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, true)))
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);
1228 lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
1229 struct dt_object *dt, struct thandle *th,
1230 int comp_idx, int stripe_idx,
1231 struct lod_obj_stripe_cb_data *data)
1233 if (data->locd_declare)
1234 return lod_sub_declare_attr_set(env, dt, data->locd_attr, th);
1236 if (data->locd_attr->la_valid & LA_LAYOUT_VERSION) {
1237 CDEBUG(D_LAYOUT, DFID": set layout version: %u, comp_idx: %d\n",
1238 PFID(lu_object_fid(&dt->do_lu)),
1239 data->locd_attr->la_layout_version, comp_idx);
1242 return lod_sub_attr_set(env, dt, data->locd_attr, th);
1246 * Implementation of dt_object_operations::do_declare_attr_set.
1248 * If the object is striped, then apply the changes to all the stripes.
1250 * \see dt_object_operations::do_declare_attr_set() in the API description
1253 static int lod_declare_attr_set(const struct lu_env *env,
1254 struct dt_object *dt,
1255 const struct lu_attr *attr,
1258 struct dt_object *next = dt_object_child(dt);
1259 struct lod_object *lo = lod_dt_obj(dt);
1264 * declare setattr on the local object
1266 rc = lod_sub_declare_attr_set(env, next, attr, th);
1270 /* osp_declare_attr_set() ignores all attributes other than
1271 * UID, GID, PROJID, and size, and osp_attr_set() ignores all
1272 * but UID, GID and PROJID. Declaration of size attr setting
1273 * happens through lod_declare_init_size(), and not through
1274 * this function. Therefore we need not load striping unless
1275 * ownership is changing. This should save memory and (we hope)
1276 * speed up rename().
1278 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1279 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1282 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1285 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID | LA_MODE |
1286 LA_ATIME | LA_MTIME | LA_CTIME |
1291 * load striping information, notice we don't do this when object
1292 * is being initialized as we don't need this information till
1293 * few specific cases like destroy, chown
1295 rc = lod_striping_load(env, lo);
1299 if (!lod_obj_is_striped(dt))
1303 * if object is striped declare changes on the stripes
1305 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1306 LASSERT(lo->ldo_stripe);
1307 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1308 if (lo->ldo_stripe[i] == NULL)
1310 if (!dt_object_exists(lo->ldo_stripe[i]))
1312 rc = lod_sub_declare_attr_set(env, lo->ldo_stripe[i],
1318 struct lod_obj_stripe_cb_data data = { { 0 } };
1320 data.locd_attr = attr;
1321 data.locd_declare = true;
1322 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1323 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1329 if (!dt_object_exists(next) || dt_object_remote(next) ||
1330 !S_ISREG(attr->la_mode))
1333 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1334 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
1338 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) ||
1339 CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1340 struct lod_thread_info *info = lod_env_info(env);
1341 struct lu_buf *buf = &info->lti_buf;
1343 buf->lb_buf = info->lti_ea_store;
1344 buf->lb_len = info->lti_ea_store_size;
1345 rc = lod_sub_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1346 LU_XATTR_REPLACE, th);
1353 * Implementation of dt_object_operations::do_attr_set.
1355 * If the object is striped, then apply the changes to all or subset of
1356 * the stripes depending on the object type and specific attributes.
1358 * \see dt_object_operations::do_attr_set() in the API description for details.
1360 static int lod_attr_set(const struct lu_env *env,
1361 struct dt_object *dt,
1362 const struct lu_attr *attr,
1365 struct dt_object *next = dt_object_child(dt);
1366 struct lod_object *lo = lod_dt_obj(dt);
1371 * apply changes to the local object
1373 rc = lod_sub_attr_set(env, next, attr, th);
1377 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1378 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1381 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1384 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE | LA_PROJID |
1385 LA_ATIME | LA_MTIME | LA_CTIME |
1390 /* FIXME: a tricky case in the code path of mdd_layout_change():
1391 * the in-memory striping information has been freed in lod_xattr_set()
1392 * due to layout change. It has to load stripe here again. It only
1393 * changes flags of layout so declare_attr_set() is still accurate */
1394 rc = lod_striping_load(env, lo);
1398 if (!lod_obj_is_striped(dt))
1402 * if object is striped, apply changes to all the stripes
1404 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1405 LASSERT(lo->ldo_stripe);
1406 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1407 if (unlikely(lo->ldo_stripe[i] == NULL))
1410 if ((dt_object_exists(lo->ldo_stripe[i]) == 0))
1413 rc = lod_sub_attr_set(env, lo->ldo_stripe[i], attr, th);
1418 struct lod_obj_stripe_cb_data data = { { 0 } };
1420 data.locd_attr = attr;
1421 data.locd_declare = false;
1422 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1423 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1429 if (!dt_object_exists(next) || dt_object_remote(next) ||
1430 !S_ISREG(attr->la_mode))
1433 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1434 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
1438 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE)) {
1439 struct lod_thread_info *info = lod_env_info(env);
1440 struct lu_buf *buf = &info->lti_buf;
1441 struct ost_id *oi = &info->lti_ostid;
1442 struct lu_fid *fid = &info->lti_fid;
1443 struct lov_mds_md_v1 *lmm;
1444 struct lov_ost_data_v1 *objs;
1447 rc = lod_get_lov_ea(env, lo);
1451 buf->lb_buf = info->lti_ea_store;
1452 buf->lb_len = info->lti_ea_store_size;
1453 lmm = info->lti_ea_store;
1454 magic = le32_to_cpu(lmm->lmm_magic);
1455 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1456 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1457 struct lov_comp_md_entry_v1 *lcme =
1458 &lcm->lcm_entries[0];
1460 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
1461 magic = le32_to_cpu(lmm->lmm_magic);
1464 if (magic == LOV_MAGIC_V1)
1465 objs = &(lmm->lmm_objects[0]);
1467 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1468 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1469 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1471 fid_to_ostid(fid, oi);
1472 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1474 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1475 LU_XATTR_REPLACE, th);
1476 } else if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1477 struct lod_thread_info *info = lod_env_info(env);
1478 struct lu_buf *buf = &info->lti_buf;
1479 struct lov_comp_md_v1 *lcm;
1480 struct lov_comp_md_entry_v1 *lcme;
1482 rc = lod_get_lov_ea(env, lo);
1486 buf->lb_buf = info->lti_ea_store;
1487 buf->lb_len = info->lti_ea_store_size;
1489 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1 &&
1490 le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_SEL)
1493 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1494 lcme = &lcm->lcm_entries[0];
1495 le64_add_cpu(&lcme->lcme_extent.e_start, 1);
1496 le64_add_cpu(&lcme->lcme_extent.e_end, -1);
1498 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1499 LU_XATTR_REPLACE, th);
1506 * Implementation of dt_object_operations::do_xattr_get.
1508 * If LOV EA is requested from the root object and it's not
1509 * found, then return default striping for the filesystem.
1511 * \see dt_object_operations::do_xattr_get() in the API description for details.
1513 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1514 struct lu_buf *buf, const char *name)
1516 struct lod_thread_info *info = lod_env_info(env);
1517 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1522 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1523 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1524 struct lmv_mds_md_v1 *lmv1;
1525 struct lmv_foreign_md *lfm;
1528 if (rc > (typeof(rc))sizeof(*lmv1))
1531 /* short (<= sizeof(struct lmv_mds_md_v1)) foreign LMV case */
1532 /* XXX empty foreign LMV is not allowed */
1533 if (rc <= offsetof(typeof(*lfm), lfm_value))
1534 RETURN(rc = rc > 0 ? -EINVAL : rc);
1536 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1537 BUILD_BUG_ON(sizeof(*lmv1) > sizeof(info->lti_key));
1539 /* lti_buf is large enough for *lmv1 or a short
1540 * (<= sizeof(struct lmv_mds_md_v1)) foreign LMV
1542 info->lti_buf.lb_buf = info->lti_key;
1543 info->lti_buf.lb_len = sizeof(*lmv1);
1544 rc = dt_xattr_get(env, dt_object_child(dt),
1545 &info->lti_buf, name);
1546 if (unlikely(rc <= offsetof(typeof(*lfm),
1548 RETURN(rc = rc > 0 ? -EINVAL : rc);
1550 lfm = info->lti_buf.lb_buf;
1551 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN)
1554 if (unlikely(rc != sizeof(*lmv1)))
1555 RETURN(rc = rc > 0 ? -EINVAL : rc);
1557 lmv1 = info->lti_buf.lb_buf;
1558 /* The on-disk LMV EA only contains header, but the
1559 * returned LMV EA size should contain the space for
1560 * the FIDs of all shards of the striped directory. */
1561 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1562 rc = lmv_mds_md_size(
1563 le32_to_cpu(lmv1->lmv_stripe_count),
1564 le32_to_cpu(lmv1->lmv_magic));
1567 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1570 if (rc != sizeof(*lmv1))
1571 RETURN(rc = rc > 0 ? -EINVAL : rc);
1573 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1577 RETURN(rc = rc1 != 0 ? rc1 : rc);
1580 if ((rc > 0) && buf->lb_buf && strcmp(name, XATTR_NAME_LOV) == 0) {
1581 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1583 if (lcm->lcm_magic == cpu_to_le32(LOV_MAGIC_SEL))
1584 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
1587 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1591 * XXX: Only used by lfsck
1593 * lod returns default striping on the real root of the device
1594 * this is like the root stores default striping for the whole
1595 * filesystem. historically we've been using a different approach
1596 * and store it in the config.
1598 dt_root_get(env, dev->lod_child, &info->lti_fid);
1599 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1601 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1602 struct lov_user_md *lum = buf->lb_buf;
1603 struct lov_desc *desc = &dev->lod_ost_descs.ltd_lov_desc;
1605 if (buf->lb_buf == NULL) {
1607 } else if (buf->lb_len >= sizeof(*lum)) {
1608 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1609 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1610 lmm_oi_set_id(&lum->lmm_oi, 0);
1611 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1612 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1613 lum->lmm_stripe_size = cpu_to_le32(
1614 desc->ld_default_stripe_size);
1615 lum->lmm_stripe_count = cpu_to_le16(
1616 desc->ld_default_stripe_count);
1617 lum->lmm_stripe_offset = cpu_to_le16(
1618 desc->ld_default_stripe_offset);
1631 * Checks that the magic of the stripe is sane.
1633 * \param[in] lod lod device
1634 * \param[in] lum a buffer storing LMV EA to verify
1636 * \retval 0 if the EA is sane
1637 * \retval negative otherwise
1639 static int lod_verify_md_striping(struct lod_device *lod,
1640 const struct lmv_user_md_v1 *lum)
1642 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1643 CERROR("%s: invalid lmv_user_md: magic = %x, "
1644 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1645 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1646 (int)le32_to_cpu(lum->lum_stripe_offset),
1647 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1655 * Initialize LMV EA for a slave.
1657 * Initialize slave's LMV EA from the master's LMV EA.
1659 * \param[in] master_lmv a buffer containing master's EA
1660 * \param[out] slave_lmv a buffer where slave's EA will be stored
1663 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1664 const struct lmv_mds_md_v1 *master_lmv)
1666 *slave_lmv = *master_lmv;
1667 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1673 * Generate LMV EA from the object passed as \a dt. The object must have
1674 * the stripes created and initialized.
1676 * \param[in] env execution environment
1677 * \param[in] dt object
1678 * \param[out] lmv_buf buffer storing generated LMV EA
1680 * \retval 0 on success
1681 * \retval negative if failed
1683 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1684 struct lu_buf *lmv_buf)
1686 struct lod_thread_info *info = lod_env_info(env);
1687 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1688 struct lod_object *lo = lod_dt_obj(dt);
1689 struct lmv_mds_md_v1 *lmm1;
1691 int type = LU_SEQ_RANGE_ANY;
1696 LASSERT(lo->ldo_dir_striped != 0);
1697 LASSERT(lo->ldo_dir_stripe_count > 0);
1698 stripe_count = lo->ldo_dir_stripe_count;
1699 /* Only store the LMV EA heahder on the disk. */
1700 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1701 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1705 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1708 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1709 memset(lmm1, 0, sizeof(*lmm1));
1710 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1711 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1712 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1713 lmm1->lmv_layout_version = cpu_to_le32(lo->ldo_dir_layout_version);
1714 if (lod_is_layout_changing(lo)) {
1715 lmm1->lmv_migrate_hash = cpu_to_le32(lo->ldo_dir_migrate_hash);
1716 lmm1->lmv_migrate_offset =
1717 cpu_to_le32(lo->ldo_dir_migrate_offset);
1719 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1724 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1725 lmv_buf->lb_buf = info->lti_ea_store;
1726 lmv_buf->lb_len = sizeof(*lmm1);
1732 * Create in-core represenation for a striped directory.
1734 * Parse the buffer containing LMV EA and instantiate LU objects
1735 * representing the stripe objects. The pointers to the objects are
1736 * stored in ldo_stripe field of \a lo. This function is used when
1737 * we need to access an already created object (i.e. load from a disk).
1739 * \param[in] env execution environment
1740 * \param[in] lo lod object
1741 * \param[in] buf buffer containing LMV EA
1743 * \retval 0 on success
1744 * \retval negative if failed
1746 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1747 const struct lu_buf *buf)
1749 struct lod_thread_info *info = lod_env_info(env);
1750 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1751 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1752 struct dt_object **stripe;
1753 union lmv_mds_md *lmm = buf->lb_buf;
1754 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1755 struct lu_fid *fid = &info->lti_fid;
1760 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1762 /* XXX may be useless as not called for foreign LMV ?? */
1763 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_FOREIGN)
1766 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1767 lo->ldo_dir_slave_stripe = 1;
1771 if (!lmv_is_sane(lmv1))
1774 LASSERT(lo->ldo_stripe == NULL);
1775 OBD_ALLOC_PTR_ARRAY(stripe, le32_to_cpu(lmv1->lmv_stripe_count));
1779 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1780 struct dt_device *tgt_dt;
1781 struct dt_object *dto;
1782 int type = LU_SEQ_RANGE_ANY;
1785 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1786 if (!fid_is_sane(fid)) {
1791 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1795 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1796 tgt_dt = lod->lod_child;
1798 struct lod_tgt_desc *tgt;
1800 tgt = LTD_TGT(ltd, idx);
1802 GOTO(out, rc = -ESTALE);
1803 tgt_dt = tgt->ltd_tgt;
1806 dto = dt_locate_at(env, tgt_dt, fid,
1807 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1810 GOTO(out, rc = PTR_ERR(dto));
1815 lo->ldo_stripe = stripe;
1816 lo->ldo_is_foreign = 0;
1817 lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1818 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1819 lo->ldo_dir_layout_version = le32_to_cpu(lmv1->lmv_layout_version);
1820 lo->ldo_dir_migrate_offset = le32_to_cpu(lmv1->lmv_migrate_offset);
1821 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv1->lmv_migrate_hash);
1822 lo->ldo_dir_hash_type = le32_to_cpu(lmv1->lmv_hash_type);
1824 lod_striping_free_nolock(env, lo);
1830 * Declare create a striped directory.
1832 * Declare creating a striped directory with a given stripe pattern on the
1833 * specified MDTs. A striped directory is represented as a regular directory
1834 * - an index listing all the stripes. The stripes point back to the master
1835 * object with ".." and LinkEA. The master object gets LMV EA which
1836 * identifies it as a striped directory. The function allocates FIDs
1839 * \param[in] env execution environment
1840 * \param[in] dt object
1841 * \param[in] attr attributes to initialize the objects with
1842 * \param[in] dof type of objects to be created
1843 * \param[in] th transaction handle
1845 * \retval 0 on success
1846 * \retval negative if failed
1848 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1849 struct dt_object *dt,
1850 struct lu_attr *attr,
1851 struct dt_object_format *dof,
1854 struct lod_thread_info *info = lod_env_info(env);
1855 struct lu_buf lmv_buf;
1856 struct lu_buf slave_lmv_buf;
1857 struct lmv_mds_md_v1 *lmm;
1858 struct lmv_mds_md_v1 *slave_lmm = NULL;
1859 struct dt_insert_rec *rec = &info->lti_dt_rec;
1860 struct lod_object *lo = lod_dt_obj(dt);
1865 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1868 lmm = lmv_buf.lb_buf;
1870 OBD_ALLOC_PTR(slave_lmm);
1871 if (slave_lmm == NULL)
1872 GOTO(out, rc = -ENOMEM);
1874 lod_prep_slave_lmv_md(slave_lmm, lmm);
1875 slave_lmv_buf.lb_buf = slave_lmm;
1876 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1878 if (!dt_try_as_dir(env, dt_object_child(dt), false))
1879 GOTO(out, rc = -EINVAL);
1881 rec->rec_type = S_IFDIR;
1882 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1883 struct dt_object *dto = lo->ldo_stripe[i];
1884 char *stripe_name = info->lti_key;
1885 struct lu_name *sname;
1886 struct linkea_data ldata = { NULL };
1887 struct lu_buf linkea_buf;
1889 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
1893 /* directory split skip create for existing stripes */
1894 if (!(lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) {
1895 rc = lod_sub_declare_create(env, dto, attr, NULL, dof,
1900 if (!dt_try_as_dir(env, dto, false))
1901 GOTO(out, rc = -EINVAL);
1903 rc = lod_sub_declare_ref_add(env, dto, th);
1907 rec->rec_fid = lu_object_fid(&dto->do_lu);
1908 rc = lod_sub_declare_insert(env, dto,
1909 (const struct dt_rec *)rec,
1910 (const struct dt_key *)dot,
1915 /* master stripe FID will be put to .. */
1916 rec->rec_fid = lu_object_fid(&dt->do_lu);
1917 rc = lod_sub_declare_insert(env, dto,
1918 (const struct dt_rec *)rec,
1919 (const struct dt_key *)dotdot,
1924 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1926 snprintf(stripe_name, sizeof(info->lti_key),
1928 PFID(lu_object_fid(&dto->do_lu)),
1931 snprintf(stripe_name, sizeof(info->lti_key),
1933 PFID(lu_object_fid(&dto->do_lu)), i);
1935 sname = lod_name_get(env, stripe_name,
1936 strlen(stripe_name));
1937 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1938 sname, lu_object_fid(&dt->do_lu));
1942 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1943 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1944 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
1945 XATTR_NAME_LINK, 0, th);
1949 rec->rec_fid = lu_object_fid(&dto->do_lu);
1950 rc = lod_sub_declare_insert(env, dt_object_child(dt),
1951 (const struct dt_rec *)rec,
1952 (const struct dt_key *)stripe_name, th);
1956 rc = lod_sub_declare_ref_add(env, dt_object_child(dt),
1962 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1963 cfs_fail_val != i) {
1964 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1966 slave_lmm->lmv_master_mdt_index =
1969 slave_lmm->lmv_master_mdt_index =
1971 rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
1972 XATTR_NAME_LMV, 0, th);
1978 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
1979 &lmv_buf, XATTR_NAME_LMV, 0, th);
1983 if (slave_lmm != NULL)
1984 OBD_FREE_PTR(slave_lmm);
1990 * Allocate a striping on a predefined set of MDTs.
1992 * Allocates new striping using the MDT index range provided by the data from
1993 * the lum_obejcts contained in the lmv_user_md passed to this method if
1994 * \a is_specific is true; or allocates new layout starting from MDT index in
1995 * lo->ldo_dir_stripe_offset. The exact order of MDTs is not important and
1996 * varies depending on MDT status. The number of stripes needed and stripe
1997 * offset are taken from the object. If that number cannot be met, then the
1998 * function returns an error and then it's the caller's responsibility to
1999 * release the stripes allocated. All the internal structures are protected,
2000 * but no concurrent allocation is allowed on the same objects.
2002 * \param[in] env execution environment for this thread
2003 * \param[in] lo LOD object
2004 * \param[out] stripes striping created
2005 * \param[out] mdt_indices MDT indices of striping created
2006 * \param[in] is_specific true if the MDTs are provided by lum; false if
2007 * only the starting MDT index is provided
2009 * \retval positive stripes allocated, including the first stripe allocated
2011 * \retval negative errno on failure
2013 static int lod_mdt_alloc_specific(const struct lu_env *env,
2014 struct lod_object *lo,
2015 struct dt_object **stripes,
2016 __u32 *mdt_indices, bool is_specific)
2018 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
2019 struct lu_tgt_descs *ltd = &lod->lod_mdt_descs;
2020 struct lu_tgt_desc *tgt = NULL;
2021 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2022 struct dt_device *tgt_dt = NULL;
2023 struct lu_fid fid = { 0 };
2024 struct dt_object *dto;
2026 u32 stripe_count = lo->ldo_dir_stripe_count;
2032 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
2033 if (!is_specific && stripe_count > 1)
2034 /* Set the start index for the 2nd stripe allocation */
2035 mdt_indices[1] = (mdt_indices[0] + 1) %
2036 (lod->lod_remote_mdt_count + 1);
2038 for (; stripe_idx < stripe_count; stripe_idx++) {
2039 /* Try to find next avaible target */
2040 idx = mdt_indices[stripe_idx];
2041 for (j = 0; j < lod->lod_remote_mdt_count;
2042 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
2043 bool already_allocated = false;
2046 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
2047 idx, lod->lod_remote_mdt_count + 1, stripe_idx);
2049 if (likely(!is_specific &&
2050 !CFS_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
2051 /* check whether the idx already exists
2052 * in current allocated array */
2053 for (k = 0; k < stripe_idx; k++) {
2054 if (mdt_indices[k] == idx) {
2055 already_allocated = true;
2060 if (already_allocated)
2064 /* Sigh, this index is not in the bitmap, let's check
2065 * next available target */
2066 if (!test_bit(idx, ltd->ltd_tgt_bitmap) &&
2067 idx != master_index)
2070 if (idx == master_index) {
2071 /* Allocate the FID locally */
2072 tgt_dt = lod->lod_child;
2073 rc = dt_fid_alloc(env, tgt_dt, &fid, NULL,
2080 /* check the status of the OSP */
2081 tgt = LTD_TGT(ltd, idx);
2085 tgt_dt = tgt->ltd_tgt;
2086 if (!tgt->ltd_active)
2087 /* this OSP doesn't feel well */
2090 rc = dt_fid_alloc(env, tgt_dt, &fid, NULL, NULL);
2097 /* Can not allocate more stripes */
2098 if (j == lod->lod_remote_mdt_count) {
2099 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
2100 lod2obd(lod)->obd_name, stripe_count,
2105 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
2106 idx, stripe_idx, PFID(&fid));
2107 mdt_indices[stripe_idx] = idx;
2108 /* Set the start index for next stripe allocation */
2109 if (!is_specific && stripe_idx < stripe_count - 1) {
2111 * for large dir test, put all other slaves on one
2112 * remote MDT, otherwise we may save too many local
2113 * slave locks which will exceed RS_MAX_LOCKS.
2115 if (unlikely(CFS_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)))
2117 mdt_indices[stripe_idx + 1] = (idx + 1) %
2118 (lod->lod_remote_mdt_count + 1);
2120 /* tgt_dt and fid must be ready after search avaible OSP
2121 * in the above loop */
2122 LASSERT(tgt_dt != NULL);
2123 LASSERT(fid_is_sane(&fid));
2125 /* fail a remote stripe FID allocation */
2126 if (stripe_idx && CFS_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_FID))
2129 dto = dt_locate_at(env, tgt_dt, &fid,
2130 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
2137 stripes[stripe_idx] = dto;
2143 for (j = 1; j < stripe_idx; j++) {
2144 LASSERT(stripes[j] != NULL);
2145 dt_object_put(env, stripes[j]);
2151 static int lod_prep_md_striped_create(const struct lu_env *env,
2152 struct dt_object *dt,
2153 struct lu_attr *attr,
2154 const struct lmv_user_md_v1 *lum,
2155 struct dt_object_format *dof,
2158 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
2159 struct lod_object *lo = lod_dt_obj(dt);
2160 struct dt_object **stripes;
2161 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2162 struct lu_fid fid = { 0 };
2169 /* The lum has been verifed in lod_verify_md_striping */
2170 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
2171 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
2173 stripe_count = lo->ldo_dir_stripe_count;
2175 OBD_ALLOC_PTR_ARRAY(stripes, stripe_count);
2179 /* Allocate the first stripe locally */
2180 rc = dt_fid_alloc(env, lod->lod_child, &fid, NULL, NULL);
2184 stripes[0] = dt_locate_at(env, lod->lod_child, &fid,
2185 dt->do_lu.lo_dev->ld_site->ls_top_dev, &conf);
2186 if (IS_ERR(stripes[0]))
2187 GOTO(out, rc = PTR_ERR(stripes[0]));
2189 if (lo->ldo_dir_stripe_offset == LMV_OFFSET_DEFAULT) {
2190 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
2191 rc = lod_mdt_alloc_qos(env, lo, stripes, 1, stripe_count);
2193 rc = lod_mdt_alloc_rr(env, lo, stripes, 1,
2197 bool is_specific = false;
2199 OBD_ALLOC_PTR_ARRAY(idx_array, stripe_count);
2201 GOTO(out, rc = -ENOMEM);
2203 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
2205 for (i = 0; i < stripe_count; i++)
2207 le32_to_cpu(lum->lum_objects[i].lum_mds);
2210 /* stripe 0 is local */
2212 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
2213 rc = lod_mdt_alloc_specific(env, lo, stripes, idx_array,
2215 OBD_FREE_PTR_ARRAY(idx_array, stripe_count);
2223 lo->ldo_dir_striped = 1;
2224 lo->ldo_stripe = stripes;
2225 lo->ldo_dir_stripe_count = rc;
2226 lo->ldo_dir_stripes_allocated = stripe_count;
2228 lo->ldo_dir_stripe_loaded = 1;
2230 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2232 lod_striping_free(env, lo);
2238 if (!IS_ERR_OR_NULL(stripes[0]))
2239 dt_object_put(env, stripes[0]);
2240 for (i = 1; i < stripe_count; i++)
2241 LASSERT(!stripes[i]);
2242 OBD_FREE_PTR_ARRAY(stripes, stripe_count);
2249 * Alloc cached foreign LOV
2251 * \param[in] lo object
2252 * \param[in] size size of foreign LOV
2254 * \retval 0 on success
2255 * \retval negative if failed
2257 int lod_alloc_foreign_lov(struct lod_object *lo, size_t size)
2259 OBD_ALLOC_LARGE(lo->ldo_foreign_lov, size);
2260 if (lo->ldo_foreign_lov == NULL)
2262 lo->ldo_foreign_lov_size = size;
2263 lo->ldo_is_foreign = 1;
2269 * Free cached foreign LOV
2271 * \param[in] lo object
2273 void lod_free_foreign_lov(struct lod_object *lo)
2275 if (lo->ldo_foreign_lov != NULL)
2276 OBD_FREE_LARGE(lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
2277 lo->ldo_foreign_lov = NULL;
2278 lo->ldo_foreign_lov_size = 0;
2279 lo->ldo_is_foreign = 0;
2284 * Alloc cached foreign LMV
2286 * \param[in] lo object
2287 * \param[in] size size of foreign LMV
2289 * \retval 0 on success
2290 * \retval negative if failed
2292 int lod_alloc_foreign_lmv(struct lod_object *lo, size_t size)
2294 OBD_ALLOC_LARGE(lo->ldo_foreign_lmv, size);
2295 if (lo->ldo_foreign_lmv == NULL)
2297 lo->ldo_foreign_lmv_size = size;
2298 lo->ldo_is_foreign = 1;
2305 * Free cached foreign LMV
2307 * \param[in] lo object
2309 void lod_free_foreign_lmv(struct lod_object *lo)
2311 if (lo->ldo_foreign_lmv != NULL)
2312 OBD_FREE_LARGE(lo->ldo_foreign_lmv, lo->ldo_foreign_lmv_size);
2313 lo->ldo_foreign_lmv = NULL;
2314 lo->ldo_foreign_lmv_size = 0;
2315 lo->ldo_is_foreign = 0;
2319 * Declare create striped md object.
2321 * The function declares intention to create a striped directory. This is a
2322 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2323 * is to verify pattern \a lum_buf is good. Check that function for the details.
2325 * \param[in] env execution environment
2326 * \param[in] dt object
2327 * \param[in] attr attributes to initialize the objects with
2328 * \param[in] lum_buf a pattern specifying the number of stripes and
2330 * \param[in] dof type of objects to be created
2331 * \param[in] th transaction handle
2333 * \retval 0 on success
2334 * \retval negative if failed
2337 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2338 struct dt_object *dt,
2339 struct lu_attr *attr,
2340 const struct lu_buf *lum_buf,
2341 struct dt_object_format *dof,
2344 struct lod_object *lo = lod_dt_obj(dt);
2345 struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
2349 LASSERT(lum != NULL);
2352 "lum magic=%x hash=%x count=%u offset=%d inherit=%u rr=%u\n",
2353 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_hash_type),
2354 le32_to_cpu(lum->lum_stripe_count),
2355 (int)le32_to_cpu(lum->lum_stripe_offset),
2356 lum->lum_max_inherit, lum->lum_max_inherit_rr);
2358 if (lo->ldo_dir_stripe_count == 0) {
2359 if (lo->ldo_is_foreign) {
2360 rc = lod_alloc_foreign_lmv(lo, lum_buf->lb_len);
2363 memcpy(lo->ldo_foreign_lmv, lum, lum_buf->lb_len);
2364 lo->ldo_dir_stripe_loaded = 1;
2369 /* prepare dir striped objects */
2370 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2372 /* failed to create striping, let's reset
2373 * config so that others don't get confused */
2374 lod_striping_free(env, lo);
2382 * Set or replace striped directory layout, and LFSCK may set layout on a plain
2383 * directory, so don't check stripe count.
2385 * \param[in] env execution environment
2386 * \param[in] dt target object
2387 * \param[in] lmv_buf LMV buf which contains source stripe FIDs
2388 * \param[in] fl set or replace
2389 * \param[in] th transaction handle
2391 * \retval 0 on success
2392 * \retval negative if failed
2394 static int lod_dir_layout_set(const struct lu_env *env,
2395 struct dt_object *dt,
2396 const struct lu_buf *lmv_buf,
2400 struct dt_object *next = dt_object_child(dt);
2401 struct lod_object *lo = lod_dt_obj(dt);
2402 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2403 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
2404 struct lmv_mds_md_v1 *slave_lmv;
2405 struct lu_buf slave_buf;
2411 if (!lmv_is_sane2(lmv))
2414 /* adjust hash for dir merge, which may not be set in user command */
2415 if (lmv_is_merging(lmv) &&
2416 !(lmv->lmv_migrate_hash & LMV_HASH_TYPE_MASK))
2417 lmv->lmv_merge_hash |=
2418 lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern &
2421 LMV_DEBUG(D_INFO, lmv, "set");
2423 rc = lod_sub_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV, fl, th);
2427 /* directory restripe may update stripe LMV directly */
2428 if (!lo->ldo_dir_stripe_count)
2431 lo->ldo_dir_hash_type = le32_to_cpu(lmv->lmv_hash_type);
2432 lo->ldo_dir_migrate_offset = le32_to_cpu(lmv->lmv_migrate_offset);
2433 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_migrate_hash);
2434 lo->ldo_dir_layout_version = le32_to_cpu(lmv->lmv_layout_version);
2436 OBD_ALLOC_PTR(slave_lmv);
2440 lod_prep_slave_lmv_md(slave_lmv, lmv);
2441 slave_buf.lb_buf = slave_lmv;
2442 slave_buf.lb_len = sizeof(*slave_lmv);
2444 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2445 if (!lo->ldo_stripe[i])
2448 if (!dt_object_exists(lo->ldo_stripe[i]))
2451 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], &slave_buf,
2452 XATTR_NAME_LMV, fl, th);
2457 OBD_FREE_PTR(slave_lmv);
2463 * Implementation of dt_object_operations::do_declare_xattr_set.
2465 * Used with regular (non-striped) objects. Basically it
2466 * initializes the striping information and applies the
2467 * change to all the stripes.
2469 * \see dt_object_operations::do_declare_xattr_set() in the API description
2472 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2473 struct dt_object *dt,
2474 const struct lu_buf *buf,
2475 const char *name, int fl,
2478 struct dt_object *next = dt_object_child(dt);
2479 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2480 struct lod_object *lo = lod_dt_obj(dt);
2485 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2486 struct lmv_user_md_v1 *lum;
2488 LASSERT(buf != NULL && buf->lb_buf != NULL);
2490 rc = lod_verify_md_striping(d, lum);
2493 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2494 rc = lod_verify_striping(env, d, lo, buf, false);
2499 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2503 /* Note: Do not set LinkEA on sub-stripes, otherwise
2504 * it will confuse the fid2path process(see mdt_path_current()).
2505 * The linkEA between master and sub-stripes is set in
2506 * lod_xattr_set_lmv(). */
2507 if (strcmp(name, XATTR_NAME_LINK) == 0)
2510 /* set xattr to each stripes, if needed */
2511 rc = lod_striping_load(env, lo);
2515 if (lo->ldo_dir_stripe_count == 0)
2518 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2519 if (!lo->ldo_stripe[i])
2522 if (!dt_object_exists(lo->ldo_stripe[i]))
2525 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2535 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2536 struct lod_object *lo,
2537 struct dt_object *dt, struct thandle *th,
2538 int comp_idx, int stripe_idx,
2539 struct lod_obj_stripe_cb_data *data)
2541 struct lod_thread_info *info = lod_env_info(env);
2542 struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
2543 struct filter_fid *ff = &info->lti_ff;
2544 struct lu_buf *buf = &info->lti_buf;
2548 buf->lb_len = sizeof(*ff);
2549 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2557 * locd_buf is set if it's called by dir migration, which doesn't check
2560 if (data->locd_buf) {
2561 memset(ff, 0, sizeof(*ff));
2562 ff->ff_parent = *(struct lu_fid *)data->locd_buf->lb_buf;
2564 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
2566 if (lu_fid_eq(lod_object_fid(lo), &ff->ff_parent) &&
2567 ff->ff_layout.ol_comp_id == comp->llc_id)
2570 memset(ff, 0, sizeof(*ff));
2571 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2574 /* rewrite filter_fid */
2575 ff->ff_parent.f_ver = stripe_idx;
2576 ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
2577 ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
2578 ff->ff_layout.ol_comp_id = comp->llc_id;
2579 ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
2580 ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
2581 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
2583 if (data->locd_declare)
2584 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2585 LU_XATTR_REPLACE, th);
2587 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2588 LU_XATTR_REPLACE, th);
2594 * Reset parent FID on OST object
2596 * Replace parent FID with @dt object FID, which is only called during migration
2597 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2598 * the FID is changed.
2600 * \param[in] env execution environment
2601 * \param[in] dt dt_object whose stripes's parent FID will be reset
2602 * \parem[in] th thandle
2603 * \param[in] declare if it is declare
2605 * \retval 0 if reset succeeds
2606 * \retval negative errno if reset fails
2608 static int lod_replace_parent_fid(const struct lu_env *env,
2609 struct dt_object *dt,
2610 const struct lu_buf *buf,
2611 struct thandle *th, bool declare)
2613 struct lod_object *lo = lod_dt_obj(dt);
2614 struct lod_thread_info *info = lod_env_info(env);
2615 struct filter_fid *ff;
2616 struct lod_obj_stripe_cb_data data = { { 0 } };
2620 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2622 /* set xattr to each stripes, if needed */
2623 rc = lod_striping_load(env, lo);
2627 if (!lod_obj_is_striped(dt))
2630 if (info->lti_ea_store_size < sizeof(*ff)) {
2631 rc = lod_ea_store_resize(info, sizeof(*ff));
2636 data.locd_declare = declare;
2637 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2638 data.locd_buf = buf;
2639 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2644 __u16 lod_comp_entry_stripe_count(struct lod_object *lo,
2645 int comp_idx, bool is_dir)
2647 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2648 struct lod_layout_component *entry;
2653 entry = &lo->ldo_comp_entries[comp_idx];
2654 if (lod_comp_inited(entry))
2655 return entry->llc_stripe_count;
2656 else if ((__u16)-1 == entry->llc_stripe_count)
2657 return lod->lod_ost_count;
2659 return lod_get_stripe_count(lod, lo, comp_idx,
2660 entry->llc_stripe_count,
2661 entry->llc_pattern &
2662 LOV_PATTERN_OVERSTRIPING);
2665 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2667 int magic, size = 0, i;
2668 struct lod_layout_component *comp_entries;
2670 bool is_composite, is_foreign = false;
2673 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2674 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2676 lo->ldo_def_striping->lds_def_striping_is_composite;
2678 comp_cnt = lo->ldo_comp_cnt;
2679 comp_entries = lo->ldo_comp_entries;
2680 is_composite = lo->ldo_is_composite;
2681 is_foreign = lo->ldo_is_foreign;
2685 return lo->ldo_foreign_lov_size;
2687 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2689 size = sizeof(struct lov_comp_md_v1) +
2690 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2691 LASSERT(size % sizeof(__u64) == 0);
2694 for (i = 0; i < comp_cnt; i++) {
2697 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2698 stripe_count = lod_comp_entry_stripe_count(lo, i, is_dir);
2699 if (!is_dir && is_composite)
2700 lod_comp_shrink_stripe_count(&comp_entries[i],
2703 size += lov_user_md_size(stripe_count, magic);
2704 LASSERT(size % sizeof(__u64) == 0);
2710 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2711 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2714 * \param[in] env execution environment
2715 * \param[in] dt dt_object to add components on
2716 * \param[in] buf buffer contains components to be added
2717 * \parem[in] th thandle
2719 * \retval 0 on success
2720 * \retval negative errno on failure
2722 static int lod_declare_layout_add(const struct lu_env *env,
2723 struct dt_object *dt,
2724 const struct lu_buf *buf,
2727 struct lod_thread_info *info = lod_env_info(env);
2728 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2729 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2730 struct dt_object *next = dt_object_child(dt);
2731 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
2732 struct lod_object *lo = lod_dt_obj(dt);
2733 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2735 int i, rc, array_cnt, old_array_cnt;
2738 LASSERT(lo->ldo_is_composite);
2740 if (lo->ldo_flr_state != LCM_FL_NONE)
2743 rc = lod_verify_striping(env, d, lo, buf, false);
2747 magic = comp_v1->lcm_magic;
2748 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2749 lustre_swab_lov_comp_md_v1(comp_v1);
2750 magic = comp_v1->lcm_magic;
2753 if (magic != LOV_USER_MAGIC_COMP_V1)
2756 mutex_lock(&lo->ldo_layout_mutex);
2758 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2759 OBD_ALLOC_PTR_ARRAY(comp_array, array_cnt);
2760 if (comp_array == NULL) {
2761 mutex_unlock(&lo->ldo_layout_mutex);
2766 memcpy(comp_array, lo->ldo_comp_entries,
2767 sizeof(*comp_array) * lo->ldo_comp_cnt);
2769 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2770 struct lov_user_md_v1 *v1;
2771 struct lu_extent *ext;
2773 v1 = (struct lov_user_md *)((char *)comp_v1 +
2774 comp_v1->lcm_entries[i].lcme_offset);
2775 ext = &comp_v1->lcm_entries[i].lcme_extent;
2777 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2778 lod_comp->llc_extent.e_start = ext->e_start;
2779 lod_comp->llc_extent.e_end = ext->e_end;
2780 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2781 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2783 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2784 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2785 lod_adjust_stripe_info(lod_comp, desc, 0);
2787 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2788 struct lov_user_md_v3 *v3 = (typeof(*v3) *) v1;
2790 if (v3->lmm_pool_name[0] != '\0' &&
2791 !lov_pool_is_ignored(v3->lmm_pool_name)) {
2792 rc = lod_set_pool(&lod_comp->llc_pool,
2800 old_array = lo->ldo_comp_entries;
2801 old_array_cnt = lo->ldo_comp_cnt;
2803 lo->ldo_comp_entries = comp_array;
2804 lo->ldo_comp_cnt = array_cnt;
2806 /* No need to increase layout generation here, it will be increased
2807 * later when generating component ID for the new components */
2809 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2810 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2811 XATTR_NAME_LOV, 0, th);
2813 lo->ldo_comp_entries = old_array;
2814 lo->ldo_comp_cnt = old_array_cnt;
2818 OBD_FREE_PTR_ARRAY(old_array, old_array_cnt);
2820 LASSERT(lo->ldo_mirror_count == 1);
2821 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2823 mutex_unlock(&lo->ldo_layout_mutex);
2828 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2829 lod_comp = &comp_array[i];
2830 if (lod_comp->llc_pool != NULL) {
2831 OBD_FREE(lod_comp->llc_pool,
2832 strlen(lod_comp->llc_pool) + 1);
2833 lod_comp->llc_pool = NULL;
2836 OBD_FREE_PTR_ARRAY(comp_array, array_cnt);
2837 mutex_unlock(&lo->ldo_layout_mutex);
2843 * lod_last_non_stale_mirror() - Check if a mirror is the last non-stale mirror.
2844 * @mirror_id: Mirror id to be checked.
2847 * This function checks if a mirror with specified @mirror_id is the last
2848 * non-stale mirror of a LOD object @lo.
2850 * Return: true or false.
2853 bool lod_last_non_stale_mirror(__u16 mirror_id, struct lod_object *lo)
2855 struct lod_layout_component *lod_comp;
2856 bool has_stale_flag;
2859 for (i = 0; i < lo->ldo_mirror_count; i++) {
2860 if (lo->ldo_mirrors[i].lme_id == mirror_id ||
2861 lo->ldo_mirrors[i].lme_stale)
2864 has_stale_flag = false;
2865 lod_foreach_mirror_comp(lod_comp, lo, i) {
2866 if (lod_comp->llc_flags & LCME_FL_STALE) {
2867 has_stale_flag = true;
2871 if (!has_stale_flag)
2879 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2880 * the '$field' can only be 'flags' now. The xattr value is binary
2881 * lov_comp_md_v1 which contains the component ID(s) and the value of
2882 * the field to be modified.
2883 * Please update allowed_lustre_lov macro if $field groks more values
2886 * \param[in] env execution environment
2887 * \param[in] dt dt_object to be modified
2888 * \param[in] op operation string, like "set.flags"
2889 * \param[in] buf buffer contains components to be set
2890 * \parem[in] th thandle
2892 * \retval 0 on success
2893 * \retval negative errno on failure
2895 static int lod_declare_layout_set(const struct lu_env *env,
2896 struct dt_object *dt,
2897 char *op, const struct lu_buf *buf,
2900 struct lod_layout_component *lod_comp;
2901 struct lod_thread_info *info = lod_env_info(env);
2902 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2903 struct lod_object *lo = lod_dt_obj(dt);
2904 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2907 bool changed = false;
2910 /* Please update allowed_lustre_lov macro if op
2911 * groks more values in the future
2913 if (strcmp(op, "set.flags") != 0) {
2914 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2915 lod2obd(d)->obd_name, op);
2919 magic = comp_v1->lcm_magic;
2920 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2921 lustre_swab_lov_comp_md_v1(comp_v1);
2922 magic = comp_v1->lcm_magic;
2925 if (magic != LOV_USER_MAGIC_COMP_V1)
2928 if (comp_v1->lcm_entry_count == 0) {
2929 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2930 lod2obd(d)->obd_name);
2934 mutex_lock(&lo->ldo_layout_mutex);
2935 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2936 __u32 id = comp_v1->lcm_entries[i].lcme_id;
2937 __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
2938 __u32 mirror_flag = flags & LCME_MIRROR_FLAGS;
2939 __u16 mirror_id = mirror_id_of(id);
2940 bool neg = flags & LCME_FL_NEG;
2942 if (flags & LCME_FL_INIT) {
2944 lod_striping_free_nolock(env, lo);
2945 mutex_unlock(&lo->ldo_layout_mutex);
2949 flags &= ~(LCME_MIRROR_FLAGS | LCME_FL_NEG);
2950 for (j = 0; j < lo->ldo_comp_cnt; j++) {
2951 lod_comp = &lo->ldo_comp_entries[j];
2953 /* lfs only put one flag in each entry */
2954 if ((flags && id != lod_comp->llc_id) ||
2955 (mirror_flag && mirror_id !=
2956 mirror_id_of(lod_comp->llc_id)))
2961 lod_comp->llc_flags &= ~flags;
2963 lod_comp->llc_flags &= ~mirror_flag;
2966 if ((flags & LCME_FL_STALE) &&
2967 lod_last_non_stale_mirror(mirror_id,
2970 &lo->ldo_layout_mutex);
2973 lod_comp->llc_flags |= flags;
2976 lod_comp->llc_flags |= mirror_flag;
2977 if (mirror_flag & LCME_FL_NOSYNC)
2978 lod_comp->llc_timestamp =
2979 ktime_get_real_seconds();
2985 mutex_unlock(&lo->ldo_layout_mutex);
2988 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
2989 lod2obd(d)->obd_name);
2993 lod_obj_inc_layout_gen(lo);
2995 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2996 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
2997 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3002 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
3003 * and the xattr value is a unique component ID or a special lcme_id.
3005 * \param[in] env execution environment
3006 * \param[in] dt dt_object to be operated on
3007 * \param[in] buf buffer contains component ID or lcme_id
3008 * \parem[in] th thandle
3010 * \retval 0 on success
3011 * \retval negative errno on failure
3013 static int lod_declare_layout_del(const struct lu_env *env,
3014 struct dt_object *dt,
3015 const struct lu_buf *buf,
3018 struct lod_thread_info *info = lod_env_info(env);
3019 struct dt_object *next = dt_object_child(dt);
3020 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3021 struct lod_object *lo = lod_dt_obj(dt);
3022 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3023 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3024 __u32 magic, id, flags, neg_flags = 0;
3028 LASSERT(lo->ldo_is_composite);
3030 if (lo->ldo_flr_state != LCM_FL_NONE)
3033 magic = comp_v1->lcm_magic;
3034 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
3035 lustre_swab_lov_comp_md_v1(comp_v1);
3036 magic = comp_v1->lcm_magic;
3039 if (magic != LOV_USER_MAGIC_COMP_V1)
3042 id = comp_v1->lcm_entries[0].lcme_id;
3043 flags = comp_v1->lcm_entries[0].lcme_flags;
3045 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
3046 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
3047 lod2obd(d)->obd_name, id, flags);
3051 if (id != LCME_ID_INVAL && flags != 0) {
3052 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
3053 lod2obd(d)->obd_name);
3057 if (id == LCME_ID_INVAL && !flags) {
3058 CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
3059 lod2obd(d)->obd_name);
3063 if (flags & LCME_FL_NEG) {
3064 neg_flags = flags & ~LCME_FL_NEG;
3068 mutex_lock(&lo->ldo_layout_mutex);
3070 left = lo->ldo_comp_cnt;
3072 mutex_unlock(&lo->ldo_layout_mutex);
3076 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3077 struct lod_layout_component *lod_comp;
3079 lod_comp = &lo->ldo_comp_entries[i];
3081 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
3083 else if (flags && !(flags & lod_comp->llc_flags))
3085 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
3088 if (left != (i + 1)) {
3089 CDEBUG(D_LAYOUT, "%s: this deletion will create "
3090 "a hole.\n", lod2obd(d)->obd_name);
3091 mutex_unlock(&lo->ldo_layout_mutex);
3096 /* Mark the component as deleted */
3097 lod_comp->llc_id = LCME_ID_INVAL;
3099 /* Not instantiated component */
3100 if (lod_comp->llc_stripe == NULL)
3103 LASSERT(lod_comp->llc_stripe_count > 0);
3104 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
3105 struct dt_object *obj = lod_comp->llc_stripe[j];
3109 rc = lod_sub_declare_destroy(env, obj, th);
3111 mutex_unlock(&lo->ldo_layout_mutex);
3117 LASSERTF(left >= 0, "left = %d\n", left);
3118 if (left == lo->ldo_comp_cnt) {
3119 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
3120 lod2obd(d)->obd_name, id);
3121 mutex_unlock(&lo->ldo_layout_mutex);
3125 mutex_unlock(&lo->ldo_layout_mutex);
3127 memset(attr, 0, sizeof(*attr));
3128 attr->la_valid = LA_SIZE;
3129 rc = lod_sub_declare_attr_set(env, next, attr, th);
3134 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3135 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
3136 XATTR_NAME_LOV, 0, th);
3138 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
3145 * Declare layout add/set/del operations issued by special xattr names:
3147 * XATTR_LUSTRE_LOV.add add component(s) to existing file
3148 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
3149 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
3151 * \param[in] env execution environment
3152 * \param[in] dt object
3153 * \param[in] name name of xattr
3154 * \param[in] buf lu_buf contains xattr value
3155 * \param[in] th transaction handle
3157 * \retval 0 on success
3158 * \retval negative if failed
3160 static int lod_declare_modify_layout(const struct lu_env *env,
3161 struct dt_object *dt,
3163 const struct lu_buf *buf,
3166 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3167 struct lod_object *lo = lod_dt_obj(dt);
3169 int rc, len = strlen(XATTR_LUSTRE_LOV);
3172 LASSERT(dt_object_exists(dt));
3174 if (strlen(name) <= len || name[len] != '.') {
3175 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
3176 lod2obd(d)->obd_name, name);
3181 rc = lod_striping_load(env, lo);
3185 /* the layout to be modified must be a composite layout */
3186 if (!lo->ldo_is_composite) {
3187 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
3188 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3189 GOTO(unlock, rc = -EINVAL);
3192 op = (char *)name + len;
3193 if (strcmp(op, "add") == 0) {
3194 rc = lod_declare_layout_add(env, dt, buf, th);
3195 } else if (strcmp(op, "del") == 0) {
3196 rc = lod_declare_layout_del(env, dt, buf, th);
3197 } else if (strncmp(op, "set", strlen("set")) == 0) {
3198 rc = lod_declare_layout_set(env, dt, op, buf, th);
3200 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
3201 lod2obd(d)->obd_name, name);
3202 GOTO(unlock, rc = -ENOTSUPP);
3206 lod_striping_free(env, lo);
3212 * Convert a plain file lov_mds_md to a composite layout.
3214 * \param[in,out] info the thread info::lti_ea_store buffer contains little
3215 * endian plain file layout
3217 * \retval 0 on success, <0 on failure
3219 static int lod_layout_convert(struct lod_thread_info *info)
3221 struct lov_mds_md *lmm = info->lti_ea_store;
3222 struct lov_mds_md *lmm_save;
3223 struct lov_comp_md_v1 *lcm;
3224 struct lov_comp_md_entry_v1 *lcme;
3230 /* realloc buffer to a composite layout which contains one component */
3231 blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3232 le32_to_cpu(lmm->lmm_magic));
3233 size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
3235 OBD_ALLOC_LARGE(lmm_save, blob_size);
3237 GOTO(out, rc = -ENOMEM);
3239 memcpy(lmm_save, lmm, blob_size);
3241 if (info->lti_ea_store_size < size) {
3242 rc = lod_ea_store_resize(info, size);
3247 lcm = info->lti_ea_store;
3248 memset(lcm, 0, sizeof(*lcm) + sizeof(*lcme));
3249 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
3250 lcm->lcm_size = cpu_to_le32(size);
3251 lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
3252 lmm_save->lmm_layout_gen));
3253 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
3254 lcm->lcm_entry_count = cpu_to_le16(1);
3256 lcme = &lcm->lcm_entries[0];
3257 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
3258 lcme->lcme_extent.e_start = 0;
3259 lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
3260 lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
3261 lcme->lcme_size = cpu_to_le32(blob_size);
3263 memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
3268 OBD_FREE_LARGE(lmm_save, blob_size);
3273 * Merge layouts to form a mirrored file.
3275 static int lod_declare_layout_merge(const struct lu_env *env,
3276 struct dt_object *dt, const struct lu_buf *mbuf,
3279 struct lod_thread_info *info = lod_env_info(env);
3280 struct lu_attr *layout_attr = &info->lti_layout_attr;
3281 struct lu_buf *buf = &info->lti_buf;
3282 struct lod_object *lo = lod_dt_obj(dt);
3283 struct lov_comp_md_v1 *lcm;
3284 struct lov_comp_md_v1 *cur_lcm;
3285 struct lov_comp_md_v1 *merge_lcm;
3286 struct lov_comp_md_entry_v1 *lcme;
3287 struct lov_mds_md_v1 *lmm;
3290 __u16 cur_entry_count;
3291 __u16 merge_entry_count;
3293 __u16 mirror_id = 0;
3300 merge_lcm = mbuf->lb_buf;
3301 if (mbuf->lb_len < sizeof(*merge_lcm))
3304 /* must be an existing layout from disk */
3305 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
3308 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
3310 /* do not allow to merge two mirrored files */
3311 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
3314 /* verify the target buffer */
3315 rc = lod_get_lov_ea(env, lo);
3317 RETURN(rc ? : -ENODATA);
3319 cur_lcm = info->lti_ea_store;
3320 switch (le32_to_cpu(cur_lcm->lcm_magic)) {
3323 rc = lod_layout_convert(info);
3325 case LOV_MAGIC_COMP_V1:
3335 /* info->lti_ea_store could be reallocated in lod_layout_convert() */
3336 cur_lcm = info->lti_ea_store;
3337 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
3339 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
3340 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
3341 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
3344 /* size of new layout */
3345 size = le32_to_cpu(cur_lcm->lcm_size) +
3346 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
3348 memset(buf, 0, sizeof(*buf));
3349 lu_buf_alloc(buf, size);
3350 if (buf->lb_buf == NULL)
3354 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
3356 offset = sizeof(*lcm) +
3357 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
3358 for (i = 0; i < cur_entry_count; i++) {
3359 struct lov_comp_md_entry_v1 *cur_lcme;
3361 lcme = &lcm->lcm_entries[i];
3362 cur_lcme = &cur_lcm->lcm_entries[i];
3364 lcme->lcme_offset = cpu_to_le32(offset);
3365 memcpy((char *)lcm + offset,
3366 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
3367 le32_to_cpu(lcme->lcme_size));
3369 offset += le32_to_cpu(lcme->lcme_size);
3371 if (mirror_count == 1 &&
3372 mirror_id_of(le32_to_cpu(lcme->lcme_id)) == 0) {
3373 /* Add mirror from a non-flr file, create new mirror ID.
3374 * Otherwise, keep existing mirror's component ID, used
3375 * for mirror extension.
3377 id = pflr_id(1, i + 1);
3378 lcme->lcme_id = cpu_to_le32(id);
3381 id = max(le32_to_cpu(lcme->lcme_id), id);
3384 mirror_id = mirror_id_of(id) + 1;
3386 /* check if first entry in new layout is DOM */
3387 lmm = (struct lov_mds_md_v1 *)((char *)merge_lcm +
3388 merge_lcm->lcm_entries[0].lcme_offset);
3389 merge_has_dom = lov_pattern(le32_to_cpu(lmm->lmm_pattern)) ==
3392 for (i = 0; i < merge_entry_count; i++) {
3393 struct lov_comp_md_entry_v1 *merge_lcme;
3395 merge_lcme = &merge_lcm->lcm_entries[i];
3396 lcme = &lcm->lcm_entries[cur_entry_count + i];
3398 *lcme = *merge_lcme;
3399 lcme->lcme_offset = cpu_to_le32(offset);
3400 if (merge_has_dom && i == 0)
3401 lcme->lcme_flags |= cpu_to_le32(LCME_FL_STALE);
3403 id = pflr_id(mirror_id, i + 1);
3404 lcme->lcme_id = cpu_to_le32(id);
3406 memcpy((char *)lcm + offset,
3407 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
3408 le32_to_cpu(lcme->lcme_size));
3410 offset += le32_to_cpu(lcme->lcme_size);
3413 /* fixup layout information */
3414 lcm->lcm_size = cpu_to_le32(size);
3415 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
3416 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
3417 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NONE)
3418 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
3420 rc = lod_striping_reload(env, lo, buf, 0);
3424 lod_obj_inc_layout_gen(lo);
3425 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3427 /* transfer layout version to OST objects. */
3428 if (lo->ldo_mirror_count > 1) {
3429 struct lod_obj_stripe_cb_data data = { {0} };
3431 layout_attr->la_valid = LA_LAYOUT_VERSION;
3432 layout_attr->la_layout_version = 0;
3433 data.locd_attr = layout_attr;
3434 data.locd_declare = true;
3435 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
3436 rc = lod_obj_for_each_stripe(env, lo, th, &data);
3441 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
3442 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3450 * Split layouts, just set the LOVEA with the layout from mbuf.
3452 static int lod_declare_layout_split(const struct lu_env *env,
3453 struct dt_object *dt, const struct lu_buf *mbuf,
3456 struct lod_thread_info *info = lod_env_info(env);
3457 struct lu_attr *layout_attr = &info->lti_layout_attr;
3458 struct lod_object *lo = lod_dt_obj(dt);
3459 struct lov_comp_md_v1 *lcm = mbuf->lb_buf;
3463 rc = lod_striping_reload(env, lo, mbuf, LVF_ALL_STALE);
3467 lod_obj_inc_layout_gen(lo);
3468 /* fix on-disk layout gen */
3469 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3471 /* transfer layout version to OST objects. */
3472 if (lo->ldo_mirror_count > 1) {
3473 struct lod_obj_stripe_cb_data data = { {0} };
3475 layout_attr->la_valid = LA_LAYOUT_VERSION;
3476 layout_attr->la_layout_version = 0;
3477 data.locd_attr = layout_attr;
3478 data.locd_declare = true;
3479 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
3480 rc = lod_obj_for_each_stripe(env, lo, th, &data);
3485 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), mbuf,
3486 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3490 static int lod_layout_declare_or_purge_mirror(const struct lu_env *env,
3491 struct dt_object *dt, const struct lu_buf *buf,
3492 struct thandle *th, bool declare)
3494 struct lod_thread_info *info = lod_env_info(env);
3495 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3496 struct lod_object *lo = lod_dt_obj(dt);
3497 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3498 struct lov_comp_md_entry_v1 *entry;
3499 struct lov_mds_md_v1 *lmm;
3500 struct dt_object **sub_objs = NULL;
3501 int rc = 0, i, k, array_count = 0;
3506 * other ops (like lod_declare_destroy) could destroying sub objects
3509 mutex_lock(&lo->ldo_layout_mutex);
3512 /* prepare sub-objects array */
3513 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
3514 entry = &comp_v1->lcm_entries[i];
3516 if (!(entry->lcme_flags & LCME_FL_INIT))
3519 lmm = (struct lov_mds_md_v1 *)
3520 ((char *)comp_v1 + entry->lcme_offset);
3521 array_count += lmm->lmm_stripe_count;
3523 OBD_ALLOC_PTR_ARRAY(sub_objs, array_count);
3524 if (sub_objs == NULL) {
3525 mutex_unlock(&lo->ldo_layout_mutex);
3530 k = 0; /* sub_objs index */
3531 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
3532 struct lov_ost_data_v1 *objs;
3533 struct lu_object *o, *n;
3534 struct dt_object *dto;
3535 struct lu_device *nd;
3536 struct lov_mds_md_v3 *v3;
3540 entry = &comp_v1->lcm_entries[i];
3542 if (!(entry->lcme_flags & LCME_FL_INIT))
3545 lmm = (struct lov_mds_md_v1 *)
3546 ((char *)comp_v1 + entry->lcme_offset);
3547 v3 = (struct lov_mds_md_v3 *)lmm;
3548 if (lmm->lmm_magic == LOV_MAGIC_V3)
3549 objs = &v3->lmm_objects[0];
3551 objs = &lmm->lmm_objects[0];
3553 for (j = 0; j < lmm->lmm_stripe_count; j++) {
3554 idx = objs[j].l_ost_idx;
3555 rc = ostid_to_fid(&info->lti_fid, &objs[j].l_ost_oi,
3560 if (!fid_is_sane(&info->lti_fid)) {
3561 CERROR("%s: sub-object insane fid "DFID"\n",
3562 lod2obd(d)->obd_name,
3563 PFID(&info->lti_fid));
3564 GOTO(out, rc = -EINVAL);
3567 lod_getref(&d->lod_ost_descs);
3569 rc = validate_lod_and_idx(d, idx);
3571 lod_putref(d, &d->lod_ost_descs);
3575 nd = &OST_TGT(d, idx)->ltd_tgt->dd_lu_dev;
3576 lod_putref(d, &d->lod_ost_descs);
3578 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
3580 GOTO(out, rc = PTR_ERR(o));
3582 n = lu_object_locate(o->lo_header, nd->ld_type);
3584 lu_object_put(env, n);
3585 GOTO(out, rc = -ENOENT);
3588 dto = container_of(n, struct dt_object, do_lu);
3591 rc = lod_sub_declare_destroy(env, dto, th);
3592 dt_object_put(env, dto);
3597 * collect to-be-destroyed sub objects, the
3598 * reference would be released after actual
3604 } /* for each stripe */
3605 } /* for each component in the mirror */
3610 /* destroy the sub objects */
3611 for (; i < k; i++) {
3612 rc = lod_sub_destroy(env, sub_objs[i], th);
3615 dt_object_put(env, sub_objs[i]);
3619 * if a sub object destroy failed, we'd release sub objects
3620 * reference get from above sub_objs collection.
3623 dt_object_put(env, sub_objs[i]);
3625 OBD_FREE_PTR_ARRAY(sub_objs, array_count);
3627 mutex_unlock(&lo->ldo_layout_mutex);
3633 * Purge layouts, delete sub objects in the mirror stored in the vic_buf,
3634 * and set the LOVEA with the layout from mbuf.
3636 static int lod_declare_layout_purge(const struct lu_env *env,
3637 struct dt_object *dt, const struct lu_buf *buf,
3640 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3641 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3646 if (le32_to_cpu(comp_v1->lcm_magic) != LOV_MAGIC_COMP_V1) {
3647 CERROR("%s: invalid layout magic %#x != %#x\n",
3648 lod2obd(d)->obd_name, le32_to_cpu(comp_v1->lcm_magic),
3653 if (cpu_to_le32(LOV_MAGIC_COMP_V1) != LOV_MAGIC_COMP_V1)
3654 lustre_swab_lov_comp_md_v1(comp_v1);
3656 /* from now on, @buf contains cpu endian data */
3658 if (comp_v1->lcm_mirror_count != 0) {
3659 CERROR("%s: can only purge one mirror from "DFID"\n",
3660 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3664 /* delcare sub objects deletion in the mirror stored in @buf */
3665 rc = lod_layout_declare_or_purge_mirror(env, dt, buf, th, true);
3669 /* delete sub objects from the mirror stored in @buf */
3670 static int lod_layout_purge(const struct lu_env *env, struct dt_object *dt,
3671 const struct lu_buf *buf, struct thandle *th)
3676 rc = lod_layout_declare_or_purge_mirror(env, dt, buf, th, false);
3681 * Implementation of dt_object_operations::do_declare_xattr_set.
3683 * \see dt_object_operations::do_declare_xattr_set() in the API description
3686 * the extension to the API:
3687 * - declaring LOVEA requests striping creation
3688 * - LU_XATTR_REPLACE means layout swap
3690 static int lod_declare_xattr_set(const struct lu_env *env,
3691 struct dt_object *dt,
3692 const struct lu_buf *buf,
3693 const char *name, int fl,
3696 struct dt_object *next = dt_object_child(dt);
3697 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3702 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
3703 if ((S_ISREG(mode) || mode == 0) &&
3704 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE | LU_XATTR_SPLIT |
3706 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3707 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
3709 * this is a request to create object's striping.
3711 * allow to declare predefined striping on a new (!mode) object
3712 * which is supposed to be replay of regular file creation
3713 * (when LOV setting is declared)
3715 * LU_XATTR_REPLACE is set to indicate a layout swap
3717 if (dt_object_exists(dt)) {
3718 rc = dt_attr_get(env, next, attr);
3722 memset(attr, 0, sizeof(*attr));
3723 attr->la_valid = LA_TYPE | LA_MODE;
3724 attr->la_mode = S_IFREG;
3726 rc = lod_declare_striped_create(env, dt, attr, buf, th);
3727 } else if (fl & LU_XATTR_MERGE) {
3728 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3729 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3730 rc = lod_declare_layout_merge(env, dt, buf, th);
3731 } else if (fl & LU_XATTR_SPLIT) {
3732 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3733 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3734 rc = lod_declare_layout_split(env, dt, buf, th);
3735 } else if (fl & LU_XATTR_PURGE) {
3736 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3737 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3738 rc = lod_declare_layout_purge(env, dt, buf, th);
3739 } else if (S_ISREG(mode) &&
3740 strlen(name) >= sizeof(XATTR_LUSTRE_LOV) + 3 &&
3741 allowed_lustre_lov(name)) {
3743 * this is a request to modify object's striping.
3744 * add/set/del component(s).
3746 if (!dt_object_exists(dt))
3749 rc = lod_declare_modify_layout(env, dt, name, buf, th);
3750 } else if (S_ISDIR(mode)) {
3751 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
3752 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3753 rc = lod_replace_parent_fid(env, dt, buf, th, true);
3755 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
3762 * Apply xattr changes to the object.
3764 * Applies xattr changes to the object and the stripes if the latter exist.
3766 * \param[in] env execution environment
3767 * \param[in] dt object
3768 * \param[in] buf buffer pointing to the new value of xattr
3769 * \param[in] name name of xattr
3770 * \param[in] fl flags
3771 * \param[in] th transaction handle
3773 * \retval 0 on success
3774 * \retval negative if failed
3776 static int lod_xattr_set_internal(const struct lu_env *env,
3777 struct dt_object *dt,
3778 const struct lu_buf *buf,
3779 const char *name, int fl,
3782 struct dt_object *next = dt_object_child(dt);
3783 struct lod_object *lo = lod_dt_obj(dt);
3788 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3789 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3792 /* Note: Do not set LinkEA on sub-stripes, otherwise
3793 * it will confuse the fid2path process(see mdt_path_current()).
3794 * The linkEA between master and sub-stripes is set in
3795 * lod_xattr_set_lmv(). */
3796 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
3799 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3800 if (!lo->ldo_stripe[i])
3803 if (!dt_object_exists(lo->ldo_stripe[i]))
3806 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
3816 * Delete an extended attribute.
3818 * Deletes specified xattr from the object and the stripes if the latter exist.
3820 * \param[in] env execution environment
3821 * \param[in] dt object
3822 * \param[in] name name of xattr
3823 * \param[in] th transaction handle
3825 * \retval 0 on success
3826 * \retval negative if failed
3828 static int lod_xattr_del_internal(const struct lu_env *env,
3829 struct dt_object *dt,
3830 const char *name, struct thandle *th)
3832 struct dt_object *next = dt_object_child(dt);
3833 struct lod_object *lo = lod_dt_obj(dt);
3839 rc = lod_sub_xattr_del(env, next, name, th);
3840 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3843 if (lo->ldo_dir_stripe_count == 0)
3846 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3847 if (!lo->ldo_stripe[i])
3850 if (!dt_object_exists(lo->ldo_stripe[i]))
3853 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3862 * Set default striping on a directory.
3864 * Sets specified striping on a directory object unless it matches the default
3865 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3866 * EA. This striping will be used when regular file is being created in this
3869 * \param[in] env execution environment
3870 * \param[in] dt the striped object
3871 * \param[in] buf buffer with the striping
3872 * \param[in] name name of EA
3873 * \param[in] fl xattr flag (see OSD API description)
3874 * \param[in] th transaction handle
3876 * \retval 0 on success
3877 * \retval negative if failed
3879 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
3880 struct dt_object *dt,
3881 const struct lu_buf *buf,
3882 const char *name, int fl,
3885 struct lov_user_md_v1 *lum;
3886 struct lov_user_md_v3 *v3 = NULL;
3887 const char *pool_name = NULL;
3892 LASSERT(buf != NULL && buf->lb_buf != NULL);
3895 switch (lum->lmm_magic) {
3896 case LOV_USER_MAGIC_SPECIFIC:
3897 case LOV_USER_MAGIC_V3:
3899 if (lov_pool_is_reserved(v3->lmm_pool_name))
3900 memset(v3->lmm_pool_name, 0, sizeof(v3->lmm_pool_name));
3901 else if (v3->lmm_pool_name[0] != '\0')
3902 pool_name = v3->lmm_pool_name;
3904 case LOV_USER_MAGIC_V1:
3905 /* if { size, offset, count } = { 0, -1, 0 } and no pool
3906 * (i.e. all default values specified) then delete default
3907 * striping from dir. */
3909 "set default striping: sz %u # %u offset %d %s %s\n",
3910 (unsigned)lum->lmm_stripe_size,
3911 (unsigned)lum->lmm_stripe_count,
3912 (int)lum->lmm_stripe_offset,
3913 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
3915 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
3916 lum->lmm_stripe_count,
3917 lum->lmm_stripe_offset,
3920 case LOV_USER_MAGIC_COMP_V1:
3922 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lum;
3923 struct lov_comp_md_entry_v1 *lcme;
3926 comp_cnt = le16_to_cpu(lcm->lcm_entry_count);
3927 for (i = 0; i < comp_cnt; i++) {
3928 lcme = &lcm->lcm_entries[i];
3929 if (lcme->lcme_flags & cpu_to_le32(LCME_FL_EXTENSION)) {
3930 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
3939 CERROR("Invalid magic %x\n", lum->lmm_magic);
3944 rc = lod_xattr_del_internal(env, dt, name, th);
3948 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3954 static int lod_get_default_lov_striping(const struct lu_env *env,
3955 struct lod_object *lo,
3956 struct lod_default_striping *lds,
3957 struct dt_allocation_hint *ah);
3960 * Helper function to convert compound layout to compound layout with
3963 * Copy lcm_entries array of \a src to \a tgt. Replace lov_user_md_v1
3964 * components of \a src with lov_user_md_v3 using \a pool.
3966 * \param[in] src source layout
3967 * \param[in] pool pool to use in \a tgt
3968 * \param[out] tgt target layout
3970 static void embed_pool_to_comp_v1(const struct lov_comp_md_v1 *src,
3972 struct lov_comp_md_v1 *tgt)
3975 struct lov_user_md_v1 *lum;
3976 struct lov_user_md_v3 *lum3;
3977 struct lov_comp_md_entry_v1 *entry;
3981 entry = tgt->lcm_entries;
3983 for (i = 0; i < le16_to_cpu(src->lcm_entry_count); i++, entry++) {
3984 *entry = src->lcm_entries[i];
3985 offset = le32_to_cpu(src->lcm_entries[i].lcme_offset);
3986 entry->lcme_offset = cpu_to_le32(offset + shift);
3988 lum = (struct lov_user_md_v1 *)((char *)src + offset);
3989 lum3 = (struct lov_user_md_v3 *)((char *)tgt + offset + shift);
3990 *(struct lov_user_md_v1 *)lum3 = *lum;
3991 if (lum->lmm_pattern == cpu_to_le32(LOV_PATTERN_MDT)) {
3992 lum3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
3994 lum3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
3995 entry->lcme_size = cpu_to_le32(sizeof(*lum3));
3996 strlcpy(lum3->lmm_pool_name, pool,
3997 sizeof(lum3->lmm_pool_name));
3998 shift += sizeof(*lum3) - sizeof(*lum);
4004 * Set default striping on a directory.
4006 * Sets specified striping on a directory object unless it matches the default
4007 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
4008 * EA. This striping will be used when regular file is being created in this
4010 * If current default striping includes a pool but specifed striping
4011 * does not - retain the pool if it exists.
4013 * \param[in] env execution environment
4014 * \param[in] dt the striped object
4015 * \param[in] buf buffer with the striping
4016 * \param[in] name name of EA
4017 * \param[in] fl xattr flag (see OSD API description)
4018 * \param[in] th transaction handle
4020 * \retval 0 on success
4021 * \retval negative if failed
4023 static int lod_xattr_set_default_lov_on_dir(const struct lu_env *env,
4024 struct dt_object *dt,
4025 const struct lu_buf *buf,
4026 const char *name, int fl,
4029 struct lod_default_striping *lds = lod_lds_buf_get(env);
4030 struct lov_user_md_v1 *v1 = buf->lb_buf;
4031 char pool[LOV_MAXPOOLNAME + 1];
4037 /* get existing striping config */
4038 rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds, NULL);
4042 memset(pool, 0, sizeof(pool));
4043 if (lds->lds_def_striping_set == 1)
4044 lod_layout_get_pool(lds->lds_def_comp_entries,
4045 lds->lds_def_comp_cnt, pool,
4048 is_del = LOVEA_DELETE_VALUES(v1->lmm_stripe_size,
4049 v1->lmm_stripe_count,
4050 v1->lmm_stripe_offset,
4053 /* Retain the pool name if it is not given */
4054 if (v1->lmm_magic == LOV_USER_MAGIC_V1 && pool[0] != '\0' &&
4056 struct lod_thread_info *info = lod_env_info(env);
4057 struct lov_user_md_v3 *v3 = info->lti_ea_store;
4059 memset(v3, 0, sizeof(*v3));
4060 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4061 v3->lmm_pattern = cpu_to_le32(v1->lmm_pattern);
4062 v3->lmm_stripe_count = cpu_to_le32(v1->lmm_stripe_count);
4063 v3->lmm_stripe_offset = cpu_to_le32(v1->lmm_stripe_offset);
4064 v3->lmm_stripe_size = cpu_to_le32(v1->lmm_stripe_size);
4066 strlcpy(v3->lmm_pool_name, pool, sizeof(v3->lmm_pool_name));
4068 info->lti_buf.lb_buf = v3;
4069 info->lti_buf.lb_len = sizeof(*v3);
4070 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4072 } else if (v1->lmm_magic == LOV_USER_MAGIC_COMP_V1 &&
4073 pool[0] != '\0' && !is_del) {
4075 * try to retain the pool from default layout if the
4076 * specified component layout does not provide pool
4079 struct lod_thread_info *info = lod_env_info(env);
4080 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
4081 struct lov_comp_md_v1 *comp_v1p;
4082 struct lov_user_md_v1 *lum;
4086 struct lov_comp_md_entry_v1 *entry;
4089 entry_count = le16_to_cpu(comp_v1->lcm_entry_count);
4090 size = sizeof(*comp_v1) +
4091 entry_count * sizeof(comp_v1->lcm_entries[0]);
4092 entry = comp_v1->lcm_entries;
4093 for (i = 0; i < entry_count; i++, entry++) {
4094 offset = le32_to_cpu(entry->lcme_offset);
4095 lum = (struct lov_user_md_v1 *)((char *)comp_v1 +
4097 if (le32_to_cpu(lum->lmm_magic) != LOV_USER_MAGIC_V1)
4098 /* the i-th component includes pool info */
4100 if (lum->lmm_pattern == cpu_to_le32(LOV_PATTERN_MDT))
4101 size += sizeof(struct lov_user_md_v1);
4103 size += sizeof(struct lov_user_md_v3);
4106 if (i == entry_count) {
4108 * re-compose the layout to include the pool for
4111 if (info->lti_ea_store_size < size)
4112 rc = lod_ea_store_resize(info, size);
4115 comp_v1p = info->lti_ea_store;
4116 *comp_v1p = *comp_v1;
4117 comp_v1p->lcm_size = cpu_to_le32(size);
4118 embed_pool_to_comp_v1(comp_v1, pool, comp_v1p);
4120 info->lti_buf.lb_buf = comp_v1p;
4121 info->lti_buf.lb_len = size;
4122 rc = lod_xattr_set_lov_on_dir(env, dt,
4127 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl,
4131 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl, th);
4134 if (lds->lds_def_striping_set == 1 && lds->lds_def_comp_entries != NULL)
4135 lod_free_def_comp_entries(lds);
4141 * Set default striping on a directory object.
4143 * Sets specified striping on a directory object unless it matches the default
4144 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
4145 * EA. This striping will be used when a new directory is being created in the
4148 * \param[in] env execution environment
4149 * \param[in] dt the striped object
4150 * \param[in] buf buffer with the striping
4151 * \param[in] name name of EA
4152 * \param[in] fl xattr flag (see OSD API description)
4153 * \param[in] th transaction handle
4155 * \retval 0 on success
4156 * \retval negative if failed
4158 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
4159 struct dt_object *dt,
4160 const struct lu_buf *buf,
4161 const char *name, int fl,
4164 struct lmv_user_md_v1 *lum;
4169 LASSERT(buf != NULL && buf->lb_buf != NULL);
4173 "set default stripe_count # %u stripe_offset %d hash %u\n",
4174 le32_to_cpu(lum->lum_stripe_count),
4175 (int)le32_to_cpu(lum->lum_stripe_offset),
4176 le32_to_cpu(lum->lum_hash_type));
4178 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
4179 le32_to_cpu(lum->lum_stripe_offset)) &&
4180 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
4181 rc = lod_xattr_del_internal(env, dt, name, th);
4185 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4194 * Turn directory into a striped directory.
4196 * During replay the client sends the striping created before MDT
4197 * failure, then the layer above LOD sends this defined striping
4198 * using ->do_xattr_set(), so LOD uses this method to replay creation
4199 * of the stripes. Notice the original information for the striping
4200 * (#stripes, FIDs, etc) was transferred in declare path.
4202 * \param[in] env execution environment
4203 * \param[in] dt the striped object
4204 * \param[in] buf not used currently
4205 * \param[in] name not used currently
4206 * \param[in] fl xattr flag (see OSD API description)
4207 * \param[in] th transaction handle
4209 * \retval 0 on success
4210 * \retval negative if failed
4212 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
4213 const struct lu_buf *buf, const char *name,
4214 int fl, struct thandle *th)
4216 struct lod_object *lo = lod_dt_obj(dt);
4217 struct lod_thread_info *info = lod_env_info(env);
4218 struct lu_attr *attr = &info->lti_attr;
4219 struct dt_object_format *dof = &info->lti_format;
4220 struct lu_buf lmv_buf;
4221 struct lu_buf slave_lmv_buf;
4222 struct lmv_mds_md_v1 *lmm;
4223 struct lmv_mds_md_v1 *slave_lmm = NULL;
4224 struct dt_insert_rec *rec = &info->lti_dt_rec;
4229 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4232 /* The stripes are supposed to be allocated in declare phase,
4233 * if there are no stripes being allocated, it will skip */
4234 if (lo->ldo_dir_stripe_count == 0) {
4235 if (lo->ldo_is_foreign) {
4236 rc = lod_sub_xattr_set(env, dt_object_child(dt), buf,
4237 XATTR_NAME_LMV, fl, th);
4244 rc = dt_attr_get(env, dt_object_child(dt), attr);
4248 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME | LA_FLAGS |
4249 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
4250 dof->dof_type = DFT_DIR;
4252 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
4255 lmm = lmv_buf.lb_buf;
4257 OBD_ALLOC_PTR(slave_lmm);
4258 if (slave_lmm == NULL)
4261 lod_prep_slave_lmv_md(slave_lmm, lmm);
4262 slave_lmv_buf.lb_buf = slave_lmm;
4263 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
4265 rec->rec_type = S_IFDIR;
4266 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4267 struct dt_object *dto = lo->ldo_stripe[i];
4268 char *stripe_name = info->lti_key;
4269 struct lu_name *sname;
4270 struct linkea_data ldata = { NULL };
4271 struct lu_buf linkea_buf;
4273 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
4277 /* fail a remote stripe creation */
4278 if (i && CFS_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_CREATE))
4281 /* don't create stripe if:
4282 * 1. it's source stripe of migrating directory
4283 * 2. it's existed stripe of splitting directory
4285 if ((lod_is_migrating(lo) && i >= lo->ldo_dir_migrate_offset) ||
4286 (lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) {
4287 if (!dt_object_exists(dto))
4288 GOTO(out, rc = -EINVAL);
4290 dt_write_lock(env, dto, DT_TGT_CHILD);
4291 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
4293 dt_write_unlock(env, dto);
4297 rc = lod_sub_ref_add(env, dto, th);
4298 dt_write_unlock(env, dto);
4302 rec->rec_fid = lu_object_fid(&dto->do_lu);
4303 rc = lod_sub_insert(env, dto,
4304 (const struct dt_rec *)rec,
4305 (const struct dt_key *)dot, th);
4310 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
4311 cfs_fail_val != i) {
4312 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
4314 slave_lmm->lmv_master_mdt_index =
4317 slave_lmm->lmv_master_mdt_index =
4320 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
4321 XATTR_NAME_LMV, 0, th);
4326 /* don't insert stripe if it's existed stripe of splitting
4327 * directory (this directory is striped).
4328 * NB, plain directory will insert itself as the first
4331 if (lod_is_splitting(lo) && lo->ldo_dir_split_offset > 1 &&
4332 lo->ldo_dir_split_offset > i)
4335 rec->rec_fid = lu_object_fid(&dt->do_lu);
4336 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
4337 (const struct dt_key *)dotdot, th);
4341 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
4343 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4344 PFID(lu_object_fid(&dto->do_lu)), i + 1);
4346 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4347 PFID(lu_object_fid(&dto->do_lu)), i);
4349 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
4350 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
4351 sname, lu_object_fid(&dt->do_lu));
4355 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
4356 linkea_buf.lb_len = ldata.ld_leh->leh_len;
4357 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
4358 XATTR_NAME_LINK, 0, th);
4362 rec->rec_fid = lu_object_fid(&dto->do_lu);
4363 rc = lod_sub_insert(env, dt_object_child(dt),
4364 (const struct dt_rec *)rec,
4365 (const struct dt_key *)stripe_name, th);
4369 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
4374 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
4375 rc = lod_sub_xattr_set(env, dt_object_child(dt),
4376 &lmv_buf, XATTR_NAME_LMV, fl, th);
4378 if (slave_lmm != NULL)
4379 OBD_FREE_PTR(slave_lmm);
4385 * Helper function to declare/execute creation of a striped directory
4387 * Called in declare/create object path, prepare striping for a directory
4388 * and prepare defaults data striping for the objects to be created in
4389 * that directory. Notice the function calls "declaration" or "execution"
4390 * methods depending on \a declare param. This is a consequence of the
4391 * current approach while we don't have natural distributed transactions:
4392 * we basically execute non-local updates in the declare phase. So, the
4393 * arguments for the both phases are the same and this is the reason for
4394 * this function to exist.
4396 * \param[in] env execution environment
4397 * \param[in] dt object
4398 * \param[in] attr attributes the stripes will be created with
4399 * \param[in] lmu lmv_user_md if MDT indices are specified
4400 * \param[in] dof format of stripes (see OSD API description)
4401 * \param[in] th transaction handle
4402 * \param[in] declare where to call "declare" or "execute" methods
4404 * \retval 0 on success
4405 * \retval negative if failed
4407 static int lod_dir_striping_create_internal(const struct lu_env *env,
4408 struct dt_object *dt,
4409 struct lu_attr *attr,
4410 const struct lu_buf *lmu,
4411 struct dt_object_format *dof,
4415 struct lod_thread_info *info = lod_env_info(env);
4416 struct lod_object *lo = lod_dt_obj(dt);
4417 const struct lod_default_striping *lds = lo->ldo_def_striping;
4421 LASSERT(ergo(lds != NULL,
4422 lds->lds_def_striping_set ||
4423 lds->lds_dir_def_striping_set));
4425 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
4426 lo->ldo_dir_stripe_offset)) {
4428 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
4429 int stripe_count = lo->ldo_dir_stripe_count;
4431 if (info->lti_ea_store_size < sizeof(*v1)) {
4432 rc = lod_ea_store_resize(info, sizeof(*v1));
4435 v1 = info->lti_ea_store;
4438 memset(v1, 0, sizeof(*v1));
4439 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
4440 v1->lum_stripe_count = cpu_to_le32(stripe_count);
4441 v1->lum_stripe_offset =
4442 cpu_to_le32(lo->ldo_dir_stripe_offset);
4444 info->lti_buf.lb_buf = v1;
4445 info->lti_buf.lb_len = sizeof(*v1);
4446 lmu = &info->lti_buf;
4450 rc = lod_declare_xattr_set_lmv(env, dt, attr, lmu, dof,
4453 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
4458 /* foreign LMV EA case */
4460 struct lmv_foreign_md *lfm = lmu->lb_buf;
4462 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
4463 rc = lod_declare_xattr_set_lmv(env, dt, attr,
4467 if (lo->ldo_is_foreign) {
4468 LASSERT(lo->ldo_foreign_lmv != NULL &&
4469 lo->ldo_foreign_lmv_size > 0);
4470 info->lti_buf.lb_buf = lo->ldo_foreign_lmv;
4471 info->lti_buf.lb_len = lo->ldo_foreign_lmv_size;
4472 lmu = &info->lti_buf;
4473 rc = lod_xattr_set_lmv(env, dt, lmu,
4474 XATTR_NAME_LMV, 0, th);
4479 /* Transfer default LMV striping from the parent */
4480 if (lds != NULL && lds->lds_dir_def_striping_set &&
4481 lds->lds_dir_def_max_inherit != LMV_INHERIT_END &&
4482 lds->lds_dir_def_max_inherit != LMV_INHERIT_NONE &&
4483 !(LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
4484 lds->lds_dir_def_stripe_offset) &&
4485 le32_to_cpu(lds->lds_dir_def_hash_type) !=
4486 LMV_HASH_TYPE_UNKNOWN)) {
4487 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
4489 if (info->lti_ea_store_size < sizeof(*v1)) {
4490 rc = lod_ea_store_resize(info, sizeof(*v1));
4493 v1 = info->lti_ea_store;
4496 memset(v1, 0, sizeof(*v1));
4497 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
4498 v1->lum_stripe_count =
4499 cpu_to_le32(lds->lds_dir_def_stripe_count);
4500 v1->lum_stripe_offset =
4501 cpu_to_le32(lds->lds_dir_def_stripe_offset);
4503 cpu_to_le32(lds->lds_dir_def_hash_type);
4504 v1->lum_max_inherit =
4505 lmv_inherit_next(lds->lds_dir_def_max_inherit);
4506 v1->lum_max_inherit_rr =
4507 lmv_inherit_rr_next(lds->lds_dir_def_max_inherit_rr);
4509 info->lti_buf.lb_buf = v1;
4510 info->lti_buf.lb_len = sizeof(*v1);
4512 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4513 XATTR_NAME_DEFAULT_LMV,
4516 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
4518 XATTR_NAME_DEFAULT_LMV, 0,
4524 /* Transfer default LOV striping from the parent */
4525 if (lds != NULL && lds->lds_def_striping_set &&
4526 lds->lds_def_comp_cnt != 0) {
4527 struct lov_mds_md *lmm;
4528 int lmm_size = lod_comp_md_size(lo, true);
4530 if (info->lti_ea_store_size < lmm_size) {
4531 rc = lod_ea_store_resize(info, lmm_size);
4535 lmm = info->lti_ea_store;
4537 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
4541 info->lti_buf.lb_buf = lmm;
4542 info->lti_buf.lb_len = lmm_size;
4545 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4546 XATTR_NAME_LOV, 0, th);
4548 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4549 XATTR_NAME_LOV, 0, th);
4554 /* ldo_def_striping is not allocated, clear after use, in case directory
4555 * layout is changed later.
4558 lo->ldo_def_striping = NULL;
4563 static int lod_declare_dir_striping_create(const struct lu_env *env,
4564 struct dt_object *dt,
4565 struct lu_attr *attr,
4567 struct dt_object_format *dof,
4570 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
4574 static int lod_dir_striping_create(const struct lu_env *env,
4575 struct dt_object *dt,
4576 struct lu_attr *attr,
4577 struct dt_object_format *dof,
4580 return lod_dir_striping_create_internal(env, dt, attr, NULL, dof, th,
4585 * Make LOV EA for striped object.
4587 * Generate striping information and store it in the LOV EA of the given
4588 * object. The caller must ensure nobody else is calling the function
4589 * against the object concurrently. The transaction must be started.
4590 * FLDB service must be running as well; it's used to map FID to the target,
4591 * which is stored in LOV EA.
4593 * \param[in] env execution environment for this thread
4594 * \param[in] lo LOD object
4595 * \param[in] th transaction handle
4597 * \retval 0 if LOV EA is stored successfully
4598 * \retval negative error number on failure
4600 static int lod_generate_and_set_lovea(const struct lu_env *env,
4601 struct lod_object *lo,
4604 struct lod_thread_info *info = lod_env_info(env);
4605 struct dt_object *next = dt_object_child(&lo->ldo_obj);
4606 struct lov_mds_md_v1 *lmm;
4612 if (lo->ldo_comp_cnt == 0 && !lo->ldo_is_foreign) {
4613 lod_striping_free_nolock(env, lo);
4614 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
4618 lmm_size = lod_comp_md_size(lo, false);
4619 if (info->lti_ea_store_size < lmm_size) {
4620 rc = lod_ea_store_resize(info, lmm_size);
4624 lmm = info->lti_ea_store;
4626 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
4630 info->lti_buf.lb_buf = lmm;
4631 info->lti_buf.lb_len = lmm_size;
4632 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
4633 XATTR_NAME_LOV, 0, th);
4637 static __u32 lod_gen_component_id(struct lod_object *lo,
4638 int mirror_id, int comp_idx);
4641 * Repeat an existing component
4643 * Creates a new layout by replicating an existing component. Uses striping
4644 * policy from previous component as a template for the striping for the new
4647 * New component starts with zero length, will be extended (or removed) before
4648 * returning layout to client.
4650 * NB: Reallocates layout components array (lo->ldo_comp_entries), invalidating
4651 * any pre-existing pointers to components. Handle with care.
4653 * \param[in] env execution environment for this thread
4654 * \param[in,out] lo object to update the layout of
4655 * \param[in] index index of component to copy
4657 * \retval 0 on success
4658 * \retval negative errno on error
4660 static int lod_layout_repeat_comp(const struct lu_env *env,
4661 struct lod_object *lo, int index)
4663 struct lod_layout_component *lod_comp;
4664 struct lod_layout_component *new_comp = NULL;
4665 struct lod_layout_component *comp_array;
4666 int rc = 0, i, new_cnt = lo->ldo_comp_cnt + 1;
4671 lod_comp = &lo->ldo_comp_entries[index];
4672 LASSERT(lod_comp_inited(lod_comp) && lod_comp->llc_id != LCME_ID_INVAL);
4674 CDEBUG(D_LAYOUT, "repeating component %d\n", index);
4676 OBD_ALLOC_PTR_ARRAY(comp_array, new_cnt);
4677 if (comp_array == NULL)
4678 GOTO(out, rc = -ENOMEM);
4680 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4681 memcpy(&comp_array[i + offset], &lo->ldo_comp_entries[i],
4682 sizeof(*comp_array));
4684 /* Duplicate this component in to the next slot */
4686 new_comp = &comp_array[i + 1];
4687 memcpy(&comp_array[i + 1], &lo->ldo_comp_entries[i],
4688 sizeof(*comp_array));
4689 /* We must now skip this new component when copying */
4694 /* Set up copied component */
4695 new_comp->llc_flags &= ~LCME_FL_INIT;
4696 new_comp->llc_stripe = NULL;
4697 new_comp->llc_stripes_allocated = 0;
4698 new_comp->llc_ost_indices = NULL;
4699 new_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4700 /* for uninstantiated components, layout gen stores default stripe
4702 new_comp->llc_layout_gen = lod_comp->llc_stripe_offset;
4703 /* This makes the repeated component zero-length, placed at the end of
4704 * the preceding component */
4705 new_comp->llc_extent.e_start = new_comp->llc_extent.e_end;
4706 new_comp->llc_timestamp = lod_comp->llc_timestamp;
4707 new_comp->llc_pool = NULL;
4709 rc = lod_set_pool(&new_comp->llc_pool, lod_comp->llc_pool);
4713 if (new_comp->llc_ostlist.op_array) {
4714 __u32 *op_array = NULL;
4716 OBD_ALLOC(op_array, new_comp->llc_ostlist.op_size);
4718 GOTO(out, rc = -ENOMEM);
4719 memcpy(op_array, &new_comp->llc_ostlist.op_array,
4720 new_comp->llc_ostlist.op_size);
4721 new_comp->llc_ostlist.op_array = op_array;
4724 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
4725 lo->ldo_comp_entries = comp_array;
4726 lo->ldo_comp_cnt = new_cnt;
4728 /* Generate an id for the new component */
4729 mirror_id = mirror_id_of(new_comp->llc_id);
4730 new_comp->llc_id = LCME_ID_INVAL;
4731 new_comp->llc_id = lod_gen_component_id(lo, mirror_id, index + 1);
4732 if (new_comp->llc_id == LCME_ID_INVAL)
4733 GOTO(out, rc = -ERANGE);
4738 OBD_FREE_PTR_ARRAY(comp_array, new_cnt);
4743 static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
4747 /* clear memory region that will be used for layout change */
4748 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
4749 info->lti_count = 0;
4751 if (info->lti_comp_size >= comp_cnt)
4754 if (info->lti_comp_size > 0) {
4755 OBD_FREE_PTR_ARRAY(info->lti_comp_idx, info->lti_comp_size);
4756 info->lti_comp_size = 0;
4759 OBD_ALLOC_PTR_ARRAY(info->lti_comp_idx, comp_cnt);
4760 if (!info->lti_comp_idx)
4763 info->lti_comp_size = comp_cnt;
4768 * Prepare new layout minus deleted components
4770 * Removes components marked for deletion (LCME_ID_INVAL) by copying to a new
4771 * layout and skipping those components. Removes stripe objects if any exist.
4774 * Reallocates layout components array (lo->ldo_comp_entries), invalidating
4775 * any pre-existing pointers to components.
4777 * Caller is responsible for updating mirror end (ldo_mirror[].lme_end).
4779 * \param[in] env execution environment for this thread
4780 * \param[in,out] lo object to update the layout of
4781 * \param[in] th transaction handle for this operation
4783 * \retval # of components deleted
4784 * \retval negative errno on error
4786 static int lod_layout_del_prep_layout(const struct lu_env *env,
4787 struct lod_object *lo,
4790 struct lod_layout_component *lod_comp;
4791 struct lod_thread_info *info = lod_env_info(env);
4792 int rc = 0, i, j, deleted = 0;
4796 LASSERT(lo->ldo_is_composite);
4797 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
4799 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
4803 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4804 lod_comp = &lo->ldo_comp_entries[i];
4806 if (lod_comp->llc_id != LCME_ID_INVAL) {
4807 /* Build array of things to keep */
4808 info->lti_comp_idx[info->lti_count++] = i;
4812 lod_obj_set_pool(lo, i, NULL);
4813 if (lod_comp->llc_ostlist.op_array) {
4814 OBD_FREE(lod_comp->llc_ostlist.op_array,
4815 lod_comp->llc_ostlist.op_size);
4816 lod_comp->llc_ostlist.op_array = NULL;
4817 lod_comp->llc_ostlist.op_size = 0;
4821 CDEBUG(D_LAYOUT, "deleting comp %d, left %d\n", i,
4822 lo->ldo_comp_cnt - deleted);
4824 /* No striping info for this component */
4825 if (lod_comp->llc_stripe == NULL)
4828 LASSERT(lod_comp->llc_stripe_count > 0);
4829 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4830 struct dt_object *obj = lod_comp->llc_stripe[j];
4835 /* components which are not init have no sub objects
4837 if (lod_comp_inited(lod_comp)) {
4838 rc = lod_sub_destroy(env, obj, th);
4843 lu_object_put(env, &obj->do_lu);
4844 lod_comp->llc_stripe[j] = NULL;
4846 OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe,
4847 lod_comp->llc_stripes_allocated);
4848 lod_comp->llc_stripe = NULL;
4849 OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices,
4850 lod_comp->llc_stripes_allocated);
4851 lod_comp->llc_ost_indices = NULL;
4852 lod_comp->llc_stripes_allocated = 0;
4855 /* info->lti_count has the amount of left components */
4856 LASSERTF(info->lti_count >= 0 && info->lti_count < lo->ldo_comp_cnt,
4857 "left = %d, lo->ldo_comp_cnt %d\n", (int)info->lti_count,
4858 (int)lo->ldo_comp_cnt);
4860 if (info->lti_count > 0) {
4861 struct lod_layout_component *comp_array;
4863 OBD_ALLOC_PTR_ARRAY(comp_array, info->lti_count);
4864 if (comp_array == NULL)
4865 GOTO(out, rc = -ENOMEM);
4867 for (i = 0; i < info->lti_count; i++) {
4868 memcpy(&comp_array[i],
4869 &lo->ldo_comp_entries[info->lti_comp_idx[i]],
4870 sizeof(*comp_array));
4873 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
4874 lo->ldo_comp_entries = comp_array;
4875 lo->ldo_comp_cnt = info->lti_count;
4877 lod_free_comp_entries(lo);
4882 return rc ? rc : deleted;
4886 * Delete layout component(s)
4888 * This function sets up the layout data in the env and does the setattrs
4889 * required to write out the new layout. The layout itself is modified in
4890 * lod_layout_del_prep_layout.
4892 * \param[in] env execution environment for this thread
4893 * \param[in] dt object
4894 * \param[in] th transaction handle
4896 * \retval 0 on success
4897 * \retval negative error number on failure
4899 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
4902 struct lod_object *lo = lod_dt_obj(dt);
4903 struct dt_object *next = dt_object_child(dt);
4904 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4907 LASSERT(lo->ldo_mirror_count == 1);
4909 mutex_lock(&lo->ldo_layout_mutex);
4911 rc = lod_layout_del_prep_layout(env, lo, th);
4915 /* Only do this if we didn't delete all components */
4916 if (lo->ldo_comp_cnt > 0) {
4917 lo->ldo_mirrors[0].lme_end = lo->ldo_comp_cnt - 1;
4918 lod_obj_inc_layout_gen(lo);
4921 LASSERT(dt_object_exists(dt));
4922 rc = dt_attr_get(env, next, attr);
4926 if (attr->la_size > 0) {
4928 attr->la_valid = LA_SIZE;
4929 rc = lod_sub_attr_set(env, next, attr, th);
4934 rc = lod_generate_and_set_lovea(env, lo, th);
4938 lod_striping_free_nolock(env, lo);
4940 mutex_unlock(&lo->ldo_layout_mutex);
4947 * Implementation of dt_object_operations::do_xattr_set.
4949 * Sets specified extended attribute on the object. Three types of EAs are
4951 * LOV EA - stores striping for a regular file or default striping (when set
4953 * LMV EA - stores a marker for the striped directories
4954 * DMV EA - stores default directory striping
4956 * When striping is applied to a non-striped existing object (this is called
4957 * late striping), then LOD notices the caller wants to turn the object into a
4958 * striped one. The stripe objects are created and appropriate EA is set:
4959 * LOV EA storing all the stripes directly or LMV EA storing just a small header
4960 * with striping configuration.
4962 * \see dt_object_operations::do_xattr_set() in the API description for details.
4964 static int lod_xattr_set(const struct lu_env *env,
4965 struct dt_object *dt, const struct lu_buf *buf,
4966 const char *name, int fl, struct thandle *th)
4968 struct dt_object *next = dt_object_child(dt);
4969 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
4970 struct lod_object *lo = lod_dt_obj(dt);
4971 struct lod_obj_stripe_cb_data data = { {0} };
4976 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4977 !strcmp(name, XATTR_NAME_LMV)) {
4979 case LU_XATTR_CREATE:
4980 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
4983 case LU_XATTR_REPLACE:
4984 rc = lod_dir_layout_set(env, dt, buf, fl, th);
4991 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4992 strcmp(name, XATTR_NAME_LOV) == 0) {
4993 rc = lod_xattr_set_default_lov_on_dir(env, dt, buf, name, fl,
4996 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4997 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
4999 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
5002 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
5003 (strcmp(name, XATTR_NAME_LOV) == 0 ||
5004 strcmp(name, XATTR_LUSTRE_LOV) == 0 ||
5005 allowed_lustre_lov(name))) {
5006 /* in case of lov EA swap, just set it
5007 * if not, it is a replay so check striping match what we
5008 * already have during req replay, declare_xattr_set()
5009 * defines striping, then create() does the work */
5010 if (fl & LU_XATTR_REPLACE) {
5011 /* free stripes, then update disk */
5012 lod_striping_free(env, lod_dt_obj(dt));
5014 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
5015 } else if (fl & LU_XATTR_SPLIT) {
5016 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
5020 rc = lod_striping_reload(env, lo, buf, LVF_ALL_STALE);
5024 if (lo->ldo_mirror_count > 1 &&
5025 layout_attr->la_valid & LA_LAYOUT_VERSION) {
5027 layout_attr->la_layout_version =
5029 data.locd_attr = layout_attr;
5030 data.locd_declare = false;
5031 data.locd_stripe_cb =
5032 lod_obj_stripe_attr_set_cb;
5033 rc = lod_obj_for_each_stripe(env, lo, th,
5038 } else if (fl & LU_XATTR_PURGE) {
5039 rc = lod_layout_purge(env, dt, buf, th);
5040 } else if (dt_object_remote(dt)) {
5041 /* This only happens during migration, see
5042 * mdd_migrate_create(), in which Master MDT will
5043 * create a remote target object, and only set
5044 * (migrating) stripe EA on the remote object,
5045 * and does not need creating each stripes. */
5046 rc = lod_sub_xattr_set(env, next, buf, name,
5048 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
5049 /* delete component(s) */
5050 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
5051 rc = lod_layout_del(env, dt, th);
5054 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
5055 * it's going to create create file with specified
5056 * component(s), the striping must have not being
5057 * cached in this case;
5059 * Otherwise, it's going to add/change component(s) to
5060 * an existing file, the striping must have been cached
5063 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
5064 !strcmp(name, XATTR_NAME_LOV),
5065 !lod_dt_obj(dt)->ldo_comp_cached));
5067 rc = lod_striped_create(env, dt, NULL, NULL, th);
5071 if (fl & LU_XATTR_MERGE && lo->ldo_mirror_count > 1 &&
5072 layout_attr->la_valid & LA_LAYOUT_VERSION) {
5073 /* mirror merge exec phase */
5074 layout_attr->la_layout_version =
5076 data.locd_attr = layout_attr;
5077 data.locd_declare = false;
5078 data.locd_stripe_cb =
5079 lod_obj_stripe_attr_set_cb;
5080 rc = lod_obj_for_each_stripe(env, lo, th,
5087 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
5088 rc = lod_replace_parent_fid(env, dt, buf, th, false);
5093 /* then all other xattr */
5094 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
5100 * Implementation of dt_object_operations::do_declare_xattr_del.
5102 * \see dt_object_operations::do_declare_xattr_del() in the API description
5105 static int lod_declare_xattr_del(const struct lu_env *env,
5106 struct dt_object *dt, const char *name,
5109 struct lod_object *lo = lod_dt_obj(dt);
5110 struct dt_object *next = dt_object_child(dt);
5115 rc = lod_sub_declare_xattr_del(env, next, name, th);
5119 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
5122 /* NB: don't delete stripe LMV, because when we do this, normally we
5123 * will remove stripes, besides, if directory LMV is corrupt, this will
5124 * prevent deleting its LMV and fixing it (via LFSCK).
5126 if (!strcmp(name, XATTR_NAME_LMV))
5129 rc = lod_striping_load(env, lo);
5133 if (lo->ldo_dir_stripe_count == 0)
5136 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5137 struct dt_object *dto = lo->ldo_stripe[i];
5142 if (!dt_object_exists(dto))
5145 rc = lod_sub_declare_xattr_del(env, dto, name, th);
5154 * Implementation of dt_object_operations::do_xattr_del.
5156 * If EA storing a regular striping is being deleted, then release
5157 * all the references to the stripe objects in core.
5159 * \see dt_object_operations::do_xattr_del() in the API description for details.
5161 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
5162 const char *name, struct thandle *th)
5168 if (!strcmp(name, XATTR_NAME_LOV) || !strcmp(name, XATTR_NAME_LMV))
5169 lod_striping_free(env, lod_dt_obj(dt));
5171 rc = lod_xattr_del_internal(env, dt, name, th);
5177 * Implementation of dt_object_operations::do_xattr_list.
5179 * \see dt_object_operations::do_xattr_list() in the API description
5182 static int lod_xattr_list(const struct lu_env *env,
5183 struct dt_object *dt, const struct lu_buf *buf)
5185 return dt_xattr_list(env, dt_object_child(dt), buf);
5188 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
5190 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
5194 * Copy OST list from layout provided by user.
5196 * \param[in] lod_comp layout_component to be filled
5197 * \param[in] v3 LOV EA V3 user data
5199 * \retval 0 on success
5200 * \retval negative if failed
5202 int lod_comp_copy_ost_lists(struct lod_layout_component *lod_comp,
5203 struct lov_user_md_v3 *v3)
5209 if (v3->lmm_stripe_offset == LOV_OFFSET_DEFAULT)
5210 v3->lmm_stripe_offset = v3->lmm_objects[0].l_ost_idx;
5212 if (lod_comp->llc_ostlist.op_array) {
5213 if (lod_comp->llc_ostlist.op_size >=
5214 v3->lmm_stripe_count * sizeof(__u32)) {
5215 lod_comp->llc_ostlist.op_count =
5216 v3->lmm_stripe_count;
5219 OBD_FREE(lod_comp->llc_ostlist.op_array,
5220 lod_comp->llc_ostlist.op_size);
5223 /* copy ost list from lmm */
5224 lod_comp->llc_ostlist.op_count = v3->lmm_stripe_count;
5225 lod_comp->llc_ostlist.op_size = v3->lmm_stripe_count * sizeof(__u32);
5226 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
5227 lod_comp->llc_ostlist.op_size);
5228 if (!lod_comp->llc_ostlist.op_array)
5231 for (j = 0; j < v3->lmm_stripe_count; j++) {
5232 lod_comp->llc_ostlist.op_array[j] =
5233 v3->lmm_objects[j].l_ost_idx;
5241 * Get default striping.
5243 * \param[in] env execution environment
5244 * \param[in] lo object
5245 * \param[out] lds default striping
5247 * \retval 0 on success
5248 * \retval negative if failed
5250 static int lod_get_default_lov_striping(const struct lu_env *env,
5251 struct lod_object *lo,
5252 struct lod_default_striping *lds,
5253 struct dt_allocation_hint *dah)
5255 struct lod_thread_info *info = lod_env_info(env);
5256 struct lov_user_md_v1 *v1 = NULL;
5257 struct lov_user_md_v3 *v3 = NULL;
5258 struct lov_comp_md_v1 *lcm = NULL;
5260 int append_stripe_count = dah != NULL ? dah->dah_append_stripe_count : 0;
5261 const char *append_pool = (dah != NULL &&
5262 dah->dah_append_pool != NULL &&
5263 dah->dah_append_pool[0] != '\0') ?
5264 dah->dah_append_pool : NULL;
5265 __u16 entry_count = 1;
5266 __u16 mirror_count = 0;
5267 bool want_composite = false;
5272 lds->lds_def_striping_set = 0;
5274 rc = lod_get_lov_ea(env, lo);
5278 if (rc < (typeof(rc))sizeof(struct lov_user_md))
5281 magic = *(__u32 *)info->lti_ea_store;
5282 if (magic == __swab32(LOV_USER_MAGIC_V1)) {
5283 lustre_swab_lov_user_md_v1(info->lti_ea_store);
5284 } else if (magic == __swab32(LOV_USER_MAGIC_V3)) {
5285 lustre_swab_lov_user_md_v3(info->lti_ea_store);
5286 } else if (magic == __swab32(LOV_USER_MAGIC_SPECIFIC)) {
5287 v3 = (struct lov_user_md_v3 *)info->lti_ea_store;
5288 lustre_swab_lov_user_md_v3(v3);
5289 lustre_swab_lov_user_md_objects(v3->lmm_objects,
5290 v3->lmm_stripe_count);
5291 } else if (magic == __swab32(LOV_USER_MAGIC_COMP_V1) ||
5292 magic == __swab32(LOV_USER_MAGIC_SEL)) {
5293 lustre_swab_lov_comp_md_v1(info->lti_ea_store);
5299 case LOV_USER_MAGIC_SPECIFIC:
5300 v1 = info->lti_ea_store;
5302 case LOV_MAGIC_COMP_V1:
5304 lcm = info->lti_ea_store;
5305 entry_count = lcm->lcm_entry_count;
5306 if (entry_count == 0)
5309 mirror_count = lcm->lcm_mirror_count + 1;
5310 want_composite = true;
5316 if (append_stripe_count != 0 || append_pool != NULL) {
5319 want_composite = false;
5322 /* realloc default comp entries if necessary */
5323 rc = lod_def_striping_comp_resize(lds, entry_count);
5327 lds->lds_def_comp_cnt = entry_count;
5328 lds->lds_def_striping_is_composite = want_composite;
5329 lds->lds_def_mirror_cnt = mirror_count;
5331 for (i = 0; i < entry_count; i++) {
5332 struct lod_layout_component *llc = &lds->lds_def_comp_entries[i];
5336 * reset llc values, llc_stripes is always NULL in the
5337 * default striping template, llc_pool will be reset
5338 * later below using lod_set_pool().
5340 * XXX At this point llc_pool may point to valid (!)
5341 * kmalloced strings from previous RPCs.
5343 memset(llc, 0, offsetof(typeof(*llc), llc_pool));
5346 v1 = (struct lov_user_md *)((char *)lcm +
5347 lcm->lcm_entries[i].lcme_offset);
5349 if (want_composite) {
5350 llc->llc_extent = lcm->lcm_entries[i].lcme_extent;
5351 /* We only inherit certain flags from the layout */
5352 llc->llc_flags = lcm->lcm_entries[i].lcme_flags &
5353 LCME_TEMPLATE_FLAGS;
5357 CDEBUG(D_LAYOUT, DFID" magic = %#08x, pattern = %#x, stripe_count = %hu, stripe_size = %u, stripe_offset = %hu, append_pool = '%s', append_stripe_count = %d\n",
5358 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
5361 v1->lmm_stripe_count,
5362 v1->lmm_stripe_size,
5363 v1->lmm_stripe_offset,
5365 append_stripe_count);
5367 if (!lov_pattern_supported(v1->lmm_pattern) &&
5368 !(v1->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
5369 lod_free_def_comp_entries(lds);
5373 llc->llc_stripe_count = v1->lmm_stripe_count;
5374 llc->llc_stripe_size = v1->lmm_stripe_size;
5375 llc->llc_stripe_offset = v1->lmm_stripe_offset;
5376 llc->llc_pattern = v1->lmm_pattern;
5378 if (append_stripe_count != 0 || append_pool != NULL)
5379 llc->llc_pattern = LOV_PATTERN_RAID0;
5381 if (append_stripe_count != 0)
5382 llc->llc_stripe_count = append_stripe_count;
5385 if (append_pool != NULL) {
5387 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
5388 /* XXX: sanity check here */
5389 v3 = (struct lov_user_md_v3 *)v1;
5390 if (v3->lmm_pool_name[0] != '\0')
5391 pool = v3->lmm_pool_name;
5394 lod_set_pool(&llc->llc_pool, pool);
5396 if (append_stripe_count != 0 || append_pool != NULL) {
5397 /* Ignore specific striping for append. */
5398 } else if (v1->lmm_magic == LOV_USER_MAGIC_SPECIFIC) {
5399 v3 = (struct lov_user_md_v3 *)v1;
5400 rc = lod_comp_copy_ost_lists(llc, v3);
5403 } else if (llc->llc_ostlist.op_array &&
5404 llc->llc_ostlist.op_count) {
5405 for (j = 0; j < llc->llc_ostlist.op_count; j++)
5406 llc->llc_ostlist.op_array[j] = -1;
5407 llc->llc_ostlist.op_count = 0;
5411 lds->lds_def_striping_set = 1;
5415 static inline void lod_lum2lds(struct lod_default_striping *lds,
5416 const struct lmv_user_md *lum)
5418 lds->lds_dir_def_stripe_count = le32_to_cpu(lum->lum_stripe_count);
5419 lds->lds_dir_def_stripe_offset = le32_to_cpu(lum->lum_stripe_offset);
5420 lds->lds_dir_def_hash_type = le32_to_cpu(lum->lum_hash_type);
5421 lds->lds_dir_def_max_inherit = lum->lum_max_inherit;
5422 lds->lds_dir_def_max_inherit_rr = lum->lum_max_inherit_rr;
5423 lds->lds_dir_def_striping_set = 1;
5427 * Get default directory striping.
5429 * \param[in] env execution environment
5430 * \param[in] lo object
5431 * \param[out] lds default striping
5433 * \retval 0 on success
5434 * \retval negative if failed
5436 static int lod_get_default_lmv_striping(const struct lu_env *env,
5437 struct lod_object *lo,
5438 struct lod_default_striping *lds)
5440 struct lmv_user_md *lmu;
5443 lds->lds_dir_def_striping_set = 0;
5445 rc = lod_get_default_lmv_ea(env, lo);
5449 if (rc >= (int)sizeof(*lmu)) {
5450 struct lod_thread_info *info = lod_env_info(env);
5452 lmu = info->lti_ea_store;
5453 lod_lum2lds(lds, lmu);
5460 * Get default striping in the object.
5462 * Get object default striping and default directory striping.
5464 * \param[in] env execution environment
5465 * \param[in] lo object
5466 * \param[out] lds default striping
5468 * \retval 0 on success
5469 * \retval negative if failed
5471 static int lod_get_default_striping(const struct lu_env *env,
5472 struct lod_object *lo,
5473 struct dt_allocation_hint *ah,
5474 struct lod_default_striping *lds)
5478 rc = lod_get_default_lov_striping(env, lo, lds, NULL);
5479 if (lds->lds_def_striping_set) {
5480 struct lod_thread_info *info = lod_env_info(env);
5481 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5483 rc = lod_verify_striping(env, d, lo, &info->lti_buf, false);
5485 lds->lds_def_striping_set = 0;
5488 if (ah->dah_eadata_is_dmv) {
5489 lod_lum2lds(lds, ah->dah_eadata);
5490 } else if (ah->dah_dmv_imp_inherit) {
5491 lds->lds_dir_def_striping_set = 0;
5493 rc1 = lod_get_default_lmv_striping(env, lo, lds);
5494 if (rc == 0 && rc1 < 0)
5502 * Apply default striping on object.
5504 * If object striping pattern is not set, set to the one in default striping.
5505 * The default striping is from parent or fs.
5507 * \param[in] lo new object
5508 * \param[in] lds default striping
5509 * \param[in] mode new object's mode
5511 static void lod_striping_from_default(struct lod_object *lo,
5512 const struct lod_default_striping *lds,
5515 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5518 if (lds->lds_def_striping_set && S_ISREG(mode)) {
5519 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
5521 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
5522 lds->lds_def_comp_cnt);
5526 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
5527 if (lds->lds_def_mirror_cnt > 1)
5528 lo->ldo_flr_state = LCM_FL_RDONLY;
5530 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5531 struct lod_layout_component *obj_comp =
5532 &lo->ldo_comp_entries[i];
5533 struct lod_layout_component *def_comp =
5534 &lds->lds_def_comp_entries[i];
5537 "inherit "DFID" file layout from default: flags=%#x size=%hu nr=%u offset=%u pattern=%#x pool=%s\n",
5538 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
5539 def_comp->llc_flags,
5540 def_comp->llc_stripe_size,
5541 def_comp->llc_stripe_count,
5542 def_comp->llc_stripe_offset,
5543 def_comp->llc_pattern,
5544 def_comp->llc_pool ?: "");
5546 *obj_comp = *def_comp;
5547 if (def_comp->llc_pool != NULL) {
5548 /* pointer was copied from def_comp */
5549 obj_comp->llc_pool = NULL;
5550 lod_obj_set_pool(lo, i, def_comp->llc_pool);
5554 if (def_comp->llc_ostlist.op_array &&
5555 def_comp->llc_ostlist.op_count) {
5556 OBD_ALLOC(obj_comp->llc_ostlist.op_array,
5557 obj_comp->llc_ostlist.op_size);
5558 if (!obj_comp->llc_ostlist.op_array)
5560 memcpy(obj_comp->llc_ostlist.op_array,
5561 def_comp->llc_ostlist.op_array,
5562 obj_comp->llc_ostlist.op_size);
5563 } else if (def_comp->llc_ostlist.op_array) {
5564 obj_comp->llc_ostlist.op_array = NULL;
5568 * Don't initialize these fields for plain layout
5569 * (v1/v3) here, they are inherited in the order of
5570 * 'parent' -> 'fs default (root)' -> 'global default
5571 * values for stripe_count & stripe_size'.
5573 * see lod_ah_init().
5575 if (!lo->ldo_is_composite)
5578 lod_adjust_stripe_info(obj_comp, desc, 0);
5580 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
5581 if (lo->ldo_dir_stripe_count == 0)
5582 lo->ldo_dir_stripe_count =
5583 lds->lds_dir_def_stripe_count;
5584 if (lo->ldo_dir_stripe_offset == -1)
5585 lo->ldo_dir_stripe_offset =
5586 lds->lds_dir_def_stripe_offset;
5587 if (lo->ldo_dir_hash_type == LMV_HASH_TYPE_UNKNOWN)
5588 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
5591 "inherit "DFID" dir layout from default: count=%hu offset=%u hash_type=%x\n",
5592 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
5593 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
5594 lo->ldo_dir_hash_type);
5598 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root,
5599 const char *append_pool)
5601 struct lod_layout_component *lod_comp;
5603 if (lo->ldo_comp_cnt == 0)
5606 if (lo->ldo_is_composite)
5609 lod_comp = &lo->ldo_comp_entries[0];
5611 if (lod_comp->llc_stripe_count <= 0 ||
5612 lod_comp->llc_stripe_size <= 0)
5615 if (from_root && (lod_comp->llc_pool == NULL ||
5616 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
5619 if (append_pool && append_pool[0])
5626 * Implementation of dt_object_operations::do_ah_init.
5628 * This method is used to make a decision on the striping configuration for the
5629 * object being created. It can be taken from the \a parent object if it exists,
5630 * or filesystem's default. The resulting configuration (number of stripes,
5631 * stripe size/offset, pool name, hash_type, etc.) is stored in the object
5632 * itself and will be used by the methods like ->doo_declare_create().
5634 * \see dt_object_operations::do_ah_init() in the API description for details.
5636 static void lod_ah_init(const struct lu_env *env,
5637 struct dt_allocation_hint *ah,
5638 struct dt_object *parent,
5639 struct dt_object *child,
5642 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
5643 struct lod_thread_info *info = lod_env_info(env);
5644 struct lod_default_striping *lds = lod_lds_buf_get(env);
5645 struct dt_object *nextp = NULL;
5646 struct dt_object *nextc;
5647 struct lod_object *lp = NULL;
5648 struct lod_object *lc;
5649 struct lov_desc *desc;
5650 struct lod_layout_component *lod_comp;
5656 if (ah->dah_append_stripe_count == -1)
5657 ah->dah_append_stripe_count =
5658 d->lod_ost_descs.ltd_lov_desc.ld_tgt_count;
5660 if (likely(parent)) {
5661 nextp = dt_object_child(parent);
5662 lp = lod_dt_obj(parent);
5665 nextc = dt_object_child(child);
5666 lc = lod_dt_obj(child);
5668 LASSERT(!lod_obj_is_striped(child));
5669 /* default layout template may have been set on the regular file
5670 * when this is called from mdd_create_data() */
5671 if (S_ISREG(child_mode))
5672 lod_free_comp_entries(lc);
5674 if (!dt_object_exists(nextc))
5675 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
5677 if (S_ISDIR(child_mode)) {
5678 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
5680 /* other default values are 0 */
5681 lc->ldo_dir_stripe_offset = -1;
5683 /* no default striping configuration is needed for
5686 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5687 le32_to_cpu(lum1->lum_magic) == LMV_MAGIC_FOREIGN) {
5688 lc->ldo_is_foreign = true;
5689 /* keep stripe_count 0 and stripe_offset -1 */
5690 CDEBUG(D_INFO, "no default striping for foreign dir\n");
5694 if (likely(lp != NULL))
5695 lod_get_default_striping(env, lp, ah, lds);
5697 /* It should always honour the specified stripes */
5698 /* Note: old client (< 2.7)might also do lfs mkdir, whose EA
5699 * will have old magic. In this case, we should ignore the
5700 * stripe count and try to create dir by default stripe.
5702 if (ah->dah_eadata && ah->dah_eadata_len &&
5703 !ah->dah_eadata_is_dmv &&
5704 (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
5705 le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC)) {
5706 lc->ldo_dir_stripe_count =
5707 le32_to_cpu(lum1->lum_stripe_count);
5708 lc->ldo_dir_stripe_offset =
5709 le32_to_cpu(lum1->lum_stripe_offset);
5710 lc->ldo_dir_hash_type =
5711 le32_to_cpu(lum1->lum_hash_type);
5713 "set dirstripe: count %hu, offset %d, hash %x\n",
5714 lc->ldo_dir_stripe_count,
5715 (int)lc->ldo_dir_stripe_offset,
5716 lc->ldo_dir_hash_type);
5718 if (d->lod_mdt_descs.ltd_lmv_desc.ld_active_tgt_count &&
5719 lc->ldo_dir_stripe_count < 2 &&
5720 lum1->lum_max_inherit != LMV_INHERIT_NONE) {
5721 /* when filesystem-wide default LMV is set, dirs
5722 * will be created on MDT by space usage, but if
5723 * dir is created with "lfs mkdir -c 1 ...", its
5724 * subdirs should be kept on the same MDT. To
5725 * guarantee this, set default LMV for such dir.
5727 lds->lds_dir_def_stripe_count =
5728 le32_to_cpu(lum1->lum_stripe_count);
5729 /* if "-1" stripe offset is set, save current
5730 * MDT index in default LMV.
5732 if (le32_to_cpu(lum1->lum_stripe_offset) ==
5734 lds->lds_dir_def_stripe_offset =
5735 lod2lu_dev(d)->ld_site->ld_seq_site->ss_node_id;
5737 lds->lds_dir_def_stripe_offset =
5738 le32_to_cpu(lum1->lum_stripe_offset);
5739 lds->lds_dir_def_hash_type =
5740 le32_to_cpu(lum1->lum_hash_type);
5741 lds->lds_dir_def_max_inherit =
5742 lum1->lum_max_inherit;
5743 /* it will be decreased by 1 later in setting */
5744 if (lum1->lum_max_inherit >= LMV_INHERIT_END &&
5745 lum1->lum_max_inherit < LMV_INHERIT_MAX)
5746 lds->lds_dir_def_max_inherit++;
5747 lds->lds_dir_def_max_inherit_rr =
5748 lum1->lum_max_inherit_rr;
5749 lds->lds_dir_def_striping_set = 1;
5750 /* don't inherit LOV from ROOT */
5751 if (lds->lds_def_striping_set &&
5752 fid_is_root(lod_object_fid(lp)))
5753 lds->lds_def_striping_set = 0;
5754 lc->ldo_def_striping = lds;
5755 } else if (lds->lds_def_striping_set &&
5756 !fid_is_root(lod_object_fid(lp))) {
5757 /* don't inherit default LMV for "lfs mkdir" */
5758 lds->lds_dir_def_striping_set = 0;
5759 lc->ldo_def_striping = lds;
5762 /* inherit default striping except ROOT */
5763 if ((lds->lds_def_striping_set ||
5764 lds->lds_dir_def_striping_set) &&
5765 !fid_is_root(lod_object_fid(lp)))
5766 lc->ldo_def_striping = lds;
5768 /* transfer defaults LMV to new directory */
5769 lod_striping_from_default(lc, lds, child_mode);
5771 /* set count 0 to create normal directory */
5772 if (lc->ldo_dir_stripe_count == 1)
5773 lc->ldo_dir_stripe_count = 0;
5775 /* do not save default LMV on server */
5776 if (ah->dah_dmv_imp_inherit) {
5777 lds->lds_dir_def_striping_set = 0;
5778 if (!lds->lds_def_striping_set)
5779 lc->ldo_def_striping = NULL;
5783 /* shrink the stripe count to max_mdt_stripecount if it is -1
5784 * and max_mdt_stripecount is not 0
5786 if (lc->ldo_dir_stripe_count == (__u16)(-1) &&
5787 d->lod_max_mdt_stripecount)
5788 lc->ldo_dir_stripe_count = d->lod_max_mdt_stripecount;
5790 /* shrink the stripe_count to the avaible MDT count */
5791 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
5792 !CFS_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)) {
5793 lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
5794 if (lc->ldo_dir_stripe_count == 1)
5795 lc->ldo_dir_stripe_count = 0;
5798 if (!lmv_is_known_hash_type(lc->ldo_dir_hash_type))
5799 lc->ldo_dir_hash_type =
5800 (lc->ldo_dir_hash_type & LMV_HASH_FLAG_KNOWN) |
5801 d->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
5803 /* make sure all fscrypt metadata stays on same mdt */
5804 if (child->do_lu.lo_header->loh_attr & LOHA_FSCRYPT_MD) {
5805 lc->ldo_dir_stripe_count = 0;
5806 lds->lds_dir_def_stripe_offset =
5807 lod2lu_dev(d)->ld_site->ld_seq_site->ss_node_id;
5808 lds->lds_dir_def_striping_set = 1;
5809 lc->ldo_def_striping = lds;
5812 CDEBUG(D_INFO, "final dir stripe_count=%hu offset=%d hash=%u\n",
5813 lc->ldo_dir_stripe_count,
5814 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
5819 /* child object regular file*/
5821 if (!lod_object_will_be_striped(S_ISREG(child_mode),
5822 lu_object_fid(&child->do_lu)))
5825 /* If object is going to be striped over OSTs, transfer default
5826 * striping information to the child, so that we can use it
5827 * during declaration and creation.
5829 * Try from the parent first.
5831 if (likely(lp != NULL)) {
5832 rc = lod_get_default_lov_striping(env, lp, lds, ah);
5833 if (rc == 0 && lds->lds_def_striping_set) {
5834 rc = lod_verify_striping(env, d, lp, &info->lti_buf,
5837 lod_striping_from_default(lc, lds, child_mode);
5841 /* Initialize lod_device::lod_md_root object reference */
5842 if (d->lod_md_root == NULL) {
5843 struct dt_object *root;
5844 struct lod_object *lroot;
5846 lu_root_fid(&info->lti_fid);
5847 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
5848 if (!IS_ERR(root)) {
5849 lroot = lod_dt_obj(root);
5851 spin_lock(&d->lod_lock);
5852 if (d->lod_md_root != NULL)
5853 dt_object_put(env, &d->lod_md_root->ldo_obj);
5854 d->lod_md_root = lroot;
5855 spin_unlock(&d->lod_lock);
5859 /* try inherit layout from the root object (fs default) when:
5860 * - parent does not have default layout; or
5861 * - parent has plain(v1/v3) default layout, and some attributes
5862 * are not specified in the default layout;
5864 if (d->lod_md_root != NULL &&
5865 lod_need_inherit_more(lc, true, ah->dah_append_pool)) {
5866 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds,
5868 if (rc || !lds->lds_def_striping_set)
5871 rc = lod_verify_striping(env, d, d->lod_md_root, &info->lti_buf,
5876 if (lc->ldo_comp_cnt == 0) {
5877 lod_striping_from_default(lc, lds, child_mode);
5878 } else if (!lds->lds_def_striping_is_composite) {
5879 struct lod_layout_component *def_comp;
5881 LASSERT(!lc->ldo_is_composite);
5882 lod_comp = &lc->ldo_comp_entries[0];
5883 def_comp = &lds->lds_def_comp_entries[0];
5885 if (lod_comp->llc_stripe_count <= 0)
5886 lod_comp->llc_stripe_count =
5887 def_comp->llc_stripe_count;
5888 if (lod_comp->llc_stripe_size <= 0)
5889 lod_comp->llc_stripe_size =
5890 def_comp->llc_stripe_size;
5891 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT &&
5892 (!lod_comp->llc_pool || !lod_comp->llc_pool[0]))
5893 lod_comp->llc_stripe_offset =
5894 def_comp->llc_stripe_offset;
5895 if (lod_comp->llc_pool == NULL)
5896 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
5901 * fs default striping may not be explicitly set, or historically set
5902 * in config log, use them.
5904 if (lod_need_inherit_more(lc, false, ah->dah_append_pool)) {
5905 if (lc->ldo_comp_cnt == 0) {
5906 rc = lod_alloc_comp_entries(lc, 0, 1);
5908 /* fail to allocate memory, will create a
5909 * non-striped file. */
5911 lc->ldo_is_composite = 0;
5912 lod_comp = &lc->ldo_comp_entries[0];
5913 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
5915 LASSERT(!lc->ldo_is_composite);
5916 lod_comp = &lc->ldo_comp_entries[0];
5917 desc = &d->lod_ost_descs.ltd_lov_desc;
5918 lod_adjust_stripe_info(lod_comp, desc,
5919 ah->dah_append_stripe_count);
5920 if (ah->dah_append_pool && ah->dah_append_pool[0])
5921 lod_obj_set_pool(lc, 0, ah->dah_append_pool);
5928 * Size initialization on late striping.
5930 * Propagate the size of a truncated object to a deferred striping.
5931 * This function handles a special case when truncate was done on a
5932 * non-striped object and now while the striping is being created
5933 * we can't lose that size, so we have to propagate it to the stripes
5936 * \param[in] env execution environment
5937 * \param[in] dt object
5938 * \param[in] th transaction handle
5940 * \retval 0 on success
5941 * \retval negative if failed
5943 static int lod_declare_init_size(const struct lu_env *env,
5944 struct dt_object *dt, struct thandle *th)
5946 struct dt_object *next = dt_object_child(dt);
5947 struct lod_object *lo = lod_dt_obj(dt);
5948 struct dt_object **objects = NULL;
5949 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
5950 uint64_t size, offs;
5951 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
5952 struct lu_extent size_ext;
5955 if (!lod_obj_is_striped(dt))
5958 rc = dt_attr_get(env, next, attr);
5959 LASSERT(attr->la_valid & LA_SIZE);
5963 size = attr->la_size;
5967 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
5968 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5969 struct lod_layout_component *lod_comp;
5970 struct lu_extent *extent;
5972 lod_comp = &lo->ldo_comp_entries[i];
5974 if (lod_comp->llc_stripe == NULL)
5977 extent = &lod_comp->llc_extent;
5978 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
5979 if (!lo->ldo_is_composite ||
5980 lu_extent_is_overlapped(extent, &size_ext)) {
5981 objects = lod_comp->llc_stripe;
5982 stripe_count = lod_comp->llc_stripe_count;
5983 stripe_size = lod_comp->llc_stripe_size;
5986 if (stripe_count == 0)
5989 LASSERT(objects != NULL && stripe_size != 0);
5990 do_div(size, stripe_size);
5991 stripe = do_div(size, stripe_count);
5992 LASSERT(objects[stripe] != NULL);
5994 size = size * stripe_size;
5995 offs = attr->la_size;
5996 size += do_div(offs, stripe_size);
5998 attr->la_valid = LA_SIZE;
5999 attr->la_size = size;
6001 rc = lod_sub_declare_attr_set(env, objects[stripe],
6010 * Declare creation of striped object.
6012 * The function declares creation stripes for a regular object. The function
6013 * also declares whether the stripes will be created with non-zero size if
6014 * previously size was set non-zero on the master object. If object \a dt is
6015 * not local, then only fully defined striping can be applied in \a lovea.
6016 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
6019 * \param[in] env execution environment
6020 * \param[in] dt object
6021 * \param[in] attr attributes the stripes will be created with
6022 * \param[in] lovea a buffer containing striping description
6023 * \param[in] th transaction handle
6025 * \retval 0 on success
6026 * \retval negative if failed
6028 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
6029 struct lu_attr *attr,
6030 const struct lu_buf *lovea, struct thandle *th)
6032 struct lod_thread_info *info = lod_env_info(env);
6033 struct dt_object *next = dt_object_child(dt);
6034 struct lod_object *lo = lod_dt_obj(dt);
6038 if (CFS_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
6039 GOTO(out, rc = -ENOMEM);
6041 if (!dt_object_remote(next)) {
6042 /* choose OST and generate appropriate objects */
6043 rc = lod_prepare_create(env, lo, attr, lovea, th);
6048 * declare storage for striping data
6050 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6052 /* LOD can not choose OST objects for remote objects, i.e.
6053 * stripes must be ready before that. Right now, it can only
6054 * happen during migrate, i.e. migrate process needs to create
6055 * remote regular file (mdd_migrate_create), then the migrate
6056 * process will provide stripeEA. */
6057 LASSERT(lovea != NULL);
6058 info->lti_buf = *lovea;
6061 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
6062 XATTR_NAME_LOV, 0, th);
6067 * if striping is created with local object's size > 0,
6068 * we have to propagate this size to specific object
6069 * the case is possible only when local object was created previously
6071 if (dt_object_exists(next))
6072 rc = lod_declare_init_size(env, dt, th);
6075 /* failed to create striping or to set initial size, let's reset
6076 * config so that others don't get confused */
6078 lod_striping_free(env, lo);
6084 * Whether subdirectories under \a dt should be created on MDTs by space QoS
6086 * If LMV_HASH_FLAG_SPACE is set on directory default layout, its subdirectories
6087 * should be created on MDT by space QoS.
6089 * \param[in] env execution environment
6090 * \param[in] dev lu device
6091 * \param[in] dt object
6093 * \retval 1 if directory should create subdir by space usage
6095 * \retval -ev if failed
6097 static inline int dt_object_qos_mkdir(const struct lu_env *env,
6098 struct lu_device *dev,
6099 struct dt_object *dt)
6101 struct lod_thread_info *info = lod_env_info(env);
6102 struct lu_object *obj;
6103 struct lod_object *lo;
6104 struct lmv_user_md *lmu;
6107 obj = lu_object_find_slice(env, dev, lu_object_fid(&dt->do_lu), NULL);
6109 return PTR_ERR(obj);
6111 lo = lu2lod_obj(obj);
6113 rc = lod_get_default_lmv_ea(env, lo);
6114 dt_object_put(env, dt);
6118 if (rc < (int)sizeof(*lmu))
6121 lmu = info->lti_ea_store;
6122 return le32_to_cpu(lmu->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
6126 * Implementation of dt_object_operations::do_declare_create.
6128 * The method declares creation of a new object. If the object will be striped,
6129 * then helper functions are called to find FIDs for the stripes, declare
6130 * creation of the stripes and declare initialization of the striping
6131 * information to be stored in the master object.
6133 * \see dt_object_operations::do_declare_create() in the API description
6136 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
6137 struct lu_attr *attr,
6138 struct dt_allocation_hint *hint,
6139 struct dt_object_format *dof, struct thandle *th)
6141 struct dt_object *next = dt_object_child(dt);
6142 struct lod_object *lo = lod_dt_obj(dt);
6151 * first of all, we declare creation of local object
6153 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
6158 * it's lod_ah_init() that has decided the object will be striped
6160 if (dof->dof_type == DFT_REGULAR) {
6161 /* callers don't want stripes */
6162 /* XXX: all tricky interactions with ->ah_make_hint() decided
6163 * to use striping, then ->declare_create() behaving differently
6164 * should be cleaned */
6165 if (dof->u.dof_reg.striped != 0)
6166 rc = lod_declare_striped_create(env, dt, attr,
6168 } else if (dof->dof_type == DFT_DIR) {
6169 struct seq_server_site *ss;
6170 struct lu_buf buf = { NULL };
6171 struct lu_buf *lmu = NULL;
6173 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
6175 /* If the parent has default stripeEA, and client
6176 * did not find it before sending create request,
6177 * then MDT will return -EREMOTE, and client will
6178 * retrieve the default stripeEA and re-create the
6181 * Note: if dah_eadata != NULL, it means creating the
6182 * striped directory with specified stripeEA, then it
6183 * should ignore the default stripeEA */
6184 if (hint != NULL && hint->dah_eadata == NULL) {
6185 if (CFS_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
6186 GOTO(out, rc = -EREMOTE);
6188 if (lo->ldo_dir_stripe_offset != LMV_OFFSET_DEFAULT &&
6189 lo->ldo_dir_stripe_offset != ss->ss_node_id) {
6190 struct lod_device *lod;
6191 struct lu_tgt_desc *mdt = NULL;
6192 bool found_mdt = false;
6194 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6195 lod_foreach_mdt(lod, mdt) {
6196 if (mdt->ltd_index ==
6197 lo->ldo_dir_stripe_offset) {
6203 /* If the MDT indicated by stripe_offset can be
6204 * found, then tell client to resend the create
6205 * request to the correct MDT, otherwise return
6206 * error to client */
6208 GOTO(out, rc = -EREMOTE);
6210 GOTO(out, rc = -EINVAL);
6212 } else if (hint && hint->dah_eadata) {
6214 lmu->lb_buf = (void *)hint->dah_eadata;
6215 lmu->lb_len = hint->dah_eadata_len;
6218 rc = lod_declare_dir_striping_create(env, dt, attr, lmu, dof,
6222 /* failed to create striping or to set initial size, let's reset
6223 * config so that others don't get confused */
6225 lod_striping_free(env, lo);
6230 * Generate component ID for new created component.
6232 * \param[in] lo LOD object
6233 * \param[in] comp_idx index of ldo_comp_entries
6235 * \retval component ID on success
6236 * \retval LCME_ID_INVAL on failure
6238 static __u32 lod_gen_component_id(struct lod_object *lo,
6239 int mirror_id, int comp_idx)
6241 struct lod_layout_component *lod_comp;
6242 __u32 id, start, end;
6245 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
6247 lod_obj_inc_layout_gen(lo);
6248 id = lo->ldo_layout_gen;
6249 if (likely(id <= SEQ_ID_MAX))
6250 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
6252 /* Layout generation wraps, need to check collisions. */
6253 start = id & SEQ_ID_MASK;
6256 for (id = start; id <= end; id++) {
6257 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6258 lod_comp = &lo->ldo_comp_entries[i];
6259 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
6262 /* Found the ununsed ID */
6263 if (i == lo->ldo_comp_cnt)
6264 RETURN(pflr_id(mirror_id, id));
6266 if (end == LCME_ID_MAX) {
6268 end = min(lo->ldo_layout_gen & LCME_ID_MASK,
6269 (__u32)(LCME_ID_MAX - 1));
6273 RETURN(LCME_ID_INVAL);
6277 * Creation of a striped regular object.
6279 * The function is called to create the stripe objects for a regular
6280 * striped file. This can happen at the initial object creation or
6281 * when the caller asks LOD to do so using ->do_xattr_set() method
6282 * (so called late striping). Notice all the information are already
6283 * prepared in the form of the list of objects (ldo_stripe field).
6284 * This is done during declare phase.
6286 * \param[in] env execution environment
6287 * \param[in] dt object
6288 * \param[in] attr attributes the stripes will be created with
6289 * \param[in] dof format of stripes (see OSD API description)
6290 * \param[in] th transaction handle
6292 * \retval 0 on success
6293 * \retval negative if failed
6295 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
6296 struct lu_attr *attr, struct dt_object_format *dof,
6299 struct lod_layout_component *lod_comp;
6300 struct lod_object *lo = lod_dt_obj(dt);
6305 mutex_lock(&lo->ldo_layout_mutex);
6307 LASSERT((lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL) ||
6308 lo->ldo_is_foreign);
6310 mirror_id = 0; /* non-flr file's mirror_id is 0 */
6311 if (lo->ldo_mirror_count > 1) {
6312 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6313 lod_comp = &lo->ldo_comp_entries[i];
6314 if (lod_comp->llc_id != LCME_ID_INVAL &&
6315 mirror_id_of(lod_comp->llc_id) > mirror_id)
6316 mirror_id = mirror_id_of(lod_comp->llc_id);
6320 /* create all underlying objects */
6321 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6322 lod_comp = &lo->ldo_comp_entries[i];
6324 if (lod_comp->llc_id == LCME_ID_INVAL) {
6325 /* only the component of FLR layout with more than 1
6326 * mirror has mirror ID in its component ID.
6328 if (lod_comp->llc_extent.e_start == 0 &&
6329 lo->ldo_mirror_count > 1)
6332 lod_comp->llc_id = lod_gen_component_id(lo,
6334 if (lod_comp->llc_id == LCME_ID_INVAL)
6335 GOTO(out, rc = -ERANGE);
6338 if (lod_comp_inited(lod_comp))
6341 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
6342 lod_comp_set_init(lod_comp);
6344 if (lov_pattern(lod_comp->llc_pattern) == LOV_PATTERN_MDT)
6345 lod_comp_set_init(lod_comp);
6347 if (lod_comp->llc_stripe == NULL)
6350 LASSERT(lod_comp->llc_stripe_count);
6351 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6352 struct dt_object *object = lod_comp->llc_stripe[j];
6353 LASSERT(object != NULL);
6354 rc = lod_sub_create(env, object, attr, NULL, dof, th);
6358 lod_comp_set_init(lod_comp);
6361 rc = lod_fill_mirrors(lo);
6365 lo->ldo_comp_cached = 1;
6367 rc = lod_generate_and_set_lovea(env, lo, th);
6371 mutex_unlock(&lo->ldo_layout_mutex);
6376 lod_striping_free_nolock(env, lo);
6377 mutex_unlock(&lo->ldo_layout_mutex);
6382 static inline bool lod_obj_is_dom(struct dt_object *dt)
6384 struct lod_object *lo = lod_dt_obj(dt);
6386 if (!dt_object_exists(dt_object_child(dt)))
6389 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
6392 if (!lo->ldo_comp_cnt)
6395 return (lov_pattern(lo->ldo_comp_entries[0].llc_pattern) ==
6400 * Implementation of dt_object_operations::do_create.
6402 * If any of preceeding methods (like ->do_declare_create(),
6403 * ->do_ah_init(), etc) chose to create a striped object,
6404 * then this method will create the master and the stripes.
6406 * \see dt_object_operations::do_create() in the API description for details.
6408 static int lod_create(const struct lu_env *env, struct dt_object *dt,
6409 struct lu_attr *attr, struct dt_allocation_hint *hint,
6410 struct dt_object_format *dof, struct thandle *th)
6415 /* create local object */
6416 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
6420 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
6421 (lod_obj_is_striped(dt) || lod_obj_is_dom(dt)) &&
6422 dof->u.dof_reg.striped != 0) {
6423 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
6424 rc = lod_striped_create(env, dt, attr, dof, th);
6431 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
6432 struct dt_object *dt, struct thandle *th,
6433 int comp_idx, int stripe_idx,
6434 struct lod_obj_stripe_cb_data *data)
6436 if (data->locd_declare)
6437 return lod_sub_declare_destroy(env, dt, th);
6439 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
6440 stripe_idx == cfs_fail_val)
6441 return lod_sub_destroy(env, dt, th);
6447 * Implementation of dt_object_operations::do_declare_destroy.
6449 * If the object is a striped directory, then the function declares reference
6450 * removal from the master object (this is an index) to the stripes and declares
6451 * destroy of all the stripes. In all the cases, it declares an intention to
6452 * destroy the object itself.
6454 * \see dt_object_operations::do_declare_destroy() in the API description
6457 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
6460 struct dt_object *next = dt_object_child(dt);
6461 struct lod_object *lo = lod_dt_obj(dt);
6462 struct lod_thread_info *info = lod_env_info(env);
6463 struct dt_object *stripe;
6464 char *stripe_name = info->lti_key;
6470 * load striping information, notice we don't do this when object
6471 * is being initialized as we don't need this information till
6472 * few specific cases like destroy, chown
6474 rc = lod_striping_load(env, lo);
6478 /* declare destroy for all underlying objects */
6479 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6480 rc = next->do_ops->do_index_try(env, next,
6481 &dt_directory_features);
6485 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6486 stripe = lo->ldo_stripe[i];
6490 rc = lod_sub_declare_ref_del(env, next, th);
6494 snprintf(stripe_name, sizeof(info->lti_key),
6496 PFID(lu_object_fid(&stripe->do_lu)), i);
6497 rc = lod_sub_declare_delete(env, next,
6498 (const struct dt_key *)stripe_name, th);
6505 * we declare destroy for the local object
6507 rc = lod_sub_declare_destroy(env, next, th);
6511 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
6512 CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
6515 if (!lod_obj_is_striped(dt))
6518 /* declare destroy all striped objects */
6519 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6520 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6521 stripe = lo->ldo_stripe[i];
6525 if (!dt_object_exists(stripe))
6528 rc = lod_sub_declare_ref_del(env, stripe, th);
6532 rc = lod_sub_declare_destroy(env, stripe, th);
6537 struct lod_obj_stripe_cb_data data = { { 0 } };
6539 data.locd_declare = true;
6540 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
6541 rc = lod_obj_for_each_stripe(env, lo, th, &data);
6548 * Implementation of dt_object_operations::do_destroy.
6550 * If the object is a striped directory, then the function removes references
6551 * from the master object (this is an index) to the stripes and destroys all
6552 * the stripes. In all the cases, the function destroys the object itself.
6554 * \see dt_object_operations::do_destroy() in the API description for details.
6556 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
6559 struct dt_object *next = dt_object_child(dt);
6560 struct lod_object *lo = lod_dt_obj(dt);
6561 struct lod_thread_info *info = lod_env_info(env);
6562 char *stripe_name = info->lti_key;
6563 struct dt_object *stripe;
6569 /* destroy sub-stripe of master object */
6570 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6571 rc = next->do_ops->do_index_try(env, next,
6572 &dt_directory_features);
6576 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6577 stripe = lo->ldo_stripe[i];
6581 rc = lod_sub_ref_del(env, next, th);
6585 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
6586 PFID(lu_object_fid(&stripe->do_lu)), i);
6588 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
6589 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
6590 PFID(lu_object_fid(&stripe->do_lu)));
6592 rc = lod_sub_delete(env, next,
6593 (const struct dt_key *)stripe_name, th);
6599 rc = lod_sub_destroy(env, next, th);
6603 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
6604 CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
6607 if (!lod_obj_is_striped(dt))
6610 /* destroy all striped objects */
6611 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6612 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6613 stripe = lo->ldo_stripe[i];
6617 if (!dt_object_exists(stripe))
6620 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
6621 i == cfs_fail_val) {
6622 dt_write_lock(env, stripe, DT_TGT_CHILD);
6623 rc = lod_sub_ref_del(env, stripe, th);
6624 dt_write_unlock(env, stripe);
6628 rc = lod_sub_destroy(env, stripe, th);
6634 struct lod_obj_stripe_cb_data data = { { 0 } };
6636 data.locd_declare = false;
6637 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
6638 rc = lod_obj_for_each_stripe(env, lo, th, &data);
6645 * Implementation of dt_object_operations::do_declare_ref_add.
6647 * \see dt_object_operations::do_declare_ref_add() in the API description
6650 static int lod_declare_ref_add(const struct lu_env *env,
6651 struct dt_object *dt, struct thandle *th)
6653 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
6657 * Implementation of dt_object_operations::do_ref_add.
6659 * \see dt_object_operations::do_ref_add() in the API description for details.
6661 static int lod_ref_add(const struct lu_env *env,
6662 struct dt_object *dt, struct thandle *th)
6664 return lod_sub_ref_add(env, dt_object_child(dt), th);
6668 * Implementation of dt_object_operations::do_declare_ref_del.
6670 * \see dt_object_operations::do_declare_ref_del() in the API description
6673 static int lod_declare_ref_del(const struct lu_env *env,
6674 struct dt_object *dt, struct thandle *th)
6676 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
6680 * Implementation of dt_object_operations::do_ref_del
6682 * \see dt_object_operations::do_ref_del() in the API description for details.
6684 static int lod_ref_del(const struct lu_env *env,
6685 struct dt_object *dt, struct thandle *th)
6687 return lod_sub_ref_del(env, dt_object_child(dt), th);
6691 * Implementation of dt_object_operations::do_object_sync.
6693 * \see dt_object_operations::do_object_sync() in the API description
6696 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
6697 __u64 start, __u64 end)
6699 return dt_object_sync(env, dt_object_child(dt), start, end);
6703 * Implementation of dt_object_operations::do_object_unlock.
6705 * Used to release LDLM lock(s).
6707 * \see dt_object_operations::do_object_unlock() in the API description
6710 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
6711 struct ldlm_enqueue_info *einfo,
6712 union ldlm_policy_data *policy)
6714 struct lod_object *lo = lod_dt_obj(dt);
6715 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
6716 int slave_locks_size;
6720 if (slave_locks == NULL)
6723 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
6724 /* Note: for remote lock for single stripe dir, MDT will cancel
6725 * the lock by lockh directly */
6726 LASSERT(!dt_object_remote(dt_object_child(dt)));
6728 /* locks were unlocked in MDT layer */
6729 for (i = 0; i < slave_locks->ha_count; i++)
6730 LASSERT(!lustre_handle_is_used(&slave_locks->ha_handles[i]));
6733 * NB, ha_count may not equal to ldo_dir_stripe_count, because dir
6734 * layout may change, e.g., shrink dir layout after migration.
6736 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6737 if (lo->ldo_stripe[i])
6738 dt_invalidate(env, lo->ldo_stripe[i]);
6741 slave_locks_size = offsetof(typeof(*slave_locks),
6742 ha_handles[slave_locks->ha_count]);
6743 OBD_FREE(slave_locks, slave_locks_size);
6744 einfo->ei_cbdata = NULL;
6750 * Implementation of dt_object_operations::do_object_lock.
6752 * Used to get LDLM lock on the non-striped and striped objects.
6754 * \see dt_object_operations::do_object_lock() in the API description
6757 static int lod_object_lock(const struct lu_env *env,
6758 struct dt_object *dt,
6759 struct lustre_handle *lh,
6760 struct ldlm_enqueue_info *einfo,
6761 union ldlm_policy_data *policy)
6763 struct lod_object *lo = lod_dt_obj(dt);
6764 int slave_locks_size;
6765 struct lustre_handle_array *slave_locks = NULL;
6770 /* remote object lock */
6771 if (!einfo->ei_enq_slave) {
6772 LASSERT(dt_object_remote(dt));
6773 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
6777 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
6780 rc = lod_striping_load(env, lo);
6785 if (lo->ldo_dir_stripe_count <= 1)
6788 slave_locks_size = offsetof(typeof(*slave_locks),
6789 ha_handles[lo->ldo_dir_stripe_count]);
6790 /* Freed in lod_object_unlock */
6791 OBD_ALLOC(slave_locks, slave_locks_size);
6794 slave_locks->ha_count = lo->ldo_dir_stripe_count;
6796 /* striped directory lock */
6797 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6798 struct lustre_handle lockh;
6799 struct ldlm_res_id *res_id;
6800 struct dt_object *stripe;
6802 stripe = lo->ldo_stripe[i];
6806 res_id = &lod_env_info(env)->lti_res_id;
6807 fid_build_reg_res_name(lu_object_fid(&stripe->do_lu), res_id);
6808 einfo->ei_res_id = res_id;
6810 if (dt_object_remote(stripe)) {
6811 set_bit(i, (void *)slave_locks->ha_map);
6812 rc = dt_object_lock(env, stripe, &lockh, einfo, policy);
6814 struct ldlm_namespace *ns = einfo->ei_namespace;
6815 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
6816 ldlm_completion_callback completion = einfo->ei_cb_cp;
6817 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
6819 if (einfo->ei_mode == LCK_PW ||
6820 einfo->ei_mode == LCK_EX)
6821 dlmflags |= LDLM_FL_COS_INCOMPAT;
6823 LASSERT(ns != NULL);
6824 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS,
6825 policy, einfo->ei_mode,
6826 &dlmflags, blocking,
6828 NULL, 0, LVB_T_NONE,
6833 ldlm_lock_decref_and_cancel(
6834 &slave_locks->ha_handles[i],
6836 OBD_FREE(slave_locks, slave_locks_size);
6839 slave_locks->ha_handles[i] = lockh;
6841 einfo->ei_cbdata = slave_locks;
6847 * Implementation of dt_object_operations::do_invalidate.
6849 * \see dt_object_operations::do_invalidate() in the API description for details
6851 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
6853 return dt_invalidate(env, dt_object_child(dt));
6856 static int lod_declare_instantiate_components(const struct lu_env *env,
6857 struct lod_object *lo,
6861 struct lod_thread_info *info = lod_env_info(env);
6866 LASSERT(info->lti_count < lo->ldo_comp_cnt);
6868 for (i = 0; i < info->lti_count; i++) {
6869 rc = lod_qos_prep_create(env, lo, NULL, th,
6870 info->lti_comp_idx[i], reserve);
6876 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6877 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
6878 &info->lti_buf, XATTR_NAME_LOV, 0, th);
6885 * Check OSTs for an existing component for further extension
6887 * Checks if OSTs are still healthy and not out of space. Gets free space
6888 * on OSTs (relative to allocation watermark rmb_low) and compares to
6889 * the proposed new_end for this component.
6891 * Decides whether or not to extend a component on its current OSTs.
6893 * \param[in] env execution environment for this thread
6894 * \param[in] lo object we're checking
6895 * \param[in] index index of this component
6896 * \param[in] extension_size extension size for this component
6897 * \param[in] extent layout extent for requested operation
6898 * \param[in] comp_extent extension component extent
6899 * \param[in] write if this is write operation
6901 * \retval true - OK to extend on current OSTs
6902 * \retval false - do not extend on current OSTs
6904 static bool lod_sel_osts_allowed(const struct lu_env *env,
6905 struct lod_object *lo,
6906 int index, __u64 reserve,
6907 struct lu_extent *extent,
6908 struct lu_extent *comp_extent, int write)
6910 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[index];
6911 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6912 struct lod_thread_info *tinfo = lod_env_info(env);
6913 struct obd_statfs *sfs = &tinfo->lti_osfs;
6914 __u64 available = 0;
6920 LASSERT(lod_comp->llc_stripe_count != 0);
6922 lod_getref(&lod->lod_ost_descs);
6923 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
6924 int index = lod_comp->llc_ost_indices[i];
6925 struct lod_tgt_desc *ost = OST_TGT(lod, index);
6926 struct obd_statfs_info info = { 0 };
6927 int j, repeated = 0;
6931 /* Get the number of times this OST repeats in this component.
6932 * Note: inter-component repeats are not counted as this is
6933 * considered as a rare case: we try to not repeat OST in other
6934 * components if possible. */
6935 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6936 if (index != lod_comp->llc_ost_indices[j])
6939 /* already handled */
6945 if (j < lod_comp->llc_stripe_count)
6948 if (!test_bit(index, lod->lod_ost_bitmap)) {
6949 CDEBUG(D_LAYOUT, "ost %d no longer present\n", index);
6954 rc = dt_statfs_info(env, ost->ltd_tgt, sfs, &info);
6956 CDEBUG(D_LAYOUT, "statfs failed for ost %d, error %d\n",
6962 if (sfs->os_state & OS_STATFS_ENOSPC ||
6963 sfs->os_state & OS_STATFS_READONLY ||
6964 sfs->os_state & OS_STATFS_DEGRADED) {
6965 CDEBUG(D_LAYOUT, "ost %d is not availble for SEL "
6966 "extension, state %u\n", index, sfs->os_state);
6972 available = sfs->os_bavail * sfs->os_bsize;
6973 /* 'available' is relative to the allocation threshold */
6974 available -= (__u64) info.os_reserved_mb_low << 20;
6976 CDEBUG(D_LAYOUT, "ost %d lowwm: %d highwm: %d, "
6977 "%llu %% blocks available, %llu %% blocks free\n",
6978 index, info.os_reserved_mb_low, info.os_reserved_mb_high,
6979 (100ull * sfs->os_bavail) / sfs->os_blocks,
6980 (100ull * sfs->os_bfree) / sfs->os_blocks);
6982 if (reserve * repeated > available) {
6984 CDEBUG(D_LAYOUT, "low space on ost %d, available %llu "
6985 "< extension size %llu repeated %d\n", index,
6986 available, reserve, repeated);
6990 lod_putref(lod, &lod->lod_ost_descs);
6996 * Adjust extents after component removal
6998 * When we remove an extension component, we move the start of the next
6999 * component to match the start of the extension component, so no space is left
7002 * \param[in] env execution environment for this thread
7003 * \param[in] lo object
7004 * \param[in] max_comp layout component
7005 * \param[in] index index of this component
7007 * \retval 0 on success
7008 * \retval negative errno on error
7010 static void lod_sel_adjust_extents(const struct lu_env *env,
7011 struct lod_object *lo,
7012 int max_comp, int index)
7014 struct lod_layout_component *lod_comp = NULL;
7015 struct lod_layout_component *next = NULL;
7016 struct lod_layout_component *prev = NULL;
7017 __u64 new_start = 0;
7021 /* Extension space component */
7022 lod_comp = &lo->ldo_comp_entries[index];
7023 next = &lo->ldo_comp_entries[index + 1];
7024 prev = &lo->ldo_comp_entries[index - 1];
7026 LASSERT(lod_comp != NULL && prev != NULL && next != NULL);
7027 LASSERT(lod_comp->llc_flags & LCME_FL_EXTENSION);
7029 /* Previous is being removed */
7030 if (prev && prev->llc_id == LCME_ID_INVAL)
7031 new_start = prev->llc_extent.e_start;
7033 new_start = lod_comp->llc_extent.e_start;
7035 for (i = index + 1; i < max_comp; i++) {
7036 lod_comp = &lo->ldo_comp_entries[i];
7038 start = lod_comp->llc_extent.e_start;
7039 lod_comp->llc_extent.e_start = new_start;
7041 /* We only move zero length extendable components */
7042 if (!(start == lod_comp->llc_extent.e_end))
7045 LASSERT(!(lod_comp->llc_flags & LCME_FL_INIT));
7047 lod_comp->llc_extent.e_end = new_start;
7051 /* Calculate the proposed 'new end' for a component we're extending */
7052 static __u64 lod_extension_new_end(__u64 extension_size, __u64 extent_end,
7053 __u32 stripe_size, __u64 component_end,
7054 __u64 extension_end)
7058 LASSERT(extension_size != 0 && stripe_size != 0);
7060 /* Round up to extension size */
7061 if (extent_end == OBD_OBJECT_EOF) {
7062 new_end = OBD_OBJECT_EOF;
7064 /* Add at least extension_size to the previous component_end,
7065 * covering the req layout extent */
7066 new_end = max(extent_end - component_end, extension_size);
7067 new_end = roundup(new_end, extension_size);
7068 new_end += component_end;
7070 /* Component end must be min stripe size aligned */
7071 if (new_end % stripe_size) {
7072 CDEBUG(D_LAYOUT, "new component end is not aligned "
7073 "by the stripe size %u: [%llu, %llu) ext size "
7074 "%llu new end %llu, aligning\n",
7075 stripe_size, component_end, extent_end,
7076 extension_size, new_end);
7077 new_end = roundup(new_end, stripe_size);
7081 if (new_end < extent_end)
7082 new_end = OBD_OBJECT_EOF;
7085 /* Don't extend past the end of the extension component */
7086 if (new_end > extension_end)
7087 new_end = extension_end;
7093 * Calculate the exact reservation (per-OST extension_size) on the OSTs being
7094 * instantiated. It needs to be calculated in advance and taken into account at
7095 * the instantiation time, because otherwise lod_statfs_and_check() may consider
7096 * an OST as OK, but SEL needs its extension_size to fit the free space and the
7097 * OST may turn out to be low-on-space, thus inappropriate OST may be used and
7100 * \param[in] lod_comp lod component we are checking
7102 * \retval size to reserved on each OST of lod_comp's stripe.
7104 static __u64 lod_sel_stripe_reserved(struct lod_layout_component *lod_comp)
7106 /* extension_size is file level, so we must divide by stripe count to
7107 * compare it to available space on a single OST */
7108 return lod_comp->llc_stripe_size * SEL_UNIT_SIZE /
7109 lod_comp->llc_stripe_count;
7112 /* As lod_sel_handler() could be re-entered for the same component several
7113 * times, this is the data for the next call. Fields could be changed to
7114 * component indexes when needed, (e.g. if there is no need to instantiate
7115 * all the previous components up to the current position) to tell the caller
7116 * where to start over from. */
7123 * Process extent updates for a particular layout component
7125 * Handle layout updates for a particular extension space component touched by
7126 * a layout update operation. Core function of self-extending PFL feature.
7128 * In general, this function processes exactly *one* stage of an extension
7129 * operation, modifying the layout accordingly, then returns to the caller.
7130 * The caller is responsible for restarting processing with the new layout,
7131 * which may repeatedly return to this function until the extension updates
7134 * This function does one of a few things to the layout:
7135 * 1. Extends the component before the current extension space component to
7136 * allow it to accomodate the requested operation (if space/policy permit that
7137 * component to continue on its current OSTs)
7139 * 2. If extension of the existing component fails, we do one of two things:
7140 * a. If there is a component after the extension space, we remove the
7141 * extension space component, move the start of the next component down
7142 * accordingly, then notify the caller to restart processing w/the new
7144 * b. If there is no following component, we try repeating the current
7145 * component, creating a new component using the current one as a
7146 * template (keeping its stripe properties but not specific striping),
7147 * and try assigning striping for this component. If there is sufficient
7148 * free space on the OSTs chosen for this component, it is instantiated
7149 * and i/o continues there.
7151 * If there is not sufficient space on the new OSTs, we remove this new
7152 * component & extend the current component.
7154 * Note further that uninited components followed by extension space can be zero
7155 * length meaning that we will try to extend them before initializing them, and
7156 * if that fails, they will be removed without initialization.
7158 * 3. If we extend to/beyond the end of an extension space component, that
7159 * component is exhausted (all of its range has been given to real components),
7160 * so we remove it and restart processing.
7162 * \param[in] env execution environment for this thread
7163 * \param[in,out] lo object to update the layout of
7164 * \param[in] extent layout extent for requested operation, update
7165 * layout to fit this operation
7166 * \param[in] th transaction handle for this operation
7167 * \param[in,out] max_comp the highest comp for the portion of the layout
7168 * we are operating on (For FLR, the chosen
7169 * replica). Updated because we may remove
7171 * \param[in] index index of the extension space component we're
7173 * \param[in] write if this is write op
7174 * \param[in,out] force if the extension is to be forced; set here
7175 to force it on the 2nd call for the same
7178 * \retval 0 on success
7179 * \retval negative errno on error
7181 static int lod_sel_handler(const struct lu_env *env,
7182 struct lod_object *lo,
7183 struct lu_extent *extent,
7184 struct thandle *th, int *max_comp,
7185 int index, int write,
7186 struct sel_data *sd)
7188 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7189 struct lod_thread_info *info = lod_env_info(env);
7190 struct lod_layout_component *lod_comp;
7191 struct lod_layout_component *prev;
7192 struct lod_layout_component *next = NULL;
7193 __u64 extension_size, reserve;
7200 /* First component cannot be extension space */
7202 CERROR("%s: "DFID" first component cannot be extension space\n",
7203 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
7207 lod_comp = &lo->ldo_comp_entries[index];
7208 prev = &lo->ldo_comp_entries[index - 1];
7209 if ((index + 1) < *max_comp)
7210 next = &lo->ldo_comp_entries[index + 1];
7212 /* extension size uses the stripe size field as KiB */
7213 extension_size = lod_comp->llc_stripe_size * SEL_UNIT_SIZE;
7215 CDEBUG(D_LAYOUT, "prev start %llu, extension start %llu, extension end"
7216 " %llu, extension size %llu\n", prev->llc_extent.e_start,
7217 lod_comp->llc_extent.e_start, lod_comp->llc_extent.e_end,
7220 /* Two extension space components cannot be adjacent & extension space
7221 * components cannot be init */
7222 if ((prev->llc_flags & LCME_FL_EXTENSION) ||
7223 !(ergo(next, !(next->llc_flags & LCME_FL_EXTENSION))) ||
7224 lod_comp_inited(lod_comp)) {
7225 CERROR("%s: "DFID" invalid extension space components\n",
7226 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
7230 reserve = lod_sel_stripe_reserved(lod_comp);
7232 if (!prev->llc_stripe) {
7233 CDEBUG(D_LAYOUT, "Previous component not inited\n");
7234 info->lti_count = 1;
7235 info->lti_comp_idx[0] = index - 1;
7236 rc = lod_declare_instantiate_components(env, lo, th, reserve);
7237 /* ENOSPC tells us we can't use this component. If there is
7238 * a next or we are repeating, we either spill over (next) or
7239 * extend the original comp (repeat). Otherwise, return the
7240 * error to the user. */
7241 if (rc == -ENOSPC && (next || sd->sd_repeat))
7247 if (sd->sd_force == 0 && rc == 0)
7248 rc = !lod_sel_osts_allowed(env, lo, index - 1, reserve, extent,
7249 &lod_comp->llc_extent, write);
7251 repeated = !!(sd->sd_repeat);
7255 /* Extend previous component */
7257 new_end = lod_extension_new_end(extension_size, extent->e_end,
7258 prev->llc_stripe_size,
7259 prev->llc_extent.e_end,
7260 lod_comp->llc_extent.e_end);
7262 CDEBUG(D_LAYOUT, "new end %llu\n", new_end);
7263 lod_comp->llc_extent.e_start = new_end;
7264 prev->llc_extent.e_end = new_end;
7266 if (prev->llc_extent.e_end == lod_comp->llc_extent.e_end) {
7267 CDEBUG(D_LAYOUT, "Extension component exhausted\n");
7268 lod_comp->llc_id = LCME_ID_INVAL;
7272 /* rc == 1, failed to extend current component */
7275 /* Normal 'spillover' case - Remove the extension
7276 * space component & bring down the start of the next
7278 lod_comp->llc_id = LCME_ID_INVAL;
7280 if (!(prev->llc_flags & LCME_FL_INIT)) {
7281 prev->llc_id = LCME_ID_INVAL;
7284 lod_sel_adjust_extents(env, lo, *max_comp, index);
7285 } else if (lod_comp_inited(prev)) {
7286 /* If there is no next, and the previous component is
7287 * INIT'ed, try repeating the previous component. */
7288 LASSERT(repeated == 0);
7289 rc = lod_layout_repeat_comp(env, lo, index - 1);
7293 /* The previous component is a repeated component.
7294 * Record this so we don't keep trying to repeat it. */
7297 /* If the previous component is not INIT'ed, this may
7298 * be a component we have just instantiated but failed
7299 * to extend. Or even a repeated component we failed
7300 * to prepare a striping for. Do not repeat but instead
7301 * remove the repeated component & force the extention
7302 * of the original one */
7305 prev->llc_id = LCME_ID_INVAL;
7312 rc = lod_layout_del_prep_layout(env, lo, NULL);
7315 LASSERTF(-rc == change,
7316 "number deleted %d != requested %d\n", -rc,
7319 *max_comp = *max_comp + change;
7321 /* lod_del_prep_layout reallocates ldo_comp_entries, so we must
7322 * refresh these pointers before using them */
7323 lod_comp = &lo->ldo_comp_entries[index];
7324 prev = &lo->ldo_comp_entries[index - 1];
7325 CDEBUG(D_LAYOUT, "After extent updates: prev start %llu, current start "
7326 "%llu, current end %llu max_comp %d ldo_comp_cnt %d\n",
7327 prev->llc_extent.e_start, lod_comp->llc_extent.e_start,
7328 lod_comp->llc_extent.e_end, *max_comp, lo->ldo_comp_cnt);
7330 /* Layout changed successfully */
7335 * Declare layout extent updates
7337 * Handles extensions. Identifies extension components touched by current
7338 * operation and passes them to processing function.
7340 * Restarts with updated layouts from the processing function until the current
7341 * operation no longer touches an extension space component.
7343 * \param[in] env execution environment for this thread
7344 * \param[in,out] lo object to update the layout of
7345 * \param[in] extent layout extent for requested operation, update layout to
7346 * fit this operation
7347 * \param[in] th transaction handle for this operation
7348 * \param[in] pick identifies chosen mirror for FLR layouts
7349 * \param[in] write if this is write op
7351 * \retval 1 on layout changed, 0 on no change
7352 * \retval negative errno on error
7354 static int lod_declare_update_extents(const struct lu_env *env,
7355 struct lod_object *lo, struct lu_extent *extent,
7356 struct thandle *th, int pick, int write)
7358 struct lod_thread_info *info = lod_env_info(env);
7359 struct lod_layout_component *lod_comp;
7360 bool layout_changed = false;
7361 struct sel_data sd = { 0 };
7369 /* This makes us work on the components of the chosen mirror */
7370 start_index = lo->ldo_mirrors[pick].lme_start;
7371 max_comp = lo->ldo_mirrors[pick].lme_end + 1;
7372 if (lo->ldo_flr_state == LCM_FL_NONE)
7373 LASSERT(start_index == 0 && max_comp == lo->ldo_comp_cnt);
7375 CDEBUG(D_LAYOUT, "extent->e_start %llu, extent->e_end %llu\n",
7376 extent->e_start, extent->e_end);
7377 for (i = start_index; i < max_comp; i++) {
7378 lod_comp = &lo->ldo_comp_entries[i];
7380 /* We've passed all components of interest */
7381 if (lod_comp->llc_extent.e_start >= extent->e_end)
7384 if (lod_comp->llc_flags & LCME_FL_EXTENSION) {
7385 layout_changed = true;
7386 rc = lod_sel_handler(env, lo, extent, th, &max_comp,
7391 /* Nothing has changed behind the prev one */
7397 /* We may have added or removed components. If so, we must update the
7398 * start & ends of all the mirrors after the current one, and the end
7399 * of the current mirror. */
7400 change = max_comp - 1 - lo->ldo_mirrors[pick].lme_end;
7402 lo->ldo_mirrors[pick].lme_end += change;
7403 for (i = pick + 1; i < lo->ldo_mirror_count; i++) {
7404 lo->ldo_mirrors[i].lme_start += change;
7405 lo->ldo_mirrors[i].lme_end += change;
7411 /* The amount of components has changed, adjust the lti_comp_idx */
7412 rc2 = lod_layout_data_init(info, lo->ldo_comp_cnt);
7414 return rc < 0 ? rc : rc2 < 0 ? rc2 : layout_changed;
7417 /* If striping is already instantiated or INIT'ed DOM? */
7418 static bool lod_is_instantiation_needed(struct lod_layout_component *comp)
7420 return !(((lov_pattern(comp->llc_pattern) == LOV_PATTERN_MDT) &&
7421 lod_comp_inited(comp)) || comp->llc_stripe);
7425 * Declare layout update for a non-FLR layout.
7427 * \param[in] env execution environment for this thread
7428 * \param[in,out] lo object to update the layout of
7429 * \param[in] layout layout intent for requested operation, "update" is
7430 * a process of reacting to this
7431 * \param[in] buf buffer containing lov ea (see comment on usage inline)
7432 * \param[in] th transaction handle for this operation
7434 * \retval 0 on success
7435 * \retval negative errno on error
7437 static int lod_declare_update_plain(const struct lu_env *env,
7438 struct lod_object *lo, struct layout_intent *layout,
7439 const struct lu_buf *buf, struct thandle *th)
7441 struct lod_thread_info *info = lod_env_info(env);
7442 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7443 struct lod_layout_component *lod_comp;
7444 struct lov_comp_md_v1 *comp_v1 = NULL;
7445 bool layout_changed = false;
7446 bool replay = false;
7450 LASSERT(lo->ldo_flr_state == LCM_FL_NONE);
7453 * In case the client is passing lovea, which only happens during
7454 * the replay of layout intent write RPC for now, we may need to
7455 * parse the lovea and apply new layout configuration.
7457 if (buf && buf->lb_len) {
7458 struct lov_user_md_v1 *v1 = buf->lb_buf;
7460 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
7461 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
7462 LOV_MAGIC_COMP_V1)) {
7463 CERROR("%s: the replay buffer of layout extend "
7464 "(magic %#x) does not contain expected "
7465 "composite layout.\n",
7466 lod2obd(d)->obd_name, v1->lmm_magic);
7467 GOTO(out, rc = -EINVAL);
7470 rc = lod_use_defined_striping(env, lo, buf);
7473 lo->ldo_comp_cached = 1;
7475 rc = lod_get_lov_ea(env, lo);
7478 /* old on-disk EA is stored in info->lti_buf */
7479 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
7481 layout_changed = true;
7483 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
7487 /* non replay path */
7488 rc = lod_striping_load(env, lo);
7493 /* Make sure defined layout covers the requested write range. */
7494 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
7495 if (lo->ldo_comp_cnt > 1 &&
7496 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
7497 lod_comp->llc_extent.e_end < layout->li_extent.e_end) {
7498 CDEBUG_LIMIT(replay ? D_ERROR : D_LAYOUT,
7499 "%s: the defined layout [0, %#llx) does not "
7500 "covers the write range "DEXT"\n",
7501 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
7502 PEXT(&layout->li_extent));
7503 GOTO(out, rc = -EINVAL);
7506 CDEBUG(D_LAYOUT, "%s: "DFID": update components "DEXT"\n",
7507 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
7508 PEXT(&layout->li_extent));
7511 rc = lod_declare_update_extents(env, lo, &layout->li_extent,
7512 th, 0, layout->li_opc == LAYOUT_INTENT_WRITE);
7516 layout_changed = true;
7520 * Iterate ld->ldo_comp_entries, find the component whose extent under
7521 * the write range and not instantianted.
7523 for (i = 0; i < lo->ldo_comp_cnt; i++) {
7524 lod_comp = &lo->ldo_comp_entries[i];
7526 if (lod_comp->llc_extent.e_start >= layout->li_extent.e_end)
7530 /* If striping is instantiated or INIT'ed DOM skip */
7531 if (!lod_is_instantiation_needed(lod_comp))
7535 * In replay path, lod_comp is the EA passed by
7536 * client replay buffer, comp_v1 is the pre-recovery
7537 * on-disk EA, we'd sift out those components which
7538 * were init-ed in the on-disk EA.
7540 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
7545 * this component hasn't instantiated in normal path, or during
7546 * replay it needs replay the instantiation.
7549 /* A released component is being extended */
7550 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
7551 GOTO(out, rc = -EINVAL);
7553 LASSERT(info->lti_comp_idx != NULL);
7554 info->lti_comp_idx[info->lti_count++] = i;
7555 layout_changed = true;
7558 if (!layout_changed)
7561 lod_obj_inc_layout_gen(lo);
7562 rc = lod_declare_instantiate_components(env, lo, th, 0);
7566 lod_striping_free(env, lo);
7570 static inline int lod_comp_index(struct lod_object *lo,
7571 struct lod_layout_component *lod_comp)
7573 LASSERT(lod_comp >= lo->ldo_comp_entries &&
7574 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
7576 return lod_comp - lo->ldo_comp_entries;
7580 * Stale other mirrors by writing extent.
7582 static int lod_stale_components(const struct lu_env *env, struct lod_object *lo,
7583 int primary, struct lu_extent *extent,
7586 struct lod_layout_component *pri_comp, *lod_comp;
7587 struct lu_extent pri_extent;
7592 /* The writing extent decides which components in the primary
7593 * are affected... */
7594 CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
7597 lod_foreach_mirror_comp(pri_comp, lo, primary) {
7598 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
7601 CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
7602 lod_comp_index(lo, pri_comp),
7603 PEXT(&pri_comp->llc_extent));
7605 pri_extent.e_start = pri_comp->llc_extent.e_start;
7606 pri_extent.e_end = pri_comp->llc_extent.e_end;
7608 for (i = 0; i < lo->ldo_mirror_count; i++) {
7611 rc = lod_declare_update_extents(env, lo, &pri_extent,
7613 /* if update_extents changed the layout, it may have
7614 * reallocated the component array, so start over to
7615 * avoid using stale pointers */
7621 /* ... and then stale other components that are
7622 * overlapping with primary components */
7623 lod_foreach_mirror_comp(lod_comp, lo, i) {
7624 if (!lu_extent_is_overlapped(
7626 &lod_comp->llc_extent))
7629 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
7630 i, lod_comp_index(lo, lod_comp));
7632 lod_comp->llc_flags |= LCME_FL_STALE;
7633 lo->ldo_mirrors[i].lme_stale = 1;
7642 * check an OST's availability
7643 * \param[in] env execution environment
7644 * \param[in] lo lod object
7645 * \param[in] dt dt object
7646 * \param[in] index mirror index
7648 * \retval negative if failed
7649 * \retval 1 if \a dt is available
7650 * \retval 0 if \a dt is not available
7652 static inline int lod_check_ost_avail(const struct lu_env *env,
7653 struct lod_object *lo,
7654 struct dt_object *dt, int index)
7656 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7657 struct lod_tgt_desc *ost;
7659 int type = LU_SEQ_RANGE_OST;
7662 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &idx, &type);
7664 CERROR("%s: can't locate "DFID":rc = %d\n",
7665 lod2obd(lod)->obd_name, PFID(lu_object_fid(&dt->do_lu)),
7670 ost = OST_TGT(lod, idx);
7671 if (ost->ltd_statfs.os_state &
7672 (OS_STATFS_READONLY | OS_STATFS_ENOSPC | OS_STATFS_ENOINO |
7673 OS_STATFS_NOPRECREATE) ||
7674 ost->ltd_active == 0) {
7675 CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail, rc = %d\n",
7676 PFID(lod_object_fid(lo)), index, idx, rc);
7684 * Pick primary mirror for write
7685 * \param[in] env execution environment
7686 * \param[in] lo object
7687 * \param[in] extent write range
7689 static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo,
7690 struct lu_extent *extent)
7692 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7693 unsigned int seq = 0;
7694 struct lod_layout_component *lod_comp;
7696 int picked = -1, second_pick = -1, third_pick = -1;
7699 if (CFS_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
7700 get_random_bytes(&seq, sizeof(seq));
7701 seq %= lo->ldo_mirror_count;
7705 * Pick a mirror as the primary, and check the availability of OSTs.
7707 * This algo can be revised later after knowing the topology of
7710 lod_qos_statfs_update(env, lod, &lod->lod_ost_descs);
7712 rc = lod_fill_mirrors(lo);
7716 for (i = 0; i < lo->ldo_mirror_count; i++) {
7717 bool ost_avail = true;
7718 int index = (i + seq) % lo->ldo_mirror_count;
7720 if (lo->ldo_mirrors[index].lme_stale) {
7721 CDEBUG(D_LAYOUT, DFID": mirror %d stale\n",
7722 PFID(lod_object_fid(lo)), index);
7726 /* 2nd pick is for the primary mirror containing unavail OST */
7727 if (lo->ldo_mirrors[index].lme_prefer && second_pick < 0)
7728 second_pick = index;
7730 /* 3rd pick is for non-primary mirror containing unavail OST */
7731 if (second_pick < 0 && third_pick < 0)
7735 * we found a non-primary 1st pick, we'd like to find a
7736 * potential pirmary mirror.
7738 if (picked >= 0 && !lo->ldo_mirrors[index].lme_prefer)
7741 /* check the availability of OSTs */
7742 lod_foreach_mirror_comp(lod_comp, lo, index) {
7743 if (!lod_comp_inited(lod_comp) || !lod_comp->llc_stripe)
7746 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
7747 struct dt_object *dt = lod_comp->llc_stripe[j];
7749 rc = lod_check_ost_avail(env, lo, dt, index);
7756 } /* for all dt object in one component */
7759 } /* for all components in a mirror */
7762 * the OSTs where allocated objects locates in the components
7763 * of the mirror are available.
7768 /* this mirror has all OSTs available */
7772 * primary with all OSTs are available, this is the perfect
7775 if (lo->ldo_mirrors[index].lme_prefer)
7777 } /* for all mirrors */
7779 /* failed to pick a sound mirror, lower our expectation */
7781 picked = second_pick;
7783 picked = third_pick;
7790 static int lod_prepare_resync_mirror(const struct lu_env *env,
7791 struct lod_object *lo,
7794 struct lod_thread_info *info = lod_env_info(env);
7795 struct lod_layout_component *lod_comp;
7796 bool neg = !!(MIRROR_ID_NEG & mirror_id);
7799 mirror_id &= ~MIRROR_ID_NEG;
7801 for (i = 0; i < lo->ldo_mirror_count; i++) {
7802 if ((!neg && lo->ldo_mirrors[i].lme_id != mirror_id) ||
7803 (neg && lo->ldo_mirrors[i].lme_id == mirror_id))
7806 lod_foreach_mirror_comp(lod_comp, lo, i) {
7807 if (lod_comp_inited(lod_comp))
7810 info->lti_comp_idx[info->lti_count++] =
7811 lod_comp_index(lo, lod_comp);
7819 * figure out the components should be instantiated for resync.
7821 static int lod_prepare_resync(const struct lu_env *env, struct lod_object *lo,
7822 struct lu_extent *extent)
7824 struct lod_thread_info *info = lod_env_info(env);
7825 struct lod_layout_component *lod_comp;
7826 unsigned int need_sync = 0;
7830 DFID": instantiate all stale components in "DEXT"\n",
7831 PFID(lod_object_fid(lo)), PEXT(extent));
7834 * instantiate all components within this extent, even non-stale
7837 for (i = 0; i < lo->ldo_mirror_count; i++) {
7838 if (!lo->ldo_mirrors[i].lme_stale)
7841 lod_foreach_mirror_comp(lod_comp, lo, i) {
7842 if (!lu_extent_is_overlapped(extent,
7843 &lod_comp->llc_extent))
7848 if (lod_comp_inited(lod_comp))
7851 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
7852 i, lod_comp_index(lo, lod_comp));
7853 info->lti_comp_idx[info->lti_count++] =
7854 lod_comp_index(lo, lod_comp);
7858 return need_sync ? 0 : -EALREADY;
7861 static int lod_declare_update_rdonly(const struct lu_env *env,
7862 struct lod_object *lo, struct md_layout_change *mlc,
7865 struct lod_thread_info *info = lod_env_info(env);
7866 struct lu_attr *layout_attr = &info->lti_layout_attr;
7867 struct lod_layout_component *lod_comp;
7868 struct lu_extent extent = { 0 };
7872 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
7873 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
7874 mlc->mlc_opc == MD_LAYOUT_RESYNC);
7875 LASSERT(lo->ldo_mirror_count > 0);
7877 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7878 struct layout_intent *layout = mlc->mlc_intent;
7879 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
7882 extent = layout->li_extent;
7883 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
7884 PFID(lod_object_fid(lo)), PEXT(&extent));
7886 picked = lod_primary_pick(env, lo, &extent);
7890 CDEBUG(D_LAYOUT, DFID": picked mirror id %u as primary\n",
7891 PFID(lod_object_fid(lo)),
7892 lo->ldo_mirrors[picked].lme_id);
7894 /* Update extents of primary before staling */
7895 rc = lod_declare_update_extents(env, lo, &extent, th, picked,
7900 if (layout->li_opc == LAYOUT_INTENT_TRUNC) {
7902 * trunc transfers [0, size) in the intent extent, we'd
7903 * stale components overlapping [size, eof).
7905 extent.e_start = extent.e_end;
7906 extent.e_end = OBD_OBJECT_EOF;
7909 /* stale overlapping components from other mirrors */
7910 rc = lod_stale_components(env, lo, picked, &extent, th);
7914 /* restore truncate intent extent */
7915 if (layout->li_opc == LAYOUT_INTENT_TRUNC)
7916 extent.e_end = extent.e_start;
7918 /* instantiate components for the picked mirror, start from 0 */
7921 lod_foreach_mirror_comp(lod_comp, lo, picked) {
7922 if (!lu_extent_is_overlapped(&extent,
7923 &lod_comp->llc_extent))
7926 if (!lod_is_instantiation_needed(lod_comp))
7929 info->lti_comp_idx[info->lti_count++] =
7930 lod_comp_index(lo, lod_comp);
7933 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
7934 } else { /* MD_LAYOUT_RESYNC */
7938 * could contain multiple non-stale mirrors, so we need to
7939 * prep uninited all components assuming any non-stale mirror
7940 * could be picked as the primary mirror.
7942 if (mlc->mlc_mirror_id == 0) {
7944 for (i = 0; i < lo->ldo_mirror_count; i++) {
7945 if (lo->ldo_mirrors[i].lme_stale)
7948 lod_foreach_mirror_comp(lod_comp, lo, i) {
7949 if (!lod_comp_inited(lod_comp))
7953 lod_comp->llc_extent.e_end)
7955 lod_comp->llc_extent.e_end;
7958 rc = lod_prepare_resync(env, lo, &extent);
7962 /* mirror write, try to init its all components */
7963 rc = lod_prepare_resync_mirror(env, lo,
7964 mlc->mlc_mirror_id);
7969 /* change the file state to SYNC_PENDING */
7970 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
7973 /* Reset the layout version once it's becoming too large.
7974 * This way it can make sure that the layout version is
7975 * monotonously increased in this writing era. */
7976 lod_obj_inc_layout_gen(lo);
7978 rc = lod_declare_instantiate_components(env, lo, th, 0);
7982 layout_attr->la_valid = LA_LAYOUT_VERSION;
7983 layout_attr->la_layout_version = 0;
7984 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
7985 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
7986 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
7992 lod_striping_free(env, lo);
7996 static int lod_declare_update_write_pending(const struct lu_env *env,
7997 struct lod_object *lo, struct md_layout_change *mlc,
8000 struct lod_thread_info *info = lod_env_info(env);
8001 struct lu_attr *layout_attr = &info->lti_layout_attr;
8002 struct lod_layout_component *lod_comp;
8003 struct lu_extent extent = { 0 };
8009 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
8010 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
8011 mlc->mlc_opc == MD_LAYOUT_RESYNC);
8013 /* look for the first preferred mirror */
8014 for (i = 0; i < lo->ldo_mirror_count; i++) {
8015 if (lo->ldo_mirrors[i].lme_stale)
8017 if (lo->ldo_mirrors[i].lme_prefer == 0)
8024 /* no primary, use any in-sync */
8025 for (i = 0; i < lo->ldo_mirror_count; i++) {
8026 if (lo->ldo_mirrors[i].lme_stale)
8032 CERROR(DFID ": doesn't have a primary mirror\n",
8033 PFID(lod_object_fid(lo)));
8034 GOTO(out, rc = -ENODATA);
8038 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
8039 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
8041 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
8043 /* for LAYOUT_WRITE opc, it has to do the following operations:
8044 * 1. stale overlapping componets from stale mirrors;
8045 * 2. instantiate components of the primary mirror;
8046 * 3. transfter layout version to all objects of the primary;
8048 * for LAYOUT_RESYNC opc, it will do:
8049 * 1. instantiate components of all stale mirrors;
8050 * 2. transfer layout version to all objects to close write era. */
8052 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
8053 struct layout_intent *layout = mlc->mlc_intent;
8054 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
8056 LASSERT(mlc->mlc_intent != NULL);
8058 extent = mlc->mlc_intent->li_extent;
8060 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
8061 PFID(lod_object_fid(lo)), PEXT(&extent));
8063 /* 1. Update extents of primary before staling */
8064 rc = lod_declare_update_extents(env, lo, &extent, th, primary,
8069 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC) {
8071 * trunc transfers [0, size) in the intent extent, we'd
8072 * stale components overlapping [size, eof).
8074 extent.e_start = extent.e_end;
8075 extent.e_end = OBD_OBJECT_EOF;
8078 /* 2. stale overlapping components */
8079 rc = lod_stale_components(env, lo, primary, &extent, th);
8083 /* 3. find the components which need instantiating.
8084 * instantiate [0, mlc->mlc_intent->e_end) */
8086 /* restore truncate intent extent */
8087 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC)
8088 extent.e_end = extent.e_start;
8091 lod_foreach_mirror_comp(lod_comp, lo, primary) {
8092 if (!lu_extent_is_overlapped(&extent,
8093 &lod_comp->llc_extent))
8096 if (!lod_is_instantiation_needed(lod_comp))
8099 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
8100 primary, lod_comp_index(lo, lod_comp));
8101 info->lti_comp_idx[info->lti_count++] =
8102 lod_comp_index(lo, lod_comp);
8104 } else { /* MD_LAYOUT_RESYNC */
8105 if (mlc->mlc_mirror_id == 0) {
8107 lod_foreach_mirror_comp(lod_comp, lo, primary) {
8108 if (!lod_comp_inited(lod_comp))
8111 extent.e_end = lod_comp->llc_extent.e_end;
8114 rc = lod_prepare_resync(env, lo, &extent);
8118 /* mirror write, try to init its all components */
8119 rc = lod_prepare_resync_mirror(env, lo,
8120 mlc->mlc_mirror_id);
8125 /* change the file state to SYNC_PENDING */
8126 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
8129 rc = lod_declare_instantiate_components(env, lo, th, 0);
8133 lod_obj_inc_layout_gen(lo);
8135 /* 3. transfer layout version to OST objects.
8136 * transfer new layout version to OST objects so that stale writes
8137 * can be denied. It also ends an era of writing by setting
8138 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
8139 * send write RPC; only resync RPCs could do it. */
8140 layout_attr->la_valid = LA_LAYOUT_VERSION;
8141 layout_attr->la_layout_version = 0;
8142 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
8143 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
8144 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
8149 lod_striping_free(env, lo);
8153 static int lod_declare_update_sync_pending(const struct lu_env *env,
8154 struct lod_object *lo, struct md_layout_change *mlc,
8157 struct lod_thread_info *info = lod_env_info(env);
8158 struct lu_attr *layout_attr = &info->lti_layout_attr;
8159 unsigned sync_components = 0;
8160 unsigned resync_components = 0;
8165 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
8166 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
8167 mlc->mlc_opc == MD_LAYOUT_WRITE);
8169 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
8170 PFID(lod_object_fid(lo)), mlc->mlc_opc);
8172 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
8173 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
8174 PFID(lod_object_fid(lo)));
8176 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
8177 return lod_declare_update_write_pending(env, lo, mlc, th);
8180 /* MD_LAYOUT_RESYNC_DONE */
8182 for (i = 0; i < lo->ldo_comp_cnt; i++) {
8183 struct lod_layout_component *lod_comp;
8186 lod_comp = &lo->ldo_comp_entries[i];
8188 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
8193 for (j = 0; j < mlc->mlc_resync_count; j++) {
8194 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
8197 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
8198 lod_comp->llc_flags &= ~LCME_FL_STALE;
8199 resync_components++;
8205 for (i = 0; i < mlc->mlc_resync_count; i++) {
8206 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
8209 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
8210 "or already synced\n", PFID(lod_object_fid(lo)),
8211 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
8212 GOTO(out, rc = -EINVAL);
8215 if (!sync_components || (mlc->mlc_resync_count && !resync_components)) {
8216 CDEBUG(D_LAYOUT, DFID": no mirror in sync\n",
8217 PFID(lod_object_fid(lo)));
8219 /* tend to return an error code here to prevent
8220 * the MDT from setting SoM attribute */
8221 GOTO(out, rc = -EINVAL);
8224 CDEBUG(D_LAYOUT, DFID": synced %u resynced %u/%zu components\n",
8225 PFID(lod_object_fid(lo)),
8226 sync_components, resync_components, mlc->mlc_resync_count);
8228 lo->ldo_flr_state = LCM_FL_RDONLY;
8229 lod_obj_inc_layout_gen(lo);
8231 layout_attr->la_valid = LA_LAYOUT_VERSION;
8232 layout_attr->la_layout_version = 0;
8233 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
8237 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
8238 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
8239 &info->lti_buf, XATTR_NAME_LOV, 0, th);
8244 lod_striping_free(env, lo);
8248 typedef int (*mlc_handler)(const struct lu_env *env, struct dt_object *dt,
8249 const struct md_layout_change *mlc,
8250 struct thandle *th);
8253 * Attach stripes after target's for migrating directory. NB, we
8254 * only need to declare this, the actual work is done inside
8255 * lod_xattr_set_lmv().
8257 * \param[in] env execution environment
8258 * \param[in] dt target object
8259 * \param[in] mlc layout change data
8260 * \param[in] th transaction handle
8262 * \retval 0 on success
8263 * \retval negative if failed
8265 static int lod_dir_declare_layout_attach(const struct lu_env *env,
8266 struct dt_object *dt,
8267 const struct md_layout_change *mlc,
8270 struct lod_thread_info *info = lod_env_info(env);
8271 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
8272 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
8273 struct lod_object *lo = lod_dt_obj(dt);
8274 struct dt_object *next = dt_object_child(dt);
8275 struct dt_object_format *dof = &info->lti_format;
8276 struct lmv_mds_md_v1 *lmv = mlc->mlc_buf.lb_buf;
8277 struct dt_object **stripes;
8278 __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
8279 struct lu_fid *fid = &info->lti_fid;
8280 struct lod_tgt_desc *tgt;
8281 struct dt_object *dto;
8282 struct dt_device *tgt_dt;
8283 int type = LU_SEQ_RANGE_ANY;
8284 struct dt_insert_rec *rec = &info->lti_dt_rec;
8285 char *stripe_name = info->lti_key;
8286 struct lu_name *sname;
8287 struct linkea_data ldata = { NULL };
8288 struct lu_buf linkea_buf;
8295 if (!lmv_is_sane(lmv))
8298 if (!dt_try_as_dir(env, dt, false))
8301 dof->dof_type = DFT_DIR;
8303 OBD_ALLOC_PTR_ARRAY(stripes, (lo->ldo_dir_stripe_count + stripe_count));
8307 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
8308 stripes[i] = lo->ldo_stripe[i];
8310 rec->rec_type = S_IFDIR;
8312 for (i = 0; i < stripe_count; i++) {
8314 &lmv->lmv_stripe_fids[i]);
8315 if (!fid_is_sane(fid))
8318 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
8322 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
8323 tgt_dt = lod->lod_child;
8325 tgt = LTD_TGT(ltd, idx);
8327 GOTO(out, rc = -ESTALE);
8328 tgt_dt = tgt->ltd_tgt;
8331 dto = dt_locate_at(env, tgt_dt, fid,
8332 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
8335 GOTO(out, rc = PTR_ERR(dto));
8337 stripes[i + lo->ldo_dir_stripe_count] = dto;
8339 if (!dt_try_as_dir(env, dto, true))
8340 GOTO(out, rc = -ENOTDIR);
8342 rc = lod_sub_declare_ref_add(env, dto, th);
8346 rec->rec_fid = lu_object_fid(&dto->do_lu);
8347 rc = lod_sub_declare_insert(env, dto,
8348 (const struct dt_rec *)rec,
8349 (const struct dt_key *)dot, th);
8353 rc = lod_sub_declare_insert(env, dto,
8354 (const struct dt_rec *)rec,
8355 (const struct dt_key *)dotdot, th);
8359 rc = lod_sub_declare_xattr_set(env, dto, &mlc->mlc_buf,
8360 XATTR_NAME_LMV, 0, th);
8364 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
8365 PFID(lu_object_fid(&dto->do_lu)),
8366 i + lo->ldo_dir_stripe_count);
8368 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
8369 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
8370 sname, lu_object_fid(&dt->do_lu));
8374 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
8375 linkea_buf.lb_len = ldata.ld_leh->leh_len;
8376 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
8377 XATTR_NAME_LINK, 0, th);
8381 rc = lod_sub_declare_insert(env, next,
8382 (const struct dt_rec *)rec,
8383 (const struct dt_key *)stripe_name,
8388 rc = lod_sub_declare_ref_add(env, next, th);
8394 OBD_FREE_PTR_ARRAY(lo->ldo_stripe,
8395 lo->ldo_dir_stripes_allocated);
8396 lo->ldo_stripe = stripes;
8397 lo->ldo_is_foreign = 0;
8398 lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count;
8399 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type);
8400 lo->ldo_dir_stripe_count += stripe_count;
8401 lo->ldo_dir_stripes_allocated += stripe_count;
8403 /* plain directory split creates target as a plain directory, while
8404 * after source attached as the first stripe, it becomes a striped
8405 * directory, set correct do_index_ops, otherwise it can't be unlinked.
8407 dt->do_index_ops = &lod_striped_index_ops;
8411 i = lo->ldo_dir_stripe_count;
8412 while (i < lo->ldo_dir_stripe_count + stripe_count && stripes[i])
8413 dt_object_put(env, stripes[i++]);
8415 OBD_FREE_PTR_ARRAY(stripes, stripe_count + lo->ldo_dir_stripe_count);
8419 static int lod_dir_declare_layout_detach(const struct lu_env *env,
8420 struct dt_object *dt,
8421 const struct md_layout_change *unused,
8424 struct lod_thread_info *info = lod_env_info(env);
8425 struct lod_object *lo = lod_dt_obj(dt);
8426 struct dt_object *next = dt_object_child(dt);
8427 char *stripe_name = info->lti_key;
8428 struct dt_object *dto;
8432 if (!dt_try_as_dir(env, dt, true))
8435 if (!lo->ldo_dir_stripe_count)
8436 return lod_sub_declare_delete(env, next,
8437 (const struct dt_key *)dotdot, th);
8439 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8440 dto = lo->ldo_stripe[i];
8444 if (!dt_try_as_dir(env, dto, true))
8447 rc = lod_sub_declare_delete(env, dto,
8448 (const struct dt_key *)dotdot, th);
8452 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8453 PFID(lu_object_fid(&dto->do_lu)), i);
8455 rc = lod_sub_declare_delete(env, next,
8456 (const struct dt_key *)stripe_name, th);
8460 rc = lod_sub_declare_ref_del(env, next, th);
8468 static int dt_dir_is_empty(const struct lu_env *env,
8469 struct dt_object *obj)
8472 const struct dt_it_ops *iops;
8477 if (!dt_try_as_dir(env, obj, true))
8480 iops = &obj->do_index_ops->dio_it;
8481 it = iops->init(env, obj, LUDA_64BITHASH);
8483 RETURN(PTR_ERR(it));
8485 rc = iops->get(env, it, (const struct dt_key *)"");
8489 for (rc = 0, i = 0; rc == 0 && i < 3; ++i)
8490 rc = iops->next(env, it);
8496 /* Huh? Index contains no zero key? */
8501 iops->fini(env, it);
8506 static int lod_dir_declare_layout_shrink(const struct lu_env *env,
8507 struct dt_object *dt,
8508 const struct md_layout_change *mlc,
8511 struct lod_thread_info *info = lod_env_info(env);
8512 struct lod_object *lo = lod_dt_obj(dt);
8513 struct dt_object *next = dt_object_child(dt);
8514 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
8515 char *stripe_name = info->lti_key;
8516 struct lu_buf *lmv_buf = &info->lti_buf;
8517 __u32 final_stripe_count;
8518 struct dt_object *dto;
8524 if (!dt_try_as_dir(env, dt, true))
8527 /* shouldn't be called on plain directory */
8528 LASSERT(lo->ldo_dir_stripe_count);
8530 lmv_buf->lb_buf = &info->lti_lmv.lmv_md_v1;
8531 lmv_buf->lb_len = sizeof(info->lti_lmv.lmv_md_v1);
8533 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
8534 LASSERT(final_stripe_count &&
8535 final_stripe_count < lo->ldo_dir_stripe_count);
8537 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8538 dto = lo->ldo_stripe[i];
8542 if (i < final_stripe_count) {
8543 rc = lod_sub_declare_xattr_set(env, dto, lmv_buf,
8545 LU_XATTR_REPLACE, th);
8552 rc = dt_dir_is_empty(env, dto);
8556 rc = lod_sub_declare_ref_del(env, dto, th);
8560 rc = lod_sub_declare_destroy(env, dto, th);
8564 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8565 PFID(lu_object_fid(&dto->do_lu)), i);
8567 rc = lod_sub_declare_delete(env, next,
8568 (const struct dt_key *)stripe_name, th);
8572 rc = lod_sub_declare_ref_del(env, next, th);
8577 rc = lod_sub_declare_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
8578 LU_XATTR_REPLACE, th);
8583 * Allocate stripes for split directory.
8585 * \param[in] env execution environment
8586 * \param[in] dt target object
8587 * \param[in] mlc layout change data
8588 * \param[in] th transaction handle
8590 * \retval 0 on success
8591 * \retval negative if failed
8593 static int lod_dir_declare_layout_split(const struct lu_env *env,
8594 struct dt_object *dt,
8595 const struct md_layout_change *mlc,
8598 struct lod_thread_info *info = lod_env_info(env);
8599 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
8600 struct lod_object *lo = lod_dt_obj(dt);
8601 struct dt_object_format *dof = &info->lti_format;
8602 struct lmv_user_md_v1 *lum = mlc->mlc_spec->u.sp_ea.eadata;
8603 struct dt_object **stripes;
8611 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
8612 LASSERT(le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT);
8614 saved_count = lo->ldo_dir_stripes_allocated;
8615 stripe_count = le32_to_cpu(lum->lum_stripe_count);
8616 if (stripe_count <= saved_count)
8619 dof->dof_type = DFT_DIR;
8621 OBD_ALLOC(stripes, sizeof(*stripes) * stripe_count);
8625 for (i = 0; i < lo->ldo_dir_stripes_allocated; i++)
8626 stripes[i] = lo->ldo_stripe[i];
8628 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
8629 rc = lod_mdt_alloc_qos(env, lo, stripes, saved_count, stripe_count);
8631 rc = lod_mdt_alloc_rr(env, lo, stripes, saved_count,
8634 OBD_FREE(stripes, sizeof(*stripes) * stripe_count);
8638 LASSERT(rc > saved_count);
8639 OBD_FREE(lo->ldo_stripe,
8640 sizeof(*stripes) * lo->ldo_dir_stripes_allocated);
8641 lo->ldo_stripe = stripes;
8642 lo->ldo_is_foreign = 0;
8643 lo->ldo_dir_striped = 1;
8644 lo->ldo_dir_stripe_count = rc;
8645 lo->ldo_dir_stripes_allocated = stripe_count;
8646 lo->ldo_dir_split_hash = lo->ldo_dir_hash_type;
8647 lo->ldo_dir_hash_type = le32_to_cpu(lum->lum_hash_type);
8648 if (!lmv_is_known_hash_type(lo->ldo_dir_hash_type))
8649 lo->ldo_dir_hash_type =
8650 lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
8651 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_SPLIT | LMV_HASH_FLAG_MIGRATION;
8652 lo->ldo_dir_split_offset = saved_count;
8653 lo->ldo_dir_layout_version++;
8654 lo->ldo_dir_stripe_loaded = 1;
8656 rc = lod_dir_declare_create_stripes(env, dt, mlc->mlc_attr, dof, th);
8658 lod_striping_free(env, lo);
8664 * detach all stripes from dir master object, NB, stripes are not destroyed, but
8665 * deleted from it's parent namespace, this function is called in two places:
8666 * 1. mdd_migrate_mdt() detach stripes from source, and attach them to
8668 * 2. mdd_dir_layout_update() detach stripe before turning 1-stripe directory to
8669 * a plain directory.
8671 * \param[in] env execution environment
8672 * \param[in] dt target object
8673 * \param[in] mlc layout change data
8674 * \param[in] th transaction handle
8676 * \retval 0 on success
8677 * \retval negative if failed
8679 static int lod_dir_layout_detach(const struct lu_env *env,
8680 struct dt_object *dt,
8681 const struct md_layout_change *mlc,
8684 struct lod_thread_info *info = lod_env_info(env);
8685 struct lod_object *lo = lod_dt_obj(dt);
8686 struct dt_object *next = dt_object_child(dt);
8687 char *stripe_name = info->lti_key;
8688 struct dt_object *dto;
8694 if (!lo->ldo_dir_stripe_count) {
8695 /* plain directory delete .. */
8696 rc = lod_sub_delete(env, next,
8697 (const struct dt_key *)dotdot, th);
8701 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8702 dto = lo->ldo_stripe[i];
8706 rc = lod_sub_delete(env, dto,
8707 (const struct dt_key *)dotdot, th);
8711 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8712 PFID(lu_object_fid(&dto->do_lu)), i);
8714 rc = lod_sub_delete(env, next,
8715 (const struct dt_key *)stripe_name, th);
8719 rc = lod_sub_ref_del(env, next, th);
8724 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8725 dto = lo->ldo_stripe[i];
8727 dt_object_put(env, dto);
8729 OBD_FREE_PTR_ARRAY(lo->ldo_stripe, lo->ldo_dir_stripes_allocated);
8730 lo->ldo_stripe = NULL;
8731 lo->ldo_dir_stripes_allocated = 0;
8732 lo->ldo_dir_stripe_count = 0;
8733 dt->do_index_ops = &lod_index_ops;
8738 static int lod_dir_layout_shrink(const struct lu_env *env,
8739 struct dt_object *dt,
8740 const struct md_layout_change *mlc,
8743 struct lod_thread_info *info = lod_env_info(env);
8744 struct lod_object *lo = lod_dt_obj(dt);
8745 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
8746 struct dt_object *next = dt_object_child(dt);
8747 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
8748 __u32 final_stripe_count;
8749 char *stripe_name = info->lti_key;
8750 struct dt_object *dto;
8751 struct lu_buf *lmv_buf = &info->lti_buf;
8752 struct lmv_mds_md_v1 *lmv = &info->lti_lmv.lmv_md_v1;
8754 int type = LU_SEQ_RANGE_ANY;
8760 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
8762 lmv_buf->lb_buf = lmv;
8763 lmv_buf->lb_len = sizeof(*lmv);
8764 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
8765 lmv->lmv_stripe_count = cpu_to_le32(final_stripe_count);
8766 lmv->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type) &
8767 cpu_to_le32(LMV_HASH_TYPE_MASK |
8768 LMV_HASH_FLAG_FIXED);
8769 lmv->lmv_layout_version =
8770 cpu_to_le32(lo->ldo_dir_layout_version + 1);
8771 lmv->lmv_migrate_offset = 0;
8772 lmv->lmv_migrate_hash = 0;
8774 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8775 dto = lo->ldo_stripe[i];
8779 if (i < final_stripe_count) {
8780 rc = lod_fld_lookup(env, lod,
8781 lu_object_fid(&dto->do_lu),
8786 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
8787 rc = lod_sub_xattr_set(env, dto, lmv_buf,
8789 LU_XATTR_REPLACE, th);
8796 dt_write_lock(env, dto, DT_TGT_CHILD);
8797 rc = lod_sub_ref_del(env, dto, th);
8798 dt_write_unlock(env, dto);
8802 rc = lod_sub_destroy(env, dto, th);
8806 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8807 PFID(lu_object_fid(&dto->do_lu)), i);
8809 rc = lod_sub_delete(env, next,
8810 (const struct dt_key *)stripe_name, th);
8814 rc = lod_sub_ref_del(env, next, th);
8819 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &mdtidx,
8824 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
8825 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
8826 rc = lod_sub_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
8827 LU_XATTR_REPLACE, th);
8831 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
8832 dto = lo->ldo_stripe[i];
8834 dt_object_put(env, dto);
8836 lo->ldo_dir_stripe_count = final_stripe_count;
8841 static mlc_handler dir_mlc_declare_ops[MD_LAYOUT_MAX] = {
8842 [MD_LAYOUT_ATTACH] = lod_dir_declare_layout_attach,
8843 [MD_LAYOUT_DETACH] = lod_dir_declare_layout_detach,
8844 [MD_LAYOUT_SHRINK] = lod_dir_declare_layout_shrink,
8845 [MD_LAYOUT_SPLIT] = lod_dir_declare_layout_split,
8848 static mlc_handler dir_mlc_ops[MD_LAYOUT_MAX] = {
8849 [MD_LAYOUT_DETACH] = lod_dir_layout_detach,
8850 [MD_LAYOUT_SHRINK] = lod_dir_layout_shrink,
8853 static int lod_declare_layout_change(const struct lu_env *env,
8854 struct dt_object *dt, struct md_layout_change *mlc,
8857 struct lod_thread_info *info = lod_env_info(env);
8858 struct lod_object *lo = lod_dt_obj(dt);
8863 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
8864 LASSERT(dir_mlc_declare_ops[mlc->mlc_opc]);
8865 rc = dir_mlc_declare_ops[mlc->mlc_opc](env, dt, mlc, th);
8869 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
8870 dt_object_remote(dt_object_child(dt)))
8873 rc = lod_striping_load(env, lo);
8877 LASSERT(lo->ldo_comp_cnt > 0);
8879 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
8883 switch (lo->ldo_flr_state) {
8885 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
8889 rc = lod_declare_update_rdonly(env, lo, mlc, th);
8891 case LCM_FL_WRITE_PENDING:
8892 rc = lod_declare_update_write_pending(env, lo, mlc, th);
8894 case LCM_FL_SYNC_PENDING:
8895 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
8906 * Instantiate layout component objects which covers the intent write offset.
8908 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
8909 struct md_layout_change *mlc, struct thandle *th)
8911 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
8912 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
8913 struct lod_object *lo = lod_dt_obj(dt);
8918 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
8919 LASSERT(dir_mlc_ops[mlc->mlc_opc]);
8920 rc = dir_mlc_ops[mlc->mlc_opc](env, dt, mlc, th);
8924 rc = lod_striped_create(env, dt, attr, NULL, th);
8925 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
8926 layout_attr->la_layout_version |= lo->ldo_layout_gen;
8927 rc = lod_attr_set(env, dt, layout_attr, th);
8933 const struct dt_object_operations lod_obj_ops = {
8934 .do_read_lock = lod_read_lock,
8935 .do_write_lock = lod_write_lock,
8936 .do_read_unlock = lod_read_unlock,
8937 .do_write_unlock = lod_write_unlock,
8938 .do_write_locked = lod_write_locked,
8939 .do_attr_get = lod_attr_get,
8940 .do_declare_attr_set = lod_declare_attr_set,
8941 .do_attr_set = lod_attr_set,
8942 .do_xattr_get = lod_xattr_get,
8943 .do_declare_xattr_set = lod_declare_xattr_set,
8944 .do_xattr_set = lod_xattr_set,
8945 .do_declare_xattr_del = lod_declare_xattr_del,
8946 .do_xattr_del = lod_xattr_del,
8947 .do_xattr_list = lod_xattr_list,
8948 .do_ah_init = lod_ah_init,
8949 .do_declare_create = lod_declare_create,
8950 .do_create = lod_create,
8951 .do_declare_destroy = lod_declare_destroy,
8952 .do_destroy = lod_destroy,
8953 .do_index_try = lod_index_try,
8954 .do_declare_ref_add = lod_declare_ref_add,
8955 .do_ref_add = lod_ref_add,
8956 .do_declare_ref_del = lod_declare_ref_del,
8957 .do_ref_del = lod_ref_del,
8958 .do_object_sync = lod_object_sync,
8959 .do_object_lock = lod_object_lock,
8960 .do_object_unlock = lod_object_unlock,
8961 .do_invalidate = lod_invalidate,
8962 .do_declare_layout_change = lod_declare_layout_change,
8963 .do_layout_change = lod_layout_change,
8967 * Implementation of dt_body_operations::dbo_read.
8969 * \see dt_body_operations::dbo_read() in the API description for details.
8971 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
8972 struct lu_buf *buf, loff_t *pos)
8974 struct dt_object *next = dt_object_child(dt);
8976 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
8977 S_ISLNK(dt->do_lu.lo_header->loh_attr));
8978 return next->do_body_ops->dbo_read(env, next, buf, pos);
8982 * Implementation of dt_body_operations::dbo_declare_write.
8984 * \see dt_body_operations::dbo_declare_write() in the API description
8987 static ssize_t lod_declare_write(const struct lu_env *env,
8988 struct dt_object *dt,
8989 const struct lu_buf *buf, loff_t pos,
8992 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
8996 * Implementation of dt_body_operations::dbo_write.
8998 * \see dt_body_operations::dbo_write() in the API description for details.
9000 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
9001 const struct lu_buf *buf, loff_t *pos,
9004 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
9005 S_ISLNK(dt->do_lu.lo_header->loh_attr));
9006 return lod_sub_write(env, dt_object_child(dt), buf, pos, th);
9009 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
9010 __u64 start, __u64 end, struct thandle *th)
9012 if (dt_object_remote(dt))
9015 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
9018 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
9019 __u64 start, __u64 end, struct thandle *th)
9021 if (dt_object_remote(dt))
9024 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
9025 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
9029 * different type of files use the same body_ops because object may be created
9030 * in OUT, where there is no chance to set correct body_ops for each type, so
9031 * body_ops themselves will check file type inside, see lod_read/write/punch for
9034 static const struct dt_body_operations lod_body_ops = {
9035 .dbo_read = lod_read,
9036 .dbo_declare_write = lod_declare_write,
9037 .dbo_write = lod_write,
9038 .dbo_declare_punch = lod_declare_punch,
9039 .dbo_punch = lod_punch,
9043 * Implementation of lu_object_operations::loo_object_init.
9045 * The function determines the type and the index of the target device using
9046 * sequence of the object's FID. Then passes control down to the
9047 * corresponding device:
9048 * OSD for the local objects, OSP for remote
9050 * \see lu_object_operations::loo_object_init() in the API description
9053 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
9054 const struct lu_object_conf *conf)
9056 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
9057 struct lu_device *cdev = NULL;
9058 struct lu_object *cobj;
9059 struct lod_tgt_descs *ltd = NULL;
9060 struct lod_tgt_desc *tgt;
9062 int type = LU_SEQ_RANGE_ANY;
9066 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
9070 if (type == LU_SEQ_RANGE_MDT &&
9071 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
9072 cdev = &lod->lod_child->dd_lu_dev;
9073 } else if (type == LU_SEQ_RANGE_MDT) {
9074 ltd = &lod->lod_mdt_descs;
9076 } else if (type == LU_SEQ_RANGE_OST) {
9077 ltd = &lod->lod_ost_descs;
9084 if (ltd->ltd_tgts_size > idx &&
9085 test_bit(idx, ltd->ltd_tgt_bitmap)) {
9086 tgt = LTD_TGT(ltd, idx);
9088 LASSERT(tgt != NULL);
9089 LASSERT(tgt->ltd_tgt != NULL);
9091 cdev = &(tgt->ltd_tgt->dd_lu_dev);
9093 lod_putref(lod, ltd);
9096 if (unlikely(cdev == NULL))
9099 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
9100 if (unlikely(cobj == NULL))
9103 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
9105 lu_object_add(lo, cobj);
9112 * Release resources associated with striping.
9114 * If the object is striped (regular or directory), then release
9115 * the stripe objects references and free the ldo_stripe array.
9117 * \param[in] env execution environment
9118 * \param[in] lo object
9120 void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo)
9122 struct lod_layout_component *lod_comp;
9123 __u32 obj_attr = lo->ldo_obj.do_lu.lo_header->loh_attr;
9126 if (unlikely(lo->ldo_is_foreign)) {
9127 if (S_ISREG(obj_attr)) {
9128 lod_free_foreign_lov(lo);
9129 lo->ldo_comp_cached = 0;
9130 } else if (S_ISDIR(obj_attr)) {
9131 lod_free_foreign_lmv(lo);
9132 lo->ldo_dir_stripe_loaded = 0;
9134 } else if (lo->ldo_stripe != NULL) {
9135 LASSERT(lo->ldo_comp_entries == NULL);
9136 LASSERT(lo->ldo_dir_stripes_allocated > 0);
9138 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
9139 if (lo->ldo_stripe[i])
9140 dt_object_put(env, lo->ldo_stripe[i]);
9143 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
9144 OBD_FREE(lo->ldo_stripe, j);
9145 lo->ldo_stripe = NULL;
9146 lo->ldo_dir_stripes_allocated = 0;
9147 lo->ldo_dir_stripe_loaded = 0;
9148 lo->ldo_dir_stripe_count = 0;
9149 lo->ldo_obj.do_index_ops = NULL;
9150 } else if (lo->ldo_comp_entries != NULL) {
9151 for (i = 0; i < lo->ldo_comp_cnt; i++) {
9152 /* free lod_layout_component::llc_stripe array */
9153 lod_comp = &lo->ldo_comp_entries[i];
9155 if (lod_comp->llc_stripe == NULL)
9157 LASSERT(lod_comp->llc_stripes_allocated != 0);
9158 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
9159 if (lod_comp->llc_stripe[j] != NULL)
9161 &lod_comp->llc_stripe[j]->do_lu);
9163 OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe,
9164 lod_comp->llc_stripes_allocated);
9165 lod_comp->llc_stripe = NULL;
9166 OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices,
9167 lod_comp->llc_stripes_allocated);
9168 lod_comp->llc_ost_indices = NULL;
9169 lod_comp->llc_stripes_allocated = 0;
9171 lod_free_comp_entries(lo);
9172 lo->ldo_comp_cached = 0;
9176 void lod_striping_free(const struct lu_env *env, struct lod_object *lo)
9178 mutex_lock(&lo->ldo_layout_mutex);
9179 lod_striping_free_nolock(env, lo);
9180 mutex_unlock(&lo->ldo_layout_mutex);
9184 * Implementation of lu_object_operations::loo_object_free.
9186 * \see lu_object_operations::loo_object_free() in the API description
9189 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
9191 struct lod_object *lo = lu2lod_obj(o);
9193 /* release all underlying object pinned */
9194 lod_striping_free(env, lo);
9196 /* lo doesn't contain a lu_object_header, so we don't need call_rcu */
9197 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
9201 * Implementation of lu_object_operations::loo_object_release.
9203 * \see lu_object_operations::loo_object_release() in the API description
9206 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
9208 /* XXX: shouldn't we release everything here in case if object
9209 * creation failed before? */
9213 * Implementation of lu_object_operations::loo_object_print.
9215 * \see lu_object_operations::loo_object_print() in the API description
9218 static int lod_object_print(const struct lu_env *env, void *cookie,
9219 lu_printer_t p, const struct lu_object *l)
9221 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
9223 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
9226 const struct lu_object_operations lod_lu_obj_ops = {
9227 .loo_object_init = lod_object_init,
9228 .loo_object_free = lod_object_free,
9229 .loo_object_release = lod_object_release,
9230 .loo_object_print = lod_object_print,