Whamcloud - gitweb
LU-8475 target: use slab allocation
[fs/lustre-release.git] / lustre / lod / lod_object.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,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License version 2 for more details.  A copy is
14  * included in the COPYING file that accompanied this code.
15  *
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright  2009 Sun Microsystems, Inc. All rights reserved
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2012, 2016, Intel Corporation.
27  */
28 /*
29  * lustre/lod/lod_object.c
30  *
31  * This file contains implementations of methods for the OSD API
32  * for the Logical Object Device (LOD) layer, which provides a virtual
33  * local OSD object interface to the MDD layer, and abstracts the
34  * addressing of local (OSD) and remote (OSP) objects. The API is
35  * described in the file lustre/include/dt_object.h and in
36  * Documentation/osd-api.txt.
37  *
38  * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
39  */
40
41 #define DEBUG_SUBSYSTEM S_MDS
42
43 #include <obd.h>
44 #include <obd_class.h>
45 #include <obd_support.h>
46
47 #include <lustre_fid.h>
48 #include <lustre_linkea.h>
49 #include <lustre_lmv.h>
50 #include <uapi/linux/lustre/lustre_param.h>
51 #include <lustre_swab.h>
52 #include <uapi/linux/lustre/lustre_ver.h>
53 #include <lprocfs_status.h>
54 #include <md_object.h>
55
56 #include "lod_internal.h"
57
58 static const char dot[] = ".";
59 static const char dotdot[] = "..";
60
61 /**
62  * Implementation of dt_index_operations::dio_lookup
63  *
64  * Used with regular (non-striped) objects.
65  *
66  * \see dt_index_operations::dio_lookup() in the API description for details.
67  */
68 static int lod_lookup(const struct lu_env *env, struct dt_object *dt,
69                       struct dt_rec *rec, const struct dt_key *key)
70 {
71         struct dt_object *next = dt_object_child(dt);
72         return next->do_index_ops->dio_lookup(env, next, rec, key);
73 }
74
75 /**
76  * Implementation of dt_index_operations::dio_declare_insert.
77  *
78  * Used with regular (non-striped) objects.
79  *
80  * \see dt_index_operations::dio_declare_insert() in the API description
81  * for details.
82  */
83 static int lod_declare_insert(const struct lu_env *env, struct dt_object *dt,
84                               const struct dt_rec *rec,
85                               const struct dt_key *key, struct thandle *th)
86 {
87         return lod_sub_declare_insert(env, dt_object_child(dt), rec, key, th);
88 }
89
90 /**
91  * Implementation of dt_index_operations::dio_insert.
92  *
93  * Used with regular (non-striped) objects
94  *
95  * \see dt_index_operations::dio_insert() in the API description for details.
96  */
97 static int lod_insert(const struct lu_env *env, struct dt_object *dt,
98                       const struct dt_rec *rec, const struct dt_key *key,
99                       struct thandle *th, int ign)
100 {
101         return lod_sub_insert(env, dt_object_child(dt), rec, key, th, ign);
102 }
103
104 /**
105  * Implementation of dt_index_operations::dio_declare_delete.
106  *
107  * Used with regular (non-striped) objects.
108  *
109  * \see dt_index_operations::dio_declare_delete() in the API description
110  * for details.
111  */
112 static int lod_declare_delete(const struct lu_env *env, struct dt_object *dt,
113                               const struct dt_key *key, struct thandle *th)
114 {
115         return lod_sub_declare_delete(env, dt_object_child(dt), key, th);
116 }
117
118 /**
119  * Implementation of dt_index_operations::dio_delete.
120  *
121  * Used with regular (non-striped) objects.
122  *
123  * \see dt_index_operations::dio_delete() in the API description for details.
124  */
125 static int lod_delete(const struct lu_env *env, struct dt_object *dt,
126                       const struct dt_key *key, struct thandle *th)
127 {
128         return lod_sub_delete(env, dt_object_child(dt), key, th);
129 }
130
131 /**
132  * Implementation of dt_it_ops::init.
133  *
134  * Used with regular (non-striped) objects.
135  *
136  * \see dt_it_ops::init() in the API description for details.
137  */
138 static struct dt_it *lod_it_init(const struct lu_env *env,
139                                  struct dt_object *dt, __u32 attr)
140 {
141         struct dt_object        *next = dt_object_child(dt);
142         struct lod_it           *it = &lod_env_info(env)->lti_it;
143         struct dt_it            *it_next;
144
145         it_next = next->do_index_ops->dio_it.init(env, next, attr);
146         if (IS_ERR(it_next))
147                 return it_next;
148
149         /* currently we do not use more than one iterator per thread
150          * so we store it in thread info. if at some point we need
151          * more active iterators in a single thread, we can allocate
152          * additional ones */
153         LASSERT(it->lit_obj == NULL);
154
155         it->lit_it = it_next;
156         it->lit_obj = next;
157
158         return (struct dt_it *)it;
159 }
160
161 #define LOD_CHECK_IT(env, it)                                   \
162 do {                                                            \
163         LASSERT((it)->lit_obj != NULL);                         \
164         LASSERT((it)->lit_it != NULL);                          \
165 } while (0)
166
167 /**
168  * Implementation of dt_index_operations::dio_it.fini.
169  *
170  * Used with regular (non-striped) objects.
171  *
172  * \see dt_index_operations::dio_it.fini() in the API description for details.
173  */
174 static void lod_it_fini(const struct lu_env *env, struct dt_it *di)
175 {
176         struct lod_it *it = (struct lod_it *)di;
177
178         LOD_CHECK_IT(env, it);
179         it->lit_obj->do_index_ops->dio_it.fini(env, it->lit_it);
180
181         /* the iterator not in use any more */
182         it->lit_obj = NULL;
183         it->lit_it = NULL;
184 }
185
186 /**
187  * Implementation of dt_it_ops::get.
188  *
189  * Used with regular (non-striped) objects.
190  *
191  * \see dt_it_ops::get() in the API description for details.
192  */
193 static int lod_it_get(const struct lu_env *env, struct dt_it *di,
194                       const struct dt_key *key)
195 {
196         const struct lod_it *it = (const struct lod_it *)di;
197
198         LOD_CHECK_IT(env, it);
199         return it->lit_obj->do_index_ops->dio_it.get(env, it->lit_it, key);
200 }
201
202 /**
203  * Implementation of dt_it_ops::put.
204  *
205  * Used with regular (non-striped) objects.
206  *
207  * \see dt_it_ops::put() in the API description for details.
208  */
209 static void lod_it_put(const struct lu_env *env, struct dt_it *di)
210 {
211         struct lod_it *it = (struct lod_it *)di;
212
213         LOD_CHECK_IT(env, it);
214         return it->lit_obj->do_index_ops->dio_it.put(env, it->lit_it);
215 }
216
217 /**
218  * Implementation of dt_it_ops::next.
219  *
220  * Used with regular (non-striped) objects
221  *
222  * \see dt_it_ops::next() in the API description for details.
223  */
224 static int lod_it_next(const struct lu_env *env, struct dt_it *di)
225 {
226         struct lod_it *it = (struct lod_it *)di;
227
228         LOD_CHECK_IT(env, it);
229         return it->lit_obj->do_index_ops->dio_it.next(env, it->lit_it);
230 }
231
232 /**
233  * Implementation of dt_it_ops::key.
234  *
235  * Used with regular (non-striped) objects.
236  *
237  * \see dt_it_ops::key() in the API description for details.
238  */
239 static struct dt_key *lod_it_key(const struct lu_env *env,
240                                  const struct dt_it *di)
241 {
242         const struct lod_it *it = (const struct lod_it *)di;
243
244         LOD_CHECK_IT(env, it);
245         return it->lit_obj->do_index_ops->dio_it.key(env, it->lit_it);
246 }
247
248 /**
249  * Implementation of dt_it_ops::key_size.
250  *
251  * Used with regular (non-striped) objects.
252  *
253  * \see dt_it_ops::key_size() in the API description for details.
254  */
255 static int lod_it_key_size(const struct lu_env *env, const struct dt_it *di)
256 {
257         struct lod_it *it = (struct lod_it *)di;
258
259         LOD_CHECK_IT(env, it);
260         return it->lit_obj->do_index_ops->dio_it.key_size(env, it->lit_it);
261 }
262
263 /**
264  * Implementation of dt_it_ops::rec.
265  *
266  * Used with regular (non-striped) objects.
267  *
268  * \see dt_it_ops::rec() in the API description for details.
269  */
270 static int lod_it_rec(const struct lu_env *env, const struct dt_it *di,
271                       struct dt_rec *rec, __u32 attr)
272 {
273         const struct lod_it *it = (const struct lod_it *)di;
274
275         LOD_CHECK_IT(env, it);
276         return it->lit_obj->do_index_ops->dio_it.rec(env, it->lit_it, rec,
277                                                      attr);
278 }
279
280 /**
281  * Implementation of dt_it_ops::rec_size.
282  *
283  * Used with regular (non-striped) objects.
284  *
285  * \see dt_it_ops::rec_size() in the API description for details.
286  */
287 static int lod_it_rec_size(const struct lu_env *env, const struct dt_it *di,
288                            __u32 attr)
289 {
290         const struct lod_it *it = (const struct lod_it *)di;
291
292         LOD_CHECK_IT(env, it);
293         return it->lit_obj->do_index_ops->dio_it.rec_size(env, it->lit_it,
294                                                           attr);
295 }
296
297 /**
298  * Implementation of dt_it_ops::store.
299  *
300  * Used with regular (non-striped) objects.
301  *
302  * \see dt_it_ops::store() in the API description for details.
303  */
304 static __u64 lod_it_store(const struct lu_env *env, const struct dt_it *di)
305 {
306         const struct lod_it *it = (const struct lod_it *)di;
307
308         LOD_CHECK_IT(env, it);
309         return it->lit_obj->do_index_ops->dio_it.store(env, it->lit_it);
310 }
311
312 /**
313  * Implementation of dt_it_ops::load.
314  *
315  * Used with regular (non-striped) objects.
316  *
317  * \see dt_it_ops::load() in the API description for details.
318  */
319 static int lod_it_load(const struct lu_env *env, const struct dt_it *di,
320                        __u64 hash)
321 {
322         const struct lod_it *it = (const struct lod_it *)di;
323
324         LOD_CHECK_IT(env, it);
325         return it->lit_obj->do_index_ops->dio_it.load(env, it->lit_it, hash);
326 }
327
328 /**
329  * Implementation of dt_it_ops::key_rec.
330  *
331  * Used with regular (non-striped) objects.
332  *
333  * \see dt_it_ops::rec() in the API description for details.
334  */
335 static int lod_it_key_rec(const struct lu_env *env, const struct dt_it *di,
336                           void *key_rec)
337 {
338         const struct lod_it *it = (const struct lod_it *)di;
339
340         LOD_CHECK_IT(env, it);
341         return it->lit_obj->do_index_ops->dio_it.key_rec(env, it->lit_it,
342                                                          key_rec);
343 }
344
345 static struct dt_index_operations lod_index_ops = {
346         .dio_lookup             = lod_lookup,
347         .dio_declare_insert     = lod_declare_insert,
348         .dio_insert             = lod_insert,
349         .dio_declare_delete     = lod_declare_delete,
350         .dio_delete             = lod_delete,
351         .dio_it = {
352                 .init           = lod_it_init,
353                 .fini           = lod_it_fini,
354                 .get            = lod_it_get,
355                 .put            = lod_it_put,
356                 .next           = lod_it_next,
357                 .key            = lod_it_key,
358                 .key_size       = lod_it_key_size,
359                 .rec            = lod_it_rec,
360                 .rec_size       = lod_it_rec_size,
361                 .store          = lod_it_store,
362                 .load           = lod_it_load,
363                 .key_rec        = lod_it_key_rec,
364         }
365 };
366
367 /**
368  * Implementation of dt_it_ops::init.
369  *
370  * Used with striped objects. Internally just initializes the iterator
371  * on the first stripe.
372  *
373  * \see dt_it_ops::init() in the API description for details.
374  */
375 static struct dt_it *lod_striped_it_init(const struct lu_env *env,
376                                          struct dt_object *dt, __u32 attr)
377 {
378         struct lod_object       *lo = lod_dt_obj(dt);
379         struct dt_object        *next;
380         struct lod_it           *it = &lod_env_info(env)->lti_it;
381         struct dt_it            *it_next;
382         ENTRY;
383
384         LASSERT(lo->ldo_dir_stripe_count > 0);
385         next = lo->ldo_stripe[0];
386         LASSERT(next != NULL);
387         LASSERT(next->do_index_ops != NULL);
388
389         it_next = next->do_index_ops->dio_it.init(env, next, attr);
390         if (IS_ERR(it_next))
391                 return it_next;
392
393         /* currently we do not use more than one iterator per thread
394          * so we store it in thread info. if at some point we need
395          * more active iterators in a single thread, we can allocate
396          * additional ones */
397         LASSERT(it->lit_obj == NULL);
398
399         it->lit_stripe_index = 0;
400         it->lit_attr = attr;
401         it->lit_it = it_next;
402         it->lit_obj = dt;
403
404         return (struct dt_it *)it;
405 }
406
407 #define LOD_CHECK_STRIPED_IT(env, it, lo)                               \
408 do {                                                                    \
409         LASSERT((it)->lit_obj != NULL);                                 \
410         LASSERT((it)->lit_it != NULL);                                  \
411         LASSERT((lo)->ldo_dir_stripe_count > 0);                        \
412         LASSERT((it)->lit_stripe_index < (lo)->ldo_dir_stripe_count);   \
413 } while (0)
414
415 /**
416  * Implementation of dt_it_ops::fini.
417  *
418  * Used with striped objects.
419  *
420  * \see dt_it_ops::fini() in the API description for details.
421  */
422 static void lod_striped_it_fini(const struct lu_env *env, struct dt_it *di)
423 {
424         struct lod_it           *it = (struct lod_it *)di;
425         struct lod_object       *lo = lod_dt_obj(it->lit_obj);
426         struct dt_object        *next;
427
428         /* If lit_it == NULL, then it means the sub_it has been finished,
429          * which only happens in failure cases, see lod_striped_it_next() */
430         if (it->lit_it != NULL) {
431                 LOD_CHECK_STRIPED_IT(env, it, lo);
432
433                 next = lo->ldo_stripe[it->lit_stripe_index];
434                 LASSERT(next != NULL);
435                 LASSERT(next->do_index_ops != NULL);
436
437                 next->do_index_ops->dio_it.fini(env, it->lit_it);
438         }
439
440         /* the iterator not in use any more */
441         it->lit_obj = NULL;
442         it->lit_it = NULL;
443         it->lit_stripe_index = 0;
444 }
445
446 /**
447  * Implementation of dt_it_ops::get.
448  *
449  * Right now it's not used widely, only to reset the iterator to the
450  * initial position. It should be possible to implement a full version
451  * which chooses a correct stripe to be able to position with any key.
452  *
453  * \see dt_it_ops::get() in the API description for details.
454  */
455 static int lod_striped_it_get(const struct lu_env *env, struct dt_it *di,
456                               const struct dt_key *key)
457 {
458         const struct lod_it     *it = (const struct lod_it *)di;
459         struct lod_object       *lo = lod_dt_obj(it->lit_obj);
460         struct dt_object        *next;
461         ENTRY;
462
463         LOD_CHECK_STRIPED_IT(env, it, lo);
464
465         next = lo->ldo_stripe[it->lit_stripe_index];
466         LASSERT(next != NULL);
467         LASSERT(next->do_index_ops != NULL);
468
469         return next->do_index_ops->dio_it.get(env, it->lit_it, key);
470 }
471
472 /**
473  * Implementation of dt_it_ops::put.
474  *
475  * Used with striped objects.
476  *
477  * \see dt_it_ops::put() in the API description for details.
478  */
479 static void lod_striped_it_put(const struct lu_env *env, struct dt_it *di)
480 {
481         struct lod_it           *it = (struct lod_it *)di;
482         struct lod_object       *lo = lod_dt_obj(it->lit_obj);
483         struct dt_object        *next;
484
485         LOD_CHECK_STRIPED_IT(env, it, lo);
486
487         next = lo->ldo_stripe[it->lit_stripe_index];
488         LASSERT(next != NULL);
489         LASSERT(next->do_index_ops != NULL);
490
491         return next->do_index_ops->dio_it.put(env, it->lit_it);
492 }
493
494 /**
495  * Implementation of dt_it_ops::next.
496  *
497  * Used with striped objects. When the end of the current stripe is
498  * reached, the method takes the next stripe's iterator.
499  *
500  * \see dt_it_ops::next() in the API description for details.
501  */
502 static int lod_striped_it_next(const struct lu_env *env, struct dt_it *di)
503 {
504         struct lod_it           *it = (struct lod_it *)di;
505         struct lod_object       *lo = lod_dt_obj(it->lit_obj);
506         struct dt_object        *next;
507         struct dt_it            *it_next;
508         int                     rc;
509         ENTRY;
510
511         LOD_CHECK_STRIPED_IT(env, it, lo);
512
513         next = lo->ldo_stripe[it->lit_stripe_index];
514         LASSERT(next != NULL);
515         LASSERT(next->do_index_ops != NULL);
516 again:
517         rc = next->do_index_ops->dio_it.next(env, it->lit_it);
518         if (rc < 0)
519                 RETURN(rc);
520
521         if (rc == 0 && it->lit_stripe_index == 0)
522                 RETURN(rc);
523
524         if (rc == 0 && it->lit_stripe_index > 0) {
525                 struct lu_dirent *ent;
526
527                 ent = (struct lu_dirent *)lod_env_info(env)->lti_key;
528
529                 rc = next->do_index_ops->dio_it.rec(env, it->lit_it,
530                                                     (struct dt_rec *)ent,
531                                                     it->lit_attr);
532                 if (rc != 0)
533                         RETURN(rc);
534
535                 /* skip . and .. for slave stripe */
536                 if ((strncmp(ent->lde_name, ".",
537                              le16_to_cpu(ent->lde_namelen)) == 0 &&
538                      le16_to_cpu(ent->lde_namelen) == 1) ||
539                     (strncmp(ent->lde_name, "..",
540                              le16_to_cpu(ent->lde_namelen)) == 0 &&
541                      le16_to_cpu(ent->lde_namelen) == 2))
542                         goto again;
543
544                 RETURN(rc);
545         }
546
547         /* go to next stripe */
548         if (it->lit_stripe_index + 1 >= lo->ldo_dir_stripe_count)
549                 RETURN(1);
550
551         it->lit_stripe_index++;
552
553         next->do_index_ops->dio_it.put(env, it->lit_it);
554         next->do_index_ops->dio_it.fini(env, it->lit_it);
555         it->lit_it = NULL;
556
557         next = lo->ldo_stripe[it->lit_stripe_index];
558         LASSERT(next != NULL);
559         rc = next->do_ops->do_index_try(env, next, &dt_directory_features);
560         if (rc != 0)
561                 RETURN(rc);
562
563         LASSERT(next->do_index_ops != NULL);
564
565         it_next = next->do_index_ops->dio_it.init(env, next, it->lit_attr);
566         if (!IS_ERR(it_next)) {
567                 it->lit_it = it_next;
568                 goto again;
569         } else {
570                 rc = PTR_ERR(it_next);
571         }
572
573         RETURN(rc);
574 }
575
576 /**
577  * Implementation of dt_it_ops::key.
578  *
579  * Used with striped objects.
580  *
581  * \see dt_it_ops::key() in the API description for details.
582  */
583 static struct dt_key *lod_striped_it_key(const struct lu_env *env,
584                                          const struct dt_it *di)
585 {
586         const struct lod_it     *it = (const struct lod_it *)di;
587         struct lod_object       *lo = lod_dt_obj(it->lit_obj);
588         struct dt_object        *next;
589
590         LOD_CHECK_STRIPED_IT(env, it, lo);
591
592         next = lo->ldo_stripe[it->lit_stripe_index];
593         LASSERT(next != NULL);
594         LASSERT(next->do_index_ops != NULL);
595
596         return next->do_index_ops->dio_it.key(env, it->lit_it);
597 }
598
599 /**
600  * Implementation of dt_it_ops::key_size.
601  *
602  * Used with striped objects.
603  *
604  * \see dt_it_ops::size() in the API description for details.
605  */
606 static int lod_striped_it_key_size(const struct lu_env *env,
607                                    const struct dt_it *di)
608 {
609         struct lod_it           *it = (struct lod_it *)di;
610         struct lod_object       *lo = lod_dt_obj(it->lit_obj);
611         struct dt_object        *next;
612
613         LOD_CHECK_STRIPED_IT(env, it, lo);
614
615         next = lo->ldo_stripe[it->lit_stripe_index];
616         LASSERT(next != NULL);
617         LASSERT(next->do_index_ops != NULL);
618
619         return next->do_index_ops->dio_it.key_size(env, it->lit_it);
620 }
621
622 /**
623  * Implementation of dt_it_ops::rec.
624  *
625  * Used with striped objects.
626  *
627  * \see dt_it_ops::rec() in the API description for details.
628  */
629 static int lod_striped_it_rec(const struct lu_env *env, const struct dt_it *di,
630                               struct dt_rec *rec, __u32 attr)
631 {
632         const struct lod_it     *it = (const struct lod_it *)di;
633         struct lod_object       *lo = lod_dt_obj(it->lit_obj);
634         struct dt_object        *next;
635
636         LOD_CHECK_STRIPED_IT(env, it, lo);
637
638         next = lo->ldo_stripe[it->lit_stripe_index];
639         LASSERT(next != NULL);
640         LASSERT(next->do_index_ops != NULL);
641
642         return next->do_index_ops->dio_it.rec(env, it->lit_it, rec, attr);
643 }
644
645 /**
646  * Implementation of dt_it_ops::rec_size.
647  *
648  * Used with striped objects.
649  *
650  * \see dt_it_ops::rec_size() in the API description for details.
651  */
652 static int lod_striped_it_rec_size(const struct lu_env *env,
653                                    const struct dt_it *di, __u32 attr)
654 {
655         struct lod_it           *it = (struct lod_it *)di;
656         struct lod_object       *lo = lod_dt_obj(it->lit_obj);
657         struct dt_object        *next;
658
659         LOD_CHECK_STRIPED_IT(env, it, lo);
660
661         next = lo->ldo_stripe[it->lit_stripe_index];
662         LASSERT(next != NULL);
663         LASSERT(next->do_index_ops != NULL);
664
665         return next->do_index_ops->dio_it.rec_size(env, it->lit_it, attr);
666 }
667
668 /**
669  * Implementation of dt_it_ops::store.
670  *
671  * Used with striped objects.
672  *
673  * \see dt_it_ops::store() in the API description for details.
674  */
675 static __u64 lod_striped_it_store(const struct lu_env *env,
676                                   const struct dt_it *di)
677 {
678         const struct lod_it     *it = (const struct lod_it *)di;
679         struct lod_object       *lo = lod_dt_obj(it->lit_obj);
680         struct dt_object        *next;
681
682         LOD_CHECK_STRIPED_IT(env, it, lo);
683
684         next = lo->ldo_stripe[it->lit_stripe_index];
685         LASSERT(next != NULL);
686         LASSERT(next->do_index_ops != NULL);
687
688         return next->do_index_ops->dio_it.store(env, it->lit_it);
689 }
690
691 /**
692  * Implementation of dt_it_ops::load.
693  *
694  * Used with striped objects.
695  *
696  * \see dt_it_ops::load() in the API description for details.
697  */
698 static int lod_striped_it_load(const struct lu_env *env,
699                                const struct dt_it *di, __u64 hash)
700 {
701         const struct lod_it     *it = (const struct lod_it *)di;
702         struct lod_object       *lo = lod_dt_obj(it->lit_obj);
703         struct dt_object        *next;
704
705         LOD_CHECK_STRIPED_IT(env, it, lo);
706
707         next = lo->ldo_stripe[it->lit_stripe_index];
708         LASSERT(next != NULL);
709         LASSERT(next->do_index_ops != NULL);
710
711         return next->do_index_ops->dio_it.load(env, it->lit_it, hash);
712 }
713
714 static struct dt_index_operations lod_striped_index_ops = {
715         .dio_lookup             = lod_lookup,
716         .dio_declare_insert     = lod_declare_insert,
717         .dio_insert             = lod_insert,
718         .dio_declare_delete     = lod_declare_delete,
719         .dio_delete             = lod_delete,
720         .dio_it = {
721                 .init           = lod_striped_it_init,
722                 .fini           = lod_striped_it_fini,
723                 .get            = lod_striped_it_get,
724                 .put            = lod_striped_it_put,
725                 .next           = lod_striped_it_next,
726                 .key            = lod_striped_it_key,
727                 .key_size       = lod_striped_it_key_size,
728                 .rec            = lod_striped_it_rec,
729                 .rec_size       = lod_striped_it_rec_size,
730                 .store          = lod_striped_it_store,
731                 .load           = lod_striped_it_load,
732         }
733 };
734
735 /**
736  * Append the FID for each shard of the striped directory after the
737  * given LMV EA header.
738  *
739  * To simplify striped directory and the consistency verification,
740  * we only store the LMV EA header on disk, for both master object
741  * and slave objects. When someone wants to know the whole LMV EA,
742  * such as client readdir(), we can build the entrie LMV EA on the
743  * MDT side (in RAM) via iterating the sub-directory entries that
744  * are contained in the master object of the stripe directory.
745  *
746  * For the master object of the striped directroy, the valid name
747  * for each shard is composed of the ${shard_FID}:${shard_idx}.
748  *
749  * There may be holes in the LMV EA if some shards' name entries
750  * are corrupted or lost.
751  *
752  * \param[in] env       pointer to the thread context
753  * \param[in] lo        pointer to the master object of the striped directory
754  * \param[in] buf       pointer to the lu_buf which will hold the LMV EA
755  * \param[in] resize    whether re-allocate the buffer if it is not big enough
756  *
757  * \retval              positive size of the LMV EA
758  * \retval              0 for nothing to be loaded
759  * \retval              negative error number on failure
760  */
761 int lod_load_lmv_shards(const struct lu_env *env, struct lod_object *lo,
762                         struct lu_buf *buf, bool resize)
763 {
764         struct lu_dirent        *ent    =
765                         (struct lu_dirent *)lod_env_info(env)->lti_key;
766         struct lod_device       *lod    = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
767         struct dt_object        *obj    = dt_object_child(&lo->ldo_obj);
768         struct lmv_mds_md_v1    *lmv1   = buf->lb_buf;
769         struct dt_it            *it;
770         const struct dt_it_ops  *iops;
771         __u32                    stripes;
772         __u32                    magic  = le32_to_cpu(lmv1->lmv_magic);
773         size_t                   lmv1_size;
774         int                      rc;
775         ENTRY;
776
777         /* If it is not a striped directory, then load nothing. */
778         if (magic != LMV_MAGIC_V1)
779                 RETURN(0);
780
781         /* If it is in migration (or failure), then load nothing. */
782         if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
783                 RETURN(0);
784
785         stripes = le32_to_cpu(lmv1->lmv_stripe_count);
786         if (stripes < 1)
787                 RETURN(0);
788
789         rc = lmv_mds_md_size(stripes, magic);
790         if (rc < 0)
791                 RETURN(rc);
792         lmv1_size = rc;
793         if (buf->lb_len < lmv1_size) {
794                 struct lu_buf tbuf;
795
796                 if (!resize)
797                         RETURN(-ERANGE);
798
799                 tbuf = *buf;
800                 buf->lb_buf = NULL;
801                 buf->lb_len = 0;
802                 lu_buf_alloc(buf, lmv1_size);
803                 lmv1 = buf->lb_buf;
804                 if (lmv1 == NULL)
805                         RETURN(-ENOMEM);
806
807                 memcpy(buf->lb_buf, tbuf.lb_buf, tbuf.lb_len);
808         }
809
810         if (unlikely(!dt_try_as_dir(env, obj)))
811                 RETURN(-ENOTDIR);
812
813         memset(&lmv1->lmv_stripe_fids[0], 0, stripes * sizeof(struct lu_fid));
814         iops = &obj->do_index_ops->dio_it;
815         it = iops->init(env, obj, LUDA_64BITHASH);
816         if (IS_ERR(it))
817                 RETURN(PTR_ERR(it));
818
819         rc = iops->load(env, it, 0);
820         if (rc == 0)
821                 rc = iops->next(env, it);
822         else if (rc > 0)
823                 rc = 0;
824
825         while (rc == 0) {
826                 char             name[FID_LEN + 2] = "";
827                 struct lu_fid    fid;
828                 __u32            index;
829                 int              len;
830
831                 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
832                 if (rc != 0)
833                         break;
834
835                 rc = -EIO;
836
837                 fid_le_to_cpu(&fid, &ent->lde_fid);
838                 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
839                 if (ent->lde_name[0] == '.') {
840                         if (ent->lde_namelen == 1)
841                                 goto next;
842
843                         if (ent->lde_namelen == 2 && ent->lde_name[1] == '.')
844                                 goto next;
845                 }
846
847                 len = snprintf(name, sizeof(name),
848                                DFID":", PFID(&ent->lde_fid));
849                 /* The ent->lde_name is composed of ${FID}:${index} */
850                 if (ent->lde_namelen < len + 1 ||
851                     memcmp(ent->lde_name, name, len) != 0) {
852                         CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
853                                "%s: invalid shard name %.*s with the FID "DFID
854                                " for the striped directory "DFID", %s\n",
855                                lod2obd(lod)->obd_name, ent->lde_namelen,
856                                ent->lde_name, PFID(&fid),
857                                PFID(lu_object_fid(&obj->do_lu)),
858                                lod->lod_lmv_failout ? "failout" : "skip");
859
860                         if (lod->lod_lmv_failout)
861                                 break;
862
863                         goto next;
864                 }
865
866                 index = 0;
867                 do {
868                         if (ent->lde_name[len] < '0' ||
869                             ent->lde_name[len] > '9') {
870                                 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
871                                        "%s: invalid shard name %.*s with the "
872                                        "FID "DFID" for the striped directory "
873                                        DFID", %s\n",
874                                        lod2obd(lod)->obd_name, ent->lde_namelen,
875                                        ent->lde_name, PFID(&fid),
876                                        PFID(lu_object_fid(&obj->do_lu)),
877                                        lod->lod_lmv_failout ?
878                                        "failout" : "skip");
879
880                                 if (lod->lod_lmv_failout)
881                                         break;
882
883                                 goto next;
884                         }
885
886                         index = index * 10 + ent->lde_name[len++] - '0';
887                 } while (len < ent->lde_namelen);
888
889                 if (len == ent->lde_namelen) {
890                         /* Out of LMV EA range. */
891                         if (index >= stripes) {
892                                 CERROR("%s: the shard %.*s for the striped "
893                                        "directory "DFID" is out of the known "
894                                        "LMV EA range [0 - %u], failout\n",
895                                        lod2obd(lod)->obd_name, ent->lde_namelen,
896                                        ent->lde_name,
897                                        PFID(lu_object_fid(&obj->do_lu)),
898                                        stripes - 1);
899
900                                 break;
901                         }
902
903                         /* The slot has been occupied. */
904                         if (!fid_is_zero(&lmv1->lmv_stripe_fids[index])) {
905                                 struct lu_fid fid0;
906
907                                 fid_le_to_cpu(&fid0,
908                                         &lmv1->lmv_stripe_fids[index]);
909                                 CERROR("%s: both the shard "DFID" and "DFID
910                                        " for the striped directory "DFID
911                                        " claim the same LMV EA slot at the "
912                                        "index %d, failout\n",
913                                        lod2obd(lod)->obd_name,
914                                        PFID(&fid0), PFID(&fid),
915                                        PFID(lu_object_fid(&obj->do_lu)), index);
916
917                                 break;
918                         }
919
920                         /* stored as LE mode */
921                         lmv1->lmv_stripe_fids[index] = ent->lde_fid;
922
923 next:
924                         rc = iops->next(env, it);
925                 }
926         }
927
928         iops->put(env, it);
929         iops->fini(env, it);
930
931         RETURN(rc > 0 ? lmv_mds_md_size(stripes, magic) : rc);
932 }
933
934 /**
935  * Implementation of dt_object_operations::do_index_try.
936  *
937  * \see dt_object_operations::do_index_try() in the API description for details.
938  */
939 static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
940                          const struct dt_index_features *feat)
941 {
942         struct lod_object       *lo = lod_dt_obj(dt);
943         struct dt_object        *next = dt_object_child(dt);
944         int                     rc;
945         ENTRY;
946
947         LASSERT(next->do_ops);
948         LASSERT(next->do_ops->do_index_try);
949
950         rc = lod_load_striping_locked(env, lo);
951         if (rc != 0)
952                 RETURN(rc);
953
954         rc = next->do_ops->do_index_try(env, next, feat);
955         if (rc != 0)
956                 RETURN(rc);
957
958         if (lo->ldo_dir_stripe_count > 0) {
959                 int i;
960
961                 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
962                         if (dt_object_exists(lo->ldo_stripe[i]) == 0)
963                                 continue;
964                         rc = lo->ldo_stripe[i]->do_ops->do_index_try(env,
965                                                 lo->ldo_stripe[i], feat);
966                         if (rc != 0)
967                                 RETURN(rc);
968                 }
969                 dt->do_index_ops = &lod_striped_index_ops;
970         } else {
971                 dt->do_index_ops = &lod_index_ops;
972         }
973
974         RETURN(rc);
975 }
976
977 /**
978  * Implementation of dt_object_operations::do_read_lock.
979  *
980  * \see dt_object_operations::do_read_lock() in the API description for details.
981  */
982 static void lod_read_lock(const struct lu_env *env, struct dt_object *dt,
983                           unsigned role)
984 {
985         dt_read_lock(env, dt_object_child(dt), role);
986 }
987
988 /**
989  * Implementation of dt_object_operations::do_write_lock.
990  *
991  * \see dt_object_operations::do_write_lock() in the API description for
992  * details.
993  */
994 static void lod_write_lock(const struct lu_env *env, struct dt_object *dt,
995                            unsigned role)
996 {
997         dt_write_lock(env, dt_object_child(dt), role);
998 }
999
1000 /**
1001  * Implementation of dt_object_operations::do_read_unlock.
1002  *
1003  * \see dt_object_operations::do_read_unlock() in the API description for
1004  * details.
1005  */
1006 static void lod_read_unlock(const struct lu_env *env, struct dt_object *dt)
1007 {
1008         dt_read_unlock(env, dt_object_child(dt));
1009 }
1010
1011 /**
1012  * Implementation of dt_object_operations::do_write_unlock.
1013  *
1014  * \see dt_object_operations::do_write_unlock() in the API description for
1015  * details.
1016  */
1017 static void lod_write_unlock(const struct lu_env *env, struct dt_object *dt)
1018 {
1019         dt_write_unlock(env, dt_object_child(dt));
1020 }
1021
1022 /**
1023  * Implementation of dt_object_operations::do_write_locked.
1024  *
1025  * \see dt_object_operations::do_write_locked() in the API description for
1026  * details.
1027  */
1028 static int lod_write_locked(const struct lu_env *env, struct dt_object *dt)
1029 {
1030         return dt_write_locked(env, dt_object_child(dt));
1031 }
1032
1033 /**
1034  * Implementation of dt_object_operations::do_attr_get.
1035  *
1036  * \see dt_object_operations::do_attr_get() in the API description for details.
1037  */
1038 static int lod_attr_get(const struct lu_env *env,
1039                         struct dt_object *dt,
1040                         struct lu_attr *attr)
1041 {
1042         /* Note: for striped directory, client will merge attributes
1043          * from all of the sub-stripes see lmv_merge_attr(), and there
1044          * no MDD logic depend on directory nlink/size/time, so we can
1045          * always use master inode nlink and size for now. */
1046         return dt_attr_get(env, dt_object_child(dt), attr);
1047 }
1048
1049 int lod_obj_for_each_stripe(const struct lu_env *env, struct lod_object *lo,
1050                             struct thandle *th, lod_obj_stripe_cb_t cb,
1051                             struct lod_obj_stripe_cb_data *data)
1052 {
1053         struct lod_layout_component *lod_comp;
1054         int i, j, rc;
1055         ENTRY;
1056
1057         LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1058         for (i = 0; i < lo->ldo_comp_cnt; i++) {
1059                 lod_comp = &lo->ldo_comp_entries[i];
1060
1061                 if (lod_comp->llc_stripe == NULL)
1062                         continue;
1063
1064                 LASSERT(lod_comp->llc_stripe_count > 0);
1065                 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
1066                         struct dt_object *dt = lod_comp->llc_stripe[j];
1067
1068                         if (dt == NULL)
1069                                 continue;
1070                         rc = cb(env, lo, dt, th, j, data);
1071                         if (rc != 0)
1072                                 RETURN(rc);
1073                 }
1074         }
1075         RETURN(0);
1076 }
1077
1078 static inline int
1079 lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
1080                            struct dt_object *dt, struct thandle *th,
1081                            int stripe_idx, struct lod_obj_stripe_cb_data *data)
1082 {
1083         if (data->locd_declare)
1084                 return lod_sub_declare_attr_set(env, dt, data->locd_attr, th);
1085
1086         return lod_sub_attr_set(env, dt, data->locd_attr, th);
1087 }
1088
1089 /**
1090  * Implementation of dt_object_operations::do_declare_attr_set.
1091  *
1092  * If the object is striped, then apply the changes to all the stripes.
1093  *
1094  * \see dt_object_operations::do_declare_attr_set() in the API description
1095  * for details.
1096  */
1097 static int lod_declare_attr_set(const struct lu_env *env,
1098                                 struct dt_object *dt,
1099                                 const struct lu_attr *attr,
1100                                 struct thandle *th)
1101 {
1102         struct dt_object  *next = dt_object_child(dt);
1103         struct lod_object *lo = lod_dt_obj(dt);
1104         int                rc, i;
1105         ENTRY;
1106
1107         /*
1108          * declare setattr on the local object
1109          */
1110         rc = lod_sub_declare_attr_set(env, next, attr, th);
1111         if (rc)
1112                 RETURN(rc);
1113
1114         /* osp_declare_attr_set() ignores all attributes other than
1115          * UID, GID, PROJID, and size, and osp_attr_set() ignores all
1116          * but UID, GID and PROJID. Declaration of size attr setting
1117          * happens through lod_declare_init_size(), and not through
1118          * this function. Therefore we need not load striping unless
1119          * ownership is changing.  This should save memory and (we hope)
1120          * speed up rename().
1121          */
1122         if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1123                 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID)))
1124                         RETURN(rc);
1125
1126                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1127                         RETURN(0);
1128         } else {
1129                 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID | LA_MODE |
1130                                         LA_ATIME | LA_MTIME | LA_CTIME |
1131                                         LA_FLAGS)))
1132                         RETURN(rc);
1133         }
1134         /*
1135          * load striping information, notice we don't do this when object
1136          * is being initialized as we don't need this information till
1137          * few specific cases like destroy, chown
1138          */
1139         rc = lod_load_striping(env, lo);
1140         if (rc)
1141                 RETURN(rc);
1142
1143         if (!lod_obj_is_striped(dt))
1144                 RETURN(0);
1145
1146         /*
1147          * if object is striped declare changes on the stripes
1148          */
1149         if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1150                 LASSERT(lo->ldo_stripe);
1151                 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1152                         if (lo->ldo_stripe[i] == NULL)
1153                                 continue;
1154                         rc = lod_sub_declare_attr_set(env, lo->ldo_stripe[i],
1155                                                       attr, th);
1156                         if (rc != 0)
1157                                 RETURN(rc);
1158                 }
1159         } else {
1160                 struct lod_obj_stripe_cb_data data;
1161
1162                 data.locd_attr = attr;
1163                 data.locd_declare = true;
1164                 rc = lod_obj_for_each_stripe(env, lo, th,
1165                                 lod_obj_stripe_attr_set_cb, &data);
1166         }
1167
1168         if (rc)
1169                 RETURN(rc);
1170
1171         if (!dt_object_exists(next) || dt_object_remote(next) ||
1172             !S_ISREG(attr->la_mode))
1173                 RETURN(0);
1174
1175         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1176                 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
1177                 RETURN(rc);
1178         }
1179
1180         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) ||
1181             OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1182                 struct lod_thread_info *info = lod_env_info(env);
1183                 struct lu_buf *buf = &info->lti_buf;
1184
1185                 buf->lb_buf = info->lti_ea_store;
1186                 buf->lb_len = info->lti_ea_store_size;
1187                 rc = lod_sub_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1188                                                LU_XATTR_REPLACE, th);
1189         }
1190
1191         RETURN(rc);
1192 }
1193
1194 /**
1195  * Implementation of dt_object_operations::do_attr_set.
1196  *
1197  * If the object is striped, then apply the changes to all or subset of
1198  * the stripes depending on the object type and specific attributes.
1199  *
1200  * \see dt_object_operations::do_attr_set() in the API description for details.
1201  */
1202 static int lod_attr_set(const struct lu_env *env,
1203                         struct dt_object *dt,
1204                         const struct lu_attr *attr,
1205                         struct thandle *th)
1206 {
1207         struct dt_object        *next = dt_object_child(dt);
1208         struct lod_object       *lo = lod_dt_obj(dt);
1209         int                     rc, i;
1210         ENTRY;
1211
1212         /*
1213          * apply changes to the local object
1214          */
1215         rc = lod_sub_attr_set(env, next, attr, th);
1216         if (rc)
1217                 RETURN(rc);
1218
1219         if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1220                 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID)))
1221                         RETURN(rc);
1222
1223                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1224                         RETURN(0);
1225         } else {
1226                 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE | LA_PROJID |
1227                                         LA_ATIME | LA_MTIME | LA_CTIME |
1228                                         LA_FLAGS)))
1229                         RETURN(rc);
1230         }
1231
1232         if (!lod_obj_is_striped(dt))
1233                 RETURN(0);
1234
1235         /*
1236          * if object is striped, apply changes to all the stripes
1237          */
1238         if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1239                 LASSERT(lo->ldo_stripe);
1240                 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1241                         if (unlikely(lo->ldo_stripe[i] == NULL))
1242                                 continue;
1243
1244                         if ((dt_object_exists(lo->ldo_stripe[i]) == 0))
1245                                 continue;
1246
1247                         rc = lod_sub_attr_set(env, lo->ldo_stripe[i], attr, th);
1248                         if (rc != 0)
1249                                 break;
1250                 }
1251         } else {
1252                 struct lod_obj_stripe_cb_data data;
1253
1254                 data.locd_attr = attr;
1255                 data.locd_declare = false;
1256                 rc = lod_obj_for_each_stripe(env, lo, th,
1257                                 lod_obj_stripe_attr_set_cb, &data);
1258         }
1259
1260         if (rc)
1261                 RETURN(rc);
1262
1263         if (!dt_object_exists(next) || dt_object_remote(next) ||
1264             !S_ISREG(attr->la_mode))
1265                 RETURN(0);
1266
1267         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1268                 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
1269                 RETURN(rc);
1270         }
1271
1272         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE)) {
1273                 struct lod_thread_info *info = lod_env_info(env);
1274                 struct lu_buf *buf = &info->lti_buf;
1275                 struct ost_id *oi = &info->lti_ostid;
1276                 struct lu_fid *fid = &info->lti_fid;
1277                 struct lov_mds_md_v1 *lmm;
1278                 struct lov_ost_data_v1 *objs;
1279                 __u32 magic;
1280
1281                 rc = lod_get_lov_ea(env, lo);
1282                 if (rc <= 0)
1283                         RETURN(rc);
1284
1285                 buf->lb_buf = info->lti_ea_store;
1286                 buf->lb_len = info->lti_ea_store_size;
1287                 lmm = info->lti_ea_store;
1288                 magic = le32_to_cpu(lmm->lmm_magic);
1289                 if (magic == LOV_MAGIC_COMP_V1) {
1290                         struct lov_comp_md_v1 *lcm = buf->lb_buf;
1291                         struct lov_comp_md_entry_v1 *lcme =
1292                                                 &lcm->lcm_entries[0];
1293
1294                         lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
1295                         magic = le32_to_cpu(lmm->lmm_magic);
1296                 }
1297
1298                 if (magic == LOV_MAGIC_V1)
1299                         objs = &(lmm->lmm_objects[0]);
1300                 else
1301                         objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1302                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1303                 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1304                 fid->f_oid--;
1305                 fid_to_ostid(fid, oi);
1306                 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1307
1308                 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1309                                        LU_XATTR_REPLACE, th);
1310         } else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1311                 struct lod_thread_info *info = lod_env_info(env);
1312                 struct lu_buf *buf = &info->lti_buf;
1313                 struct lov_comp_md_v1 *lcm;
1314                 struct lov_comp_md_entry_v1 *lcme;
1315
1316                 rc = lod_get_lov_ea(env, lo);
1317                 if (rc <= 0)
1318                         RETURN(rc);
1319
1320                 buf->lb_buf = info->lti_ea_store;
1321                 buf->lb_len = info->lti_ea_store_size;
1322                 lcm = buf->lb_buf;
1323                 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
1324                         RETURN(-EINVAL);
1325
1326                 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1327                 lcme = &lcm->lcm_entries[0];
1328                 le64_add_cpu(&lcme->lcme_extent.e_start, 1);
1329                 le64_add_cpu(&lcme->lcme_extent.e_end, -1);
1330
1331                 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1332                                        LU_XATTR_REPLACE, th);
1333         }
1334
1335         RETURN(rc);
1336 }
1337
1338 /**
1339  * Implementation of dt_object_operations::do_xattr_get.
1340  *
1341  * If LOV EA is requested from the root object and it's not
1342  * found, then return default striping for the filesystem.
1343  *
1344  * \see dt_object_operations::do_xattr_get() in the API description for details.
1345  */
1346 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1347                          struct lu_buf *buf, const char *name)
1348 {
1349         struct lod_thread_info *info = lod_env_info(env);
1350         struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1351         int is_root;
1352         int rc;
1353         ENTRY;
1354
1355         rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1356         if (strcmp(name, XATTR_NAME_LMV) == 0) {
1357                 struct lmv_mds_md_v1    *lmv1;
1358                 int                      rc1 = 0;
1359
1360                 if (rc > (typeof(rc))sizeof(*lmv1))
1361                         RETURN(rc);
1362
1363                 if (rc < (typeof(rc))sizeof(*lmv1))
1364                         RETURN(rc = rc > 0 ? -EINVAL : rc);
1365
1366                 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1367                         CLASSERT(sizeof(*lmv1) <= sizeof(info->lti_key));
1368
1369                         info->lti_buf.lb_buf = info->lti_key;
1370                         info->lti_buf.lb_len = sizeof(*lmv1);
1371                         rc = dt_xattr_get(env, dt_object_child(dt),
1372                                           &info->lti_buf, name);
1373                         if (unlikely(rc != sizeof(*lmv1)))
1374                                 RETURN(rc = rc > 0 ? -EINVAL : rc);
1375
1376                         lmv1 = info->lti_buf.lb_buf;
1377                         /* The on-disk LMV EA only contains header, but the
1378                          * returned LMV EA size should contain the space for
1379                          * the FIDs of all shards of the striped directory. */
1380                         if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1381                                 rc = lmv_mds_md_size(
1382                                         le32_to_cpu(lmv1->lmv_stripe_count),
1383                                         LMV_MAGIC_V1);
1384                 } else {
1385                         rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1386                                                   buf, false);
1387                 }
1388
1389                 RETURN(rc = rc1 != 0 ? rc1 : rc);
1390         }
1391
1392         if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1393                 RETURN(rc);
1394
1395         /*
1396          * XXX: Only used by lfsck
1397          *
1398          * lod returns default striping on the real root of the device
1399          * this is like the root stores default striping for the whole
1400          * filesystem. historically we've been using a different approach
1401          * and store it in the config.
1402          */
1403         dt_root_get(env, dev->lod_child, &info->lti_fid);
1404         is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1405
1406         if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1407                 struct lov_user_md *lum = buf->lb_buf;
1408                 struct lov_desc    *desc = &dev->lod_desc;
1409
1410                 if (buf->lb_buf == NULL) {
1411                         rc = sizeof(*lum);
1412                 } else if (buf->lb_len >= sizeof(*lum)) {
1413                         lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1414                         lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1415                         lmm_oi_set_id(&lum->lmm_oi, 0);
1416                         lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1417                         lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1418                         lum->lmm_stripe_size = cpu_to_le32(
1419                                                 desc->ld_default_stripe_size);
1420                         lum->lmm_stripe_count = cpu_to_le16(
1421                                                 desc->ld_default_stripe_count);
1422                         lum->lmm_stripe_offset = cpu_to_le16(
1423                                                 desc->ld_default_stripe_offset);
1424                         rc = sizeof(*lum);
1425                 } else {
1426                         rc = -ERANGE;
1427                 }
1428         }
1429
1430         RETURN(rc);
1431 }
1432
1433 /**
1434  * Verify LVM EA.
1435  *
1436  * Checks that the magic of the stripe is sane.
1437  *
1438  * \param[in] lod       lod device
1439  * \param[in] lum       a buffer storing LMV EA to verify
1440  *
1441  * \retval              0 if the EA is sane
1442  * \retval              negative otherwise
1443  */
1444 static int lod_verify_md_striping(struct lod_device *lod,
1445                                   const struct lmv_user_md_v1 *lum)
1446 {
1447         if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1448                 CERROR("%s: invalid lmv_user_md: magic = %x, "
1449                        "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1450                        lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1451                        (int)le32_to_cpu(lum->lum_stripe_offset),
1452                        le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1453                 return -EINVAL;
1454         }
1455
1456         return 0;
1457 }
1458
1459 /**
1460  * Initialize LMV EA for a slave.
1461  *
1462  * Initialize slave's LMV EA from the master's LMV EA.
1463  *
1464  * \param[in] master_lmv        a buffer containing master's EA
1465  * \param[out] slave_lmv        a buffer where slave's EA will be stored
1466  *
1467  */
1468 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1469                                   const struct lmv_mds_md_v1 *master_lmv)
1470 {
1471         *slave_lmv = *master_lmv;
1472         slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1473 }
1474
1475 /**
1476  * Generate LMV EA.
1477  *
1478  * Generate LMV EA from the object passed as \a dt. The object must have
1479  * the stripes created and initialized.
1480  *
1481  * \param[in] env       execution environment
1482  * \param[in] dt        object
1483  * \param[out] lmv_buf  buffer storing generated LMV EA
1484  *
1485  * \retval              0 on success
1486  * \retval              negative if failed
1487  */
1488 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1489                            struct lu_buf *lmv_buf)
1490 {
1491         struct lod_thread_info  *info = lod_env_info(env);
1492         struct lod_device       *lod = lu2lod_dev(dt->do_lu.lo_dev);
1493         struct lod_object       *lo = lod_dt_obj(dt);
1494         struct lmv_mds_md_v1    *lmm1;
1495         int                     stripe_count;
1496         int                     type = LU_SEQ_RANGE_ANY;
1497         int                     rc;
1498         __u32                   mdtidx;
1499         ENTRY;
1500
1501         LASSERT(lo->ldo_dir_striped != 0);
1502         LASSERT(lo->ldo_dir_stripe_count > 0);
1503         stripe_count = lo->ldo_dir_stripe_count;
1504         /* Only store the LMV EA heahder on the disk. */
1505         if (info->lti_ea_store_size < sizeof(*lmm1)) {
1506                 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1507                 if (rc != 0)
1508                         RETURN(rc);
1509         } else {
1510                 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1511         }
1512
1513         lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1514         lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1515         lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1516         lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1517         rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1518                             &mdtidx, &type);
1519         if (rc != 0)
1520                 RETURN(rc);
1521
1522         lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1523         lmv_buf->lb_buf = info->lti_ea_store;
1524         lmv_buf->lb_len = sizeof(*lmm1);
1525
1526         RETURN(rc);
1527 }
1528
1529 /**
1530  * Create in-core represenation for a striped directory.
1531  *
1532  * Parse the buffer containing LMV EA and instantiate LU objects
1533  * representing the stripe objects. The pointers to the objects are
1534  * stored in ldo_stripe field of \a lo. This function is used when
1535  * we need to access an already created object (i.e. load from a disk).
1536  *
1537  * \param[in] env       execution environment
1538  * \param[in] lo        lod object
1539  * \param[in] buf       buffer containing LMV EA
1540  *
1541  * \retval              0 on success
1542  * \retval              negative if failed
1543  */
1544 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1545                            const struct lu_buf *buf)
1546 {
1547         struct lod_thread_info  *info = lod_env_info(env);
1548         struct lod_device       *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1549         struct lod_tgt_descs    *ltd = &lod->lod_mdt_descs;
1550         struct dt_object        **stripe;
1551         union lmv_mds_md        *lmm = buf->lb_buf;
1552         struct lmv_mds_md_v1    *lmv1 = &lmm->lmv_md_v1;
1553         struct lu_fid           *fid = &info->lti_fid;
1554         unsigned int            i;
1555         int                     rc = 0;
1556         ENTRY;
1557
1558         if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
1559                 RETURN(0);
1560
1561         if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1562                 lo->ldo_dir_slave_stripe = 1;
1563                 RETURN(0);
1564         }
1565
1566         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1567                 RETURN(-EINVAL);
1568
1569         if (le32_to_cpu(lmv1->lmv_stripe_count) < 1)
1570                 RETURN(0);
1571
1572         LASSERT(lo->ldo_stripe == NULL);
1573         OBD_ALLOC(stripe, sizeof(stripe[0]) *
1574                   (le32_to_cpu(lmv1->lmv_stripe_count)));
1575         if (stripe == NULL)
1576                 RETURN(-ENOMEM);
1577
1578         for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1579                 struct dt_device        *tgt_dt;
1580                 struct dt_object        *dto;
1581                 int                     type = LU_SEQ_RANGE_ANY;
1582                 __u32                   idx;
1583
1584                 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1585                 if (!fid_is_sane(fid))
1586                         GOTO(out, rc = -ESTALE);
1587
1588                 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1589                 if (rc != 0)
1590                         GOTO(out, rc);
1591
1592                 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1593                         tgt_dt = lod->lod_child;
1594                 } else {
1595                         struct lod_tgt_desc     *tgt;
1596
1597                         tgt = LTD_TGT(ltd, idx);
1598                         if (tgt == NULL)
1599                                 GOTO(out, rc = -ESTALE);
1600                         tgt_dt = tgt->ltd_tgt;
1601                 }
1602
1603                 dto = dt_locate_at(env, tgt_dt, fid,
1604                                   lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1605                                   NULL);
1606                 if (IS_ERR(dto))
1607                         GOTO(out, rc = PTR_ERR(dto));
1608
1609                 stripe[i] = dto;
1610         }
1611 out:
1612         lo->ldo_stripe = stripe;
1613         lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1614         lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1615         if (rc != 0)
1616                 lod_object_free_striping(env, lo);
1617
1618         RETURN(rc);
1619 }
1620
1621 /**
1622  * Declare create a striped directory.
1623  *
1624  * Declare creating a striped directory with a given stripe pattern on the
1625  * specified MDTs. A striped directory is represented as a regular directory
1626  * - an index listing all the stripes. The stripes point back to the master
1627  * object with ".." and LinkEA. The master object gets LMV EA which
1628  * identifies it as a striped directory. The function allocates FIDs
1629  * for all stripes.
1630  *
1631  * \param[in] env       execution environment
1632  * \param[in] dt        object
1633  * \param[in] attr      attributes to initialize the objects with
1634  * \param[in] dof       type of objects to be created
1635  * \param[in] th        transaction handle
1636  *
1637  * \retval              0 on success
1638  * \retval              negative if failed
1639  */
1640 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1641                                           struct dt_object *dt,
1642                                           struct lu_attr *attr,
1643                                           struct dt_object_format *dof,
1644                                           struct thandle *th)
1645 {
1646         struct lod_thread_info  *info = lod_env_info(env);
1647         struct lu_buf           lmv_buf;
1648         struct lu_buf           slave_lmv_buf;
1649         struct lmv_mds_md_v1    *lmm;
1650         struct lmv_mds_md_v1    *slave_lmm = NULL;
1651         struct dt_insert_rec    *rec = &info->lti_dt_rec;
1652         struct lod_object       *lo = lod_dt_obj(dt);
1653         int                     rc;
1654         __u32                   i;
1655         ENTRY;
1656
1657         rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1658         if (rc != 0)
1659                 GOTO(out, rc);
1660         lmm = lmv_buf.lb_buf;
1661
1662         OBD_ALLOC_PTR(slave_lmm);
1663         if (slave_lmm == NULL)
1664                 GOTO(out, rc = -ENOMEM);
1665
1666         lod_prep_slave_lmv_md(slave_lmm, lmm);
1667         slave_lmv_buf.lb_buf = slave_lmm;
1668         slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1669
1670         if (!dt_try_as_dir(env, dt_object_child(dt)))
1671                 GOTO(out, rc = -EINVAL);
1672
1673         rec->rec_type = S_IFDIR;
1674         for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1675                 struct dt_object        *dto = lo->ldo_stripe[i];
1676                 char                    *stripe_name = info->lti_key;
1677                 struct lu_name          *sname;
1678                 struct linkea_data       ldata          = { NULL };
1679                 struct lu_buf           linkea_buf;
1680
1681                 rc = lod_sub_declare_create(env, dto, attr, NULL, dof, th);
1682                 if (rc != 0)
1683                         GOTO(out, rc);
1684
1685                 if (!dt_try_as_dir(env, dto))
1686                         GOTO(out, rc = -EINVAL);
1687
1688                 rc = lod_sub_declare_ref_add(env, dto, th);
1689                 if (rc != 0)
1690                         GOTO(out, rc);
1691
1692                 rec->rec_fid = lu_object_fid(&dto->do_lu);
1693                 rc = lod_sub_declare_insert(env, dto,
1694                                             (const struct dt_rec *)rec,
1695                                             (const struct dt_key *)dot, th);
1696                 if (rc != 0)
1697                         GOTO(out, rc);
1698
1699                 /* master stripe FID will be put to .. */
1700                 rec->rec_fid = lu_object_fid(&dt->do_lu);
1701                 rc = lod_sub_declare_insert(env, dto,
1702                                             (const struct dt_rec *)rec,
1703                                             (const struct dt_key *)dotdot, th);
1704                 if (rc != 0)
1705                         GOTO(out, rc);
1706
1707                 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1708                     cfs_fail_val != i) {
1709                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1710                             cfs_fail_val == i)
1711                                 slave_lmm->lmv_master_mdt_index =
1712                                                         cpu_to_le32(i + 1);
1713                         else
1714                                 slave_lmm->lmv_master_mdt_index =
1715                                                         cpu_to_le32(i);
1716                         rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
1717                                                        XATTR_NAME_LMV, 0, th);
1718                         if (rc != 0)
1719                                 GOTO(out, rc);
1720                 }
1721
1722                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1723                     cfs_fail_val == i)
1724                         snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1725                                 PFID(lu_object_fid(&dto->do_lu)), i + 1);
1726                 else
1727                         snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1728                                 PFID(lu_object_fid(&dto->do_lu)), i);
1729
1730                 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
1731                 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1732                                       sname, lu_object_fid(&dt->do_lu));
1733                 if (rc != 0)
1734                         GOTO(out, rc);
1735
1736                 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1737                 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1738                 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
1739                                                XATTR_NAME_LINK, 0, th);
1740                 if (rc != 0)
1741                         GOTO(out, rc);
1742
1743                 rec->rec_fid = lu_object_fid(&dto->do_lu);
1744                 rc = lod_sub_declare_insert(env, dt_object_child(dt),
1745                                             (const struct dt_rec *)rec,
1746                                             (const struct dt_key *)stripe_name,
1747                                             th);
1748                 if (rc != 0)
1749                         GOTO(out, rc);
1750
1751                 rc = lod_sub_declare_ref_add(env, dt_object_child(dt), th);
1752                 if (rc != 0)
1753                         GOTO(out, rc);
1754         }
1755
1756         rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
1757                                        &lmv_buf, XATTR_NAME_LMV, 0, th);
1758         if (rc != 0)
1759                 GOTO(out, rc);
1760 out:
1761         if (slave_lmm != NULL)
1762                 OBD_FREE_PTR(slave_lmm);
1763
1764         RETURN(rc);
1765 }
1766
1767 static int lod_prep_md_striped_create(const struct lu_env *env,
1768                                       struct dt_object *dt,
1769                                       struct lu_attr *attr,
1770                                       const struct lmv_user_md_v1 *lum,
1771                                       struct dt_object_format *dof,
1772                                       struct thandle *th)
1773 {
1774         struct lod_device       *lod = lu2lod_dev(dt->do_lu.lo_dev);
1775         struct lod_tgt_descs    *ltd = &lod->lod_mdt_descs;
1776         struct lod_object       *lo = lod_dt_obj(dt);
1777         struct dt_object        **stripe;
1778         __u32                   stripe_count;
1779         int                     *idx_array;
1780         __u32                   master_index;
1781         int                     rc = 0;
1782         __u32                   i;
1783         __u32                   j;
1784         ENTRY;
1785
1786         /* The lum has been verifed in lod_verify_md_striping */
1787         LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
1788         LASSERT(le32_to_cpu(lum->lum_stripe_count) > 0);
1789
1790         stripe_count = le32_to_cpu(lum->lum_stripe_count);
1791
1792         OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
1793         if (idx_array == NULL)
1794                 RETURN(-ENOMEM);
1795
1796         OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_count);
1797         if (stripe == NULL)
1798                 GOTO(out_free, rc = -ENOMEM);
1799
1800         /* Start index must be the master MDT */
1801         master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1802         idx_array[0] = master_index;
1803         for (i = 0; i < stripe_count; i++) {
1804                 struct lod_tgt_desc     *tgt = NULL;
1805                 struct dt_object        *dto;
1806                 struct lu_fid           fid = { 0 };
1807                 int                     idx;
1808                 struct lu_object_conf   conf = { 0 };
1809                 struct dt_device        *tgt_dt = NULL;
1810
1811                 /* Try to find next avaible target */
1812                 idx = idx_array[i];
1813                 for (j = 0; j < lod->lod_remote_mdt_count;
1814                      j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
1815                         bool already_allocated = false;
1816                         __u32 k;
1817
1818                         CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
1819                                idx, lod->lod_remote_mdt_count + 1, i);
1820
1821                         if (likely(!OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
1822                                 /* check whether the idx already exists
1823                                  * in current allocated array */
1824                                 for (k = 0; k < i; k++) {
1825                                         if (idx_array[k] == idx) {
1826                                                 already_allocated = true;
1827                                                 break;
1828                                         }
1829                                 }
1830
1831                                 if (already_allocated)
1832                                         continue;
1833                         }
1834
1835                         /* Sigh, this index is not in the bitmap, let's check
1836                          * next available target */
1837                         if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) &&
1838                             idx != master_index)
1839                                 continue;
1840
1841                         if (idx == master_index) {
1842                                 /* Allocate the FID locally */
1843                                 rc = obd_fid_alloc(env, lod->lod_child_exp,
1844                                                    &fid, NULL);
1845                                 if (rc < 0)
1846                                         GOTO(out_put, rc);
1847                                 tgt_dt = lod->lod_child;
1848                                 break;
1849                         }
1850
1851                         /* check the status of the OSP */
1852                         tgt = LTD_TGT(ltd, idx);
1853                         if (tgt == NULL)
1854                                 continue;
1855
1856                         tgt_dt = tgt->ltd_tgt;
1857                         rc = dt_statfs(env, tgt_dt, NULL);
1858                         if (rc) {
1859                                 /* this OSP doesn't feel well */
1860                                 rc = 0;
1861                                 continue;
1862                         }
1863
1864                         rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
1865                         if (rc < 0) {
1866                                 rc = 0;
1867                                 continue;
1868                         }
1869
1870                         break;
1871                 }
1872
1873                 /* Can not allocate more stripes */
1874                 if (j == lod->lod_remote_mdt_count) {
1875                         CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
1876                                lod2obd(lod)->obd_name, stripe_count, i);
1877                         break;
1878                 }
1879
1880                 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
1881                        idx, i, PFID(&fid));
1882                 idx_array[i] = idx;
1883                 /* Set the start index for next stripe allocation */
1884                 if (i < stripe_count - 1)
1885                         idx_array[i + 1] = (idx + 1) %
1886                                            (lod->lod_remote_mdt_count + 1);
1887                 /* tgt_dt and fid must be ready after search avaible OSP
1888                  * in the above loop */
1889                 LASSERT(tgt_dt != NULL);
1890                 LASSERT(fid_is_sane(&fid));
1891                 conf.loc_flags = LOC_F_NEW;
1892                 dto = dt_locate_at(env, tgt_dt, &fid,
1893                                    dt->do_lu.lo_dev->ld_site->ls_top_dev,
1894                                    &conf);
1895                 if (IS_ERR(dto))
1896                         GOTO(out_put, rc = PTR_ERR(dto));
1897                 stripe[i] = dto;
1898         }
1899
1900         lo->ldo_dir_stripe_loaded = 1;
1901         lo->ldo_dir_striped = 1;
1902         lo->ldo_stripe = stripe;
1903         lo->ldo_dir_stripe_count = i;
1904         lo->ldo_dir_stripes_allocated = stripe_count;
1905
1906         if (lo->ldo_dir_stripe_count == 0)
1907                 GOTO(out_put, rc = -ENOSPC);
1908
1909         rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
1910         if (rc != 0)
1911                 GOTO(out_put, rc);
1912
1913 out_put:
1914         if (rc < 0) {
1915                 for (i = 0; i < stripe_count; i++)
1916                         if (stripe[i] != NULL)
1917                                 dt_object_put(env, stripe[i]);
1918                 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_count);
1919                 lo->ldo_dir_stripe_count = 0;
1920                 lo->ldo_dir_stripes_allocated = 0;
1921                 lo->ldo_stripe = NULL;
1922         }
1923
1924 out_free:
1925         OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
1926
1927         RETURN(rc);
1928 }
1929
1930 /**
1931  * Declare create striped md object.
1932  *
1933  * The function declares intention to create a striped directory. This is a
1934  * wrapper for lod_prep_md_striped_create(). The only additional functionality
1935  * is to verify pattern \a lum_buf is good. Check that function for the details.
1936  *
1937  * \param[in] env       execution environment
1938  * \param[in] dt        object
1939  * \param[in] attr      attributes to initialize the objects with
1940  * \param[in] lum_buf   a pattern specifying the number of stripes and
1941  *                      MDT to start from
1942  * \param[in] dof       type of objects to be created
1943  * \param[in] th        transaction handle
1944  *
1945  * \retval              0 on success
1946  * \retval              negative if failed
1947  *
1948  */
1949 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
1950                                      struct dt_object *dt,
1951                                      struct lu_attr *attr,
1952                                      const struct lu_buf *lum_buf,
1953                                      struct dt_object_format *dof,
1954                                      struct thandle *th)
1955 {
1956         struct lod_object       *lo = lod_dt_obj(dt);
1957         struct lod_device       *lod = lu2lod_dev(dt->do_lu.lo_dev);
1958         struct lmv_user_md_v1   *lum;
1959         int                     rc;
1960         ENTRY;
1961
1962         lum = lum_buf->lb_buf;
1963         LASSERT(lum != NULL);
1964
1965         CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
1966                le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
1967                (int)le32_to_cpu(lum->lum_stripe_offset));
1968
1969         if (le32_to_cpu(lum->lum_stripe_count) == 0)
1970                 GOTO(out, rc = 0);
1971
1972         rc = lod_verify_md_striping(lod, lum);
1973         if (rc != 0)
1974                 GOTO(out, rc);
1975
1976         /* prepare dir striped objects */
1977         rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
1978         if (rc != 0) {
1979                 /* failed to create striping, let's reset
1980                  * config so that others don't get confused */
1981                 lod_object_free_striping(env, lo);
1982                 GOTO(out, rc);
1983         }
1984 out:
1985         RETURN(rc);
1986 }
1987
1988 /**
1989  * Implementation of dt_object_operations::do_declare_xattr_set.
1990  *
1991  * Used with regular (non-striped) objects. Basically it
1992  * initializes the striping information and applies the
1993  * change to all the stripes.
1994  *
1995  * \see dt_object_operations::do_declare_xattr_set() in the API description
1996  * for details.
1997  */
1998 static int lod_dir_declare_xattr_set(const struct lu_env *env,
1999                                      struct dt_object *dt,
2000                                      const struct lu_buf *buf,
2001                                      const char *name, int fl,
2002                                      struct thandle *th)
2003 {
2004         struct dt_object        *next = dt_object_child(dt);
2005         struct lod_device       *d = lu2lod_dev(dt->do_lu.lo_dev);
2006         struct lod_object       *lo = lod_dt_obj(dt);
2007         int                     i;
2008         int                     rc;
2009         ENTRY;
2010
2011         if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2012                 struct lmv_user_md_v1 *lum;
2013
2014                 LASSERT(buf != NULL && buf->lb_buf != NULL);
2015                 lum = buf->lb_buf;
2016                 rc = lod_verify_md_striping(d, lum);
2017                 if (rc != 0)
2018                         RETURN(rc);
2019         } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2020                 rc = lod_verify_striping(d, buf, false, 0);
2021                 if (rc != 0)
2022                         RETURN(rc);
2023         }
2024
2025         rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2026         if (rc != 0)
2027                 RETURN(rc);
2028
2029         /* Note: Do not set LinkEA on sub-stripes, otherwise
2030          * it will confuse the fid2path process(see mdt_path_current()).
2031          * The linkEA between master and sub-stripes is set in
2032          * lod_xattr_set_lmv(). */
2033         if (strcmp(name, XATTR_NAME_LINK) == 0)
2034                 RETURN(0);
2035
2036         /* set xattr to each stripes, if needed */
2037         rc = lod_load_striping(env, lo);
2038         if (rc != 0)
2039                 RETURN(rc);
2040
2041         if (lo->ldo_dir_stripe_count == 0)
2042                 RETURN(0);
2043
2044         for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2045                 LASSERT(lo->ldo_stripe[i]);
2046
2047                 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2048                                                buf, name, fl, th);
2049                 if (rc != 0)
2050                         break;
2051         }
2052
2053         RETURN(rc);
2054 }
2055
2056 static int
2057 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2058                                      struct lod_object *lo,
2059                                      struct dt_object *dt, struct thandle *th,
2060                                      int stripe_idx,
2061                                      struct lod_obj_stripe_cb_data *data)
2062 {
2063         struct lod_thread_info *info = lod_env_info(env);
2064         struct filter_fid *ff = &info->lti_ff;
2065         struct lu_buf *buf = &info->lti_buf;
2066         int rc;
2067
2068         buf->lb_buf = ff;
2069         buf->lb_len = sizeof(*ff);
2070         rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2071         if (rc == -ENODATA)
2072                 return 0;
2073
2074         if (rc < 0)
2075                 return rc;
2076
2077         ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2078         ff->ff_parent.f_ver = stripe_idx;
2079         fid_cpu_to_le(&ff->ff_parent, &ff->ff_parent);
2080         if (data->locd_declare)
2081                 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2082                                                LU_XATTR_REPLACE, th);
2083         else
2084                 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2085                                        LU_XATTR_REPLACE, th);
2086
2087         return rc;
2088 }
2089
2090 /**
2091  * Reset parent FID on OST object
2092  *
2093  * Replace parent FID with @dt object FID, which is only called during migration
2094  * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2095  * the FID is changed.
2096  *
2097  * \param[in] env execution environment
2098  * \param[in] dt dt_object whose stripes's parent FID will be reset
2099  * \parem[in] th thandle
2100  * \param[in] declare if it is declare
2101  *
2102  * \retval      0 if reset succeeds
2103  * \retval      negative errno if reset fails
2104  */
2105 static int lod_replace_parent_fid(const struct lu_env *env,
2106                                   struct dt_object *dt,
2107                                   struct thandle *th, bool declare)
2108 {
2109         struct lod_object *lo = lod_dt_obj(dt);
2110         struct lod_thread_info  *info = lod_env_info(env);
2111         struct lu_buf *buf = &info->lti_buf;
2112         struct filter_fid *ff;
2113         struct lod_obj_stripe_cb_data data;
2114         int rc;
2115         ENTRY;
2116
2117         LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2118
2119         /* set xattr to each stripes, if needed */
2120         rc = lod_load_striping(env, lo);
2121         if (rc != 0)
2122                 RETURN(rc);
2123
2124         if (!lod_obj_is_striped(dt))
2125                 RETURN(0);
2126
2127         if (info->lti_ea_store_size < sizeof(*ff)) {
2128                 rc = lod_ea_store_resize(info, sizeof(*ff));
2129                 if (rc != 0)
2130                         RETURN(rc);
2131         }
2132
2133         buf->lb_buf = info->lti_ea_store;
2134         buf->lb_len = info->lti_ea_store_size;
2135
2136         data.locd_declare = declare;
2137         rc = lod_obj_for_each_stripe(env, lo, th,
2138                                      lod_obj_stripe_replace_parent_fid_cb,
2139                                      &data);
2140
2141         RETURN(rc);
2142 }
2143
2144 inline __u16 lod_comp_entry_stripe_count(struct lod_object *lo,
2145                                          struct lod_layout_component *entry,
2146                                          bool is_dir)
2147 {
2148         struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2149
2150         if (is_dir)
2151                 return  0;
2152         else if (lod_comp_inited(entry))
2153                 return entry->llc_stripe_count;
2154         else if ((__u16)-1 == entry->llc_stripe_count)
2155                 return lod->lod_desc.ld_tgt_count;
2156         else
2157                 return lod_get_stripe_count(lod, lo, entry->llc_stripe_count);
2158 }
2159
2160 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2161 {
2162         int magic, size = 0, i;
2163         struct lod_layout_component *comp_entries;
2164         __u16 comp_cnt;
2165         bool is_composite;
2166
2167         if (is_dir) {
2168                 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2169                 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2170                 is_composite =
2171                         lo->ldo_def_striping->lds_def_striping_is_composite;
2172         } else {
2173                 comp_cnt = lo->ldo_comp_cnt;
2174                 comp_entries = lo->ldo_comp_entries;
2175                 is_composite = lo->ldo_is_composite;
2176         }
2177
2178
2179         LASSERT(comp_cnt != 0 && comp_entries != NULL);
2180         if (is_composite) {
2181                 size = sizeof(struct lov_comp_md_v1) +
2182                        sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2183                 LASSERT(size % sizeof(__u64) == 0);
2184         }
2185
2186         for (i = 0; i < comp_cnt; i++) {
2187                 __u16 stripe_count;
2188
2189                 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2190                 stripe_count = lod_comp_entry_stripe_count(lo, &comp_entries[i],
2191                                                            is_dir);
2192                 if (!is_dir && is_composite)
2193                         lod_comp_shrink_stripe_count(&comp_entries[i],
2194                                                      &stripe_count);
2195
2196                 size += lov_user_md_size(stripe_count, magic);
2197                 LASSERT(size % sizeof(__u64) == 0);
2198         }
2199         return size;
2200 }
2201
2202 /**
2203  * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2204  * the xattr value is binary lov_comp_md_v1 which contains component(s)
2205  * to be added.
2206   *
2207  * \param[in] env       execution environment
2208  * \param[in] dt        dt_object to add components on
2209  * \param[in] buf       buffer contains components to be added
2210  * \parem[in] th        thandle
2211  *
2212  * \retval      0 on success
2213  * \retval      negative errno on failure
2214  */
2215 static int lod_declare_layout_add(const struct lu_env *env,
2216                                   struct dt_object *dt,
2217                                   const struct lu_buf *buf,
2218                                   struct thandle *th)
2219 {
2220         struct lod_thread_info  *info = lod_env_info(env);
2221         struct lod_layout_component *comp_array, *lod_comp;
2222         struct lod_device       *d = lu2lod_dev(dt->do_lu.lo_dev);
2223         struct dt_object *next = dt_object_child(dt);
2224         struct lov_desc         *desc = &d->lod_desc;
2225         struct lod_object       *lo = lod_dt_obj(dt);
2226         struct lov_user_md_v3   *v3;
2227         struct lov_comp_md_v1   *comp_v1 = buf->lb_buf;
2228         __u32   magic;
2229         __u64   prev_end;
2230         int     i, rc, array_cnt;
2231         ENTRY;
2232
2233         LASSERT(lo->ldo_is_composite);
2234
2235         prev_end = lo->ldo_comp_entries[lo->ldo_comp_cnt - 1].llc_extent.e_end;
2236         rc = lod_verify_striping(d, buf, false, prev_end);
2237         if (rc != 0)
2238                 RETURN(rc);
2239
2240         magic = comp_v1->lcm_magic;
2241         if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2242                 lustre_swab_lov_comp_md_v1(comp_v1);
2243                 magic = comp_v1->lcm_magic;
2244         }
2245
2246         if (magic != LOV_USER_MAGIC_COMP_V1)
2247                 RETURN(-EINVAL);
2248
2249         array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2250         OBD_ALLOC(comp_array, sizeof(*comp_array) * array_cnt);
2251         if (comp_array == NULL)
2252                 RETURN(-ENOMEM);
2253
2254         memcpy(comp_array, lo->ldo_comp_entries,
2255                sizeof(*comp_array) * lo->ldo_comp_cnt);
2256
2257         for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2258                 struct lov_user_md_v1 *v1;
2259                 struct lu_extent *ext;
2260
2261                 v1 = (struct lov_user_md *)((char *)comp_v1 +
2262                                 comp_v1->lcm_entries[i].lcme_offset);
2263                 ext = &comp_v1->lcm_entries[i].lcme_extent;
2264
2265                 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2266                 lod_comp->llc_extent.e_start = ext->e_start;
2267                 lod_comp->llc_extent.e_end = ext->e_end;
2268                 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2269
2270                 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2271                 if (!lod_comp->llc_stripe_count ||
2272                     lod_comp->llc_stripe_count == (__u16)-1)
2273                         lod_comp->llc_stripe_count =
2274                                 desc->ld_default_stripe_count;
2275                 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2276                 if (!lod_comp->llc_stripe_size)
2277                         lod_comp->llc_stripe_size =
2278                                 desc->ld_default_stripe_size;
2279
2280                 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2281                         v3 = (struct lov_user_md_v3 *) v1;
2282                         if (v3->lmm_pool_name[0] != '\0') {
2283                                 rc = lod_set_pool(&lod_comp->llc_pool,
2284                                                   v3->lmm_pool_name);
2285                                 if (rc)
2286                                         GOTO(error, rc);
2287                         }
2288                 }
2289         }
2290
2291         OBD_FREE(lo->ldo_comp_entries, sizeof(*lod_comp) * lo->ldo_comp_cnt);
2292         lo->ldo_comp_entries = comp_array;
2293         lo->ldo_comp_cnt = array_cnt;
2294         /* No need to increase layout generation here, it will be increased
2295          * later when generating component ID for the new components */
2296
2297         info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2298         rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2299                                               XATTR_NAME_LOV, 0, th);
2300         if (rc)
2301                 GOTO(error, rc);
2302
2303         RETURN(0);
2304
2305 error:
2306         for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2307                 lod_comp = &comp_array[i];
2308                 if (lod_comp->llc_pool != NULL) {
2309                         OBD_FREE(lod_comp->llc_pool,
2310                                  strlen(lod_comp->llc_pool) + 1);
2311                         lod_comp->llc_pool = NULL;
2312                 }
2313         }
2314         OBD_FREE(comp_array, sizeof(*comp_array) * array_cnt);
2315         RETURN(rc);
2316 }
2317
2318 /**
2319  * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2320  * the '$field' can only be 'flags' now. The xattr value is binary
2321  * lov_comp_md_v1 which contains the component ID(s) and the value of
2322  * the field to be modified.
2323  *
2324  * \param[in] env       execution environment
2325  * \param[in] dt        dt_object to be modified
2326  * \param[in] op        operation string, like "set.flags"
2327  * \param[in] buf       buffer contains components to be set
2328  * \parem[in] th        thandle
2329  *
2330  * \retval      0 on success
2331  * \retval      negative errno on failure
2332  */
2333 static int lod_declare_layout_set(const struct lu_env *env,
2334                                   struct dt_object *dt,
2335                                   char *op, const struct lu_buf *buf,
2336                                   struct thandle *th)
2337 {
2338         struct lod_layout_component     *lod_comp;
2339         struct lod_thread_info  *info = lod_env_info(env);
2340         struct lod_device       *d = lu2lod_dev(dt->do_lu.lo_dev);
2341         struct lod_object       *lo = lod_dt_obj(dt);
2342         struct lov_comp_md_v1   *comp_v1 = buf->lb_buf;
2343         __u32   magic, id;
2344         int     i, j, rc;
2345         bool    changed = false;
2346         ENTRY;
2347
2348         if (strcmp(op, "set.flags") != 0) {
2349                 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2350                        lod2obd(d)->obd_name, op);
2351                 RETURN(-ENOTSUPP);
2352         }
2353
2354         magic = comp_v1->lcm_magic;
2355         if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2356                 lustre_swab_lov_comp_md_v1(comp_v1);
2357                 magic = comp_v1->lcm_magic;
2358         }
2359
2360         if (magic != LOV_USER_MAGIC_COMP_V1)
2361                 RETURN(-EINVAL);
2362
2363         if (comp_v1->lcm_entry_count == 0) {
2364                 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2365                        lod2obd(d)->obd_name);
2366                 RETURN(-EINVAL);
2367         }
2368
2369         for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2370                 id = comp_v1->lcm_entries[i].lcme_id;
2371
2372                 for (j = 0; j < lo->ldo_comp_cnt; j++) {
2373                         lod_comp = &lo->ldo_comp_entries[j];
2374                         if (id == lod_comp->llc_id || id == LCME_ID_ALL) {
2375                                 lod_comp->llc_flags =
2376                                         comp_v1->lcm_entries[i].lcme_flags;
2377                                 changed = true;
2378                         }
2379                 }
2380         }
2381
2382         if (!changed) {
2383                 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
2384                        lod2obd(d)->obd_name);
2385                 RETURN(-EINVAL);
2386         }
2387
2388         lod_obj_inc_layout_gen(lo);
2389
2390         info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2391         rc = lod_sub_declare_xattr_set(env, dt, &info->lti_buf,
2392                                        XATTR_NAME_LOV, 0, th);
2393         RETURN(rc);
2394 }
2395
2396 /**
2397  * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
2398  * and the xattr value is a unique component ID or a special lcme_id.
2399  *
2400  * \param[in] env       execution environment
2401  * \param[in] dt        dt_object to be operated on
2402  * \param[in] buf       buffer contains component ID or lcme_id
2403  * \parem[in] th        thandle
2404  *
2405  * \retval      0 on success
2406  * \retval      negative errno on failure
2407  */
2408 static int lod_declare_layout_del(const struct lu_env *env,
2409                                   struct dt_object *dt,
2410                                   const struct lu_buf *buf,
2411                                   struct thandle *th)
2412 {
2413         struct lod_thread_info  *info = lod_env_info(env);
2414         struct dt_object *next = dt_object_child(dt);
2415         struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2416         struct lod_object *lo = lod_dt_obj(dt);
2417         struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2418         struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2419         __u32 magic, id, flags, neg_flags = 0;
2420         int rc, i, j, left;
2421         ENTRY;
2422
2423         LASSERT(lo->ldo_is_composite);
2424
2425         rc = lod_verify_striping(d, buf, false, 0);
2426         if (rc != 0)
2427                 RETURN(rc);
2428
2429         magic = comp_v1->lcm_magic;
2430         if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2431                 lustre_swab_lov_comp_md_v1(comp_v1);
2432                 magic = comp_v1->lcm_magic;
2433         }
2434
2435         if (magic != LOV_USER_MAGIC_COMP_V1)
2436                 RETURN(-EINVAL);
2437
2438         id = comp_v1->lcm_entries[0].lcme_id;
2439         flags = comp_v1->lcm_entries[0].lcme_flags;
2440
2441         if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
2442                 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
2443                        lod2obd(d)->obd_name, id, flags);
2444                 RETURN(-EINVAL);
2445         }
2446
2447         if (id != LCME_ID_INVAL && flags != 0) {
2448                 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
2449                        lod2obd(d)->obd_name);
2450                 RETURN(-EINVAL);
2451         }
2452
2453         if (flags & LCME_FL_NEG) {
2454                 neg_flags = flags & ~LCME_FL_NEG;
2455                 flags = 0;
2456         }
2457
2458         left = lo->ldo_comp_cnt;
2459         if (left <= 0)
2460                 RETURN(-EINVAL);
2461
2462         for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
2463                 struct lod_layout_component *lod_comp;
2464
2465                 lod_comp = &lo->ldo_comp_entries[i];
2466
2467                 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
2468                         continue;
2469                 else if (flags && !(flags & lod_comp->llc_flags))
2470                         continue;
2471                 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
2472                         continue;
2473
2474                 if (left != (i + 1)) {
2475                         CDEBUG(D_LAYOUT, "%s: this deletion will create "
2476                                "a hole.\n", lod2obd(d)->obd_name);
2477                         RETURN(-EINVAL);
2478                 }
2479                 left--;
2480
2481                 /* Mark the component as deleted */
2482                 lod_comp->llc_id = LCME_ID_INVAL;
2483
2484                 /* Not instantiated component */
2485                 if (lod_comp->llc_stripe == NULL)
2486                         continue;
2487
2488                 LASSERT(lod_comp->llc_stripe_count > 0);
2489                 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
2490                         struct dt_object *obj = lod_comp->llc_stripe[j];
2491
2492                         if (obj == NULL)
2493                                 continue;
2494                         rc = lod_sub_declare_destroy(env, obj, th);
2495                         if (rc)
2496                                 RETURN(rc);
2497                 }
2498         }
2499
2500         LASSERTF(left >= 0, "left = %d\n", left);
2501         if (left == lo->ldo_comp_cnt) {
2502                 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
2503                        lod2obd(d)->obd_name, id);
2504                 RETURN(-EINVAL);
2505         }
2506
2507         memset(attr, 0, sizeof(*attr));
2508         attr->la_valid = LA_SIZE;
2509         rc = lod_sub_declare_attr_set(env, next, attr, th);
2510         if (rc)
2511                 RETURN(rc);
2512
2513         if (left > 0) {
2514                 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2515                 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2516                                                XATTR_NAME_LOV, 0, th);
2517         } else {
2518                 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
2519         }
2520
2521         RETURN(rc);
2522 }
2523
2524 /**
2525  * Declare layout add/set/del operations issued by special xattr names:
2526  *
2527  * XATTR_LUSTRE_LOV.add         add component(s) to existing file
2528  * XATTR_LUSTRE_LOV.del         delete component(s) from existing file
2529  * XATTR_LUSTRE_LOV.set.$field  set specified field of certain component(s)
2530  *
2531  * \param[in] env       execution environment
2532  * \param[in] dt        object
2533  * \param[in] name      name of xattr
2534  * \param[in] buf       lu_buf contains xattr value
2535  * \param[in] th        transaction handle
2536  *
2537  * \retval              0 on success
2538  * \retval              negative if failed
2539  */
2540 static int lod_declare_modify_layout(const struct lu_env *env,
2541                                      struct dt_object *dt,
2542                                      const char *name,
2543                                      const struct lu_buf *buf,
2544                                      struct thandle *th)
2545 {
2546         struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2547         struct lod_object *lo = lod_dt_obj(dt);
2548         struct dt_object *next = dt_object_child(&lo->ldo_obj);
2549         char *op;
2550         int rc, len = strlen(XATTR_LUSTRE_LOV);
2551         ENTRY;
2552
2553         LASSERT(dt_object_exists(dt));
2554
2555         if (strlen(name) <= len || name[len] != '.') {
2556                 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
2557                        lod2obd(d)->obd_name, name);
2558                 RETURN(-EINVAL);
2559         }
2560         len++;
2561
2562         dt_write_lock(env, next, 0);
2563         rc = lod_load_striping_locked(env, lo);
2564         if (rc)
2565                 GOTO(unlock, rc);
2566
2567         /* the layout to be modified must be a composite layout */
2568         if (!lo->ldo_is_composite) {
2569                 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
2570                        lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
2571                 GOTO(unlock, rc = -EINVAL);
2572         }
2573
2574         op = (char *)name + len;
2575         if (strcmp(op, "add") == 0) {
2576                 rc = lod_declare_layout_add(env, dt, buf, th);
2577         } else if (strcmp(op, "del") == 0) {
2578                 rc = lod_declare_layout_del(env, dt, buf, th);
2579         } else if (strncmp(op, "set", strlen("set")) == 0) {
2580                 rc = lod_declare_layout_set(env, dt, op, buf, th);
2581         } else  {
2582                 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
2583                        lod2obd(d)->obd_name, name);
2584                 GOTO(unlock, rc = -ENOTSUPP);
2585         }
2586 unlock:
2587         if (rc)
2588                 lod_object_free_striping(env, lo);
2589         dt_write_unlock(env, next);
2590
2591         RETURN(rc);
2592 }
2593
2594 /**
2595  * Implementation of dt_object_operations::do_declare_xattr_set.
2596  *
2597  * \see dt_object_operations::do_declare_xattr_set() in the API description
2598  * for details.
2599  *
2600  * the extension to the API:
2601  *   - declaring LOVEA requests striping creation
2602  *   - LU_XATTR_REPLACE means layout swap
2603  */
2604 static int lod_declare_xattr_set(const struct lu_env *env,
2605                                  struct dt_object *dt,
2606                                  const struct lu_buf *buf,
2607                                  const char *name, int fl,
2608                                  struct thandle *th)
2609 {
2610         struct dt_object *next = dt_object_child(dt);
2611         struct lu_attr   *attr = &lod_env_info(env)->lti_attr;
2612         __u32             mode;
2613         int               rc;
2614         ENTRY;
2615
2616         mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
2617         if ((S_ISREG(mode) || mode == 0) && !(fl & LU_XATTR_REPLACE) &&
2618             (strcmp(name, XATTR_NAME_LOV) == 0 ||
2619              strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
2620                 /*
2621                  * this is a request to create object's striping.
2622                  *
2623                  * allow to declare predefined striping on a new (!mode) object
2624                  * which is supposed to be replay of regular file creation
2625                  * (when LOV setting is declared)
2626                  *
2627                  * LU_XATTR_REPLACE is set to indicate a layout swap
2628                  */
2629                 if (dt_object_exists(dt)) {
2630                         rc = dt_attr_get(env, next, attr);
2631                         if (rc)
2632                                 RETURN(rc);
2633                 } else {
2634                         memset(attr, 0, sizeof(*attr));
2635                         attr->la_valid = LA_TYPE | LA_MODE;
2636                         attr->la_mode = S_IFREG;
2637                 }
2638                 rc = lod_declare_striped_create(env, dt, attr, buf, th);
2639         } else if (S_ISREG(mode) &&
2640                    strlen(name) > strlen(XATTR_LUSTRE_LOV) + 1 &&
2641                    strncmp(name, XATTR_LUSTRE_LOV,
2642                            strlen(XATTR_LUSTRE_LOV)) == 0) {
2643                 /*
2644                  * this is a request to modify object's striping.
2645                  * add/set/del component(s).
2646                  */
2647                 if (!dt_object_exists(dt))
2648                         RETURN(-ENOENT);
2649
2650                 rc = lod_declare_modify_layout(env, dt, name, buf, th);
2651         } else if (S_ISDIR(mode)) {
2652                 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
2653         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
2654                 rc = lod_replace_parent_fid(env, dt, th, true);
2655         } else {
2656                 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2657         }
2658
2659         RETURN(rc);
2660 }
2661
2662 /**
2663  * Apply xattr changes to the object.
2664  *
2665  * Applies xattr changes to the object and the stripes if the latter exist.
2666  *
2667  * \param[in] env       execution environment
2668  * \param[in] dt        object
2669  * \param[in] buf       buffer pointing to the new value of xattr
2670  * \param[in] name      name of xattr
2671  * \param[in] fl        flags
2672  * \param[in] th        transaction handle
2673  *
2674  * \retval              0 on success
2675  * \retval              negative if failed
2676  */
2677 static int lod_xattr_set_internal(const struct lu_env *env,
2678                                   struct dt_object *dt,
2679                                   const struct lu_buf *buf,
2680                                   const char *name, int fl,
2681                                   struct thandle *th)
2682 {
2683         struct dt_object        *next = dt_object_child(dt);
2684         struct lod_object       *lo = lod_dt_obj(dt);
2685         int                     rc;
2686         int                     i;
2687         ENTRY;
2688
2689         rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
2690         if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2691                 RETURN(rc);
2692
2693         /* Note: Do not set LinkEA on sub-stripes, otherwise
2694          * it will confuse the fid2path process(see mdt_path_current()).
2695          * The linkEA between master and sub-stripes is set in
2696          * lod_xattr_set_lmv(). */
2697         if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
2698                 RETURN(0);
2699
2700         for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2701                 LASSERT(lo->ldo_stripe[i]);
2702
2703                 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
2704                                        fl, th);
2705                 if (rc != 0)
2706                         break;
2707         }
2708
2709         RETURN(rc);
2710 }
2711
2712 /**
2713  * Delete an extended attribute.
2714  *
2715  * Deletes specified xattr from the object and the stripes if the latter exist.
2716  *
2717  * \param[in] env       execution environment
2718  * \param[in] dt        object
2719  * \param[in] name      name of xattr
2720  * \param[in] th        transaction handle
2721  *
2722  * \retval              0 on success
2723  * \retval              negative if failed
2724  */
2725 static int lod_xattr_del_internal(const struct lu_env *env,
2726                                   struct dt_object *dt,
2727                                   const char *name, struct thandle *th)
2728 {
2729         struct dt_object        *next = dt_object_child(dt);
2730         struct lod_object       *lo = lod_dt_obj(dt);
2731         int                     rc;
2732         int                     i;
2733         ENTRY;
2734
2735         rc = lod_sub_xattr_del(env, next, name, th);
2736         if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2737                 RETURN(rc);
2738
2739         if (lo->ldo_dir_stripe_count == 0)
2740                 RETURN(rc);
2741
2742         for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2743                 LASSERT(lo->ldo_stripe[i]);
2744
2745                 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
2746                 if (rc != 0)
2747                         break;
2748         }
2749
2750         RETURN(rc);
2751 }
2752
2753 /**
2754  * Set default striping on a directory.
2755  *
2756  * Sets specified striping on a directory object unless it matches the default
2757  * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
2758  * EA. This striping will be used when regular file is being created in this
2759  * directory.
2760  *
2761  * \param[in] env       execution environment
2762  * \param[in] dt        the striped object
2763  * \param[in] buf       buffer with the striping
2764  * \param[in] name      name of EA
2765  * \param[in] fl        xattr flag (see OSD API description)
2766  * \param[in] th        transaction handle
2767  *
2768  * \retval              0 on success
2769  * \retval              negative if failed
2770  */
2771 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
2772                                     struct dt_object *dt,
2773                                     const struct lu_buf *buf,
2774                                     const char *name, int fl,
2775                                     struct thandle *th)
2776 {
2777         struct lov_user_md_v1   *lum;
2778         struct lov_user_md_v3   *v3 = NULL;
2779         const char              *pool_name = NULL;
2780         int                      rc;
2781         bool                     is_del;
2782         ENTRY;
2783
2784         LASSERT(buf != NULL && buf->lb_buf != NULL);
2785         lum = buf->lb_buf;
2786
2787         switch (lum->lmm_magic) {
2788         case LOV_USER_MAGIC_V3:
2789                 v3 = buf->lb_buf;
2790                 if (v3->lmm_pool_name[0] != '\0')
2791                         pool_name = v3->lmm_pool_name;
2792                 /* fall through */
2793         case LOV_USER_MAGIC_V1:
2794                 /* if { size, offset, count } = { 0, -1, 0 } and no pool
2795                  * (i.e. all default values specified) then delete default
2796                  * striping from dir. */
2797                 CDEBUG(D_LAYOUT,
2798                        "set default striping: sz %u # %u offset %d %s %s\n",
2799                        (unsigned)lum->lmm_stripe_size,
2800                        (unsigned)lum->lmm_stripe_count,
2801                        (int)lum->lmm_stripe_offset,
2802                        v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
2803
2804                 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
2805                                              lum->lmm_stripe_count,
2806                                              lum->lmm_stripe_offset,
2807                                              pool_name);
2808                 break;
2809         case LOV_USER_MAGIC_COMP_V1:
2810                 is_del = false;
2811                 break;
2812         default:
2813                 CERROR("Invalid magic %x\n", lum->lmm_magic);
2814                 RETURN(-EINVAL);
2815         }
2816
2817         if (is_del) {
2818                 rc = lod_xattr_del_internal(env, dt, name, th);
2819                 if (rc == -ENODATA)
2820                         rc = 0;
2821         } else {
2822                 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
2823         }
2824
2825         RETURN(rc);
2826 }
2827
2828 /**
2829  * Set default striping on a directory object.
2830  *
2831  * Sets specified striping on a directory object unless it matches the default
2832  * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
2833  * EA. This striping will be used when a new directory is being created in the
2834  * directory.
2835  *
2836  * \param[in] env       execution environment
2837  * \param[in] dt        the striped object
2838  * \param[in] buf       buffer with the striping
2839  * \param[in] name      name of EA
2840  * \param[in] fl        xattr flag (see OSD API description)
2841  * \param[in] th        transaction handle
2842  *
2843  * \retval              0 on success
2844  * \retval              negative if failed
2845  */
2846 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
2847                                             struct dt_object *dt,
2848                                             const struct lu_buf *buf,
2849                                             const char *name, int fl,
2850                                             struct thandle *th)
2851 {
2852         struct lmv_user_md_v1   *lum;
2853         int                      rc;
2854         ENTRY;
2855
2856         LASSERT(buf != NULL && buf->lb_buf != NULL);
2857         lum = buf->lb_buf;
2858
2859         CDEBUG(D_OTHER, "set default stripe_count # %u stripe_offset %d\n",
2860               le32_to_cpu(lum->lum_stripe_count),
2861               (int)le32_to_cpu(lum->lum_stripe_offset));
2862
2863         if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
2864                                  le32_to_cpu(lum->lum_stripe_offset)) &&
2865                                 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
2866                 rc = lod_xattr_del_internal(env, dt, name, th);
2867                 if (rc == -ENODATA)
2868                         rc = 0;
2869         } else {
2870                 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
2871                 if (rc != 0)
2872                         RETURN(rc);
2873         }
2874
2875         RETURN(rc);
2876 }
2877
2878 /**
2879  * Turn directory into a striped directory.
2880  *
2881  * During replay the client sends the striping created before MDT
2882  * failure, then the layer above LOD sends this defined striping
2883  * using ->do_xattr_set(), so LOD uses this method to replay creation
2884  * of the stripes. Notice the original information for the striping
2885  * (#stripes, FIDs, etc) was transferred in declare path.
2886  *
2887  * \param[in] env       execution environment
2888  * \param[in] dt        the striped object
2889  * \param[in] buf       not used currently
2890  * \param[in] name      not used currently
2891  * \param[in] fl        xattr flag (see OSD API description)
2892  * \param[in] th        transaction handle
2893  *
2894  * \retval              0 on success
2895  * \retval              negative if failed
2896  */
2897 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
2898                              const struct lu_buf *buf, const char *name,
2899                              int fl, struct thandle *th)
2900 {
2901         struct lod_object       *lo = lod_dt_obj(dt);
2902         struct lod_thread_info  *info = lod_env_info(env);
2903         struct lu_attr          *attr = &info->lti_attr;
2904         struct dt_object_format *dof = &info->lti_format;
2905         struct lu_buf           lmv_buf;
2906         struct lu_buf           slave_lmv_buf;
2907         struct lmv_mds_md_v1    *lmm;
2908         struct lmv_mds_md_v1    *slave_lmm = NULL;
2909         struct dt_insert_rec    *rec = &info->lti_dt_rec;
2910         int                     i;
2911         int                     rc;
2912         ENTRY;
2913
2914         if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
2915                 RETURN(-ENOTDIR);
2916
2917         /* The stripes are supposed to be allocated in declare phase,
2918          * if there are no stripes being allocated, it will skip */
2919         if (lo->ldo_dir_stripe_count == 0)
2920                 RETURN(0);
2921
2922         rc = dt_attr_get(env, dt_object_child(dt), attr);
2923         if (rc != 0)
2924                 RETURN(rc);
2925
2926         attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME |
2927                          LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
2928         dof->dof_type = DFT_DIR;
2929
2930         rc = lod_prep_lmv_md(env, dt, &lmv_buf);
2931         if (rc != 0)
2932                 RETURN(rc);
2933         lmm = lmv_buf.lb_buf;
2934
2935         OBD_ALLOC_PTR(slave_lmm);
2936         if (slave_lmm == NULL)
2937                 RETURN(-ENOMEM);
2938
2939         lod_prep_slave_lmv_md(slave_lmm, lmm);
2940         slave_lmv_buf.lb_buf = slave_lmm;
2941         slave_lmv_buf.lb_len = sizeof(*slave_lmm);
2942
2943         rec->rec_type = S_IFDIR;
2944         for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2945                 struct dt_object *dto;
2946                 char             *stripe_name = info->lti_key;
2947                 struct lu_name          *sname;
2948                 struct linkea_data       ldata          = { NULL };
2949                 struct lu_buf            linkea_buf;
2950
2951                 dto = lo->ldo_stripe[i];
2952
2953                 dt_write_lock(env, dto, MOR_TGT_CHILD);
2954                 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
2955                 if (rc != 0) {
2956                         dt_write_unlock(env, dto);
2957                         GOTO(out, rc);
2958                 }
2959
2960                 rc = lod_sub_ref_add(env, dto, th);
2961                 dt_write_unlock(env, dto);
2962                 if (rc != 0)
2963                         GOTO(out, rc);
2964
2965                 rec->rec_fid = lu_object_fid(&dto->do_lu);
2966                 rc = lod_sub_insert(env, dto, (const struct dt_rec *)rec,
2967                                     (const struct dt_key *)dot, th, 0);
2968                 if (rc != 0)
2969                         GOTO(out, rc);
2970
2971                 rec->rec_fid = lu_object_fid(&dt->do_lu);
2972                 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
2973                                     (const struct dt_key *)dotdot, th, 0);
2974                 if (rc != 0)
2975                         GOTO(out, rc);
2976
2977                 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
2978                     cfs_fail_val != i) {
2979                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
2980                             cfs_fail_val == i)
2981                                 slave_lmm->lmv_master_mdt_index =
2982                                                         cpu_to_le32(i + 1);
2983                         else
2984                                 slave_lmm->lmv_master_mdt_index =
2985                                                         cpu_to_le32(i);
2986
2987                         rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
2988                                                XATTR_NAME_LMV, fl, th);
2989                         if (rc != 0)
2990                                 GOTO(out, rc);
2991                 }
2992
2993                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
2994                     cfs_fail_val == i)
2995                         snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2996                                  PFID(lu_object_fid(&dto->do_lu)), i + 1);
2997                 else
2998                         snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2999                                  PFID(lu_object_fid(&dto->do_lu)), i);
3000
3001                 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
3002                 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
3003                                       sname, lu_object_fid(&dt->do_lu));
3004                 if (rc != 0)
3005                         GOTO(out, rc);
3006
3007                 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
3008                 linkea_buf.lb_len = ldata.ld_leh->leh_len;
3009                 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
3010                                        XATTR_NAME_LINK, 0, th);
3011                 if (rc != 0)
3012                         GOTO(out, rc);
3013
3014                 rec->rec_fid = lu_object_fid(&dto->do_lu);
3015                 rc = lod_sub_insert(env, dt_object_child(dt),
3016                                     (const struct dt_rec *)rec,
3017                                     (const struct dt_key *)stripe_name, th, 0);
3018                 if (rc != 0)
3019                         GOTO(out, rc);
3020
3021                 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
3022                 if (rc != 0)
3023                         GOTO(out, rc);
3024         }
3025
3026         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
3027                 rc = lod_sub_xattr_set(env, dt_object_child(dt),
3028                                        &lmv_buf, XATTR_NAME_LMV, fl, th);
3029 out:
3030         if (slave_lmm != NULL)
3031                 OBD_FREE_PTR(slave_lmm);
3032
3033         RETURN(rc);
3034 }
3035
3036 /**
3037  * Helper function to declare/execute creation of a striped directory
3038  *
3039  * Called in declare/create object path, prepare striping for a directory
3040  * and prepare defaults data striping for the objects to be created in
3041  * that directory. Notice the function calls "declaration" or "execution"
3042  * methods depending on \a declare param. This is a consequence of the
3043  * current approach while we don't have natural distributed transactions:
3044  * we basically execute non-local updates in the declare phase. So, the
3045  * arguments for the both phases are the same and this is the reason for
3046  * this function to exist.
3047  *
3048  * \param[in] env       execution environment
3049  * \param[in] dt        object
3050  * \param[in] attr      attributes the stripes will be created with
3051  * \param[in] dof       format of stripes (see OSD API description)
3052  * \param[in] th        transaction handle
3053  * \param[in] declare   where to call "declare" or "execute" methods
3054  *
3055  * \retval              0 on success
3056  * \retval              negative if failed
3057  */
3058 static int lod_dir_striping_create_internal(const struct lu_env *env,
3059                                             struct dt_object *dt,
3060                                             struct lu_attr *attr,
3061                                             struct dt_object_format *dof,
3062                                             struct thandle *th,
3063                                             bool declare)
3064 {
3065         struct lod_thread_info *info = lod_env_info(env);
3066         struct lod_object *lo = lod_dt_obj(dt);
3067         const struct lod_default_striping *lds = lo->ldo_def_striping;
3068         int rc;
3069         ENTRY;
3070
3071         LASSERT(ergo(lds != NULL,
3072                      lds->lds_def_striping_set ||
3073                      lds->lds_dir_def_striping_set));
3074
3075         if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
3076                                  lo->ldo_dir_stripe_offset)) {
3077                 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3078                 int stripe_count = lo->ldo_dir_stripe_count;
3079
3080                 if (info->lti_ea_store_size < sizeof(*v1)) {
3081                         rc = lod_ea_store_resize(info, sizeof(*v1));
3082                         if (rc != 0)
3083                                 RETURN(rc);
3084                         v1 = info->lti_ea_store;
3085                 }
3086
3087                 memset(v1, 0, sizeof(*v1));
3088                 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3089                 v1->lum_stripe_count = cpu_to_le32(stripe_count);
3090                 v1->lum_stripe_offset =
3091                                 cpu_to_le32(lo->ldo_dir_stripe_offset);
3092
3093                 info->lti_buf.lb_buf = v1;
3094                 info->lti_buf.lb_len = sizeof(*v1);
3095
3096                 if (declare)
3097                         rc = lod_declare_xattr_set_lmv(env, dt, attr,
3098                                                        &info->lti_buf, dof, th);
3099                 else
3100                         rc = lod_xattr_set_lmv(env, dt, &info->lti_buf,
3101                                                XATTR_NAME_LMV, 0, th);
3102                 if (rc != 0)
3103                         RETURN(rc);
3104         }
3105
3106         /* Transfer default LMV striping from the parent */
3107         if (lds != NULL && lds->lds_dir_def_striping_set &&
3108             !LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
3109                                  lds->lds_dir_def_stripe_offset)) {
3110                 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3111
3112                 if (info->lti_ea_store_size < sizeof(*v1)) {
3113                         rc = lod_ea_store_resize(info, sizeof(*v1));
3114                         if (rc != 0)
3115                                 RETURN(rc);
3116                         v1 = info->lti_ea_store;
3117                 }
3118
3119                 memset(v1, 0, sizeof(*v1));
3120                 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3121                 v1->lum_stripe_count =
3122                         cpu_to_le32(lds->lds_dir_def_stripe_count);
3123                 v1->lum_stripe_offset =
3124                         cpu_to_le32(lds->lds_dir_def_stripe_offset);
3125                 v1->lum_hash_type =
3126                         cpu_to_le32(lds->lds_dir_def_hash_type);
3127
3128                 info->lti_buf.lb_buf = v1;
3129                 info->lti_buf.lb_len = sizeof(*v1);
3130                 if (declare)
3131                         rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3132                                                        XATTR_NAME_DEFAULT_LMV,
3133                                                        0, th);
3134                 else
3135                         rc = lod_xattr_set_default_lmv_on_dir(env, dt,
3136                                                   &info->lti_buf,
3137                                                   XATTR_NAME_DEFAULT_LMV, 0,
3138                                                   th);
3139                 if (rc != 0)
3140                         RETURN(rc);
3141         }
3142
3143         /* Transfer default LOV striping from the parent */
3144         if (lds != NULL && lds->lds_def_striping_set &&
3145             lds->lds_def_comp_cnt != 0) {
3146                 struct lov_mds_md *lmm;
3147                 int lmm_size = lod_comp_md_size(lo, true);
3148
3149                 if (info->lti_ea_store_size < lmm_size) {
3150                         rc = lod_ea_store_resize(info, lmm_size);
3151                         if (rc != 0)
3152                                 RETURN(rc);
3153                 }
3154                 lmm = info->lti_ea_store;
3155
3156                 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
3157                 if (rc != 0)
3158                         RETURN(rc);
3159
3160                 info->lti_buf.lb_buf = lmm;
3161                 info->lti_buf.lb_len = lmm_size;
3162
3163                 if (declare)
3164                         rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3165                                                        XATTR_NAME_LOV, 0, th);
3166                 else
3167                         rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
3168                                                       XATTR_NAME_LOV, 0, th);
3169                 if (rc != 0)
3170                         RETURN(rc);
3171         }
3172
3173         RETURN(0);
3174 }
3175
3176 static int lod_declare_dir_striping_create(const struct lu_env *env,
3177                                            struct dt_object *dt,
3178                                            struct lu_attr *attr,
3179                                            struct dt_object_format *dof,
3180                                            struct thandle *th)
3181 {
3182         return lod_dir_striping_create_internal(env, dt, attr, dof, th, true);
3183 }
3184
3185 static int lod_dir_striping_create(const struct lu_env *env,
3186                                    struct dt_object *dt,
3187                                    struct lu_attr *attr,
3188                                    struct dt_object_format *dof,
3189                                    struct thandle *th)
3190 {
3191         return lod_dir_striping_create_internal(env, dt, attr, dof, th, false);
3192 }
3193
3194 /**
3195  * Make LOV EA for striped object.
3196  *
3197  * Generate striping information and store it in the LOV EA of the given
3198  * object. The caller must ensure nobody else is calling the function
3199  * against the object concurrently. The transaction must be started.
3200  * FLDB service must be running as well; it's used to map FID to the target,
3201  * which is stored in LOV EA.
3202  *
3203  * \param[in] env               execution environment for this thread
3204  * \param[in] lo                LOD object
3205  * \param[in] th                transaction handle
3206  *
3207  * \retval                      0 if LOV EA is stored successfully
3208  * \retval                      negative error number on failure
3209  */
3210 static int lod_generate_and_set_lovea(const struct lu_env *env,
3211                                       struct lod_object *lo,
3212                                       struct thandle *th)
3213 {
3214         struct lod_thread_info  *info = lod_env_info(env);
3215         struct dt_object        *next = dt_object_child(&lo->ldo_obj);
3216         struct lov_mds_md_v1    *lmm;
3217         int                      rc, lmm_size;
3218         ENTRY;
3219
3220         LASSERT(lo);
3221
3222         if (lo->ldo_comp_cnt == 0) {
3223                 lod_object_free_striping(env, lo);
3224                 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
3225                 RETURN(rc);
3226         }
3227
3228         lmm_size = lod_comp_md_size(lo, false);
3229         if (info->lti_ea_store_size < lmm_size) {
3230                 rc = lod_ea_store_resize(info, lmm_size);
3231                 if (rc)
3232                         RETURN(rc);
3233         }
3234         lmm = info->lti_ea_store;
3235
3236         rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
3237         if (rc)
3238                 RETURN(rc);
3239
3240         info->lti_buf.lb_buf = lmm;
3241         info->lti_buf.lb_len = lmm_size;
3242         rc = lod_sub_xattr_set(env, next, &info->lti_buf,
3243                                XATTR_NAME_LOV, 0, th);
3244         RETURN(rc);
3245 }
3246
3247 /**
3248  * Delete layout component(s)
3249  *
3250  * \param[in] env       execution environment for this thread
3251  * \param[in] dt        object
3252  * \param[in] th        transaction handle
3253  *
3254  * \retval      0 on success
3255  * \retval      negative error number on failure
3256  */
3257 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
3258                           struct thandle *th)
3259 {
3260         struct lod_layout_component     *lod_comp;
3261         struct lod_object       *lo = lod_dt_obj(dt);
3262         struct dt_object        *next = dt_object_child(dt);
3263         struct lu_attr  *attr = &lod_env_info(env)->lti_attr;
3264         int     rc, i, j, left;
3265
3266         LASSERT(lo->ldo_is_composite);
3267         LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
3268
3269         left = lo->ldo_comp_cnt;
3270         for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3271                 lod_comp = &lo->ldo_comp_entries[i];
3272
3273                 if (lod_comp->llc_id != LCME_ID_INVAL)
3274                         break;
3275                 left--;
3276
3277                 /* Not instantiated component */
3278                 if (lod_comp->llc_stripe == NULL)
3279                         continue;
3280
3281                 LASSERT(lod_comp->llc_stripe_count > 0);
3282                 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
3283                         struct dt_object *obj = lod_comp->llc_stripe[j];
3284
3285                         if (obj == NULL)
3286                                 continue;
3287                         rc = lod_sub_destroy(env, obj, th);
3288                         if (rc)
3289                                 GOTO(out, rc);
3290
3291                         lu_object_put(env, &obj->do_lu);
3292                         lod_comp->llc_stripe[j] = NULL;
3293                 }
3294                 OBD_FREE(lod_comp->llc_stripe, sizeof(struct dt_object *) *
3295                                         lod_comp->llc_stripes_allocated);
3296                 lod_comp->llc_stripe = NULL;
3297                 lod_comp->llc_stripes_allocated = 0;
3298                 lod_obj_set_pool(lo, i, NULL);
3299                 if (lod_comp->llc_ostlist.op_array) {
3300                         OBD_FREE(lod_comp->llc_ostlist.op_array,
3301                                  lod_comp->llc_ostlist.op_size);
3302                         lod_comp->llc_ostlist.op_array = NULL;
3303                         lod_comp->llc_ostlist.op_size = 0;
3304                 }
3305         }
3306
3307         LASSERTF(left >= 0 && left < lo->ldo_comp_cnt, "left = %d\n", left);
3308         if (left > 0) {
3309                 struct lod_layout_component     *comp_array;
3310
3311                 OBD_ALLOC(comp_array, sizeof(*comp_array) * left);
3312                 if (comp_array == NULL)
3313                         GOTO(out, rc = -ENOMEM);
3314
3315                 memcpy(&comp_array[0], &lo->ldo_comp_entries[0],
3316                        sizeof(*comp_array) * left);
3317
3318                 OBD_FREE(lo->ldo_comp_entries,
3319                          sizeof(*comp_array) * lo->ldo_comp_cnt);
3320                 lo->ldo_comp_entries = comp_array;
3321                 lo->ldo_comp_cnt = left;
3322                 lod_obj_inc_layout_gen(lo);
3323         } else {
3324                 lod_free_comp_entries(lo);
3325         }
3326
3327         LASSERT(dt_object_exists(dt));
3328         rc = dt_attr_get(env, next, attr);
3329         if (rc)
3330                 GOTO(out, rc);
3331
3332         if (attr->la_size > 0) {
3333                 attr->la_size = 0;
3334                 attr->la_valid = LA_SIZE;
3335                 rc = lod_sub_attr_set(env, next, attr, th);
3336                 if (rc)
3337                         GOTO(out, rc);
3338         }
3339
3340         rc = lod_generate_and_set_lovea(env, lo, th);
3341         EXIT;
3342 out:
3343         if (rc)
3344                 lod_object_free_striping(env, lo);
3345         return rc;
3346 }
3347
3348 /**
3349  * Implementation of dt_object_operations::do_xattr_set.
3350  *
3351  * Sets specified extended attribute on the object. Three types of EAs are
3352  * special:
3353  *   LOV EA - stores striping for a regular file or default striping (when set
3354  *            on a directory)
3355  *   LMV EA - stores a marker for the striped directories
3356  *   DMV EA - stores default directory striping
3357  *
3358  * When striping is applied to a non-striped existing object (this is called
3359  * late striping), then LOD notices the caller wants to turn the object into a
3360  * striped one. The stripe objects are created and appropriate EA is set:
3361  * LOV EA storing all the stripes directly or LMV EA storing just a small header
3362  * with striping configuration.
3363  *
3364  * \see dt_object_operations::do_xattr_set() in the API description for details.
3365  */
3366 static int lod_xattr_set(const struct lu_env *env,
3367                          struct dt_object *dt, const struct lu_buf *buf,
3368                          const char *name, int fl, struct thandle *th)
3369 {
3370         struct dt_object        *next = dt_object_child(dt);
3371         int                      rc;
3372         ENTRY;
3373
3374         if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
3375             strcmp(name, XATTR_NAME_LMV) == 0) {
3376                 struct lmv_mds_md_v1 *lmm = buf->lb_buf;
3377
3378                 if (lmm != NULL && le32_to_cpu(lmm->lmv_hash_type) &
3379                                                 LMV_HASH_FLAG_MIGRATION)
3380                         rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3381                 else
3382                         rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
3383
3384                 RETURN(rc);
3385         }
3386
3387         if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
3388             strcmp(name, XATTR_NAME_LOV) == 0) {
3389                 /* default LOVEA */
3390                 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl, th);
3391                 RETURN(rc);
3392         } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
3393                    strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
3394                 /* default LMVEA */
3395                 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
3396                                                       th);
3397                 RETURN(rc);
3398         } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
3399                    (!strcmp(name, XATTR_NAME_LOV) ||
3400                     !strncmp(name, XATTR_LUSTRE_LOV,
3401                              strlen(XATTR_LUSTRE_LOV)))) {
3402                 /* in case of lov EA swap, just set it
3403                  * if not, it is a replay so check striping match what we
3404                  * already have during req replay, declare_xattr_set()
3405                  * defines striping, then create() does the work */
3406                 if (fl & LU_XATTR_REPLACE) {
3407                         /* free stripes, then update disk */
3408                         lod_object_free_striping(env, lod_dt_obj(dt));
3409
3410                         rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3411                 } else if (dt_object_remote(dt)) {
3412                         /* This only happens during migration, see
3413                          * mdd_migrate_create(), in which Master MDT will
3414                          * create a remote target object, and only set
3415                          * (migrating) stripe EA on the remote object,
3416                          * and does not need creating each stripes. */
3417                         rc = lod_sub_xattr_set(env, next, buf, name,
3418                                                       fl, th);
3419                 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
3420                         /* delete component(s) */
3421                         LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
3422                         rc = lod_layout_del(env, dt, th);
3423                 } else {
3424                         /*
3425                          * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
3426                          * it's going to create create file with specified
3427                          * component(s), the striping must have not being
3428                          * cached in this case;
3429                          *
3430                          * Otherwise, it's going to add/change component(s) to
3431                          * an existing file, the striping must have been cached
3432                          * in this case.
3433                          */
3434                         LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
3435                                      !strcmp(name, XATTR_NAME_LOV),
3436                                 !lod_dt_obj(dt)->ldo_comp_cached));
3437
3438                         rc = lod_striped_create(env, dt, NULL, NULL, th);
3439                 }
3440                 RETURN(rc);
3441         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3442                 rc = lod_replace_parent_fid(env, dt, th, false);
3443
3444                 RETURN(rc);
3445         }
3446
3447         /* then all other xattr */
3448         rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3449
3450         RETURN(rc);
3451 }
3452
3453 /**
3454  * Implementation of dt_object_operations::do_declare_xattr_del.
3455  *
3456  * \see dt_object_operations::do_declare_xattr_del() in the API description
3457  * for details.
3458  */
3459 static int lod_declare_xattr_del(const struct lu_env *env,
3460                                  struct dt_object *dt, const char *name,
3461                                  struct thandle *th)
3462 {
3463         struct lod_object       *lo = lod_dt_obj(dt);
3464         int                     rc;
3465         int                     i;
3466         ENTRY;
3467
3468         rc = lod_sub_declare_xattr_del(env, dt_object_child(dt), name, th);
3469         if (rc != 0)
3470                 RETURN(rc);
3471
3472         if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3473                 RETURN(0);
3474
3475         /* set xattr to each stripes, if needed */
3476         rc = lod_load_striping(env, lo);
3477         if (rc != 0)
3478                 RETURN(rc);
3479
3480         if (lo->ldo_dir_stripe_count == 0)
3481                 RETURN(0);
3482
3483         for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3484                 LASSERT(lo->ldo_stripe[i]);
3485                 rc = lod_sub_declare_xattr_del(env, lo->ldo_stripe[i],
3486                                                name, th);
3487                 if (rc != 0)
3488                         break;
3489         }
3490
3491         RETURN(rc);
3492 }
3493
3494 /**
3495  * Implementation of dt_object_operations::do_xattr_del.
3496  *
3497  * If EA storing a regular striping is being deleted, then release
3498  * all the references to the stripe objects in core.
3499  *
3500  * \see dt_object_operations::do_xattr_del() in the API description for details.
3501  */
3502 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
3503                          const char *name, struct thandle *th)
3504 {
3505         struct dt_object        *next = dt_object_child(dt);
3506         struct lod_object       *lo = lod_dt_obj(dt);
3507         int                     rc;
3508         int                     i;
3509         ENTRY;
3510
3511         if (!strcmp(name, XATTR_NAME_LOV))
3512                 lod_object_free_striping(env, lod_dt_obj(dt));
3513
3514         rc = lod_sub_xattr_del(env, next, name, th);
3515         if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3516                 RETURN(rc);
3517
3518         if (lo->ldo_dir_stripe_count == 0)
3519                 RETURN(0);
3520
3521         for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3522                 LASSERT(lo->ldo_stripe[i]);
3523
3524                 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3525                 if (rc != 0)
3526                         break;
3527         }
3528
3529         RETURN(rc);
3530 }
3531
3532 /**
3533  * Implementation of dt_object_operations::do_xattr_list.
3534  *
3535  * \see dt_object_operations::do_xattr_list() in the API description
3536  * for details.
3537  */
3538 static int lod_xattr_list(const struct lu_env *env,
3539                           struct dt_object *dt, const struct lu_buf *buf)
3540 {
3541         return dt_xattr_list(env, dt_object_child(dt), buf);
3542 }
3543
3544 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
3545 {
3546         return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
3547 }
3548
3549
3550 /**
3551  * Get default striping.
3552  *
3553  * \param[in] env               execution environment
3554  * \param[in] lo                object
3555  * \param[out] lds              default striping
3556  *
3557  * \retval              0 on success
3558  * \retval              negative if failed
3559  */
3560 static int lod_get_default_lov_striping(const struct lu_env *env,
3561                                         struct lod_object *lo,
3562                                         struct lod_default_striping *lds)
3563 {
3564         struct lod_thread_info *info = lod_env_info(env);
3565         struct lov_user_md_v1 *v1 = NULL;
3566         struct lov_user_md_v3 *v3 = NULL;
3567         struct lov_comp_md_v1 *comp_v1 = NULL;
3568         __u16   comp_cnt;
3569         bool    composite;
3570         int     rc, i;
3571         ENTRY;
3572
3573         lds->lds_def_striping_set = 0;
3574
3575         rc = lod_get_lov_ea(env, lo);
3576         if (rc < 0)
3577                 RETURN(rc);
3578
3579         if (rc < (typeof(rc))sizeof(struct lov_user_md))
3580                 RETURN(0);
3581
3582         v1 = info->lti_ea_store;
3583         if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
3584                 lustre_swab_lov_user_md_v1(v1);
3585         } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
3586                 v3 = (struct lov_user_md_v3 *)v1;
3587                 lustre_swab_lov_user_md_v3(v3);
3588         } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
3589                 comp_v1 = (struct lov_comp_md_v1 *)v1;
3590                 lustre_swab_lov_comp_md_v1(comp_v1);
3591         }
3592
3593         if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1 &&
3594             v1->lmm_magic != LOV_MAGIC_COMP_V1)
3595                 RETURN(-ENOTSUPP);
3596
3597         if (v1->lmm_magic == LOV_MAGIC_COMP_V1) {
3598                 comp_v1 = (struct lov_comp_md_v1 *)v1;
3599                 comp_cnt = comp_v1->lcm_entry_count;
3600                 if (comp_cnt == 0)
3601                         RETURN(-EINVAL);
3602                 composite = true;
3603         } else {
3604                 comp_cnt = 1;
3605                 composite = false;
3606         }
3607
3608         /* realloc default comp entries if necessary */
3609         rc = lod_def_striping_comp_resize(lds, comp_cnt);
3610         if (rc < 0)
3611                 RETURN(rc);
3612
3613         lds->lds_def_comp_cnt = comp_cnt;
3614         lds->lds_def_striping_is_composite = composite ? 1 : 0;
3615
3616         for (i = 0; i < comp_cnt; i++) {
3617                 struct lod_layout_component *lod_comp;
3618                 struct lu_extent *ext;
3619                 char *pool;
3620
3621                 lod_comp = &lds->lds_def_comp_entries[i];
3622                 /*
3623                  * reset lod_comp values, llc_stripes is always NULL in
3624                  * the default striping template, llc_pool will be reset
3625                  * later below.
3626                  */
3627                 memset(lod_comp, 0, offsetof(typeof(*lod_comp), llc_pool));
3628
3629                 if (composite) {
3630                         v1 = (struct lov_user_md *)((char *)comp_v1 +
3631                                         comp_v1->lcm_entries[i].lcme_offset);
3632                         ext = &comp_v1->lcm_entries[i].lcme_extent;
3633                         lod_comp->llc_extent = *ext;
3634                 }
3635
3636                 if (v1->lmm_pattern != LOV_PATTERN_RAID0 &&
3637                     v1->lmm_pattern != 0) {
3638                         lod_free_def_comp_entries(lds);
3639                         RETURN(-EINVAL);
3640                 }
3641
3642                 CDEBUG(D_LAYOUT, DFID" stripe_count=%d stripe_size=%d "
3643                        "stripe_offset=%d\n",
3644                        PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
3645                        (int)v1->lmm_stripe_count, (int)v1->lmm_stripe_size,
3646                        (int)v1->lmm_stripe_offset);
3647
3648                 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
3649                 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
3650                 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
3651
3652                 pool = NULL;
3653                 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
3654                         /* XXX: sanity check here */
3655                         v3 = (struct lov_user_md_v3 *) v1;
3656                         if (v3->lmm_pool_name[0] != '\0')
3657                                 pool = v3->lmm_pool_name;
3658                 }
3659                 lod_set_def_pool(lds, i, pool);
3660         }
3661
3662         lds->lds_def_striping_set = 1;
3663         RETURN(rc);
3664 }
3665
3666 /**
3667  * Get default directory striping.
3668  *
3669  * \param[in] env               execution environment
3670  * \param[in] lo                object
3671  * \param[out] lds              default striping
3672  *
3673  * \retval              0 on success
3674  * \retval              negative if failed
3675  */
3676 static int lod_get_default_lmv_striping(const struct lu_env *env,
3677                                         struct lod_object *lo,
3678                                         struct lod_default_striping *lds)
3679 {
3680         struct lod_thread_info  *info = lod_env_info(env);
3681         struct lmv_user_md_v1   *v1 = NULL;
3682         int                      rc;
3683         ENTRY;
3684
3685         lds->lds_dir_def_striping_set = 0;
3686         rc = lod_get_default_lmv_ea(env, lo);
3687         if (rc < 0)
3688                 RETURN(rc);
3689
3690         if (rc < (typeof(rc))sizeof(struct lmv_user_md))
3691                 RETURN(0);
3692
3693         v1 = info->lti_ea_store;
3694
3695         lds->lds_dir_def_stripe_count = le32_to_cpu(v1->lum_stripe_count);
3696         lds->lds_dir_def_stripe_offset = le32_to_cpu(v1->lum_stripe_offset);
3697         lds->lds_dir_def_hash_type = le32_to_cpu(v1->lum_hash_type);
3698         lds->lds_dir_def_striping_set = 1;
3699
3700         RETURN(0);
3701 }
3702
3703 /**
3704  * Get default striping in the object.
3705  *
3706  * Get object default striping and default directory striping.
3707  *
3708  * \param[in] env               execution environment
3709  * \param[in] lo                object
3710  * \param[out] lds              default striping
3711  *
3712  * \retval              0 on success
3713  * \retval              negative if failed
3714  */
3715 static int lod_get_default_striping(const struct lu_env *env,
3716                                     struct lod_object *lo,
3717                                     struct lod_default_striping *lds)
3718 {
3719         int rc, rc1;
3720
3721         rc = lod_get_default_lov_striping(env, lo, lds);
3722         rc1 = lod_get_default_lmv_striping(env, lo, lds);
3723         if (rc == 0 && rc1 < 0)
3724                 rc = rc1;
3725
3726         return rc;
3727 }
3728
3729 /**
3730  * Apply default striping on object.
3731  *
3732  * If object striping pattern is not set, set to the one in default striping.
3733  * The default striping is from parent or fs.
3734  *
3735  * \param[in] lo                new object
3736  * \param[in] lds               default striping
3737  * \param[in] mode              new object's mode
3738  */
3739 static void lod_striping_from_default(struct lod_object *lo,
3740                                       const struct lod_default_striping *lds,
3741                                       umode_t mode)
3742 {
3743         struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
3744         struct lov_desc *desc = &d->lod_desc;
3745         int i, rc;
3746
3747         if (lds->lds_def_striping_set && S_ISREG(mode)) {
3748                 rc = lod_alloc_comp_entries(lo, lds->lds_def_comp_cnt);
3749                 if (rc != 0)
3750                         return;
3751
3752                 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
3753
3754                 for (i = 0; i < lo->ldo_comp_cnt; i++) {
3755                         struct lod_layout_component *obj_comp =
3756                                                 &lo->ldo_comp_entries[i];
3757                         struct lod_layout_component *def_comp =
3758                                                 &lds->lds_def_comp_entries[i];
3759
3760                         CDEBUG(D_LAYOUT, "Inherite from default: size:%hu "
3761                                "nr:%u offset:%u %s\n",
3762                                def_comp->llc_stripe_size,
3763                                def_comp->llc_stripe_count,
3764                                def_comp->llc_stripe_offset,
3765                                def_comp->llc_pool ?: "");
3766
3767                         *obj_comp = *def_comp;
3768                         if (def_comp->llc_pool != NULL) {
3769                                 /* pointer was copied from def_comp */
3770                                 obj_comp->llc_pool = NULL;
3771                                 lod_obj_set_pool(lo, i, def_comp->llc_pool);
3772                         }
3773
3774                         /*
3775                          * Don't initialize these fields for plain layout
3776                          * (v1/v3) here, they are inherited in the order of
3777                          * 'parent' -> 'fs default (root)' -> 'global default
3778                          * values for stripe_count & stripe_size'.
3779                          *
3780                          * see lod_ah_init().
3781                          */
3782                         if (!lo->ldo_is_composite)
3783                                 continue;
3784
3785                         if (obj_comp->llc_stripe_count <= 0)
3786                                 obj_comp->llc_stripe_count =
3787                                         desc->ld_default_stripe_count;
3788                         if (obj_comp->llc_stripe_size <= 0)
3789                                 obj_comp->llc_stripe_size =
3790                                         desc->ld_default_stripe_size;
3791                 }
3792         } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
3793                 if (lo->ldo_dir_stripe_count == 0)
3794                         lo->ldo_dir_stripe_count =
3795                                 lds->lds_dir_def_stripe_count;
3796                 if (lo->ldo_dir_stripe_offset == -1)
3797                         lo->ldo_dir_stripe_offset =
3798                                 lds->lds_dir_def_stripe_offset;
3799                 if (lo->ldo_dir_hash_type == 0)
3800                         lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
3801
3802                 CDEBUG(D_LAYOUT, "striping from default dir: count:%hu, "
3803                        "offset:%u, hash_type:%u\n",
3804                        lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
3805                        lo->ldo_dir_hash_type);
3806         }
3807 }
3808
3809 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root)
3810 {
3811         struct lod_layout_component *lod_comp;
3812
3813         if (lo->ldo_comp_cnt == 0)
3814                 return true;
3815
3816         if (lo->ldo_is_composite)
3817                 return false;
3818
3819         lod_comp = &lo->ldo_comp_entries[0];
3820
3821         if (lod_comp->llc_stripe_count <= 0 ||
3822             lod_comp->llc_stripe_size <= 0)
3823                 return true;
3824
3825         if (from_root && (lod_comp->llc_pool == NULL ||
3826                           lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
3827                 return true;
3828
3829         return false;
3830 }
3831
3832 /**
3833  * Implementation of dt_object_operations::do_ah_init.
3834  *
3835  * This method is used to make a decision on the striping configuration for the
3836  * object being created. It can be taken from the \a parent object if it exists,
3837  * or filesystem's default. The resulting configuration (number of stripes,
3838  * stripe size/offset, pool name, etc) is stored in the object itself and will
3839  * be used by the methods like ->doo_declare_create().
3840  *
3841  * \see dt_object_operations::do_ah_init() in the API description for details.
3842  */
3843 static void lod_ah_init(const struct lu_env *env,
3844                         struct dt_allocation_hint *ah,
3845                         struct dt_object *parent,
3846                         struct dt_object *child,
3847                         umode_t child_mode)
3848 {
3849         struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
3850         struct lod_thread_info *info = lod_env_info(env);
3851         struct lod_default_striping *lds = &info->lti_def_striping;
3852         struct dt_object *nextp = NULL;
3853         struct dt_object *nextc;
3854         struct lod_object *lp = NULL;
3855         struct lod_object *lc;
3856         struct lov_desc *desc;
3857         struct lod_layout_component *lod_comp;
3858         int rc;
3859         ENTRY;
3860
3861         LASSERT(child);
3862
3863         if (likely(parent)) {
3864                 nextp = dt_object_child(parent);
3865                 lp = lod_dt_obj(parent);
3866         }
3867
3868         nextc = dt_object_child(child);
3869         lc = lod_dt_obj(child);
3870
3871         LASSERT(!lod_obj_is_striped(child));
3872         /* default layout template may have been set on the regular file
3873          * when this is called from mdd_create_data() */
3874         if (S_ISREG(child_mode))
3875                 lod_free_comp_entries(lc);
3876
3877         if (!dt_object_exists(nextc))
3878                 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
3879
3880         if (S_ISDIR(child_mode)) {
3881                 /* other default values are 0 */
3882                 lc->ldo_dir_stripe_offset = -1;
3883
3884                 /* get default striping from parent object */
3885                 if (likely(lp != NULL))
3886                         lod_get_default_striping(env, lp, lds);
3887
3888                 /* set child default striping info, default value is NULL */
3889                 if (lds->lds_def_striping_set || lds->lds_dir_def_striping_set)
3890                         lc->ldo_def_striping = lds;
3891
3892                 /* It should always honour the specified stripes */
3893                 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
3894                     lod_verify_md_striping(d, ah->dah_eadata) == 0) {
3895                         const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
3896
3897                         lc->ldo_dir_stripe_count =
3898                                 le32_to_cpu(lum1->lum_stripe_count);
3899                         lc->ldo_dir_stripe_offset =
3900                                 le32_to_cpu(lum1->lum_stripe_offset);
3901                         lc->ldo_dir_hash_type =
3902                                 le32_to_cpu(lum1->lum_hash_type);
3903                         CDEBUG(D_INFO,
3904                                "set dirstripe: count %hu, offset %d, hash %u\n",
3905                                 lc->ldo_dir_stripe_count,
3906                                 (int)lc->ldo_dir_stripe_offset,
3907                                 lc->ldo_dir_hash_type);
3908                 } else {
3909                         /* transfer defaults LMV to new directory */
3910                         lod_striping_from_default(lc, lds, child_mode);
3911                 }
3912
3913                 /* shrink the stripe_count to the avaible MDT count */
3914                 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
3915                     !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))
3916                         lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
3917
3918                 /* Directory will be striped only if stripe_count > 1, if
3919                  * stripe_count == 1, let's reset stripe_count = 0 to avoid
3920                  * create single master stripe and also help to unify the
3921                  * stripe handling of directories and files */
3922                 if (lc->ldo_dir_stripe_count == 1)
3923                         lc->ldo_dir_stripe_count = 0;
3924
3925                 CDEBUG(D_INFO, "final dir stripe [%hu %d %u]\n",
3926                        lc->ldo_dir_stripe_count,
3927                        (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
3928
3929                 RETURN_EXIT;
3930         }
3931
3932         /* child object regular file*/
3933
3934         if (!lod_object_will_be_striped(S_ISREG(child_mode),
3935                                         lu_object_fid(&child->do_lu)))
3936                 RETURN_EXIT;
3937
3938         /* If object is going to be striped over OSTs, transfer default
3939          * striping information to the child, so that we can use it
3940          * during declaration and creation.
3941          *
3942          * Try from the parent first.
3943          */
3944         if (likely(lp != NULL)) {
3945                 rc = lod_get_default_lov_striping(env, lp, lds);
3946                 if (rc == 0)
3947                         lod_striping_from_default(lc, lds, child_mode);
3948         }
3949
3950         /* Initialize lod_device::lod_md_root object reference */
3951         if (d->lod_md_root == NULL) {
3952                 struct dt_object *root;
3953                 struct lod_object *lroot;
3954
3955                 lu_root_fid(&info->lti_fid);
3956                 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
3957                 if (!IS_ERR(root)) {
3958                         lroot = lod_dt_obj(root);
3959
3960                         spin_lock(&d->lod_lock);
3961                         if (d->lod_md_root != NULL)
3962                                 dt_object_put(env, &d->lod_md_root->ldo_obj);
3963                         d->lod_md_root = lroot;
3964                         spin_unlock(&d->lod_lock);
3965                 }
3966         }
3967
3968         /* try inherit layout from the root object (fs default) when:
3969          *  - parent does not have default layout; or
3970          *  - parent has plain(v1/v3) default layout, and some attributes
3971          *    are not specified in the default layout;
3972          */
3973         if (d->lod_md_root != NULL && lod_need_inherit_more(lc, true)) {
3974                 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds);
3975                 if (rc)
3976                         goto out;
3977                 if (lc->ldo_comp_cnt == 0) {
3978                         lod_striping_from_default(lc, lds, child_mode);
3979                 } else if (!lds->lds_def_striping_is_composite) {
3980                         struct lod_layout_component *def_comp;
3981
3982                         LASSERT(!lc->ldo_is_composite);
3983                         lod_comp = &lc->ldo_comp_entries[0];
3984                         def_comp = &lds->lds_def_comp_entries[0];
3985
3986                         if (lod_comp->llc_stripe_count <= 0)
3987                                 lod_comp->llc_stripe_count =
3988                                         def_comp->llc_stripe_count;
3989                         if (lod_comp->llc_stripe_size <= 0)
3990                                 lod_comp->llc_stripe_size =
3991                                         def_comp->llc_stripe_size;
3992                         if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT)
3993                                 lod_comp->llc_stripe_offset =
3994                                         def_comp->llc_stripe_offset;
3995                         if (lod_comp->llc_pool == NULL)
3996                                 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
3997                 }
3998         }
3999 out:
4000         /*
4001          * fs default striping may not be explicitly set, or historically set
4002          * in config log, use them.
4003          */
4004         if (lod_need_inherit_more(lc, false)) {
4005
4006                 if (lc->ldo_comp_cnt == 0) {
4007                         rc = lod_alloc_comp_entries(lc, 1);
4008                         if (rc)
4009                                 /* fail to allocate memory, will create a
4010                                  * non-striped file. */
4011                                 RETURN_EXIT;
4012                         lc->ldo_is_composite = 0;
4013                         lod_comp = &lc->ldo_comp_entries[0];
4014                         lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4015                 }
4016                 LASSERT(!lc->ldo_is_composite);
4017                 lod_comp = &lc->ldo_comp_entries[0];
4018                 desc = &d->lod_desc;
4019                 if (lod_comp->llc_stripe_count <= 0)
4020                         lod_comp->llc_stripe_count =
4021                                 desc->ld_default_stripe_count;
4022                 if (lod_comp->llc_stripe_size <= 0)
4023                         lod_comp->llc_stripe_size =
4024                                 desc->ld_default_stripe_size;
4025         }
4026
4027         EXIT;
4028 }
4029
4030 #define ll_do_div64(aaa,bbb)    do_div((aaa), (bbb))
4031 /**
4032  * Size initialization on late striping.
4033  *
4034  * Propagate the size of a truncated object to a deferred striping.
4035  * This function handles a special case when truncate was done on a
4036  * non-striped object and now while the striping is being created
4037  * we can't lose that size, so we have to propagate it to the stripes
4038  * being created.
4039  *
4040  * \param[in] env       execution environment
4041  * \param[in] dt        object
4042  * \param[in] th        transaction handle
4043  *
4044  * \retval              0 on success
4045  * \retval              negative if failed
4046  */
4047 static int lod_declare_init_size(const struct lu_env *env,
4048                                  struct dt_object *dt, struct thandle *th)
4049 {
4050         struct dt_object        *next = dt_object_child(dt);
4051         struct lod_object       *lo = lod_dt_obj(dt);
4052         struct dt_object        **objects = NULL;
4053         struct lu_attr  *attr = &lod_env_info(env)->lti_attr;
4054         uint64_t        size, offs;
4055         int     i, rc, stripe, stripe_count = 0, stripe_size = 0;
4056         ENTRY;
4057
4058         if (!lod_obj_is_striped(dt))
4059                 RETURN(0);
4060
4061         rc = dt_attr_get(env, next, attr);
4062         LASSERT(attr->la_valid & LA_SIZE);
4063         if (rc)
4064                 RETURN(rc);
4065
4066         size = attr->la_size;
4067         if (size == 0)
4068                 RETURN(0);
4069
4070         for (i = 0; i < lo->ldo_comp_cnt; i++) {
4071                 struct lod_layout_component *lod_comp;
4072                 struct lu_extent *extent;
4073
4074                 lod_comp = &lo->ldo_comp_entries[i];
4075
4076                 if (lod_comp->llc_stripe == NULL)
4077                         continue;
4078
4079                 extent = &lod_comp->llc_extent;
4080                 CDEBUG(D_INFO, "%lld [%lld, %lld)\n",
4081                        size, extent->e_start, extent->e_end);
4082                 if (!lo->ldo_is_composite ||
4083                     (size >= extent->e_start && size < extent->e_end)) {
4084                         objects = lod_comp->llc_stripe;
4085                         stripe_count = lod_comp->llc_stripe_count;
4086                         stripe_size = lod_comp->llc_stripe_size;
4087                         break;
4088                 }
4089         }
4090
4091         if (stripe_count == 0)
4092                 RETURN(0);
4093
4094         LASSERT(objects != NULL && stripe_size != 0);
4095
4096         /* ll_do_div64(a, b) returns a % b, and a = a / b */
4097         ll_do_div64(size, (__u64)stripe_size);
4098         stripe = ll_do_div64(size, (__u64)stripe_count);
4099         LASSERT(objects[stripe] != NULL);
4100
4101         size = size * stripe_size;
4102         offs = attr->la_size;
4103         size += ll_do_div64(offs, stripe_size);
4104
4105         attr->la_valid = LA_SIZE;
4106         attr->la_size = size;
4107
4108         rc = lod_sub_declare_attr_set(env, objects[stripe], attr, th);
4109
4110         RETURN(rc);
4111 }
4112
4113 /**
4114  * Declare creation of striped object.
4115  *
4116  * The function declares creation stripes for a regular object. The function
4117  * also declares whether the stripes will be created with non-zero size if
4118  * previously size was set non-zero on the master object. If object \a dt is
4119  * not local, then only fully defined striping can be applied in \a lovea.
4120  * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
4121  * for the details.
4122  *
4123  * \param[in] env       execution environment
4124  * \param[in] dt        object
4125  * \param[in] attr      attributes the stripes will be created with
4126  * \param[in] lovea     a buffer containing striping description
4127  * \param[in] th        transaction handle
4128  *
4129  * \retval              0 on success
4130  * \retval              negative if failed
4131  */
4132 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
4133                                struct lu_attr *attr,
4134                                const struct lu_buf *lovea, struct thandle *th)
4135 {
4136         struct lod_thread_info  *info = lod_env_info(env);
4137         struct dt_object        *next = dt_object_child(dt);
4138         struct lod_object       *lo = lod_dt_obj(dt);
4139         int                      rc;
4140         ENTRY;
4141
4142         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
4143                 GOTO(out, rc = -ENOMEM);
4144
4145         if (!dt_object_remote(next)) {
4146                 /* choose OST and generate appropriate objects */
4147                 rc = lod_prepare_create(env, lo, attr, lovea, th);
4148                 if (rc)
4149                         GOTO(out, rc);
4150
4151                 /*
4152                  * declare storage for striping data
4153                  */
4154                 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
4155         } else {
4156                 /* LOD can not choose OST objects for remote objects, i.e.
4157                  * stripes must be ready before that. Right now, it can only
4158                  * happen during migrate, i.e. migrate process needs to create
4159                  * remote regular file (mdd_migrate_create), then the migrate
4160                  * process will provide stripeEA. */
4161                 LASSERT(lovea != NULL);
4162                 info->lti_buf = *lovea;
4163         }
4164
4165         rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
4166                                        XATTR_NAME_LOV, 0, th);
4167         if (rc)
4168                 GOTO(out, rc);
4169
4170         /*
4171          * if striping is created with local object's size > 0,
4172          * we have to propagate this size to specific object
4173          * the case is possible only when local object was created previously
4174          */
4175         if (dt_object_exists(next))
4176                 rc = lod_declare_init_size(env, dt, th);
4177
4178 out:
4179         /* failed to create striping or to set initial size, let's reset
4180          * config so that others don't get confused */
4181         if (rc)
4182                 lod_object_free_striping(env, lo);
4183
4184         RETURN(rc);
4185 }
4186
4187 /**
4188  * Implementation of dt_object_operations::do_declare_create.
4189  *
4190  * The method declares creation of a new object. If the object will be striped,
4191  * then helper functions are called to find FIDs for the stripes, declare
4192  * creation of the stripes and declare initialization of the striping
4193  * information to be stored in the master object.
4194  *
4195  * \see dt_object_operations::do_declare_create() in the API description
4196  * for details.
4197  */
4198 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
4199                               struct lu_attr *attr,
4200                               struct dt_allocation_hint *hint,
4201                               struct dt_object_format *dof, struct thandle *th)
4202 {
4203         struct dt_object   *next = dt_object_child(dt);
4204         struct lod_object  *lo = lod_dt_obj(dt);
4205         int                 rc;
4206         ENTRY;
4207
4208         LASSERT(dof);
4209         LASSERT(attr);
4210         LASSERT(th);
4211
4212         /*
4213          * first of all, we declare creation of local object
4214          */
4215         rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
4216         if (rc != 0)
4217                 GOTO(out, rc);
4218
4219         /*
4220          * it's lod_ah_init() that has decided the object will be striped
4221          */
4222         if (dof->dof_type == DFT_REGULAR) {
4223                 /* callers don't want stripes */
4224                 /* XXX: all tricky interactions with ->ah_make_hint() decided
4225                  * to use striping, then ->declare_create() behaving differently
4226                  * should be cleaned */
4227                 if (dof->u.dof_reg.striped != 0)
4228                         rc = lod_declare_striped_create(env, dt, attr,
4229                                                         NULL, th);
4230         } else if (dof->dof_type == DFT_DIR) {
4231                 struct seq_server_site *ss;
4232
4233                 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
4234
4235                 /* If the parent has default stripeEA, and client
4236                  * did not find it before sending create request,
4237                  * then MDT will return -EREMOTE, and client will
4238                  * retrieve the default stripeEA and re-create the
4239                  * sub directory.
4240                  *
4241                  * Note: if dah_eadata != NULL, it means creating the
4242                  * striped directory with specified stripeEA, then it
4243                  * should ignore the default stripeEA */
4244                 if (hint != NULL && hint->dah_eadata == NULL) {
4245                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
4246                                 GOTO(out, rc = -EREMOTE);
4247
4248                         if (lo->ldo_dir_stripe_offset == -1) {
4249                                 /* child and parent should be in the same MDT */
4250                                 if (hint->dah_parent != NULL &&
4251                                     dt_object_remote(hint->dah_parent))
4252                                         GOTO(out, rc = -EREMOTE);
4253                         } else if (lo->ldo_dir_stripe_offset !=
4254                                    ss->ss_node_id) {
4255                                 struct lod_device *lod;
4256                                 struct lod_tgt_descs *ltd;
4257                                 struct lod_tgt_desc *tgt = NULL;
4258                                 bool found_mdt = false;
4259                                 int i;
4260
4261                                 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
4262                                 ltd = &lod->lod_mdt_descs;
4263                                 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
4264                                         tgt = LTD_TGT(ltd, i);
4265                                         if (tgt->ltd_index ==
4266                                                 lo->ldo_dir_stripe_offset) {
4267                                                 found_mdt = true;
4268                                                 break;
4269                                         }
4270                                 }
4271
4272                                 /* If the MDT indicated by stripe_offset can be
4273                                  * found, then tell client to resend the create
4274                                  * request to the correct MDT, otherwise return
4275                                  * error to client */
4276                                 if (found_mdt)
4277                                         GOTO(out, rc = -EREMOTE);
4278                                 else
4279                                         GOTO(out, rc = -EINVAL);
4280                         }
4281                 }
4282
4283                 rc = lod_declare_dir_striping_create(env, dt, attr, dof, th);
4284         }
4285 out:
4286         /* failed to create striping or to set initial size, let's reset
4287          * config so that others don't get confused */
4288         if (rc)
4289                 lod_object_free_striping(env, lo);
4290         RETURN(rc);
4291 }
4292
4293 /**
4294  * Creation of a striped regular object.
4295  *
4296  * The function is called to create the stripe objects for a regular
4297  * striped file. This can happen at the initial object creation or
4298  * when the caller asks LOD to do so using ->do_xattr_set() method
4299  * (so called late striping). Notice all the information are already
4300  * prepared in the form of the list of objects (ldo_stripe field).
4301  * This is done during declare phase.
4302  *
4303  * \param[in] env       execution environment
4304  * \param[in] dt        object
4305  * \param[in] attr      attributes the stripes will be created with
4306  * \param[in] dof       format of stripes (see OSD API description)
4307  * \param[in] th        transaction handle
4308  *
4309  * \retval              0 on success
4310  * \retval              negative if failed
4311  */
4312 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
4313                        struct lu_attr *attr, struct dt_object_format *dof,
4314                        struct thandle *th)
4315 {
4316         struct lod_layout_component     *lod_comp;
4317         struct lod_object       *lo = lod_dt_obj(dt);
4318         int     rc = 0, i, j;
4319         ENTRY;
4320
4321         LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
4322
4323         /* create all underlying objects */
4324         for (i = 0; i < lo->ldo_comp_cnt; i++) {
4325                 lod_comp = &lo->ldo_comp_entries[i];
4326
4327                 if (lod_comp_inited(lod_comp))
4328                         continue;
4329
4330                 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
4331                         lod_comp_set_init(lod_comp);
4332
4333                 if (lod_comp->llc_stripe == NULL)
4334                         continue;
4335
4336                 LASSERT(lod_comp->llc_stripe_count);
4337                 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4338                         struct dt_object *object = lod_comp->llc_stripe[j];
4339                         LASSERT(object != NULL);
4340                         rc = lod_sub_create(env, object, attr, NULL, dof, th);
4341                         if (rc)
4342                                 break;
4343                 }
4344                 lod_comp_set_init(lod_comp);
4345         }
4346
4347         if (rc == 0)
4348                 rc = lod_generate_and_set_lovea(env, lo, th);
4349
4350         if (rc == 0)
4351                 lo->ldo_comp_cached = 1;
4352         else
4353                 lod_object_free_striping(env, lo);
4354
4355         RETURN(rc);
4356 }
4357
4358 /**
4359  * Implementation of dt_object_operations::do_create.
4360  *
4361  * If any of preceeding methods (like ->do_declare_create(),
4362  * ->do_ah_init(), etc) chose to create a striped object,
4363  * then this method will create the master and the stripes.
4364  *
4365  * \see dt_object_operations::do_create() in the API description for details.
4366  */
4367 static int lod_create(const struct lu_env *env, struct dt_object *dt,
4368                       struct lu_attr *attr, struct dt_allocation_hint *hint,
4369                       struct dt_object_format *dof, struct thandle *th)
4370 {
4371         int                 rc;
4372         ENTRY;
4373
4374         /* create local object */
4375         rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
4376         if (rc != 0)
4377                 RETURN(rc);
4378
4379         if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
4380             lod_obj_is_striped(dt) && dof->u.dof_reg.striped != 0) {
4381                 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
4382                 rc = lod_striped_create(env, dt, attr, dof, th);
4383         }
4384
4385         RETURN(rc);
4386 }
4387
4388 static inline int
4389 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
4390                           struct dt_object *dt, struct thandle *th,
4391                           int stripe_idx, struct lod_obj_stripe_cb_data *data)
4392 {
4393         if (data->locd_declare)
4394                 return lod_sub_declare_destroy(env, dt, th);
4395         else if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
4396                  stripe_idx == cfs_fail_val)
4397                 return lod_sub_destroy(env, dt, th);
4398         else
4399                 return 0;
4400 }
4401
4402 /**
4403  * Implementation of dt_object_operations::do_declare_destroy.
4404  *
4405  * If the object is a striped directory, then the function declares reference
4406  * removal from the master object (this is an index) to the stripes and declares
4407  * destroy of all the stripes. In all the cases, it declares an intention to
4408  * destroy the object itself.
4409  *
4410  * \see dt_object_operations::do_declare_destroy() in the API description
4411  * for details.
4412  */
4413 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
4414                                struct thandle *th)
4415 {
4416         struct dt_object   *next = dt_object_child(dt);
4417         struct lod_object  *lo = lod_dt_obj(dt);
4418         struct lod_thread_info *info = lod_env_info(env);
4419         char               *stripe_name = info->lti_key;
4420         int                 rc, i;
4421         ENTRY;
4422
4423         /*
4424          * load striping information, notice we don't do this when object
4425          * is being initialized as we don't need this information till
4426          * few specific cases like destroy, chown
4427          */
4428         rc = lod_load_striping(env, lo);
4429         if (rc)
4430                 RETURN(rc);
4431
4432         /* declare destroy for all underlying objects */
4433         if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4434                 rc = next->do_ops->do_index_try(env, next,
4435                                                 &dt_directory_features);
4436                 if (rc != 0)
4437                         RETURN(rc);
4438
4439                 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4440                         rc = lod_sub_declare_ref_del(env, next, th);
4441                         if (rc != 0)
4442                                 RETURN(rc);
4443
4444                         snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4445                                 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
4446                                 i);
4447                         rc = lod_sub_declare_delete(env, next,
4448                                         (const struct dt_key *)stripe_name, th);
4449                         if (rc != 0)
4450                                 RETURN(rc);
4451                 }
4452         }
4453
4454         /*
4455          * we declare destroy for the local object
4456          */
4457         rc = lod_sub_declare_destroy(env, next, th);
4458         if (rc)
4459                 RETURN(rc);
4460
4461         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
4462             OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
4463                 RETURN(0);
4464
4465         if (!lod_obj_is_striped(dt))
4466                 RETURN(0);
4467
4468         /* declare destroy all striped objects */
4469         if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4470                 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4471                         if (lo->ldo_stripe[i] == NULL)
4472                                 continue;
4473
4474                         rc = lod_sub_declare_ref_del(env, lo->ldo_stripe[i],
4475                                                      th);
4476
4477                         rc = lod_sub_declare_destroy(env, lo->ldo_stripe[i],
4478                                                      th);
4479                         if (rc != 0)
4480                                 break;
4481                 }
4482         } else {
4483                 struct lod_obj_stripe_cb_data data;
4484
4485                 data.locd_declare = true;
4486                 rc = lod_obj_for_each_stripe(env, lo, th,
4487                                 lod_obj_stripe_destroy_cb, &data);
4488         }
4489
4490         RETURN(rc);
4491 }
4492
4493 /**
4494  * Implementation of dt_object_operations::do_destroy.
4495  *
4496  * If the object is a striped directory, then the function removes references
4497  * from the master object (this is an index) to the stripes and destroys all
4498  * the stripes. In all the cases, the function destroys the object itself.
4499  *
4500  * \see dt_object_operations::do_destroy() in the API description for details.
4501  */
4502 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
4503                        struct thandle *th)
4504 {
4505         struct dt_object  *next = dt_object_child(dt);
4506         struct lod_object *lo = lod_dt_obj(dt);
4507         struct lod_thread_info *info = lod_env_info(env);
4508         char               *stripe_name = info->lti_key;
4509         unsigned int       i;
4510         int                rc;
4511         ENTRY;
4512
4513         /* destroy sub-stripe of master object */
4514         if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4515                 rc = next->do_ops->do_index_try(env, next,
4516                                                 &dt_directory_features);
4517                 if (rc != 0)
4518                         RETURN(rc);
4519
4520                 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4521                         rc = lod_sub_ref_del(env, next, th);
4522                         if (rc != 0)
4523                                 RETURN(rc);
4524
4525                         snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4526                                 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
4527                                 i);
4528
4529                         CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
4530                                PFID(lu_object_fid(&dt->do_lu)), stripe_name,
4531                                PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)));
4532
4533                         rc = lod_sub_delete(env, next,
4534                                        (const struct dt_key *)stripe_name, th);
4535                         if (rc != 0)
4536                                 RETURN(rc);
4537                 }
4538         }
4539
4540         rc = lod_sub_destroy(env, next, th);
4541         if (rc != 0)
4542                 RETURN(rc);
4543
4544         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
4545             OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
4546                 RETURN(0);
4547
4548         if (!lod_obj_is_striped(dt))
4549                 RETURN(0);
4550
4551         /* destroy all striped objects */
4552         if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4553                 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4554                         if (lo->ldo_stripe[i] == NULL)
4555                                 continue;
4556                         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
4557                             i == cfs_fail_val) {
4558                                 dt_write_lock(env, lo->ldo_stripe[i],
4559                                               MOR_TGT_CHILD);
4560                                 rc = lod_sub_ref_del(env, lo->ldo_stripe[i],
4561                                                      th);
4562                                 dt_write_unlock(env, lo->ldo_stripe[i]);
4563                                 if (rc != 0)
4564                                         break;
4565
4566                                 rc = lod_sub_destroy(env, lo->ldo_stripe[i],
4567                                                      th);
4568                                 if (rc != 0)
4569                                         break;
4570                         }
4571                 }
4572         } else {
4573                 struct lod_obj_stripe_cb_data data;
4574
4575                 data.locd_declare = false;
4576                 rc = lod_obj_for_each_stripe(env, lo, th,
4577                                 lod_obj_stripe_destroy_cb, &data);
4578         }
4579
4580         RETURN(rc);
4581 }
4582
4583 /**
4584  * Implementation of dt_object_operations::do_declare_ref_add.
4585  *
4586  * \see dt_object_operations::do_declare_ref_add() in the API description
4587  * for details.
4588  */
4589 static int lod_declare_ref_add(const struct lu_env *env,
4590                                struct dt_object *dt, struct thandle *th)
4591 {
4592         return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
4593 }
4594
4595 /**
4596  * Implementation of dt_object_operations::do_ref_add.
4597  *
4598  * \see dt_object_operations::do_ref_add() in the API description for details.
4599  */
4600 static int lod_ref_add(const struct lu_env *env,
4601                        struct dt_object *dt, struct thandle *th)
4602 {
4603         return lod_sub_ref_add(env, dt_object_child(dt), th);
4604 }
4605
4606 /**
4607  * Implementation of dt_object_operations::do_declare_ref_del.
4608  *
4609  * \see dt_object_operations::do_declare_ref_del() in the API description
4610  * for details.
4611  */
4612 static int lod_declare_ref_del(const struct lu_env *env,
4613                                struct dt_object *dt, struct thandle *th)
4614 {
4615         return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
4616 }
4617
4618 /**
4619  * Implementation of dt_object_operations::do_ref_del
4620  *
4621  * \see dt_object_operations::do_ref_del() in the API description for details.
4622  */
4623 static int lod_ref_del(const struct lu_env *env,
4624                        struct dt_object *dt, struct thandle *th)
4625 {
4626         return lod_sub_ref_del(env, dt_object_child(dt), th);
4627 }
4628
4629 /**
4630  * Implementation of dt_object_operations::do_object_sync.
4631  *
4632  * \see dt_object_operations::do_object_sync() in the API description
4633  * for details.
4634  */
4635 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
4636                            __u64 start, __u64 end)
4637 {
4638         return dt_object_sync(env, dt_object_child(dt), start, end);
4639 }
4640
4641 /**
4642  * Release LDLM locks on the stripes of a striped directory.
4643  *
4644  * Iterates over all the locks taken on the stripe objects and
4645  * cancel them.
4646  *
4647  * \param[in] env       execution environment
4648  * \param[in] dt        striped object
4649  * \param[in] einfo     lock description
4650  * \param[in] policy    data describing requested lock
4651  *
4652  * \retval              0 on success
4653  * \retval              negative if failed
4654  */
4655 static int lod_object_unlock_internal(const struct lu_env *env,
4656                                       struct dt_object *dt,
4657                                       struct ldlm_enqueue_info *einfo,
4658                                       union ldlm_policy_data *policy)
4659 {
4660         struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
4661         int                     rc = 0;
4662         int                     i;
4663         ENTRY;
4664
4665         if (slave_locks == NULL)
4666                 RETURN(0);
4667
4668         for (i = 1; i < slave_locks->count; i++) {
4669                 if (lustre_handle_is_used(&slave_locks->handles[i]))
4670                         ldlm_lock_decref_and_cancel(&slave_locks->handles[i],
4671                                                     einfo->ei_mode);
4672         }
4673
4674         RETURN(rc);
4675 }
4676
4677 /**
4678  * Implementation of dt_object_operations::do_object_unlock.
4679  *
4680  * Used to release LDLM lock(s).
4681  *
4682  * \see dt_object_operations::do_object_unlock() in the API description
4683  * for details.
4684  */
4685 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
4686                              struct ldlm_enqueue_info *einfo,
4687                              union ldlm_policy_data *policy)
4688 {
4689         struct lod_object *lo = lod_dt_obj(dt);
4690         struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
4691         int slave_locks_size;
4692         int i;
4693         ENTRY;
4694
4695         if (slave_locks == NULL)
4696                 RETURN(0);
4697
4698         LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
4699         LASSERT(lo->ldo_dir_stripe_count > 1);
4700         /* Note: for remote lock for single stripe dir, MDT will cancel
4701          * the lock by lockh directly */
4702         LASSERT(!dt_object_remote(dt_object_child(dt)));
4703
4704         /* locks were unlocked in MDT layer */
4705         for (i = 1; i < slave_locks->count; i++) {
4706                 LASSERT(!lustre_handle_is_used(&slave_locks->handles[i]));
4707                 dt_invalidate(env, lo->ldo_stripe[i]);
4708         }
4709
4710         slave_locks_size = sizeof(*slave_locks) + slave_locks->count *
4711                            sizeof(slave_locks->handles[0]);
4712         OBD_FREE(slave_locks, slave_locks_size);
4713         einfo->ei_cbdata = NULL;
4714
4715         RETURN(0);
4716 }
4717
4718 /**
4719  * Implementation of dt_object_operations::do_object_lock.
4720  *
4721  * Used to get LDLM lock on the non-striped and striped objects.
4722  *
4723  * \see dt_object_operations::do_object_lock() in the API description
4724  * for details.
4725  */
4726 static int lod_object_lock(const struct lu_env *env,
4727                            struct dt_object *dt,
4728                            struct lustre_handle *lh,
4729                            struct ldlm_enqueue_info *einfo,
4730                            union ldlm_policy_data *policy)
4731 {
4732         struct lod_object       *lo = lod_dt_obj(dt);
4733         int                     rc = 0;
4734         int                     i;
4735         int                     slave_locks_size;
4736         struct lustre_handle_array *slave_locks = NULL;
4737         ENTRY;
4738
4739         /* remote object lock */
4740         if (!einfo->ei_enq_slave) {
4741                 LASSERT(dt_object_remote(dt));
4742                 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
4743                                       policy);
4744         }
4745
4746         if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4747                 GOTO(out, rc = -ENOTDIR);
4748
4749         rc = lod_load_striping(env, lo);
4750         if (rc != 0)
4751                 GOTO(out, rc);
4752
4753         /* No stripes */
4754         if (lo->ldo_dir_stripe_count <= 1) {
4755                 /*
4756                  * NB, ei_cbdata stores pointer to slave locks, if no locks
4757                  * taken, make sure it's set to NULL, otherwise MDT will try to
4758                  * unlock them.
4759                  */
4760                 einfo->ei_cbdata = NULL;
4761                 GOTO(out, rc = 0);
4762         }
4763
4764         slave_locks_size = sizeof(*slave_locks) + lo->ldo_dir_stripe_count *
4765                            sizeof(slave_locks->handles[0]);
4766         /* Freed in lod_object_unlock */
4767         OBD_ALLOC(slave_locks, slave_locks_size);
4768         if (slave_locks == NULL)
4769                 GOTO(out, rc = -ENOMEM);
4770         slave_locks->count = lo->ldo_dir_stripe_count;
4771
4772         /* striped directory lock */
4773         for (i = 1; i < lo->ldo_dir_stripe_count; i++) {
4774                 struct lustre_handle    lockh;
4775                 struct ldlm_res_id      *res_id;
4776
4777                 res_id = &lod_env_info(env)->lti_res_id;
4778                 fid_build_reg_res_name(lu_object_fid(&lo->ldo_stripe[i]->do_lu),
4779                                        res_id);
4780                 einfo->ei_res_id = res_id;
4781
4782                 LASSERT(lo->ldo_stripe[i] != NULL);
4783                 if (likely(dt_object_remote(lo->ldo_stripe[i]))) {
4784                         rc = dt_object_lock(env, lo->ldo_stripe[i], &lockh,
4785                                             einfo, policy);
4786                 } else {
4787                         struct ldlm_namespace *ns = einfo->ei_namespace;
4788                         ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
4789                         ldlm_completion_callback completion = einfo->ei_cb_cp;
4790                         __u64   dlmflags = LDLM_FL_ATOMIC_CB;
4791
4792                         if (einfo->ei_mode == LCK_PW ||
4793                             einfo->ei_mode == LCK_EX)
4794                                 dlmflags |= LDLM_FL_COS_INCOMPAT;
4795
4796                         /* This only happens if there are mulitple stripes
4797                          * on the master MDT, i.e. except stripe0, there are
4798                          * other stripes on the Master MDT as well, Only
4799                          * happens in the test case right now. */
4800                         LASSERT(ns != NULL);
4801                         rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS,
4802                                                     policy, einfo->ei_mode,
4803                                                     &dlmflags, blocking,
4804                                                     completion, NULL,
4805                                                     NULL, 0, LVB_T_NONE,
4806                                                     NULL, &lockh);
4807                 }
4808                 if (rc != 0)
4809                         break;
4810                 slave_locks->handles[i] = lockh;
4811         }
4812         einfo->ei_cbdata = slave_locks;
4813
4814         if (rc != 0 && slave_locks != NULL) {
4815                 lod_object_unlock_internal(env, dt, einfo, policy);
4816                 OBD_FREE(slave_locks, slave_locks_size);
4817         }
4818         EXIT;
4819 out:
4820         if (rc != 0)
4821                 einfo->ei_cbdata = NULL;
4822         RETURN(rc);
4823 }
4824
4825 /**
4826  * Implementation of dt_object_operations::do_invalidate.
4827  *
4828  * \see dt_object_operations::do_invalidate() in the API description for details
4829  */
4830 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
4831 {
4832         return dt_invalidate(env, dt_object_child(dt));
4833 }
4834
4835 static int lod_declare_layout_change(const struct lu_env *env,
4836                                      struct dt_object *dt,
4837                                      struct layout_intent *layout,
4838                                      const struct lu_buf *buf,
4839                                      struct thandle *th)
4840 {
4841         struct lod_thread_info  *info = lod_env_info(env);
4842         struct lod_object *lo = lod_dt_obj(dt);
4843         struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
4844         struct dt_object *next = dt_object_child(dt);
4845         struct ost_pool *inuse = &info->lti_inuse_osts;
4846         struct lod_layout_component *lod_comp;
4847         struct lov_comp_md_v1 *comp_v1 = NULL;
4848         bool replay = false;
4849         bool need_create = false;
4850         int i, rc;
4851         ENTRY;
4852
4853         if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
4854             dt_object_remote(next))
4855                 RETURN(-EINVAL);
4856
4857         dt_write_lock(env, next, 0);
4858         /*
4859          * In case the client is passing lovea, which only happens during
4860          * the replay of layout intent write RPC for now, we may need to
4861          * parse the lovea and apply new layout configuration.
4862          */
4863         if (buf && buf->lb_len)  {
4864                 struct lov_user_md_v1 *v1 = buf->lb_buf;
4865
4866                 if (v1->lmm_magic != (LOV_MAGIC_DEF | LOV_MAGIC_COMP_V1) &&
4867                     v1->lmm_magic !=
4868                                 __swab32(LOV_MAGIC_DEF | LOV_MAGIC_COMP_V1)) {
4869                         CERROR("%s: the replay buffer of layout extend "
4870                                "(magic %#x) does not contain expected "
4871                                "composite layout.\n",
4872                                lod2obd(d)->obd_name, v1->lmm_magic);
4873                         GOTO(out, rc = -EINVAL);
4874                 }
4875
4876                 lod_object_free_striping(env, lo);
4877                 rc = lod_use_defined_striping(env, lo, buf);
4878                 if (rc)
4879                         GOTO(out, rc);
4880
4881                 rc = lod_get_lov_ea(env, lo);
4882                 if (rc <= 0)
4883                         GOTO(out, rc);
4884                 /* old on-disk EA is stored in info->lti_buf */
4885                 comp_v1 = (struct lov_comp_md_v1 *)&info->lti_buf.lb_buf;
4886                 replay = true;
4887         } else {
4888                 /* non replay path */
4889                 rc = lod_load_striping_locked(env, lo);
4890                 if (rc)
4891                         GOTO(out, rc);
4892
4893                 /* Prepare inuse array for composite file */
4894                 rc = lod_prepare_inuse(env, lo);
4895                 if (rc)
4896                         GOTO(out, rc);
4897         }
4898
4899         /* Make sure defined layout covers the requested write range. */
4900         lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
4901         if (lo->ldo_comp_cnt > 1 &&
4902             lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
4903             lod_comp->llc_extent.e_end < layout->li_end) {
4904                 CDEBUG(replay ? D_ERROR : D_LAYOUT,
4905                        "%s: the defined layout [0, %#llx) does not covers "
4906                        "the write range [%#llx, %#llx).\n",
4907                        lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
4908                        layout->li_start, layout->li_end);
4909                 GOTO(out, rc = -EINVAL);
4910         }
4911
4912         /*
4913          * Iterate ld->ldo_comp_entries, find the component whose extent under
4914          * the write range and not instantianted.
4915          */
4916         for (i = 0; i < lo->ldo_comp_cnt; i++) {
4917                 lod_comp = &lo->ldo_comp_entries[i];
4918
4919                 if (lod_comp->llc_extent.e_start >= layout->li_end)
4920                         break;
4921
4922                 if (!replay) {
4923                         if (lod_comp_inited(lod_comp))
4924                                 continue;
4925                 } else {
4926                         /**
4927                          * In replay path, lod_comp is the EA passed by
4928                          * client replay buffer,  comp_v1 is the pre-recovery
4929                          * on-disk EA, we'd sift out those components which
4930                          * were init-ed in the on-disk EA.
4931                          */
4932                         if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
4933                             LCME_FL_INIT)
4934                                 continue;
4935                 }
4936                 /*
4937                  * this component hasn't instantiated in normal path, or during
4938                  * replay it needs replay the instantiation.
4939                  */
4940
4941                 /* A released component is being extended */
4942                 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
4943                         GOTO(out, rc = -EINVAL);
4944
4945                 need_create = true;
4946
4947                 rc = lod_qos_prep_create(env, lo, NULL, th, i, inuse);
4948                 if (rc)
4949                         break;
4950         }
4951
4952         if (need_create)
4953                 lod_obj_inc_layout_gen(lo);
4954         else
4955                 GOTO(unlock, rc = -EALREADY);
4956
4957         if (!rc) {
4958                 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
4959                 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
4960                                                XATTR_NAME_LOV, 0, th);
4961         }
4962 out:
4963         if (rc)
4964                 lod_object_free_striping(env, lo);
4965
4966 unlock:
4967         dt_write_unlock(env, next);
4968
4969         RETURN(rc);
4970 }
4971
4972 /**
4973  * Instantiate layout component objects which covers the intent write offset.
4974  */
4975 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
4976                              struct layout_intent *layout,
4977                              const struct lu_buf *buf, struct thandle *th)
4978 {
4979         struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4980
4981         RETURN(lod_striped_create(env, dt, attr, NULL, th));
4982 }
4983
4984 struct dt_object_operations lod_obj_ops = {
4985         .do_read_lock           = lod_read_lock,
4986         .do_write_lock          = lod_write_lock,
4987         .do_read_unlock         = lod_read_unlock,
4988         .do_write_unlock        = lod_write_unlock,
4989         .do_write_locked        = lod_write_locked,
4990         .do_attr_get            = lod_attr_get,
4991         .do_declare_attr_set    = lod_declare_attr_set,
4992         .do_attr_set            = lod_attr_set,
4993         .do_xattr_get           = lod_xattr_get,
4994         .do_declare_xattr_set   = lod_declare_xattr_set,
4995         .do_xattr_set           = lod_xattr_set,
4996         .do_declare_xattr_del   = lod_declare_xattr_del,
4997         .do_xattr_del           = lod_xattr_del,
4998         .do_xattr_list          = lod_xattr_list,
4999         .do_ah_init             = lod_ah_init,
5000         .do_declare_create      = lod_declare_create,
5001         .do_create              = lod_create,
5002         .do_declare_destroy     = lod_declare_destroy,
5003         .do_destroy             = lod_destroy,
5004         .do_index_try           = lod_index_try,
5005         .do_declare_ref_add     = lod_declare_ref_add,
5006         .do_ref_add             = lod_ref_add,
5007         .do_declare_ref_del     = lod_declare_ref_del,
5008         .do_ref_del             = lod_ref_del,
5009         .do_object_sync         = lod_object_sync,
5010         .do_object_lock         = lod_object_lock,
5011         .do_object_unlock       = lod_object_unlock,
5012         .do_invalidate          = lod_invalidate,
5013         .do_declare_layout_change = lod_declare_layout_change,
5014         .do_layout_change       = lod_layout_change,
5015 };
5016
5017 /**
5018  * Implementation of dt_body_operations::dbo_read.
5019  *
5020  * \see dt_body_operations::dbo_read() in the API description for details.
5021  */
5022 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
5023                         struct lu_buf *buf, loff_t *pos)
5024 {
5025         struct dt_object *next = dt_object_child(dt);
5026
5027         LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
5028                 S_ISLNK(dt->do_lu.lo_header->loh_attr));
5029         return next->do_body_ops->dbo_read(env, next, buf, pos);
5030 }
5031
5032 /**
5033  * Implementation of dt_body_operations::dbo_declare_write.
5034  *
5035  * \see dt_body_operations::dbo_declare_write() in the API description
5036  * for details.
5037  */
5038 static ssize_t lod_declare_write(const struct lu_env *env,
5039                                  struct dt_object *dt,
5040                                  const struct lu_buf *buf, loff_t pos,
5041                                  struct thandle *th)
5042 {
5043         return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
5044 }
5045
5046 /**
5047  * Implementation of dt_body_operations::dbo_write.
5048  *
5049  * \see dt_body_operations::dbo_write() in the API description for details.
5050  */
5051 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
5052                          const struct lu_buf *buf, loff_t *pos,
5053                          struct thandle *th, int iq)
5054 {
5055         LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
5056                 S_ISLNK(dt->do_lu.lo_header->loh_attr));
5057         return lod_sub_write(env, dt_object_child(dt), buf, pos, th, iq);
5058 }
5059
5060 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
5061                              __u64 start, __u64 end, struct thandle *th)
5062 {
5063         if (dt_object_remote(dt))
5064                 return -ENOTSUPP;
5065
5066         return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
5067 }
5068
5069 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
5070                      __u64 start, __u64 end, struct thandle *th)
5071 {
5072         if (dt_object_remote(dt))
5073                 return -ENOTSUPP;
5074
5075         LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
5076         return lod_sub_punch(env, dt_object_child(dt), start, end, th);
5077 }
5078
5079 /*
5080  * different type of files use the same body_ops because object may be created
5081  * in OUT, where there is no chance to set correct body_ops for each type, so
5082  * body_ops themselves will check file type inside, see lod_read/write/punch for
5083  * details.
5084  */
5085 const struct dt_body_operations lod_body_ops = {
5086         .dbo_read               = lod_read,
5087         .dbo_declare_write      = lod_declare_write,
5088         .dbo_write              = lod_write,
5089         .dbo_declare_punch      = lod_declare_punch,
5090         .dbo_punch              = lod_punch,
5091 };
5092
5093 /**
5094  * Implementation of lu_object_operations::loo_object_init.
5095  *
5096  * The function determines the type and the index of the target device using
5097  * sequence of the object's FID. Then passes control down to the
5098  * corresponding device:
5099  *  OSD for the local objects, OSP for remote
5100  *
5101  * \see lu_object_operations::loo_object_init() in the API description
5102  * for details.
5103  */
5104 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
5105                            const struct lu_object_conf *conf)
5106 {
5107         struct lod_device       *lod    = lu2lod_dev(lo->lo_dev);
5108         struct lu_device        *cdev   = NULL;
5109         struct lu_object        *cobj;
5110         struct lod_tgt_descs    *ltd    = NULL;
5111         struct lod_tgt_desc     *tgt;
5112         u32                      idx    = 0;
5113         int                      type   = LU_SEQ_RANGE_ANY;
5114         int                      rc;
5115         ENTRY;
5116
5117         rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
5118         if (rc != 0) {
5119                 /* Note: Sometimes, it will Return EAGAIN here, see
5120                  * ptrlpc_import_delay_req(), which might confuse
5121                  * lu_object_find_at() and make it wait there incorrectly.
5122                  * so we convert it to EIO here.*/
5123                 if (rc == -EAGAIN)
5124                         rc = -EIO;
5125
5126                 RETURN(rc);
5127         }
5128
5129         if (type == LU_SEQ_RANGE_MDT &&
5130             idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
5131                 cdev = &lod->lod_child->dd_lu_dev;
5132         } else if (type == LU_SEQ_RANGE_MDT) {
5133                 ltd = &lod->lod_mdt_descs;
5134                 lod_getref(ltd);
5135         } else if (type == LU_SEQ_RANGE_OST) {
5136                 ltd = &lod->lod_ost_descs;
5137                 lod_getref(ltd);
5138         } else {
5139                 LBUG();
5140         }
5141
5142         if (ltd != NULL) {
5143                 if (ltd->ltd_tgts_size > idx &&
5144                     cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
5145                         tgt = LTD_TGT(ltd, idx);
5146
5147                         LASSERT(tgt != NULL);
5148                         LASSERT(tgt->ltd_tgt != NULL);
5149
5150                         cdev = &(tgt->ltd_tgt->dd_lu_dev);
5151                 }
5152                 lod_putref(lod, ltd);
5153         }
5154
5155         if (unlikely(cdev == NULL))
5156                 RETURN(-ENOENT);
5157
5158         cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
5159         if (unlikely(cobj == NULL))
5160                 RETURN(-ENOMEM);
5161
5162         lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
5163
5164         lu_object_add(lo, cobj);
5165
5166         RETURN(0);
5167 }
5168
5169 /**
5170  *
5171  * Release resources associated with striping.
5172  *
5173  * If the object is striped (regular or directory), then release
5174  * the stripe objects references and free the ldo_stripe array.
5175  *
5176  * \param[in] env       execution environment
5177  * \param[in] lo        object
5178  */
5179 void lod_object_free_striping(const struct lu_env *env, struct lod_object *lo)
5180 {
5181         struct lod_layout_component *lod_comp;
5182         int i, j;
5183
5184         if (lo->ldo_stripe != NULL) {
5185                 LASSERT(lo->ldo_comp_entries == NULL);
5186                 LASSERT(lo->ldo_dir_stripes_allocated > 0);
5187
5188                 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5189                         if (lo->ldo_stripe[i])
5190                                 dt_object_put(env, lo->ldo_stripe[i]);
5191                 }
5192
5193                 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
5194                 OBD_FREE(lo->ldo_stripe, j);
5195                 lo->ldo_stripe = NULL;
5196                 lo->ldo_dir_stripes_allocated = 0;
5197                 lo->ldo_dir_stripe_loaded = 0;
5198                 lo->ldo_dir_stripe_count = 0;
5199         } else if (lo->ldo_comp_entries != NULL) {
5200                 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5201                         /* free lod_layout_component::llc_stripe array */
5202                         lod_comp = &lo->ldo_comp_entries[i];
5203
5204                         if (lod_comp->llc_stripe == NULL)
5205                                 continue;
5206                         LASSERT(lod_comp->llc_stripes_allocated != 0);
5207                         for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
5208                                 if (lod_comp->llc_stripe[j] != NULL)
5209                                         lu_object_put(env,
5210                                                &lod_comp->llc_stripe[j]->do_lu);
5211                         }
5212                         OBD_FREE(lod_comp->llc_stripe,
5213                                  sizeof(struct dt_object *) *
5214                                  lod_comp->llc_stripes_allocated);
5215                         lod_comp->llc_stripe = NULL;
5216                         lod_comp->llc_stripes_allocated = 0;
5217                 }
5218                 lod_free_comp_entries(lo);
5219                 lo->ldo_comp_cached = 0;
5220         }
5221 }
5222
5223 /**
5224  * Implementation of lu_object_operations::loo_object_free.
5225  *
5226  * \see lu_object_operations::loo_object_free() in the API description
5227  * for details.
5228  */
5229 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
5230 {
5231         struct lod_object *lo = lu2lod_obj(o);
5232
5233         /* release all underlying object pinned */
5234         lod_object_free_striping(env, lo);
5235         lu_object_fini(o);
5236         OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
5237 }
5238
5239 /**
5240  * Implementation of lu_object_operations::loo_object_release.
5241  *
5242  * \see lu_object_operations::loo_object_release() in the API description
5243  * for details.
5244  */
5245 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
5246 {
5247         /* XXX: shouldn't we release everything here in case if object
5248          * creation failed before? */
5249 }
5250
5251 /**
5252  * Implementation of lu_object_operations::loo_object_print.
5253  *
5254  * \see lu_object_operations::loo_object_print() in the API description
5255  * for details.
5256  */
5257 static int lod_object_print(const struct lu_env *env, void *cookie,
5258                             lu_printer_t p, const struct lu_object *l)
5259 {
5260         struct lod_object *o = lu2lod_obj((struct lu_object *) l);
5261
5262         return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
5263 }
5264
5265 struct lu_object_operations lod_lu_obj_ops = {
5266         .loo_object_init        = lod_object_init,
5267         .loo_object_free        = lod_object_free,
5268         .loo_object_release     = lod_object_release,
5269         .loo_object_print       = lod_object_print,
5270 };