc)
1912                 RETURN(rc);
1913
1914         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1915         if (IS_ERR(tgt))
1916                 RETURN(PTR_ERR(tgt));
1917
1918         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1919                 (int)op_data->op_namelen, op_data->op_name,
1920                 PFID(&op_data->op_fid1), tgt->ltd_idx);
1921
1922         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1923         if (rc != 0)
1924                 RETURN(rc);
1925
1926         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1927         LASSERT(body != NULL);
1928
1929         if (body->mbo_valid & OBD_MD_MDS) {
1930                 struct lu_fid rid = body->mbo_fid1;
1931                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1932                        PFID(&rid));
1933
1934                 tgt = lmv_find_target(lmv, &rid);
1935                 if (IS_ERR(tgt)) {
1936                         ptlrpc_req_finished(*preq);
1937                         preq = NULL;
1938                         RETURN(PTR_ERR(tgt));
1939                 }
1940
1941                 op_data->op_fid1 = rid;
1942                 op_data->op_valid |= OBD_MD_FLCROSSREF;
1943                 op_data->op_namelen = 0;
1944                 op_data->op_name = NULL;
1945                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1946                 ptlrpc_req_finished(*preq);
1947                 *preq = req;
1948         }
1949
1950         RETURN(rc);
1951 }
1952
1953 #define md_op_data_fid(op_data, fl)                     \
1954         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1955          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1956          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1957          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1958          NULL)
1959
1960 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1961                             struct md_op_data *op_data,
1962                             __u32 op_tgt, ldlm_mode_t mode, int bits, int flag)
1963 {
1964         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
1965         struct obd_device      *obd = exp->exp_obd;
1966         struct lmv_obd         *lmv = &obd->u.lmv;
1967         ldlm_policy_data_t      policy = {{ 0 }};
1968         int                     rc = 0;
1969         ENTRY;
1970
1971         if (!fid_is_sane(fid))
1972                 RETURN(0);
1973
1974         if (tgt == NULL) {
1975                 tgt = lmv_find_target(lmv, fid);
1976                 if (IS_ERR(tgt))
1977                         RETURN(PTR_ERR(tgt));
1978         }
1979
1980         if (tgt->ltd_idx != op_tgt) {
1981                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1982                 policy.l_inodebits.bits = bits;
1983                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1984                                       mode, LCF_ASYNC, NULL);
1985         } else {
1986                 CDEBUG(D_INODE,
1987                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
1988                        op_tgt, PFID(fid));
1989                 op_data->op_flags |= flag;
1990                 rc = 0;
1991         }
1992
1993         RETURN(rc);
1994 }
1995
1996 /*
1997  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1998  * op_data->op_fid2
1999  */
2000 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2001                     struct ptlrpc_request **request)
2002 {
2003         struct obd_device       *obd = exp->exp_obd;
2004         struct lmv_obd          *lmv = &obd->u.lmv;
2005         struct lmv_tgt_desc     *tgt;
2006         int                      rc;
2007         ENTRY;
2008
2009         rc = lmv_check_connect(obd);
2010         if (rc)
2011                 RETURN(rc);
2012
2013         LASSERT(op_data->op_namelen != 0);
2014
2015         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2016                PFID(&op_data->op_fid2), (int)op_data->op_namelen,
2017                op_data->op_name, PFID(&op_data->op_fid1));
2018
2019         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2020         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2021         op_data->op_cap = cfs_curproc_cap_pack();
2022         if (op_data->op_mea2 != NULL) {
2023                 struct lmv_stripe_md    *lsm = op_data->op_mea2;
2024                 const struct lmv_oinfo  *oinfo;
2025
2026                 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2027                                                 op_data->op_namelen);
2028                 if (IS_ERR(oinfo))
2029                         RETURN(PTR_ERR(oinfo));
2030
2031                 op_data->op_fid2 = oinfo->lmo_fid;
2032         }
2033
2034         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2035         if (IS_ERR(tgt))
2036                 RETURN(PTR_ERR(tgt));
2037
2038         /*
2039          * Cancel UPDATE lock on child (fid1).
2040          */
2041         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2042         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2043                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2044         if (rc != 0)
2045                 RETURN(rc);
2046
2047         rc = md_link(tgt->ltd_exp, op_data, request);
2048
2049         RETURN(rc);
2050 }
2051
2052 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2053                       const char *old, size_t oldlen,
2054                       const char *new, size_t newlen,
2055                       struct ptlrpc_request **request)
2056 {
2057         struct obd_device       *obd = exp->exp_obd;
2058         struct lmv_obd          *lmv = &obd->u.lmv;
2059         struct lmv_tgt_desc     *src_tgt;
2060         int                     rc;
2061         ENTRY;
2062
2063         LASSERT(oldlen != 0);
2064
2065         CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2066                (int)oldlen, old, PFID(&op_data->op_fid1),
2067                op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2068                (int)newlen, new, PFID(&op_data->op_fid2),
2069                op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2070
2071         rc = lmv_check_connect(obd);
2072         if (rc)
2073                 RETURN(rc);
2074
2075         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2076         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2077         op_data->op_cap = cfs_curproc_cap_pack();
2078         if (op_data->op_cli_flags & CLI_MIGRATE) {
2079                 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2080                          PFID(&op_data->op_fid3));
2081                 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2082                 if (rc)
2083                         RETURN(rc);
2084                 src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid3);
2085         } else {
2086                 if (op_data->op_mea1 != NULL) {
2087                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2088
2089                         src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2090                                                              oldlen,
2091                                                              &op_data->op_fid1,
2092                                                              &op_data->op_mds);
2093                         if (IS_ERR(src_tgt))
2094                                 RETURN(PTR_ERR(src_tgt));
2095                 } else {
2096                         src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2097                         if (IS_ERR(src_tgt))
2098                                 RETURN(PTR_ERR(src_tgt));
2099
2100                         op_data->op_mds = src_tgt->ltd_idx;
2101                 }
2102
2103                 if (op_data->op_mea2) {
2104                         struct lmv_stripe_md    *lsm = op_data->op_mea2;
2105                         const struct lmv_oinfo  *oinfo;
2106
2107                         oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2108                         if (IS_ERR(oinfo))
2109                                 RETURN(PTR_ERR(oinfo));
2110
2111                         op_data->op_fid2 = oinfo->lmo_fid;
2112                 }
2113         }
2114         if (IS_ERR(src_tgt))
2115                 RETURN(PTR_ERR(src_tgt));
2116
2117         /*
2118          * LOOKUP lock on src child (fid3) should also be cancelled for
2119          * src_tgt in mdc_rename.
2120          */
2121         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2122
2123         /*
2124          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2125          * own target.
2126          */
2127         rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2128                               LCK_EX, MDS_INODELOCK_UPDATE,
2129                               MF_MDC_CANCEL_FID2);
2130
2131         if (rc != 0)
2132                 RETURN(rc);
2133         /*
2134          * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2135          */
2136         if (fid_is_sane(&op_data->op_fid3)) {
2137                 struct lmv_tgt_desc *tgt;
2138
2139                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2140                 if (IS_ERR(tgt))
2141                         RETURN(PTR_ERR(tgt));
2142
2143                 /* Cancel LOOKUP lock on its parent */
2144                 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2145                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2146                                       MF_MDC_CANCEL_FID3);
2147                 if (rc != 0)
2148                         RETURN(rc);
2149
2150                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2151                                       LCK_EX, MDS_INODELOCK_FULL,
2152                                       MF_MDC_CANCEL_FID3);
2153                 if (rc != 0)
2154                         RETURN(rc);
2155         }
2156
2157         /*
2158          * Cancel all the locks on tgt child (fid4).
2159          */
2160         if (fid_is_sane(&op_data->op_fid4))
2161                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2162                                       LCK_EX, MDS_INODELOCK_FULL,
2163                                       MF_MDC_CANCEL_FID4);
2164
2165         CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2166                op_data->op_mds, PFID(&op_data->op_fid2));
2167
2168         rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2169                        request);
2170
2171         RETURN(rc);
2172 }
2173
2174 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2175                         void *ea, size_t ealen, void *ea2, size_t ea2len,
2176                         struct ptlrpc_request **request,
2177                         struct md_open_data **mod)
2178 {
2179         struct obd_device       *obd = exp->exp_obd;
2180         struct lmv_obd          *lmv = &obd->u.lmv;
2181         struct lmv_tgt_desc     *tgt;
2182         int                      rc = 0;
2183         ENTRY;
2184
2185         rc = lmv_check_connect(obd);
2186         if (rc)
2187                 RETURN(rc);
2188
2189         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2190                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2191
2192         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2193         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2194         if (IS_ERR(tgt))
2195                 RETURN(PTR_ERR(tgt));
2196
2197         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2198                         ea2len, request, mod);
2199
2200         RETURN(rc);
2201 }
2202
2203 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2204                      struct obd_capa *oc, struct ptlrpc_request **request)
2205 {
2206         struct obd_device       *obd = exp->exp_obd;
2207         struct lmv_obd          *lmv = &obd->u.lmv;
2208         struct lmv_tgt_desc     *tgt;
2209         int                      rc;
2210         ENTRY;
2211
2212         rc = lmv_check_connect(obd);
2213         if (rc != 0)
2214                 RETURN(rc);
2215
2216         tgt = lmv_find_target(lmv, fid);
2217         if (IS_ERR(tgt))
2218                 RETURN(PTR_ERR(tgt));
2219
2220         rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2221         RETURN(rc);
2222 }
2223
2224 /**
2225  * Get current minimum entry from striped directory
2226  *
2227  * This function will search the dir entry, whose hash value is the
2228  * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2229  * only being called for striped directory.
2230  *
2231  * \param[in] exp               export of LMV
2232  * \param[in] op_data           parameters transferred beween client MD stack
2233  *                              stripe_information will be included in this
2234  *                              parameter
2235  * \param[in] cb_op             ldlm callback being used in enqueue in
2236  *                              mdc_read_page
2237  * \param[in] hash_offset       the hash value, which is used to locate
2238  *                              minum(closet) dir entry
2239  * \param[in|out] stripe_offset the caller use this to indicate the stripe
2240  *                              index of last entry, so to avoid hash conflict
2241  *                              between stripes. It will also be used to
2242  *                              return the stripe index of current dir entry.
2243  * \param[in|out] entp          the minum entry and it also is being used
2244  *                              to input the last dir entry to resolve the
2245  *                              hash conflict
2246  *
2247  * \param[out] ppage            the page which holds the minum entry
2248  *
2249  * \retval                      = 0 get the entry successfully
2250  *                              negative errno (< 0) does not get the entry
2251  */
2252 static int lmv_get_min_striped_entry(struct obd_export *exp,
2253                                      struct md_op_data *op_data,
2254                                      struct md_callback *cb_op,
2255                                      __u64 hash_offset, int *stripe_offset,
2256                                      struct lu_dirent **entp,
2257                                      struct page **ppage)
2258 {
2259         struct obd_device       *obd = exp->exp_obd;
2260         struct lmv_obd          *lmv = &obd->u.lmv;
2261         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2262         struct lmv_tgt_desc     *tgt;
2263         int                     stripe_count;
2264         struct lu_dirent        *min_ent = NULL;
2265         struct page             *min_page = NULL;
2266         int                     min_idx = 0;
2267         int                     i;
2268         int                     rc = 0;
2269         ENTRY;
2270
2271         stripe_count = lsm->lsm_md_stripe_count;
2272         for (i = 0; i < stripe_count; i++) {
2273                 struct lu_dirent        *ent = NULL;
2274                 struct page             *page = NULL;
2275                 struct lu_dirpage       *dp;
2276                 __u64                   stripe_hash = hash_offset;
2277
2278                 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2279                 if (IS_ERR(tgt))
2280                         GOTO(out, rc = PTR_ERR(tgt));
2281
2282                 /* op_data will be shared by each stripe, so we need
2283                  * reset these value for each stripe */
2284                 op_data->op_stripe_offset = i;
2285                 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2286                 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2287                 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2288 next:
2289                 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2290                                   &page);
2291                 if (rc != 0)
2292                         GOTO(out, rc);
2293
2294                 dp = page_address(page);
2295                 for (ent = lu_dirent_start(dp); ent != NULL;
2296                      ent = lu_dirent_next(ent)) {
2297                         /* Skip dummy entry */
2298                         if (le16_to_cpu(ent->lde_namelen) == 0)
2299                                 continue;
2300
2301                         if (le64_to_cpu(ent->lde_hash) < hash_offset)
2302                                 continue;
2303
2304                         if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2305                             (*entp == ent || i < *stripe_offset))
2306                                 continue;
2307
2308                         /* skip . and .. for other stripes */
2309                         if (i != 0 &&
2310                             (strncmp(ent->lde_name, ".",
2311                                      le16_to_cpu(ent->lde_namelen)) == 0 ||
2312                              strncmp(ent->lde_name, "..",
2313                                      le16_to_cpu(ent->lde_namelen)) == 0))
2314                                 continue;
2315                         break;
2316                 }
2317
2318                 if (ent == NULL) {
2319                         stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2320
2321                         kunmap(page);
2322                         page_cache_release(page);
2323                         page = NULL;
2324
2325                         /* reach the end of current stripe, go to next stripe */
2326                         if (stripe_hash == MDS_DIR_END_OFF)
2327                                 continue;
2328                         else
2329                                 goto next;
2330                 }
2331
2332                 if (min_ent != NULL) {
2333                         if (le64_to_cpu(min_ent->lde_hash) >
2334                             le64_to_cpu(ent->lde_hash)) {
2335                                 min_ent = ent;
2336                                 kunmap(min_page);
2337                                 page_cache_release(min_page);
2338                                 min_idx = i;
2339                                 min_page = page;
2340                         } else {
2341                                 kunmap(page);
2342                                 page_cache_release(page);
2343                                 page = NULL;
2344                         }
2345                 } else {
2346                         min_ent = ent;
2347                         min_page = page;
2348                         min_idx = i;
2349                 }
2350         }
2351
2352 out:
2353         if (*ppage != NULL) {
2354                 kunmap(*ppage);
2355                 page_cache_release(*ppage);
2356         }
2357         *stripe_offset = min_idx;
2358         *entp = min_ent;
2359         *ppage = min_page;
2360         RETURN(rc);
2361 }
2362
2363 /**
2364  * Build dir entry page from a striped directory
2365  *
2366  * This function gets one entry by @offset from a striped directory. It will
2367  * read entries from all of stripes, and choose one closest to the required
2368  * offset(&offset). A few notes
2369  * 1. skip . and .. for non-zero stripes, because there can only have one .
2370  * and .. in a directory.
2371  * 2. op_data will be shared by all of stripes, instead of allocating new
2372  * one, so need to restore before reusing.
2373  * 3. release the entry page if that is not being chosen.
2374  *
2375  * \param[in] exp       obd export refer to LMV
2376  * \param[in] op_data   hold those MD parameters of read_entry
2377  * \param[in] cb_op     ldlm callback being used in enqueue in mdc_read_entry
2378  * \param[out] ldp      the entry being read
2379  * \param[out] ppage    the page holding the entry. Note: because the entry
2380  *                      will be accessed in upper layer, so we need hold the
2381  *                      page until the usages of entry is finished, see
2382  *                      ll_dir_entry_next.
2383  *
2384  * retval               =0 if get entry successfully
2385  *                      <0 cannot get entry
2386  */
2387 static int lmv_read_striped_page(struct obd_export *exp,
2388                                  struct md_op_data *op_data,
2389                                  struct md_callback *cb_op,
2390                                  __u64 offset, struct page **ppage)
2391 {
2392         struct obd_device       *obd = exp->exp_obd;
2393         struct lu_fid           master_fid = op_data->op_fid1;
2394         struct inode            *master_inode = op_data->op_data;
2395         __u64                   hash_offset = offset;
2396         struct lu_dirpage       *dp;
2397         struct page             *min_ent_page = NULL;
2398         struct page             *ent_page = NULL;
2399         struct lu_dirent        *ent;
2400         void                    *area;
2401         int                     ent_idx = 0;
2402         struct lu_dirent        *min_ent = NULL;
2403         struct lu_dirent        *last_ent;
2404         int                     left_bytes;
2405         int                     rc;
2406         ENTRY;
2407
2408         rc = lmv_check_connect(obd);
2409         if (rc)
2410                 RETURN(rc);
2411
2412         /* Allocate a page and read entries from all of stripes and fill
2413          * the page by hash order */
2414         ent_page = alloc_page(GFP_KERNEL);
2415         if (ent_page == NULL)
2416                 RETURN(-ENOMEM);
2417
2418         /* Initialize the entry page */
2419         dp = kmap(ent_page);
2420         memset(dp, 0, sizeof(*dp));
2421         dp->ldp_hash_start = cpu_to_le64(offset);
2422         dp->ldp_flags |= LDF_COLLIDE;
2423
2424         area = dp + 1;
2425         left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2426         ent = area;
2427         last_ent = ent;
2428         do {
2429                 __u16   ent_size;
2430
2431                 /* Find the minum entry from all sub-stripes */
2432                 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2433                                                &ent_idx, &min_ent,
2434                                                &min_ent_page);
2435                 if (rc != 0)
2436                         GOTO(out, rc);
2437
2438                 /* If it can not get minum entry, it means it already reaches
2439                  * the end of this directory */
2440                 if (min_ent == NULL) {
2441                         last_ent->lde_reclen = 0;
2442                         hash_offset = MDS_DIR_END_OFF;
2443                         GOTO(out, rc);
2444                 }
2445
2446                 ent_size = le16_to_cpu(min_ent->lde_reclen);
2447
2448                 /* the last entry lde_reclen is 0, but it might not
2449                  * the end of this entry of this temporay entry */
2450                 if (ent_size == 0)
2451                         ent_size = lu_dirent_calc_size(
2452                                         le16_to_cpu(min_ent->lde_namelen),
2453                                         le32_to_cpu(min_ent->lde_attrs));
2454                 if (ent_size > left_bytes) {
2455                         last_ent->lde_reclen = cpu_to_le16(0);
2456                         hash_offset = le64_to_cpu(min_ent->lde_hash);
2457                         GOTO(out, rc);
2458                 }
2459
2460                 memcpy(ent, min_ent, ent_size);
2461
2462                 /* Replace . with master FID and Replace .. with the parent FID
2463                  * of master object */
2464                 if (strncmp(ent->lde_name, ".",
2465                             le16_to_cpu(ent->lde_namelen)) == 0 &&
2466                     le16_to_cpu(ent->lde_namelen) == 1)
2467                         fid_cpu_to_le(&ent->lde_fid, &master_fid);
2468                 else if (strncmp(ent->lde_name, "..",
2469                                    le16_to_cpu(ent->lde_namelen)) == 0 &&
2470                            le16_to_cpu(ent->lde_namelen) == 2)
2471                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2472
2473                 left_bytes -= ent_size;
2474                 ent->lde_reclen = cpu_to_le16(ent_size);
2475                 last_ent = ent;
2476                 ent = (void *)ent + ent_size;
2477                 hash_offset = le64_to_cpu(min_ent->lde_hash);
2478                 if (hash_offset == MDS_DIR_END_OFF) {
2479                         last_ent->lde_reclen = 0;
2480                         break;
2481                 }
2482         } while (1);
2483 out:
2484         if (min_ent_page != NULL) {
2485                 kunmap(min_ent_page);
2486                 page_cache_release(min_ent_page);
2487         }
2488
2489         if (unlikely(rc != 0)) {
2490                 __free_page(ent_page);
2491                 ent_page = NULL;
2492         } else {
2493                 if (ent == area)
2494                         dp->ldp_flags |= LDF_EMPTY;
2495                 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2496                 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2497         }
2498
2499         /* We do not want to allocate md_op_data during each
2500          * dir entry reading, so op_data will be shared by every stripe,
2501          * then we need to restore it back to original value before
2502          * return to the upper layer */
2503         op_data->op_fid1 = master_fid;
2504         op_data->op_fid2 = master_fid;
2505         op_data->op_data = master_inode;
2506
2507         *ppage = ent_page;
2508
2509         RETURN(rc);
2510 }
2511
2512 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2513                   struct md_callback *cb_op, __u64 offset,
2514                   struct page **ppage)
2515 {
2516         struct obd_device       *obd = exp->exp_obd;
2517         struct lmv_obd          *lmv = &obd->u.lmv;
2518         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2519         struct lmv_tgt_desc     *tgt;
2520         int                     rc;
2521         ENTRY;
2522
2523         rc = lmv_check_connect(obd);
2524         if (rc != 0)
2525                 RETURN(rc);
2526
2527         if (unlikely(lsm != NULL)) {
2528                 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2529                 RETURN(rc);
2530         }
2531
2532         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2533         if (IS_ERR(tgt))
2534                 RETURN(PTR_ERR(tgt));
2535
2536         rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2537
2538         RETURN(rc);
2539 }
2540
2541 /**
2542  * Unlink a file/directory
2543  *
2544  * Unlink a file or directory under the parent dir. The unlink request
2545  * usually will be sent to the MDT where the child is located, but if
2546  * the client does not have the child FID then request will be sent to the
2547  * MDT where the parent is located.
2548  *
2549  * If the parent is a striped directory then it also needs to locate which
2550  * stripe the name of the child is located, and replace the parent FID
2551  * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2552  * it will walk through all of sub-stripes until the child is being
2553  * unlinked finally.
2554  *
2555  * \param[in] exp       export refer to LMV
2556  * \param[in] op_data   different parameters transferred beween client
2557  *                      MD stacks, name, namelen, FIDs etc.
2558  *                      op_fid1 is the parent FID, op_fid2 is the child
2559  *                      FID.
2560  * \param[out] request  point to the request of unlink.
2561  *
2562  * retval               0 if succeed
2563  *                      negative errno if failed.
2564  */
2565 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2566                       struct ptlrpc_request **request)
2567 {
2568         struct obd_device       *obd = exp->exp_obd;
2569         struct lmv_obd          *lmv = &obd->u.lmv;
2570         struct lmv_tgt_desc     *tgt = NULL;
2571         struct lmv_tgt_desc     *parent_tgt = NULL;
2572         struct mdt_body         *body;
2573         int                     rc;
2574         int                     stripe_index = 0;
2575         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2576         ENTRY;
2577
2578         rc = lmv_check_connect(obd);
2579         if (rc)
2580                 RETURN(rc);
2581 retry_unlink:
2582         /* For striped dir, we need to locate the parent as well */
2583         if (lsm != NULL) {
2584                 struct lmv_tgt_desc *tmp;
2585
2586                 LASSERT(op_data->op_name != NULL &&
2587                         op_data->op_namelen != 0);
2588
2589                 tmp = lmv_locate_target_for_name(lmv, lsm,
2590                                                  op_data->op_name,
2591                                                  op_data->op_namelen,
2592                                                  &op_data->op_fid1,
2593                                                  &op_data->op_mds);
2594
2595                 /* return -EBADFD means unknown hash type, might
2596                  * need try all sub-stripe here */
2597                 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2598                         RETURN(PTR_ERR(tmp));
2599
2600                 /* Note: both migrating dir and unknown hash dir need to
2601                  * try all of sub-stripes, so we need start search the
2602                  * name from stripe 0, but migrating dir is already handled
2603                  * inside lmv_locate_target_for_name(), so we only check
2604                  * unknown hash type directory here */
2605                 if (!lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
2606                         struct lmv_oinfo *oinfo;
2607
2608                         oinfo = &lsm->lsm_md_oinfo[stripe_index];
2609
2610                         op_data->op_fid1 = oinfo->lmo_fid;
2611                         op_data->op_mds = oinfo->lmo_mds;
2612                 }
2613         }
2614
2615 try_next_stripe:
2616         /* Send unlink requests to the MDT where the child is located */
2617         if (likely(!fid_is_zero(&op_data->op_fid2)))
2618                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2619         else if (lsm != NULL)
2620                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2621         else
2622                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2623
2624         if (IS_ERR(tgt))
2625                 RETURN(PTR_ERR(tgt));
2626
2627         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2628         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2629         op_data->op_cap = cfs_curproc_cap_pack();
2630
2631         /*
2632          * If child's fid is given, cancel unused locks for it if it is from
2633          * another export than parent.
2634          *
2635          * LOOKUP lock for child (fid3) should also be cancelled on parent
2636          * tgt_tgt in mdc_unlink().
2637          */
2638         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2639
2640         /*
2641          * Cancel FULL locks on child (fid3).
2642          */
2643         parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2644         if (IS_ERR(parent_tgt))
2645                 RETURN(PTR_ERR(parent_tgt));
2646
2647         if (parent_tgt != tgt) {
2648                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2649                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2650                                       MF_MDC_CANCEL_FID3);
2651         }
2652
2653         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2654                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2655         if (rc != 0)
2656                 RETURN(rc);
2657
2658         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2659                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2660
2661         rc = md_unlink(tgt->ltd_exp, op_data, request);
2662         if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2663                 RETURN(rc);
2664
2665         /* Try next stripe if it is needed. */
2666         if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2667                 struct lmv_oinfo *oinfo;
2668
2669                 stripe_index++;
2670                 if (stripe_index >= lsm->lsm_md_stripe_count)
2671                         RETURN(rc);
2672
2673                 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2674
2675                 op_data->op_fid1 = oinfo->lmo_fid;
2676                 op_data->op_mds = oinfo->lmo_mds;
2677
2678                 ptlrpc_req_finished(*request);
2679                 *request = NULL;
2680
2681                 goto try_next_stripe;
2682         }
2683
2684         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2685         if (body == NULL)
2686                 RETURN(-EPROTO);
2687
2688         /* Not cross-ref case, just get out of here. */
2689         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2690                 RETURN(rc);
2691
2692         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2693                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2694
2695         /* This is a remote object, try remote MDT, Note: it may
2696          * try more than 1 time here, Considering following case
2697          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2698          * 1. Initially A does not know where remote1 is, it send
2699          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2700          *    resend unlink RPC to MDT1 (retry 1st time).
2701          *
2702          * 2. During the unlink RPC in flight,
2703          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2704          *    and create new remote1, but on MDT0
2705          *
2706          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2707          *    /mnt/lustre, then lookup get fid of remote1, and find
2708          *    it is remote dir again, and replay -EREMOTE again.
2709          *
2710          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2711          *
2712          * In theory, it might try unlimited time here, but it should
2713          * be very rare case.  */
2714         op_data->op_fid2 = body->mbo_fid1;
2715         ptlrpc_req_finished(*request);
2716         *request = NULL;
2717
2718         goto retry_unlink;
2719 }
2720
2721 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2722 {
2723         struct lmv_obd *lmv = &obd->u.lmv;
2724         int rc = 0;
2725
2726         switch (stage) {
2727         case OBD_CLEANUP_EARLY:
2728                 /* XXX: here should be calling obd_precleanup() down to
2729                  * stack. */
2730                 break;
2731         case OBD_CLEANUP_EXPORTS:
2732                 fld_client_proc_fini(&lmv->lmv_fld);
2733                 lprocfs_obd_cleanup(obd);
2734                 lprocfs_free_md_stats(obd);
2735                 break;
2736         default:
2737                 break;
2738         }
2739         RETURN(rc);
2740 }
2741
2742 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2743                         __u32 keylen, void *key, __u32 *vallen, void *val,
2744                         struct lov_stripe_md *lsm)
2745 {
2746         struct obd_device       *obd;
2747         struct lmv_obd          *lmv;
2748         int                      rc = 0;
2749         ENTRY;
2750
2751         obd = class_exp2obd(exp);
2752         if (obd == NULL) {
2753                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2754                        exp->exp_handle.h_cookie);
2755                 RETURN(-EINVAL);
2756         }
2757
2758         lmv = &obd->u.lmv;
2759         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2760                 int i;
2761
2762                 rc = lmv_check_connect(obd);
2763                 if (rc)
2764                         RETURN(rc);
2765
2766                 LASSERT(*vallen == sizeof(__u32));
2767                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2768                         struct lmv_tgt_desc *tgt = lmv->tgts[i];
2769                         /*
2770                          * All tgts should be connected when this gets called.
2771                          */
2772                         if (tgt == NULL || tgt->ltd_exp == NULL)
2773                                 continue;
2774
2775                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2776                                           vallen, val, NULL))
2777                                 RETURN(0);
2778                 }
2779                 RETURN(-EINVAL);
2780         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2781                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2782                    KEY_IS(KEY_MAX_COOKIESIZE) ||
2783                    KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2784                    KEY_IS(KEY_CONN_DATA)) {
2785                 rc = lmv_check_connect(obd);
2786                 if (rc)
2787                         RETURN(rc);
2788
2789                 /*
2790                  * Forwarding this request to first MDS, it should know LOV
2791                  * desc.
2792                  */
2793                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2794                                   vallen, val, NULL);
2795                 if (!rc && KEY_IS(KEY_CONN_DATA))
2796                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2797                 RETURN(rc);
2798         } else if (KEY_IS(KEY_TGT_COUNT)) {
2799                 *((int *)val) = lmv->desc.ld_tgt_count;
2800                 RETURN(0);
2801         }
2802
2803         CDEBUG(D_IOCTL, "Invalid key\n");
2804         RETURN(-EINVAL);
2805 }
2806
2807 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2808                        obd_count keylen, void *key, obd_count vallen,
2809                        void *val, struct ptlrpc_request_set *set)
2810 {
2811         struct lmv_tgt_desc    *tgt = NULL;
2812         struct obd_device      *obd;
2813         struct lmv_obd         *lmv;
2814         int rc = 0;
2815         ENTRY;
2816
2817         obd = class_exp2obd(exp);
2818         if (obd == NULL) {
2819                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2820                        exp->exp_handle.h_cookie);
2821                 RETURN(-EINVAL);
2822         }
2823         lmv = &obd->u.lmv;
2824
2825         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2826                 int i, err = 0;
2827
2828                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2829                         tgt = lmv->tgts[i];
2830
2831                         if (tgt == NULL || tgt->ltd_exp == NULL)
2832                                 continue;
2833
2834                         err = obd_set_info_async(env, tgt->ltd_exp,
2835                                                  keylen, key, vallen, val, set);
2836                         if (err && rc == 0)
2837                                 rc = err;
2838                 }
2839
2840                 RETURN(rc);
2841         }
2842
2843         RETURN(-EINVAL);
2844 }
2845
2846 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2847                           struct lmv_mds_md_v1 *lmm1)
2848 {
2849         int     cplen;
2850         int     i;
2851
2852         lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2853         lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2854         lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
2855         lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
2856         cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
2857                         sizeof(lmm1->lmv_pool_name));
2858         if (cplen >= sizeof(lmm1->lmv_pool_name))
2859                 return -E2BIG;
2860
2861         for (i = 0; i < lsm->lsm_md_stripe_count; i++)
2862                 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
2863                               &lsm->lsm_md_oinfo[i].lmo_fid);
2864         return 0;
2865 }
2866
2867 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
2868                 int stripe_count)
2869 {
2870         int     lmm_size = 0;
2871         bool    allocated = false;
2872         int     rc = 0;
2873         ENTRY;
2874
2875         LASSERT(lmmp != NULL);
2876         /* Free lmm */
2877         if (*lmmp != NULL && lsm == NULL) {
2878                 int stripe_count;
2879
2880                 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
2881                 lmm_size = lmv_mds_md_size(stripe_count,
2882                                            le32_to_cpu((*lmmp)->lmv_magic));
2883                 if (lmm_size == 0)
2884                         RETURN(-EINVAL);
2885                 OBD_FREE(*lmmp, lmm_size);
2886                 *lmmp = NULL;
2887                 RETURN(0);
2888         }
2889
2890         /* Alloc lmm */
2891         if (*lmmp == NULL && lsm == NULL) {
2892                 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
2893                 LASSERT(lmm_size > 0);
2894                 OBD_ALLOC(*lmmp, lmm_size);
2895                 if (*lmmp == NULL)
2896                         RETURN(-ENOMEM);
2897                 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
2898                 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
2899                 RETURN(lmm_size);
2900         }
2901
2902         /* pack lmm */
2903         LASSERT(lsm != NULL);
2904         lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
2905         if (*lmmp == NULL) {
2906                 OBD_ALLOC(*lmmp, lmm_size);
2907                 if (*lmmp == NULL)
2908                         RETURN(-ENOMEM);
2909                 allocated = true;
2910         }
2911
2912         switch (lsm->lsm_md_magic) {
2913         case LMV_MAGIC_V1:
2914                 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
2915                 break;
2916         default:
2917                 rc = -EINVAL;
2918                 break;
2919         }
2920
2921         if (rc != 0 && allocated) {
2922                 OBD_FREE(*lmmp, lmm_size);
2923                 *lmmp = NULL;
2924         }
2925
2926         RETURN(lmm_size);
2927 }
2928 EXPORT_SYMBOL(lmv_pack_md);
2929
2930 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2931                             const struct lmv_mds_md_v1 *lmm1)
2932 {
2933         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
2934         int             stripe_count;
2935         int             cplen;
2936         int             i;
2937         int             rc = 0;
2938         ENTRY;
2939
2940         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2941         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2942         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2943         if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
2944                 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
2945         else
2946                 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2947         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2948         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2949                         sizeof(lsm->lsm_md_pool_name));
2950
2951         if (cplen >= sizeof(lsm->lsm_md_pool_name))
2952                 RETURN(-E2BIG);
2953
2954         CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2955                "layout_version %d\n", lsm->lsm_md_stripe_count,
2956                lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2957                lsm->lsm_md_layout_version);
2958
2959         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2960         for (i = 0; i < stripe_count; i++) {
2961                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2962                               &lmm1->lmv_stripe_fids[i]);
2963                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2964                                     &lsm->lsm_md_oinfo[i].lmo_mds);
2965                 if (rc != 0)
2966                         RETURN(rc);
2967                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
2968                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
2969         }
2970
2971         RETURN(rc);
2972 }
2973
2974 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
2975                   const union lmv_mds_md *lmm, int stripe_count)
2976 {
2977         struct lmv_stripe_md     *lsm;
2978         int                      lsm_size;
2979         int                      rc;
2980         bool                     allocated = false;
2981         ENTRY;
2982
2983         LASSERT(lsmp != NULL);
2984
2985         lsm = *lsmp;
2986         /* Free memmd */
2987         if (lsm != NULL && lmm == NULL) {
2988                 int i;
2989                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
2990                         /* For migrating inode, the master stripe and master
2991                          * object will be the same, so do not need iput, see
2992                          * ll_update_lsm_md */
2993                         if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
2994                               i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
2995                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
2996                 }
2997                 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
2998                 OBD_FREE(lsm, lsm_size);
2999                 *lsmp = NULL;
3000                 RETURN(0);
3001         }
3002
3003         /* Alloc memmd */
3004         if (lsm == NULL && lmm == NULL) {
3005                 lsm_size = lmv_stripe_md_size(stripe_count);
3006                 OBD_ALLOC(lsm, lsm_size);
3007                 if (lsm == NULL)
3008                         RETURN(-ENOMEM);
3009                 lsm->lsm_md_stripe_count = stripe_count;
3010                 *lsmp = lsm;
3011                 RETURN(0);
3012         }
3013
3014         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3015                 RETURN(-EPERM);
3016
3017         /* Unpack memmd */
3018         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3019             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3020                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3021                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3022                        -EIO);
3023                 RETURN(-EIO);
3024         }
3025
3026         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3027                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3028         else
3029                 /**
3030                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3031                  * stripecount should be 0 then.
3032                  */
3033                 lsm_size = lmv_stripe_md_size(0);
3034
3035         lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3036         if (lsm == NULL) {
3037                 OBD_ALLOC(lsm, lsm_size);
3038                 if (lsm == NULL)
3039                         RETURN(-ENOMEM);
3040                 allocated = true;
3041                 *lsmp = lsm;
3042         }
3043
3044         switch (le32_to_cpu(lmm->lmv_magic)) {
3045         case LMV_MAGIC_V1:
3046                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3047                 break;
3048         default:
3049                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3050                        le32_to_cpu(lmm->lmv_magic));
3051                 rc = -EINVAL;
3052                 break;
3053         }
3054
3055         if (rc != 0 && allocated) {
3056                 OBD_FREE(lsm, lsm_size);
3057                 *lsmp = NULL;
3058                 lsm_size = rc;
3059         }
3060         RETURN(lsm_size);
3061 }
3062
3063 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
3064 {
3065         return lmv_unpack_md(NULL, lsmp, NULL, stripes);
3066 }
3067 EXPORT_SYMBOL(lmv_alloc_memmd);
3068
3069 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3070 {
3071         lmv_unpack_md(NULL, &lsm, NULL, 0);
3072 }
3073 EXPORT_SYMBOL(lmv_free_memmd);
3074
3075 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
3076                  struct lov_mds_md *lmm, int disk_len)
3077 {
3078         return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
3079                              (union lmv_mds_md *)lmm, disk_len);
3080 }
3081
3082 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
3083                struct lov_stripe_md *lsm)
3084 {
3085         struct obd_device               *obd = exp->exp_obd;
3086         struct lmv_obd                  *lmv_obd = &obd->u.lmv;
3087         const struct lmv_stripe_md      *lmv = (struct lmv_stripe_md *)lsm;
3088         int                             stripe_count;
3089
3090         if (lmmp == NULL) {
3091                 if (lsm != NULL)
3092                         stripe_count = lmv->lsm_md_stripe_count;
3093                 else
3094                         stripe_count = lmv_obd->desc.ld_tgt_count;
3095
3096                 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
3097         }
3098
3099         return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
3100 }
3101
3102 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3103                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
3104                              ldlm_cancel_flags_t flags, void *opaque)
3105 {
3106         struct obd_device       *obd = exp->exp_obd;
3107         struct lmv_obd          *lmv = &obd->u.lmv;
3108         int                      rc = 0;
3109         int                      err;
3110         __u32                    i;
3111         ENTRY;
3112
3113         LASSERT(fid != NULL);
3114
3115         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3116                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3117
3118                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3119                         continue;
3120
3121                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3122                                        opaque);
3123                 if (!rc)
3124                         rc = err;
3125         }
3126         RETURN(rc);
3127 }
3128
3129 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3130                       __u64 *bits)
3131 {
3132         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3133         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3134         int                      rc;
3135         ENTRY;
3136
3137         if (tgt == NULL || tgt->ltd_exp == NULL)
3138                 RETURN(-EINVAL);
3139         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3140         RETURN(rc);
3141 }
3142
3143 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3144                            const struct lu_fid *fid, ldlm_type_t type,
3145                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
3146                            struct lustre_handle *lockh)
3147 {
3148         struct obd_device       *obd = exp->exp_obd;
3149         struct lmv_obd          *lmv = &obd->u.lmv;
3150         ldlm_mode_t             rc;
3151         int                     tgt;
3152         int                     i;
3153         ENTRY;
3154
3155         CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3156
3157         /*
3158          * With DNE every object can have two locks in different namespaces:
3159          * lookup lock in space of MDT storing direntry and update/open lock in
3160          * space of MDT storing inode.  Try the MDT that the FID maps to first,
3161          * since this can be easily found, and only try others if that fails.
3162          */
3163         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3164              i < lmv->desc.ld_tgt_count;
3165              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3166                 if (tgt < 0) {
3167                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3168                                obd->obd_name, PFID(fid), tgt);
3169                         tgt = 0;
3170                 }
3171
3172                 if (lmv->tgts[tgt] == NULL ||
3173                     lmv->tgts[tgt]->ltd_exp == NULL ||
3174                     lmv->tgts[tgt]->ltd_active == 0)
3175                         continue;
3176
3177                 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3178                                    type, policy, mode, lockh);
3179                 if (rc)
3180                         RETURN(rc);
3181         }
3182
3183         RETURN(0);
3184 }
3185
3186 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3187                       struct obd_export *dt_exp, struct obd_export *md_exp,
3188                       struct lustre_md *md)
3189 {
3190         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3191         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3192
3193         if (tgt == NULL || tgt->ltd_exp == NULL)
3194                 RETURN(-EINVAL);
3195
3196         return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3197 }
3198
3199 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3200 {
3201         struct obd_device       *obd = exp->exp_obd;
3202         struct lmv_obd          *lmv = &obd->u.lmv;
3203         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3204         ENTRY;
3205
3206         if (md->lmv != NULL) {
3207                 lmv_free_memmd(md->lmv);
3208                 md->lmv = NULL;
3209         }
3210         if (tgt == NULL || tgt->ltd_exp == NULL)
3211                 RETURN(-EINVAL);
3212         RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3213 }
3214
3215 int lmv_set_open_replay_data(struct obd_export *exp,
3216                              struct obd_client_handle *och,
3217                              struct lookup_intent *it)
3218 {
3219         struct obd_device       *obd = exp->exp_obd;
3220         struct lmv_obd          *lmv = &obd->u.lmv;
3221         struct lmv_tgt_desc     *tgt;
3222         ENTRY;
3223
3224         tgt = lmv_find_target(lmv, &och->och_fid);
3225         if (IS_ERR(tgt))
3226                 RETURN(PTR_ERR(tgt));
3227
3228         RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3229 }
3230
3231 int lmv_clear_open_replay_data(struct obd_export *exp,
3232                                struct obd_client_handle *och)
3233 {
3234         struct obd_device       *obd = exp->exp_obd;
3235         struct lmv_obd          *lmv = &obd->u.lmv;
3236         struct lmv_tgt_desc     *tgt;
3237         ENTRY;
3238
3239         tgt = lmv_find_target(lmv, &och->och_fid);
3240         if (IS_ERR(tgt))
3241                 RETURN(PTR_ERR(tgt));
3242
3243         RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3244 }
3245
3246 static int lmv_get_remote_perm(struct obd_export *exp,
3247                                const struct lu_fid *fid,
3248                                struct obd_capa *oc, __u32 suppgid,
3249                                struct ptlrpc_request **request)
3250 {
3251         struct obd_device       *obd = exp->exp_obd;
3252         struct lmv_obd          *lmv = &obd->u.lmv;
3253         struct lmv_tgt_desc     *tgt;
3254         int                      rc;
3255         ENTRY;
3256
3257         rc = lmv_check_connect(obd);
3258         if (rc)
3259                 RETURN(rc);
3260
3261         tgt = lmv_find_target(lmv, fid);
3262         if (IS_ERR(tgt))
3263                 RETURN(PTR_ERR(tgt));
3264
3265         rc = md_get_remote_perm(tgt->ltd_exp, fid, oc, suppgid, request);
3266         RETURN(rc);
3267 }
3268
3269 static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3270                           renew_capa_cb_t cb)
3271 {
3272         struct obd_device       *obd = exp->exp_obd;
3273         struct lmv_obd          *lmv = &obd->u.lmv;
3274         struct lmv_tgt_desc     *tgt;
3275         int                      rc;
3276         ENTRY;
3277
3278         rc = lmv_check_connect(obd);
3279         if (rc)
3280                 RETURN(rc);
3281
3282         tgt = lmv_find_target(lmv, &oc->c_capa.lc_fid);
3283         if (IS_ERR(tgt))
3284                 RETURN(PTR_ERR(tgt));
3285
3286         rc = md_renew_capa(tgt->ltd_exp, oc, cb);
3287         RETURN(rc);
3288 }
3289
3290 int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
3291                     const struct req_msg_field *field, struct obd_capa **oc)
3292 {
3293         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3294         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3295
3296         if (tgt == NULL || tgt->ltd_exp == NULL)
3297                 RETURN(-EINVAL);
3298         return md_unpack_capa(tgt->ltd_exp, req, field, oc);
3299 }
3300
3301 int lmv_intent_getattr_async(struct obd_export *exp,
3302                              struct md_enqueue_info *minfo,
3303                              struct ldlm_enqueue_info *einfo)
3304 {
3305         struct md_op_data       *op_data = &minfo->mi_data;
3306         struct obd_device       *obd = exp->exp_obd;
3307         struct lmv_obd          *lmv = &obd->u.lmv;
3308         struct lmv_tgt_desc     *tgt = NULL;
3309         int                      rc;
3310         ENTRY;
3311
3312         rc = lmv_check_connect(obd);
3313         if (rc)
3314                 RETURN(rc);
3315
3316         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3317         if (IS_ERR(tgt))
3318                 RETURN(PTR_ERR(tgt));
3319
3320         rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3321         RETURN(rc);
3322 }
3323
3324 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3325                         struct lu_fid *fid, __u64 *bits)
3326 {
3327         struct obd_device       *obd = exp->exp_obd;
3328         struct lmv_obd          *lmv = &obd->u.lmv;
3329         struct lmv_tgt_desc     *tgt;
3330         int                      rc;
3331         ENTRY;
3332
3333         rc = lmv_check_connect(obd);
3334         if (rc)
3335                 RETURN(rc);
3336
3337         tgt = lmv_find_target(lmv, fid);
3338         if (IS_ERR(tgt))
3339                 RETURN(PTR_ERR(tgt));
3340
3341         rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3342         RETURN(rc);
3343 }
3344
3345 int lmv_get_fid_from_lsm(struct obd_export *exp,
3346                          const struct lmv_stripe_md *lsm,
3347                          const char *name, int namelen, struct lu_fid *fid)
3348 {
3349         const struct lmv_oinfo *oinfo;
3350
3351         LASSERT(lsm != NULL);
3352         oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3353         if (IS_ERR(oinfo))
3354                 return PTR_ERR(oinfo);
3355
3356         *fid = oinfo->lmo_fid;
3357
3358         RETURN(0);
3359 }
3360
3361 /**
3362  * For lmv, only need to send request to master MDT, and the master MDT will
3363  * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3364  * we directly fetch data from the slave MDTs.
3365  */
3366 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3367                  struct obd_quotactl *oqctl)
3368 {
3369         struct obd_device   *obd = class_exp2obd(exp);
3370         struct lmv_obd      *lmv = &obd->u.lmv;
3371         struct lmv_tgt_desc *tgt = lmv->tgts[0];
3372         int                  rc = 0;
3373         __u32                i;
3374         __u64                curspace, curinodes;
3375         ENTRY;
3376
3377         if (tgt == NULL ||
3378             tgt->ltd_exp == NULL ||
3379             !tgt->ltd_active ||
3380             lmv->desc.ld_tgt_count == 0) {
3381                 CERROR("master lmv inactive\n");
3382                 RETURN(-EIO);
3383         }
3384
3385         if (oqctl->qc_cmd != Q_GETOQUOTA) {
3386                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3387                 RETURN(rc);
3388         }
3389
3390         curspace = curinodes = 0;
3391         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3392                 int err;
3393                 tgt = lmv->tgts[i];
3394
3395                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3396                         continue;
3397
3398                 err = obd_quotactl(tgt->ltd_exp, oqctl);
3399                 if (err) {
3400                         CERROR("getquota on mdt %d failed. %d\n", i, err);
3401                         if (!rc)
3402                                 rc = err;
3403                 } else {
3404                         curspace += oqctl->qc_dqblk.dqb_curspace;
3405                         curinodes += oqctl->qc_dqblk.dqb_curinodes;
3406                 }
3407         }
3408         oqctl->qc_dqblk.dqb_curspace = curspace;
3409         oqctl->qc_dqblk.dqb_curinodes = curinodes;
3410
3411         RETURN(rc);
3412 }
3413
3414 int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp,
3415                    struct obd_quotactl *oqctl)
3416 {
3417         struct obd_device       *obd = class_exp2obd(exp);
3418         struct lmv_obd          *lmv = &obd->u.lmv;
3419         struct lmv_tgt_desc     *tgt;
3420         __u32                    i;
3421         int                      rc = 0;
3422         ENTRY;
3423
3424         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3425                 int err;
3426                 tgt = lmv->tgts[i];
3427                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
3428                         CERROR("lmv idx %d inactive\n", i);
3429                         RETURN(-EIO);
3430                 }
3431
3432                 err = obd_quotacheck(tgt->ltd_exp, oqctl);
3433                 if (err && !rc)
3434                         rc = err;
3435         }
3436
3437         RETURN(rc);
3438 }
3439
3440 int lmv_update_lsm_md(struct obd_export *exp, struct lmv_stripe_md *lsm,
3441                       struct mdt_body *body, ldlm_blocking_callback cb_blocking)
3442 {
3443         return lmv_revalidate_slaves(exp, body, lsm, cb_blocking, 0);
3444 }
3445
3446 int lmv_merge_attr(struct obd_export *exp, const struct lmv_stripe_md *lsm,
3447                    struct cl_attr *attr)
3448 {
3449         int i;
3450
3451         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3452                 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3453
3454                 CDEBUG(D_INFO, ""DFID" size %llu, nlink %u, atime %lu ctime"
3455                        "%lu, mtime %lu.\n", PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3456                        i_size_read(inode), inode->i_nlink,
3457                        LTIME_S(inode->i_atime), LTIME_S(inode->i_ctime),
3458                        LTIME_S(inode->i_mtime));
3459
3460                 /* for slave stripe, it needs to subtract nlink for . and .. */
3461                 if (i != 0)
3462                         attr->cat_nlink += inode->i_nlink - 2;
3463                 else
3464                         attr->cat_nlink = inode->i_nlink;
3465
3466                 attr->cat_size += i_size_read(inode);
3467
3468                 if (attr->cat_atime < LTIME_S(inode->i_atime))
3469                         attr->cat_atime = LTIME_S(inode->i_atime);
3470
3471                 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3472                         attr->cat_ctime = LTIME_S(inode->i_ctime);
3473
3474                 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3475                         attr->cat_mtime = LTIME_S(inode->i_mtime);
3476         }
3477         return 0;
3478 }
3479
3480 struct obd_ops lmv_obd_ops = {
3481         .o_owner                = THIS_MODULE,
3482         .o_setup                = lmv_setup,
3483         .o_cleanup              = lmv_cleanup,
3484         .o_precleanup           = lmv_precleanup,
3485         .o_process_config       = lmv_process_config,
3486         .o_connect              = lmv_connect,
3487         .o_disconnect           = lmv_disconnect,
3488         .o_statfs               = lmv_statfs,
3489         .o_get_info             = lmv_get_info,
3490         .o_set_info_async       = lmv_set_info_async,
3491         .o_packmd               = lmv_packmd,
3492         .o_unpackmd             = lmv_unpackmd,
3493         .o_notify               = lmv_notify,
3494         .o_get_uuid             = lmv_get_uuid,
3495         .o_iocontrol            = lmv_iocontrol,
3496         .o_quotacheck           = lmv_quotacheck,
3497         .o_quotactl             = lmv_quotactl
3498 };
3499
3500 struct md_ops lmv_md_ops = {
3501         .m_getstatus            = lmv_getstatus,
3502         .m_null_inode           = lmv_null_inode,
3503         .m_find_cbdata          = lmv_find_cbdata,
3504         .m_close                = lmv_close,
3505         .m_create               = lmv_create,
3506         .m_done_writing         = lmv_done_writing,
3507         .m_enqueue              = lmv_enqueue,
3508         .m_getattr              = lmv_getattr,
3509         .m_getxattr             = lmv_getxattr,
3510         .m_getattr_name         = lmv_getattr_name,
3511         .m_intent_lock          = lmv_intent_lock,
3512         .m_link                 = lmv_link,
3513         .m_rename               = lmv_rename,
3514         .m_setattr              = lmv_setattr,
3515         .m_setxattr             = lmv_setxattr,
3516         .m_fsync                = lmv_fsync,
3517         .m_read_page            = lmv_read_page,
3518         .m_unlink               = lmv_unlink,
3519         .m_init_ea_size         = lmv_init_ea_size,
3520         .m_cancel_unused        = lmv_cancel_unused,
3521         .m_set_lock_data        = lmv_set_lock_data,
3522         .m_lock_match           = lmv_lock_match,
3523         .m_get_lustre_md        = lmv_get_lustre_md,
3524         .m_free_lustre_md       = lmv_free_lustre_md,
3525         .m_update_lsm_md        = lmv_update_lsm_md,
3526         .m_merge_attr           = lmv_merge_attr,
3527         .m_set_open_replay_data = lmv_set_open_replay_data,
3528         .m_clear_open_replay_data = lmv_clear_open_replay_data,
3529         .m_renew_capa           = lmv_renew_capa,
3530         .m_unpack_capa          = lmv_unpack_capa,
3531         .m_get_remote_perm      = lmv_get_remote_perm,
3532         .m_intent_getattr_async = lmv_intent_getattr_async,
3533         .m_revalidate_lock      = lmv_revalidate_lock,
3534         .m_get_fid_from_lsm     = lmv_get_fid_from_lsm,
3535 };
3536
3537 int __init lmv_init(void)
3538 {
3539         return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3540                                    LUSTRE_LMV_NAME, NULL);
3541 }
3542
3543 static void lmv_exit(void)
3544 {
3545         class_unregister_type(LUSTRE_LMV_NAME);
3546 }
3547
3548 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3549 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3550 MODULE_LICENSE("GPL");
3551
3552 module_init(lmv_init);
3553 module_exit(lmv_exit);