4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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.
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
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * lustre/lod/lod_object.c
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.
38 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
41 #define DEBUG_SUBSYSTEM S_MDS
43 #include <linux/random.h>
46 #include <obd_class.h>
47 #include <obd_support.h>
49 #include <lustre_fid.h>
50 #include <lustre_linkea.h>
51 #include <lustre_lmv.h>
52 #include <uapi/linux/lustre/lustre_param.h>
53 #include <lustre_swab.h>
54 #include <uapi/linux/lustre/lustre_ver.h>
55 #include <lprocfs_status.h>
56 #include <md_object.h>
58 #include "lod_internal.h"
60 static const char dot[] = ".";
61 static const char dotdot[] = "..";
64 * Implementation of dt_index_operations::dio_lookup
66 * Used with regular (non-striped) objects.
68 * \see dt_index_operations::dio_lookup() in the API description for details.
70 static int lod_lookup(const struct lu_env *env, struct dt_object *dt,
71 struct dt_rec *rec, const struct dt_key *key)
73 struct dt_object *next = dt_object_child(dt);
74 return next->do_index_ops->dio_lookup(env, next, rec, key);
78 * Implementation of dt_index_operations::dio_declare_insert.
80 * Used with regular (non-striped) objects.
82 * \see dt_index_operations::dio_declare_insert() in the API description
85 static int lod_declare_insert(const struct lu_env *env, struct dt_object *dt,
86 const struct dt_rec *rec,
87 const struct dt_key *key, struct thandle *th)
89 return lod_sub_declare_insert(env, dt_object_child(dt), rec, key, th);
93 * Implementation of dt_index_operations::dio_insert.
95 * Used with regular (non-striped) objects
97 * \see dt_index_operations::dio_insert() in the API description for details.
99 static int lod_insert(const struct lu_env *env, struct dt_object *dt,
100 const struct dt_rec *rec, const struct dt_key *key,
103 return lod_sub_insert(env, dt_object_child(dt), rec, key, th);
107 * Implementation of dt_index_operations::dio_declare_delete.
109 * Used with regular (non-striped) objects.
111 * \see dt_index_operations::dio_declare_delete() in the API description
114 static int lod_declare_delete(const struct lu_env *env, struct dt_object *dt,
115 const struct dt_key *key, struct thandle *th)
117 return lod_sub_declare_delete(env, dt_object_child(dt), key, th);
121 * Implementation of dt_index_operations::dio_delete.
123 * Used with regular (non-striped) objects.
125 * \see dt_index_operations::dio_delete() in the API description for details.
127 static int lod_delete(const struct lu_env *env, struct dt_object *dt,
128 const struct dt_key *key, struct thandle *th)
130 return lod_sub_delete(env, dt_object_child(dt), key, th);
134 * Implementation of dt_it_ops::init.
136 * Used with regular (non-striped) objects.
138 * \see dt_it_ops::init() in the API description for details.
140 static struct dt_it *lod_it_init(const struct lu_env *env,
141 struct dt_object *dt, __u32 attr)
143 struct dt_object *next = dt_object_child(dt);
144 struct lod_it *it = &lod_env_info(env)->lti_it;
145 struct dt_it *it_next;
147 it_next = next->do_index_ops->dio_it.init(env, next, attr);
151 /* currently we do not use more than one iterator per thread
152 * so we store it in thread info. if at some point we need
153 * more active iterators in a single thread, we can allocate
155 LASSERT(it->lit_obj == NULL);
157 it->lit_it = it_next;
160 return (struct dt_it *)it;
163 #define LOD_CHECK_IT(env, it) \
165 LASSERT((it)->lit_obj != NULL); \
166 LASSERT((it)->lit_it != NULL); \
170 * Implementation of dt_index_operations::dio_it.fini.
172 * Used with regular (non-striped) objects.
174 * \see dt_index_operations::dio_it.fini() in the API description for details.
176 static void lod_it_fini(const struct lu_env *env, struct dt_it *di)
178 struct lod_it *it = (struct lod_it *)di;
180 LOD_CHECK_IT(env, it);
181 it->lit_obj->do_index_ops->dio_it.fini(env, it->lit_it);
183 /* the iterator not in use any more */
189 * Implementation of dt_it_ops::get.
191 * Used with regular (non-striped) objects.
193 * \see dt_it_ops::get() in the API description for details.
195 static int lod_it_get(const struct lu_env *env, struct dt_it *di,
196 const struct dt_key *key)
198 const struct lod_it *it = (const struct lod_it *)di;
200 LOD_CHECK_IT(env, it);
201 return it->lit_obj->do_index_ops->dio_it.get(env, it->lit_it, key);
205 * Implementation of dt_it_ops::put.
207 * Used with regular (non-striped) objects.
209 * \see dt_it_ops::put() in the API description for details.
211 static void lod_it_put(const struct lu_env *env, struct dt_it *di)
213 struct lod_it *it = (struct lod_it *)di;
215 LOD_CHECK_IT(env, it);
216 return it->lit_obj->do_index_ops->dio_it.put(env, it->lit_it);
220 * Implementation of dt_it_ops::next.
222 * Used with regular (non-striped) objects
224 * \see dt_it_ops::next() in the API description for details.
226 static int lod_it_next(const struct lu_env *env, struct dt_it *di)
228 struct lod_it *it = (struct lod_it *)di;
230 LOD_CHECK_IT(env, it);
231 return it->lit_obj->do_index_ops->dio_it.next(env, it->lit_it);
235 * Implementation of dt_it_ops::key.
237 * Used with regular (non-striped) objects.
239 * \see dt_it_ops::key() in the API description for details.
241 static struct dt_key *lod_it_key(const struct lu_env *env,
242 const struct dt_it *di)
244 const struct lod_it *it = (const struct lod_it *)di;
246 LOD_CHECK_IT(env, it);
247 return it->lit_obj->do_index_ops->dio_it.key(env, it->lit_it);
251 * Implementation of dt_it_ops::key_size.
253 * Used with regular (non-striped) objects.
255 * \see dt_it_ops::key_size() in the API description for details.
257 static int lod_it_key_size(const struct lu_env *env, const struct dt_it *di)
259 struct lod_it *it = (struct lod_it *)di;
261 LOD_CHECK_IT(env, it);
262 return it->lit_obj->do_index_ops->dio_it.key_size(env, it->lit_it);
266 * Implementation of dt_it_ops::rec.
268 * Used with regular (non-striped) objects.
270 * \see dt_it_ops::rec() in the API description for details.
272 static int lod_it_rec(const struct lu_env *env, const struct dt_it *di,
273 struct dt_rec *rec, __u32 attr)
275 const struct lod_it *it = (const struct lod_it *)di;
277 LOD_CHECK_IT(env, it);
278 return it->lit_obj->do_index_ops->dio_it.rec(env, it->lit_it, rec,
283 * Implementation of dt_it_ops::rec_size.
285 * Used with regular (non-striped) objects.
287 * \see dt_it_ops::rec_size() in the API description for details.
289 static int lod_it_rec_size(const struct lu_env *env, const struct dt_it *di,
292 const struct lod_it *it = (const struct lod_it *)di;
294 LOD_CHECK_IT(env, it);
295 return it->lit_obj->do_index_ops->dio_it.rec_size(env, it->lit_it,
300 * Implementation of dt_it_ops::store.
302 * Used with regular (non-striped) objects.
304 * \see dt_it_ops::store() in the API description for details.
306 static __u64 lod_it_store(const struct lu_env *env, const struct dt_it *di)
308 const struct lod_it *it = (const struct lod_it *)di;
310 LOD_CHECK_IT(env, it);
311 return it->lit_obj->do_index_ops->dio_it.store(env, it->lit_it);
315 * Implementation of dt_it_ops::load.
317 * Used with regular (non-striped) objects.
319 * \see dt_it_ops::load() in the API description for details.
321 static int lod_it_load(const struct lu_env *env, const struct dt_it *di,
324 const struct lod_it *it = (const struct lod_it *)di;
326 LOD_CHECK_IT(env, it);
327 return it->lit_obj->do_index_ops->dio_it.load(env, it->lit_it, hash);
331 * Implementation of dt_it_ops::key_rec.
333 * Used with regular (non-striped) objects.
335 * \see dt_it_ops::rec() in the API description for details.
337 static int lod_it_key_rec(const struct lu_env *env, const struct dt_it *di,
340 const struct lod_it *it = (const struct lod_it *)di;
342 LOD_CHECK_IT(env, it);
343 return it->lit_obj->do_index_ops->dio_it.key_rec(env, it->lit_it,
347 static const struct dt_index_operations lod_index_ops = {
348 .dio_lookup = lod_lookup,
349 .dio_declare_insert = lod_declare_insert,
350 .dio_insert = lod_insert,
351 .dio_declare_delete = lod_declare_delete,
352 .dio_delete = lod_delete,
360 .key_size = lod_it_key_size,
362 .rec_size = lod_it_rec_size,
363 .store = lod_it_store,
365 .key_rec = lod_it_key_rec,
370 * Implementation of dt_index_operations::dio_lookup
372 * Used with striped directories.
374 * \see dt_index_operations::dio_lookup() in the API description for details.
376 static int lod_striped_lookup(const struct lu_env *env, struct dt_object *dt,
377 struct dt_rec *rec, const struct dt_key *key)
379 struct lod_object *lo = lod_dt_obj(dt);
380 struct dt_object *next;
381 const char *name = (const char *)key;
383 LASSERT(lo->ldo_dir_stripe_count > 0);
385 if (strcmp(name, dot) == 0) {
386 struct lu_fid *fid = (struct lu_fid *)rec;
388 *fid = *lod_object_fid(lo);
392 if (strcmp(name, dotdot) == 0) {
393 next = dt_object_child(dt);
397 index = __lmv_name_to_stripe_index(lo->ldo_dir_hash_type,
398 lo->ldo_dir_stripe_count,
399 lo->ldo_dir_migrate_hash,
400 lo->ldo_dir_migrate_offset,
401 name, strlen(name), true);
405 next = lo->ldo_stripe[index];
406 if (!next || !dt_object_exists(next))
410 return next->do_index_ops->dio_lookup(env, next, rec, key);
414 * Implementation of dt_it_ops::init.
416 * Used with striped objects. Internally just initializes the iterator
417 * on the first stripe.
419 * \see dt_it_ops::init() in the API description for details.
421 static struct dt_it *lod_striped_it_init(const struct lu_env *env,
422 struct dt_object *dt, __u32 attr)
424 struct lod_object *lo = lod_dt_obj(dt);
425 struct dt_object *next;
426 struct lod_it *it = &lod_env_info(env)->lti_it;
427 struct dt_it *it_next;
430 LASSERT(lo->ldo_dir_stripe_count > 0);
433 next = lo->ldo_stripe[index];
434 if (next && dt_object_exists(next))
436 } while (++index < lo->ldo_dir_stripe_count);
438 /* no valid stripe */
439 if (!next || !dt_object_exists(next))
440 return ERR_PTR(-ENODEV);
442 LASSERT(next->do_index_ops != NULL);
444 it_next = next->do_index_ops->dio_it.init(env, next, attr);
448 /* currently we do not use more than one iterator per thread
449 * so we store it in thread info. if at some point we need
450 * more active iterators in a single thread, we can allocate
452 LASSERT(it->lit_obj == NULL);
454 it->lit_stripe_index = index;
456 it->lit_it = it_next;
459 return (struct dt_it *)it;
462 #define LOD_CHECK_STRIPED_IT(env, it, lo) \
464 LASSERT((it)->lit_obj != NULL); \
465 LASSERT((it)->lit_it != NULL); \
466 LASSERT((lo)->ldo_dir_stripe_count > 0); \
467 LASSERT((it)->lit_stripe_index < (lo)->ldo_dir_stripe_count); \
471 * Implementation of dt_it_ops::fini.
473 * Used with striped objects.
475 * \see dt_it_ops::fini() in the API description for details.
477 static void lod_striped_it_fini(const struct lu_env *env, struct dt_it *di)
479 struct lod_it *it = (struct lod_it *)di;
480 struct lod_object *lo = lod_dt_obj(it->lit_obj);
481 struct dt_object *next;
483 /* If lit_it == NULL, then it means the sub_it has been finished,
484 * which only happens in failure cases, see lod_striped_it_next() */
485 if (it->lit_it != NULL) {
486 LOD_CHECK_STRIPED_IT(env, it, lo);
488 next = lo->ldo_stripe[it->lit_stripe_index];
490 LASSERT(next->do_index_ops != NULL);
491 next->do_index_ops->dio_it.fini(env, it->lit_it);
495 /* the iterator not in use any more */
498 it->lit_stripe_index = 0;
502 * Implementation of dt_it_ops::get.
504 * Right now it's not used widely, only to reset the iterator to the
505 * initial position. It should be possible to implement a full version
506 * which chooses a correct stripe to be able to position with any key.
508 * \see dt_it_ops::get() in the API description for details.
510 static int lod_striped_it_get(const struct lu_env *env, struct dt_it *di,
511 const struct dt_key *key)
513 const struct lod_it *it = (const struct lod_it *)di;
514 struct lod_object *lo = lod_dt_obj(it->lit_obj);
515 struct dt_object *next;
517 LOD_CHECK_STRIPED_IT(env, it, lo);
519 next = lo->ldo_stripe[it->lit_stripe_index];
520 LASSERT(next != NULL);
521 LASSERT(dt_object_exists(next));
522 LASSERT(next->do_index_ops != NULL);
524 return next->do_index_ops->dio_it.get(env, it->lit_it, key);
528 * Implementation of dt_it_ops::put.
530 * Used with striped objects.
532 * \see dt_it_ops::put() in the API description for details.
534 static void lod_striped_it_put(const struct lu_env *env, struct dt_it *di)
536 struct lod_it *it = (struct lod_it *)di;
537 struct lod_object *lo = lod_dt_obj(it->lit_obj);
538 struct dt_object *next;
541 * If lit_it == NULL, then it means the sub_it has been finished,
542 * which only happens in failure cases, see lod_striped_it_next()
547 LOD_CHECK_STRIPED_IT(env, it, lo);
549 next = lo->ldo_stripe[it->lit_stripe_index];
550 LASSERT(next != NULL);
551 LASSERT(next->do_index_ops != NULL);
553 return next->do_index_ops->dio_it.put(env, it->lit_it);
557 * Implementation of dt_it_ops::next.
559 * Used with striped objects. When the end of the current stripe is
560 * reached, the method takes the next stripe's iterator.
562 * \see dt_it_ops::next() in the API description for details.
564 static int lod_striped_it_next(const struct lu_env *env, struct dt_it *di)
566 struct lod_it *it = (struct lod_it *)di;
567 struct lod_object *lo = lod_dt_obj(it->lit_obj);
568 struct dt_object *next;
569 struct dt_it *it_next;
575 LOD_CHECK_STRIPED_IT(env, it, lo);
577 next = lo->ldo_stripe[it->lit_stripe_index];
578 LASSERT(next != NULL);
579 LASSERT(dt_object_exists(next));
580 LASSERT(next->do_index_ops != NULL);
582 rc = next->do_index_ops->dio_it.next(env, it->lit_it);
586 if (rc == 0 && it->lit_stripe_index == 0)
589 if (rc == 0 && it->lit_stripe_index > 0) {
590 struct lu_dirent *ent;
592 ent = (struct lu_dirent *)lod_env_info(env)->lti_key;
594 rc = next->do_index_ops->dio_it.rec(env, it->lit_it,
595 (struct dt_rec *)ent,
600 /* skip . and .. for slave stripe */
601 if ((strncmp(ent->lde_name, ".",
602 le16_to_cpu(ent->lde_namelen)) == 0 &&
603 le16_to_cpu(ent->lde_namelen) == 1) ||
604 (strncmp(ent->lde_name, "..",
605 le16_to_cpu(ent->lde_namelen)) == 0 &&
606 le16_to_cpu(ent->lde_namelen) == 2))
612 next->do_index_ops->dio_it.put(env, it->lit_it);
613 next->do_index_ops->dio_it.fini(env, it->lit_it);
616 /* go to next stripe */
617 index = it->lit_stripe_index;
618 while (++index < lo->ldo_dir_stripe_count) {
619 next = lo->ldo_stripe[index];
623 if (!dt_object_exists(next))
626 rc = next->do_ops->do_index_try(env, next,
627 &dt_directory_features);
631 LASSERT(next->do_index_ops != NULL);
633 it_next = next->do_index_ops->dio_it.init(env, next,
636 RETURN(PTR_ERR(it_next));
638 rc = next->do_index_ops->dio_it.get(env, it_next,
639 (const struct dt_key *)"");
641 RETURN(rc == 0 ? -EIO : rc);
643 it->lit_it = it_next;
644 it->lit_stripe_index = index;
653 * Implementation of dt_it_ops::key.
655 * Used with striped objects.
657 * \see dt_it_ops::key() in the API description for details.
659 static struct dt_key *lod_striped_it_key(const struct lu_env *env,
660 const struct dt_it *di)
662 const struct lod_it *it = (const struct lod_it *)di;
663 struct lod_object *lo = lod_dt_obj(it->lit_obj);
664 struct dt_object *next;
666 LOD_CHECK_STRIPED_IT(env, it, lo);
668 next = lo->ldo_stripe[it->lit_stripe_index];
669 LASSERT(next != NULL);
670 LASSERT(next->do_index_ops != NULL);
672 return next->do_index_ops->dio_it.key(env, it->lit_it);
676 * Implementation of dt_it_ops::key_size.
678 * Used with striped objects.
680 * \see dt_it_ops::size() in the API description for details.
682 static int lod_striped_it_key_size(const struct lu_env *env,
683 const struct dt_it *di)
685 struct lod_it *it = (struct lod_it *)di;
686 struct lod_object *lo = lod_dt_obj(it->lit_obj);
687 struct dt_object *next;
689 LOD_CHECK_STRIPED_IT(env, it, lo);
691 next = lo->ldo_stripe[it->lit_stripe_index];
692 LASSERT(next != NULL);
693 LASSERT(next->do_index_ops != NULL);
695 return next->do_index_ops->dio_it.key_size(env, it->lit_it);
699 * Implementation of dt_it_ops::rec.
701 * Used with striped objects.
703 * \see dt_it_ops::rec() in the API description for details.
705 static int lod_striped_it_rec(const struct lu_env *env, const struct dt_it *di,
706 struct dt_rec *rec, __u32 attr)
708 const struct lod_it *it = (const struct lod_it *)di;
709 struct lod_object *lo = lod_dt_obj(it->lit_obj);
710 struct dt_object *next;
712 LOD_CHECK_STRIPED_IT(env, it, lo);
714 next = lo->ldo_stripe[it->lit_stripe_index];
715 LASSERT(next != NULL);
716 LASSERT(next->do_index_ops != NULL);
718 return next->do_index_ops->dio_it.rec(env, it->lit_it, rec, attr);
722 * Implementation of dt_it_ops::rec_size.
724 * Used with striped objects.
726 * \see dt_it_ops::rec_size() in the API description for details.
728 static int lod_striped_it_rec_size(const struct lu_env *env,
729 const struct dt_it *di, __u32 attr)
731 struct lod_it *it = (struct lod_it *)di;
732 struct lod_object *lo = lod_dt_obj(it->lit_obj);
733 struct dt_object *next;
735 LOD_CHECK_STRIPED_IT(env, it, lo);
737 next = lo->ldo_stripe[it->lit_stripe_index];
738 LASSERT(next != NULL);
739 LASSERT(next->do_index_ops != NULL);
741 return next->do_index_ops->dio_it.rec_size(env, it->lit_it, attr);
745 * Implementation of dt_it_ops::store.
747 * Used with striped objects.
749 * \see dt_it_ops::store() in the API description for details.
751 static __u64 lod_striped_it_store(const struct lu_env *env,
752 const struct dt_it *di)
754 const struct lod_it *it = (const struct lod_it *)di;
755 struct lod_object *lo = lod_dt_obj(it->lit_obj);
756 struct dt_object *next;
758 LOD_CHECK_STRIPED_IT(env, it, lo);
760 next = lo->ldo_stripe[it->lit_stripe_index];
761 LASSERT(next != NULL);
762 LASSERT(next->do_index_ops != NULL);
764 return next->do_index_ops->dio_it.store(env, it->lit_it);
768 * Implementation of dt_it_ops::load.
770 * Used with striped objects.
772 * \see dt_it_ops::load() in the API description for details.
774 static int lod_striped_it_load(const struct lu_env *env,
775 const struct dt_it *di, __u64 hash)
777 const struct lod_it *it = (const struct lod_it *)di;
778 struct lod_object *lo = lod_dt_obj(it->lit_obj);
779 struct dt_object *next;
781 LOD_CHECK_STRIPED_IT(env, it, lo);
783 next = lo->ldo_stripe[it->lit_stripe_index];
784 LASSERT(next != NULL);
785 LASSERT(next->do_index_ops != NULL);
787 return next->do_index_ops->dio_it.load(env, it->lit_it, hash);
790 static const struct dt_index_operations lod_striped_index_ops = {
791 .dio_lookup = lod_striped_lookup,
792 .dio_declare_insert = lod_declare_insert,
793 .dio_insert = lod_insert,
794 .dio_declare_delete = lod_declare_delete,
795 .dio_delete = lod_delete,
797 .init = lod_striped_it_init,
798 .fini = lod_striped_it_fini,
799 .get = lod_striped_it_get,
800 .put = lod_striped_it_put,
801 .next = lod_striped_it_next,
802 .key = lod_striped_it_key,
803 .key_size = lod_striped_it_key_size,
804 .rec = lod_striped_it_rec,
805 .rec_size = lod_striped_it_rec_size,
806 .store = lod_striped_it_store,
807 .load = lod_striped_it_load,
812 * Append the FID for each shard of the striped directory after the
813 * given LMV EA header.
815 * To simplify striped directory and the consistency verification,
816 * we only store the LMV EA header on disk, for both master object
817 * and slave objects. When someone wants to know the whole LMV EA,
818 * such as client readdir(), we can build the entrie LMV EA on the
819 * MDT side (in RAM) via iterating the sub-directory entries that
820 * are contained in the master object of the stripe directory.
822 * For the master object of the striped directroy, the valid name
823 * for each shard is composed of the ${shard_FID}:${shard_idx}.
825 * There may be holes in the LMV EA if some shards' name entries
826 * are corrupted or lost.
828 * \param[in] env pointer to the thread context
829 * \param[in] lo pointer to the master object of the striped directory
830 * \param[in] buf pointer to the lu_buf which will hold the LMV EA
831 * \param[in] resize whether re-allocate the buffer if it is not big enough
833 * \retval positive size of the LMV EA
834 * \retval 0 for nothing to be loaded
835 * \retval negative error number on failure
837 int lod_load_lmv_shards(const struct lu_env *env, struct lod_object *lo,
838 struct lu_buf *buf, bool resize)
840 struct lu_dirent *ent =
841 (struct lu_dirent *)lod_env_info(env)->lti_key;
842 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
843 struct dt_object *obj = dt_object_child(&lo->ldo_obj);
844 struct lmv_mds_md_v1 *lmv1 = buf->lb_buf;
846 const struct dt_it_ops *iops;
848 __u32 magic = le32_to_cpu(lmv1->lmv_magic);
853 if (magic != LMV_MAGIC_V1)
856 stripes = le32_to_cpu(lmv1->lmv_stripe_count);
860 rc = lmv_mds_md_size(stripes, magic);
864 if (buf->lb_len < lmv1_size) {
873 lu_buf_alloc(buf, lmv1_size);
878 memcpy(buf->lb_buf, tbuf.lb_buf, tbuf.lb_len);
881 if (unlikely(!dt_try_as_dir(env, obj)))
884 memset(&lmv1->lmv_stripe_fids[0], 0, stripes * sizeof(struct lu_fid));
885 iops = &obj->do_index_ops->dio_it;
886 it = iops->init(env, obj, LUDA_64BITHASH);
890 rc = iops->load(env, it, 0);
892 rc = iops->next(env, it);
897 char name[FID_LEN + 2] = "";
902 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
908 fid_le_to_cpu(&fid, &ent->lde_fid);
909 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
910 if (ent->lde_name[0] == '.') {
911 if (ent->lde_namelen == 1)
914 if (ent->lde_namelen == 2 && ent->lde_name[1] == '.')
918 len = scnprintf(name, sizeof(name),
919 DFID":", PFID(&ent->lde_fid));
920 /* The ent->lde_name is composed of ${FID}:${index} */
921 if (ent->lde_namelen < len + 1 ||
922 memcmp(ent->lde_name, name, len) != 0) {
923 CDEBUG_LIMIT(lod->lod_lmv_failout ? D_ERROR : D_INFO,
924 "%s: invalid shard name %.*s with the FID "DFID" for the striped directory "DFID", %s\n",
925 lod2obd(lod)->obd_name, ent->lde_namelen,
926 ent->lde_name, PFID(&fid),
927 PFID(lu_object_fid(&obj->do_lu)),
928 lod->lod_lmv_failout ? "failout" : "skip");
930 if (lod->lod_lmv_failout)
938 if (ent->lde_name[len] < '0' ||
939 ent->lde_name[len] > '9') {
940 CDEBUG_LIMIT(lod->lod_lmv_failout ?
942 "%s: invalid shard name %.*s with the FID "DFID" for the striped directory "DFID", %s\n",
943 lod2obd(lod)->obd_name,
945 ent->lde_name, PFID(&fid),
946 PFID(lu_object_fid(&obj->do_lu)),
947 lod->lod_lmv_failout ?
950 if (lod->lod_lmv_failout)
956 index = index * 10 + ent->lde_name[len++] - '0';
957 } while (len < ent->lde_namelen);
959 if (len == ent->lde_namelen) {
960 /* Out of LMV EA range. */
961 if (index >= stripes) {
962 CERROR("%s: the shard %.*s for the striped "
963 "directory "DFID" is out of the known "
964 "LMV EA range [0 - %u], failout\n",
965 lod2obd(lod)->obd_name, ent->lde_namelen,
967 PFID(lu_object_fid(&obj->do_lu)),
973 /* The slot has been occupied. */
974 if (!fid_is_zero(&lmv1->lmv_stripe_fids[index])) {
978 &lmv1->lmv_stripe_fids[index]);
979 CERROR("%s: both the shard "DFID" and "DFID
980 " for the striped directory "DFID
981 " claim the same LMV EA slot at the "
982 "index %d, failout\n",
983 lod2obd(lod)->obd_name,
984 PFID(&fid0), PFID(&fid),
985 PFID(lu_object_fid(&obj->do_lu)), index);
990 /* stored as LE mode */
991 lmv1->lmv_stripe_fids[index] = ent->lde_fid;
994 rc = iops->next(env, it);
1001 RETURN(rc > 0 ? lmv_mds_md_size(stripes, magic) : rc);
1005 * Implementation of dt_object_operations::do_index_try.
1007 * \see dt_object_operations::do_index_try() in the API description for details.
1009 static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
1010 const struct dt_index_features *feat)
1012 struct lod_object *lo = lod_dt_obj(dt);
1013 struct dt_object *next = dt_object_child(dt);
1017 LASSERT(next->do_ops);
1018 LASSERT(next->do_ops->do_index_try);
1020 rc = lod_striping_load(env, lo);
1024 rc = next->do_ops->do_index_try(env, next, feat);
1028 if (lo->ldo_dir_stripe_count > 0) {
1031 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1032 if (!lo->ldo_stripe[i])
1034 if (!dt_object_exists(lo->ldo_stripe[i]))
1036 rc = lo->ldo_stripe[i]->do_ops->do_index_try(env,
1037 lo->ldo_stripe[i], feat);
1041 dt->do_index_ops = &lod_striped_index_ops;
1043 dt->do_index_ops = &lod_index_ops;
1050 * Implementation of dt_object_operations::do_read_lock.
1052 * \see dt_object_operations::do_read_lock() in the API description for details.
1054 static void lod_read_lock(const struct lu_env *env, struct dt_object *dt,
1057 dt_read_lock(env, dt_object_child(dt), role);
1061 * Implementation of dt_object_operations::do_write_lock.
1063 * \see dt_object_operations::do_write_lock() in the API description for
1066 static void lod_write_lock(const struct lu_env *env, struct dt_object *dt,
1069 dt_write_lock(env, dt_object_child(dt), role);
1073 * Implementation of dt_object_operations::do_read_unlock.
1075 * \see dt_object_operations::do_read_unlock() in the API description for
1078 static void lod_read_unlock(const struct lu_env *env, struct dt_object *dt)
1080 dt_read_unlock(env, dt_object_child(dt));
1084 * Implementation of dt_object_operations::do_write_unlock.
1086 * \see dt_object_operations::do_write_unlock() in the API description for
1089 static void lod_write_unlock(const struct lu_env *env, struct dt_object *dt)
1091 dt_write_unlock(env, dt_object_child(dt));
1095 * Implementation of dt_object_operations::do_write_locked.
1097 * \see dt_object_operations::do_write_locked() in the API description for
1100 static int lod_write_locked(const struct lu_env *env, struct dt_object *dt)
1102 return dt_write_locked(env, dt_object_child(dt));
1106 * Implementation of dt_object_operations::do_attr_get.
1108 * \see dt_object_operations::do_attr_get() in the API description for details.
1110 static int lod_attr_get(const struct lu_env *env,
1111 struct dt_object *dt,
1112 struct lu_attr *attr)
1114 /* Note: for striped directory, client will merge attributes
1115 * from all of the sub-stripes see lmv_merge_attr(), and there
1116 * no MDD logic depend on directory nlink/size/time, so we can
1117 * always use master inode nlink and size for now. */
1118 return dt_attr_get(env, dt_object_child(dt), attr);
1121 void lod_adjust_stripe_size(struct lod_layout_component *comp,
1122 __u32 def_stripe_size)
1124 __u64 comp_end = comp->llc_extent.e_end;
1126 /* Choose stripe size if not set. Note that default stripe size can't
1127 * be used as is, because it must be multiplier of given component end.
1128 * - first check if default stripe size can be used
1129 * - if not than select the lowest set bit from component end and use
1130 * that value as stripe size
1132 if (!comp->llc_stripe_size) {
1133 if (comp_end == LUSTRE_EOF || !(comp_end % def_stripe_size))
1134 comp->llc_stripe_size = def_stripe_size;
1136 comp->llc_stripe_size = comp_end & ~(comp_end - 1);
1138 if (comp_end != LUSTRE_EOF &&
1139 comp_end & (LOV_MIN_STRIPE_SIZE - 1)) {
1140 CWARN("Component end %llu is not a multiple of min size %u\n",
1141 comp_end, LOV_MIN_STRIPE_SIZE);
1142 comp_end = round_up(comp_end, LOV_MIN_STRIPE_SIZE);
1144 /* check stripe size is multiplier of comp_end */
1145 if (comp_end != LUSTRE_EOF &&
1146 comp_end != comp->llc_extent.e_start &&
1147 comp_end % comp->llc_stripe_size) {
1148 /* fix that even for defined stripe size but warn
1149 * about the problem, that must not happen
1151 CWARN("Component end %llu is not aligned by the stripe size %u\n",
1152 comp_end, comp->llc_stripe_size);
1153 comp->llc_stripe_size = comp_end & ~(comp_end - 1);
1158 static inline void lod_adjust_stripe_info(struct lod_layout_component *comp,
1159 struct lov_desc *desc,
1162 if (comp->llc_pattern != LOV_PATTERN_MDT) {
1163 if (append_stripes) {
1164 comp->llc_stripe_count = append_stripes;
1165 } else if (!comp->llc_stripe_count) {
1166 comp->llc_stripe_count =
1167 desc->ld_default_stripe_count;
1171 lod_adjust_stripe_size(comp, desc->ld_default_stripe_size);
1174 int lod_obj_for_each_stripe(const struct lu_env *env, struct lod_object *lo,
1176 struct lod_obj_stripe_cb_data *data)
1178 struct lod_layout_component *lod_comp;
1182 mutex_lock(&lo->ldo_layout_mutex);
1183 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1184 lod_comp = &lo->ldo_comp_entries[i];
1186 if (lod_comp->llc_stripe == NULL)
1189 /* has stripe but not inited yet, this component has been
1190 * declared to be created, but hasn't created yet.
1192 if (!lod_comp_inited(lod_comp))
1195 if (data->locd_comp_skip_cb &&
1196 data->locd_comp_skip_cb(env, lo, i, data))
1199 if (data->locd_comp_cb) {
1200 rc = data->locd_comp_cb(env, lo, i, data);
1205 /* could used just to do sth about component, not each
1208 if (!data->locd_stripe_cb)
1211 LASSERT(lod_comp->llc_stripe_count > 0);
1212 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
1213 struct dt_object *dt = lod_comp->llc_stripe[j];
1217 rc = data->locd_stripe_cb(env, lo, dt, th, i, j, data);
1223 mutex_unlock(&lo->ldo_layout_mutex);
1227 static bool lod_obj_attr_set_comp_skip_cb(const struct lu_env *env,
1228 struct lod_object *lo, int comp_idx,
1229 struct lod_obj_stripe_cb_data *data)
1231 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[comp_idx];
1232 bool skipped = false;
1234 if (!(data->locd_attr->la_valid & LA_LAYOUT_VERSION))
1237 switch (lo->ldo_flr_state) {
1238 case LCM_FL_WRITE_PENDING: {
1241 /* skip stale components */
1242 if (lod_comp->llc_flags & LCME_FL_STALE) {
1247 /* skip valid and overlapping components, therefore any
1248 * attempts to write overlapped components will never succeed
1249 * because client will get EINPROGRESS. */
1250 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1254 if (lo->ldo_comp_entries[i].llc_flags & LCME_FL_STALE)
1257 if (lu_extent_is_overlapped(&lod_comp->llc_extent,
1258 &lo->ldo_comp_entries[i].llc_extent)) {
1266 case LCM_FL_SYNC_PENDING:
1269 LASSERTF(0, "impossible: %d\n", lo->ldo_flr_state);
1273 CDEBUG(D_LAYOUT, DFID": %s to set component %x to version: %u\n",
1274 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
1275 skipped ? "skipped" : "chose", lod_comp->llc_id,
1276 data->locd_attr->la_layout_version);
1282 lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
1283 struct dt_object *dt, struct thandle *th,
1284 int comp_idx, int stripe_idx,
1285 struct lod_obj_stripe_cb_data *data)
1287 if (data->locd_declare)
1288 return lod_sub_declare_attr_set(env, dt, data->locd_attr, th);
1290 if (data->locd_attr->la_valid & LA_LAYOUT_VERSION) {
1291 CDEBUG(D_LAYOUT, DFID": set layout version: %u, comp_idx: %d\n",
1292 PFID(lu_object_fid(&dt->do_lu)),
1293 data->locd_attr->la_layout_version, comp_idx);
1296 return lod_sub_attr_set(env, dt, data->locd_attr, th);
1300 * Implementation of dt_object_operations::do_declare_attr_set.
1302 * If the object is striped, then apply the changes to all the stripes.
1304 * \see dt_object_operations::do_declare_attr_set() in the API description
1307 static int lod_declare_attr_set(const struct lu_env *env,
1308 struct dt_object *dt,
1309 const struct lu_attr *attr,
1312 struct dt_object *next = dt_object_child(dt);
1313 struct lod_object *lo = lod_dt_obj(dt);
1318 * declare setattr on the local object
1320 rc = lod_sub_declare_attr_set(env, next, attr, th);
1324 /* osp_declare_attr_set() ignores all attributes other than
1325 * UID, GID, PROJID, and size, and osp_attr_set() ignores all
1326 * but UID, GID and PROJID. Declaration of size attr setting
1327 * happens through lod_declare_init_size(), and not through
1328 * this function. Therefore we need not load striping unless
1329 * ownership is changing. This should save memory and (we hope)
1330 * speed up rename().
1332 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1333 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1336 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1339 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID | LA_MODE |
1340 LA_ATIME | LA_MTIME | LA_CTIME |
1345 * load striping information, notice we don't do this when object
1346 * is being initialized as we don't need this information till
1347 * few specific cases like destroy, chown
1349 rc = lod_striping_load(env, lo);
1353 if (!lod_obj_is_striped(dt))
1357 * if object is striped declare changes on the stripes
1359 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1360 LASSERT(lo->ldo_stripe);
1361 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1362 if (lo->ldo_stripe[i] == NULL)
1364 if (!dt_object_exists(lo->ldo_stripe[i]))
1366 rc = lod_sub_declare_attr_set(env, lo->ldo_stripe[i],
1372 struct lod_obj_stripe_cb_data data = { { 0 } };
1374 data.locd_attr = attr;
1375 data.locd_declare = true;
1376 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1377 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1383 if (!dt_object_exists(next) || dt_object_remote(next) ||
1384 !S_ISREG(attr->la_mode))
1387 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1388 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
1392 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) ||
1393 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1394 struct lod_thread_info *info = lod_env_info(env);
1395 struct lu_buf *buf = &info->lti_buf;
1397 buf->lb_buf = info->lti_ea_store;
1398 buf->lb_len = info->lti_ea_store_size;
1399 rc = lod_sub_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1400 LU_XATTR_REPLACE, th);
1407 * Implementation of dt_object_operations::do_attr_set.
1409 * If the object is striped, then apply the changes to all or subset of
1410 * the stripes depending on the object type and specific attributes.
1412 * \see dt_object_operations::do_attr_set() in the API description for details.
1414 static int lod_attr_set(const struct lu_env *env,
1415 struct dt_object *dt,
1416 const struct lu_attr *attr,
1419 struct dt_object *next = dt_object_child(dt);
1420 struct lod_object *lo = lod_dt_obj(dt);
1425 * apply changes to the local object
1427 rc = lod_sub_attr_set(env, next, attr, th);
1431 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1432 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1435 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1438 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE | LA_PROJID |
1439 LA_ATIME | LA_MTIME | LA_CTIME |
1444 /* FIXME: a tricky case in the code path of mdd_layout_change():
1445 * the in-memory striping information has been freed in lod_xattr_set()
1446 * due to layout change. It has to load stripe here again. It only
1447 * changes flags of layout so declare_attr_set() is still accurate */
1448 rc = lod_striping_load(env, lo);
1452 if (!lod_obj_is_striped(dt))
1456 * if object is striped, apply changes to all the stripes
1458 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1459 LASSERT(lo->ldo_stripe);
1460 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1461 if (unlikely(lo->ldo_stripe[i] == NULL))
1464 if ((dt_object_exists(lo->ldo_stripe[i]) == 0))
1467 rc = lod_sub_attr_set(env, lo->ldo_stripe[i], attr, th);
1472 struct lod_obj_stripe_cb_data data = { { 0 } };
1474 data.locd_attr = attr;
1475 data.locd_declare = false;
1476 data.locd_comp_skip_cb = lod_obj_attr_set_comp_skip_cb;
1477 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1478 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1484 if (!dt_object_exists(next) || dt_object_remote(next) ||
1485 !S_ISREG(attr->la_mode))
1488 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1489 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
1493 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE)) {
1494 struct lod_thread_info *info = lod_env_info(env);
1495 struct lu_buf *buf = &info->lti_buf;
1496 struct ost_id *oi = &info->lti_ostid;
1497 struct lu_fid *fid = &info->lti_fid;
1498 struct lov_mds_md_v1 *lmm;
1499 struct lov_ost_data_v1 *objs;
1502 rc = lod_get_lov_ea(env, lo);
1506 buf->lb_buf = info->lti_ea_store;
1507 buf->lb_len = info->lti_ea_store_size;
1508 lmm = info->lti_ea_store;
1509 magic = le32_to_cpu(lmm->lmm_magic);
1510 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1511 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1512 struct lov_comp_md_entry_v1 *lcme =
1513 &lcm->lcm_entries[0];
1515 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
1516 magic = le32_to_cpu(lmm->lmm_magic);
1519 if (magic == LOV_MAGIC_V1)
1520 objs = &(lmm->lmm_objects[0]);
1522 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1523 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1524 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1526 fid_to_ostid(fid, oi);
1527 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1529 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1530 LU_XATTR_REPLACE, th);
1531 } else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1532 struct lod_thread_info *info = lod_env_info(env);
1533 struct lu_buf *buf = &info->lti_buf;
1534 struct lov_comp_md_v1 *lcm;
1535 struct lov_comp_md_entry_v1 *lcme;
1537 rc = lod_get_lov_ea(env, lo);
1541 buf->lb_buf = info->lti_ea_store;
1542 buf->lb_len = info->lti_ea_store_size;
1544 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1 &&
1545 le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_SEL)
1548 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1549 lcme = &lcm->lcm_entries[0];
1550 le64_add_cpu(&lcme->lcme_extent.e_start, 1);
1551 le64_add_cpu(&lcme->lcme_extent.e_end, -1);
1553 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1554 LU_XATTR_REPLACE, th);
1561 * Implementation of dt_object_operations::do_xattr_get.
1563 * If LOV EA is requested from the root object and it's not
1564 * found, then return default striping for the filesystem.
1566 * \see dt_object_operations::do_xattr_get() in the API description for details.
1568 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1569 struct lu_buf *buf, const char *name)
1571 struct lod_thread_info *info = lod_env_info(env);
1572 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1577 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1578 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1579 struct lmv_mds_md_v1 *lmv1;
1580 struct lmv_foreign_md *lfm;
1583 if (rc > (typeof(rc))sizeof(*lmv1))
1586 /* short (<= sizeof(struct lmv_mds_md_v1)) foreign LMV case */
1587 /* XXX empty foreign LMV is not allowed */
1588 if (rc <= offsetof(typeof(*lfm), lfm_value))
1589 RETURN(rc = rc > 0 ? -EINVAL : rc);
1591 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1592 BUILD_BUG_ON(sizeof(*lmv1) > sizeof(info->lti_key));
1594 /* lti_buf is large enough for *lmv1 or a short
1595 * (<= sizeof(struct lmv_mds_md_v1)) foreign LMV
1597 info->lti_buf.lb_buf = info->lti_key;
1598 info->lti_buf.lb_len = sizeof(*lmv1);
1599 rc = dt_xattr_get(env, dt_object_child(dt),
1600 &info->lti_buf, name);
1601 if (unlikely(rc <= offsetof(typeof(*lfm),
1603 RETURN(rc = rc > 0 ? -EINVAL : rc);
1605 lfm = info->lti_buf.lb_buf;
1606 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN)
1609 if (unlikely(rc != sizeof(*lmv1)))
1610 RETURN(rc = rc > 0 ? -EINVAL : rc);
1612 lmv1 = info->lti_buf.lb_buf;
1613 /* The on-disk LMV EA only contains header, but the
1614 * returned LMV EA size should contain the space for
1615 * the FIDs of all shards of the striped directory. */
1616 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1617 rc = lmv_mds_md_size(
1618 le32_to_cpu(lmv1->lmv_stripe_count),
1619 le32_to_cpu(lmv1->lmv_magic));
1622 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1625 if (rc != sizeof(*lmv1))
1626 RETURN(rc = rc > 0 ? -EINVAL : rc);
1628 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1632 RETURN(rc = rc1 != 0 ? rc1 : rc);
1635 if ((rc > 0) && buf->lb_buf && strcmp(name, XATTR_NAME_LOV) == 0) {
1636 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1638 if (lcm->lcm_magic == cpu_to_le32(LOV_MAGIC_SEL))
1639 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
1642 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1646 * XXX: Only used by lfsck
1648 * lod returns default striping on the real root of the device
1649 * this is like the root stores default striping for the whole
1650 * filesystem. historically we've been using a different approach
1651 * and store it in the config.
1653 dt_root_get(env, dev->lod_child, &info->lti_fid);
1654 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1656 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1657 struct lov_user_md *lum = buf->lb_buf;
1658 struct lov_desc *desc = &dev->lod_ost_descs.ltd_lov_desc;
1660 if (buf->lb_buf == NULL) {
1662 } else if (buf->lb_len >= sizeof(*lum)) {
1663 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1664 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1665 lmm_oi_set_id(&lum->lmm_oi, 0);
1666 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1667 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1668 lum->lmm_stripe_size = cpu_to_le32(
1669 desc->ld_default_stripe_size);
1670 lum->lmm_stripe_count = cpu_to_le16(
1671 desc->ld_default_stripe_count);
1672 lum->lmm_stripe_offset = cpu_to_le16(
1673 desc->ld_default_stripe_offset);
1686 * Checks that the magic of the stripe is sane.
1688 * \param[in] lod lod device
1689 * \param[in] lum a buffer storing LMV EA to verify
1691 * \retval 0 if the EA is sane
1692 * \retval negative otherwise
1694 static int lod_verify_md_striping(struct lod_device *lod,
1695 const struct lmv_user_md_v1 *lum)
1697 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1698 CERROR("%s: invalid lmv_user_md: magic = %x, "
1699 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1700 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1701 (int)le32_to_cpu(lum->lum_stripe_offset),
1702 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1710 * Initialize LMV EA for a slave.
1712 * Initialize slave's LMV EA from the master's LMV EA.
1714 * \param[in] master_lmv a buffer containing master's EA
1715 * \param[out] slave_lmv a buffer where slave's EA will be stored
1718 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1719 const struct lmv_mds_md_v1 *master_lmv)
1721 *slave_lmv = *master_lmv;
1722 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1728 * Generate LMV EA from the object passed as \a dt. The object must have
1729 * the stripes created and initialized.
1731 * \param[in] env execution environment
1732 * \param[in] dt object
1733 * \param[out] lmv_buf buffer storing generated LMV EA
1735 * \retval 0 on success
1736 * \retval negative if failed
1738 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1739 struct lu_buf *lmv_buf)
1741 struct lod_thread_info *info = lod_env_info(env);
1742 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1743 struct lod_object *lo = lod_dt_obj(dt);
1744 struct lmv_mds_md_v1 *lmm1;
1746 int type = LU_SEQ_RANGE_ANY;
1751 LASSERT(lo->ldo_dir_striped != 0);
1752 LASSERT(lo->ldo_dir_stripe_count > 0);
1753 stripe_count = lo->ldo_dir_stripe_count;
1754 /* Only store the LMV EA heahder on the disk. */
1755 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1756 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1760 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1763 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1764 memset(lmm1, 0, sizeof(*lmm1));
1765 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1766 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1767 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1768 lmm1->lmv_layout_version = cpu_to_le32(lo->ldo_dir_layout_version);
1769 if (lod_is_layout_changing(lo)) {
1770 lmm1->lmv_migrate_hash = cpu_to_le32(lo->ldo_dir_migrate_hash);
1771 lmm1->lmv_migrate_offset =
1772 cpu_to_le32(lo->ldo_dir_migrate_offset);
1774 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1779 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1780 lmv_buf->lb_buf = info->lti_ea_store;
1781 lmv_buf->lb_len = sizeof(*lmm1);
1787 * Create in-core represenation for a striped directory.
1789 * Parse the buffer containing LMV EA and instantiate LU objects
1790 * representing the stripe objects. The pointers to the objects are
1791 * stored in ldo_stripe field of \a lo. This function is used when
1792 * we need to access an already created object (i.e. load from a disk).
1794 * \param[in] env execution environment
1795 * \param[in] lo lod object
1796 * \param[in] buf buffer containing LMV EA
1798 * \retval 0 on success
1799 * \retval negative if failed
1801 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1802 const struct lu_buf *buf)
1804 struct lod_thread_info *info = lod_env_info(env);
1805 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1806 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1807 struct dt_object **stripe;
1808 union lmv_mds_md *lmm = buf->lb_buf;
1809 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1810 struct lu_fid *fid = &info->lti_fid;
1815 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1817 /* XXX may be useless as not called for foreign LMV ?? */
1818 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_FOREIGN)
1821 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1822 lo->ldo_dir_slave_stripe = 1;
1826 if (!lmv_is_sane(lmv1))
1829 LASSERT(lo->ldo_stripe == NULL);
1830 OBD_ALLOC_PTR_ARRAY(stripe, le32_to_cpu(lmv1->lmv_stripe_count));
1834 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1835 struct dt_device *tgt_dt;
1836 struct dt_object *dto;
1837 int type = LU_SEQ_RANGE_ANY;
1840 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1841 if (!fid_is_sane(fid)) {
1846 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1850 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1851 tgt_dt = lod->lod_child;
1853 struct lod_tgt_desc *tgt;
1855 tgt = LTD_TGT(ltd, idx);
1857 GOTO(out, rc = -ESTALE);
1858 tgt_dt = tgt->ltd_tgt;
1861 dto = dt_locate_at(env, tgt_dt, fid,
1862 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1865 GOTO(out, rc = PTR_ERR(dto));
1870 lo->ldo_stripe = stripe;
1871 lo->ldo_is_foreign = 0;
1872 lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1873 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1874 lo->ldo_dir_layout_version = le32_to_cpu(lmv1->lmv_layout_version);
1875 lo->ldo_dir_migrate_offset = le32_to_cpu(lmv1->lmv_migrate_offset);
1876 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv1->lmv_migrate_hash);
1877 lo->ldo_dir_hash_type = le32_to_cpu(lmv1->lmv_hash_type);
1879 lod_striping_free_nolock(env, lo);
1885 * Declare create a striped directory.
1887 * Declare creating a striped directory with a given stripe pattern on the
1888 * specified MDTs. A striped directory is represented as a regular directory
1889 * - an index listing all the stripes. The stripes point back to the master
1890 * object with ".." and LinkEA. The master object gets LMV EA which
1891 * identifies it as a striped directory. The function allocates FIDs
1894 * \param[in] env execution environment
1895 * \param[in] dt object
1896 * \param[in] attr attributes to initialize the objects with
1897 * \param[in] dof type of objects to be created
1898 * \param[in] th transaction handle
1900 * \retval 0 on success
1901 * \retval negative if failed
1903 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1904 struct dt_object *dt,
1905 struct lu_attr *attr,
1906 struct dt_object_format *dof,
1909 struct lod_thread_info *info = lod_env_info(env);
1910 struct lu_buf lmv_buf;
1911 struct lu_buf slave_lmv_buf;
1912 struct lmv_mds_md_v1 *lmm;
1913 struct lmv_mds_md_v1 *slave_lmm = NULL;
1914 struct dt_insert_rec *rec = &info->lti_dt_rec;
1915 struct lod_object *lo = lod_dt_obj(dt);
1920 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1923 lmm = lmv_buf.lb_buf;
1925 OBD_ALLOC_PTR(slave_lmm);
1926 if (slave_lmm == NULL)
1927 GOTO(out, rc = -ENOMEM);
1929 lod_prep_slave_lmv_md(slave_lmm, lmm);
1930 slave_lmv_buf.lb_buf = slave_lmm;
1931 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1933 if (!dt_try_as_dir(env, dt_object_child(dt)))
1934 GOTO(out, rc = -EINVAL);
1936 rec->rec_type = S_IFDIR;
1937 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1938 struct dt_object *dto = lo->ldo_stripe[i];
1939 char *stripe_name = info->lti_key;
1940 struct lu_name *sname;
1941 struct linkea_data ldata = { NULL };
1942 struct lu_buf linkea_buf;
1944 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
1948 /* directory split skip create for existing stripes */
1949 if (!(lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) {
1950 rc = lod_sub_declare_create(env, dto, attr, NULL, dof,
1955 if (!dt_try_as_dir(env, dto))
1956 GOTO(out, rc = -EINVAL);
1958 rc = lod_sub_declare_ref_add(env, dto, th);
1962 rec->rec_fid = lu_object_fid(&dto->do_lu);
1963 rc = lod_sub_declare_insert(env, dto,
1964 (const struct dt_rec *)rec,
1965 (const struct dt_key *)dot,
1970 /* master stripe FID will be put to .. */
1971 rec->rec_fid = lu_object_fid(&dt->do_lu);
1972 rc = lod_sub_declare_insert(env, dto,
1973 (const struct dt_rec *)rec,
1974 (const struct dt_key *)dotdot,
1979 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1981 snprintf(stripe_name, sizeof(info->lti_key),
1983 PFID(lu_object_fid(&dto->do_lu)),
1986 snprintf(stripe_name, sizeof(info->lti_key),
1988 PFID(lu_object_fid(&dto->do_lu)), i);
1990 sname = lod_name_get(env, stripe_name,
1991 strlen(stripe_name));
1992 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1993 sname, lu_object_fid(&dt->do_lu));
1997 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1998 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1999 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
2000 XATTR_NAME_LINK, 0, th);
2004 rec->rec_fid = lu_object_fid(&dto->do_lu);
2005 rc = lod_sub_declare_insert(env, dt_object_child(dt),
2006 (const struct dt_rec *)rec,
2007 (const struct dt_key *)stripe_name, th);
2011 rc = lod_sub_declare_ref_add(env, dt_object_child(dt),
2017 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
2018 cfs_fail_val != i) {
2019 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
2021 slave_lmm->lmv_master_mdt_index =
2024 slave_lmm->lmv_master_mdt_index =
2026 rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
2027 XATTR_NAME_LMV, 0, th);
2033 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
2034 &lmv_buf, XATTR_NAME_LMV, 0, th);
2038 if (slave_lmm != NULL)
2039 OBD_FREE_PTR(slave_lmm);
2045 * Allocate a striping on a predefined set of MDTs.
2047 * Allocates new striping using the MDT index range provided by the data from
2048 * the lum_obejcts contained in the lmv_user_md passed to this method if
2049 * \a is_specific is true; or allocates new layout starting from MDT index in
2050 * lo->ldo_dir_stripe_offset. The exact order of MDTs is not important and
2051 * varies depending on MDT status. The number of stripes needed and stripe
2052 * offset are taken from the object. If that number cannot be met, then the
2053 * function returns an error and then it's the caller's responsibility to
2054 * release the stripes allocated. All the internal structures are protected,
2055 * but no concurrent allocation is allowed on the same objects.
2057 * \param[in] env execution environment for this thread
2058 * \param[in] lo LOD object
2059 * \param[out] stripes striping created
2060 * \param[out] mdt_indices MDT indices of striping created
2061 * \param[in] is_specific true if the MDTs are provided by lum; false if
2062 * only the starting MDT index is provided
2064 * \retval positive stripes allocated, including the first stripe allocated
2066 * \retval negative errno on failure
2068 static int lod_mdt_alloc_specific(const struct lu_env *env,
2069 struct lod_object *lo,
2070 struct dt_object **stripes,
2071 __u32 *mdt_indices, bool is_specific)
2073 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
2074 struct lu_tgt_descs *ltd = &lod->lod_mdt_descs;
2075 struct lu_tgt_desc *tgt = NULL;
2076 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2077 struct dt_device *tgt_dt = NULL;
2078 struct lu_fid fid = { 0 };
2079 struct dt_object *dto;
2081 u32 stripe_count = lo->ldo_dir_stripe_count;
2087 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
2088 if (!is_specific && stripe_count > 1)
2089 /* Set the start index for the 2nd stripe allocation */
2090 mdt_indices[1] = (mdt_indices[0] + 1) %
2091 (lod->lod_remote_mdt_count + 1);
2093 for (; stripe_idx < stripe_count; stripe_idx++) {
2094 /* Try to find next avaible target */
2095 idx = mdt_indices[stripe_idx];
2096 for (j = 0; j < lod->lod_remote_mdt_count;
2097 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
2098 bool already_allocated = false;
2101 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
2102 idx, lod->lod_remote_mdt_count + 1, stripe_idx);
2104 if (likely(!is_specific &&
2105 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
2106 /* check whether the idx already exists
2107 * in current allocated array */
2108 for (k = 0; k < stripe_idx; k++) {
2109 if (mdt_indices[k] == idx) {
2110 already_allocated = true;
2115 if (already_allocated)
2119 /* Sigh, this index is not in the bitmap, let's check
2120 * next available target */
2121 if (!test_bit(idx, ltd->ltd_tgt_bitmap) &&
2122 idx != master_index)
2125 if (idx == master_index) {
2126 /* Allocate the FID locally */
2127 tgt_dt = lod->lod_child;
2128 rc = dt_fid_alloc(env, tgt_dt, &fid, NULL,
2135 /* check the status of the OSP */
2136 tgt = LTD_TGT(ltd, idx);
2140 tgt_dt = tgt->ltd_tgt;
2141 if (!tgt->ltd_active)
2142 /* this OSP doesn't feel well */
2145 rc = dt_fid_alloc(env, tgt_dt, &fid, NULL, NULL);
2152 /* Can not allocate more stripes */
2153 if (j == lod->lod_remote_mdt_count) {
2154 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
2155 lod2obd(lod)->obd_name, stripe_count,
2160 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
2161 idx, stripe_idx, PFID(&fid));
2162 mdt_indices[stripe_idx] = idx;
2163 /* Set the start index for next stripe allocation */
2164 if (!is_specific && stripe_idx < stripe_count - 1) {
2166 * for large dir test, put all other slaves on one
2167 * remote MDT, otherwise we may save too many local
2168 * slave locks which will exceed RS_MAX_LOCKS.
2170 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)))
2172 mdt_indices[stripe_idx + 1] = (idx + 1) %
2173 (lod->lod_remote_mdt_count + 1);
2175 /* tgt_dt and fid must be ready after search avaible OSP
2176 * in the above loop */
2177 LASSERT(tgt_dt != NULL);
2178 LASSERT(fid_is_sane(&fid));
2180 /* fail a remote stripe FID allocation */
2181 if (stripe_idx && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_FID))
2184 dto = dt_locate_at(env, tgt_dt, &fid,
2185 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
2192 stripes[stripe_idx] = dto;
2198 for (j = 1; j < stripe_idx; j++) {
2199 LASSERT(stripes[j] != NULL);
2200 dt_object_put(env, stripes[j]);
2206 static int lod_prep_md_striped_create(const struct lu_env *env,
2207 struct dt_object *dt,
2208 struct lu_attr *attr,
2209 const struct lmv_user_md_v1 *lum,
2210 struct dt_object_format *dof,
2213 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
2214 struct lod_object *lo = lod_dt_obj(dt);
2215 struct dt_object **stripes;
2216 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2217 struct lu_fid fid = { 0 };
2224 /* The lum has been verifed in lod_verify_md_striping */
2225 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
2226 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
2228 stripe_count = lo->ldo_dir_stripe_count;
2230 OBD_ALLOC_PTR_ARRAY(stripes, stripe_count);
2234 /* Allocate the first stripe locally */
2235 rc = dt_fid_alloc(env, lod->lod_child, &fid, NULL, NULL);
2239 stripes[0] = dt_locate_at(env, lod->lod_child, &fid,
2240 dt->do_lu.lo_dev->ld_site->ls_top_dev, &conf);
2241 if (IS_ERR(stripes[0]))
2242 GOTO(out, rc = PTR_ERR(stripes[0]));
2244 if (lo->ldo_dir_stripe_offset == LMV_OFFSET_DEFAULT) {
2245 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
2246 rc = lod_mdt_alloc_qos(env, lo, stripes, 1, stripe_count);
2248 rc = lod_mdt_alloc_rr(env, lo, stripes, 1,
2252 bool is_specific = false;
2254 OBD_ALLOC_PTR_ARRAY(idx_array, stripe_count);
2256 GOTO(out, rc = -ENOMEM);
2258 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
2260 for (i = 0; i < stripe_count; i++)
2262 le32_to_cpu(lum->lum_objects[i].lum_mds);
2265 /* stripe 0 is local */
2267 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
2268 rc = lod_mdt_alloc_specific(env, lo, stripes, idx_array,
2270 OBD_FREE_PTR_ARRAY(idx_array, stripe_count);
2278 lo->ldo_dir_striped = 1;
2279 lo->ldo_stripe = stripes;
2280 lo->ldo_dir_stripe_count = rc;
2281 lo->ldo_dir_stripes_allocated = stripe_count;
2283 lo->ldo_dir_stripe_loaded = 1;
2285 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2287 lod_striping_free(env, lo);
2293 if (!IS_ERR_OR_NULL(stripes[0]))
2294 dt_object_put(env, stripes[0]);
2295 for (i = 1; i < stripe_count; i++)
2296 LASSERT(!stripes[i]);
2297 OBD_FREE_PTR_ARRAY(stripes, stripe_count);
2304 * Alloc cached foreign LOV
2306 * \param[in] lo object
2307 * \param[in] size size of foreign LOV
2309 * \retval 0 on success
2310 * \retval negative if failed
2312 int lod_alloc_foreign_lov(struct lod_object *lo, size_t size)
2314 OBD_ALLOC_LARGE(lo->ldo_foreign_lov, size);
2315 if (lo->ldo_foreign_lov == NULL)
2317 lo->ldo_foreign_lov_size = size;
2318 lo->ldo_is_foreign = 1;
2324 * Free cached foreign LOV
2326 * \param[in] lo object
2328 void lod_free_foreign_lov(struct lod_object *lo)
2330 if (lo->ldo_foreign_lov != NULL)
2331 OBD_FREE_LARGE(lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
2332 lo->ldo_foreign_lov = NULL;
2333 lo->ldo_foreign_lov_size = 0;
2334 lo->ldo_is_foreign = 0;
2339 * Alloc cached foreign LMV
2341 * \param[in] lo object
2342 * \param[in] size size of foreign LMV
2344 * \retval 0 on success
2345 * \retval negative if failed
2347 int lod_alloc_foreign_lmv(struct lod_object *lo, size_t size)
2349 OBD_ALLOC_LARGE(lo->ldo_foreign_lmv, size);
2350 if (lo->ldo_foreign_lmv == NULL)
2352 lo->ldo_foreign_lmv_size = size;
2353 lo->ldo_is_foreign = 1;
2360 * Free cached foreign LMV
2362 * \param[in] lo object
2364 void lod_free_foreign_lmv(struct lod_object *lo)
2366 if (lo->ldo_foreign_lmv != NULL)
2367 OBD_FREE_LARGE(lo->ldo_foreign_lmv, lo->ldo_foreign_lmv_size);
2368 lo->ldo_foreign_lmv = NULL;
2369 lo->ldo_foreign_lmv_size = 0;
2370 lo->ldo_is_foreign = 0;
2374 * Declare create striped md object.
2376 * The function declares intention to create a striped directory. This is a
2377 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2378 * is to verify pattern \a lum_buf is good. Check that function for the details.
2380 * \param[in] env execution environment
2381 * \param[in] dt object
2382 * \param[in] attr attributes to initialize the objects with
2383 * \param[in] lum_buf a pattern specifying the number of stripes and
2385 * \param[in] dof type of objects to be created
2386 * \param[in] th transaction handle
2388 * \retval 0 on success
2389 * \retval negative if failed
2392 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2393 struct dt_object *dt,
2394 struct lu_attr *attr,
2395 const struct lu_buf *lum_buf,
2396 struct dt_object_format *dof,
2399 struct lod_object *lo = lod_dt_obj(dt);
2400 struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
2404 LASSERT(lum != NULL);
2406 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
2407 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
2408 (int)le32_to_cpu(lum->lum_stripe_offset));
2410 if (lo->ldo_dir_stripe_count == 0) {
2411 if (lo->ldo_is_foreign) {
2412 rc = lod_alloc_foreign_lmv(lo, lum_buf->lb_len);
2415 memcpy(lo->ldo_foreign_lmv, lum, lum_buf->lb_len);
2416 lo->ldo_dir_stripe_loaded = 1;
2421 /* prepare dir striped objects */
2422 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2424 /* failed to create striping, let's reset
2425 * config so that others don't get confused */
2426 lod_striping_free(env, lo);
2434 * Set or replace striped directory layout, and LFSCK may set layout on a plain
2435 * directory, so don't check stripe count.
2437 * \param[in] env execution environment
2438 * \param[in] dt target object
2439 * \param[in] buf LMV buf which contains source stripe fids
2440 * \param[in] fl set or replace
2441 * \param[in] th transaction handle
2443 * \retval 0 on success
2444 * \retval negative if failed
2446 static int lod_dir_layout_set(const struct lu_env *env,
2447 struct dt_object *dt,
2448 const struct lu_buf *buf,
2452 struct dt_object *next = dt_object_child(dt);
2453 struct lod_object *lo = lod_dt_obj(dt);
2454 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2455 struct lmv_mds_md_v1 *lmv = buf->lb_buf;
2456 struct lmv_mds_md_v1 *slave_lmv;
2457 struct lu_buf slave_buf;
2463 if (!lmv_is_sane2(lmv))
2466 /* adjust hash for dir merge, which may not be set in user command */
2467 if (lmv_is_merging(lmv) &&
2468 !(lmv->lmv_migrate_hash & LMV_HASH_TYPE_MASK))
2469 lmv->lmv_merge_hash |=
2470 lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern &
2473 LMV_DEBUG(D_INFO, lmv, "set");
2475 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LMV, fl, th);
2479 /* directory restripe may update stripe LMV directly */
2480 if (!lo->ldo_dir_stripe_count)
2483 lo->ldo_dir_hash_type = le32_to_cpu(lmv->lmv_hash_type);
2484 lo->ldo_dir_migrate_offset = le32_to_cpu(lmv->lmv_migrate_offset);
2485 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_migrate_hash);
2486 lo->ldo_dir_layout_version = le32_to_cpu(lmv->lmv_layout_version);
2488 OBD_ALLOC_PTR(slave_lmv);
2492 lod_prep_slave_lmv_md(slave_lmv, lmv);
2493 slave_buf.lb_buf = slave_lmv;
2494 slave_buf.lb_len = sizeof(*slave_lmv);
2496 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2497 if (!lo->ldo_stripe[i])
2500 if (!dt_object_exists(lo->ldo_stripe[i]))
2503 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], &slave_buf,
2504 XATTR_NAME_LMV, fl, th);
2509 OBD_FREE_PTR(slave_lmv);
2515 * Implementation of dt_object_operations::do_declare_xattr_set.
2517 * Used with regular (non-striped) objects. Basically it
2518 * initializes the striping information and applies the
2519 * change to all the stripes.
2521 * \see dt_object_operations::do_declare_xattr_set() in the API description
2524 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2525 struct dt_object *dt,
2526 const struct lu_buf *buf,
2527 const char *name, int fl,
2530 struct dt_object *next = dt_object_child(dt);
2531 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2532 struct lod_object *lo = lod_dt_obj(dt);
2537 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2538 struct lmv_user_md_v1 *lum;
2540 LASSERT(buf != NULL && buf->lb_buf != NULL);
2542 rc = lod_verify_md_striping(d, lum);
2545 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2546 rc = lod_verify_striping(env, d, lo, buf, false);
2551 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2555 /* Note: Do not set LinkEA on sub-stripes, otherwise
2556 * it will confuse the fid2path process(see mdt_path_current()).
2557 * The linkEA between master and sub-stripes is set in
2558 * lod_xattr_set_lmv(). */
2559 if (strcmp(name, XATTR_NAME_LINK) == 0)
2562 /* set xattr to each stripes, if needed */
2563 rc = lod_striping_load(env, lo);
2567 if (lo->ldo_dir_stripe_count == 0)
2570 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2571 if (!lo->ldo_stripe[i])
2574 if (!dt_object_exists(lo->ldo_stripe[i]))
2577 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2587 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2588 struct lod_object *lo,
2589 struct dt_object *dt, struct thandle *th,
2590 int comp_idx, int stripe_idx,
2591 struct lod_obj_stripe_cb_data *data)
2593 struct lod_thread_info *info = lod_env_info(env);
2594 struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
2595 struct filter_fid *ff = &info->lti_ff;
2596 struct lu_buf *buf = &info->lti_buf;
2600 buf->lb_len = sizeof(*ff);
2601 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2609 * locd_buf is set if it's called by dir migration, which doesn't check
2612 if (data->locd_buf) {
2613 memset(ff, 0, sizeof(*ff));
2614 ff->ff_parent = *(struct lu_fid *)data->locd_buf->lb_buf;
2616 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
2618 if (lu_fid_eq(lod_object_fid(lo), &ff->ff_parent) &&
2619 ff->ff_layout.ol_comp_id == comp->llc_id)
2622 memset(ff, 0, sizeof(*ff));
2623 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2626 /* rewrite filter_fid */
2627 ff->ff_parent.f_ver = stripe_idx;
2628 ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
2629 ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
2630 ff->ff_layout.ol_comp_id = comp->llc_id;
2631 ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
2632 ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
2633 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
2635 if (data->locd_declare)
2636 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2637 LU_XATTR_REPLACE, th);
2639 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2640 LU_XATTR_REPLACE, th);
2646 * Reset parent FID on OST object
2648 * Replace parent FID with @dt object FID, which is only called during migration
2649 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2650 * the FID is changed.
2652 * \param[in] env execution environment
2653 * \param[in] dt dt_object whose stripes's parent FID will be reset
2654 * \parem[in] th thandle
2655 * \param[in] declare if it is declare
2657 * \retval 0 if reset succeeds
2658 * \retval negative errno if reset fails
2660 static int lod_replace_parent_fid(const struct lu_env *env,
2661 struct dt_object *dt,
2662 const struct lu_buf *buf,
2663 struct thandle *th, bool declare)
2665 struct lod_object *lo = lod_dt_obj(dt);
2666 struct lod_thread_info *info = lod_env_info(env);
2667 struct filter_fid *ff;
2668 struct lod_obj_stripe_cb_data data = { { 0 } };
2672 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2674 /* set xattr to each stripes, if needed */
2675 rc = lod_striping_load(env, lo);
2679 if (!lod_obj_is_striped(dt))
2682 if (info->lti_ea_store_size < sizeof(*ff)) {
2683 rc = lod_ea_store_resize(info, sizeof(*ff));
2688 data.locd_declare = declare;
2689 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2690 data.locd_buf = buf;
2691 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2696 __u16 lod_comp_entry_stripe_count(struct lod_object *lo,
2697 int comp_idx, bool is_dir)
2699 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2700 struct lod_layout_component *entry;
2705 entry = &lo->ldo_comp_entries[comp_idx];
2706 if (lod_comp_inited(entry))
2707 return entry->llc_stripe_count;
2708 else if ((__u16)-1 == entry->llc_stripe_count)
2709 return lod->lod_ost_count;
2711 return lod_get_stripe_count(lod, lo, comp_idx,
2712 entry->llc_stripe_count,
2713 entry->llc_pattern &
2714 LOV_PATTERN_OVERSTRIPING);
2717 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2719 int magic, size = 0, i;
2720 struct lod_layout_component *comp_entries;
2722 bool is_composite, is_foreign = false;
2725 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2726 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2728 lo->ldo_def_striping->lds_def_striping_is_composite;
2730 comp_cnt = lo->ldo_comp_cnt;
2731 comp_entries = lo->ldo_comp_entries;
2732 is_composite = lo->ldo_is_composite;
2733 is_foreign = lo->ldo_is_foreign;
2737 return lo->ldo_foreign_lov_size;
2739 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2741 size = sizeof(struct lov_comp_md_v1) +
2742 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2743 LASSERT(size % sizeof(__u64) == 0);
2746 for (i = 0; i < comp_cnt; i++) {
2749 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2750 stripe_count = lod_comp_entry_stripe_count(lo, i, is_dir);
2751 if (!is_dir && is_composite)
2752 lod_comp_shrink_stripe_count(&comp_entries[i],
2755 size += lov_user_md_size(stripe_count, magic);
2756 LASSERT(size % sizeof(__u64) == 0);
2762 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2763 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2766 * \param[in] env execution environment
2767 * \param[in] dt dt_object to add components on
2768 * \param[in] buf buffer contains components to be added
2769 * \parem[in] th thandle
2771 * \retval 0 on success
2772 * \retval negative errno on failure
2774 static int lod_declare_layout_add(const struct lu_env *env,
2775 struct dt_object *dt,
2776 const struct lu_buf *buf,
2779 struct lod_thread_info *info = lod_env_info(env);
2780 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2781 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2782 struct dt_object *next = dt_object_child(dt);
2783 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
2784 struct lod_object *lo = lod_dt_obj(dt);
2785 struct lov_user_md_v3 *v3;
2786 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2788 int i, rc, array_cnt, old_array_cnt;
2791 LASSERT(lo->ldo_is_composite);
2793 if (lo->ldo_flr_state != LCM_FL_NONE)
2796 rc = lod_verify_striping(env, d, lo, buf, false);
2800 magic = comp_v1->lcm_magic;
2801 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2802 lustre_swab_lov_comp_md_v1(comp_v1);
2803 magic = comp_v1->lcm_magic;
2806 if (magic != LOV_USER_MAGIC_COMP_V1)
2809 mutex_lock(&lo->ldo_layout_mutex);
2811 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2812 OBD_ALLOC_PTR_ARRAY(comp_array, array_cnt);
2813 if (comp_array == NULL) {
2814 mutex_unlock(&lo->ldo_layout_mutex);
2819 memcpy(comp_array, lo->ldo_comp_entries,
2820 sizeof(*comp_array) * lo->ldo_comp_cnt);
2822 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2823 struct lov_user_md_v1 *v1;
2824 struct lu_extent *ext;
2826 v1 = (struct lov_user_md *)((char *)comp_v1 +
2827 comp_v1->lcm_entries[i].lcme_offset);
2828 ext = &comp_v1->lcm_entries[i].lcme_extent;
2830 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2831 lod_comp->llc_extent.e_start = ext->e_start;
2832 lod_comp->llc_extent.e_end = ext->e_end;
2833 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2834 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2836 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2837 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2838 lod_adjust_stripe_info(lod_comp, desc, 0);
2840 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2841 v3 = (struct lov_user_md_v3 *) v1;
2842 if (v3->lmm_pool_name[0] != '\0') {
2843 rc = lod_set_pool(&lod_comp->llc_pool,
2851 old_array = lo->ldo_comp_entries;
2852 old_array_cnt = lo->ldo_comp_cnt;
2854 lo->ldo_comp_entries = comp_array;
2855 lo->ldo_comp_cnt = array_cnt;
2857 /* No need to increase layout generation here, it will be increased
2858 * later when generating component ID for the new components */
2860 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2861 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2862 XATTR_NAME_LOV, 0, th);
2864 lo->ldo_comp_entries = old_array;
2865 lo->ldo_comp_cnt = old_array_cnt;
2869 OBD_FREE_PTR_ARRAY(old_array, old_array_cnt);
2871 LASSERT(lo->ldo_mirror_count == 1);
2872 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2874 mutex_unlock(&lo->ldo_layout_mutex);
2879 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2880 lod_comp = &comp_array[i];
2881 if (lod_comp->llc_pool != NULL) {
2882 OBD_FREE(lod_comp->llc_pool,
2883 strlen(lod_comp->llc_pool) + 1);
2884 lod_comp->llc_pool = NULL;
2887 OBD_FREE_PTR_ARRAY(comp_array, array_cnt);
2888 mutex_unlock(&lo->ldo_layout_mutex);
2894 * lod_last_non_stale_mirror() - Check if a mirror is the last non-stale mirror.
2895 * @mirror_id: Mirror id to be checked.
2898 * This function checks if a mirror with specified @mirror_id is the last
2899 * non-stale mirror of a LOD object @lo.
2901 * Return: true or false.
2904 bool lod_last_non_stale_mirror(__u16 mirror_id, struct lod_object *lo)
2906 struct lod_layout_component *lod_comp;
2907 bool has_stale_flag;
2910 for (i = 0; i < lo->ldo_mirror_count; i++) {
2911 if (lo->ldo_mirrors[i].lme_id == mirror_id ||
2912 lo->ldo_mirrors[i].lme_stale)
2915 has_stale_flag = false;
2916 lod_foreach_mirror_comp(lod_comp, lo, i) {
2917 if (lod_comp->llc_flags & LCME_FL_STALE) {
2918 has_stale_flag = true;
2922 if (!has_stale_flag)
2930 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2931 * the '$field' can only be 'flags' now. The xattr value is binary
2932 * lov_comp_md_v1 which contains the component ID(s) and the value of
2933 * the field to be modified.
2934 * Please update allowed_lustre_lov macro if $field groks more values
2937 * \param[in] env execution environment
2938 * \param[in] dt dt_object to be modified
2939 * \param[in] op operation string, like "set.flags"
2940 * \param[in] buf buffer contains components to be set
2941 * \parem[in] th thandle
2943 * \retval 0 on success
2944 * \retval negative errno on failure
2946 static int lod_declare_layout_set(const struct lu_env *env,
2947 struct dt_object *dt,
2948 char *op, const struct lu_buf *buf,
2951 struct lod_layout_component *lod_comp;
2952 struct lod_thread_info *info = lod_env_info(env);
2953 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2954 struct lod_object *lo = lod_dt_obj(dt);
2955 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2958 bool changed = false;
2961 /* Please update allowed_lustre_lov macro if op
2962 * groks more values in the future
2964 if (strcmp(op, "set.flags") != 0) {
2965 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2966 lod2obd(d)->obd_name, op);
2970 magic = comp_v1->lcm_magic;
2971 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2972 lustre_swab_lov_comp_md_v1(comp_v1);
2973 magic = comp_v1->lcm_magic;
2976 if (magic != LOV_USER_MAGIC_COMP_V1)
2979 if (comp_v1->lcm_entry_count == 0) {
2980 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2981 lod2obd(d)->obd_name);
2985 mutex_lock(&lo->ldo_layout_mutex);
2986 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2987 __u32 id = comp_v1->lcm_entries[i].lcme_id;
2988 __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
2989 __u32 mirror_flag = flags & LCME_MIRROR_FLAGS;
2990 __u16 mirror_id = mirror_id_of(id);
2991 bool neg = flags & LCME_FL_NEG;
2993 if (flags & LCME_FL_INIT) {
2995 lod_striping_free_nolock(env, lo);
2996 mutex_unlock(&lo->ldo_layout_mutex);
3000 flags &= ~(LCME_MIRROR_FLAGS | LCME_FL_NEG);
3001 for (j = 0; j < lo->ldo_comp_cnt; j++) {
3002 lod_comp = &lo->ldo_comp_entries[j];
3004 /* lfs only put one flag in each entry */
3005 if ((flags && id != lod_comp->llc_id) ||
3006 (mirror_flag && mirror_id !=
3007 mirror_id_of(lod_comp->llc_id)))
3012 lod_comp->llc_flags &= ~flags;
3014 lod_comp->llc_flags &= ~mirror_flag;
3017 if ((flags & LCME_FL_STALE) &&
3018 lod_last_non_stale_mirror(mirror_id,
3021 &lo->ldo_layout_mutex);
3024 lod_comp->llc_flags |= flags;
3027 lod_comp->llc_flags |= mirror_flag;
3028 if (mirror_flag & LCME_FL_NOSYNC)
3029 lod_comp->llc_timestamp =
3030 ktime_get_real_seconds();
3036 mutex_unlock(&lo->ldo_layout_mutex);
3039 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
3040 lod2obd(d)->obd_name);
3044 lod_obj_inc_layout_gen(lo);
3046 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3047 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
3048 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3053 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
3054 * and the xattr value is a unique component ID or a special lcme_id.
3056 * \param[in] env execution environment
3057 * \param[in] dt dt_object to be operated on
3058 * \param[in] buf buffer contains component ID or lcme_id
3059 * \parem[in] th thandle
3061 * \retval 0 on success
3062 * \retval negative errno on failure
3064 static int lod_declare_layout_del(const struct lu_env *env,
3065 struct dt_object *dt,
3066 const struct lu_buf *buf,
3069 struct lod_thread_info *info = lod_env_info(env);
3070 struct dt_object *next = dt_object_child(dt);
3071 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3072 struct lod_object *lo = lod_dt_obj(dt);
3073 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3074 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3075 __u32 magic, id, flags, neg_flags = 0;
3079 LASSERT(lo->ldo_is_composite);
3081 if (lo->ldo_flr_state != LCM_FL_NONE)
3084 magic = comp_v1->lcm_magic;
3085 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
3086 lustre_swab_lov_comp_md_v1(comp_v1);
3087 magic = comp_v1->lcm_magic;
3090 if (magic != LOV_USER_MAGIC_COMP_V1)
3093 id = comp_v1->lcm_entries[0].lcme_id;
3094 flags = comp_v1->lcm_entries[0].lcme_flags;
3096 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
3097 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
3098 lod2obd(d)->obd_name, id, flags);
3102 if (id != LCME_ID_INVAL && flags != 0) {
3103 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
3104 lod2obd(d)->obd_name);
3108 if (id == LCME_ID_INVAL && !flags) {
3109 CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
3110 lod2obd(d)->obd_name);
3114 if (flags & LCME_FL_NEG) {
3115 neg_flags = flags & ~LCME_FL_NEG;
3119 mutex_lock(&lo->ldo_layout_mutex);
3121 left = lo->ldo_comp_cnt;
3123 mutex_unlock(&lo->ldo_layout_mutex);
3127 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3128 struct lod_layout_component *lod_comp;
3130 lod_comp = &lo->ldo_comp_entries[i];
3132 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
3134 else if (flags && !(flags & lod_comp->llc_flags))
3136 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
3139 if (left != (i + 1)) {
3140 CDEBUG(D_LAYOUT, "%s: this deletion will create "
3141 "a hole.\n", lod2obd(d)->obd_name);
3142 mutex_unlock(&lo->ldo_layout_mutex);
3147 /* Mark the component as deleted */
3148 lod_comp->llc_id = LCME_ID_INVAL;
3150 /* Not instantiated component */
3151 if (lod_comp->llc_stripe == NULL)
3154 LASSERT(lod_comp->llc_stripe_count > 0);
3155 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
3156 struct dt_object *obj = lod_comp->llc_stripe[j];
3160 rc = lod_sub_declare_destroy(env, obj, th);
3162 mutex_unlock(&lo->ldo_layout_mutex);
3168 LASSERTF(left >= 0, "left = %d\n", left);
3169 if (left == lo->ldo_comp_cnt) {
3170 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
3171 lod2obd(d)->obd_name, id);
3172 mutex_unlock(&lo->ldo_layout_mutex);
3176 mutex_unlock(&lo->ldo_layout_mutex);
3178 memset(attr, 0, sizeof(*attr));
3179 attr->la_valid = LA_SIZE;
3180 rc = lod_sub_declare_attr_set(env, next, attr, th);
3185 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3186 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
3187 XATTR_NAME_LOV, 0, th);
3189 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
3196 * Declare layout add/set/del operations issued by special xattr names:
3198 * XATTR_LUSTRE_LOV.add add component(s) to existing file
3199 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
3200 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
3202 * \param[in] env execution environment
3203 * \param[in] dt object
3204 * \param[in] name name of xattr
3205 * \param[in] buf lu_buf contains xattr value
3206 * \param[in] th transaction handle
3208 * \retval 0 on success
3209 * \retval negative if failed
3211 static int lod_declare_modify_layout(const struct lu_env *env,
3212 struct dt_object *dt,
3214 const struct lu_buf *buf,
3217 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3218 struct lod_object *lo = lod_dt_obj(dt);
3220 int rc, len = strlen(XATTR_LUSTRE_LOV);
3223 LASSERT(dt_object_exists(dt));
3225 if (strlen(name) <= len || name[len] != '.') {
3226 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
3227 lod2obd(d)->obd_name, name);
3232 rc = lod_striping_load(env, lo);
3236 /* the layout to be modified must be a composite layout */
3237 if (!lo->ldo_is_composite) {
3238 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
3239 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3240 GOTO(unlock, rc = -EINVAL);
3243 op = (char *)name + len;
3244 if (strcmp(op, "add") == 0) {
3245 rc = lod_declare_layout_add(env, dt, buf, th);
3246 } else if (strcmp(op, "del") == 0) {
3247 rc = lod_declare_layout_del(env, dt, buf, th);
3248 } else if (strncmp(op, "set", strlen("set")) == 0) {
3249 rc = lod_declare_layout_set(env, dt, op, buf, th);
3251 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
3252 lod2obd(d)->obd_name, name);
3253 GOTO(unlock, rc = -ENOTSUPP);
3257 lod_striping_free(env, lo);
3263 * Convert a plain file lov_mds_md to a composite layout.
3265 * \param[in,out] info the thread info::lti_ea_store buffer contains little
3266 * endian plain file layout
3268 * \retval 0 on success, <0 on failure
3270 static int lod_layout_convert(struct lod_thread_info *info)
3272 struct lov_mds_md *lmm = info->lti_ea_store;
3273 struct lov_mds_md *lmm_save;
3274 struct lov_comp_md_v1 *lcm;
3275 struct lov_comp_md_entry_v1 *lcme;
3281 /* realloc buffer to a composite layout which contains one component */
3282 blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3283 le32_to_cpu(lmm->lmm_magic));
3284 size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
3286 OBD_ALLOC_LARGE(lmm_save, blob_size);
3288 GOTO(out, rc = -ENOMEM);
3290 memcpy(lmm_save, lmm, blob_size);
3292 if (info->lti_ea_store_size < size) {
3293 rc = lod_ea_store_resize(info, size);
3298 lcm = info->lti_ea_store;
3299 memset(lcm, 0, sizeof(*lcm) + sizeof(*lcme));
3300 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
3301 lcm->lcm_size = cpu_to_le32(size);
3302 lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
3303 lmm_save->lmm_layout_gen));
3304 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
3305 lcm->lcm_entry_count = cpu_to_le16(1);
3307 lcme = &lcm->lcm_entries[0];
3308 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
3309 lcme->lcme_extent.e_start = 0;
3310 lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
3311 lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
3312 lcme->lcme_size = cpu_to_le32(blob_size);
3314 memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
3319 OBD_FREE_LARGE(lmm_save, blob_size);
3324 * Merge layouts to form a mirrored file.
3326 static int lod_declare_layout_merge(const struct lu_env *env,
3327 struct dt_object *dt, const struct lu_buf *mbuf,
3330 struct lod_thread_info *info = lod_env_info(env);
3331 struct lu_attr *layout_attr = &info->lti_layout_attr;
3332 struct lu_buf *buf = &info->lti_buf;
3333 struct lod_object *lo = lod_dt_obj(dt);
3334 struct lov_comp_md_v1 *lcm;
3335 struct lov_comp_md_v1 *cur_lcm;
3336 struct lov_comp_md_v1 *merge_lcm;
3337 struct lov_comp_md_entry_v1 *lcme;
3338 struct lov_mds_md_v1 *lmm;
3341 __u16 cur_entry_count;
3342 __u16 merge_entry_count;
3344 __u16 mirror_id = 0;
3351 merge_lcm = mbuf->lb_buf;
3352 if (mbuf->lb_len < sizeof(*merge_lcm))
3355 /* must be an existing layout from disk */
3356 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
3359 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
3361 /* do not allow to merge two mirrored files */
3362 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
3365 /* verify the target buffer */
3366 rc = lod_get_lov_ea(env, lo);
3368 RETURN(rc ? : -ENODATA);
3370 cur_lcm = info->lti_ea_store;
3371 switch (le32_to_cpu(cur_lcm->lcm_magic)) {
3374 rc = lod_layout_convert(info);
3376 case LOV_MAGIC_COMP_V1:
3386 /* info->lti_ea_store could be reallocated in lod_layout_convert() */
3387 cur_lcm = info->lti_ea_store;
3388 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
3390 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
3391 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
3392 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
3395 /* size of new layout */
3396 size = le32_to_cpu(cur_lcm->lcm_size) +
3397 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
3399 memset(buf, 0, sizeof(*buf));
3400 lu_buf_alloc(buf, size);
3401 if (buf->lb_buf == NULL)
3405 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
3407 offset = sizeof(*lcm) +
3408 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
3409 for (i = 0; i < cur_entry_count; i++) {
3410 struct lov_comp_md_entry_v1 *cur_lcme;
3412 lcme = &lcm->lcm_entries[i];
3413 cur_lcme = &cur_lcm->lcm_entries[i];
3415 lcme->lcme_offset = cpu_to_le32(offset);
3416 memcpy((char *)lcm + offset,
3417 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
3418 le32_to_cpu(lcme->lcme_size));
3420 offset += le32_to_cpu(lcme->lcme_size);
3422 if (mirror_count == 1 &&
3423 mirror_id_of(le32_to_cpu(lcme->lcme_id)) == 0) {
3424 /* Add mirror from a non-flr file, create new mirror ID.
3425 * Otherwise, keep existing mirror's component ID, used
3426 * for mirror extension.
3428 id = pflr_id(1, i + 1);
3429 lcme->lcme_id = cpu_to_le32(id);
3432 id = max(le32_to_cpu(lcme->lcme_id), id);
3435 mirror_id = mirror_id_of(id) + 1;
3437 /* check if first entry in new layout is DOM */
3438 lmm = (struct lov_mds_md_v1 *)((char *)merge_lcm +
3439 merge_lcm->lcm_entries[0].lcme_offset);
3440 merge_has_dom = lov_pattern(le32_to_cpu(lmm->lmm_pattern)) ==
3443 for (i = 0; i < merge_entry_count; i++) {
3444 struct lov_comp_md_entry_v1 *merge_lcme;
3446 merge_lcme = &merge_lcm->lcm_entries[i];
3447 lcme = &lcm->lcm_entries[cur_entry_count + i];
3449 *lcme = *merge_lcme;
3450 lcme->lcme_offset = cpu_to_le32(offset);
3451 if (merge_has_dom && i == 0)
3452 lcme->lcme_flags |= cpu_to_le32(LCME_FL_STALE);
3454 id = pflr_id(mirror_id, i + 1);
3455 lcme->lcme_id = cpu_to_le32(id);
3457 memcpy((char *)lcm + offset,
3458 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
3459 le32_to_cpu(lcme->lcme_size));
3461 offset += le32_to_cpu(lcme->lcme_size);
3464 /* fixup layout information */
3465 lcm->lcm_size = cpu_to_le32(size);
3466 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
3467 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
3468 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NONE)
3469 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
3471 rc = lod_striping_reload(env, lo, buf, 0);
3475 lod_obj_inc_layout_gen(lo);
3476 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3478 /* transfer layout version to OST objects. */
3479 if (lo->ldo_mirror_count > 1) {
3480 struct lod_obj_stripe_cb_data data = { {0} };
3482 layout_attr->la_valid = LA_LAYOUT_VERSION;
3483 layout_attr->la_layout_version = 0;
3484 data.locd_attr = layout_attr;
3485 data.locd_declare = true;
3486 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
3487 rc = lod_obj_for_each_stripe(env, lo, th, &data);
3492 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
3493 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3501 * Split layouts, just set the LOVEA with the layout from mbuf.
3503 static int lod_declare_layout_split(const struct lu_env *env,
3504 struct dt_object *dt, const struct lu_buf *mbuf,
3507 struct lod_thread_info *info = lod_env_info(env);
3508 struct lu_attr *layout_attr = &info->lti_layout_attr;
3509 struct lod_object *lo = lod_dt_obj(dt);
3510 struct lov_comp_md_v1 *lcm = mbuf->lb_buf;
3514 rc = lod_striping_reload(env, lo, mbuf, LVF_ALL_STALE);
3518 lod_obj_inc_layout_gen(lo);
3519 /* fix on-disk layout gen */
3520 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3523 /* transfer layout version to OST objects. */
3524 if (lo->ldo_mirror_count > 1) {
3525 struct lod_obj_stripe_cb_data data = { {0} };
3527 layout_attr->la_valid = LA_LAYOUT_VERSION;
3528 layout_attr->la_layout_version = 0;
3529 data.locd_attr = layout_attr;
3530 data.locd_declare = true;
3531 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
3532 rc = lod_obj_for_each_stripe(env, lo, th, &data);
3537 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), mbuf,
3538 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3542 static int lod_layout_declare_or_purge_mirror(const struct lu_env *env,
3543 struct dt_object *dt, const struct lu_buf *buf,
3544 struct thandle *th, bool declare)
3546 struct lod_thread_info *info = lod_env_info(env);
3547 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3548 struct lod_object *lo = lod_dt_obj(dt);
3549 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3550 struct lov_comp_md_entry_v1 *entry;
3551 struct lov_mds_md_v1 *lmm;
3552 struct dt_object **sub_objs = NULL;
3553 int rc = 0, i, k, array_count = 0;
3558 * other ops (like lod_declare_destroy) could destroying sub objects
3561 mutex_lock(&lo->ldo_layout_mutex);
3564 /* prepare sub-objects array */
3565 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
3566 entry = &comp_v1->lcm_entries[i];
3568 if (!(entry->lcme_flags & LCME_FL_INIT))
3571 lmm = (struct lov_mds_md_v1 *)
3572 ((char *)comp_v1 + entry->lcme_offset);
3573 array_count += lmm->lmm_stripe_count;
3575 OBD_ALLOC_PTR_ARRAY(sub_objs, array_count);
3576 if (sub_objs == NULL) {
3577 mutex_unlock(&lo->ldo_layout_mutex);
3582 k = 0; /* sub_objs index */
3583 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
3584 struct lov_ost_data_v1 *objs;
3585 struct lu_object *o, *n;
3586 struct dt_object *dto;
3587 struct lu_device *nd;
3588 struct lov_mds_md_v3 *v3;
3592 entry = &comp_v1->lcm_entries[i];
3594 if (!(entry->lcme_flags & LCME_FL_INIT))
3597 lmm = (struct lov_mds_md_v1 *)
3598 ((char *)comp_v1 + entry->lcme_offset);
3599 v3 = (struct lov_mds_md_v3 *)lmm;
3600 if (lmm->lmm_magic == LOV_MAGIC_V3)
3601 objs = &v3->lmm_objects[0];
3603 objs = &lmm->lmm_objects[0];
3605 for (j = 0; j < lmm->lmm_stripe_count; j++) {
3606 idx = objs[j].l_ost_idx;
3607 rc = ostid_to_fid(&info->lti_fid, &objs[j].l_ost_oi,
3612 if (!fid_is_sane(&info->lti_fid)) {
3613 CERROR("%s: sub-object insane fid "DFID"\n",
3614 lod2obd(d)->obd_name,
3615 PFID(&info->lti_fid));
3616 GOTO(out, rc = -EINVAL);
3619 lod_getref(&d->lod_ost_descs);
3621 rc = validate_lod_and_idx(d, idx);
3623 lod_putref(d, &d->lod_ost_descs);
3627 nd = &OST_TGT(d, idx)->ltd_tgt->dd_lu_dev;
3628 lod_putref(d, &d->lod_ost_descs);
3630 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
3632 GOTO(out, rc = PTR_ERR(o));
3634 n = lu_object_locate(o->lo_header, nd->ld_type);
3636 lu_object_put(env, n);
3637 GOTO(out, rc = -ENOENT);
3640 dto = container_of(n, struct dt_object, do_lu);
3643 rc = lod_sub_declare_destroy(env, dto, th);
3644 dt_object_put(env, dto);
3649 * collect to-be-destroyed sub objects, the
3650 * reference would be released after actual
3656 } /* for each stripe */
3657 } /* for each component in the mirror */
3662 /* destroy the sub objects */
3663 for (; i < k; i++) {
3664 rc = lod_sub_destroy(env, sub_objs[i], th);
3667 dt_object_put(env, sub_objs[i]);
3671 * if a sub object destroy failed, we'd release sub objects
3672 * reference get from above sub_objs collection.
3675 dt_object_put(env, sub_objs[i]);
3677 OBD_FREE_PTR_ARRAY(sub_objs, array_count);
3679 mutex_unlock(&lo->ldo_layout_mutex);
3685 * Purge layouts, delete sub objects in the mirror stored in the vic_buf,
3686 * and set the LOVEA with the layout from mbuf.
3688 static int lod_declare_layout_purge(const struct lu_env *env,
3689 struct dt_object *dt, const struct lu_buf *buf,
3692 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3693 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3698 if (le32_to_cpu(comp_v1->lcm_magic) != LOV_MAGIC_COMP_V1) {
3699 CERROR("%s: invalid layout magic %#x != %#x\n",
3700 lod2obd(d)->obd_name, le32_to_cpu(comp_v1->lcm_magic),
3705 if (cpu_to_le32(LOV_MAGIC_COMP_V1) != LOV_MAGIC_COMP_V1)
3706 lustre_swab_lov_comp_md_v1(comp_v1);
3708 /* from now on, @buf contains cpu endian data */
3710 if (comp_v1->lcm_mirror_count != 0) {
3711 CERROR("%s: can only purge one mirror from "DFID"\n",
3712 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3716 /* delcare sub objects deletion in the mirror stored in @buf */
3717 rc = lod_layout_declare_or_purge_mirror(env, dt, buf, th, true);
3721 /* delete sub objects from the mirror stored in @buf */
3722 static int lod_layout_purge(const struct lu_env *env, struct dt_object *dt,
3723 const struct lu_buf *buf, struct thandle *th)
3728 rc = lod_layout_declare_or_purge_mirror(env, dt, buf, th, false);
3733 * Implementation of dt_object_operations::do_declare_xattr_set.
3735 * \see dt_object_operations::do_declare_xattr_set() in the API description
3738 * the extension to the API:
3739 * - declaring LOVEA requests striping creation
3740 * - LU_XATTR_REPLACE means layout swap
3742 static int lod_declare_xattr_set(const struct lu_env *env,
3743 struct dt_object *dt,
3744 const struct lu_buf *buf,
3745 const char *name, int fl,
3748 struct dt_object *next = dt_object_child(dt);
3749 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3754 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
3755 if ((S_ISREG(mode) || mode == 0) &&
3756 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE | LU_XATTR_SPLIT |
3758 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3759 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
3761 * this is a request to create object's striping.
3763 * allow to declare predefined striping on a new (!mode) object
3764 * which is supposed to be replay of regular file creation
3765 * (when LOV setting is declared)
3767 * LU_XATTR_REPLACE is set to indicate a layout swap
3769 if (dt_object_exists(dt)) {
3770 rc = dt_attr_get(env, next, attr);
3774 memset(attr, 0, sizeof(*attr));
3775 attr->la_valid = LA_TYPE | LA_MODE;
3776 attr->la_mode = S_IFREG;
3778 rc = lod_declare_striped_create(env, dt, attr, buf, th);
3779 } else if (fl & LU_XATTR_MERGE) {
3780 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3781 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3782 rc = lod_declare_layout_merge(env, dt, buf, th);
3783 } else if (fl & LU_XATTR_SPLIT) {
3784 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3785 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3786 rc = lod_declare_layout_split(env, dt, buf, th);
3787 } else if (fl & LU_XATTR_PURGE) {
3788 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3789 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3790 rc = lod_declare_layout_purge(env, dt, buf, th);
3791 } else if (S_ISREG(mode) &&
3792 strlen(name) >= sizeof(XATTR_LUSTRE_LOV) + 3 &&
3793 allowed_lustre_lov(name)) {
3795 * this is a request to modify object's striping.
3796 * add/set/del component(s).
3798 if (!dt_object_exists(dt))
3801 rc = lod_declare_modify_layout(env, dt, name, buf, th);
3802 } else if (S_ISDIR(mode)) {
3803 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
3804 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3805 rc = lod_replace_parent_fid(env, dt, buf, th, true);
3807 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
3814 * Apply xattr changes to the object.
3816 * Applies xattr changes to the object and the stripes if the latter exist.
3818 * \param[in] env execution environment
3819 * \param[in] dt object
3820 * \param[in] buf buffer pointing to the new value of xattr
3821 * \param[in] name name of xattr
3822 * \param[in] fl flags
3823 * \param[in] th transaction handle
3825 * \retval 0 on success
3826 * \retval negative if failed
3828 static int lod_xattr_set_internal(const struct lu_env *env,
3829 struct dt_object *dt,
3830 const struct lu_buf *buf,
3831 const char *name, int fl,
3834 struct dt_object *next = dt_object_child(dt);
3835 struct lod_object *lo = lod_dt_obj(dt);
3840 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3841 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3844 /* Note: Do not set LinkEA on sub-stripes, otherwise
3845 * it will confuse the fid2path process(see mdt_path_current()).
3846 * The linkEA between master and sub-stripes is set in
3847 * lod_xattr_set_lmv(). */
3848 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
3851 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3852 if (!lo->ldo_stripe[i])
3855 if (!dt_object_exists(lo->ldo_stripe[i]))
3858 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
3868 * Delete an extended attribute.
3870 * Deletes specified xattr from the object and the stripes if the latter exist.
3872 * \param[in] env execution environment
3873 * \param[in] dt object
3874 * \param[in] name name of xattr
3875 * \param[in] th transaction handle
3877 * \retval 0 on success
3878 * \retval negative if failed
3880 static int lod_xattr_del_internal(const struct lu_env *env,
3881 struct dt_object *dt,
3882 const char *name, struct thandle *th)
3884 struct dt_object *next = dt_object_child(dt);
3885 struct lod_object *lo = lod_dt_obj(dt);
3891 rc = lod_sub_xattr_del(env, next, name, th);
3892 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3895 if (lo->ldo_dir_stripe_count == 0)
3898 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3899 if (!lo->ldo_stripe[i])
3902 if (!dt_object_exists(lo->ldo_stripe[i]))
3905 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3914 * Set default striping on a directory.
3916 * Sets specified striping on a directory object unless it matches the default
3917 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3918 * EA. This striping will be used when regular file is being created in this
3921 * \param[in] env execution environment
3922 * \param[in] dt the striped object
3923 * \param[in] buf buffer with the striping
3924 * \param[in] name name of EA
3925 * \param[in] fl xattr flag (see OSD API description)
3926 * \param[in] th transaction handle
3928 * \retval 0 on success
3929 * \retval negative if failed
3931 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
3932 struct dt_object *dt,
3933 const struct lu_buf *buf,
3934 const char *name, int fl,
3937 struct lov_user_md_v1 *lum;
3938 struct lov_user_md_v3 *v3 = NULL;
3939 const char *pool_name = NULL;
3944 LASSERT(buf != NULL && buf->lb_buf != NULL);
3947 switch (lum->lmm_magic) {
3948 case LOV_USER_MAGIC_SPECIFIC:
3949 case LOV_USER_MAGIC_V3:
3951 if (v3->lmm_pool_name[0] != '\0')
3952 pool_name = v3->lmm_pool_name;
3954 case LOV_USER_MAGIC_V1:
3955 /* if { size, offset, count } = { 0, -1, 0 } and no pool
3956 * (i.e. all default values specified) then delete default
3957 * striping from dir. */
3959 "set default striping: sz %u # %u offset %d %s %s\n",
3960 (unsigned)lum->lmm_stripe_size,
3961 (unsigned)lum->lmm_stripe_count,
3962 (int)lum->lmm_stripe_offset,
3963 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
3965 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
3966 lum->lmm_stripe_count,
3967 lum->lmm_stripe_offset,
3970 case LOV_USER_MAGIC_COMP_V1:
3972 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lum;
3973 struct lov_comp_md_entry_v1 *lcme;
3976 comp_cnt = le16_to_cpu(lcm->lcm_entry_count);
3977 for (i = 0; i < comp_cnt; i++) {
3978 lcme = &lcm->lcm_entries[i];
3979 if (lcme->lcme_flags & cpu_to_le32(LCME_FL_EXTENSION)) {
3980 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
3989 CERROR("Invalid magic %x\n", lum->lmm_magic);
3994 rc = lod_xattr_del_internal(env, dt, name, th);
3998 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4004 static int lod_get_default_lov_striping(const struct lu_env *env,
4005 struct lod_object *lo,
4006 struct lod_default_striping *lds,
4007 struct dt_allocation_hint *ah);
4010 * Helper function to convert compound layout to compound layout with
4013 * Copy lcm_entries array of \a src to \a tgt. Replace lov_user_md_v1
4014 * components of \a src with lov_user_md_v3 using \a pool.
4016 * \param[in] src source layout
4017 * \param[in] pool pool to use in \a tgt
4018 * \param[out] tgt target layout
4020 static void embed_pool_to_comp_v1(const struct lov_comp_md_v1 *src,
4022 struct lov_comp_md_v1 *tgt)
4025 struct lov_user_md_v1 *lum;
4026 struct lov_user_md_v3 *lum3;
4027 struct lov_comp_md_entry_v1 *entry;
4031 entry = tgt->lcm_entries;
4033 for (i = 0; i < le16_to_cpu(src->lcm_entry_count); i++, entry++) {
4034 *entry = src->lcm_entries[i];
4035 offset = le32_to_cpu(src->lcm_entries[i].lcme_offset);
4036 entry->lcme_offset = cpu_to_le32(offset + shift);
4038 lum = (struct lov_user_md_v1 *)((char *)src + offset);
4039 lum3 = (struct lov_user_md_v3 *)((char *)tgt + offset + shift);
4040 *(struct lov_user_md_v1 *)lum3 = *lum;
4041 if (lum->lmm_pattern == cpu_to_le32(LOV_PATTERN_MDT)) {
4042 lum3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
4044 lum3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4045 entry->lcme_size = cpu_to_le32(sizeof(*lum3));
4046 strlcpy(lum3->lmm_pool_name, pool,
4047 sizeof(lum3->lmm_pool_name));
4048 shift += sizeof(*lum3) - sizeof(*lum);
4054 * Set default striping on a directory.
4056 * Sets specified striping on a directory object unless it matches the default
4057 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
4058 * EA. This striping will be used when regular file is being created in this
4060 * If current default striping includes a pool but specifed striping
4061 * does not - retain the pool if it exists.
4063 * \param[in] env execution environment
4064 * \param[in] dt the striped object
4065 * \param[in] buf buffer with the striping
4066 * \param[in] name name of EA
4067 * \param[in] fl xattr flag (see OSD API description)
4068 * \param[in] th transaction handle
4070 * \retval 0 on success
4071 * \retval negative if failed
4073 static int lod_xattr_set_default_lov_on_dir(const struct lu_env *env,
4074 struct dt_object *dt,
4075 const struct lu_buf *buf,
4076 const char *name, int fl,
4079 struct lod_default_striping *lds = lod_lds_buf_get(env);
4080 struct lov_user_md_v1 *v1 = buf->lb_buf;
4081 char pool[LOV_MAXPOOLNAME + 1];
4087 /* get existing striping config */
4088 rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds, NULL);
4092 memset(pool, 0, sizeof(pool));
4093 if (lds->lds_def_striping_set == 1)
4094 lod_layout_get_pool(lds->lds_def_comp_entries,
4095 lds->lds_def_comp_cnt, pool,
4098 is_del = LOVEA_DELETE_VALUES(v1->lmm_stripe_size,
4099 v1->lmm_stripe_count,
4100 v1->lmm_stripe_offset,
4103 /* Retain the pool name if it is not given */
4104 if (v1->lmm_magic == LOV_USER_MAGIC_V1 && pool[0] != '\0' &&
4106 struct lod_thread_info *info = lod_env_info(env);
4107 struct lov_user_md_v3 *v3 = info->lti_ea_store;
4109 memset(v3, 0, sizeof(*v3));
4110 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4111 v3->lmm_pattern = cpu_to_le32(v1->lmm_pattern);
4112 v3->lmm_stripe_count = cpu_to_le32(v1->lmm_stripe_count);
4113 v3->lmm_stripe_offset = cpu_to_le32(v1->lmm_stripe_offset);
4114 v3->lmm_stripe_size = cpu_to_le32(v1->lmm_stripe_size);
4116 strlcpy(v3->lmm_pool_name, pool, sizeof(v3->lmm_pool_name));
4118 info->lti_buf.lb_buf = v3;
4119 info->lti_buf.lb_len = sizeof(*v3);
4120 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4122 } else if (v1->lmm_magic == LOV_USER_MAGIC_COMP_V1 &&
4123 pool[0] != '\0' && !is_del) {
4125 * try to retain the pool from default layout if the
4126 * specified component layout does not provide pool
4129 struct lod_thread_info *info = lod_env_info(env);
4130 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
4131 struct lov_comp_md_v1 *comp_v1p;
4132 struct lov_user_md_v1 *lum;
4136 struct lov_comp_md_entry_v1 *entry;
4139 entry_count = le16_to_cpu(comp_v1->lcm_entry_count);
4140 size = sizeof(*comp_v1) +
4141 entry_count * sizeof(comp_v1->lcm_entries[0]);
4142 entry = comp_v1->lcm_entries;
4143 for (i = 0; i < entry_count; i++, entry++) {
4144 offset = le32_to_cpu(entry->lcme_offset);
4145 lum = (struct lov_user_md_v1 *)((char *)comp_v1 +
4147 if (le32_to_cpu(lum->lmm_magic) != LOV_USER_MAGIC_V1)
4148 /* the i-th component includes pool info */
4150 if (lum->lmm_pattern == cpu_to_le32(LOV_PATTERN_MDT))
4151 size += sizeof(struct lov_user_md_v1);
4153 size += sizeof(struct lov_user_md_v3);
4156 if (i == entry_count) {
4158 * re-compose the layout to include the pool for
4161 if (info->lti_ea_store_size < size)
4162 rc = lod_ea_store_resize(info, size);
4165 comp_v1p = info->lti_ea_store;
4166 *comp_v1p = *comp_v1;
4167 comp_v1p->lcm_size = cpu_to_le32(size);
4168 embed_pool_to_comp_v1(comp_v1, pool, comp_v1p);
4170 info->lti_buf.lb_buf = comp_v1p;
4171 info->lti_buf.lb_len = size;
4172 rc = lod_xattr_set_lov_on_dir(env, dt,
4177 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl,
4181 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl, th);
4184 if (lds->lds_def_striping_set == 1 && lds->lds_def_comp_entries != NULL)
4185 lod_free_def_comp_entries(lds);
4191 * Set default striping on a directory object.
4193 * Sets specified striping on a directory object unless it matches the default
4194 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
4195 * EA. This striping will be used when a new directory is being created in the
4198 * \param[in] env execution environment
4199 * \param[in] dt the striped object
4200 * \param[in] buf buffer with the striping
4201 * \param[in] name name of EA
4202 * \param[in] fl xattr flag (see OSD API description)
4203 * \param[in] th transaction handle
4205 * \retval 0 on success
4206 * \retval negative if failed
4208 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
4209 struct dt_object *dt,
4210 const struct lu_buf *buf,
4211 const char *name, int fl,
4214 struct lmv_user_md_v1 *lum;
4219 LASSERT(buf != NULL && buf->lb_buf != NULL);
4223 "set default stripe_count # %u stripe_offset %d hash %u\n",
4224 le32_to_cpu(lum->lum_stripe_count),
4225 (int)le32_to_cpu(lum->lum_stripe_offset),
4226 le32_to_cpu(lum->lum_hash_type));
4228 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
4229 le32_to_cpu(lum->lum_stripe_offset)) &&
4230 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
4231 rc = lod_xattr_del_internal(env, dt, name, th);
4235 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4244 * Turn directory into a striped directory.
4246 * During replay the client sends the striping created before MDT
4247 * failure, then the layer above LOD sends this defined striping
4248 * using ->do_xattr_set(), so LOD uses this method to replay creation
4249 * of the stripes. Notice the original information for the striping
4250 * (#stripes, FIDs, etc) was transferred in declare path.
4252 * \param[in] env execution environment
4253 * \param[in] dt the striped object
4254 * \param[in] buf not used currently
4255 * \param[in] name not used currently
4256 * \param[in] fl xattr flag (see OSD API description)
4257 * \param[in] th transaction handle
4259 * \retval 0 on success
4260 * \retval negative if failed
4262 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
4263 const struct lu_buf *buf, const char *name,
4264 int fl, struct thandle *th)
4266 struct lod_object *lo = lod_dt_obj(dt);
4267 struct lod_thread_info *info = lod_env_info(env);
4268 struct lu_attr *attr = &info->lti_attr;
4269 struct dt_object_format *dof = &info->lti_format;
4270 struct lu_buf lmv_buf;
4271 struct lu_buf slave_lmv_buf;
4272 struct lmv_mds_md_v1 *lmm;
4273 struct lmv_mds_md_v1 *slave_lmm = NULL;
4274 struct dt_insert_rec *rec = &info->lti_dt_rec;
4279 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4282 /* The stripes are supposed to be allocated in declare phase,
4283 * if there are no stripes being allocated, it will skip */
4284 if (lo->ldo_dir_stripe_count == 0) {
4285 if (lo->ldo_is_foreign) {
4286 rc = lod_sub_xattr_set(env, dt_object_child(dt), buf,
4287 XATTR_NAME_LMV, fl, th);
4294 rc = dt_attr_get(env, dt_object_child(dt), attr);
4298 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME | LA_FLAGS |
4299 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
4300 dof->dof_type = DFT_DIR;
4302 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
4305 lmm = lmv_buf.lb_buf;
4307 OBD_ALLOC_PTR(slave_lmm);
4308 if (slave_lmm == NULL)
4311 lod_prep_slave_lmv_md(slave_lmm, lmm);
4312 slave_lmv_buf.lb_buf = slave_lmm;
4313 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
4315 rec->rec_type = S_IFDIR;
4316 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4317 struct dt_object *dto = lo->ldo_stripe[i];
4318 char *stripe_name = info->lti_key;
4319 struct lu_name *sname;
4320 struct linkea_data ldata = { NULL };
4321 struct lu_buf linkea_buf;
4323 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
4327 /* fail a remote stripe creation */
4328 if (i && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_CREATE))
4331 /* don't create stripe if:
4332 * 1. it's source stripe of migrating directory
4333 * 2. it's existed stripe of splitting directory
4335 if ((lod_is_migrating(lo) && i >= lo->ldo_dir_migrate_offset) ||
4336 (lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) {
4337 if (!dt_object_exists(dto))
4338 GOTO(out, rc = -EINVAL);
4340 dt_write_lock(env, dto, DT_TGT_CHILD);
4341 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
4343 dt_write_unlock(env, dto);
4347 rc = lod_sub_ref_add(env, dto, th);
4348 dt_write_unlock(env, dto);
4352 rec->rec_fid = lu_object_fid(&dto->do_lu);
4353 rc = lod_sub_insert(env, dto,
4354 (const struct dt_rec *)rec,
4355 (const struct dt_key *)dot, th);
4360 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
4361 cfs_fail_val != i) {
4362 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
4364 slave_lmm->lmv_master_mdt_index =
4367 slave_lmm->lmv_master_mdt_index =
4370 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
4371 XATTR_NAME_LMV, 0, th);
4376 /* don't insert stripe if it's existed stripe of splitting
4377 * directory (this directory is striped).
4378 * NB, plain directory will insert itself as the first
4381 if (lod_is_splitting(lo) && lo->ldo_dir_split_offset > 1 &&
4382 lo->ldo_dir_split_offset > i)
4385 rec->rec_fid = lu_object_fid(&dt->do_lu);
4386 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
4387 (const struct dt_key *)dotdot, th);
4391 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
4393 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4394 PFID(lu_object_fid(&dto->do_lu)), i + 1);
4396 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4397 PFID(lu_object_fid(&dto->do_lu)), i);
4399 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
4400 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
4401 sname, lu_object_fid(&dt->do_lu));
4405 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
4406 linkea_buf.lb_len = ldata.ld_leh->leh_len;
4407 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
4408 XATTR_NAME_LINK, 0, th);
4412 rec->rec_fid = lu_object_fid(&dto->do_lu);
4413 rc = lod_sub_insert(env, dt_object_child(dt),
4414 (const struct dt_rec *)rec,
4415 (const struct dt_key *)stripe_name, th);
4419 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
4424 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
4425 rc = lod_sub_xattr_set(env, dt_object_child(dt),
4426 &lmv_buf, XATTR_NAME_LMV, fl, th);
4428 if (slave_lmm != NULL)
4429 OBD_FREE_PTR(slave_lmm);
4435 * Helper function to declare/execute creation of a striped directory
4437 * Called in declare/create object path, prepare striping for a directory
4438 * and prepare defaults data striping for the objects to be created in
4439 * that directory. Notice the function calls "declaration" or "execution"
4440 * methods depending on \a declare param. This is a consequence of the
4441 * current approach while we don't have natural distributed transactions:
4442 * we basically execute non-local updates in the declare phase. So, the
4443 * arguments for the both phases are the same and this is the reason for
4444 * this function to exist.
4446 * \param[in] env execution environment
4447 * \param[in] dt object
4448 * \param[in] attr attributes the stripes will be created with
4449 * \param[in] lmu lmv_user_md if MDT indices are specified
4450 * \param[in] dof format of stripes (see OSD API description)
4451 * \param[in] th transaction handle
4452 * \param[in] declare where to call "declare" or "execute" methods
4454 * \retval 0 on success
4455 * \retval negative if failed
4457 static int lod_dir_striping_create_internal(const struct lu_env *env,
4458 struct dt_object *dt,
4459 struct lu_attr *attr,
4460 const struct lu_buf *lmu,
4461 struct dt_object_format *dof,
4465 struct lod_thread_info *info = lod_env_info(env);
4466 struct lod_object *lo = lod_dt_obj(dt);
4467 const struct lod_default_striping *lds = lo->ldo_def_striping;
4471 LASSERT(ergo(lds != NULL,
4472 lds->lds_def_striping_set ||
4473 lds->lds_dir_def_striping_set));
4475 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
4476 lo->ldo_dir_stripe_offset)) {
4478 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
4479 int stripe_count = lo->ldo_dir_stripe_count;
4481 if (info->lti_ea_store_size < sizeof(*v1)) {
4482 rc = lod_ea_store_resize(info, sizeof(*v1));
4485 v1 = info->lti_ea_store;
4488 memset(v1, 0, sizeof(*v1));
4489 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
4490 v1->lum_stripe_count = cpu_to_le32(stripe_count);
4491 v1->lum_stripe_offset =
4492 cpu_to_le32(lo->ldo_dir_stripe_offset);
4494 info->lti_buf.lb_buf = v1;
4495 info->lti_buf.lb_len = sizeof(*v1);
4496 lmu = &info->lti_buf;
4500 rc = lod_declare_xattr_set_lmv(env, dt, attr, lmu, dof,
4503 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
4508 /* foreign LMV EA case */
4510 struct lmv_foreign_md *lfm = lmu->lb_buf;
4512 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
4513 rc = lod_declare_xattr_set_lmv(env, dt, attr,
4517 if (lo->ldo_is_foreign) {
4518 LASSERT(lo->ldo_foreign_lmv != NULL &&
4519 lo->ldo_foreign_lmv_size > 0);
4520 info->lti_buf.lb_buf = lo->ldo_foreign_lmv;
4521 info->lti_buf.lb_len = lo->ldo_foreign_lmv_size;
4522 lmu = &info->lti_buf;
4523 rc = lod_xattr_set_lmv(env, dt, lmu,
4524 XATTR_NAME_LMV, 0, th);
4529 /* Transfer default LMV striping from the parent */
4530 if (lds != NULL && lds->lds_dir_def_striping_set &&
4531 lds->lds_dir_def_max_inherit != LMV_INHERIT_END &&
4532 lds->lds_dir_def_max_inherit != LMV_INHERIT_NONE &&
4533 !(LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
4534 lds->lds_dir_def_stripe_offset) &&
4535 le32_to_cpu(lds->lds_dir_def_hash_type) !=
4536 LMV_HASH_TYPE_UNKNOWN)) {
4537 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
4539 if (info->lti_ea_store_size < sizeof(*v1)) {
4540 rc = lod_ea_store_resize(info, sizeof(*v1));
4543 v1 = info->lti_ea_store;
4546 memset(v1, 0, sizeof(*v1));
4547 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
4548 v1->lum_stripe_count =
4549 cpu_to_le32(lds->lds_dir_def_stripe_count);
4550 v1->lum_stripe_offset =
4551 cpu_to_le32(lds->lds_dir_def_stripe_offset);
4553 cpu_to_le32(lds->lds_dir_def_hash_type);
4554 v1->lum_max_inherit =
4555 lmv_inherit_next(lds->lds_dir_def_max_inherit);
4556 v1->lum_max_inherit_rr =
4557 lmv_inherit_rr_next(lds->lds_dir_def_max_inherit_rr);
4559 info->lti_buf.lb_buf = v1;
4560 info->lti_buf.lb_len = sizeof(*v1);
4562 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4563 XATTR_NAME_DEFAULT_LMV,
4566 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
4568 XATTR_NAME_DEFAULT_LMV, 0,
4574 /* Transfer default LOV striping from the parent */
4575 if (lds != NULL && lds->lds_def_striping_set &&
4576 lds->lds_def_comp_cnt != 0) {
4577 struct lov_mds_md *lmm;
4578 int lmm_size = lod_comp_md_size(lo, true);
4580 if (info->lti_ea_store_size < lmm_size) {
4581 rc = lod_ea_store_resize(info, lmm_size);
4585 lmm = info->lti_ea_store;
4587 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
4591 info->lti_buf.lb_buf = lmm;
4592 info->lti_buf.lb_len = lmm_size;
4595 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4596 XATTR_NAME_LOV, 0, th);
4598 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4599 XATTR_NAME_LOV, 0, th);
4604 /* ldo_def_striping is not allocated, clear after use, in case directory
4605 * layout is changed later.
4608 lo->ldo_def_striping = NULL;
4613 static int lod_declare_dir_striping_create(const struct lu_env *env,
4614 struct dt_object *dt,
4615 struct lu_attr *attr,
4617 struct dt_object_format *dof,
4620 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
4624 static int lod_dir_striping_create(const struct lu_env *env,
4625 struct dt_object *dt,
4626 struct lu_attr *attr,
4627 struct dt_object_format *dof,
4630 return lod_dir_striping_create_internal(env, dt, attr, NULL, dof, th,
4635 * Make LOV EA for striped object.
4637 * Generate striping information and store it in the LOV EA of the given
4638 * object. The caller must ensure nobody else is calling the function
4639 * against the object concurrently. The transaction must be started.
4640 * FLDB service must be running as well; it's used to map FID to the target,
4641 * which is stored in LOV EA.
4643 * \param[in] env execution environment for this thread
4644 * \param[in] lo LOD object
4645 * \param[in] th transaction handle
4647 * \retval 0 if LOV EA is stored successfully
4648 * \retval negative error number on failure
4650 static int lod_generate_and_set_lovea(const struct lu_env *env,
4651 struct lod_object *lo,
4654 struct lod_thread_info *info = lod_env_info(env);
4655 struct dt_object *next = dt_object_child(&lo->ldo_obj);
4656 struct lov_mds_md_v1 *lmm;
4662 if (lo->ldo_comp_cnt == 0 && !lo->ldo_is_foreign) {
4663 lod_striping_free_nolock(env, lo);
4664 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
4668 lmm_size = lod_comp_md_size(lo, false);
4669 if (info->lti_ea_store_size < lmm_size) {
4670 rc = lod_ea_store_resize(info, lmm_size);
4674 lmm = info->lti_ea_store;
4676 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
4680 info->lti_buf.lb_buf = lmm;
4681 info->lti_buf.lb_len = lmm_size;
4682 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
4683 XATTR_NAME_LOV, 0, th);
4687 static __u32 lod_gen_component_id(struct lod_object *lo,
4688 int mirror_id, int comp_idx);
4691 * Repeat an existing component
4693 * Creates a new layout by replicating an existing component. Uses striping
4694 * policy from previous component as a template for the striping for the new
4697 * New component starts with zero length, will be extended (or removed) before
4698 * returning layout to client.
4700 * NB: Reallocates layout components array (lo->ldo_comp_entries), invalidating
4701 * any pre-existing pointers to components. Handle with care.
4703 * \param[in] env execution environment for this thread
4704 * \param[in,out] lo object to update the layout of
4705 * \param[in] index index of component to copy
4707 * \retval 0 on success
4708 * \retval negative errno on error
4710 static int lod_layout_repeat_comp(const struct lu_env *env,
4711 struct lod_object *lo, int index)
4713 struct lod_layout_component *lod_comp;
4714 struct lod_layout_component *new_comp = NULL;
4715 struct lod_layout_component *comp_array;
4716 int rc = 0, i, new_cnt = lo->ldo_comp_cnt + 1;
4721 lod_comp = &lo->ldo_comp_entries[index];
4722 LASSERT(lod_comp_inited(lod_comp) && lod_comp->llc_id != LCME_ID_INVAL);
4724 CDEBUG(D_LAYOUT, "repeating component %d\n", index);
4726 OBD_ALLOC_PTR_ARRAY(comp_array, new_cnt);
4727 if (comp_array == NULL)
4728 GOTO(out, rc = -ENOMEM);
4730 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4731 memcpy(&comp_array[i + offset], &lo->ldo_comp_entries[i],
4732 sizeof(*comp_array));
4734 /* Duplicate this component in to the next slot */
4736 new_comp = &comp_array[i + 1];
4737 memcpy(&comp_array[i + 1], &lo->ldo_comp_entries[i],
4738 sizeof(*comp_array));
4739 /* We must now skip this new component when copying */
4744 /* Set up copied component */
4745 new_comp->llc_flags &= ~LCME_FL_INIT;
4746 new_comp->llc_stripe = NULL;
4747 new_comp->llc_stripes_allocated = 0;
4748 new_comp->llc_ost_indices = NULL;
4749 new_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4750 /* for uninstantiated components, layout gen stores default stripe
4752 new_comp->llc_layout_gen = lod_comp->llc_stripe_offset;
4753 /* This makes the repeated component zero-length, placed at the end of
4754 * the preceding component */
4755 new_comp->llc_extent.e_start = new_comp->llc_extent.e_end;
4756 new_comp->llc_timestamp = lod_comp->llc_timestamp;
4757 new_comp->llc_pool = NULL;
4759 rc = lod_set_pool(&new_comp->llc_pool, lod_comp->llc_pool);
4763 if (new_comp->llc_ostlist.op_array) {
4764 __u32 *op_array = NULL;
4766 OBD_ALLOC(op_array, new_comp->llc_ostlist.op_size);
4768 GOTO(out, rc = -ENOMEM);
4769 memcpy(op_array, &new_comp->llc_ostlist.op_array,
4770 new_comp->llc_ostlist.op_size);
4771 new_comp->llc_ostlist.op_array = op_array;
4774 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
4775 lo->ldo_comp_entries = comp_array;
4776 lo->ldo_comp_cnt = new_cnt;
4778 /* Generate an id for the new component */
4779 mirror_id = mirror_id_of(new_comp->llc_id);
4780 new_comp->llc_id = LCME_ID_INVAL;
4781 new_comp->llc_id = lod_gen_component_id(lo, mirror_id, index + 1);
4782 if (new_comp->llc_id == LCME_ID_INVAL)
4783 GOTO(out, rc = -ERANGE);
4788 OBD_FREE_PTR_ARRAY(comp_array, new_cnt);
4793 static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
4797 /* clear memory region that will be used for layout change */
4798 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
4799 info->lti_count = 0;
4801 if (info->lti_comp_size >= comp_cnt)
4804 if (info->lti_comp_size > 0) {
4805 OBD_FREE_PTR_ARRAY(info->lti_comp_idx, info->lti_comp_size);
4806 info->lti_comp_size = 0;
4809 OBD_ALLOC_PTR_ARRAY(info->lti_comp_idx, comp_cnt);
4810 if (!info->lti_comp_idx)
4813 info->lti_comp_size = comp_cnt;
4818 * Prepare new layout minus deleted components
4820 * Removes components marked for deletion (LCME_ID_INVAL) by copying to a new
4821 * layout and skipping those components. Removes stripe objects if any exist.
4824 * Reallocates layout components array (lo->ldo_comp_entries), invalidating
4825 * any pre-existing pointers to components.
4827 * Caller is responsible for updating mirror end (ldo_mirror[].lme_end).
4829 * \param[in] env execution environment for this thread
4830 * \param[in,out] lo object to update the layout of
4831 * \param[in] th transaction handle for this operation
4833 * \retval # of components deleted
4834 * \retval negative errno on error
4836 static int lod_layout_del_prep_layout(const struct lu_env *env,
4837 struct lod_object *lo,
4840 struct lod_layout_component *lod_comp;
4841 struct lod_thread_info *info = lod_env_info(env);
4842 int rc = 0, i, j, deleted = 0;
4846 LASSERT(lo->ldo_is_composite);
4847 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
4849 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
4853 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4854 lod_comp = &lo->ldo_comp_entries[i];
4856 if (lod_comp->llc_id != LCME_ID_INVAL) {
4857 /* Build array of things to keep */
4858 info->lti_comp_idx[info->lti_count++] = i;
4862 lod_obj_set_pool(lo, i, NULL);
4863 if (lod_comp->llc_ostlist.op_array) {
4864 OBD_FREE(lod_comp->llc_ostlist.op_array,
4865 lod_comp->llc_ostlist.op_size);
4866 lod_comp->llc_ostlist.op_array = NULL;
4867 lod_comp->llc_ostlist.op_size = 0;
4871 CDEBUG(D_LAYOUT, "deleting comp %d, left %d\n", i,
4872 lo->ldo_comp_cnt - deleted);
4874 /* No striping info for this component */
4875 if (lod_comp->llc_stripe == NULL)
4878 LASSERT(lod_comp->llc_stripe_count > 0);
4879 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4880 struct dt_object *obj = lod_comp->llc_stripe[j];
4885 /* components which are not init have no sub objects
4887 if (lod_comp_inited(lod_comp)) {
4888 rc = lod_sub_destroy(env, obj, th);
4893 lu_object_put(env, &obj->do_lu);
4894 lod_comp->llc_stripe[j] = NULL;
4896 OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe,
4897 lod_comp->llc_stripes_allocated);
4898 lod_comp->llc_stripe = NULL;
4899 OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices,
4900 lod_comp->llc_stripes_allocated);
4901 lod_comp->llc_ost_indices = NULL;
4902 lod_comp->llc_stripes_allocated = 0;
4905 /* info->lti_count has the amount of left components */
4906 LASSERTF(info->lti_count >= 0 && info->lti_count < lo->ldo_comp_cnt,
4907 "left = %d, lo->ldo_comp_cnt %d\n", (int)info->lti_count,
4908 (int)lo->ldo_comp_cnt);
4910 if (info->lti_count > 0) {
4911 struct lod_layout_component *comp_array;
4913 OBD_ALLOC_PTR_ARRAY(comp_array, info->lti_count);
4914 if (comp_array == NULL)
4915 GOTO(out, rc = -ENOMEM);
4917 for (i = 0; i < info->lti_count; i++) {
4918 memcpy(&comp_array[i],
4919 &lo->ldo_comp_entries[info->lti_comp_idx[i]],
4920 sizeof(*comp_array));
4923 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
4924 lo->ldo_comp_entries = comp_array;
4925 lo->ldo_comp_cnt = info->lti_count;
4927 lod_free_comp_entries(lo);
4932 return rc ? rc : deleted;
4936 * Delete layout component(s)
4938 * This function sets up the layout data in the env and does the setattrs
4939 * required to write out the new layout. The layout itself is modified in
4940 * lod_layout_del_prep_layout.
4942 * \param[in] env execution environment for this thread
4943 * \param[in] dt object
4944 * \param[in] th transaction handle
4946 * \retval 0 on success
4947 * \retval negative error number on failure
4949 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
4952 struct lod_object *lo = lod_dt_obj(dt);
4953 struct dt_object *next = dt_object_child(dt);
4954 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4957 LASSERT(lo->ldo_mirror_count == 1);
4959 mutex_lock(&lo->ldo_layout_mutex);
4961 rc = lod_layout_del_prep_layout(env, lo, th);
4965 /* Only do this if we didn't delete all components */
4966 if (lo->ldo_comp_cnt > 0) {
4967 lo->ldo_mirrors[0].lme_end = lo->ldo_comp_cnt - 1;
4968 lod_obj_inc_layout_gen(lo);
4971 LASSERT(dt_object_exists(dt));
4972 rc = dt_attr_get(env, next, attr);
4976 if (attr->la_size > 0) {
4978 attr->la_valid = LA_SIZE;
4979 rc = lod_sub_attr_set(env, next, attr, th);
4984 rc = lod_generate_and_set_lovea(env, lo, th);
4988 lod_striping_free_nolock(env, lo);
4990 mutex_unlock(&lo->ldo_layout_mutex);
4997 * Implementation of dt_object_operations::do_xattr_set.
4999 * Sets specified extended attribute on the object. Three types of EAs are
5001 * LOV EA - stores striping for a regular file or default striping (when set
5003 * LMV EA - stores a marker for the striped directories
5004 * DMV EA - stores default directory striping
5006 * When striping is applied to a non-striped existing object (this is called
5007 * late striping), then LOD notices the caller wants to turn the object into a
5008 * striped one. The stripe objects are created and appropriate EA is set:
5009 * LOV EA storing all the stripes directly or LMV EA storing just a small header
5010 * with striping configuration.
5012 * \see dt_object_operations::do_xattr_set() in the API description for details.
5014 static int lod_xattr_set(const struct lu_env *env,
5015 struct dt_object *dt, const struct lu_buf *buf,
5016 const char *name, int fl, struct thandle *th)
5018 struct dt_object *next = dt_object_child(dt);
5019 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
5020 struct lod_object *lo = lod_dt_obj(dt);
5021 struct lod_obj_stripe_cb_data data = { {0} };
5026 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
5027 !strcmp(name, XATTR_NAME_LMV)) {
5029 case LU_XATTR_CREATE:
5030 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
5033 case LU_XATTR_REPLACE:
5034 rc = lod_dir_layout_set(env, dt, buf, fl, th);
5041 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
5042 strcmp(name, XATTR_NAME_LOV) == 0) {
5043 rc = lod_xattr_set_default_lov_on_dir(env, dt, buf, name, fl,
5046 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
5047 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
5049 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
5052 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
5053 (strcmp(name, XATTR_NAME_LOV) == 0 ||
5054 strcmp(name, XATTR_LUSTRE_LOV) == 0 ||
5055 allowed_lustre_lov(name))) {
5056 /* in case of lov EA swap, just set it
5057 * if not, it is a replay so check striping match what we
5058 * already have during req replay, declare_xattr_set()
5059 * defines striping, then create() does the work */
5060 if (fl & LU_XATTR_REPLACE) {
5061 /* free stripes, then update disk */
5062 lod_striping_free(env, lod_dt_obj(dt));
5064 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
5065 } else if (fl & LU_XATTR_SPLIT) {
5066 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
5070 rc = lod_striping_reload(env, lo, buf, LVF_ALL_STALE);
5074 if (lo->ldo_mirror_count > 1 &&
5075 layout_attr->la_valid & LA_LAYOUT_VERSION) {
5077 layout_attr->la_layout_version =
5079 data.locd_attr = layout_attr;
5080 data.locd_declare = false;
5081 data.locd_stripe_cb =
5082 lod_obj_stripe_attr_set_cb;
5083 rc = lod_obj_for_each_stripe(env, lo, th,
5088 } else if (fl & LU_XATTR_PURGE) {
5089 rc = lod_layout_purge(env, dt, buf, th);
5090 } else if (dt_object_remote(dt)) {
5091 /* This only happens during migration, see
5092 * mdd_migrate_create(), in which Master MDT will
5093 * create a remote target object, and only set
5094 * (migrating) stripe EA on the remote object,
5095 * and does not need creating each stripes. */
5096 rc = lod_sub_xattr_set(env, next, buf, name,
5098 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
5099 /* delete component(s) */
5100 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
5101 rc = lod_layout_del(env, dt, th);
5104 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
5105 * it's going to create create file with specified
5106 * component(s), the striping must have not being
5107 * cached in this case;
5109 * Otherwise, it's going to add/change component(s) to
5110 * an existing file, the striping must have been cached
5113 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
5114 !strcmp(name, XATTR_NAME_LOV),
5115 !lod_dt_obj(dt)->ldo_comp_cached));
5117 rc = lod_striped_create(env, dt, NULL, NULL, th);
5121 if (fl & LU_XATTR_MERGE && lo->ldo_mirror_count > 1 &&
5122 layout_attr->la_valid & LA_LAYOUT_VERSION) {
5123 /* mirror merge exec phase */
5124 layout_attr->la_layout_version =
5126 data.locd_attr = layout_attr;
5127 data.locd_declare = false;
5128 data.locd_stripe_cb =
5129 lod_obj_stripe_attr_set_cb;
5130 rc = lod_obj_for_each_stripe(env, lo, th,
5137 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
5138 rc = lod_replace_parent_fid(env, dt, buf, th, false);
5143 /* then all other xattr */
5144 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
5150 * Implementation of dt_object_operations::do_declare_xattr_del.
5152 * \see dt_object_operations::do_declare_xattr_del() in the API description
5155 static int lod_declare_xattr_del(const struct lu_env *env,
5156 struct dt_object *dt, const char *name,
5159 struct lod_object *lo = lod_dt_obj(dt);
5160 struct dt_object *next = dt_object_child(dt);
5165 rc = lod_sub_declare_xattr_del(env, next, name, th);
5169 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
5172 /* NB: don't delete stripe LMV, because when we do this, normally we
5173 * will remove stripes, besides, if directory LMV is corrupt, this will
5174 * prevent deleting its LMV and fixing it (via LFSCK).
5176 if (!strcmp(name, XATTR_NAME_LMV))
5179 rc = lod_striping_load(env, lo);
5183 if (lo->ldo_dir_stripe_count == 0)
5186 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5187 struct dt_object *dto = lo->ldo_stripe[i];
5192 if (!dt_object_exists(dto))
5195 rc = lod_sub_declare_xattr_del(env, dto, name, th);
5204 * Implementation of dt_object_operations::do_xattr_del.
5206 * If EA storing a regular striping is being deleted, then release
5207 * all the references to the stripe objects in core.
5209 * \see dt_object_operations::do_xattr_del() in the API description for details.
5211 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
5212 const char *name, struct thandle *th)
5218 if (!strcmp(name, XATTR_NAME_LOV) || !strcmp(name, XATTR_NAME_LMV))
5219 lod_striping_free(env, lod_dt_obj(dt));
5221 rc = lod_xattr_del_internal(env, dt, name, th);
5227 * Implementation of dt_object_operations::do_xattr_list.
5229 * \see dt_object_operations::do_xattr_list() in the API description
5232 static int lod_xattr_list(const struct lu_env *env,
5233 struct dt_object *dt, const struct lu_buf *buf)
5235 return dt_xattr_list(env, dt_object_child(dt), buf);
5238 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
5240 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
5244 * Copy OST list from layout provided by user.
5246 * \param[in] lod_comp layout_component to be filled
5247 * \param[in] v3 LOV EA V3 user data
5249 * \retval 0 on success
5250 * \retval negative if failed
5252 int lod_comp_copy_ost_lists(struct lod_layout_component *lod_comp,
5253 struct lov_user_md_v3 *v3)
5259 if (v3->lmm_stripe_offset == LOV_OFFSET_DEFAULT)
5260 v3->lmm_stripe_offset = v3->lmm_objects[0].l_ost_idx;
5262 if (lod_comp->llc_ostlist.op_array) {
5263 if (lod_comp->llc_ostlist.op_size >=
5264 v3->lmm_stripe_count * sizeof(__u32)) {
5265 lod_comp->llc_ostlist.op_count =
5266 v3->lmm_stripe_count;
5269 OBD_FREE(lod_comp->llc_ostlist.op_array,
5270 lod_comp->llc_ostlist.op_size);
5273 /* copy ost list from lmm */
5274 lod_comp->llc_ostlist.op_count = v3->lmm_stripe_count;
5275 lod_comp->llc_ostlist.op_size = v3->lmm_stripe_count * sizeof(__u32);
5276 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
5277 lod_comp->llc_ostlist.op_size);
5278 if (!lod_comp->llc_ostlist.op_array)
5281 for (j = 0; j < v3->lmm_stripe_count; j++) {
5282 lod_comp->llc_ostlist.op_array[j] =
5283 v3->lmm_objects[j].l_ost_idx;
5291 * Get default striping.
5293 * \param[in] env execution environment
5294 * \param[in] lo object
5295 * \param[out] lds default striping
5297 * \retval 0 on success
5298 * \retval negative if failed
5300 static int lod_get_default_lov_striping(const struct lu_env *env,
5301 struct lod_object *lo,
5302 struct lod_default_striping *lds,
5303 struct dt_allocation_hint *ah)
5305 struct lod_thread_info *info = lod_env_info(env);
5306 struct lov_user_md_v1 *v1 = NULL;
5307 struct lov_user_md_v3 *v3 = NULL;
5308 struct lov_comp_md_v1 *comp_v1 = NULL;
5316 lds->lds_def_striping_set = 0;
5318 rc = lod_get_lov_ea(env, lo);
5322 if (rc < (typeof(rc))sizeof(struct lov_user_md))
5325 v1 = info->lti_ea_store;
5326 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
5327 lustre_swab_lov_user_md_v1(v1);
5328 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
5329 v3 = (struct lov_user_md_v3 *)v1;
5330 lustre_swab_lov_user_md_v3(v3);
5331 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_SPECIFIC)) {
5332 v3 = (struct lov_user_md_v3 *)v1;
5333 lustre_swab_lov_user_md_v3(v3);
5334 lustre_swab_lov_user_md_objects(v3->lmm_objects,
5335 v3->lmm_stripe_count);
5336 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_COMP_V1) ||
5337 v1->lmm_magic == __swab32(LOV_USER_MAGIC_SEL)) {
5338 comp_v1 = (struct lov_comp_md_v1 *)v1;
5339 lustre_swab_lov_comp_md_v1(comp_v1);
5342 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1 &&
5343 v1->lmm_magic != LOV_MAGIC_COMP_V1 &&
5344 v1->lmm_magic != LOV_MAGIC_SEL &&
5345 v1->lmm_magic != LOV_USER_MAGIC_SPECIFIC)
5348 if ((v1->lmm_magic == LOV_MAGIC_COMP_V1 ||
5349 v1->lmm_magic == LOV_MAGIC_SEL) &&
5350 !(ah && ah->dah_append_stripes)) {
5351 comp_v1 = (struct lov_comp_md_v1 *)v1;
5352 comp_cnt = comp_v1->lcm_entry_count;
5355 mirror_cnt = comp_v1->lcm_mirror_count + 1;
5363 /* realloc default comp entries if necessary */
5364 rc = lod_def_striping_comp_resize(lds, comp_cnt);
5368 lds->lds_def_comp_cnt = comp_cnt;
5369 lds->lds_def_striping_is_composite = composite;
5370 lds->lds_def_mirror_cnt = mirror_cnt;
5372 for (i = 0; i < comp_cnt; i++) {
5373 struct lod_layout_component *lod_comp;
5376 lod_comp = &lds->lds_def_comp_entries[i];
5378 * reset lod_comp values, llc_stripes is always NULL in
5379 * the default striping template, llc_pool will be reset
5382 memset(lod_comp, 0, offsetof(typeof(*lod_comp), llc_pool));
5385 v1 = (struct lov_user_md *)((char *)comp_v1 +
5386 comp_v1->lcm_entries[i].lcme_offset);
5387 lod_comp->llc_extent =
5388 comp_v1->lcm_entries[i].lcme_extent;
5389 /* We only inherit certain flags from the layout */
5390 lod_comp->llc_flags =
5391 comp_v1->lcm_entries[i].lcme_flags &
5392 LCME_TEMPLATE_FLAGS;
5395 if (!lov_pattern_supported(v1->lmm_pattern) &&
5396 !(v1->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
5397 lod_free_def_comp_entries(lds);
5401 CDEBUG(D_LAYOUT, DFID" stripe_count=%d stripe_size=%d stripe_offset=%d append_stripes=%d\n",
5402 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
5403 (int)v1->lmm_stripe_count, (int)v1->lmm_stripe_size,
5404 (int)v1->lmm_stripe_offset,
5405 ah ? ah->dah_append_stripes : 0);
5407 if (ah && ah->dah_append_stripes)
5408 lod_comp->llc_stripe_count = ah->dah_append_stripes;
5410 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
5411 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
5412 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
5413 lod_comp->llc_pattern = v1->lmm_pattern;
5416 if (ah && ah->dah_append_pool && ah->dah_append_pool[0]) {
5417 pool = ah->dah_append_pool;
5418 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
5419 /* XXX: sanity check here */
5420 v3 = (struct lov_user_md_v3 *) v1;
5421 if (v3->lmm_pool_name[0] != '\0')
5422 pool = v3->lmm_pool_name;
5424 lod_set_def_pool(lds, i, pool);
5425 if (v1->lmm_magic == LOV_USER_MAGIC_SPECIFIC) {
5426 v3 = (struct lov_user_md_v3 *)v1;
5427 rc = lod_comp_copy_ost_lists(lod_comp, v3);
5430 } else if (lod_comp->llc_ostlist.op_array &&
5431 lod_comp->llc_ostlist.op_count) {
5432 for (j = 0; j < lod_comp->llc_ostlist.op_count; j++)
5433 lod_comp->llc_ostlist.op_array[j] = -1;
5434 lod_comp->llc_ostlist.op_count = 0;
5438 lds->lds_def_striping_set = 1;
5443 * Get default directory striping.
5445 * \param[in] env execution environment
5446 * \param[in] lo object
5447 * \param[out] lds default striping
5449 * \retval 0 on success
5450 * \retval negative if failed
5452 static int lod_get_default_lmv_striping(const struct lu_env *env,
5453 struct lod_object *lo,
5454 struct lod_default_striping *lds)
5456 struct lmv_user_md *lmu;
5459 lds->lds_dir_def_striping_set = 0;
5461 rc = lod_get_default_lmv_ea(env, lo);
5465 if (rc >= (int)sizeof(*lmu)) {
5466 struct lod_thread_info *info = lod_env_info(env);
5468 lmu = info->lti_ea_store;
5470 lds->lds_dir_def_stripe_count =
5471 le32_to_cpu(lmu->lum_stripe_count);
5472 lds->lds_dir_def_stripe_offset =
5473 le32_to_cpu(lmu->lum_stripe_offset);
5474 lds->lds_dir_def_hash_type =
5475 le32_to_cpu(lmu->lum_hash_type);
5476 lds->lds_dir_def_max_inherit = lmu->lum_max_inherit;
5477 lds->lds_dir_def_max_inherit_rr = lmu->lum_max_inherit_rr;
5478 lds->lds_dir_def_striping_set = 1;
5485 * Get default striping in the object.
5487 * Get object default striping and default directory striping.
5489 * \param[in] env execution environment
5490 * \param[in] lo object
5491 * \param[out] lds default striping
5493 * \retval 0 on success
5494 * \retval negative if failed
5496 static int lod_get_default_striping(const struct lu_env *env,
5497 struct lod_object *lo,
5498 struct lod_default_striping *lds)
5502 rc = lod_get_default_lov_striping(env, lo, lds, NULL);
5503 if (lds->lds_def_striping_set) {
5504 struct lod_thread_info *info = lod_env_info(env);
5505 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5507 rc = lod_verify_striping(env, d, lo, &info->lti_buf, false);
5509 lds->lds_def_striping_set = 0;
5512 rc1 = lod_get_default_lmv_striping(env, lo, lds);
5513 if (rc == 0 && rc1 < 0)
5520 * Apply default striping on object.
5522 * If object striping pattern is not set, set to the one in default striping.
5523 * The default striping is from parent or fs.
5525 * \param[in] lo new object
5526 * \param[in] lds default striping
5527 * \param[in] mode new object's mode
5529 static void lod_striping_from_default(struct lod_object *lo,
5530 const struct lod_default_striping *lds,
5533 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5536 if (lds->lds_def_striping_set && S_ISREG(mode)) {
5537 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
5539 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
5540 lds->lds_def_comp_cnt);
5544 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
5545 if (lds->lds_def_mirror_cnt > 1)
5546 lo->ldo_flr_state = LCM_FL_RDONLY;
5548 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5549 struct lod_layout_component *obj_comp =
5550 &lo->ldo_comp_entries[i];
5551 struct lod_layout_component *def_comp =
5552 &lds->lds_def_comp_entries[i];
5554 CDEBUG(D_LAYOUT, "Inherit from default: flags=%#x "
5555 "size=%hu nr=%u offset=%u pattern=%#x pool=%s\n",
5556 def_comp->llc_flags,
5557 def_comp->llc_stripe_size,
5558 def_comp->llc_stripe_count,
5559 def_comp->llc_stripe_offset,
5560 def_comp->llc_pattern,
5561 def_comp->llc_pool ?: "");
5563 *obj_comp = *def_comp;
5564 if (def_comp->llc_pool != NULL) {
5565 /* pointer was copied from def_comp */
5566 obj_comp->llc_pool = NULL;
5567 lod_obj_set_pool(lo, i, def_comp->llc_pool);
5571 if (def_comp->llc_ostlist.op_array &&
5572 def_comp->llc_ostlist.op_count) {
5573 OBD_ALLOC(obj_comp->llc_ostlist.op_array,
5574 obj_comp->llc_ostlist.op_size);
5575 if (!obj_comp->llc_ostlist.op_array)
5577 memcpy(obj_comp->llc_ostlist.op_array,
5578 def_comp->llc_ostlist.op_array,
5579 obj_comp->llc_ostlist.op_size);
5580 } else if (def_comp->llc_ostlist.op_array) {
5581 obj_comp->llc_ostlist.op_array = NULL;
5585 * Don't initialize these fields for plain layout
5586 * (v1/v3) here, they are inherited in the order of
5587 * 'parent' -> 'fs default (root)' -> 'global default
5588 * values for stripe_count & stripe_size'.
5590 * see lod_ah_init().
5592 if (!lo->ldo_is_composite)
5595 lod_adjust_stripe_info(obj_comp, desc, 0);
5597 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
5598 if (lo->ldo_dir_stripe_count == 0)
5599 lo->ldo_dir_stripe_count =
5600 lds->lds_dir_def_stripe_count;
5601 if (lo->ldo_dir_stripe_offset == -1)
5602 lo->ldo_dir_stripe_offset =
5603 lds->lds_dir_def_stripe_offset;
5604 if (lo->ldo_dir_hash_type == 0)
5605 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
5607 CDEBUG(D_LAYOUT, "striping from default dir: count:%hu, "
5608 "offset:%u, hash_type:%u\n",
5609 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
5610 lo->ldo_dir_hash_type);
5614 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root,
5617 struct lod_layout_component *lod_comp;
5619 if (lo->ldo_comp_cnt == 0)
5622 if (lo->ldo_is_composite)
5625 lod_comp = &lo->ldo_comp_entries[0];
5627 if (lod_comp->llc_stripe_count <= 0 ||
5628 lod_comp->llc_stripe_size <= 0)
5631 if (from_root && (lod_comp->llc_pool == NULL ||
5632 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
5635 if (append_pool && append_pool[0])
5642 * Implementation of dt_object_operations::do_ah_init.
5644 * This method is used to make a decision on the striping configuration for the
5645 * object being created. It can be taken from the \a parent object if it exists,
5646 * or filesystem's default. The resulting configuration (number of stripes,
5647 * stripe size/offset, pool name, etc) is stored in the object itself and will
5648 * be used by the methods like ->doo_declare_create().
5650 * \see dt_object_operations::do_ah_init() in the API description for details.
5652 static void lod_ah_init(const struct lu_env *env,
5653 struct dt_allocation_hint *ah,
5654 struct dt_object *parent,
5655 struct dt_object *child,
5658 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
5659 struct lod_thread_info *info = lod_env_info(env);
5660 struct lod_default_striping *lds = lod_lds_buf_get(env);
5661 struct dt_object *nextp = NULL;
5662 struct dt_object *nextc;
5663 struct lod_object *lp = NULL;
5664 struct lod_object *lc;
5665 struct lov_desc *desc;
5666 struct lod_layout_component *lod_comp;
5672 if (ah->dah_append_stripes == -1)
5673 ah->dah_append_stripes =
5674 d->lod_ost_descs.ltd_lov_desc.ld_tgt_count;
5676 if (likely(parent)) {
5677 nextp = dt_object_child(parent);
5678 lp = lod_dt_obj(parent);
5681 nextc = dt_object_child(child);
5682 lc = lod_dt_obj(child);
5684 LASSERT(!lod_obj_is_striped(child));
5685 /* default layout template may have been set on the regular file
5686 * when this is called from mdd_create_data() */
5687 if (S_ISREG(child_mode))
5688 lod_free_comp_entries(lc);
5690 if (!dt_object_exists(nextc))
5691 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
5693 if (S_ISDIR(child_mode)) {
5694 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
5696 /* other default values are 0 */
5697 lc->ldo_dir_stripe_offset = -1;
5699 /* no default striping configuration is needed for
5702 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5703 le32_to_cpu(lum1->lum_magic) == LMV_MAGIC_FOREIGN) {
5704 lc->ldo_is_foreign = true;
5705 /* keep stripe_count 0 and stripe_offset -1 */
5706 CDEBUG(D_INFO, "no default striping for foreign dir\n");
5710 if (likely(lp != NULL))
5711 lod_get_default_striping(env, lp, lds);
5713 /* It should always honour the specified stripes */
5714 /* Note: old client (< 2.7)might also do lfs mkdir, whose EA
5715 * will have old magic. In this case, we should ignore the
5716 * stripe count and try to create dir by default stripe.
5718 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5719 (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
5720 le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC)) {
5721 lc->ldo_dir_stripe_count =
5722 le32_to_cpu(lum1->lum_stripe_count);
5723 lc->ldo_dir_stripe_offset =
5724 le32_to_cpu(lum1->lum_stripe_offset);
5725 lc->ldo_dir_hash_type =
5726 le32_to_cpu(lum1->lum_hash_type);
5728 "set dirstripe: count %hu, offset %d, hash %u\n",
5729 lc->ldo_dir_stripe_count,
5730 (int)lc->ldo_dir_stripe_offset,
5731 lc->ldo_dir_hash_type);
5733 if (d->lod_mdt_descs.ltd_lmv_desc.ld_active_tgt_count &&
5734 lc->ldo_dir_stripe_count < 2 &&
5735 lum1->lum_max_inherit != LMV_INHERIT_NONE) {
5736 /* when filesystem-wide default LMV is set, dirs
5737 * will be created on MDT by space usage, but if
5738 * dir is created with "lfs mkdir -c 1 ...", its
5739 * subdirs should be kept on the same MDT. To
5740 * guarantee this, set default LMV for such dir.
5742 lds->lds_dir_def_stripe_count =
5743 le32_to_cpu(lum1->lum_stripe_count);
5744 /* if "-1" stripe offset is set, save current
5745 * MDT index in default LMV.
5747 if (le32_to_cpu(lum1->lum_stripe_offset) ==
5749 lds->lds_dir_def_stripe_offset =
5750 lod2lu_dev(d)->ld_site->ld_seq_site->ss_node_id;
5752 lds->lds_dir_def_stripe_offset =
5753 le32_to_cpu(lum1->lum_stripe_offset);
5754 lds->lds_dir_def_hash_type =
5755 le32_to_cpu(lum1->lum_hash_type);
5756 lds->lds_dir_def_max_inherit =
5757 lum1->lum_max_inherit;
5758 /* it will be decreased by 1 later in setting */
5759 if (lum1->lum_max_inherit >= LMV_INHERIT_END &&
5760 lum1->lum_max_inherit < LMV_INHERIT_MAX)
5761 lds->lds_dir_def_max_inherit++;
5762 lds->lds_dir_def_max_inherit_rr =
5763 lum1->lum_max_inherit_rr;
5764 lds->lds_dir_def_striping_set = 1;
5765 /* don't inherit LOV from ROOT */
5766 if (lds->lds_def_striping_set &&
5767 fid_is_root(lod_object_fid(lp)))
5768 lds->lds_def_striping_set = 0;
5769 lc->ldo_def_striping = lds;
5770 } else if (lds->lds_def_striping_set &&
5771 !fid_is_root(lod_object_fid(lp))) {
5772 /* don't inherit default LMV for "lfs mkdir" */
5773 lds->lds_dir_def_striping_set = 0;
5774 lc->ldo_def_striping = lds;
5777 /* inherit default striping except ROOT */
5778 if ((lds->lds_def_striping_set ||
5779 lds->lds_dir_def_striping_set) &&
5780 !fid_is_root(lod_object_fid(lp)))
5781 lc->ldo_def_striping = lds;
5783 /* transfer defaults LMV to new directory */
5784 lod_striping_from_default(lc, lds, child_mode);
5786 /* set count 0 to create normal directory */
5787 if (lc->ldo_dir_stripe_count == 1)
5788 lc->ldo_dir_stripe_count = 0;
5791 /* shrink the stripe count to max_mdt_stripecount if it is -1
5792 * and max_mdt_stripecount is not 0
5794 if (lc->ldo_dir_stripe_count == (__u16)(-1) &&
5795 d->lod_max_mdt_stripecount)
5796 lc->ldo_dir_stripe_count = d->lod_max_mdt_stripecount;
5798 /* shrink the stripe_count to the avaible MDT count */
5799 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
5800 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)) {
5801 lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
5802 if (lc->ldo_dir_stripe_count == 1)
5803 lc->ldo_dir_stripe_count = 0;
5806 if (!(lc->ldo_dir_hash_type & LMV_HASH_TYPE_MASK))
5807 lc->ldo_dir_hash_type |=
5808 d->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
5810 CDEBUG(D_INFO, "final dir stripe [%hu %d %u]\n",
5811 lc->ldo_dir_stripe_count,
5812 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
5817 /* child object regular file*/
5819 if (!lod_object_will_be_striped(S_ISREG(child_mode),
5820 lu_object_fid(&child->do_lu)))
5823 /* If object is going to be striped over OSTs, transfer default
5824 * striping information to the child, so that we can use it
5825 * during declaration and creation.
5827 * Try from the parent first.
5829 if (likely(lp != NULL)) {
5830 rc = lod_get_default_lov_striping(env, lp, lds, ah);
5831 if (rc == 0 && lds->lds_def_striping_set) {
5832 rc = lod_verify_striping(env, d, lp, &info->lti_buf,
5835 lod_striping_from_default(lc, lds, child_mode);
5839 /* Initialize lod_device::lod_md_root object reference */
5840 if (d->lod_md_root == NULL) {
5841 struct dt_object *root;
5842 struct lod_object *lroot;
5844 lu_root_fid(&info->lti_fid);
5845 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
5846 if (!IS_ERR(root)) {
5847 lroot = lod_dt_obj(root);
5849 spin_lock(&d->lod_lock);
5850 if (d->lod_md_root != NULL)
5851 dt_object_put(env, &d->lod_md_root->ldo_obj);
5852 d->lod_md_root = lroot;
5853 spin_unlock(&d->lod_lock);
5857 /* try inherit layout from the root object (fs default) when:
5858 * - parent does not have default layout; or
5859 * - parent has plain(v1/v3) default layout, and some attributes
5860 * are not specified in the default layout;
5862 if (d->lod_md_root != NULL &&
5863 lod_need_inherit_more(lc, true, ah->dah_append_pool)) {
5864 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds,
5866 if (rc || !lds->lds_def_striping_set)
5869 rc = lod_verify_striping(env, d, d->lod_md_root, &info->lti_buf,
5874 if (lc->ldo_comp_cnt == 0) {
5875 lod_striping_from_default(lc, lds, child_mode);
5876 } else if (!lds->lds_def_striping_is_composite) {
5877 struct lod_layout_component *def_comp;
5879 LASSERT(!lc->ldo_is_composite);
5880 lod_comp = &lc->ldo_comp_entries[0];
5881 def_comp = &lds->lds_def_comp_entries[0];
5883 if (lod_comp->llc_stripe_count <= 0)
5884 lod_comp->llc_stripe_count =
5885 def_comp->llc_stripe_count;
5886 if (lod_comp->llc_stripe_size <= 0)
5887 lod_comp->llc_stripe_size =
5888 def_comp->llc_stripe_size;
5889 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT &&
5890 (!lod_comp->llc_pool || !lod_comp->llc_pool[0]))
5891 lod_comp->llc_stripe_offset =
5892 def_comp->llc_stripe_offset;
5893 if (lod_comp->llc_pool == NULL)
5894 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
5899 * fs default striping may not be explicitly set, or historically set
5900 * in config log, use them.
5902 if (lod_need_inherit_more(lc, false, ah->dah_append_pool)) {
5903 if (lc->ldo_comp_cnt == 0) {
5904 rc = lod_alloc_comp_entries(lc, 0, 1);
5906 /* fail to allocate memory, will create a
5907 * non-striped file. */
5909 lc->ldo_is_composite = 0;
5910 lod_comp = &lc->ldo_comp_entries[0];
5911 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
5913 LASSERT(!lc->ldo_is_composite);
5914 lod_comp = &lc->ldo_comp_entries[0];
5915 desc = &d->lod_ost_descs.ltd_lov_desc;
5916 lod_adjust_stripe_info(lod_comp, desc, ah->dah_append_stripes);
5917 if (ah->dah_append_pool && ah->dah_append_pool[0])
5918 lod_obj_set_pool(lc, 0, ah->dah_append_pool);
5925 * Size initialization on late striping.
5927 * Propagate the size of a truncated object to a deferred striping.
5928 * This function handles a special case when truncate was done on a
5929 * non-striped object and now while the striping is being created
5930 * we can't lose that size, so we have to propagate it to the stripes
5933 * \param[in] env execution environment
5934 * \param[in] dt object
5935 * \param[in] th transaction handle
5937 * \retval 0 on success
5938 * \retval negative if failed
5940 static int lod_declare_init_size(const struct lu_env *env,
5941 struct dt_object *dt, struct thandle *th)
5943 struct dt_object *next = dt_object_child(dt);
5944 struct lod_object *lo = lod_dt_obj(dt);
5945 struct dt_object **objects = NULL;
5946 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
5947 uint64_t size, offs;
5948 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
5949 struct lu_extent size_ext;
5952 if (!lod_obj_is_striped(dt))
5955 rc = dt_attr_get(env, next, attr);
5956 LASSERT(attr->la_valid & LA_SIZE);
5960 size = attr->la_size;
5964 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
5965 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5966 struct lod_layout_component *lod_comp;
5967 struct lu_extent *extent;
5969 lod_comp = &lo->ldo_comp_entries[i];
5971 if (lod_comp->llc_stripe == NULL)
5974 extent = &lod_comp->llc_extent;
5975 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
5976 if (!lo->ldo_is_composite ||
5977 lu_extent_is_overlapped(extent, &size_ext)) {
5978 objects = lod_comp->llc_stripe;
5979 stripe_count = lod_comp->llc_stripe_count;
5980 stripe_size = lod_comp->llc_stripe_size;
5983 if (stripe_count == 0)
5986 LASSERT(objects != NULL && stripe_size != 0);
5987 do_div(size, stripe_size);
5988 stripe = do_div(size, stripe_count);
5989 LASSERT(objects[stripe] != NULL);
5991 size = size * stripe_size;
5992 offs = attr->la_size;
5993 size += do_div(offs, stripe_size);
5995 attr->la_valid = LA_SIZE;
5996 attr->la_size = size;
5998 rc = lod_sub_declare_attr_set(env, objects[stripe],
6007 * Declare creation of striped object.
6009 * The function declares creation stripes for a regular object. The function
6010 * also declares whether the stripes will be created with non-zero size if
6011 * previously size was set non-zero on the master object. If object \a dt is
6012 * not local, then only fully defined striping can be applied in \a lovea.
6013 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
6016 * \param[in] env execution environment
6017 * \param[in] dt object
6018 * \param[in] attr attributes the stripes will be created with
6019 * \param[in] lovea a buffer containing striping description
6020 * \param[in] th transaction handle
6022 * \retval 0 on success
6023 * \retval negative if failed
6025 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
6026 struct lu_attr *attr,
6027 const struct lu_buf *lovea, struct thandle *th)
6029 struct lod_thread_info *info = lod_env_info(env);
6030 struct dt_object *next = dt_object_child(dt);
6031 struct lod_object *lo = lod_dt_obj(dt);
6035 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
6036 GOTO(out, rc = -ENOMEM);
6038 if (!dt_object_remote(next)) {
6039 /* choose OST and generate appropriate objects */
6040 rc = lod_prepare_create(env, lo, attr, lovea, th);
6045 * declare storage for striping data
6047 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6049 /* LOD can not choose OST objects for remote objects, i.e.
6050 * stripes must be ready before that. Right now, it can only
6051 * happen during migrate, i.e. migrate process needs to create
6052 * remote regular file (mdd_migrate_create), then the migrate
6053 * process will provide stripeEA. */
6054 LASSERT(lovea != NULL);
6055 info->lti_buf = *lovea;
6058 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
6059 XATTR_NAME_LOV, 0, th);
6064 * if striping is created with local object's size > 0,
6065 * we have to propagate this size to specific object
6066 * the case is possible only when local object was created previously
6068 if (dt_object_exists(next))
6069 rc = lod_declare_init_size(env, dt, th);
6072 /* failed to create striping or to set initial size, let's reset
6073 * config so that others don't get confused */
6075 lod_striping_free(env, lo);
6081 * Whether subdirectories under \a dt should be created on MDTs by space QoS
6083 * If LMV_HASH_FLAG_SPACE is set on directory default layout, its subdirectories
6084 * should be created on MDT by space QoS.
6086 * \param[in] env execution environment
6087 * \param[in] dev lu device
6088 * \param[in] dt object
6090 * \retval 1 if directory should create subdir by space usage
6092 * \retval -ev if failed
6094 static inline int dt_object_qos_mkdir(const struct lu_env *env,
6095 struct lu_device *dev,
6096 struct dt_object *dt)
6098 struct lod_thread_info *info = lod_env_info(env);
6099 struct lu_object *obj;
6100 struct lod_object *lo;
6101 struct lmv_user_md *lmu;
6104 obj = lu_object_find_slice(env, dev, lu_object_fid(&dt->do_lu), NULL);
6106 return PTR_ERR(obj);
6108 lo = lu2lod_obj(obj);
6110 rc = lod_get_default_lmv_ea(env, lo);
6111 dt_object_put(env, dt);
6115 if (rc < (int)sizeof(*lmu))
6118 lmu = info->lti_ea_store;
6119 return le32_to_cpu(lmu->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
6123 * Implementation of dt_object_operations::do_declare_create.
6125 * The method declares creation of a new object. If the object will be striped,
6126 * then helper functions are called to find FIDs for the stripes, declare
6127 * creation of the stripes and declare initialization of the striping
6128 * information to be stored in the master object.
6130 * \see dt_object_operations::do_declare_create() in the API description
6133 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
6134 struct lu_attr *attr,
6135 struct dt_allocation_hint *hint,
6136 struct dt_object_format *dof, struct thandle *th)
6138 struct dt_object *next = dt_object_child(dt);
6139 struct lod_object *lo = lod_dt_obj(dt);
6148 * first of all, we declare creation of local object
6150 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
6155 * it's lod_ah_init() that has decided the object will be striped
6157 if (dof->dof_type == DFT_REGULAR) {
6158 /* callers don't want stripes */
6159 /* XXX: all tricky interactions with ->ah_make_hint() decided
6160 * to use striping, then ->declare_create() behaving differently
6161 * should be cleaned */
6162 if (dof->u.dof_reg.striped != 0)
6163 rc = lod_declare_striped_create(env, dt, attr,
6165 } else if (dof->dof_type == DFT_DIR) {
6166 struct seq_server_site *ss;
6167 struct lu_buf buf = { NULL };
6168 struct lu_buf *lmu = NULL;
6170 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
6172 /* If the parent has default stripeEA, and client
6173 * did not find it before sending create request,
6174 * then MDT will return -EREMOTE, and client will
6175 * retrieve the default stripeEA and re-create the
6178 * Note: if dah_eadata != NULL, it means creating the
6179 * striped directory with specified stripeEA, then it
6180 * should ignore the default stripeEA */
6181 if (hint != NULL && hint->dah_eadata == NULL) {
6182 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
6183 GOTO(out, rc = -EREMOTE);
6185 if (lo->ldo_dir_stripe_offset != LMV_OFFSET_DEFAULT &&
6186 lo->ldo_dir_stripe_offset != ss->ss_node_id) {
6187 struct lod_device *lod;
6188 struct lu_tgt_desc *mdt = NULL;
6189 bool found_mdt = false;
6191 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6192 lod_foreach_mdt(lod, mdt) {
6193 if (mdt->ltd_index ==
6194 lo->ldo_dir_stripe_offset) {
6200 /* If the MDT indicated by stripe_offset can be
6201 * found, then tell client to resend the create
6202 * request to the correct MDT, otherwise return
6203 * error to client */
6205 GOTO(out, rc = -EREMOTE);
6207 GOTO(out, rc = -EINVAL);
6209 } else if (hint && hint->dah_eadata) {
6211 lmu->lb_buf = (void *)hint->dah_eadata;
6212 lmu->lb_len = hint->dah_eadata_len;
6215 rc = lod_declare_dir_striping_create(env, dt, attr, lmu, dof,
6219 /* failed to create striping or to set initial size, let's reset
6220 * config so that others don't get confused */
6222 lod_striping_free(env, lo);
6227 * Generate component ID for new created component.
6229 * \param[in] lo LOD object
6230 * \param[in] comp_idx index of ldo_comp_entries
6232 * \retval component ID on success
6233 * \retval LCME_ID_INVAL on failure
6235 static __u32 lod_gen_component_id(struct lod_object *lo,
6236 int mirror_id, int comp_idx)
6238 struct lod_layout_component *lod_comp;
6239 __u32 id, start, end;
6242 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
6244 lod_obj_inc_layout_gen(lo);
6245 id = lo->ldo_layout_gen;
6246 if (likely(id <= SEQ_ID_MAX))
6247 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
6249 /* Layout generation wraps, need to check collisions. */
6250 start = id & SEQ_ID_MASK;
6253 for (id = start; id <= end; id++) {
6254 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6255 lod_comp = &lo->ldo_comp_entries[i];
6256 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
6259 /* Found the ununsed ID */
6260 if (i == lo->ldo_comp_cnt)
6261 RETURN(pflr_id(mirror_id, id));
6263 if (end == LCME_ID_MAX) {
6265 end = min(lo->ldo_layout_gen & LCME_ID_MASK,
6266 (__u32)(LCME_ID_MAX - 1));
6270 RETURN(LCME_ID_INVAL);
6274 * Creation of a striped regular object.
6276 * The function is called to create the stripe objects for a regular
6277 * striped file. This can happen at the initial object creation or
6278 * when the caller asks LOD to do so using ->do_xattr_set() method
6279 * (so called late striping). Notice all the information are already
6280 * prepared in the form of the list of objects (ldo_stripe field).
6281 * This is done during declare phase.
6283 * \param[in] env execution environment
6284 * \param[in] dt object
6285 * \param[in] attr attributes the stripes will be created with
6286 * \param[in] dof format of stripes (see OSD API description)
6287 * \param[in] th transaction handle
6289 * \retval 0 on success
6290 * \retval negative if failed
6292 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
6293 struct lu_attr *attr, struct dt_object_format *dof,
6296 struct lod_layout_component *lod_comp;
6297 struct lod_object *lo = lod_dt_obj(dt);
6302 mutex_lock(&lo->ldo_layout_mutex);
6304 LASSERT((lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL) ||
6305 lo->ldo_is_foreign);
6307 mirror_id = 0; /* non-flr file's mirror_id is 0 */
6308 if (lo->ldo_mirror_count > 1) {
6309 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6310 lod_comp = &lo->ldo_comp_entries[i];
6311 if (lod_comp->llc_id != LCME_ID_INVAL &&
6312 mirror_id_of(lod_comp->llc_id) > mirror_id)
6313 mirror_id = mirror_id_of(lod_comp->llc_id);
6317 /* create all underlying objects */
6318 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6319 lod_comp = &lo->ldo_comp_entries[i];
6321 if (lod_comp->llc_id == LCME_ID_INVAL) {
6322 /* only the component of FLR layout with more than 1
6323 * mirror has mirror ID in its component ID.
6325 if (lod_comp->llc_extent.e_start == 0 &&
6326 lo->ldo_mirror_count > 1)
6329 lod_comp->llc_id = lod_gen_component_id(lo,
6331 if (lod_comp->llc_id == LCME_ID_INVAL)
6332 GOTO(out, rc = -ERANGE);
6335 if (lod_comp_inited(lod_comp))
6338 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
6339 lod_comp_set_init(lod_comp);
6341 if (lov_pattern(lod_comp->llc_pattern) == LOV_PATTERN_MDT)
6342 lod_comp_set_init(lod_comp);
6344 if (lod_comp->llc_stripe == NULL)
6347 LASSERT(lod_comp->llc_stripe_count);
6348 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6349 struct dt_object *object = lod_comp->llc_stripe[j];
6350 LASSERT(object != NULL);
6351 rc = lod_sub_create(env, object, attr, NULL, dof, th);
6355 lod_comp_set_init(lod_comp);
6358 rc = lod_fill_mirrors(lo);
6362 lo->ldo_comp_cached = 1;
6364 rc = lod_generate_and_set_lovea(env, lo, th);
6368 mutex_unlock(&lo->ldo_layout_mutex);
6373 lod_striping_free_nolock(env, lo);
6374 mutex_unlock(&lo->ldo_layout_mutex);
6379 static inline bool lod_obj_is_dom(struct dt_object *dt)
6381 struct lod_object *lo = lod_dt_obj(dt);
6383 if (!dt_object_exists(dt_object_child(dt)))
6386 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
6389 if (!lo->ldo_comp_cnt)
6392 return (lov_pattern(lo->ldo_comp_entries[0].llc_pattern) ==
6397 * Implementation of dt_object_operations::do_create.
6399 * If any of preceeding methods (like ->do_declare_create(),
6400 * ->do_ah_init(), etc) chose to create a striped object,
6401 * then this method will create the master and the stripes.
6403 * \see dt_object_operations::do_create() in the API description for details.
6405 static int lod_create(const struct lu_env *env, struct dt_object *dt,
6406 struct lu_attr *attr, struct dt_allocation_hint *hint,
6407 struct dt_object_format *dof, struct thandle *th)
6412 /* create local object */
6413 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
6417 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
6418 (lod_obj_is_striped(dt) || lod_obj_is_dom(dt)) &&
6419 dof->u.dof_reg.striped != 0) {
6420 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
6421 rc = lod_striped_create(env, dt, attr, dof, th);
6428 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
6429 struct dt_object *dt, struct thandle *th,
6430 int comp_idx, int stripe_idx,
6431 struct lod_obj_stripe_cb_data *data)
6433 if (data->locd_declare)
6434 return lod_sub_declare_destroy(env, dt, th);
6436 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
6437 stripe_idx == cfs_fail_val)
6438 return lod_sub_destroy(env, dt, th);
6444 * Implementation of dt_object_operations::do_declare_destroy.
6446 * If the object is a striped directory, then the function declares reference
6447 * removal from the master object (this is an index) to the stripes and declares
6448 * destroy of all the stripes. In all the cases, it declares an intention to
6449 * destroy the object itself.
6451 * \see dt_object_operations::do_declare_destroy() in the API description
6454 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
6457 struct dt_object *next = dt_object_child(dt);
6458 struct lod_object *lo = lod_dt_obj(dt);
6459 struct lod_thread_info *info = lod_env_info(env);
6460 struct dt_object *stripe;
6461 char *stripe_name = info->lti_key;
6467 * load striping information, notice we don't do this when object
6468 * is being initialized as we don't need this information till
6469 * few specific cases like destroy, chown
6471 rc = lod_striping_load(env, lo);
6475 /* declare destroy for all underlying objects */
6476 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6477 rc = next->do_ops->do_index_try(env, next,
6478 &dt_directory_features);
6482 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6483 stripe = lo->ldo_stripe[i];
6487 rc = lod_sub_declare_ref_del(env, next, th);
6491 snprintf(stripe_name, sizeof(info->lti_key),
6493 PFID(lu_object_fid(&stripe->do_lu)), i);
6494 rc = lod_sub_declare_delete(env, next,
6495 (const struct dt_key *)stripe_name, th);
6502 * we declare destroy for the local object
6504 rc = lod_sub_declare_destroy(env, next, th);
6508 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
6509 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
6512 if (!lod_obj_is_striped(dt))
6515 /* declare destroy all striped objects */
6516 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6517 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6518 stripe = lo->ldo_stripe[i];
6522 if (!dt_object_exists(stripe))
6525 rc = lod_sub_declare_ref_del(env, stripe, th);
6529 rc = lod_sub_declare_destroy(env, stripe, th);
6534 struct lod_obj_stripe_cb_data data = { { 0 } };
6536 data.locd_declare = true;
6537 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
6538 rc = lod_obj_for_each_stripe(env, lo, th, &data);
6545 * Implementation of dt_object_operations::do_destroy.
6547 * If the object is a striped directory, then the function removes references
6548 * from the master object (this is an index) to the stripes and destroys all
6549 * the stripes. In all the cases, the function destroys the object itself.
6551 * \see dt_object_operations::do_destroy() in the API description for details.
6553 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
6556 struct dt_object *next = dt_object_child(dt);
6557 struct lod_object *lo = lod_dt_obj(dt);
6558 struct lod_thread_info *info = lod_env_info(env);
6559 char *stripe_name = info->lti_key;
6560 struct dt_object *stripe;
6566 /* destroy sub-stripe of master object */
6567 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6568 rc = next->do_ops->do_index_try(env, next,
6569 &dt_directory_features);
6573 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6574 stripe = lo->ldo_stripe[i];
6578 rc = lod_sub_ref_del(env, next, th);
6582 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
6583 PFID(lu_object_fid(&stripe->do_lu)), i);
6585 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
6586 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
6587 PFID(lu_object_fid(&stripe->do_lu)));
6589 rc = lod_sub_delete(env, next,
6590 (const struct dt_key *)stripe_name, th);
6596 rc = lod_sub_destroy(env, next, th);
6600 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
6601 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
6604 if (!lod_obj_is_striped(dt))
6607 /* destroy all striped objects */
6608 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6609 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6610 stripe = lo->ldo_stripe[i];
6614 if (!dt_object_exists(stripe))
6617 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
6618 i == cfs_fail_val) {
6619 dt_write_lock(env, stripe, DT_TGT_CHILD);
6620 rc = lod_sub_ref_del(env, stripe, th);
6621 dt_write_unlock(env, stripe);
6625 rc = lod_sub_destroy(env, stripe, th);
6631 struct lod_obj_stripe_cb_data data = { { 0 } };
6633 data.locd_declare = false;
6634 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
6635 rc = lod_obj_for_each_stripe(env, lo, th, &data);
6642 * Implementation of dt_object_operations::do_declare_ref_add.
6644 * \see dt_object_operations::do_declare_ref_add() in the API description
6647 static int lod_declare_ref_add(const struct lu_env *env,
6648 struct dt_object *dt, struct thandle *th)
6650 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
6654 * Implementation of dt_object_operations::do_ref_add.
6656 * \see dt_object_operations::do_ref_add() in the API description for details.
6658 static int lod_ref_add(const struct lu_env *env,
6659 struct dt_object *dt, struct thandle *th)
6661 return lod_sub_ref_add(env, dt_object_child(dt), th);
6665 * Implementation of dt_object_operations::do_declare_ref_del.
6667 * \see dt_object_operations::do_declare_ref_del() in the API description
6670 static int lod_declare_ref_del(const struct lu_env *env,
6671 struct dt_object *dt, struct thandle *th)
6673 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
6677 * Implementation of dt_object_operations::do_ref_del
6679 * \see dt_object_operations::do_ref_del() in the API description for details.
6681 static int lod_ref_del(const struct lu_env *env,
6682 struct dt_object *dt, struct thandle *th)
6684 return lod_sub_ref_del(env, dt_object_child(dt), th);
6688 * Implementation of dt_object_operations::do_object_sync.
6690 * \see dt_object_operations::do_object_sync() in the API description
6693 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
6694 __u64 start, __u64 end)
6696 return dt_object_sync(env, dt_object_child(dt), start, end);
6700 * Implementation of dt_object_operations::do_object_unlock.
6702 * Used to release LDLM lock(s).
6704 * \see dt_object_operations::do_object_unlock() in the API description
6707 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
6708 struct ldlm_enqueue_info *einfo,
6709 union ldlm_policy_data *policy)
6711 struct lod_object *lo = lod_dt_obj(dt);
6712 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
6713 int slave_locks_size;
6717 if (slave_locks == NULL)
6720 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
6721 /* Note: for remote lock for single stripe dir, MDT will cancel
6722 * the lock by lockh directly */
6723 LASSERT(!dt_object_remote(dt_object_child(dt)));
6725 /* locks were unlocked in MDT layer */
6726 for (i = 0; i < slave_locks->ha_count; i++)
6727 LASSERT(!lustre_handle_is_used(&slave_locks->ha_handles[i]));
6730 * NB, ha_count may not equal to ldo_dir_stripe_count, because dir
6731 * layout may change, e.g., shrink dir layout after migration.
6733 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6734 if (lo->ldo_stripe[i])
6735 dt_invalidate(env, lo->ldo_stripe[i]);
6738 slave_locks_size = offsetof(typeof(*slave_locks),
6739 ha_handles[slave_locks->ha_count]);
6740 OBD_FREE(slave_locks, slave_locks_size);
6741 einfo->ei_cbdata = NULL;
6747 * Implementation of dt_object_operations::do_object_lock.
6749 * Used to get LDLM lock on the non-striped and striped objects.
6751 * \see dt_object_operations::do_object_lock() in the API description
6754 static int lod_object_lock(const struct lu_env *env,
6755 struct dt_object *dt,
6756 struct lustre_handle *lh,
6757 struct ldlm_enqueue_info *einfo,
6758 union ldlm_policy_data *policy)
6760 struct lod_object *lo = lod_dt_obj(dt);
6761 int slave_locks_size;
6762 struct lustre_handle_array *slave_locks = NULL;
6767 /* remote object lock */
6768 if (!einfo->ei_enq_slave) {
6769 LASSERT(dt_object_remote(dt));
6770 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
6774 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
6777 rc = lod_striping_load(env, lo);
6782 if (lo->ldo_dir_stripe_count <= 1)
6785 slave_locks_size = offsetof(typeof(*slave_locks),
6786 ha_handles[lo->ldo_dir_stripe_count]);
6787 /* Freed in lod_object_unlock */
6788 OBD_ALLOC(slave_locks, slave_locks_size);
6791 slave_locks->ha_count = lo->ldo_dir_stripe_count;
6793 /* striped directory lock */
6794 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6795 struct lustre_handle lockh;
6796 struct ldlm_res_id *res_id;
6797 struct dt_object *stripe;
6799 stripe = lo->ldo_stripe[i];
6803 res_id = &lod_env_info(env)->lti_res_id;
6804 fid_build_reg_res_name(lu_object_fid(&stripe->do_lu), res_id);
6805 einfo->ei_res_id = res_id;
6807 if (dt_object_remote(stripe)) {
6808 set_bit(i, (void *)slave_locks->ha_map);
6809 rc = dt_object_lock(env, stripe, &lockh, einfo, policy);
6811 struct ldlm_namespace *ns = einfo->ei_namespace;
6812 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
6813 ldlm_completion_callback completion = einfo->ei_cb_cp;
6814 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
6816 if (einfo->ei_mode == LCK_PW ||
6817 einfo->ei_mode == LCK_EX)
6818 dlmflags |= LDLM_FL_COS_INCOMPAT;
6820 LASSERT(ns != NULL);
6821 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS,
6822 policy, einfo->ei_mode,
6823 &dlmflags, blocking,
6825 NULL, 0, LVB_T_NONE,
6830 ldlm_lock_decref_and_cancel(
6831 &slave_locks->ha_handles[i],
6833 OBD_FREE(slave_locks, slave_locks_size);
6836 slave_locks->ha_handles[i] = lockh;
6838 einfo->ei_cbdata = slave_locks;
6844 * Implementation of dt_object_operations::do_invalidate.
6846 * \see dt_object_operations::do_invalidate() in the API description for details
6848 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
6850 return dt_invalidate(env, dt_object_child(dt));
6853 static int lod_declare_instantiate_components(const struct lu_env *env,
6854 struct lod_object *lo,
6858 struct lod_thread_info *info = lod_env_info(env);
6863 LASSERT(info->lti_count < lo->ldo_comp_cnt);
6865 for (i = 0; i < info->lti_count; i++) {
6866 rc = lod_qos_prep_create(env, lo, NULL, th,
6867 info->lti_comp_idx[i], reserve);
6873 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6874 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
6875 &info->lti_buf, XATTR_NAME_LOV, 0, th);
6882 * Check OSTs for an existing component for further extension
6884 * Checks if OSTs are still healthy and not out of space. Gets free space
6885 * on OSTs (relative to allocation watermark rmb_low) and compares to
6886 * the proposed new_end for this component.
6888 * Decides whether or not to extend a component on its current OSTs.
6890 * \param[in] env execution environment for this thread
6891 * \param[in] lo object we're checking
6892 * \param[in] index index of this component
6893 * \param[in] extension_size extension size for this component
6894 * \param[in] extent layout extent for requested operation
6895 * \param[in] comp_extent extension component extent
6896 * \param[in] write if this is write operation
6898 * \retval true - OK to extend on current OSTs
6899 * \retval false - do not extend on current OSTs
6901 static bool lod_sel_osts_allowed(const struct lu_env *env,
6902 struct lod_object *lo,
6903 int index, __u64 reserve,
6904 struct lu_extent *extent,
6905 struct lu_extent *comp_extent, int write)
6907 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[index];
6908 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6909 struct lod_thread_info *tinfo = lod_env_info(env);
6910 struct obd_statfs *sfs = &tinfo->lti_osfs;
6911 __u64 available = 0;
6917 LASSERT(lod_comp->llc_stripe_count != 0);
6919 lod_getref(&lod->lod_ost_descs);
6920 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
6921 int index = lod_comp->llc_ost_indices[i];
6922 struct lod_tgt_desc *ost = OST_TGT(lod, index);
6923 struct obd_statfs_info info = { 0 };
6924 int j, repeated = 0;
6928 /* Get the number of times this OST repeats in this component.
6929 * Note: inter-component repeats are not counted as this is
6930 * considered as a rare case: we try to not repeat OST in other
6931 * components if possible. */
6932 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6933 if (index != lod_comp->llc_ost_indices[j])
6936 /* already handled */
6942 if (j < lod_comp->llc_stripe_count)
6945 if (!test_bit(index, lod->lod_ost_bitmap)) {
6946 CDEBUG(D_LAYOUT, "ost %d no longer present\n", index);
6951 rc = dt_statfs_info(env, ost->ltd_tgt, sfs, &info);
6953 CDEBUG(D_LAYOUT, "statfs failed for ost %d, error %d\n",
6959 if (sfs->os_state & OS_STATFS_ENOSPC ||
6960 sfs->os_state & OS_STATFS_READONLY ||
6961 sfs->os_state & OS_STATFS_DEGRADED) {
6962 CDEBUG(D_LAYOUT, "ost %d is not availble for SEL "
6963 "extension, state %u\n", index, sfs->os_state);
6969 available = sfs->os_bavail * sfs->os_bsize;
6970 /* 'available' is relative to the allocation threshold */
6971 available -= (__u64) info.os_reserved_mb_low << 20;
6973 CDEBUG(D_LAYOUT, "ost %d lowwm: %d highwm: %d, "
6974 "%llu %% blocks available, %llu %% blocks free\n",
6975 index, info.os_reserved_mb_low, info.os_reserved_mb_high,
6976 (100ull * sfs->os_bavail) / sfs->os_blocks,
6977 (100ull * sfs->os_bfree) / sfs->os_blocks);
6979 if (reserve * repeated > available) {
6981 CDEBUG(D_LAYOUT, "low space on ost %d, available %llu "
6982 "< extension size %llu repeated %d\n", index,
6983 available, reserve, repeated);
6987 lod_putref(lod, &lod->lod_ost_descs);
6993 * Adjust extents after component removal
6995 * When we remove an extension component, we move the start of the next
6996 * component to match the start of the extension component, so no space is left
6999 * \param[in] env execution environment for this thread
7000 * \param[in] lo object
7001 * \param[in] max_comp layout component
7002 * \param[in] index index of this component
7004 * \retval 0 on success
7005 * \retval negative errno on error
7007 static void lod_sel_adjust_extents(const struct lu_env *env,
7008 struct lod_object *lo,
7009 int max_comp, int index)
7011 struct lod_layout_component *lod_comp = NULL;
7012 struct lod_layout_component *next = NULL;
7013 struct lod_layout_component *prev = NULL;
7014 __u64 new_start = 0;
7018 /* Extension space component */
7019 lod_comp = &lo->ldo_comp_entries[index];
7020 next = &lo->ldo_comp_entries[index + 1];
7021 prev = &lo->ldo_comp_entries[index - 1];
7023 LASSERT(lod_comp != NULL && prev != NULL && next != NULL);
7024 LASSERT(lod_comp->llc_flags & LCME_FL_EXTENSION);
7026 /* Previous is being removed */
7027 if (prev && prev->llc_id == LCME_ID_INVAL)
7028 new_start = prev->llc_extent.e_start;
7030 new_start = lod_comp->llc_extent.e_start;
7032 for (i = index + 1; i < max_comp; i++) {
7033 lod_comp = &lo->ldo_comp_entries[i];
7035 start = lod_comp->llc_extent.e_start;
7036 lod_comp->llc_extent.e_start = new_start;
7038 /* We only move zero length extendable components */
7039 if (!(start == lod_comp->llc_extent.e_end))
7042 LASSERT(!(lod_comp->llc_flags & LCME_FL_INIT));
7044 lod_comp->llc_extent.e_end = new_start;
7048 /* Calculate the proposed 'new end' for a component we're extending */
7049 static __u64 lod_extension_new_end(__u64 extension_size, __u64 extent_end,
7050 __u32 stripe_size, __u64 component_end,
7051 __u64 extension_end)
7055 LASSERT(extension_size != 0 && stripe_size != 0);
7057 /* Round up to extension size */
7058 if (extent_end == OBD_OBJECT_EOF) {
7059 new_end = OBD_OBJECT_EOF;
7061 /* Add at least extension_size to the previous component_end,
7062 * covering the req layout extent */
7063 new_end = max(extent_end - component_end, extension_size);
7064 new_end = roundup(new_end, extension_size);
7065 new_end += component_end;
7067 /* Component end must be min stripe size aligned */
7068 if (new_end % stripe_size) {
7069 CDEBUG(D_LAYOUT, "new component end is not aligned "
7070 "by the stripe size %u: [%llu, %llu) ext size "
7071 "%llu new end %llu, aligning\n",
7072 stripe_size, component_end, extent_end,
7073 extension_size, new_end);
7074 new_end = roundup(new_end, stripe_size);
7078 if (new_end < extent_end)
7079 new_end = OBD_OBJECT_EOF;
7082 /* Don't extend past the end of the extension component */
7083 if (new_end > extension_end)
7084 new_end = extension_end;
7090 * Calculate the exact reservation (per-OST extension_size) on the OSTs being
7091 * instantiated. It needs to be calculated in advance and taken into account at
7092 * the instantiation time, because otherwise lod_statfs_and_check() may consider
7093 * an OST as OK, but SEL needs its extension_size to fit the free space and the
7094 * OST may turn out to be low-on-space, thus inappropriate OST may be used and
7097 * \param[in] lod_comp lod component we are checking
7099 * \retval size to reserved on each OST of lod_comp's stripe.
7101 static __u64 lod_sel_stripe_reserved(struct lod_layout_component *lod_comp)
7103 /* extension_size is file level, so we must divide by stripe count to
7104 * compare it to available space on a single OST */
7105 return lod_comp->llc_stripe_size * SEL_UNIT_SIZE /
7106 lod_comp->llc_stripe_count;
7109 /* As lod_sel_handler() could be re-entered for the same component several
7110 * times, this is the data for the next call. Fields could be changed to
7111 * component indexes when needed, (e.g. if there is no need to instantiate
7112 * all the previous components up to the current position) to tell the caller
7113 * where to start over from. */
7120 * Process extent updates for a particular layout component
7122 * Handle layout updates for a particular extension space component touched by
7123 * a layout update operation. Core function of self-extending PFL feature.
7125 * In general, this function processes exactly *one* stage of an extension
7126 * operation, modifying the layout accordingly, then returns to the caller.
7127 * The caller is responsible for restarting processing with the new layout,
7128 * which may repeatedly return to this function until the extension updates
7131 * This function does one of a few things to the layout:
7132 * 1. Extends the component before the current extension space component to
7133 * allow it to accomodate the requested operation (if space/policy permit that
7134 * component to continue on its current OSTs)
7136 * 2. If extension of the existing component fails, we do one of two things:
7137 * a. If there is a component after the extension space, we remove the
7138 * extension space component, move the start of the next component down
7139 * accordingly, then notify the caller to restart processing w/the new
7141 * b. If there is no following component, we try repeating the current
7142 * component, creating a new component using the current one as a
7143 * template (keeping its stripe properties but not specific striping),
7144 * and try assigning striping for this component. If there is sufficient
7145 * free space on the OSTs chosen for this component, it is instantiated
7146 * and i/o continues there.
7148 * If there is not sufficient space on the new OSTs, we remove this new
7149 * component & extend the current component.
7151 * Note further that uninited components followed by extension space can be zero
7152 * length meaning that we will try to extend them before initializing them, and
7153 * if that fails, they will be removed without initialization.
7155 * 3. If we extend to/beyond the end of an extension space component, that
7156 * component is exhausted (all of its range has been given to real components),
7157 * so we remove it and restart processing.
7159 * \param[in] env execution environment for this thread
7160 * \param[in,out] lo object to update the layout of
7161 * \param[in] extent layout extent for requested operation, update
7162 * layout to fit this operation
7163 * \param[in] th transaction handle for this operation
7164 * \param[in,out] max_comp the highest comp for the portion of the layout
7165 * we are operating on (For FLR, the chosen
7166 * replica). Updated because we may remove
7168 * \param[in] index index of the extension space component we're
7170 * \param[in] write if this is write op
7171 * \param[in,out] force if the extension is to be forced; set here
7172 to force it on the 2nd call for the same
7175 * \retval 0 on success
7176 * \retval negative errno on error
7178 static int lod_sel_handler(const struct lu_env *env,
7179 struct lod_object *lo,
7180 struct lu_extent *extent,
7181 struct thandle *th, int *max_comp,
7182 int index, int write,
7183 struct sel_data *sd)
7185 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7186 struct lod_thread_info *info = lod_env_info(env);
7187 struct lod_layout_component *lod_comp;
7188 struct lod_layout_component *prev;
7189 struct lod_layout_component *next = NULL;
7190 __u64 extension_size, reserve;
7197 /* First component cannot be extension space */
7199 CERROR("%s: "DFID" first component cannot be extension space\n",
7200 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
7204 lod_comp = &lo->ldo_comp_entries[index];
7205 prev = &lo->ldo_comp_entries[index - 1];
7206 if ((index + 1) < *max_comp)
7207 next = &lo->ldo_comp_entries[index + 1];
7209 /* extension size uses the stripe size field as KiB */
7210 extension_size = lod_comp->llc_stripe_size * SEL_UNIT_SIZE;
7212 CDEBUG(D_LAYOUT, "prev start %llu, extension start %llu, extension end"
7213 " %llu, extension size %llu\n", prev->llc_extent.e_start,
7214 lod_comp->llc_extent.e_start, lod_comp->llc_extent.e_end,
7217 /* Two extension space components cannot be adjacent & extension space
7218 * components cannot be init */
7219 if ((prev->llc_flags & LCME_FL_EXTENSION) ||
7220 !(ergo(next, !(next->llc_flags & LCME_FL_EXTENSION))) ||
7221 lod_comp_inited(lod_comp)) {
7222 CERROR("%s: "DFID" invalid extension space components\n",
7223 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
7227 reserve = lod_sel_stripe_reserved(lod_comp);
7229 if (!prev->llc_stripe) {
7230 CDEBUG(D_LAYOUT, "Previous component not inited\n");
7231 info->lti_count = 1;
7232 info->lti_comp_idx[0] = index - 1;
7233 rc = lod_declare_instantiate_components(env, lo, th, reserve);
7234 /* ENOSPC tells us we can't use this component. If there is
7235 * a next or we are repeating, we either spill over (next) or
7236 * extend the original comp (repeat). Otherwise, return the
7237 * error to the user. */
7238 if (rc == -ENOSPC && (next || sd->sd_repeat))
7244 if (sd->sd_force == 0 && rc == 0)
7245 rc = !lod_sel_osts_allowed(env, lo, index - 1, reserve, extent,
7246 &lod_comp->llc_extent, write);
7248 repeated = !!(sd->sd_repeat);
7252 /* Extend previous component */
7254 new_end = lod_extension_new_end(extension_size, extent->e_end,
7255 prev->llc_stripe_size,
7256 prev->llc_extent.e_end,
7257 lod_comp->llc_extent.e_end);
7259 CDEBUG(D_LAYOUT, "new end %llu\n", new_end);
7260 lod_comp->llc_extent.e_start = new_end;
7261 prev->llc_extent.e_end = new_end;
7263 if (prev->llc_extent.e_end == lod_comp->llc_extent.e_end) {
7264 CDEBUG(D_LAYOUT, "Extension component exhausted\n");
7265 lod_comp->llc_id = LCME_ID_INVAL;
7269 /* rc == 1, failed to extend current component */
7272 /* Normal 'spillover' case - Remove the extension
7273 * space component & bring down the start of the next
7275 lod_comp->llc_id = LCME_ID_INVAL;
7277 if (!(prev->llc_flags & LCME_FL_INIT)) {
7278 prev->llc_id = LCME_ID_INVAL;
7281 lod_sel_adjust_extents(env, lo, *max_comp, index);
7282 } else if (lod_comp_inited(prev)) {
7283 /* If there is no next, and the previous component is
7284 * INIT'ed, try repeating the previous component. */
7285 LASSERT(repeated == 0);
7286 rc = lod_layout_repeat_comp(env, lo, index - 1);
7290 /* The previous component is a repeated component.
7291 * Record this so we don't keep trying to repeat it. */
7294 /* If the previous component is not INIT'ed, this may
7295 * be a component we have just instantiated but failed
7296 * to extend. Or even a repeated component we failed
7297 * to prepare a striping for. Do not repeat but instead
7298 * remove the repeated component & force the extention
7299 * of the original one */
7302 prev->llc_id = LCME_ID_INVAL;
7309 rc = lod_layout_del_prep_layout(env, lo, NULL);
7312 LASSERTF(-rc == change,
7313 "number deleted %d != requested %d\n", -rc,
7316 *max_comp = *max_comp + change;
7318 /* lod_del_prep_layout reallocates ldo_comp_entries, so we must
7319 * refresh these pointers before using them */
7320 lod_comp = &lo->ldo_comp_entries[index];
7321 prev = &lo->ldo_comp_entries[index - 1];
7322 CDEBUG(D_LAYOUT, "After extent updates: prev start %llu, current start "
7323 "%llu, current end %llu max_comp %d ldo_comp_cnt %d\n",
7324 prev->llc_extent.e_start, lod_comp->llc_extent.e_start,
7325 lod_comp->llc_extent.e_end, *max_comp, lo->ldo_comp_cnt);
7327 /* Layout changed successfully */
7332 * Declare layout extent updates
7334 * Handles extensions. Identifies extension components touched by current
7335 * operation and passes them to processing function.
7337 * Restarts with updated layouts from the processing function until the current
7338 * operation no longer touches an extension space component.
7340 * \param[in] env execution environment for this thread
7341 * \param[in,out] lo object to update the layout of
7342 * \param[in] extent layout extent for requested operation, update layout to
7343 * fit this operation
7344 * \param[in] th transaction handle for this operation
7345 * \param[in] pick identifies chosen mirror for FLR layouts
7346 * \param[in] write if this is write op
7348 * \retval 1 on layout changed, 0 on no change
7349 * \retval negative errno on error
7351 static int lod_declare_update_extents(const struct lu_env *env,
7352 struct lod_object *lo, struct lu_extent *extent,
7353 struct thandle *th, int pick, int write)
7355 struct lod_thread_info *info = lod_env_info(env);
7356 struct lod_layout_component *lod_comp;
7357 bool layout_changed = false;
7358 struct sel_data sd = { 0 };
7366 /* This makes us work on the components of the chosen mirror */
7367 start_index = lo->ldo_mirrors[pick].lme_start;
7368 max_comp = lo->ldo_mirrors[pick].lme_end + 1;
7369 if (lo->ldo_flr_state == LCM_FL_NONE)
7370 LASSERT(start_index == 0 && max_comp == lo->ldo_comp_cnt);
7372 CDEBUG(D_LAYOUT, "extent->e_start %llu, extent->e_end %llu\n",
7373 extent->e_start, extent->e_end);
7374 for (i = start_index; i < max_comp; i++) {
7375 lod_comp = &lo->ldo_comp_entries[i];
7377 /* We've passed all components of interest */
7378 if (lod_comp->llc_extent.e_start >= extent->e_end)
7381 if (lod_comp->llc_flags & LCME_FL_EXTENSION) {
7382 layout_changed = true;
7383 rc = lod_sel_handler(env, lo, extent, th, &max_comp,
7388 /* Nothing has changed behind the prev one */
7394 /* We may have added or removed components. If so, we must update the
7395 * start & ends of all the mirrors after the current one, and the end
7396 * of the current mirror. */
7397 change = max_comp - 1 - lo->ldo_mirrors[pick].lme_end;
7399 lo->ldo_mirrors[pick].lme_end += change;
7400 for (i = pick + 1; i < lo->ldo_mirror_count; i++) {
7401 lo->ldo_mirrors[i].lme_start += change;
7402 lo->ldo_mirrors[i].lme_end += change;
7408 /* The amount of components has changed, adjust the lti_comp_idx */
7409 rc2 = lod_layout_data_init(info, lo->ldo_comp_cnt);
7411 return rc < 0 ? rc : rc2 < 0 ? rc2 : layout_changed;
7414 /* If striping is already instantiated or INIT'ed DOM? */
7415 static bool lod_is_instantiation_needed(struct lod_layout_component *comp)
7417 return !(((lov_pattern(comp->llc_pattern) == LOV_PATTERN_MDT) &&
7418 lod_comp_inited(comp)) || comp->llc_stripe);
7422 * Declare layout update for a non-FLR layout.
7424 * \param[in] env execution environment for this thread
7425 * \param[in,out] lo object to update the layout of
7426 * \param[in] layout layout intent for requested operation, "update" is
7427 * a process of reacting to this
7428 * \param[in] buf buffer containing lov ea (see comment on usage inline)
7429 * \param[in] th transaction handle for this operation
7431 * \retval 0 on success
7432 * \retval negative errno on error
7434 static int lod_declare_update_plain(const struct lu_env *env,
7435 struct lod_object *lo, struct layout_intent *layout,
7436 const struct lu_buf *buf, struct thandle *th)
7438 struct lod_thread_info *info = lod_env_info(env);
7439 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7440 struct lod_layout_component *lod_comp;
7441 struct lov_comp_md_v1 *comp_v1 = NULL;
7442 bool layout_changed = false;
7443 bool replay = false;
7447 LASSERT(lo->ldo_flr_state == LCM_FL_NONE);
7450 * In case the client is passing lovea, which only happens during
7451 * the replay of layout intent write RPC for now, we may need to
7452 * parse the lovea and apply new layout configuration.
7454 if (buf && buf->lb_len) {
7455 struct lov_user_md_v1 *v1 = buf->lb_buf;
7457 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
7458 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
7459 LOV_MAGIC_COMP_V1)) {
7460 CERROR("%s: the replay buffer of layout extend "
7461 "(magic %#x) does not contain expected "
7462 "composite layout.\n",
7463 lod2obd(d)->obd_name, v1->lmm_magic);
7464 GOTO(out, rc = -EINVAL);
7467 rc = lod_use_defined_striping(env, lo, buf);
7470 lo->ldo_comp_cached = 1;
7472 rc = lod_get_lov_ea(env, lo);
7475 /* old on-disk EA is stored in info->lti_buf */
7476 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
7478 layout_changed = true;
7480 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
7484 /* non replay path */
7485 rc = lod_striping_load(env, lo);
7490 /* Make sure defined layout covers the requested write range. */
7491 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
7492 if (lo->ldo_comp_cnt > 1 &&
7493 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
7494 lod_comp->llc_extent.e_end < layout->li_extent.e_end) {
7495 CDEBUG_LIMIT(replay ? D_ERROR : D_LAYOUT,
7496 "%s: the defined layout [0, %#llx) does not "
7497 "covers the write range "DEXT"\n",
7498 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
7499 PEXT(&layout->li_extent));
7500 GOTO(out, rc = -EINVAL);
7503 CDEBUG(D_LAYOUT, "%s: "DFID": update components "DEXT"\n",
7504 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
7505 PEXT(&layout->li_extent));
7508 rc = lod_declare_update_extents(env, lo, &layout->li_extent,
7509 th, 0, layout->li_opc == LAYOUT_INTENT_WRITE);
7513 layout_changed = true;
7517 * Iterate ld->ldo_comp_entries, find the component whose extent under
7518 * the write range and not instantianted.
7520 for (i = 0; i < lo->ldo_comp_cnt; i++) {
7521 lod_comp = &lo->ldo_comp_entries[i];
7523 if (lod_comp->llc_extent.e_start >= layout->li_extent.e_end)
7527 /* If striping is instantiated or INIT'ed DOM skip */
7528 if (!lod_is_instantiation_needed(lod_comp))
7532 * In replay path, lod_comp is the EA passed by
7533 * client replay buffer, comp_v1 is the pre-recovery
7534 * on-disk EA, we'd sift out those components which
7535 * were init-ed in the on-disk EA.
7537 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
7542 * this component hasn't instantiated in normal path, or during
7543 * replay it needs replay the instantiation.
7546 /* A released component is being extended */
7547 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
7548 GOTO(out, rc = -EINVAL);
7550 LASSERT(info->lti_comp_idx != NULL);
7551 info->lti_comp_idx[info->lti_count++] = i;
7552 layout_changed = true;
7555 if (!layout_changed)
7558 lod_obj_inc_layout_gen(lo);
7559 rc = lod_declare_instantiate_components(env, lo, th, 0);
7563 lod_striping_free(env, lo);
7567 static inline int lod_comp_index(struct lod_object *lo,
7568 struct lod_layout_component *lod_comp)
7570 LASSERT(lod_comp >= lo->ldo_comp_entries &&
7571 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
7573 return lod_comp - lo->ldo_comp_entries;
7577 * Stale other mirrors by writing extent.
7579 static int lod_stale_components(const struct lu_env *env, struct lod_object *lo,
7580 int primary, struct lu_extent *extent,
7583 struct lod_layout_component *pri_comp, *lod_comp;
7584 struct lu_extent pri_extent;
7589 /* The writing extent decides which components in the primary
7590 * are affected... */
7591 CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
7594 lod_foreach_mirror_comp(pri_comp, lo, primary) {
7595 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
7598 CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
7599 lod_comp_index(lo, pri_comp),
7600 PEXT(&pri_comp->llc_extent));
7602 pri_extent.e_start = pri_comp->llc_extent.e_start;
7603 pri_extent.e_end = pri_comp->llc_extent.e_end;
7605 for (i = 0; i < lo->ldo_mirror_count; i++) {
7608 rc = lod_declare_update_extents(env, lo, &pri_extent,
7610 /* if update_extents changed the layout, it may have
7611 * reallocated the component array, so start over to
7612 * avoid using stale pointers */
7618 /* ... and then stale other components that are
7619 * overlapping with primary components */
7620 lod_foreach_mirror_comp(lod_comp, lo, i) {
7621 if (!lu_extent_is_overlapped(
7623 &lod_comp->llc_extent))
7626 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
7627 i, lod_comp_index(lo, lod_comp));
7629 lod_comp->llc_flags |= LCME_FL_STALE;
7630 lo->ldo_mirrors[i].lme_stale = 1;
7639 * check an OST's availability
7640 * \param[in] env execution environment
7641 * \param[in] lo lod object
7642 * \param[in] dt dt object
7643 * \param[in] index mirror index
7645 * \retval negative if failed
7646 * \retval 1 if \a dt is available
7647 * \retval 0 if \a dt is not available
7649 static inline int lod_check_ost_avail(const struct lu_env *env,
7650 struct lod_object *lo,
7651 struct dt_object *dt, int index)
7653 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7654 struct lod_tgt_desc *ost;
7656 int type = LU_SEQ_RANGE_OST;
7659 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &idx, &type);
7661 CERROR("%s: can't locate "DFID":rc = %d\n",
7662 lod2obd(lod)->obd_name, PFID(lu_object_fid(&dt->do_lu)),
7667 ost = OST_TGT(lod, idx);
7668 if (ost->ltd_statfs.os_state &
7669 (OS_STATFS_READONLY | OS_STATFS_ENOSPC | OS_STATFS_ENOINO |
7670 OS_STATFS_NOPRECREATE) ||
7671 ost->ltd_active == 0) {
7672 CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail, rc = %d\n",
7673 PFID(lod_object_fid(lo)), index, idx, rc);
7681 * Pick primary mirror for write
7682 * \param[in] env execution environment
7683 * \param[in] lo object
7684 * \param[in] extent write range
7686 static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo,
7687 struct lu_extent *extent)
7689 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7690 unsigned int seq = 0;
7691 struct lod_layout_component *lod_comp;
7693 int picked = -1, second_pick = -1, third_pick = -1;
7696 if (OBD_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
7697 get_random_bytes(&seq, sizeof(seq));
7698 seq %= lo->ldo_mirror_count;
7702 * Pick a mirror as the primary, and check the availability of OSTs.
7704 * This algo can be revised later after knowing the topology of
7707 lod_qos_statfs_update(env, lod, &lod->lod_ost_descs);
7709 rc = lod_fill_mirrors(lo);
7713 for (i = 0; i < lo->ldo_mirror_count; i++) {
7714 bool ost_avail = true;
7715 int index = (i + seq) % lo->ldo_mirror_count;
7717 if (lo->ldo_mirrors[index].lme_stale) {
7718 CDEBUG(D_LAYOUT, DFID": mirror %d stale\n",
7719 PFID(lod_object_fid(lo)), index);
7723 /* 2nd pick is for the primary mirror containing unavail OST */
7724 if (lo->ldo_mirrors[index].lme_prefer && second_pick < 0)
7725 second_pick = index;
7727 /* 3rd pick is for non-primary mirror containing unavail OST */
7728 if (second_pick < 0 && third_pick < 0)
7732 * we found a non-primary 1st pick, we'd like to find a
7733 * potential pirmary mirror.
7735 if (picked >= 0 && !lo->ldo_mirrors[index].lme_prefer)
7738 /* check the availability of OSTs */
7739 lod_foreach_mirror_comp(lod_comp, lo, index) {
7740 if (!lod_comp_inited(lod_comp) || !lod_comp->llc_stripe)
7743 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
7744 struct dt_object *dt = lod_comp->llc_stripe[j];
7746 rc = lod_check_ost_avail(env, lo, dt, index);
7753 } /* for all dt object in one component */
7756 } /* for all components in a mirror */
7759 * the OSTs where allocated objects locates in the components
7760 * of the mirror are available.
7765 /* this mirror has all OSTs available */
7769 * primary with all OSTs are available, this is the perfect
7772 if (lo->ldo_mirrors[index].lme_prefer)
7774 } /* for all mirrors */
7776 /* failed to pick a sound mirror, lower our expectation */
7778 picked = second_pick;
7780 picked = third_pick;
7787 static int lod_prepare_resync_mirror(const struct lu_env *env,
7788 struct lod_object *lo,
7791 struct lod_thread_info *info = lod_env_info(env);
7792 struct lod_layout_component *lod_comp;
7793 bool neg = !!(MIRROR_ID_NEG & mirror_id);
7796 mirror_id &= ~MIRROR_ID_NEG;
7798 for (i = 0; i < lo->ldo_mirror_count; i++) {
7799 if ((!neg && lo->ldo_mirrors[i].lme_id != mirror_id) ||
7800 (neg && lo->ldo_mirrors[i].lme_id == mirror_id))
7803 lod_foreach_mirror_comp(lod_comp, lo, i) {
7804 if (lod_comp_inited(lod_comp))
7807 info->lti_comp_idx[info->lti_count++] =
7808 lod_comp_index(lo, lod_comp);
7816 * figure out the components should be instantiated for resync.
7818 static int lod_prepare_resync(const struct lu_env *env, struct lod_object *lo,
7819 struct lu_extent *extent)
7821 struct lod_thread_info *info = lod_env_info(env);
7822 struct lod_layout_component *lod_comp;
7823 unsigned int need_sync = 0;
7827 DFID": instantiate all stale components in "DEXT"\n",
7828 PFID(lod_object_fid(lo)), PEXT(extent));
7831 * instantiate all components within this extent, even non-stale
7834 for (i = 0; i < lo->ldo_mirror_count; i++) {
7835 if (!lo->ldo_mirrors[i].lme_stale)
7838 lod_foreach_mirror_comp(lod_comp, lo, i) {
7839 if (!lu_extent_is_overlapped(extent,
7840 &lod_comp->llc_extent))
7845 if (lod_comp_inited(lod_comp))
7848 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
7849 i, lod_comp_index(lo, lod_comp));
7850 info->lti_comp_idx[info->lti_count++] =
7851 lod_comp_index(lo, lod_comp);
7855 return need_sync ? 0 : -EALREADY;
7858 static int lod_declare_update_rdonly(const struct lu_env *env,
7859 struct lod_object *lo, struct md_layout_change *mlc,
7862 struct lod_thread_info *info = lod_env_info(env);
7863 struct lu_attr *layout_attr = &info->lti_layout_attr;
7864 struct lod_layout_component *lod_comp;
7865 struct lu_extent extent = { 0 };
7869 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
7870 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
7871 mlc->mlc_opc == MD_LAYOUT_RESYNC);
7872 LASSERT(lo->ldo_mirror_count > 0);
7874 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7875 struct layout_intent *layout = mlc->mlc_intent;
7876 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
7879 extent = layout->li_extent;
7880 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
7881 PFID(lod_object_fid(lo)), PEXT(&extent));
7883 picked = lod_primary_pick(env, lo, &extent);
7887 CDEBUG(D_LAYOUT, DFID": picked mirror id %u as primary\n",
7888 PFID(lod_object_fid(lo)),
7889 lo->ldo_mirrors[picked].lme_id);
7891 /* Update extents of primary before staling */
7892 rc = lod_declare_update_extents(env, lo, &extent, th, picked,
7897 if (layout->li_opc == LAYOUT_INTENT_TRUNC) {
7899 * trunc transfers [0, size) in the intent extent, we'd
7900 * stale components overlapping [size, eof).
7902 extent.e_start = extent.e_end;
7903 extent.e_end = OBD_OBJECT_EOF;
7906 /* stale overlapping components from other mirrors */
7907 rc = lod_stale_components(env, lo, picked, &extent, th);
7911 /* restore truncate intent extent */
7912 if (layout->li_opc == LAYOUT_INTENT_TRUNC)
7913 extent.e_end = extent.e_start;
7915 /* instantiate components for the picked mirror, start from 0 */
7918 lod_foreach_mirror_comp(lod_comp, lo, picked) {
7919 if (!lu_extent_is_overlapped(&extent,
7920 &lod_comp->llc_extent))
7923 if (!lod_is_instantiation_needed(lod_comp))
7926 info->lti_comp_idx[info->lti_count++] =
7927 lod_comp_index(lo, lod_comp);
7930 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
7931 } else { /* MD_LAYOUT_RESYNC */
7935 * could contain multiple non-stale mirrors, so we need to
7936 * prep uninited all components assuming any non-stale mirror
7937 * could be picked as the primary mirror.
7939 if (mlc->mlc_mirror_id == 0) {
7941 for (i = 0; i < lo->ldo_mirror_count; i++) {
7942 if (lo->ldo_mirrors[i].lme_stale)
7945 lod_foreach_mirror_comp(lod_comp, lo, i) {
7946 if (!lod_comp_inited(lod_comp))
7950 lod_comp->llc_extent.e_end)
7952 lod_comp->llc_extent.e_end;
7955 rc = lod_prepare_resync(env, lo, &extent);
7959 /* mirror write, try to init its all components */
7960 rc = lod_prepare_resync_mirror(env, lo,
7961 mlc->mlc_mirror_id);
7966 /* change the file state to SYNC_PENDING */
7967 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
7970 /* Reset the layout version once it's becoming too large.
7971 * This way it can make sure that the layout version is
7972 * monotonously increased in this writing era. */
7973 lod_obj_inc_layout_gen(lo);
7974 if (lo->ldo_layout_gen > (LCME_ID_MAX >> 1)) {
7975 __u32 layout_version;
7977 get_random_bytes(&layout_version, sizeof(layout_version));
7978 lo->ldo_layout_gen = layout_version & 0xffff;
7981 rc = lod_declare_instantiate_components(env, lo, th, 0);
7985 layout_attr->la_valid = LA_LAYOUT_VERSION;
7986 layout_attr->la_layout_version = 0; /* set current version */
7987 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
7988 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
7989 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
7995 lod_striping_free(env, lo);
7999 static int lod_declare_update_write_pending(const struct lu_env *env,
8000 struct lod_object *lo, struct md_layout_change *mlc,
8003 struct lod_thread_info *info = lod_env_info(env);
8004 struct lu_attr *layout_attr = &info->lti_layout_attr;
8005 struct lod_layout_component *lod_comp;
8006 struct lu_extent extent = { 0 };
8012 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
8013 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
8014 mlc->mlc_opc == MD_LAYOUT_RESYNC);
8016 /* look for the first preferred mirror */
8017 for (i = 0; i < lo->ldo_mirror_count; i++) {
8018 if (lo->ldo_mirrors[i].lme_stale)
8020 if (lo->ldo_mirrors[i].lme_prefer == 0)
8027 /* no primary, use any in-sync */
8028 for (i = 0; i < lo->ldo_mirror_count; i++) {
8029 if (lo->ldo_mirrors[i].lme_stale)
8035 CERROR(DFID ": doesn't have a primary mirror\n",
8036 PFID(lod_object_fid(lo)));
8037 GOTO(out, rc = -ENODATA);
8041 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
8042 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
8044 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
8046 /* for LAYOUT_WRITE opc, it has to do the following operations:
8047 * 1. stale overlapping componets from stale mirrors;
8048 * 2. instantiate components of the primary mirror;
8049 * 3. transfter layout version to all objects of the primary;
8051 * for LAYOUT_RESYNC opc, it will do:
8052 * 1. instantiate components of all stale mirrors;
8053 * 2. transfer layout version to all objects to close write era. */
8055 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
8056 struct layout_intent *layout = mlc->mlc_intent;
8057 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
8059 LASSERT(mlc->mlc_intent != NULL);
8061 extent = mlc->mlc_intent->li_extent;
8063 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
8064 PFID(lod_object_fid(lo)), PEXT(&extent));
8066 /* 1. Update extents of primary before staling */
8067 rc = lod_declare_update_extents(env, lo, &extent, th, primary,
8072 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC) {
8074 * trunc transfers [0, size) in the intent extent, we'd
8075 * stale components overlapping [size, eof).
8077 extent.e_start = extent.e_end;
8078 extent.e_end = OBD_OBJECT_EOF;
8081 /* 2. stale overlapping components */
8082 rc = lod_stale_components(env, lo, primary, &extent, th);
8086 /* 3. find the components which need instantiating.
8087 * instantiate [0, mlc->mlc_intent->e_end) */
8089 /* restore truncate intent extent */
8090 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC)
8091 extent.e_end = extent.e_start;
8094 lod_foreach_mirror_comp(lod_comp, lo, primary) {
8095 if (!lu_extent_is_overlapped(&extent,
8096 &lod_comp->llc_extent))
8099 if (!lod_is_instantiation_needed(lod_comp))
8102 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
8103 primary, lod_comp_index(lo, lod_comp));
8104 info->lti_comp_idx[info->lti_count++] =
8105 lod_comp_index(lo, lod_comp);
8107 } else { /* MD_LAYOUT_RESYNC */
8108 if (mlc->mlc_mirror_id == 0) {
8110 lod_foreach_mirror_comp(lod_comp, lo, primary) {
8111 if (!lod_comp_inited(lod_comp))
8114 extent.e_end = lod_comp->llc_extent.e_end;
8117 rc = lod_prepare_resync(env, lo, &extent);
8121 /* mirror write, try to init its all components */
8122 rc = lod_prepare_resync_mirror(env, lo,
8123 mlc->mlc_mirror_id);
8128 /* change the file state to SYNC_PENDING */
8129 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
8132 rc = lod_declare_instantiate_components(env, lo, th, 0);
8136 /* 3. transfer layout version to OST objects.
8137 * transfer new layout version to OST objects so that stale writes
8138 * can be denied. It also ends an era of writing by setting
8139 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
8140 * send write RPC; only resync RPCs could do it. */
8141 layout_attr->la_valid = LA_LAYOUT_VERSION;
8142 layout_attr->la_layout_version = 0; /* set current version */
8143 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
8144 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
8145 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
8149 lod_obj_inc_layout_gen(lo);
8152 lod_striping_free(env, lo);
8156 static int lod_declare_update_sync_pending(const struct lu_env *env,
8157 struct lod_object *lo, struct md_layout_change *mlc,
8160 struct lod_thread_info *info = lod_env_info(env);
8161 struct lu_attr *layout_attr = &info->lti_layout_attr;
8162 unsigned sync_components = 0;
8163 unsigned resync_components = 0;
8168 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
8169 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
8170 mlc->mlc_opc == MD_LAYOUT_WRITE);
8172 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
8173 PFID(lod_object_fid(lo)), mlc->mlc_opc);
8175 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
8176 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
8177 PFID(lod_object_fid(lo)));
8179 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
8180 return lod_declare_update_write_pending(env, lo, mlc, th);
8183 /* MD_LAYOUT_RESYNC_DONE */
8185 for (i = 0; i < lo->ldo_comp_cnt; i++) {
8186 struct lod_layout_component *lod_comp;
8189 lod_comp = &lo->ldo_comp_entries[i];
8191 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
8196 for (j = 0; j < mlc->mlc_resync_count; j++) {
8197 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
8200 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
8201 lod_comp->llc_flags &= ~LCME_FL_STALE;
8202 resync_components++;
8208 for (i = 0; i < mlc->mlc_resync_count; i++) {
8209 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
8212 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
8213 "or already synced\n", PFID(lod_object_fid(lo)),
8214 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
8215 GOTO(out, rc = -EINVAL);
8218 if (!sync_components || (mlc->mlc_resync_count && !resync_components)) {
8219 CDEBUG(D_LAYOUT, DFID": no mirror in sync\n",
8220 PFID(lod_object_fid(lo)));
8222 /* tend to return an error code here to prevent
8223 * the MDT from setting SoM attribute */
8224 GOTO(out, rc = -EINVAL);
8227 CDEBUG(D_LAYOUT, DFID": synced %u resynced %u/%zu components\n",
8228 PFID(lod_object_fid(lo)),
8229 sync_components, resync_components, mlc->mlc_resync_count);
8231 lo->ldo_flr_state = LCM_FL_RDONLY;
8232 lod_obj_inc_layout_gen(lo);
8234 layout_attr->la_valid = LA_LAYOUT_VERSION;
8235 layout_attr->la_layout_version = 0; /* set current version */
8236 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
8240 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
8241 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
8242 &info->lti_buf, XATTR_NAME_LOV, 0, th);
8247 lod_striping_free(env, lo);
8251 typedef int (*mlc_handler)(const struct lu_env *env, struct dt_object *dt,
8252 const struct md_layout_change *mlc,
8253 struct thandle *th);
8256 * Attach stripes after target's for migrating directory. NB, we
8257 * only need to declare this, the actual work is done inside
8258 * lod_xattr_set_lmv().
8260 * \param[in] env execution environment
8261 * \param[in] dt target object
8262 * \param[in] mlc layout change data
8263 * \param[in] th transaction handle
8265 * \retval 0 on success
8266 * \retval negative if failed
8268 static int lod_dir_declare_layout_attach(const struct lu_env *env,
8269 struct dt_object *dt,
8270 const struct md_layout_change *mlc,
8273 struct lod_thread_info *info = lod_env_info(env);
8274 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
8275 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
8276 struct lod_object *lo = lod_dt_obj(dt);
8277 struct dt_object *next = dt_object_child(dt);
8278 struct dt_object_format *dof = &info->lti_format;
8279 struct lmv_mds_md_v1 *lmv = mlc->mlc_buf.lb_buf;
8280 struct dt_object **stripes;
8281 __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
8282 struct lu_fid *fid = &info->lti_fid;
8283 struct lod_tgt_desc *tgt;
8284 struct dt_object *dto;
8285 struct dt_device *tgt_dt;
8286 int type = LU_SEQ_RANGE_ANY;
8287 struct dt_insert_rec *rec = &info->lti_dt_rec;
8288 char *stripe_name = info->lti_key;
8289 struct lu_name *sname;
8290 struct linkea_data ldata = { NULL };
8291 struct lu_buf linkea_buf;
8298 if (!lmv_is_sane(lmv))
8301 if (!dt_try_as_dir(env, dt))
8304 dof->dof_type = DFT_DIR;
8306 OBD_ALLOC_PTR_ARRAY(stripes, (lo->ldo_dir_stripe_count + stripe_count));
8310 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
8311 stripes[i] = lo->ldo_stripe[i];
8313 rec->rec_type = S_IFDIR;
8315 for (i = 0; i < stripe_count; i++) {
8317 &lmv->lmv_stripe_fids[i]);
8318 if (!fid_is_sane(fid))
8321 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
8325 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
8326 tgt_dt = lod->lod_child;
8328 tgt = LTD_TGT(ltd, idx);
8330 GOTO(out, rc = -ESTALE);
8331 tgt_dt = tgt->ltd_tgt;
8334 dto = dt_locate_at(env, tgt_dt, fid,
8335 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
8338 GOTO(out, rc = PTR_ERR(dto));
8340 stripes[i + lo->ldo_dir_stripe_count] = dto;
8342 if (!dt_try_as_dir(env, dto))
8343 GOTO(out, rc = -ENOTDIR);
8345 rc = lod_sub_declare_ref_add(env, dto, th);
8349 rec->rec_fid = lu_object_fid(&dto->do_lu);
8350 rc = lod_sub_declare_insert(env, dto,
8351 (const struct dt_rec *)rec,
8352 (const struct dt_key *)dot, th);
8356 rc = lod_sub_declare_insert(env, dto,
8357 (const struct dt_rec *)rec,
8358 (const struct dt_key *)dotdot, th);
8362 rc = lod_sub_declare_xattr_set(env, dto, &mlc->mlc_buf,
8363 XATTR_NAME_LMV, 0, th);
8367 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
8368 PFID(lu_object_fid(&dto->do_lu)),
8369 i + lo->ldo_dir_stripe_count);
8371 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
8372 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
8373 sname, lu_object_fid(&dt->do_lu));
8377 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
8378 linkea_buf.lb_len = ldata.ld_leh->leh_len;
8379 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
8380 XATTR_NAME_LINK, 0, th);
8384 rc = lod_sub_declare_insert(env, next,
8385 (const struct dt_rec *)rec,
8386 (const struct dt_key *)stripe_name,
8391 rc = lod_sub_declare_ref_add(env, next, th);
8397 OBD_FREE_PTR_ARRAY(lo->ldo_stripe,
8398 lo->ldo_dir_stripes_allocated);
8399 lo->ldo_stripe = stripes;
8400 lo->ldo_is_foreign = 0;
8401 lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count;
8402 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type);
8403 lo->ldo_dir_stripe_count += stripe_count;
8404 lo->ldo_dir_stripes_allocated += stripe_count;
8406 /* plain directory split creates target as a plain directory, while
8407 * after source attached as the first stripe, it becomes a striped
8408 * directory, set correct do_index_ops, otherwise it can't be unlinked.
8410 dt->do_index_ops = &lod_striped_index_ops;
8414 i = lo->ldo_dir_stripe_count;
8415 while (i < lo->ldo_dir_stripe_count + stripe_count && stripes[i])
8416 dt_object_put(env, stripes[i++]);
8418 OBD_FREE_PTR_ARRAY(stripes, stripe_count + lo->ldo_dir_stripe_count);
8422 static int lod_dir_declare_layout_detach(const struct lu_env *env,
8423 struct dt_object *dt,
8424 const struct md_layout_change *unused,
8427 struct lod_thread_info *info = lod_env_info(env);
8428 struct lod_object *lo = lod_dt_obj(dt);
8429 struct dt_object *next = dt_object_child(dt);
8430 char *stripe_name = info->lti_key;
8431 struct dt_object *dto;
8435 if (!dt_try_as_dir(env, dt))
8438 if (!lo->ldo_dir_stripe_count)
8439 return lod_sub_declare_delete(env, next,
8440 (const struct dt_key *)dotdot, th);
8442 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8443 dto = lo->ldo_stripe[i];
8447 if (!dt_try_as_dir(env, dto))
8450 rc = lod_sub_declare_delete(env, dto,
8451 (const struct dt_key *)dotdot, th);
8455 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8456 PFID(lu_object_fid(&dto->do_lu)), i);
8458 rc = lod_sub_declare_delete(env, next,
8459 (const struct dt_key *)stripe_name, th);
8463 rc = lod_sub_declare_ref_del(env, next, th);
8471 static int dt_dir_is_empty(const struct lu_env *env,
8472 struct dt_object *obj)
8475 const struct dt_it_ops *iops;
8480 if (!dt_try_as_dir(env, obj))
8483 iops = &obj->do_index_ops->dio_it;
8484 it = iops->init(env, obj, LUDA_64BITHASH);
8486 RETURN(PTR_ERR(it));
8488 rc = iops->get(env, it, (const struct dt_key *)"");
8492 for (rc = 0, i = 0; rc == 0 && i < 3; ++i)
8493 rc = iops->next(env, it);
8499 /* Huh? Index contains no zero key? */
8504 iops->fini(env, it);
8509 static int lod_dir_declare_layout_shrink(const struct lu_env *env,
8510 struct dt_object *dt,
8511 const struct md_layout_change *mlc,
8514 struct lod_thread_info *info = lod_env_info(env);
8515 struct lod_object *lo = lod_dt_obj(dt);
8516 struct dt_object *next = dt_object_child(dt);
8517 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
8518 char *stripe_name = info->lti_key;
8519 struct lu_buf *lmv_buf = &info->lti_buf;
8520 __u32 final_stripe_count;
8521 struct dt_object *dto;
8527 if (!dt_try_as_dir(env, dt))
8530 /* shouldn't be called on plain directory */
8531 LASSERT(lo->ldo_dir_stripe_count);
8533 lmv_buf->lb_buf = &info->lti_lmv.lmv_md_v1;
8534 lmv_buf->lb_len = sizeof(info->lti_lmv.lmv_md_v1);
8536 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
8537 LASSERT(final_stripe_count &&
8538 final_stripe_count < lo->ldo_dir_stripe_count);
8540 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8541 dto = lo->ldo_stripe[i];
8545 if (i < final_stripe_count) {
8546 rc = lod_sub_declare_xattr_set(env, dto, lmv_buf,
8548 LU_XATTR_REPLACE, th);
8555 rc = dt_dir_is_empty(env, dto);
8559 rc = lod_sub_declare_ref_del(env, dto, th);
8563 rc = lod_sub_declare_destroy(env, dto, th);
8567 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8568 PFID(lu_object_fid(&dto->do_lu)), i);
8570 rc = lod_sub_declare_delete(env, next,
8571 (const struct dt_key *)stripe_name, th);
8575 rc = lod_sub_declare_ref_del(env, next, th);
8580 rc = lod_sub_declare_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
8581 LU_XATTR_REPLACE, th);
8586 * Allocate stripes for split directory.
8588 * \param[in] env execution environment
8589 * \param[in] dt target object
8590 * \param[in] mlc layout change data
8591 * \param[in] th transaction handle
8593 * \retval 0 on success
8594 * \retval negative if failed
8596 static int lod_dir_declare_layout_split(const struct lu_env *env,
8597 struct dt_object *dt,
8598 const struct md_layout_change *mlc,
8601 struct lod_thread_info *info = lod_env_info(env);
8602 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
8603 struct lod_object *lo = lod_dt_obj(dt);
8604 struct dt_object_format *dof = &info->lti_format;
8605 struct lmv_user_md_v1 *lum = mlc->mlc_spec->u.sp_ea.eadata;
8606 struct dt_object **stripes;
8614 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
8615 LASSERT(le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT);
8617 saved_count = lo->ldo_dir_stripes_allocated;
8618 stripe_count = le32_to_cpu(lum->lum_stripe_count);
8619 if (stripe_count <= saved_count)
8622 dof->dof_type = DFT_DIR;
8624 OBD_ALLOC(stripes, sizeof(*stripes) * stripe_count);
8628 for (i = 0; i < lo->ldo_dir_stripes_allocated; i++)
8629 stripes[i] = lo->ldo_stripe[i];
8631 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
8632 rc = lod_mdt_alloc_qos(env, lo, stripes, saved_count, stripe_count);
8634 rc = lod_mdt_alloc_rr(env, lo, stripes, saved_count,
8637 OBD_FREE(stripes, sizeof(*stripes) * stripe_count);
8641 LASSERT(rc > saved_count);
8642 OBD_FREE(lo->ldo_stripe,
8643 sizeof(*stripes) * lo->ldo_dir_stripes_allocated);
8644 lo->ldo_stripe = stripes;
8645 lo->ldo_is_foreign = 0;
8646 lo->ldo_dir_striped = 1;
8647 lo->ldo_dir_stripe_count = rc;
8648 lo->ldo_dir_stripes_allocated = stripe_count;
8649 lo->ldo_dir_split_hash = lo->ldo_dir_hash_type;
8650 lo->ldo_dir_hash_type = le32_to_cpu(lum->lum_hash_type);
8651 if (!lmv_is_known_hash_type(lo->ldo_dir_hash_type))
8652 lo->ldo_dir_hash_type =
8653 lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
8654 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_SPLIT | LMV_HASH_FLAG_MIGRATION;
8655 lo->ldo_dir_split_offset = saved_count;
8656 lo->ldo_dir_layout_version++;
8657 lo->ldo_dir_stripe_loaded = 1;
8659 rc = lod_dir_declare_create_stripes(env, dt, mlc->mlc_attr, dof, th);
8661 lod_striping_free(env, lo);
8667 * detach all stripes from dir master object, NB, stripes are not destroyed, but
8668 * deleted from it's parent namespace, this function is called in two places:
8669 * 1. mdd_migrate_mdt() detach stripes from source, and attach them to
8671 * 2. mdd_dir_layout_update() detach stripe before turning 1-stripe directory to
8672 * a plain directory.
8674 * \param[in] env execution environment
8675 * \param[in] dt target object
8676 * \param[in] mlc layout change data
8677 * \param[in] th transaction handle
8679 * \retval 0 on success
8680 * \retval negative if failed
8682 static int lod_dir_layout_detach(const struct lu_env *env,
8683 struct dt_object *dt,
8684 const struct md_layout_change *mlc,
8687 struct lod_thread_info *info = lod_env_info(env);
8688 struct lod_object *lo = lod_dt_obj(dt);
8689 struct dt_object *next = dt_object_child(dt);
8690 char *stripe_name = info->lti_key;
8691 struct dt_object *dto;
8697 if (!lo->ldo_dir_stripe_count) {
8698 /* plain directory delete .. */
8699 rc = lod_sub_delete(env, next,
8700 (const struct dt_key *)dotdot, th);
8704 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8705 dto = lo->ldo_stripe[i];
8709 rc = lod_sub_delete(env, dto,
8710 (const struct dt_key *)dotdot, th);
8714 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8715 PFID(lu_object_fid(&dto->do_lu)), i);
8717 rc = lod_sub_delete(env, next,
8718 (const struct dt_key *)stripe_name, th);
8722 rc = lod_sub_ref_del(env, next, th);
8727 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8728 dto = lo->ldo_stripe[i];
8730 dt_object_put(env, dto);
8732 OBD_FREE_PTR_ARRAY(lo->ldo_stripe, lo->ldo_dir_stripes_allocated);
8733 lo->ldo_stripe = NULL;
8734 lo->ldo_dir_stripes_allocated = 0;
8735 lo->ldo_dir_stripe_count = 0;
8736 dt->do_index_ops = &lod_index_ops;
8741 static int lod_dir_layout_shrink(const struct lu_env *env,
8742 struct dt_object *dt,
8743 const struct md_layout_change *mlc,
8746 struct lod_thread_info *info = lod_env_info(env);
8747 struct lod_object *lo = lod_dt_obj(dt);
8748 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
8749 struct dt_object *next = dt_object_child(dt);
8750 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
8751 __u32 final_stripe_count;
8752 char *stripe_name = info->lti_key;
8753 struct dt_object *dto;
8754 struct lu_buf *lmv_buf = &info->lti_buf;
8755 struct lmv_mds_md_v1 *lmv = &info->lti_lmv.lmv_md_v1;
8757 int type = LU_SEQ_RANGE_ANY;
8763 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
8765 lmv_buf->lb_buf = lmv;
8766 lmv_buf->lb_len = sizeof(*lmv);
8767 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
8768 lmv->lmv_stripe_count = cpu_to_le32(final_stripe_count);
8769 lmv->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type) &
8770 cpu_to_le32(LMV_HASH_TYPE_MASK |
8771 LMV_HASH_FLAG_FIXED);
8772 lmv->lmv_layout_version =
8773 cpu_to_le32(lo->ldo_dir_layout_version + 1);
8774 lmv->lmv_migrate_offset = 0;
8775 lmv->lmv_migrate_hash = 0;
8777 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8778 dto = lo->ldo_stripe[i];
8782 if (i < final_stripe_count) {
8783 rc = lod_fld_lookup(env, lod,
8784 lu_object_fid(&dto->do_lu),
8789 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
8790 rc = lod_sub_xattr_set(env, dto, lmv_buf,
8792 LU_XATTR_REPLACE, th);
8799 dt_write_lock(env, dto, DT_TGT_CHILD);
8800 rc = lod_sub_ref_del(env, dto, th);
8801 dt_write_unlock(env, dto);
8805 rc = lod_sub_destroy(env, dto, th);
8809 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8810 PFID(lu_object_fid(&dto->do_lu)), i);
8812 rc = lod_sub_delete(env, next,
8813 (const struct dt_key *)stripe_name, th);
8817 rc = lod_sub_ref_del(env, next, th);
8822 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &mdtidx,
8827 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
8828 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
8829 rc = lod_sub_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
8830 LU_XATTR_REPLACE, th);
8834 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
8835 dto = lo->ldo_stripe[i];
8837 dt_object_put(env, dto);
8839 lo->ldo_dir_stripe_count = final_stripe_count;
8844 static mlc_handler dir_mlc_declare_ops[MD_LAYOUT_MAX] = {
8845 [MD_LAYOUT_ATTACH] = lod_dir_declare_layout_attach,
8846 [MD_LAYOUT_DETACH] = lod_dir_declare_layout_detach,
8847 [MD_LAYOUT_SHRINK] = lod_dir_declare_layout_shrink,
8848 [MD_LAYOUT_SPLIT] = lod_dir_declare_layout_split,
8851 static mlc_handler dir_mlc_ops[MD_LAYOUT_MAX] = {
8852 [MD_LAYOUT_DETACH] = lod_dir_layout_detach,
8853 [MD_LAYOUT_SHRINK] = lod_dir_layout_shrink,
8856 static int lod_declare_layout_change(const struct lu_env *env,
8857 struct dt_object *dt, struct md_layout_change *mlc,
8860 struct lod_thread_info *info = lod_env_info(env);
8861 struct lod_object *lo = lod_dt_obj(dt);
8866 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
8867 LASSERT(dir_mlc_declare_ops[mlc->mlc_opc]);
8868 rc = dir_mlc_declare_ops[mlc->mlc_opc](env, dt, mlc, th);
8872 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
8873 dt_object_remote(dt_object_child(dt)))
8876 rc = lod_striping_load(env, lo);
8880 LASSERT(lo->ldo_comp_cnt > 0);
8882 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
8886 switch (lo->ldo_flr_state) {
8888 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
8892 rc = lod_declare_update_rdonly(env, lo, mlc, th);
8894 case LCM_FL_WRITE_PENDING:
8895 rc = lod_declare_update_write_pending(env, lo, mlc, th);
8897 case LCM_FL_SYNC_PENDING:
8898 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
8909 * Instantiate layout component objects which covers the intent write offset.
8911 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
8912 struct md_layout_change *mlc, struct thandle *th)
8914 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
8915 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
8916 struct lod_object *lo = lod_dt_obj(dt);
8921 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
8922 LASSERT(dir_mlc_ops[mlc->mlc_opc]);
8923 rc = dir_mlc_ops[mlc->mlc_opc](env, dt, mlc, th);
8927 rc = lod_striped_create(env, dt, attr, NULL, th);
8928 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
8929 layout_attr->la_layout_version |= lo->ldo_layout_gen;
8930 rc = lod_attr_set(env, dt, layout_attr, th);
8936 const struct dt_object_operations lod_obj_ops = {
8937 .do_read_lock = lod_read_lock,
8938 .do_write_lock = lod_write_lock,
8939 .do_read_unlock = lod_read_unlock,
8940 .do_write_unlock = lod_write_unlock,
8941 .do_write_locked = lod_write_locked,
8942 .do_attr_get = lod_attr_get,
8943 .do_declare_attr_set = lod_declare_attr_set,
8944 .do_attr_set = lod_attr_set,
8945 .do_xattr_get = lod_xattr_get,
8946 .do_declare_xattr_set = lod_declare_xattr_set,
8947 .do_xattr_set = lod_xattr_set,
8948 .do_declare_xattr_del = lod_declare_xattr_del,
8949 .do_xattr_del = lod_xattr_del,
8950 .do_xattr_list = lod_xattr_list,
8951 .do_ah_init = lod_ah_init,
8952 .do_declare_create = lod_declare_create,
8953 .do_create = lod_create,
8954 .do_declare_destroy = lod_declare_destroy,
8955 .do_destroy = lod_destroy,
8956 .do_index_try = lod_index_try,
8957 .do_declare_ref_add = lod_declare_ref_add,
8958 .do_ref_add = lod_ref_add,
8959 .do_declare_ref_del = lod_declare_ref_del,
8960 .do_ref_del = lod_ref_del,
8961 .do_object_sync = lod_object_sync,
8962 .do_object_lock = lod_object_lock,
8963 .do_object_unlock = lod_object_unlock,
8964 .do_invalidate = lod_invalidate,
8965 .do_declare_layout_change = lod_declare_layout_change,
8966 .do_layout_change = lod_layout_change,
8970 * Implementation of dt_body_operations::dbo_read.
8972 * \see dt_body_operations::dbo_read() in the API description for details.
8974 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
8975 struct lu_buf *buf, loff_t *pos)
8977 struct dt_object *next = dt_object_child(dt);
8979 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
8980 S_ISLNK(dt->do_lu.lo_header->loh_attr));
8981 return next->do_body_ops->dbo_read(env, next, buf, pos);
8985 * Implementation of dt_body_operations::dbo_declare_write.
8987 * \see dt_body_operations::dbo_declare_write() in the API description
8990 static ssize_t lod_declare_write(const struct lu_env *env,
8991 struct dt_object *dt,
8992 const struct lu_buf *buf, loff_t pos,
8995 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
8999 * Implementation of dt_body_operations::dbo_write.
9001 * \see dt_body_operations::dbo_write() in the API description for details.
9003 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
9004 const struct lu_buf *buf, loff_t *pos,
9007 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
9008 S_ISLNK(dt->do_lu.lo_header->loh_attr));
9009 return lod_sub_write(env, dt_object_child(dt), buf, pos, th);
9012 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
9013 __u64 start, __u64 end, struct thandle *th)
9015 if (dt_object_remote(dt))
9018 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
9021 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
9022 __u64 start, __u64 end, struct thandle *th)
9024 if (dt_object_remote(dt))
9027 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
9028 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
9032 * different type of files use the same body_ops because object may be created
9033 * in OUT, where there is no chance to set correct body_ops for each type, so
9034 * body_ops themselves will check file type inside, see lod_read/write/punch for
9037 static const struct dt_body_operations lod_body_ops = {
9038 .dbo_read = lod_read,
9039 .dbo_declare_write = lod_declare_write,
9040 .dbo_write = lod_write,
9041 .dbo_declare_punch = lod_declare_punch,
9042 .dbo_punch = lod_punch,
9046 * Implementation of lu_object_operations::loo_object_init.
9048 * The function determines the type and the index of the target device using
9049 * sequence of the object's FID. Then passes control down to the
9050 * corresponding device:
9051 * OSD for the local objects, OSP for remote
9053 * \see lu_object_operations::loo_object_init() in the API description
9056 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
9057 const struct lu_object_conf *conf)
9059 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
9060 struct lu_device *cdev = NULL;
9061 struct lu_object *cobj;
9062 struct lod_tgt_descs *ltd = NULL;
9063 struct lod_tgt_desc *tgt;
9065 int type = LU_SEQ_RANGE_ANY;
9069 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
9073 if (type == LU_SEQ_RANGE_MDT &&
9074 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
9075 cdev = &lod->lod_child->dd_lu_dev;
9076 } else if (type == LU_SEQ_RANGE_MDT) {
9077 ltd = &lod->lod_mdt_descs;
9079 } else if (type == LU_SEQ_RANGE_OST) {
9080 ltd = &lod->lod_ost_descs;
9087 if (ltd->ltd_tgts_size > idx &&
9088 test_bit(idx, ltd->ltd_tgt_bitmap)) {
9089 tgt = LTD_TGT(ltd, idx);
9091 LASSERT(tgt != NULL);
9092 LASSERT(tgt->ltd_tgt != NULL);
9094 cdev = &(tgt->ltd_tgt->dd_lu_dev);
9096 lod_putref(lod, ltd);
9099 if (unlikely(cdev == NULL))
9102 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
9103 if (unlikely(cobj == NULL))
9106 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
9108 lu_object_add(lo, cobj);
9115 * Release resources associated with striping.
9117 * If the object is striped (regular or directory), then release
9118 * the stripe objects references and free the ldo_stripe array.
9120 * \param[in] env execution environment
9121 * \param[in] lo object
9123 void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo)
9125 struct lod_layout_component *lod_comp;
9126 __u32 obj_attr = lo->ldo_obj.do_lu.lo_header->loh_attr;
9129 if (unlikely(lo->ldo_is_foreign)) {
9130 if (S_ISREG(obj_attr)) {
9131 lod_free_foreign_lov(lo);
9132 lo->ldo_comp_cached = 0;
9133 } else if (S_ISDIR(obj_attr)) {
9134 lod_free_foreign_lmv(lo);
9135 lo->ldo_dir_stripe_loaded = 0;
9137 } else if (lo->ldo_stripe != NULL) {
9138 LASSERT(lo->ldo_comp_entries == NULL);
9139 LASSERT(lo->ldo_dir_stripes_allocated > 0);
9141 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
9142 if (lo->ldo_stripe[i])
9143 dt_object_put(env, lo->ldo_stripe[i]);
9146 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
9147 OBD_FREE(lo->ldo_stripe, j);
9148 lo->ldo_stripe = NULL;
9149 lo->ldo_dir_stripes_allocated = 0;
9150 lo->ldo_dir_stripe_loaded = 0;
9151 lo->ldo_dir_stripe_count = 0;
9152 } else if (lo->ldo_comp_entries != NULL) {
9153 for (i = 0; i < lo->ldo_comp_cnt; i++) {
9154 /* free lod_layout_component::llc_stripe array */
9155 lod_comp = &lo->ldo_comp_entries[i];
9157 if (lod_comp->llc_stripe == NULL)
9159 LASSERT(lod_comp->llc_stripes_allocated != 0);
9160 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
9161 if (lod_comp->llc_stripe[j] != NULL)
9163 &lod_comp->llc_stripe[j]->do_lu);
9165 OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe,
9166 lod_comp->llc_stripes_allocated);
9167 lod_comp->llc_stripe = NULL;
9168 OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices,
9169 lod_comp->llc_stripes_allocated);
9170 lod_comp->llc_ost_indices = NULL;
9171 lod_comp->llc_stripes_allocated = 0;
9173 lod_free_comp_entries(lo);
9174 lo->ldo_comp_cached = 0;
9178 void lod_striping_free(const struct lu_env *env, struct lod_object *lo)
9180 mutex_lock(&lo->ldo_layout_mutex);
9181 lod_striping_free_nolock(env, lo);
9182 mutex_unlock(&lo->ldo_layout_mutex);
9186 * Implementation of lu_object_operations::loo_object_free.
9188 * \see lu_object_operations::loo_object_free() in the API description
9191 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
9193 struct lod_object *lo = lu2lod_obj(o);
9195 /* release all underlying object pinned */
9196 lod_striping_free(env, lo);
9198 /* lo doesn't contain a lu_object_header, so we don't need call_rcu */
9199 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
9203 * Implementation of lu_object_operations::loo_object_release.
9205 * \see lu_object_operations::loo_object_release() in the API description
9208 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
9210 /* XXX: shouldn't we release everything here in case if object
9211 * creation failed before? */
9215 * Implementation of lu_object_operations::loo_object_print.
9217 * \see lu_object_operations::loo_object_print() in the API description
9220 static int lod_object_print(const struct lu_env *env, void *cookie,
9221 lu_printer_t p, const struct lu_object *l)
9223 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
9225 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
9228 const struct lu_object_operations lod_lu_obj_ops = {
9229 .loo_object_init = lod_object_init,
9230 .loo_object_free = lod_object_free,
9231 .loo_object_release = lod_object_release,
9232 .loo_object_print = lod_object_print,