Whamcloud - gitweb
LU-2675 lustre: remove lustre_lite.h
[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 /* This hash is only for testing purpose */
62 static inline unsigned int
63 lmv_hash_all_chars(unsigned int count, const char *name, int namelen)
64 {
65         unsigned int c = 0;
66         const unsigned char *p = (const unsigned char *)name;
67
68         while (--namelen >= 0)
69                 c += p[namelen];
70
71         c = c % count;
72
73         return c;
74 }
75
76 static inline unsigned int
77 lmv_hash_fnv1a(unsigned int count, const char *name, int namelen)
78 {
79         __u64   hash;
80
81         hash = lustre_hash_fnv_1a_64(name, namelen);
82
83         hash = hash % count;
84
85         return hash;
86 }
87
88 int lmv_name_to_stripe_index(__u32 lmv_hash_type, unsigned int stripe_count,
89                              const char *name, int namelen)
90 {
91         int     idx;
92         __u32   hash_type = lmv_hash_type & LMV_HASH_TYPE_MASK;
93
94         LASSERT(namelen > 0);
95         if (stripe_count <= 1)
96                 return 0;
97
98         /* for migrating object, always start from 0 stripe */
99         if (lmv_hash_type & LMV_HASH_FLAG_MIGRATION)
100                 return 0;
101
102         switch (hash_type) {
103         case LMV_HASH_TYPE_ALL_CHARS:
104                 idx = lmv_hash_all_chars(stripe_count, name, namelen);
105                 break;
106         case LMV_HASH_TYPE_FNV_1A_64:
107                 idx = lmv_hash_fnv1a(stripe_count, name, namelen);
108                 break;
109         default:
110                 idx = -EBADFD;
111                 break;
112         }
113
114         CDEBUG(D_INFO, "name %.*s hash_type %d idx %d\n", namelen, name,
115                hash_type, idx);
116
117         return idx;
118 }
119
120 static void lmv_activate_target(struct lmv_obd *lmv,
121                                 struct lmv_tgt_desc *tgt,
122                                 int activate)
123 {
124         if (tgt->ltd_active == activate)
125                 return;
126
127         tgt->ltd_active = activate;
128         lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
129 }
130
131 /**
132  * Error codes:
133  *
134  *  -EINVAL  : UUID can't be found in the LMV's target list
135  *  -ENOTCONN: The UUID is found, but the target connection is bad (!)
136  *  -EBADF   : The UUID is found, but the OBD of the wrong type (!)
137  */
138 static int lmv_set_mdc_active(struct lmv_obd *lmv,
139                               const struct obd_uuid *uuid,
140                               int activate)
141 {
142         struct lmv_tgt_desc     *tgt = NULL;
143         struct obd_device       *obd;
144         __u32                    i;
145         int                      rc = 0;
146         ENTRY;
147
148         CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
149                         lmv, uuid->uuid, activate);
150
151         spin_lock(&lmv->lmv_lock);
152         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
153                 tgt = lmv->tgts[i];
154                 if (tgt == NULL || tgt->ltd_exp == NULL)
155                         continue;
156
157                 CDEBUG(D_INFO, "Target idx %d is %s conn "LPX64"\n", i,
158                        tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
159
160                 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
161                         break;
162         }
163
164         if (i == lmv->desc.ld_tgt_count)
165                 GOTO(out_lmv_lock, rc = -EINVAL);
166
167         obd = class_exp2obd(tgt->ltd_exp);
168         if (obd == NULL)
169                 GOTO(out_lmv_lock, rc = -ENOTCONN);
170
171         CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
172                obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
173                obd->obd_type->typ_name, i);
174         LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
175
176         if (tgt->ltd_active == activate) {
177                 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
178                        activate ? "" : "in");
179                 GOTO(out_lmv_lock, rc);
180         }
181
182         CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
183                activate ? "" : "in");
184         lmv_activate_target(lmv, tgt, activate);
185         EXIT;
186
187  out_lmv_lock:
188         spin_unlock(&lmv->lmv_lock);
189         return rc;
190 }
191
192 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
193 {
194         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
195         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
196
197         return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
198 }
199
200 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
201                       enum obd_notify_event ev, void *data)
202 {
203         struct obd_connect_data *conn_data;
204         struct lmv_obd          *lmv = &obd->u.lmv;
205         struct obd_uuid         *uuid;
206         int                      rc = 0;
207         ENTRY;
208
209         if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
210                 CERROR("unexpected notification of %s %s!\n",
211                        watched->obd_type->typ_name,
212                        watched->obd_name);
213                 RETURN(-EINVAL);
214         }
215
216         uuid = &watched->u.cli.cl_target_uuid;
217         if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
218                 /*
219                  * Set MDC as active before notifying the observer, so the
220                  * observer can use the MDC normally.
221                  */
222                 rc = lmv_set_mdc_active(lmv, uuid,
223                                         ev == OBD_NOTIFY_ACTIVE);
224                 if (rc) {
225                         CERROR("%sactivation of %s failed: %d\n",
226                                ev == OBD_NOTIFY_ACTIVE ? "" : "de",
227                                uuid->uuid, rc);
228                         RETURN(rc);
229                 }
230         } else if (ev == OBD_NOTIFY_OCD) {
231                 conn_data = &watched->u.cli.cl_import->imp_connect_data;
232                 /*
233                  * XXX: Make sure that ocd_connect_flags from all targets are
234                  * the same. Otherwise one of MDTs runs wrong version or
235                  * something like this.  --umka
236                  */
237                 obd->obd_self_export->exp_connect_data = *conn_data;
238         }
239 #if 0
240         else if (ev == OBD_NOTIFY_DISCON) {
241                 /*
242                  * For disconnect event, flush fld cache for failout MDS case.
243                  */
244                 fld_client_flush(&lmv->lmv_fld);
245         }
246 #endif
247         /*
248          * Pass the notification up the chain.
249          */
250         if (obd->obd_observer)
251                 rc = obd_notify(obd->obd_observer, watched, ev, data);
252
253         RETURN(rc);
254 }
255
256 /**
257  * This is fake connect function. Its purpose is to initialize lmv and say
258  * caller that everything is okay. Real connection will be performed later.
259  */
260 static int lmv_connect(const struct lu_env *env,
261                        struct obd_export **exp, struct obd_device *obd,
262                        struct obd_uuid *cluuid, struct obd_connect_data *data,
263                        void *localdata)
264 {
265         struct lmv_obd        *lmv = &obd->u.lmv;
266         struct lustre_handle  conn = { 0 };
267         int                    rc = 0;
268         ENTRY;
269
270         /*
271          * We don't want to actually do the underlying connections more than
272          * once, so keep track.
273          */
274         lmv->refcount++;
275         if (lmv->refcount > 1) {
276                 *exp = NULL;
277                 RETURN(0);
278         }
279
280         rc = class_connect(&conn, obd, cluuid);
281         if (rc) {
282                 CERROR("class_connection() returned %d\n", rc);
283                 RETURN(rc);
284         }
285
286         *exp = class_conn2export(&conn);
287         class_export_get(*exp);
288
289         lmv->exp = *exp;
290         lmv->connected = 0;
291         lmv->cluuid = *cluuid;
292
293         if (data)
294                 lmv->conn_data = *data;
295
296         if (lmv->targets_proc_entry == NULL) {
297                 lmv->targets_proc_entry = lprocfs_seq_register("target_obds",
298                                                         obd->obd_proc_entry,
299                                                         NULL, NULL);
300                 if (IS_ERR(lmv->targets_proc_entry)) {
301                         CERROR("%s: cannot register "
302                                "/proc/fs/lustre/%s/%s/target_obds\n",
303                                obd->obd_name, obd->obd_type->typ_name,
304                                obd->obd_name);
305                         lmv->targets_proc_entry = NULL;
306                 }
307         }
308
309         /*
310          * All real clients should perform actual connection right away, because
311          * it is possible, that LMV will not have opportunity to connect targets
312          * and MDC stuff will be called directly, for instance while reading
313          * ../mdc/../kbytesfree procfs file, etc.
314          */
315         if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
316                 rc = lmv_check_connect(obd);
317
318         if (rc && lmv->targets_proc_entry != NULL)
319                 lprocfs_remove(&lmv->targets_proc_entry);
320         RETURN(rc);
321 }
322
323 static void lmv_set_timeouts(struct obd_device *obd)
324 {
325         struct lmv_obd          *lmv;
326         __u32                    i;
327
328         lmv = &obd->u.lmv;
329         if (lmv->server_timeout == 0)
330                 return;
331
332         if (lmv->connected == 0)
333                 return;
334
335         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
336                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
337
338                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
339                         continue;
340
341                 obd_set_info_async(NULL, tgt->ltd_exp, sizeof(KEY_INTERMDS),
342                                    KEY_INTERMDS, 0, NULL, NULL);
343         }
344 }
345
346 static int lmv_init_ea_size(struct obd_export *exp, int easize,
347                             int def_easize, int cookiesize, int def_cookiesize)
348 {
349         struct obd_device       *obd = exp->exp_obd;
350         struct lmv_obd          *lmv = &obd->u.lmv;
351         __u32                    i;
352         int                      rc = 0;
353         int                      change = 0;
354         ENTRY;
355
356         if (lmv->max_easize < easize) {
357                 lmv->max_easize = easize;
358                 change = 1;
359         }
360         if (lmv->max_def_easize < def_easize) {
361                 lmv->max_def_easize = def_easize;
362                 change = 1;
363         }
364         if (lmv->max_cookiesize < cookiesize) {
365                 lmv->max_cookiesize = cookiesize;
366                 change = 1;
367         }
368         if (lmv->max_def_cookiesize < def_cookiesize) {
369                 lmv->max_def_cookiesize = def_cookiesize;
370                 change = 1;
371         }
372         if (change == 0)
373                 RETURN(0);
374
375         if (lmv->connected == 0)
376                 RETURN(0);
377
378         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
379                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
380
381                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
382                         CWARN("%s: NULL export for %d\n", obd->obd_name, i);
383                         continue;
384                 }
385
386                 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize,
387                                      cookiesize, def_cookiesize);
388                 if (rc) {
389                         CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
390                                " rc = %d\n", obd->obd_name, i, rc);
391                         break;
392                 }
393         }
394         RETURN(rc);
395 }
396
397 #define MAX_STRING_SIZE 128
398
399 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
400 {
401         struct lmv_obd          *lmv = &obd->u.lmv;
402         struct obd_uuid         *cluuid = &lmv->cluuid;
403         struct obd_uuid          lmv_mdc_uuid = { "LMV_MDC_UUID" };
404         struct obd_device       *mdc_obd;
405         struct obd_export       *mdc_exp;
406         struct lu_fld_target     target;
407         int                      rc;
408         ENTRY;
409
410         mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
411                                         &obd->obd_uuid);
412         if (!mdc_obd) {
413                 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
414                 RETURN(-EINVAL);
415         }
416
417         CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
418                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
419                 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
420                 cluuid->uuid);
421
422         if (!mdc_obd->obd_set_up) {
423                 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
424                 RETURN(-EINVAL);
425         }
426
427         rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
428                          &lmv->conn_data, NULL);
429         if (rc) {
430                 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
431                 RETURN(rc);
432         }
433
434         /*
435          * Init fid sequence client for this mdc and add new fld target.
436          */
437         rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
438         if (rc)
439                 RETURN(rc);
440
441         target.ft_srv = NULL;
442         target.ft_exp = mdc_exp;
443         target.ft_idx = tgt->ltd_idx;
444
445         fld_client_add_target(&lmv->lmv_fld, &target);
446
447         rc = obd_register_observer(mdc_obd, obd);
448         if (rc) {
449                 obd_disconnect(mdc_exp);
450                 CERROR("target %s register_observer error %d\n",
451                        tgt->ltd_uuid.uuid, rc);
452                 RETURN(rc);
453         }
454
455         if (obd->obd_observer) {
456                 /*
457                  * Tell the observer about the new target.
458                  */
459                 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
460                                 OBD_NOTIFY_ACTIVE,
461                                 (void *)(tgt - lmv->tgts[0]));
462                 if (rc) {
463                         obd_disconnect(mdc_exp);
464                         RETURN(rc);
465                 }
466         }
467
468         tgt->ltd_active = 1;
469         tgt->ltd_exp = mdc_exp;
470         lmv->desc.ld_active_tgt_count++;
471
472         md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize,
473                         lmv->max_cookiesize, lmv->max_def_cookiesize);
474
475         CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
476                 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
477                 atomic_read(&obd->obd_refcount));
478
479         if (lmv->targets_proc_entry != NULL) {
480                 struct proc_dir_entry *mdc_symlink;
481
482                 LASSERT(mdc_obd->obd_type != NULL);
483                 LASSERT(mdc_obd->obd_type->typ_name != NULL);
484                 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
485                                                   lmv->targets_proc_entry,
486                                                   "../../../%s/%s",
487                                                   mdc_obd->obd_type->typ_name,
488                                                   mdc_obd->obd_name);
489                 if (mdc_symlink == NULL) {
490                         CERROR("cannot register LMV target "
491                                "/proc/fs/lustre/%s/%s/target_obds/%s\n",
492                                obd->obd_type->typ_name, obd->obd_name,
493                                mdc_obd->obd_name);
494                 }
495         }
496         RETURN(0);
497 }
498
499 static void lmv_del_target(struct lmv_obd *lmv, int index)
500 {
501         if (lmv->tgts[index] == NULL)
502                 return;
503
504         OBD_FREE_PTR(lmv->tgts[index]);
505         lmv->tgts[index] = NULL;
506         return;
507 }
508
509 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
510                            __u32 index, int gen)
511 {
512         struct lmv_obd      *lmv = &obd->u.lmv;
513         struct lmv_tgt_desc *tgt;
514         int                  orig_tgt_count = 0;
515         int                  rc = 0;
516         ENTRY;
517
518         CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
519
520         lmv_init_lock(lmv);
521
522         if (lmv->desc.ld_tgt_count == 0) {
523                 struct obd_device *mdc_obd;
524
525                 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
526                                                 &obd->obd_uuid);
527                 if (!mdc_obd) {
528                         lmv_init_unlock(lmv);
529                         CERROR("%s: Target %s not attached: rc = %d\n",
530                                obd->obd_name, uuidp->uuid, -EINVAL);
531                         RETURN(-EINVAL);
532                 }
533         }
534
535         if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
536                 tgt = lmv->tgts[index];
537                 CERROR("%s: UUID %s already assigned at LOV target index %d:"
538                        " rc = %d\n", obd->obd_name,
539                        obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
540                 lmv_init_unlock(lmv);
541                 RETURN(-EEXIST);
542         }
543
544         if (index >= lmv->tgts_size) {
545                 /* We need to reallocate the lmv target array. */
546                 struct lmv_tgt_desc **newtgts, **old = NULL;
547                 __u32 newsize = 1;
548                 __u32 oldsize = 0;
549
550                 while (newsize < index + 1)
551                         newsize = newsize << 1;
552                 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
553                 if (newtgts == NULL) {
554                         lmv_init_unlock(lmv);
555                         RETURN(-ENOMEM);
556                 }
557
558                 if (lmv->tgts_size) {
559                         memcpy(newtgts, lmv->tgts,
560                                sizeof(*newtgts) * lmv->tgts_size);
561                         old = lmv->tgts;
562                         oldsize = lmv->tgts_size;
563                 }
564
565                 lmv->tgts = newtgts;
566                 lmv->tgts_size = newsize;
567                 smp_rmb();
568                 if (old)
569                         OBD_FREE(old, sizeof(*old) * oldsize);
570
571                 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
572                        lmv->tgts_size);
573         }
574
575         OBD_ALLOC_PTR(tgt);
576         if (!tgt) {
577                 lmv_init_unlock(lmv);
578                 RETURN(-ENOMEM);
579         }
580
581         mutex_init(&tgt->ltd_fid_mutex);
582         tgt->ltd_idx = index;
583         tgt->ltd_uuid = *uuidp;
584         tgt->ltd_active = 0;
585         lmv->tgts[index] = tgt;
586         if (index >= lmv->desc.ld_tgt_count) {
587                 orig_tgt_count = lmv->desc.ld_tgt_count;
588                 lmv->desc.ld_tgt_count = index + 1;
589         }
590
591         if (lmv->connected) {
592                 rc = lmv_connect_mdc(obd, tgt);
593                 if (rc != 0) {
594                         spin_lock(&lmv->lmv_lock);
595                         if (lmv->desc.ld_tgt_count == index + 1)
596                                 lmv->desc.ld_tgt_count = orig_tgt_count;
597                         memset(tgt, 0, sizeof(*tgt));
598                         spin_unlock(&lmv->lmv_lock);
599                 } else {
600                         int easize = sizeof(struct lmv_stripe_md) +
601                                 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
602                         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
603                 }
604         }
605
606         lmv_init_unlock(lmv);
607         RETURN(rc);
608 }
609
610 int lmv_check_connect(struct obd_device *obd)
611 {
612         struct lmv_obd          *lmv = &obd->u.lmv;
613         struct lmv_tgt_desc     *tgt;
614         __u32                    i;
615         int                      rc;
616         int                      easize;
617         ENTRY;
618
619         if (lmv->connected)
620                 RETURN(0);
621
622         lmv_init_lock(lmv);
623         if (lmv->connected) {
624                 lmv_init_unlock(lmv);
625                 RETURN(0);
626         }
627
628         if (lmv->desc.ld_tgt_count == 0) {
629                 lmv_init_unlock(lmv);
630                 CERROR("%s: no targets configured.\n", obd->obd_name);
631                 RETURN(-EINVAL);
632         }
633
634         LASSERT(lmv->tgts != NULL);
635
636         if (lmv->tgts[0] == NULL) {
637                 lmv_init_unlock(lmv);
638                 CERROR("%s: no target configured for index 0.\n",
639                        obd->obd_name);
640                 RETURN(-EINVAL);
641         }
642
643         CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
644                lmv->cluuid.uuid, obd->obd_name);
645
646         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
647                 tgt = lmv->tgts[i];
648                 if (tgt == NULL)
649                         continue;
650                 rc = lmv_connect_mdc(obd, tgt);
651                 if (rc)
652                         GOTO(out_disc, rc);
653         }
654
655         lmv_set_timeouts(obd);
656         class_export_put(lmv->exp);
657         lmv->connected = 1;
658         easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
659         lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0);
660         lmv_init_unlock(lmv);
661         RETURN(0);
662
663  out_disc:
664         while (i-- > 0) {
665                 int rc2;
666                 tgt = lmv->tgts[i];
667                 if (tgt == NULL)
668                         continue;
669                 tgt->ltd_active = 0;
670                 if (tgt->ltd_exp) {
671                         --lmv->desc.ld_active_tgt_count;
672                         rc2 = obd_disconnect(tgt->ltd_exp);
673                         if (rc2) {
674                                 CERROR("LMV target %s disconnect on "
675                                        "MDC idx %d: error %d\n",
676                                        tgt->ltd_uuid.uuid, i, rc2);
677                         }
678                 }
679         }
680         class_disconnect(lmv->exp);
681         lmv_init_unlock(lmv);
682         RETURN(rc);
683 }
684
685 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
686 {
687         struct lmv_obd         *lmv = &obd->u.lmv;
688         struct obd_device      *mdc_obd;
689         int                     rc;
690         ENTRY;
691
692         LASSERT(tgt != NULL);
693         LASSERT(obd != NULL);
694
695         mdc_obd = class_exp2obd(tgt->ltd_exp);
696
697         if (mdc_obd) {
698                 mdc_obd->obd_force = obd->obd_force;
699                 mdc_obd->obd_fail = obd->obd_fail;
700                 mdc_obd->obd_no_recov = obd->obd_no_recov;
701         }
702
703         if (lmv->targets_proc_entry != NULL)
704                 lprocfs_remove_proc_entry(mdc_obd->obd_name,
705                                           lmv->targets_proc_entry);
706
707         rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
708         if (rc)
709                 CERROR("Can't finanize fids factory\n");
710
711         CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
712                tgt->ltd_exp->exp_obd->obd_name,
713                tgt->ltd_exp->exp_obd->obd_uuid.uuid);
714
715         obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
716         rc = obd_disconnect(tgt->ltd_exp);
717         if (rc) {
718                 if (tgt->ltd_active) {
719                         CERROR("Target %s disconnect error %d\n",
720                                tgt->ltd_uuid.uuid, rc);
721                 }
722         }
723
724         lmv_activate_target(lmv, tgt, 0);
725         tgt->ltd_exp = NULL;
726         RETURN(0);
727 }
728
729 static int lmv_disconnect(struct obd_export *exp)
730 {
731         struct obd_device       *obd = class_exp2obd(exp);
732         struct lmv_obd          *lmv = &obd->u.lmv;
733         int                      rc;
734         __u32                    i;
735         ENTRY;
736
737         if (!lmv->tgts)
738                 goto out_local;
739
740         /*
741          * Only disconnect the underlying layers on the final disconnect.
742          */
743         lmv->refcount--;
744         if (lmv->refcount != 0)
745                 goto out_local;
746
747         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
748                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
749                         continue;
750
751                 lmv_disconnect_mdc(obd, lmv->tgts[i]);
752         }
753
754         if (lmv->targets_proc_entry != NULL)
755                 lprocfs_remove(&lmv->targets_proc_entry);
756         else
757                 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
758                        obd->obd_type->typ_name, obd->obd_name);
759
760 out_local:
761         /*
762          * This is the case when no real connection is established by
763          * lmv_check_connect().
764          */
765         if (!lmv->connected)
766                 class_export_put(exp);
767         rc = class_disconnect(exp);
768         if (lmv->refcount == 0)
769                 lmv->connected = 0;
770         RETURN(rc);
771 }
772
773 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
774 {
775         struct obd_device       *obddev = class_exp2obd(exp);
776         struct lmv_obd          *lmv = &obddev->u.lmv;
777         struct getinfo_fid2path *gf;
778         struct lmv_tgt_desc     *tgt;
779         struct getinfo_fid2path *remote_gf = NULL;
780         int                     remote_gf_size = 0;
781         int                     rc;
782
783         gf = (struct getinfo_fid2path *)karg;
784         tgt = lmv_find_target(lmv, &gf->gf_fid);
785         if (IS_ERR(tgt))
786                 RETURN(PTR_ERR(tgt));
787
788 repeat_fid2path:
789         rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
790         if (rc != 0 && rc != -EREMOTE)
791                 GOTO(out_fid2path, rc);
792
793         /* If remote_gf != NULL, it means just building the
794          * path on the remote MDT, copy this path segement to gf */
795         if (remote_gf != NULL) {
796                 struct getinfo_fid2path *ori_gf;
797                 char *ptr;
798
799                 ori_gf = (struct getinfo_fid2path *)karg;
800                 if (strlen(ori_gf->gf_path) +
801                     strlen(gf->gf_path) > ori_gf->gf_pathlen)
802                         GOTO(out_fid2path, rc = -EOVERFLOW);
803
804                 ptr = ori_gf->gf_path;
805
806                 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
807                         strlen(ori_gf->gf_path));
808
809                 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
810                 ptr += strlen(gf->gf_path);
811                 *ptr = '/';
812         }
813
814         CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
815                tgt->ltd_exp->exp_obd->obd_name,
816                gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
817                gf->gf_linkno);
818
819         if (rc == 0)
820                 GOTO(out_fid2path, rc);
821
822         /* sigh, has to go to another MDT to do path building further */
823         if (remote_gf == NULL) {
824                 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
825                 OBD_ALLOC(remote_gf, remote_gf_size);
826                 if (remote_gf == NULL)
827                         GOTO(out_fid2path, rc = -ENOMEM);
828                 remote_gf->gf_pathlen = PATH_MAX;
829         }
830
831         if (!fid_is_sane(&gf->gf_fid)) {
832                 CERROR("%s: invalid FID "DFID": rc = %d\n",
833                        tgt->ltd_exp->exp_obd->obd_name,
834                        PFID(&gf->gf_fid), -EINVAL);
835                 GOTO(out_fid2path, rc = -EINVAL);
836         }
837
838         tgt = lmv_find_target(lmv, &gf->gf_fid);
839         if (IS_ERR(tgt))
840                 GOTO(out_fid2path, rc = -EINVAL);
841
842         remote_gf->gf_fid = gf->gf_fid;
843         remote_gf->gf_recno = -1;
844         remote_gf->gf_linkno = -1;
845         memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
846         gf = remote_gf;
847         goto repeat_fid2path;
848
849 out_fid2path:
850         if (remote_gf != NULL)
851                 OBD_FREE(remote_gf, remote_gf_size);
852         RETURN(rc);
853 }
854
855 static int lmv_hsm_req_count(struct lmv_obd *lmv,
856                              const struct hsm_user_request *hur,
857                              const struct lmv_tgt_desc *tgt_mds)
858 {
859         __u32                    i;
860         int                      nr = 0;
861         struct lmv_tgt_desc     *curr_tgt;
862
863         /* count how many requests must be sent to the given target */
864         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
865                 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
866                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
867                         nr++;
868         }
869         return nr;
870 }
871
872 static void lmv_hsm_req_build(struct lmv_obd *lmv,
873                               struct hsm_user_request *hur_in,
874                               const struct lmv_tgt_desc *tgt_mds,
875                               struct hsm_user_request *hur_out)
876 {
877         __u32                    i, nr_out;
878         struct lmv_tgt_desc     *curr_tgt;
879
880         /* build the hsm_user_request for the given target */
881         hur_out->hur_request = hur_in->hur_request;
882         nr_out = 0;
883         for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
884                 curr_tgt = lmv_find_target(lmv,
885                                            &hur_in->hur_user_item[i].hui_fid);
886                 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
887                         hur_out->hur_user_item[nr_out] =
888                                                 hur_in->hur_user_item[i];
889                         nr_out++;
890                 }
891         }
892         hur_out->hur_request.hr_itemcount = nr_out;
893         memcpy(hur_data(hur_out), hur_data(hur_in),
894                hur_in->hur_request.hr_data_len);
895 }
896
897 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
898                                  struct lustre_kernelcomm *lk, void *uarg)
899 {
900         __u32                    i;
901         int                      rc;
902         struct kkuc_ct_data     *kcd = NULL;
903         ENTRY;
904
905         /* unregister request (call from llapi_hsm_copytool_fini) */
906         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
907                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
908
909                 if (tgt == NULL || tgt->ltd_exp == NULL)
910                         continue;
911                 /* best effort: try to clean as much as possible
912                  * (continue on error) */
913                 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
914         }
915
916         /* Whatever the result, remove copytool from kuc groups.
917          * Unreached coordinators will get EPIPE on next requests
918          * and will unregister automatically.
919          */
920         rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
921         if (kcd != NULL)
922                 OBD_FREE_PTR(kcd);
923
924         RETURN(rc);
925 }
926
927 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
928                                struct lustre_kernelcomm *lk, void *uarg)
929 {
930         struct file             *filp;
931         __u32                    i, j;
932         int                      err, rc;
933         bool                     any_set = false;
934         struct kkuc_ct_data     *kcd;
935         ENTRY;
936
937         /* All or nothing: try to register to all MDS.
938          * In case of failure, unregister from previous MDS,
939          * except if it because of inactive target. */
940         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
941                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
942
943                 if (tgt == NULL || tgt->ltd_exp == NULL)
944                         continue;
945                 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
946                 if (err) {
947                         if (tgt->ltd_active) {
948                                 /* permanent error */
949                                 CERROR("%s: iocontrol MDC %s on MDT"
950                                        " idx %d cmd %x: err = %d\n",
951                                        class_exp2obd(lmv->exp)->obd_name,
952                                        tgt->ltd_uuid.uuid, i, cmd, err);
953                                 rc = err;
954                                 lk->lk_flags |= LK_FLG_STOP;
955                                 /* unregister from previous MDS */
956                                 for (j = 0; j < i; j++) {
957                                         tgt = lmv->tgts[j];
958                                         if (tgt == NULL || tgt->ltd_exp == NULL)
959                                                 continue;
960                                         obd_iocontrol(cmd, tgt->ltd_exp, len,
961                                                       lk, uarg);
962                                 }
963                                 RETURN(rc);
964                         }
965                         /* else: transient error.
966                          * kuc will register to the missing MDT
967                          * when it is back */
968                 } else {
969                         any_set = true;
970                 }
971         }
972
973         if (!any_set)
974                 /* no registration done: return error */
975                 RETURN(-ENOTCONN);
976
977         /* at least one registration done, with no failure */
978         filp = fget(lk->lk_wfd);
979         if (filp == NULL)
980                 RETURN(-EBADF);
981
982         OBD_ALLOC_PTR(kcd);
983         if (kcd == NULL) {
984                 fput(filp);
985                 RETURN(-ENOMEM);
986         }
987         kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
988         kcd->kcd_uuid = lmv->cluuid;
989         kcd->kcd_archive = lk->lk_data;
990
991         rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
992         if (rc != 0) {
993                 if (filp != NULL)
994                         fput(filp);
995                 OBD_FREE_PTR(kcd);
996         }
997
998         RETURN(rc);
999 }
1000
1001
1002
1003
1004 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
1005                          int len, void *karg, void *uarg)
1006 {
1007         struct obd_device       *obddev = class_exp2obd(exp);
1008         struct lmv_obd          *lmv = &obddev->u.lmv;
1009         struct lmv_tgt_desc     *tgt = NULL;
1010         __u32                    i = 0;
1011         int                      rc = 0;
1012         int                      set = 0;
1013         __u32                    count = lmv->desc.ld_tgt_count;
1014         ENTRY;
1015
1016         if (count == 0)
1017                 RETURN(-ENOTTY);
1018
1019         switch (cmd) {
1020         case IOC_OBD_STATFS: {
1021                 struct obd_ioctl_data *data = karg;
1022                 struct obd_device *mdc_obd;
1023                 struct obd_statfs stat_buf = {0};
1024                 __u32 index;
1025
1026                 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1027                 if ((index >= count))
1028                         RETURN(-ENODEV);
1029
1030                 tgt = lmv->tgts[index];
1031                 if (tgt == NULL || !tgt->ltd_active)
1032                         RETURN(-ENODATA);
1033
1034                 mdc_obd = class_exp2obd(tgt->ltd_exp);
1035                 if (!mdc_obd)
1036                         RETURN(-EINVAL);
1037
1038                 /* copy UUID */
1039                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
1040                                  min((int) data->ioc_plen2,
1041                                      (int) sizeof(struct obd_uuid))))
1042                         RETURN(-EFAULT);
1043
1044                 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
1045                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1046                                 0);
1047                 if (rc)
1048                         RETURN(rc);
1049                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1050                                  min((int) data->ioc_plen1,
1051                                      (int) sizeof(stat_buf))))
1052                         RETURN(-EFAULT);
1053                 break;
1054         }
1055         case OBD_IOC_QUOTACTL: {
1056                 struct if_quotactl *qctl = karg;
1057                 struct obd_quotactl *oqctl;
1058
1059                 if (qctl->qc_valid == QC_MDTIDX) {
1060                         if (count <= qctl->qc_idx)
1061                                 RETURN(-EINVAL);
1062
1063                         tgt = lmv->tgts[qctl->qc_idx];
1064                         if (tgt == NULL || tgt->ltd_exp == NULL)
1065                                 RETURN(-EINVAL);
1066                 } else if (qctl->qc_valid == QC_UUID) {
1067                         for (i = 0; i < count; i++) {
1068                                 tgt = lmv->tgts[i];
1069                                 if (tgt == NULL)
1070                                         continue;
1071                                 if (!obd_uuid_equals(&tgt->ltd_uuid,
1072                                                      &qctl->obd_uuid))
1073                                         continue;
1074
1075                                 if (tgt->ltd_exp == NULL)
1076                                         RETURN(-EINVAL);
1077
1078                                 break;
1079                         }
1080                 } else {
1081                         RETURN(-EINVAL);
1082                 }
1083
1084                 if (i >= count)
1085                         RETURN(-EAGAIN);
1086
1087                 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
1088                 OBD_ALLOC_PTR(oqctl);
1089                 if (!oqctl)
1090                         RETURN(-ENOMEM);
1091
1092                 QCTL_COPY(oqctl, qctl);
1093                 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1094                 if (rc == 0) {
1095                         QCTL_COPY(qctl, oqctl);
1096                         qctl->qc_valid = QC_MDTIDX;
1097                         qctl->obd_uuid = tgt->ltd_uuid;
1098                 }
1099                 OBD_FREE_PTR(oqctl);
1100                 break;
1101         }
1102         case OBD_IOC_CHANGELOG_SEND:
1103         case OBD_IOC_CHANGELOG_CLEAR: {
1104                 struct ioc_changelog *icc = karg;
1105
1106                 if (icc->icc_mdtindex >= count)
1107                         RETURN(-ENODEV);
1108
1109                 tgt = lmv->tgts[icc->icc_mdtindex];
1110                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1111                         RETURN(-ENODEV);
1112                 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1113                 break;
1114         }
1115         case LL_IOC_GET_CONNECT_FLAGS: {
1116                 tgt = lmv->tgts[0];
1117                 if (tgt == NULL || tgt->ltd_exp == NULL)
1118                         RETURN(-ENODATA);
1119                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1120                 break;
1121         }
1122         case LL_IOC_FID2MDTIDX: {
1123                 struct lu_fid *fid = karg;
1124                 int             mdt_index;
1125
1126                 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1127                 if (rc != 0)
1128                         RETURN(rc);
1129
1130                 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1131                  * point to user space memory for FID2MDTIDX. */
1132                 *(__u32 *)uarg = mdt_index;
1133                 break;
1134         }
1135         case OBD_IOC_FID2PATH: {
1136                 rc = lmv_fid2path(exp, len, karg, uarg);
1137                 break;
1138         }
1139         case LL_IOC_HSM_STATE_GET:
1140         case LL_IOC_HSM_STATE_SET:
1141         case LL_IOC_HSM_ACTION: {
1142                 struct md_op_data       *op_data = karg;
1143
1144                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1145                 if (IS_ERR(tgt))
1146                         RETURN(PTR_ERR(tgt));
1147
1148                 if (tgt->ltd_exp == NULL)
1149                         RETURN(-EINVAL);
1150
1151                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1152                 break;
1153         }
1154         case LL_IOC_HSM_PROGRESS: {
1155                 const struct hsm_progress_kernel *hpk = karg;
1156
1157                 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1158                 if (IS_ERR(tgt))
1159                         RETURN(PTR_ERR(tgt));
1160                 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1161                 break;
1162         }
1163         case LL_IOC_HSM_REQUEST: {
1164                 struct hsm_user_request *hur = karg;
1165                 unsigned int reqcount = hur->hur_request.hr_itemcount;
1166
1167                 if (reqcount == 0)
1168                         RETURN(0);
1169
1170                 /* if the request is about a single fid
1171                  * or if there is a single MDS, no need to split
1172                  * the request. */
1173                 if (reqcount == 1 || count == 1) {
1174                         tgt = lmv_find_target(lmv,
1175                                               &hur->hur_user_item[0].hui_fid);
1176                         if (IS_ERR(tgt))
1177                                 RETURN(PTR_ERR(tgt));
1178                         rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1179                 } else {
1180                         /* split fid list to their respective MDS */
1181                         for (i = 0; i < count; i++) {
1182                                 unsigned int            nr, reqlen;
1183                                 int                     rc1;
1184                                 struct hsm_user_request *req;
1185
1186                                 tgt = lmv->tgts[i];
1187                                 if (tgt == NULL || tgt->ltd_exp == NULL)
1188                                         continue;
1189
1190                                 nr = lmv_hsm_req_count(lmv, hur, tgt);
1191                                 if (nr == 0) /* nothing for this MDS */
1192                                         continue;
1193
1194                                 /* build a request with fids for this MDS */
1195                                 reqlen = offsetof(typeof(*hur),
1196                                                   hur_user_item[nr])
1197                                                 + hur->hur_request.hr_data_len;
1198                                 OBD_ALLOC_LARGE(req, reqlen);
1199                                 if (req == NULL)
1200                                         RETURN(-ENOMEM);
1201
1202                                 lmv_hsm_req_build(lmv, hur, tgt, req);
1203
1204                                 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1205                                                     req, uarg);
1206                                 if (rc1 != 0 && rc == 0)
1207                                         rc = rc1;
1208                                 OBD_FREE_LARGE(req, reqlen);
1209                         }
1210                 }
1211                 break;
1212         }
1213         case LL_IOC_LOV_SWAP_LAYOUTS: {
1214                 struct md_op_data       *op_data = karg;
1215                 struct lmv_tgt_desc     *tgt1, *tgt2;
1216
1217                 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1218                 if (IS_ERR(tgt1))
1219                         RETURN(PTR_ERR(tgt1));
1220
1221                 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1222                 if (IS_ERR(tgt2))
1223                         RETURN(PTR_ERR(tgt2));
1224
1225                 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1226                         RETURN(-EINVAL);
1227
1228                 /* only files on same MDT can have their layouts swapped */
1229                 if (tgt1->ltd_idx != tgt2->ltd_idx)
1230                         RETURN(-EPERM);
1231
1232                 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1233                 break;
1234         }
1235         case LL_IOC_HSM_CT_START: {
1236                 struct lustre_kernelcomm *lk = karg;
1237                 if (lk->lk_flags & LK_FLG_STOP)
1238                         rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1239                 else
1240                         rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1241                 break;
1242         }
1243         default:
1244                 for (i = 0; i < count; i++) {
1245                         struct obd_device *mdc_obd;
1246                         int err;
1247
1248                         tgt = lmv->tgts[i];
1249                         if (tgt == NULL || tgt->ltd_exp == NULL)
1250                                 continue;
1251                         /* ll_umount_begin() sets force flag but for lmv, not
1252                          * mdc. Let's pass it through */
1253                         mdc_obd = class_exp2obd(tgt->ltd_exp);
1254                         mdc_obd->obd_force = obddev->obd_force;
1255                         err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1256                         if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1257                                 RETURN(err);
1258                         } else if (err) {
1259                                 if (tgt->ltd_active) {
1260                                         CERROR("error: iocontrol MDC %s on MDT"
1261                                                " idx %d cmd %x: err = %d\n",
1262                                                tgt->ltd_uuid.uuid, i, cmd, err);
1263                                         if (!rc)
1264                                                 rc = err;
1265                                 }
1266                         } else
1267                                 set = 1;
1268                 }
1269                 if (!set && !rc)
1270                         rc = -EIO;
1271         }
1272         RETURN(rc);
1273 }
1274
1275 #if 0
1276 static int lmv_all_chars_policy(int count, const char *name,
1277                                 int len)
1278 {
1279         unsigned int c = 0;
1280
1281         while (len > 0)
1282                 c += name[--len];
1283         c = c % count;
1284         return c;
1285 }
1286
1287 static int lmv_nid_policy(struct lmv_obd *lmv)
1288 {
1289         struct obd_import *imp;
1290         __u32              id;
1291
1292         /*
1293          * XXX: To get nid we assume that underlying obd device is mdc.
1294          */
1295         imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1296         id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1297         return id % lmv->desc.ld_tgt_count;
1298 }
1299
1300 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1301                           placement_policy_t placement)
1302 {
1303         switch (placement) {
1304         case PLACEMENT_CHAR_POLICY:
1305                 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1306                                             op_data->op_name,
1307                                             op_data->op_namelen);
1308         case PLACEMENT_NID_POLICY:
1309                 return lmv_nid_policy(lmv);
1310
1311         default:
1312                 break;
1313         }
1314
1315         CERROR("Unsupported placement policy %x\n", placement);
1316         return -EINVAL;
1317 }
1318 #endif
1319
1320 /**
1321  * This is _inode_ placement policy function (not name).
1322  */
1323 static int lmv_placement_policy(struct obd_device *obd,
1324                                 struct md_op_data *op_data, u32 *mds)
1325 {
1326         struct lmv_obd          *lmv = &obd->u.lmv;
1327         ENTRY;
1328
1329         LASSERT(mds != NULL);
1330
1331         if (lmv->desc.ld_tgt_count == 1) {
1332                 *mds = 0;
1333                 RETURN(0);
1334         }
1335
1336         /**
1337          * If stripe_offset is provided during setdirstripe
1338          * (setdirstripe -i xx), xx MDS will be choosen.
1339          */
1340         if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1341                 struct lmv_user_md *lum;
1342
1343                 lum = op_data->op_data;
1344
1345                 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1346                         *mds = le32_to_cpu(lum->lum_stripe_offset);
1347                 } else {
1348                         /* -1 means default, which will be in the same MDT with
1349                          * the stripe */
1350                         *mds = op_data->op_mds;
1351                         lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1352                 }
1353         } else {
1354                 /* Allocate new fid on target according to operation type and
1355                  * parent home mds. */
1356                 *mds = op_data->op_mds;
1357         }
1358
1359         RETURN(0);
1360 }
1361
1362 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1363 {
1364         struct lmv_tgt_desc     *tgt;
1365         int                      rc;
1366         ENTRY;
1367
1368         tgt = lmv_get_target(lmv, mds, NULL);
1369         if (IS_ERR(tgt))
1370                 RETURN(PTR_ERR(tgt));
1371
1372         /*
1373          * New seq alloc and FLD setup should be atomic. Otherwise we may find
1374          * on server that seq in new allocated fid is not yet known.
1375          */
1376         mutex_lock(&tgt->ltd_fid_mutex);
1377
1378         if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1379                 GOTO(out, rc = -ENODEV);
1380
1381         /*
1382          * Asking underlying tgt layer to allocate new fid.
1383          */
1384         rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1385         if (rc > 0) {
1386                 LASSERT(fid_is_sane(fid));
1387                 rc = 0;
1388         }
1389
1390         EXIT;
1391 out:
1392         mutex_unlock(&tgt->ltd_fid_mutex);
1393         return rc;
1394 }
1395
1396 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1397                   struct lu_fid *fid, struct md_op_data *op_data)
1398 {
1399         struct obd_device     *obd = class_exp2obd(exp);
1400         struct lmv_obd        *lmv = &obd->u.lmv;
1401         u32                    mds = 0;
1402         int                    rc;
1403         ENTRY;
1404
1405         LASSERT(op_data != NULL);
1406         LASSERT(fid != NULL);
1407
1408         rc = lmv_placement_policy(obd, op_data, &mds);
1409         if (rc) {
1410                 CERROR("Can't get target for allocating fid, "
1411                        "rc %d\n", rc);
1412                 RETURN(rc);
1413         }
1414
1415         rc = __lmv_fid_alloc(lmv, fid, mds);
1416         if (rc) {
1417                 CERROR("Can't alloc new fid, rc %d\n", rc);
1418                 RETURN(rc);
1419         }
1420
1421         RETURN(rc);
1422 }
1423
1424 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1425 {
1426         struct lmv_obd  *lmv = &obd->u.lmv;
1427         struct lmv_desc *desc;
1428         int             rc;
1429         ENTRY;
1430
1431         if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1432                 CERROR("LMV setup requires a descriptor\n");
1433                 RETURN(-EINVAL);
1434         }
1435
1436         desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1437         if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1438                 CERROR("Lmv descriptor size wrong: %d > %d\n",
1439                        (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1440                 RETURN(-EINVAL);
1441         }
1442
1443         OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * 32);
1444         if (lmv->tgts == NULL)
1445                 RETURN(-ENOMEM);
1446         lmv->tgts_size = 32;
1447
1448         obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1449         lmv->desc.ld_tgt_count = 0;
1450         lmv->desc.ld_active_tgt_count = 0;
1451         lmv->max_cookiesize = 0;
1452         lmv->max_def_easize = 0;
1453         lmv->max_easize = 0;
1454         lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1455
1456         spin_lock_init(&lmv->lmv_lock);
1457         mutex_init(&lmv->init_mutex);
1458
1459 #ifdef LPROCFS
1460         obd->obd_vars = lprocfs_lmv_obd_vars;
1461         lprocfs_obd_setup(obd);
1462         lprocfs_alloc_md_stats(obd, 0);
1463         rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1464                                 0444, &lmv_proc_target_fops, obd);
1465         if (rc)
1466                 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1467                       obd->obd_name, rc);
1468 #endif
1469         rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1470                              LUSTRE_CLI_FLD_HASH_DHT);
1471         if (rc) {
1472                 CERROR("Can't init FLD, err %d\n", rc);
1473                 GOTO(out, rc);
1474         }
1475
1476         RETURN(0);
1477
1478 out:
1479         return rc;
1480 }
1481
1482 static int lmv_cleanup(struct obd_device *obd)
1483 {
1484         struct lmv_obd   *lmv = &obd->u.lmv;
1485         ENTRY;
1486
1487         fld_client_fini(&lmv->lmv_fld);
1488         if (lmv->tgts != NULL) {
1489                 int i;
1490                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1491                         if (lmv->tgts[i] == NULL)
1492                                 continue;
1493                         lmv_del_target(lmv, i);
1494                 }
1495                 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1496                 lmv->tgts_size = 0;
1497         }
1498         RETURN(0);
1499 }
1500
1501 static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf)
1502 {
1503         struct lustre_cfg       *lcfg = buf;
1504         struct obd_uuid         obd_uuid;
1505         int                     gen;
1506         __u32                   index;
1507         int                     rc;
1508         ENTRY;
1509
1510         switch (lcfg->lcfg_command) {
1511         case LCFG_ADD_MDC:
1512                 /* modify_mdc_tgts add 0:lustre-clilmv  1:lustre-MDT0000_UUID
1513                  * 2:0  3:1  4:lustre-MDT0000-mdc_UUID */
1514                 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1515                         GOTO(out, rc = -EINVAL);
1516
1517                 obd_str2uuid(&obd_uuid,  lustre_cfg_buf(lcfg, 1));
1518
1519                 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1520                         GOTO(out, rc = -EINVAL);
1521                 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1522                         GOTO(out, rc = -EINVAL);
1523                 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1524                 GOTO(out, rc);
1525         default:
1526                 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1527                 GOTO(out, rc = -EINVAL);
1528         }
1529 out:
1530         RETURN(rc);
1531 }
1532
1533 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1534                       struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1535 {
1536         struct obd_device       *obd = class_exp2obd(exp);
1537         struct lmv_obd          *lmv = &obd->u.lmv;
1538         struct obd_statfs       *temp;
1539         int                      rc = 0;
1540         __u32                    i;
1541         ENTRY;
1542
1543         rc = lmv_check_connect(obd);
1544         if (rc)
1545                 RETURN(rc);
1546
1547         OBD_ALLOC(temp, sizeof(*temp));
1548         if (temp == NULL)
1549                 RETURN(-ENOMEM);
1550
1551         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1552                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1553                         continue;
1554
1555                 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1556                                 max_age, flags);
1557                 if (rc) {
1558                         CERROR("can't stat MDS #%d (%s), error %d\n", i,
1559                                lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1560                                rc);
1561                         GOTO(out_free_temp, rc);
1562                 }
1563
1564                 if (i == 0) {
1565                         *osfs = *temp;
1566                         /* If the statfs is from mount, it will needs
1567                          * retrieve necessary information from MDT0.
1568                          * i.e. mount does not need the merged osfs
1569                          * from all of MDT.
1570                          * And also clients can be mounted as long as
1571                          * MDT0 is in service*/
1572                         if (flags & OBD_STATFS_FOR_MDT0)
1573                                 GOTO(out_free_temp, rc);
1574                 } else {
1575                         osfs->os_bavail += temp->os_bavail;
1576                         osfs->os_blocks += temp->os_blocks;
1577                         osfs->os_ffree += temp->os_ffree;
1578                         osfs->os_files += temp->os_files;
1579                 }
1580         }
1581
1582         EXIT;
1583 out_free_temp:
1584         OBD_FREE(temp, sizeof(*temp));
1585         return rc;
1586 }
1587
1588 static int lmv_getstatus(struct obd_export *exp,
1589                          struct lu_fid *fid,
1590                          struct obd_capa **pc)
1591 {
1592         struct obd_device    *obd = exp->exp_obd;
1593         struct lmv_obd       *lmv = &obd->u.lmv;
1594         int                   rc;
1595         ENTRY;
1596
1597         rc = lmv_check_connect(obd);
1598         if (rc)
1599                 RETURN(rc);
1600
1601         rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid, pc);
1602         RETURN(rc);
1603 }
1604
1605 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1606                         struct obd_capa *oc, obd_valid valid, const char *name,
1607                         const char *input, int input_size, int output_size,
1608                         int flags, struct ptlrpc_request **request)
1609 {
1610         struct obd_device      *obd = exp->exp_obd;
1611         struct lmv_obd         *lmv = &obd->u.lmv;
1612         struct lmv_tgt_desc    *tgt;
1613         int                     rc;
1614         ENTRY;
1615
1616         rc = lmv_check_connect(obd);
1617         if (rc)
1618                 RETURN(rc);
1619
1620         tgt = lmv_find_target(lmv, fid);
1621         if (IS_ERR(tgt))
1622                 RETURN(PTR_ERR(tgt));
1623
1624         rc = md_getxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1625                          input_size, output_size, flags, request);
1626
1627         RETURN(rc);
1628 }
1629
1630 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1631                         struct obd_capa *oc, obd_valid valid, const char *name,
1632                         const char *input, int input_size, int output_size,
1633                         int flags, __u32 suppgid,
1634                         struct ptlrpc_request **request)
1635 {
1636         struct obd_device      *obd = exp->exp_obd;
1637         struct lmv_obd         *lmv = &obd->u.lmv;
1638         struct lmv_tgt_desc    *tgt;
1639         int                     rc;
1640         ENTRY;
1641
1642         rc = lmv_check_connect(obd);
1643         if (rc)
1644                 RETURN(rc);
1645
1646         tgt = lmv_find_target(lmv, fid);
1647         if (IS_ERR(tgt))
1648                 RETURN(PTR_ERR(tgt));
1649
1650         rc = md_setxattr(tgt->ltd_exp, fid, oc, valid, name, input,
1651                          input_size, output_size, flags, suppgid,
1652                          request);
1653
1654         RETURN(rc);
1655 }
1656
1657 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1658                        struct ptlrpc_request **request)
1659 {
1660         struct obd_device       *obd = exp->exp_obd;
1661         struct lmv_obd          *lmv = &obd->u.lmv;
1662         struct lmv_tgt_desc     *tgt;
1663         int                      rc;
1664         ENTRY;
1665
1666         rc = lmv_check_connect(obd);
1667         if (rc)
1668                 RETURN(rc);
1669
1670         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1671         if (IS_ERR(tgt))
1672                 RETURN(PTR_ERR(tgt));
1673
1674         if (op_data->op_flags & MF_GET_MDT_IDX) {
1675                 op_data->op_mds = tgt->ltd_idx;
1676                 RETURN(0);
1677         }
1678
1679         rc = md_getattr(tgt->ltd_exp, op_data, request);
1680
1681         RETURN(rc);
1682 }
1683
1684 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1685 {
1686         struct obd_device   *obd = exp->exp_obd;
1687         struct lmv_obd      *lmv = &obd->u.lmv;
1688         __u32                i;
1689         int                  rc;
1690         ENTRY;
1691
1692         rc = lmv_check_connect(obd);
1693         if (rc)
1694                 RETURN(rc);
1695
1696         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1697
1698         /*
1699          * With DNE every object can have two locks in different namespaces:
1700          * lookup lock in space of MDT storing direntry and update/open lock in
1701          * space of MDT storing inode.
1702          */
1703         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1704                 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1705                         continue;
1706                 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1707         }
1708
1709         RETURN(0);
1710 }
1711
1712 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1713                            ldlm_iterator_t it, void *data)
1714 {
1715         struct obd_device       *obd = exp->exp_obd;
1716         struct lmv_obd          *lmv = &obd->u.lmv;
1717         int                     i;
1718         int                     tgt;
1719         int                     rc;
1720         ENTRY;
1721
1722         rc = lmv_check_connect(obd);
1723         if (rc)
1724                 RETURN(rc);
1725
1726         CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1727
1728         /*
1729          * With DNE every object can have two locks in different namespaces:
1730          * lookup lock in space of MDT storing direntry and update/open lock in
1731          * space of MDT storing inode.  Try the MDT that the FID maps to first,
1732          * since this can be easily found, and only try others if that fails.
1733          */
1734         for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1735              i < lmv->desc.ld_tgt_count;
1736              i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1737                 if (tgt < 0) {
1738                         CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1739                                obd->obd_name, PFID(fid), tgt);
1740                         tgt = 0;
1741                 }
1742
1743                 if (lmv->tgts[tgt] == NULL ||
1744                     lmv->tgts[tgt]->ltd_exp == NULL)
1745                         continue;
1746
1747                 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1748                 if (rc)
1749                         RETURN(rc);
1750         }
1751
1752         RETURN(rc);
1753 }
1754
1755
1756 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1757                      struct md_open_data *mod, struct ptlrpc_request **request)
1758 {
1759         struct obd_device     *obd = exp->exp_obd;
1760         struct lmv_obd        *lmv = &obd->u.lmv;
1761         struct lmv_tgt_desc   *tgt;
1762         int                    rc;
1763         ENTRY;
1764
1765         rc = lmv_check_connect(obd);
1766         if (rc)
1767                 RETURN(rc);
1768
1769         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1770         if (IS_ERR(tgt))
1771                 RETURN(PTR_ERR(tgt));
1772
1773         CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1774         rc = md_close(tgt->ltd_exp, op_data, mod, request);
1775         RETURN(rc);
1776 }
1777
1778 /**
1779  * Choosing the MDT by name or FID in @op_data.
1780  * For non-striped directory, it will locate MDT by fid.
1781  * For striped-directory, it will locate MDT by name. And also
1782  * it will reset op_fid1 with the FID of the choosen stripe.
1783  **/
1784 struct lmv_tgt_desc *
1785 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1786                            const char *name, int namelen, struct lu_fid *fid,
1787                            u32 *mds)
1788 {
1789         struct lmv_tgt_desc     *tgt;
1790         const struct lmv_oinfo  *oinfo;
1791
1792         oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1793         if (IS_ERR(oinfo))
1794                 RETURN(ERR_CAST(oinfo));
1795         *fid = oinfo->lmo_fid;
1796         *mds = oinfo->lmo_mds;
1797         tgt = lmv_get_target(lmv, *mds, NULL);
1798
1799         CDEBUG(D_INFO, "locate on mds %u "DFID"\n", *mds, PFID(fid));
1800         return tgt;
1801 }
1802
1803 /**
1804  * Locate mds by fid or name
1805  *
1806  * For striped directory (lsm != NULL), it will locate the stripe
1807  * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1808  * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1809  * walk through all of stripes to locate the entry.
1810  *
1811  * For normal direcotry, it will locate MDS by FID directly.
1812  * \param[in] lmv       LMV device
1813  * \param[in] op_data   client MD stack parameters, name, namelen
1814  *                      mds_num etc.
1815  * \param[in] fid       object FID used to locate MDS.
1816  *
1817  * retval               pointer to the lmv_tgt_desc if succeed.
1818  *                      ERR_PTR(errno) if failed.
1819  */
1820 struct lmv_tgt_desc*
1821 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1822                struct lu_fid *fid)
1823 {
1824         struct lmv_stripe_md    *lsm = op_data->op_mea1;
1825         struct lmv_tgt_desc     *tgt;
1826
1827         /* During creating VOLATILE file, it should honor the mdt
1828          * index if the file under striped dir is being restored, see
1829          * ct_restore(). */
1830         if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1831             (int)op_data->op_mds != -1 && lsm != NULL) {
1832                 int i;
1833                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1834                 if (IS_ERR(tgt))
1835                         return tgt;
1836
1837                 /* refill the right parent fid */
1838                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1839                         struct lmv_oinfo *oinfo;
1840
1841                         oinfo = &lsm->lsm_md_oinfo[i];
1842                         if (oinfo->lmo_mds == op_data->op_mds) {
1843                                 *fid = oinfo->lmo_fid;
1844                                 break;
1845                         }
1846                 }
1847
1848                 /* Hmm, can not find the stripe by mdt_index(op_mds) */
1849                 if (i == lsm->lsm_md_stripe_count)
1850                         tgt = ERR_PTR(-EINVAL);
1851
1852                 return tgt;
1853         }
1854
1855         if (lsm == NULL || op_data->op_namelen == 0) {
1856                 tgt = lmv_find_target(lmv, fid);
1857                 if (IS_ERR(tgt))
1858                         return tgt;
1859
1860                 op_data->op_mds = tgt->ltd_idx;
1861                 return tgt;
1862         }
1863
1864         return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1865                                           op_data->op_namelen, fid,
1866                                           &op_data->op_mds);
1867 }
1868
1869 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1870                const void *data, int datalen, int mode, __u32 uid,
1871                __u32 gid, cfs_cap_t cap_effective, __u64 rdev,
1872                struct ptlrpc_request **request)
1873 {
1874         struct obd_device       *obd = exp->exp_obd;
1875         struct lmv_obd          *lmv = &obd->u.lmv;
1876         struct lmv_tgt_desc     *tgt;
1877         int                      rc;
1878         ENTRY;
1879
1880         rc = lmv_check_connect(obd);
1881         if (rc)
1882                 RETURN(rc);
1883
1884         if (!lmv->desc.ld_active_tgt_count)
1885                 RETURN(-EIO);
1886
1887         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1888         if (IS_ERR(tgt))
1889                 RETURN(PTR_ERR(tgt));
1890
1891         CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1892                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1893                op_data->op_mds);
1894
1895         rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1896         if (rc)
1897                 RETURN(rc);
1898
1899         /* Send the create request to the MDT where the object
1900          * will be located */
1901         tgt = lmv_find_target(lmv, &op_data->op_fid2);
1902         if (IS_ERR(tgt))
1903                 RETURN(PTR_ERR(tgt));
1904
1905         op_data->op_mds = tgt->ltd_idx;
1906
1907         CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1908                PFID(&op_data->op_fid2), op_data->op_mds);
1909
1910         op_data->op_flags |= MF_MDC_CANCEL_FID1;
1911         rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1912                        cap_effective, rdev, request);
1913         if (rc == 0) {
1914                 if (*request == NULL)
1915                         RETURN(rc);
1916                 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1917         }
1918         RETURN(rc);
1919 }
1920
1921 static int lmv_done_writing(struct obd_export *exp,
1922                             struct md_op_data *op_data,
1923                             struct md_open_data *mod)
1924 {
1925         struct obd_device     *obd = exp->exp_obd;
1926         struct lmv_obd        *lmv = &obd->u.lmv;
1927         struct lmv_tgt_desc   *tgt;
1928         int                    rc;
1929         ENTRY;
1930
1931         rc = lmv_check_connect(obd);
1932         if (rc)
1933                 RETURN(rc);
1934
1935         tgt = lmv_find_target(lmv, &op_data->op_fid1);
1936         if (IS_ERR(tgt))
1937                 RETURN(PTR_ERR(tgt));
1938
1939         rc = md_done_writing(tgt->ltd_exp, op_data, mod);
1940         RETURN(rc);
1941 }
1942
1943 static int
1944 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1945             const union ldlm_policy_data *policy,
1946             struct lookup_intent *it, struct md_op_data *op_data,
1947             struct lustre_handle *lockh, __u64 extra_lock_flags)
1948 {
1949         struct obd_device        *obd = exp->exp_obd;
1950         struct lmv_obd           *lmv = &obd->u.lmv;
1951         struct lmv_tgt_desc      *tgt;
1952         int                       rc;
1953         ENTRY;
1954
1955         rc = lmv_check_connect(obd);
1956         if (rc)
1957                 RETURN(rc);
1958
1959         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1960                LL_IT2STR(it), PFID(&op_data->op_fid1));
1961
1962         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1963         if (IS_ERR(tgt))
1964                 RETURN(PTR_ERR(tgt));
1965
1966         CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n",
1967                LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1968
1969         rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1970                         extra_lock_flags);
1971
1972         RETURN(rc);
1973 }
1974
1975 static int
1976 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1977                  struct ptlrpc_request **preq)
1978 {
1979         struct ptlrpc_request   *req = NULL;
1980         struct obd_device       *obd = exp->exp_obd;
1981         struct lmv_obd          *lmv = &obd->u.lmv;
1982         struct lmv_tgt_desc     *tgt;
1983         struct mdt_body         *body;
1984         int                      rc;
1985         ENTRY;
1986
1987         rc = lmv_check_connect(obd);
1988         if (rc)
1989                 RETURN(rc);
1990
1991         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1992         if (IS_ERR(tgt))
1993                 RETURN(PTR_ERR(tgt));
1994
1995         CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1996                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1997                tgt->ltd_idx);
1998
1999         rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
2000         if (rc != 0)
2001                 RETURN(rc);
2002
2003         body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
2004         LASSERT(body != NULL);
2005
2006         if (body->mbo_valid & OBD_MD_MDS) {
2007                 struct lu_fid rid = body->mbo_fid1;
2008                 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
2009                        PFID(&rid));
2010
2011                 tgt = lmv_find_target(lmv, &rid);
2012                 if (IS_ERR(tgt)) {
2013                         ptlrpc_req_finished(*preq);
2014                         preq = NULL;
2015                         RETURN(PTR_ERR(tgt));
2016                 }
2017
2018                 op_data->op_fid1 = rid;
2019                 op_data->op_valid |= OBD_MD_FLCROSSREF;
2020                 op_data->op_namelen = 0;
2021                 op_data->op_name = NULL;
2022                 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
2023                 ptlrpc_req_finished(*preq);
2024                 *preq = req;
2025         }
2026
2027         RETURN(rc);
2028 }
2029
2030 #define md_op_data_fid(op_data, fl)                     \
2031         (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
2032          fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
2033          fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
2034          fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
2035          NULL)
2036
2037 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
2038                             struct md_op_data *op_data,
2039                             int op_tgt, ldlm_mode_t mode, int bits, int flag)
2040 {
2041         struct lu_fid          *fid = md_op_data_fid(op_data, flag);
2042         struct obd_device      *obd = exp->exp_obd;
2043         struct lmv_obd         *lmv = &obd->u.lmv;
2044         ldlm_policy_data_t      policy = {{ 0 }};
2045         int                     rc = 0;
2046         ENTRY;
2047
2048         if (!fid_is_sane(fid))
2049                 RETURN(0);
2050
2051         if (tgt == NULL) {
2052                 tgt = lmv_find_target(lmv, fid);
2053                 if (IS_ERR(tgt))
2054                         RETURN(PTR_ERR(tgt));
2055         }
2056
2057         if (tgt->ltd_idx != op_tgt) {
2058                 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
2059                 policy.l_inodebits.bits = bits;
2060                 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
2061                                       mode, LCF_ASYNC, NULL);
2062         } else {
2063                 CDEBUG(D_INODE,
2064                        "EARLY_CANCEL skip operation target %d on "DFID"\n",
2065                        op_tgt, PFID(fid));
2066                 op_data->op_flags |= flag;
2067                 rc = 0;
2068         }
2069
2070         RETURN(rc);
2071 }
2072
2073 /*
2074  * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2075  * op_data->op_fid2
2076  */
2077 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2078                     struct ptlrpc_request **request)
2079 {
2080         struct obd_device       *obd = exp->exp_obd;
2081         struct lmv_obd          *lmv = &obd->u.lmv;
2082         struct lmv_tgt_desc     *tgt;
2083         int                      rc;
2084         ENTRY;
2085
2086         rc = lmv_check_connect(obd);
2087         if (rc)
2088                 RETURN(rc);
2089
2090         LASSERT(op_data->op_namelen != 0);
2091
2092         CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2093                PFID(&op_data->op_fid2), op_data->op_namelen,
2094                op_data->op_name, PFID(&op_data->op_fid1));
2095
2096         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2097         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2098         op_data->op_cap = cfs_curproc_cap_pack();
2099         if (op_data->op_mea2 != NULL) {
2100                 struct lmv_stripe_md    *lsm = op_data->op_mea2;
2101                 const struct lmv_oinfo  *oinfo;
2102
2103                 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2104                                                 op_data->op_namelen);
2105                 if (IS_ERR(oinfo))
2106                         RETURN(PTR_ERR(oinfo));
2107
2108                 op_data->op_fid2 = oinfo->lmo_fid;
2109         }
2110
2111         tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2112         if (IS_ERR(tgt))
2113                 RETURN(PTR_ERR(tgt));
2114
2115         /*
2116          * Cancel UPDATE lock on child (fid1).
2117          */
2118         op_data->op_flags |= MF_MDC_CANCEL_FID2;
2119         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2120                               MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2121         if (rc != 0)
2122                 RETURN(rc);
2123
2124         rc = md_link(tgt->ltd_exp, op_data, request);
2125
2126         RETURN(rc);
2127 }
2128
2129 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2130                       const char *old, int oldlen, const char *new, int newlen,
2131                       struct ptlrpc_request **request)
2132 {
2133         struct obd_device       *obd = exp->exp_obd;
2134         struct lmv_obd          *lmv = &obd->u.lmv;
2135         struct lmv_tgt_desc     *src_tgt;
2136         int                     rc;
2137         ENTRY;
2138
2139         LASSERT(oldlen != 0);
2140
2141         CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2142                oldlen, old, PFID(&op_data->op_fid1),
2143                op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2144                newlen, new, PFID(&op_data->op_fid2),
2145                op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2146
2147         rc = lmv_check_connect(obd);
2148         if (rc)
2149                 RETURN(rc);
2150
2151         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2152         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2153         op_data->op_cap = cfs_curproc_cap_pack();
2154         if (op_data->op_cli_flags & CLI_MIGRATE) {
2155                 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2156                          PFID(&op_data->op_fid3));
2157                 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2158                 if (rc)
2159                         RETURN(rc);
2160                 src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid3);
2161         } else {
2162                 if (op_data->op_mea1 != NULL) {
2163                         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2164
2165                         src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2166                                                              oldlen,
2167                                                              &op_data->op_fid1,
2168                                                              &op_data->op_mds);
2169                         if (IS_ERR(src_tgt))
2170                                 RETURN(PTR_ERR(src_tgt));
2171                 } else {
2172                         src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2173                         if (IS_ERR(src_tgt))
2174                                 RETURN(PTR_ERR(src_tgt));
2175
2176                         op_data->op_mds = src_tgt->ltd_idx;
2177                 }
2178
2179                 if (op_data->op_mea2) {
2180                         struct lmv_stripe_md    *lsm = op_data->op_mea2;
2181                         const struct lmv_oinfo  *oinfo;
2182
2183                         oinfo = lsm_name_to_stripe_info(lsm, new, newlen);
2184                         if (IS_ERR(oinfo))
2185                                 RETURN(PTR_ERR(oinfo));
2186
2187                         op_data->op_fid2 = oinfo->lmo_fid;
2188                 }
2189         }
2190         if (IS_ERR(src_tgt))
2191                 RETURN(PTR_ERR(src_tgt));
2192
2193         /*
2194          * LOOKUP lock on src child (fid3) should also be cancelled for
2195          * src_tgt in mdc_rename.
2196          */
2197         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2198
2199         /*
2200          * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2201          * own target.
2202          */
2203         rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2204                               LCK_EX, MDS_INODELOCK_UPDATE,
2205                               MF_MDC_CANCEL_FID2);
2206
2207         if (rc != 0)
2208                 RETURN(rc);
2209         /*
2210          * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2211          */
2212         if (fid_is_sane(&op_data->op_fid3)) {
2213                 struct lmv_tgt_desc *tgt;
2214
2215                 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2216                 if (IS_ERR(tgt))
2217                         RETURN(PTR_ERR(tgt));
2218
2219                 /* Cancel LOOKUP lock on its parent */
2220                 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2221                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2222                                       MF_MDC_CANCEL_FID3);
2223                 if (rc != 0)
2224                         RETURN(rc);
2225
2226                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2227                                       LCK_EX, MDS_INODELOCK_FULL,
2228                                       MF_MDC_CANCEL_FID3);
2229                 if (rc != 0)
2230                         RETURN(rc);
2231         }
2232
2233         /*
2234          * Cancel all the locks on tgt child (fid4).
2235          */
2236         if (fid_is_sane(&op_data->op_fid4))
2237                 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2238                                       LCK_EX, MDS_INODELOCK_FULL,
2239                                       MF_MDC_CANCEL_FID4);
2240
2241         CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1),
2242                op_data->op_mds, PFID(&op_data->op_fid2));
2243
2244         rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2245                        request);
2246
2247         RETURN(rc);
2248 }
2249
2250 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2251                        void *ea, int ealen, void *ea2, int ea2len,
2252                        struct ptlrpc_request **request,
2253                        struct md_open_data **mod)
2254 {
2255         struct obd_device       *obd = exp->exp_obd;
2256         struct lmv_obd          *lmv = &obd->u.lmv;
2257         struct lmv_tgt_desc     *tgt;
2258         int                      rc = 0;
2259         ENTRY;
2260
2261         rc = lmv_check_connect(obd);
2262         if (rc)
2263                 RETURN(rc);
2264
2265         CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2266                PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2267
2268         op_data->op_flags |= MF_MDC_CANCEL_FID1;
2269         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2270         if (IS_ERR(tgt))
2271                 RETURN(PTR_ERR(tgt));
2272
2273         rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, ea2,
2274                         ea2len, request, mod);
2275
2276         RETURN(rc);
2277 }
2278
2279 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2280                      struct obd_capa *oc, struct ptlrpc_request **request)
2281 {
2282         struct obd_device       *obd = exp->exp_obd;
2283         struct lmv_obd          *lmv = &obd->u.lmv;
2284         struct lmv_tgt_desc     *tgt;
2285         int                      rc;
2286         ENTRY;
2287
2288         rc = lmv_check_connect(obd);
2289         if (rc != 0)
2290                 RETURN(rc);
2291
2292         tgt = lmv_find_target(lmv, fid);
2293         if (IS_ERR(tgt))
2294                 RETURN(PTR_ERR(tgt));
2295
2296         rc = md_fsync(tgt->ltd_exp, fid, oc, request);
2297         RETURN(rc);
2298 }
2299
2300 /**
2301  * Get current minimum entry from striped directory
2302  *
2303  * This function will search the dir entry, whose hash value is the
2304  * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2305  * only being called for striped directory.
2306  *
2307  * \param[in] exp               export of LMV
2308  * \param[in] op_data           parameters transferred beween client MD stack
2309  *                              stripe_information will be included in this
2310  *                              parameter
2311  * \param[in] cb_op             ldlm callback being used in enqueue in
2312  *                              mdc_read_page
2313  * \param[in] hash_offset       the hash value, which is used to locate
2314  *                              minum(closet) dir entry
2315  * \param[in|out] stripe_offset the caller use this to indicate the stripe
2316  *                              index of last entry, so to avoid hash conflict
2317  *                              between stripes. It will also be used to
2318  *                              return the stripe index of current dir entry.
2319  * \param[in|out] entp          the minum entry and it also is being used
2320  *                              to input the last dir entry to resolve the
2321  *                              hash conflict
2322  *
2323  * \param[out] ppage            the page which holds the minum entry
2324  *
2325  * \retval                      = 0 get the entry successfully
2326  *                              negative errno (< 0) does not get the entry
2327  */
2328 static int lmv_get_min_striped_entry(struct obd_export *exp,
2329                                      struct md_op_data *op_data,
2330                                      struct md_callback *cb_op,
2331                                      __u64 hash_offset, int *stripe_offset,
2332                                      struct lu_dirent **entp,
2333                                      struct page **ppage)
2334 {
2335         struct obd_device       *obd = exp->exp_obd;
2336         struct lmv_obd          *lmv = &obd->u.lmv;
2337         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2338         struct lmv_tgt_desc     *tgt;
2339         int                     stripe_count;
2340         struct lu_dirent        *min_ent = NULL;
2341         struct page             *min_page = NULL;
2342         int                     min_idx = 0;
2343         int                     i;
2344         int                     rc = 0;
2345         ENTRY;
2346
2347         stripe_count = lsm->lsm_md_stripe_count;
2348         for (i = 0; i < stripe_count; i++) {
2349                 struct lu_dirent        *ent = NULL;
2350                 struct page             *page = NULL;
2351                 struct lu_dirpage       *dp;
2352                 __u64                   stripe_hash = hash_offset;
2353
2354                 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2355                 if (IS_ERR(tgt))
2356                         GOTO(out, rc = PTR_ERR(tgt));
2357
2358                 /* op_data will be shared by each stripe, so we need
2359                  * reset these value for each stripe */
2360                 op_data->op_stripe_offset = i;
2361                 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2362                 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2363                 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2364 next:
2365                 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2366                                   &page);
2367                 if (rc != 0)
2368                         GOTO(out, rc);
2369
2370                 dp = page_address(page);
2371                 for (ent = lu_dirent_start(dp); ent != NULL;
2372                      ent = lu_dirent_next(ent)) {
2373                         /* Skip dummy entry */
2374                         if (le16_to_cpu(ent->lde_namelen) == 0)
2375                                 continue;
2376
2377                         if (le64_to_cpu(ent->lde_hash) < hash_offset)
2378                                 continue;
2379
2380                         if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2381                             (*entp == ent || i < *stripe_offset))
2382                                 continue;
2383
2384                         /* skip . and .. for other stripes */
2385                         if (i != 0 &&
2386                             (strncmp(ent->lde_name, ".",
2387                                      le16_to_cpu(ent->lde_namelen)) == 0 ||
2388                              strncmp(ent->lde_name, "..",
2389                                      le16_to_cpu(ent->lde_namelen)) == 0))
2390                                 continue;
2391                         break;
2392                 }
2393
2394                 if (ent == NULL) {
2395                         stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2396
2397                         kunmap(page);
2398                         page_cache_release(page);
2399                         page = NULL;
2400
2401                         /* reach the end of current stripe, go to next stripe */
2402                         if (stripe_hash == MDS_DIR_END_OFF)
2403                                 continue;
2404                         else
2405                                 goto next;
2406                 }
2407
2408                 if (min_ent != NULL) {
2409                         if (le64_to_cpu(min_ent->lde_hash) >
2410                             le64_to_cpu(ent->lde_hash)) {
2411                                 min_ent = ent;
2412                                 kunmap(min_page);
2413                                 page_cache_release(min_page);
2414                                 min_idx = i;
2415                                 min_page = page;
2416                         } else {
2417                                 kunmap(page);
2418                                 page_cache_release(page);
2419                                 page = NULL;
2420                         }
2421                 } else {
2422                         min_ent = ent;
2423                         min_page = page;
2424                         min_idx = i;
2425                 }
2426         }
2427
2428 out:
2429         if (*ppage != NULL) {
2430                 kunmap(*ppage);
2431                 page_cache_release(*ppage);
2432         }
2433         *stripe_offset = min_idx;
2434         *entp = min_ent;
2435         *ppage = min_page;
2436         RETURN(rc);
2437 }
2438
2439 /**
2440  * Build dir entry page from a striped directory
2441  *
2442  * This function gets one entry by @offset from a striped directory. It will
2443  * read entries from all of stripes, and choose one closest to the required
2444  * offset(&offset). A few notes
2445  * 1. skip . and .. for non-zero stripes, because there can only have one .
2446  * and .. in a directory.
2447  * 2. op_data will be shared by all of stripes, instead of allocating new
2448  * one, so need to restore before reusing.
2449  * 3. release the entry page if that is not being chosen.
2450  *
2451  * \param[in] exp       obd export refer to LMV
2452  * \param[in] op_data   hold those MD parameters of read_entry
2453  * \param[in] cb_op     ldlm callback being used in enqueue in mdc_read_entry
2454  * \param[out] ldp      the entry being read
2455  * \param[out] ppage    the page holding the entry. Note: because the entry
2456  *                      will be accessed in upper layer, so we need hold the
2457  *                      page until the usages of entry is finished, see
2458  *                      ll_dir_entry_next.
2459  *
2460  * retval               =0 if get entry successfully
2461  *                      <0 cannot get entry
2462  */
2463 static int lmv_read_striped_page(struct obd_export *exp,
2464                                  struct md_op_data *op_data,
2465                                  struct md_callback *cb_op,
2466                                  __u64 offset, struct page **ppage)
2467 {
2468         struct obd_device       *obd = exp->exp_obd;
2469         struct lu_fid           master_fid = op_data->op_fid1;
2470         struct inode            *master_inode = op_data->op_data;
2471         __u64                   hash_offset = offset;
2472         struct lu_dirpage       *dp;
2473         struct page             *min_ent_page = NULL;
2474         struct page             *ent_page = NULL;
2475         struct lu_dirent        *ent;
2476         void                    *area;
2477         int                     ent_idx = 0;
2478         struct lu_dirent        *min_ent = NULL;
2479         struct lu_dirent        *last_ent;
2480         int                     left_bytes;
2481         int                     rc;
2482         ENTRY;
2483
2484         rc = lmv_check_connect(obd);
2485         if (rc)
2486                 RETURN(rc);
2487
2488         /* Allocate a page and read entries from all of stripes and fill
2489          * the page by hash order */
2490         ent_page = alloc_page(GFP_KERNEL);
2491         if (ent_page == NULL)
2492                 RETURN(-ENOMEM);
2493
2494         /* Initialize the entry page */
2495         dp = kmap(ent_page);
2496         memset(dp, 0, sizeof(*dp));
2497         dp->ldp_hash_start = cpu_to_le64(offset);
2498         dp->ldp_flags |= LDF_COLLIDE;
2499
2500         area = dp + 1;
2501         left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2502         ent = area;
2503         last_ent = ent;
2504         do {
2505                 __u16   ent_size;
2506
2507                 /* Find the minum entry from all sub-stripes */
2508                 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2509                                                &ent_idx, &min_ent,
2510                                                &min_ent_page);
2511                 if (rc != 0)
2512                         GOTO(out, rc);
2513
2514                 /* If it can not get minum entry, it means it already reaches
2515                  * the end of this directory */
2516                 if (min_ent == NULL) {
2517                         last_ent->lde_reclen = 0;
2518                         hash_offset = MDS_DIR_END_OFF;
2519                         GOTO(out, rc);
2520                 }
2521
2522                 ent_size = le16_to_cpu(min_ent->lde_reclen);
2523
2524                 /* the last entry lde_reclen is 0, but it might not
2525                  * the end of this entry of this temporay entry */
2526                 if (ent_size == 0)
2527                         ent_size = lu_dirent_calc_size(
2528                                         le16_to_cpu(min_ent->lde_namelen),
2529                                         le32_to_cpu(min_ent->lde_attrs));
2530                 if (ent_size > left_bytes) {
2531                         last_ent->lde_reclen = cpu_to_le16(0);
2532                         hash_offset = le64_to_cpu(min_ent->lde_hash);
2533                         GOTO(out, rc);
2534                 }
2535
2536                 memcpy(ent, min_ent, ent_size);
2537
2538                 /* Replace . with master FID and Replace .. with the parent FID
2539                  * of master object */
2540                 if (strncmp(ent->lde_name, ".",
2541                             le16_to_cpu(ent->lde_namelen)) == 0 &&
2542                     le16_to_cpu(ent->lde_namelen) == 1)
2543                         fid_cpu_to_le(&ent->lde_fid, &master_fid);
2544                 else if (strncmp(ent->lde_name, "..",
2545                                    le16_to_cpu(ent->lde_namelen)) == 0 &&
2546                            le16_to_cpu(ent->lde_namelen) == 2)
2547                         fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2548
2549                 left_bytes -= ent_size;
2550                 ent->lde_reclen = cpu_to_le16(ent_size);
2551                 last_ent = ent;
2552                 ent = (void *)ent + ent_size;
2553                 hash_offset = le64_to_cpu(min_ent->lde_hash);
2554                 if (hash_offset == MDS_DIR_END_OFF) {
2555                         last_ent->lde_reclen = 0;
2556                         break;
2557                 }
2558         } while (1);
2559 out:
2560         if (min_ent_page != NULL) {
2561                 kunmap(min_ent_page);
2562                 page_cache_release(min_ent_page);
2563         }
2564
2565         if (unlikely(rc != 0)) {
2566                 __free_page(ent_page);
2567                 ent_page = NULL;
2568         } else {
2569                 if (ent == area)
2570                         dp->ldp_flags |= LDF_EMPTY;
2571                 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2572                 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2573         }
2574
2575         /* We do not want to allocate md_op_data during each
2576          * dir entry reading, so op_data will be shared by every stripe,
2577          * then we need to restore it back to original value before
2578          * return to the upper layer */
2579         op_data->op_fid1 = master_fid;
2580         op_data->op_fid2 = master_fid;
2581         op_data->op_data = master_inode;
2582
2583         *ppage = ent_page;
2584
2585         RETURN(rc);
2586 }
2587
2588 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2589                   struct md_callback *cb_op, __u64 offset,
2590                   struct page **ppage)
2591 {
2592         struct obd_device       *obd = exp->exp_obd;
2593         struct lmv_obd          *lmv = &obd->u.lmv;
2594         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2595         struct lmv_tgt_desc     *tgt;
2596         int                     rc;
2597         ENTRY;
2598
2599         rc = lmv_check_connect(obd);
2600         if (rc != 0)
2601                 RETURN(rc);
2602
2603         if (unlikely(lsm != NULL)) {
2604                 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2605                 RETURN(rc);
2606         }
2607
2608         tgt = lmv_find_target(lmv, &op_data->op_fid1);
2609         if (IS_ERR(tgt))
2610                 RETURN(PTR_ERR(tgt));
2611
2612         rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2613
2614         RETURN(rc);
2615 }
2616
2617 /**
2618  * Unlink a file/directory
2619  *
2620  * Unlink a file or directory under the parent dir. The unlink request
2621  * usually will be sent to the MDT where the child is located, but if
2622  * the client does not have the child FID then request will be sent to the
2623  * MDT where the parent is located.
2624  *
2625  * If the parent is a striped directory then it also needs to locate which
2626  * stripe the name of the child is located, and replace the parent FID
2627  * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2628  * it will walk through all of sub-stripes until the child is being
2629  * unlinked finally.
2630  *
2631  * \param[in] exp       export refer to LMV
2632  * \param[in] op_data   different parameters transferred beween client
2633  *                      MD stacks, name, namelen, FIDs etc.
2634  *                      op_fid1 is the parent FID, op_fid2 is the child
2635  *                      FID.
2636  * \param[out] request  point to the request of unlink.
2637  *
2638  * retval               0 if succeed
2639  *                      negative errno if failed.
2640  */
2641 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2642                       struct ptlrpc_request **request)
2643 {
2644         struct obd_device       *obd = exp->exp_obd;
2645         struct lmv_obd          *lmv = &obd->u.lmv;
2646         struct lmv_tgt_desc     *tgt = NULL;
2647         struct lmv_tgt_desc     *parent_tgt = NULL;
2648         struct mdt_body         *body;
2649         int                     rc;
2650         int                     stripe_index = 0;
2651         struct lmv_stripe_md    *lsm = op_data->op_mea1;
2652         ENTRY;
2653
2654         rc = lmv_check_connect(obd);
2655         if (rc)
2656                 RETURN(rc);
2657 retry_unlink:
2658         /* For striped dir, we need to locate the parent as well */
2659         if (lsm != NULL) {
2660                 struct lmv_tgt_desc *tmp;
2661
2662                 LASSERT(op_data->op_name != NULL &&
2663                         op_data->op_namelen != 0);
2664
2665                 tmp = lmv_locate_target_for_name(lmv, lsm,
2666                                                  op_data->op_name,
2667                                                  op_data->op_namelen,
2668                                                  &op_data->op_fid1,
2669                                                  &op_data->op_mds);
2670
2671                 /* return -EBADFD means unknown hash type, might
2672                  * need try all sub-stripe here */
2673                 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2674                         RETURN(PTR_ERR(tmp));
2675
2676                 /* Note: both migrating dir and unknown hash dir need to
2677                  * try all of sub-stripes, so we need start search the
2678                  * name from stripe 0, but migrating dir is already handled
2679                  * inside lmv_locate_target_for_name(), so we only check
2680                  * unknown hash type directory here */
2681                 if (!lmv_is_known_hash_type(lsm)) {
2682                         struct lmv_oinfo *oinfo;
2683
2684                         oinfo = &lsm->lsm_md_oinfo[stripe_index];
2685
2686                         op_data->op_fid1 = oinfo->lmo_fid;
2687                         op_data->op_mds = oinfo->lmo_mds;
2688                 }
2689         }
2690
2691 try_next_stripe:
2692         /* Send unlink requests to the MDT where the child is located */
2693         if (likely(!fid_is_zero(&op_data->op_fid2)))
2694                 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2695         else if (lsm != NULL)
2696                 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2697         else
2698                 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2699
2700         if (IS_ERR(tgt))
2701                 RETURN(PTR_ERR(tgt));
2702
2703         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2704         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2705         op_data->op_cap = cfs_curproc_cap_pack();
2706
2707         /*
2708          * If child's fid is given, cancel unused locks for it if it is from
2709          * another export than parent.
2710          *
2711          * LOOKUP lock for child (fid3) should also be cancelled on parent
2712          * tgt_tgt in mdc_unlink().
2713          */
2714         op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2715
2716         /*
2717          * Cancel FULL locks on child (fid3).
2718          */
2719         parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2720         if (IS_ERR(parent_tgt))
2721                 RETURN(PTR_ERR(parent_tgt));
2722
2723         if (parent_tgt != tgt) {
2724                 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2725                                       LCK_EX, MDS_INODELOCK_LOOKUP,
2726                                       MF_MDC_CANCEL_FID3);
2727         }
2728
2729         rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2730                               MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2731         if (rc != 0)
2732                 RETURN(rc);
2733
2734         CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%d\n",
2735                PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2736
2737         rc = md_unlink(tgt->ltd_exp, op_data, request);
2738         if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2739                 RETURN(rc);
2740
2741         /* Try next stripe if it is needed. */
2742         if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2743                 struct lmv_oinfo *oinfo;
2744
2745                 stripe_index++;
2746                 if (stripe_index >= lsm->lsm_md_stripe_count)
2747                         RETURN(rc);
2748
2749                 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2750
2751                 op_data->op_fid1 = oinfo->lmo_fid;
2752                 op_data->op_mds = oinfo->lmo_mds;
2753
2754                 ptlrpc_req_finished(*request);
2755                 *request = NULL;
2756
2757                 goto try_next_stripe;
2758         }
2759
2760         body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2761         if (body == NULL)
2762                 RETURN(-EPROTO);
2763
2764         /* Not cross-ref case, just get out of here. */
2765         if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2766                 RETURN(rc);
2767
2768         CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2769                exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2770
2771         /* This is a remote object, try remote MDT, Note: it may
2772          * try more than 1 time here, Considering following case
2773          * /mnt/lustre is root on MDT0, remote1 is on MDT1
2774          * 1. Initially A does not know where remote1 is, it send
2775          *    unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2776          *    resend unlink RPC to MDT1 (retry 1st time).
2777          *
2778          * 2. During the unlink RPC in flight,
2779          *    client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2780          *    and create new remote1, but on MDT0
2781          *
2782          * 3. MDT1 get unlink RPC(from A), then do remote lock on
2783          *    /mnt/lustre, then lookup get fid of remote1, and find
2784          *    it is remote dir again, and replay -EREMOTE again.
2785          *
2786          * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2787          *
2788          * In theory, it might try unlimited time here, but it should
2789          * be very rare case.  */
2790         op_data->op_fid2 = body->mbo_fid1;
2791         ptlrpc_req_finished(*request);
2792         *request = NULL;
2793
2794         goto retry_unlink;
2795 }
2796
2797 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2798 {
2799         struct lmv_obd *lmv = &obd->u.lmv;
2800         int rc = 0;
2801
2802         switch (stage) {
2803         case OBD_CLEANUP_EARLY:
2804                 /* XXX: here should be calling obd_precleanup() down to
2805                  * stack. */
2806                 break;
2807         case OBD_CLEANUP_EXPORTS:
2808                 fld_client_proc_fini(&lmv->lmv_fld);
2809                 lprocfs_obd_cleanup(obd);
2810                 lprocfs_free_md_stats(obd);
2811                 break;
2812         default:
2813                 break;
2814         }
2815         RETURN(rc);
2816 }
2817
2818 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2819                         __u32 keylen, void *key, __u32 *vallen, void *val,
2820                         struct lov_stripe_md *lsm)
2821 {
2822         struct obd_device       *obd;
2823         struct lmv_obd          *lmv;
2824         int                      rc = 0;
2825         ENTRY;
2826
2827         obd = class_exp2obd(exp);
2828         if (obd == NULL) {
2829                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2830                        exp->exp_handle.h_cookie);
2831                 RETURN(-EINVAL);
2832         }
2833
2834         lmv = &obd->u.lmv;
2835         if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2836                 int i;
2837
2838                 rc = lmv_check_connect(obd);
2839                 if (rc)
2840                         RETURN(rc);
2841
2842                 LASSERT(*vallen == sizeof(__u32));
2843                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2844                         struct lmv_tgt_desc *tgt = lmv->tgts[i];
2845                         /*
2846                          * All tgts should be connected when this gets called.
2847                          */
2848                         if (tgt == NULL || tgt->ltd_exp == NULL)
2849                                 continue;
2850
2851                         if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2852                                           vallen, val, NULL))
2853                                 RETURN(0);
2854                 }
2855                 RETURN(-EINVAL);
2856         } else if (KEY_IS(KEY_MAX_EASIZE) ||
2857                    KEY_IS(KEY_DEFAULT_EASIZE) ||
2858                    KEY_IS(KEY_MAX_COOKIESIZE) ||
2859                    KEY_IS(KEY_DEFAULT_COOKIESIZE) ||
2860                    KEY_IS(KEY_CONN_DATA)) {
2861                 rc = lmv_check_connect(obd);
2862                 if (rc)
2863                         RETURN(rc);
2864
2865                 /*
2866                  * Forwarding this request to first MDS, it should know LOV
2867                  * desc.
2868                  */
2869                 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2870                                   vallen, val, NULL);
2871                 if (!rc && KEY_IS(KEY_CONN_DATA))
2872                         exp->exp_connect_data = *(struct obd_connect_data *)val;
2873                 RETURN(rc);
2874         } else if (KEY_IS(KEY_TGT_COUNT)) {
2875                 *((int *)val) = lmv->desc.ld_tgt_count;
2876                 RETURN(0);
2877         }
2878
2879         CDEBUG(D_IOCTL, "Invalid key\n");
2880         RETURN(-EINVAL);
2881 }
2882
2883 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2884                        obd_count keylen, void *key, obd_count vallen,
2885                        void *val, struct ptlrpc_request_set *set)
2886 {
2887         struct lmv_tgt_desc    *tgt = NULL;
2888         struct obd_device      *obd;
2889         struct lmv_obd         *lmv;
2890         int rc = 0;
2891         ENTRY;
2892
2893         obd = class_exp2obd(exp);
2894         if (obd == NULL) {
2895                 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2896                        exp->exp_handle.h_cookie);
2897                 RETURN(-EINVAL);
2898         }
2899         lmv = &obd->u.lmv;
2900
2901         if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX)) {
2902                 int i, err = 0;
2903
2904                 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2905                         tgt = lmv->tgts[i];
2906
2907                         if (tgt == NULL || tgt->ltd_exp == NULL)
2908                                 continue;
2909
2910                         err = obd_set_info_async(env, tgt->ltd_exp,
2911                                                  keylen, key, vallen, val, set);
2912                         if (err && rc == 0)
2913                                 rc = err;
2914                 }
2915
2916                 RETURN(rc);
2917         }
2918
2919         RETURN(-EINVAL);
2920 }
2921
2922 static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm,
2923                           struct lmv_mds_md_v1 *lmm1)
2924 {
2925         int     cplen;
2926         int     i;
2927
2928         lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic);
2929         lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count);
2930         lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index);
2931         lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type);
2932         cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name,
2933                         sizeof(lmm1->lmv_pool_name));
2934         if (cplen >= sizeof(lmm1->lmv_pool_name))
2935                 return -E2BIG;
2936
2937         for (i = 0; i < lsm->lsm_md_stripe_count; i++)
2938                 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i],
2939                               &lsm->lsm_md_oinfo[i].lmo_fid);
2940         return 0;
2941 }
2942
2943 int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm,
2944                 int stripe_count)
2945 {
2946         int     lmm_size = 0;
2947         bool    allocated = false;
2948         int     rc = 0;
2949         ENTRY;
2950
2951         LASSERT(lmmp != NULL);
2952         /* Free lmm */
2953         if (*lmmp != NULL && lsm == NULL) {
2954                 int stripe_count;
2955
2956                 stripe_count = lmv_mds_md_stripe_count_get(*lmmp);
2957                 lmm_size = lmv_mds_md_size(stripe_count,
2958                                            le32_to_cpu((*lmmp)->lmv_magic));
2959                 if (lmm_size == 0)
2960                         RETURN(-EINVAL);
2961                 OBD_FREE(*lmmp, lmm_size);
2962                 *lmmp = NULL;
2963                 RETURN(0);
2964         }
2965
2966         /* Alloc lmm */
2967         if (*lmmp == NULL && lsm == NULL) {
2968                 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
2969                 LASSERT(lmm_size > 0);
2970                 OBD_ALLOC(*lmmp, lmm_size);
2971                 if (*lmmp == NULL)
2972                         RETURN(-ENOMEM);
2973                 lmv_mds_md_stripe_count_set(*lmmp, stripe_count);
2974                 (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC);
2975                 RETURN(lmm_size);
2976         }
2977
2978         /* pack lmm */
2979         LASSERT(lsm != NULL);
2980         lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic);
2981         if (*lmmp == NULL) {
2982                 OBD_ALLOC(*lmmp, lmm_size);
2983                 if (*lmmp == NULL)
2984                         RETURN(-ENOMEM);
2985                 allocated = true;
2986         }
2987
2988         switch (lsm->lsm_md_magic) {
2989         case LMV_MAGIC_V1:
2990                 rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1);
2991                 break;
2992         default:
2993                 rc = -EINVAL;
2994                 break;
2995         }
2996
2997         if (rc != 0 && allocated) {
2998                 OBD_FREE(*lmmp, lmm_size);
2999                 *lmmp = NULL;
3000         }
3001
3002         RETURN(lmm_size);
3003 }
3004 EXPORT_SYMBOL(lmv_pack_md);
3005
3006 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
3007                             const struct lmv_mds_md_v1 *lmm1)
3008 {
3009         struct lmv_obd  *lmv = &exp->exp_obd->u.lmv;
3010         int             stripe_count;
3011         int             cplen;
3012         int             i;
3013         int             rc = 0;
3014         ENTRY;
3015
3016         lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3017         lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3018         lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3019         if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
3020                 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
3021         else
3022                 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3023         lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3024         cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3025                         sizeof(lsm->lsm_md_pool_name));
3026
3027         if (cplen >= sizeof(lsm->lsm_md_pool_name))
3028                 RETURN(-E2BIG);
3029
3030         CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
3031                "layout_version %d\n", lsm->lsm_md_stripe_count,
3032                lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
3033                lsm->lsm_md_layout_version);
3034
3035         stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3036         for (i = 0; i < stripe_count; i++) {
3037                 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3038                               &lmm1->lmv_stripe_fids[i]);
3039                 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3040                                     &lsm->lsm_md_oinfo[i].lmo_mds);
3041                 if (rc != 0)
3042                         RETURN(rc);
3043                 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3044                        PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3045         }
3046
3047         RETURN(rc);
3048 }
3049
3050 int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3051                   const union lmv_mds_md *lmm, int stripe_count)
3052 {
3053         struct lmv_stripe_md     *lsm;
3054         int                      lsm_size;
3055         int                      rc;
3056         bool                     allocated = false;
3057         ENTRY;
3058
3059         LASSERT(lsmp != NULL);
3060
3061         lsm = *lsmp;
3062         /* Free memmd */
3063         if (lsm != NULL && lmm == NULL) {
3064                 int i;
3065                 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3066                         /* For migrating inode, the master stripe and master
3067                          * object will be the same, so do not need iput, see
3068                          * ll_update_lsm_md */
3069                         if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
3070                               i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
3071                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
3072                 }
3073                 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3074                 OBD_FREE(lsm, lsm_size);
3075                 *lsmp = NULL;
3076                 RETURN(0);
3077         }
3078
3079         /* Alloc memmd */
3080         if (lsm == NULL && lmm == NULL) {
3081                 lsm_size = lmv_stripe_md_size(stripe_count);
3082                 OBD_ALLOC(lsm, lsm_size);
3083                 if (lsm == NULL)
3084                         RETURN(-ENOMEM);
3085                 lsm->lsm_md_stripe_count = stripe_count;
3086                 *lsmp = lsm;
3087                 RETURN(0);
3088         }
3089
3090         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3091                 RETURN(-EPERM);
3092
3093         /* Unpack memmd */
3094         if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3095             le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3096                 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3097                        exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3098                        -EIO);
3099                 RETURN(-EIO);
3100         }
3101
3102         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3103                 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3104         else
3105                 /**
3106                  * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3107                  * stripecount should be 0 then.
3108                  */
3109                 lsm_size = lmv_stripe_md_size(0);
3110
3111         lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3112         if (lsm == NULL) {
3113                 OBD_ALLOC(lsm, lsm_size);
3114                 if (lsm == NULL)
3115                         RETURN(-ENOMEM);
3116                 allocated = true;
3117                 *lsmp = lsm;
3118         }
3119
3120         switch (le32_to_cpu(lmm->lmv_magic)) {
3121         case LMV_MAGIC_V1:
3122                 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3123                 break;
3124         default:
3125                 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3126                        le32_to_cpu(lmm->lmv_magic));
3127                 rc = -EINVAL;
3128                 break;
3129         }
3130
3131         if (rc != 0 && allocated) {
3132                 OBD_FREE(lsm, lsm_size);
3133                 *lsmp = NULL;
3134                 lsm_size = rc;
3135         }
3136         RETURN(lsm_size);
3137 }
3138
3139 int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes)
3140 {
3141         return lmv_unpack_md(NULL, lsmp, NULL, stripes);
3142 }
3143 EXPORT_SYMBOL(lmv_alloc_memmd);
3144
3145 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3146 {
3147         lmv_unpack_md(NULL, &lsm, NULL, 0);
3148 }
3149 EXPORT_SYMBOL(lmv_free_memmd);
3150
3151 int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
3152                  struct lov_mds_md *lmm, int disk_len)
3153 {
3154         return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp,
3155                              (union lmv_mds_md *)lmm, disk_len);
3156 }
3157
3158 int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
3159                struct lov_stripe_md *lsm)
3160 {
3161         struct obd_device               *obd = exp->exp_obd;
3162         struct lmv_obd                  *lmv_obd = &obd->u.lmv;
3163         const struct lmv_stripe_md      *lmv = (struct lmv_stripe_md *)lsm;
3164         int                             stripe_count;
3165
3166         if (lmmp == NULL) {
3167                 if (lsm != NULL)
3168                         stripe_count = lmv->lsm_md_stripe_count;
3169                 else
3170                         stripe_count = lmv_obd->desc.ld_tgt_count;
3171
3172                 return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1);
3173         }
3174
3175         return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0);
3176 }
3177
3178 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3179                              ldlm_policy_data_t *policy, ldlm_mode_t mode,
3180                              ldlm_cancel_flags_t flags, void *opaque)
3181 {
3182         struct obd_device       *obd = exp->exp_obd;
3183         struct lmv_obd          *lmv = &obd->u.lmv;
3184         int                      rc = 0;
3185         int                      err;
3186         __u32                    i;
3187         ENTRY;
3188
3189         LASSERT(fid != NULL);
3190
3191         for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3192                 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3193
3194                 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3195                         continue;
3196
3197                 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3198                                        opaque);
3199                 if (!rc)
3200                         rc = err;
3201         }
3202         RETURN(rc);
3203 }
3204
3205 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3206                       __u64 *bits)
3207 {
3208         struct lmv_obd          *lmv = &exp->exp_obd->u.lmv;
3209         struct lmv_tgt_desc     *tgt = lmv->tgts[0];
3210         int                      rc;
3211         ENTRY;
3212
3213         if (tgt == NULL || tgt->ltd_exp == NULL)
3214                 RETURN(-EINVAL);
3215         rc =  md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3216         RETURN(rc);