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);
1227 static bool lod_obj_attr_set_comp_skip_cb(const struct lu_env *env,
1228 struct lod_object *lo, int comp_idx,
1229 struct lod_obj_stripe_cb_data *data)
1231 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[comp_idx];
1232 bool skipped = false;
1234 if (!(data->locd_attr->la_valid & LA_LAYOUT_VERSION))
1237 switch (lo->ldo_flr_state) {
1238 case LCM_FL_WRITE_PENDING: {
1241 /* skip stale components */
1242 if (lod_comp->llc_flags & LCME_FL_STALE) {
1247 /* skip valid and overlapping components, therefore any
1248 * attempts to write overlapped components will never succeed
1249 * because client will get EINPROGRESS. */
1250 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1254 if (lo->ldo_comp_entries[i].llc_flags & LCME_FL_STALE)
1257 if (lu_extent_is_overlapped(&lod_comp->llc_extent,
1258 &lo->ldo_comp_entries[i].llc_extent)) {
1266 case LCM_FL_SYNC_PENDING:
1269 LASSERTF(0, "impossible: %d\n", lo->ldo_flr_state);
1273 CDEBUG(D_LAYOUT, DFID": %s to set component %x to version: %u\n",
1274 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
1275 skipped ? "skipped" : "chose", lod_comp->llc_id,
1276 data->locd_attr->la_layout_version);
1282 lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
1283 struct dt_object *dt, struct thandle *th,
1284 int comp_idx, int stripe_idx,
1285 struct lod_obj_stripe_cb_data *data)
1287 if (data->locd_declare)
1288 return lod_sub_declare_attr_set(env, dt, data->locd_attr, th);
1290 if (data->locd_attr->la_valid & LA_LAYOUT_VERSION) {
1291 CDEBUG(D_LAYOUT, DFID": set layout version: %u, comp_idx: %d\n",
1292 PFID(lu_object_fid(&dt->do_lu)),
1293 data->locd_attr->la_layout_version, comp_idx);
1296 return lod_sub_attr_set(env, dt, data->locd_attr, th);
1300 * Implementation of dt_object_operations::do_declare_attr_set.
1302 * If the object is striped, then apply the changes to all the stripes.
1304 * \see dt_object_operations::do_declare_attr_set() in the API description
1307 static int lod_declare_attr_set(const struct lu_env *env,
1308 struct dt_object *dt,
1309 const struct lu_attr *attr,
1312 struct dt_object *next = dt_object_child(dt);
1313 struct lod_object *lo = lod_dt_obj(dt);
1318 * declare setattr on the local object
1320 rc = lod_sub_declare_attr_set(env, next, attr, th);
1324 /* osp_declare_attr_set() ignores all attributes other than
1325 * UID, GID, PROJID, and size, and osp_attr_set() ignores all
1326 * but UID, GID and PROJID. Declaration of size attr setting
1327 * happens through lod_declare_init_size(), and not through
1328 * this function. Therefore we need not load striping unless
1329 * ownership is changing. This should save memory and (we hope)
1330 * speed up rename().
1332 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1333 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1336 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1339 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID | LA_MODE |
1340 LA_ATIME | LA_MTIME | LA_CTIME |
1345 * load striping information, notice we don't do this when object
1346 * is being initialized as we don't need this information till
1347 * few specific cases like destroy, chown
1349 rc = lod_striping_load(env, lo);
1353 if (!lod_obj_is_striped(dt))
1357 * if object is striped declare changes on the stripes
1359 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1360 LASSERT(lo->ldo_stripe);
1361 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1362 if (lo->ldo_stripe[i] == NULL)
1364 if (!dt_object_exists(lo->ldo_stripe[i]))
1366 rc = lod_sub_declare_attr_set(env, lo->ldo_stripe[i],
1372 struct lod_obj_stripe_cb_data data = { { 0 } };
1374 data.locd_attr = attr;
1375 data.locd_declare = true;
1376 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1377 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1383 if (!dt_object_exists(next) || dt_object_remote(next) ||
1384 !S_ISREG(attr->la_mode))
1387 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1388 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
1392 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) ||
1393 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1394 struct lod_thread_info *info = lod_env_info(env);
1395 struct lu_buf *buf = &info->lti_buf;
1397 buf->lb_buf = info->lti_ea_store;
1398 buf->lb_len = info->lti_ea_store_size;
1399 rc = lod_sub_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1400 LU_XATTR_REPLACE, th);
1407 * Implementation of dt_object_operations::do_attr_set.
1409 * If the object is striped, then apply the changes to all or subset of
1410 * the stripes depending on the object type and specific attributes.
1412 * \see dt_object_operations::do_attr_set() in the API description for details.
1414 static int lod_attr_set(const struct lu_env *env,
1415 struct dt_object *dt,
1416 const struct lu_attr *attr,
1419 struct dt_object *next = dt_object_child(dt);
1420 struct lod_object *lo = lod_dt_obj(dt);
1425 * apply changes to the local object
1427 rc = lod_sub_attr_set(env, next, attr, th);
1431 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1432 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1435 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1438 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE | LA_PROJID |
1439 LA_ATIME | LA_MTIME | LA_CTIME |
1444 /* FIXME: a tricky case in the code path of mdd_layout_change():
1445 * the in-memory striping information has been freed in lod_xattr_set()
1446 * due to layout change. It has to load stripe here again. It only
1447 * changes flags of layout so declare_attr_set() is still accurate */
1448 rc = lod_striping_load(env, lo);
1452 if (!lod_obj_is_striped(dt))
1456 * if object is striped, apply changes to all the stripes
1458 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1459 LASSERT(lo->ldo_stripe);
1460 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1461 if (unlikely(lo->ldo_stripe[i] == NULL))
1464 if ((dt_object_exists(lo->ldo_stripe[i]) == 0))
1467 rc = lod_sub_attr_set(env, lo->ldo_stripe[i], attr, th);
1472 struct lod_obj_stripe_cb_data data = { { 0 } };
1474 data.locd_attr = attr;
1475 data.locd_declare = false;
1476 data.locd_comp_skip_cb = lod_obj_attr_set_comp_skip_cb;
1477 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1478 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1484 if (!dt_object_exists(next) || dt_object_remote(next) ||
1485 !S_ISREG(attr->la_mode))
1488 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1489 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
1493 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE)) {
1494 struct lod_thread_info *info = lod_env_info(env);
1495 struct lu_buf *buf = &info->lti_buf;
1496 struct ost_id *oi = &info->lti_ostid;
1497 struct lu_fid *fid = &info->lti_fid;
1498 struct lov_mds_md_v1 *lmm;
1499 struct lov_ost_data_v1 *objs;
1502 rc = lod_get_lov_ea(env, lo);
1506 buf->lb_buf = info->lti_ea_store;
1507 buf->lb_len = info->lti_ea_store_size;
1508 lmm = info->lti_ea_store;
1509 magic = le32_to_cpu(lmm->lmm_magic);
1510 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1511 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1512 struct lov_comp_md_entry_v1 *lcme =
1513 &lcm->lcm_entries[0];
1515 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
1516 magic = le32_to_cpu(lmm->lmm_magic);
1519 if (magic == LOV_MAGIC_V1)
1520 objs = &(lmm->lmm_objects[0]);
1522 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1523 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1524 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1526 fid_to_ostid(fid, oi);
1527 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1529 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1530 LU_XATTR_REPLACE, th);
1531 } else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1532 struct lod_thread_info *info = lod_env_info(env);
1533 struct lu_buf *buf = &info->lti_buf;
1534 struct lov_comp_md_v1 *lcm;
1535 struct lov_comp_md_entry_v1 *lcme;
1537 rc = lod_get_lov_ea(env, lo);
1541 buf->lb_buf = info->lti_ea_store;
1542 buf->lb_len = info->lti_ea_store_size;
1544 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1 &&
1545 le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_SEL)
1548 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1549 lcme = &lcm->lcm_entries[0];
1550 le64_add_cpu(&lcme->lcme_extent.e_start, 1);
1551 le64_add_cpu(&lcme->lcme_extent.e_end, -1);
1553 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1554 LU_XATTR_REPLACE, th);
1561 * Implementation of dt_object_operations::do_xattr_get.
1563 * If LOV EA is requested from the root object and it's not
1564 * found, then return default striping for the filesystem.
1566 * \see dt_object_operations::do_xattr_get() in the API description for details.
1568 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1569 struct lu_buf *buf, const char *name)
1571 struct lod_thread_info *info = lod_env_info(env);
1572 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1577 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1578 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1579 struct lmv_mds_md_v1 *lmv1;
1580 struct lmv_foreign_md *lfm;
1583 if (rc > (typeof(rc))sizeof(*lmv1))
1586 /* short (<= sizeof(struct lmv_mds_md_v1)) foreign LMV case */
1587 /* XXX empty foreign LMV is not allowed */
1588 if (rc <= offsetof(typeof(*lfm), lfm_value))
1589 RETURN(rc = rc > 0 ? -EINVAL : rc);
1591 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1592 BUILD_BUG_ON(sizeof(*lmv1) > sizeof(info->lti_key));
1594 /* lti_buf is large enough for *lmv1 or a short
1595 * (<= sizeof(struct lmv_mds_md_v1)) foreign LMV
1597 info->lti_buf.lb_buf = info->lti_key;
1598 info->lti_buf.lb_len = sizeof(*lmv1);
1599 rc = dt_xattr_get(env, dt_object_child(dt),
1600 &info->lti_buf, name);
1601 if (unlikely(rc <= offsetof(typeof(*lfm),
1603 RETURN(rc = rc > 0 ? -EINVAL : rc);
1605 lfm = info->lti_buf.lb_buf;
1606 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN)
1609 if (unlikely(rc != sizeof(*lmv1)))
1610 RETURN(rc = rc > 0 ? -EINVAL : rc);
1612 lmv1 = info->lti_buf.lb_buf;
1613 /* The on-disk LMV EA only contains header, but the
1614 * returned LMV EA size should contain the space for
1615 * the FIDs of all shards of the striped directory. */
1616 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1617 rc = lmv_mds_md_size(
1618 le32_to_cpu(lmv1->lmv_stripe_count),
1619 le32_to_cpu(lmv1->lmv_magic));
1622 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1625 if (rc != sizeof(*lmv1))
1626 RETURN(rc = rc > 0 ? -EINVAL : rc);
1628 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1632 RETURN(rc = rc1 != 0 ? rc1 : rc);
1635 if ((rc > 0) && buf->lb_buf && strcmp(name, XATTR_NAME_LOV) == 0) {
1636 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1638 if (lcm->lcm_magic == cpu_to_le32(LOV_MAGIC_SEL))
1639 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
1642 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1646 * XXX: Only used by lfsck
1648 * lod returns default striping on the real root of the device
1649 * this is like the root stores default striping for the whole
1650 * filesystem. historically we've been using a different approach
1651 * and store it in the config.
1653 dt_root_get(env, dev->lod_child, &info->lti_fid);
1654 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1656 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1657 struct lov_user_md *lum = buf->lb_buf;
1658 struct lov_desc *desc = &dev->lod_ost_descs.ltd_lov_desc;
1660 if (buf->lb_buf == NULL) {
1662 } else if (buf->lb_len >= sizeof(*lum)) {
1663 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1664 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1665 lmm_oi_set_id(&lum->lmm_oi, 0);
1666 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1667 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1668 lum->lmm_stripe_size = cpu_to_le32(
1669 desc->ld_default_stripe_size);
1670 lum->lmm_stripe_count = cpu_to_le16(
1671 desc->ld_default_stripe_count);
1672 lum->lmm_stripe_offset = cpu_to_le16(
1673 desc->ld_default_stripe_offset);
1686 * Checks that the magic of the stripe is sane.
1688 * \param[in] lod lod device
1689 * \param[in] lum a buffer storing LMV EA to verify
1691 * \retval 0 if the EA is sane
1692 * \retval negative otherwise
1694 static int lod_verify_md_striping(struct lod_device *lod,
1695 const struct lmv_user_md_v1 *lum)
1697 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1698 CERROR("%s: invalid lmv_user_md: magic = %x, "
1699 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1700 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1701 (int)le32_to_cpu(lum->lum_stripe_offset),
1702 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1710 * Initialize LMV EA for a slave.
1712 * Initialize slave's LMV EA from the master's LMV EA.
1714 * \param[in] master_lmv a buffer containing master's EA
1715 * \param[out] slave_lmv a buffer where slave's EA will be stored
1718 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1719 const struct lmv_mds_md_v1 *master_lmv)
1721 *slave_lmv = *master_lmv;
1722 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1728 * Generate LMV EA from the object passed as \a dt. The object must have
1729 * the stripes created and initialized.
1731 * \param[in] env execution environment
1732 * \param[in] dt object
1733 * \param[out] lmv_buf buffer storing generated LMV EA
1735 * \retval 0 on success
1736 * \retval negative if failed
1738 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1739 struct lu_buf *lmv_buf)
1741 struct lod_thread_info *info = lod_env_info(env);
1742 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1743 struct lod_object *lo = lod_dt_obj(dt);
1744 struct lmv_mds_md_v1 *lmm1;
1746 int type = LU_SEQ_RANGE_ANY;
1751 LASSERT(lo->ldo_dir_striped != 0);
1752 LASSERT(lo->ldo_dir_stripe_count > 0);
1753 stripe_count = lo->ldo_dir_stripe_count;
1754 /* Only store the LMV EA heahder on the disk. */
1755 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1756 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1760 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1763 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1764 memset(lmm1, 0, sizeof(*lmm1));
1765 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1766 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1767 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1768 lmm1->lmv_layout_version = cpu_to_le32(lo->ldo_dir_layout_version);
1769 if (lod_is_layout_changing(lo)) {
1770 lmm1->lmv_migrate_hash = cpu_to_le32(lo->ldo_dir_migrate_hash);
1771 lmm1->lmv_migrate_offset =
1772 cpu_to_le32(lo->ldo_dir_migrate_offset);
1774 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1779 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1780 lmv_buf->lb_buf = info->lti_ea_store;
1781 lmv_buf->lb_len = sizeof(*lmm1);
1787 * Create in-core represenation for a striped directory.
1789 * Parse the buffer containing LMV EA and instantiate LU objects
1790 * representing the stripe objects. The pointers to the objects are
1791 * stored in ldo_stripe field of \a lo. This function is used when
1792 * we need to access an already created object (i.e. load from a disk).
1794 * \param[in] env execution environment
1795 * \param[in] lo lod object
1796 * \param[in] buf buffer containing LMV EA
1798 * \retval 0 on success
1799 * \retval negative if failed
1801 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1802 const struct lu_buf *buf)
1804 struct lod_thread_info *info = lod_env_info(env);
1805 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1806 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1807 struct dt_object **stripe;
1808 union lmv_mds_md *lmm = buf->lb_buf;
1809 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1810 struct lu_fid *fid = &info->lti_fid;
1815 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1817 /* XXX may be useless as not called for foreign LMV ?? */
1818 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_FOREIGN)
1821 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1822 lo->ldo_dir_slave_stripe = 1;
1826 if (!lmv_is_sane(lmv1))
1829 LASSERT(lo->ldo_stripe == NULL);
1830 OBD_ALLOC_PTR_ARRAY(stripe, le32_to_cpu(lmv1->lmv_stripe_count));
1834 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1835 struct dt_device *tgt_dt;
1836 struct dt_object *dto;
1837 int type = LU_SEQ_RANGE_ANY;
1840 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1841 if (!fid_is_sane(fid)) {
1846 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1850 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1851 tgt_dt = lod->lod_child;
1853 struct lod_tgt_desc *tgt;
1855 tgt = LTD_TGT(ltd, idx);
1857 GOTO(out, rc = -ESTALE);
1858 tgt_dt = tgt->ltd_tgt;
1861 dto = dt_locate_at(env, tgt_dt, fid,
1862 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1865 GOTO(out, rc = PTR_ERR(dto));
1870 lo->ldo_stripe = stripe;
1871 lo->ldo_is_foreign = 0;
1872 lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1873 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1874 lo->ldo_dir_layout_version = le32_to_cpu(lmv1->lmv_layout_version);
1875 lo->ldo_dir_migrate_offset = le32_to_cpu(lmv1->lmv_migrate_offset);
1876 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv1->lmv_migrate_hash);
1877 lo->ldo_dir_hash_type = le32_to_cpu(lmv1->lmv_hash_type);
1879 lod_striping_free_nolock(env, lo);
1885 * Declare create a striped directory.
1887 * Declare creating a striped directory with a given stripe pattern on the
1888 * specified MDTs. A striped directory is represented as a regular directory
1889 * - an index listing all the stripes. The stripes point back to the master
1890 * object with ".." and LinkEA. The master object gets LMV EA which
1891 * identifies it as a striped directory. The function allocates FIDs
1894 * \param[in] env execution environment
1895 * \param[in] dt object
1896 * \param[in] attr attributes to initialize the objects with
1897 * \param[in] dof type of objects to be created
1898 * \param[in] th transaction handle
1900 * \retval 0 on success
1901 * \retval negative if failed
1903 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1904 struct dt_object *dt,
1905 struct lu_attr *attr,
1906 struct dt_object_format *dof,
1909 struct lod_thread_info *info = lod_env_info(env);
1910 struct lu_buf lmv_buf;
1911 struct lu_buf slave_lmv_buf;
1912 struct lmv_mds_md_v1 *lmm;
1913 struct lmv_mds_md_v1 *slave_lmm = NULL;
1914 struct dt_insert_rec *rec = &info->lti_dt_rec;
1915 struct lod_object *lo = lod_dt_obj(dt);
1920 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1923 lmm = lmv_buf.lb_buf;
1925 OBD_ALLOC_PTR(slave_lmm);
1926 if (slave_lmm == NULL)
1927 GOTO(out, rc = -ENOMEM);
1929 lod_prep_slave_lmv_md(slave_lmm, lmm);
1930 slave_lmv_buf.lb_buf = slave_lmm;
1931 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1933 if (!dt_try_as_dir(env, dt_object_child(dt), false))
1934 GOTO(out, rc = -EINVAL);
1936 rec->rec_type = S_IFDIR;
1937 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1938 struct dt_object *dto = lo->ldo_stripe[i];
1939 char *stripe_name = info->lti_key;
1940 struct lu_name *sname;
1941 struct linkea_data ldata = { NULL };
1942 struct lu_buf linkea_buf;
1944 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
1948 /* directory split skip create for existing stripes */
1949 if (!(lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) {
1950 rc = lod_sub_declare_create(env, dto, attr, NULL, dof,
1955 if (!dt_try_as_dir(env, dto, false))
1956 GOTO(out, rc = -EINVAL);
1958 rc = lod_sub_declare_ref_add(env, dto, th);
1962 rec->rec_fid = lu_object_fid(&dto->do_lu);
1963 rc = lod_sub_declare_insert(env, dto,
1964 (const struct dt_rec *)rec,
1965 (const struct dt_key *)dot,
1970 /* master stripe FID will be put to .. */
1971 rec->rec_fid = lu_object_fid(&dt->do_lu);
1972 rc = lod_sub_declare_insert(env, dto,
1973 (const struct dt_rec *)rec,
1974 (const struct dt_key *)dotdot,
1979 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1981 snprintf(stripe_name, sizeof(info->lti_key),
1983 PFID(lu_object_fid(&dto->do_lu)),
1986 snprintf(stripe_name, sizeof(info->lti_key),
1988 PFID(lu_object_fid(&dto->do_lu)), i);
1990 sname = lod_name_get(env, stripe_name,
1991 strlen(stripe_name));
1992 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1993 sname, lu_object_fid(&dt->do_lu));
1997 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1998 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1999 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
2000 XATTR_NAME_LINK, 0, th);
2004 rec->rec_fid = lu_object_fid(&dto->do_lu);
2005 rc = lod_sub_declare_insert(env, dt_object_child(dt),
2006 (const struct dt_rec *)rec,
2007 (const struct dt_key *)stripe_name, th);
2011 rc = lod_sub_declare_ref_add(env, dt_object_child(dt),
2017 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
2018 cfs_fail_val != i) {
2019 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
2021 slave_lmm->lmv_master_mdt_index =
2024 slave_lmm->lmv_master_mdt_index =
2026 rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
2027 XATTR_NAME_LMV, 0, th);
2033 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
2034 &lmv_buf, XATTR_NAME_LMV, 0, th);
2038 if (slave_lmm != NULL)
2039 OBD_FREE_PTR(slave_lmm);
2045 * Allocate a striping on a predefined set of MDTs.
2047 * Allocates new striping using the MDT index range provided by the data from
2048 * the lum_obejcts contained in the lmv_user_md passed to this method if
2049 * \a is_specific is true; or allocates new layout starting from MDT index in
2050 * lo->ldo_dir_stripe_offset. The exact order of MDTs is not important and
2051 * varies depending on MDT status. The number of stripes needed and stripe
2052 * offset are taken from the object. If that number cannot be met, then the
2053 * function returns an error and then it's the caller's responsibility to
2054 * release the stripes allocated. All the internal structures are protected,
2055 * but no concurrent allocation is allowed on the same objects.
2057 * \param[in] env execution environment for this thread
2058 * \param[in] lo LOD object
2059 * \param[out] stripes striping created
2060 * \param[out] mdt_indices MDT indices of striping created
2061 * \param[in] is_specific true if the MDTs are provided by lum; false if
2062 * only the starting MDT index is provided
2064 * \retval positive stripes allocated, including the first stripe allocated
2066 * \retval negative errno on failure
2068 static int lod_mdt_alloc_specific(const struct lu_env *env,
2069 struct lod_object *lo,
2070 struct dt_object **stripes,
2071 __u32 *mdt_indices, bool is_specific)
2073 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
2074 struct lu_tgt_descs *ltd = &lod->lod_mdt_descs;
2075 struct lu_tgt_desc *tgt = NULL;
2076 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2077 struct dt_device *tgt_dt = NULL;
2078 struct lu_fid fid = { 0 };
2079 struct dt_object *dto;
2081 u32 stripe_count = lo->ldo_dir_stripe_count;
2087 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
2088 if (!is_specific && stripe_count > 1)
2089 /* Set the start index for the 2nd stripe allocation */
2090 mdt_indices[1] = (mdt_indices[0] + 1) %
2091 (lod->lod_remote_mdt_count + 1);
2093 for (; stripe_idx < stripe_count; stripe_idx++) {
2094 /* Try to find next avaible target */
2095 idx = mdt_indices[stripe_idx];
2096 for (j = 0; j < lod->lod_remote_mdt_count;
2097 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
2098 bool already_allocated = false;
2101 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
2102 idx, lod->lod_remote_mdt_count + 1, stripe_idx);
2104 if (likely(!is_specific &&
2105 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
2106 /* check whether the idx already exists
2107 * in current allocated array */
2108 for (k = 0; k < stripe_idx; k++) {
2109 if (mdt_indices[k] == idx) {
2110 already_allocated = true;
2115 if (already_allocated)
2119 /* Sigh, this index is not in the bitmap, let's check
2120 * next available target */
2121 if (!test_bit(idx, ltd->ltd_tgt_bitmap) &&
2122 idx != master_index)
2125 if (idx == master_index) {
2126 /* Allocate the FID locally */
2127 tgt_dt = lod->lod_child;
2128 rc = dt_fid_alloc(env, tgt_dt, &fid, NULL,
2135 /* check the status of the OSP */
2136 tgt = LTD_TGT(ltd, idx);
2140 tgt_dt = tgt->ltd_tgt;
2141 if (!tgt->ltd_active)
2142 /* this OSP doesn't feel well */
2145 rc = dt_fid_alloc(env, tgt_dt, &fid, NULL, NULL);
2152 /* Can not allocate more stripes */
2153 if (j == lod->lod_remote_mdt_count) {
2154 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
2155 lod2obd(lod)->obd_name, stripe_count,
2160 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
2161 idx, stripe_idx, PFID(&fid));
2162 mdt_indices[stripe_idx] = idx;
2163 /* Set the start index for next stripe allocation */
2164 if (!is_specific && stripe_idx < stripe_count - 1) {
2166 * for large dir test, put all other slaves on one
2167 * remote MDT, otherwise we may save too many local
2168 * slave locks which will exceed RS_MAX_LOCKS.
2170 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)))
2172 mdt_indices[stripe_idx + 1] = (idx + 1) %
2173 (lod->lod_remote_mdt_count + 1);
2175 /* tgt_dt and fid must be ready after search avaible OSP
2176 * in the above loop */
2177 LASSERT(tgt_dt != NULL);
2178 LASSERT(fid_is_sane(&fid));
2180 /* fail a remote stripe FID allocation */
2181 if (stripe_idx && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_FID))
2184 dto = dt_locate_at(env, tgt_dt, &fid,
2185 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
2192 stripes[stripe_idx] = dto;
2198 for (j = 1; j < stripe_idx; j++) {
2199 LASSERT(stripes[j] != NULL);
2200 dt_object_put(env, stripes[j]);
2206 static int lod_prep_md_striped_create(const struct lu_env *env,
2207 struct dt_object *dt,
2208 struct lu_attr *attr,
2209 const struct lmv_user_md_v1 *lum,
2210 struct dt_object_format *dof,
2213 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
2214 struct lod_object *lo = lod_dt_obj(dt);
2215 struct dt_object **stripes;
2216 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2217 struct lu_fid fid = { 0 };
2224 /* The lum has been verifed in lod_verify_md_striping */
2225 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
2226 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
2228 stripe_count = lo->ldo_dir_stripe_count;
2230 OBD_ALLOC_PTR_ARRAY(stripes, stripe_count);
2234 /* Allocate the first stripe locally */
2235 rc = dt_fid_alloc(env, lod->lod_child, &fid, NULL, NULL);
2239 stripes[0] = dt_locate_at(env, lod->lod_child, &fid,
2240 dt->do_lu.lo_dev->ld_site->ls_top_dev, &conf);
2241 if (IS_ERR(stripes[0]))
2242 GOTO(out, rc = PTR_ERR(stripes[0]));
2244 if (lo->ldo_dir_stripe_offset == LMV_OFFSET_DEFAULT) {
2245 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
2246 rc = lod_mdt_alloc_qos(env, lo, stripes, 1, stripe_count);
2248 rc = lod_mdt_alloc_rr(env, lo, stripes, 1,
2252 bool is_specific = false;
2254 OBD_ALLOC_PTR_ARRAY(idx_array, stripe_count);
2256 GOTO(out, rc = -ENOMEM);
2258 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
2260 for (i = 0; i < stripe_count; i++)
2262 le32_to_cpu(lum->lum_objects[i].lum_mds);
2265 /* stripe 0 is local */
2267 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
2268 rc = lod_mdt_alloc_specific(env, lo, stripes, idx_array,
2270 OBD_FREE_PTR_ARRAY(idx_array, stripe_count);
2278 lo->ldo_dir_striped = 1;
2279 lo->ldo_stripe = stripes;
2280 lo->ldo_dir_stripe_count = rc;
2281 lo->ldo_dir_stripes_allocated = stripe_count;
2283 lo->ldo_dir_stripe_loaded = 1;
2285 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2287 lod_striping_free(env, lo);
2293 if (!IS_ERR_OR_NULL(stripes[0]))
2294 dt_object_put(env, stripes[0]);
2295 for (i = 1; i < stripe_count; i++)
2296 LASSERT(!stripes[i]);
2297 OBD_FREE_PTR_ARRAY(stripes, stripe_count);
2304 * Alloc cached foreign LOV
2306 * \param[in] lo object
2307 * \param[in] size size of foreign LOV
2309 * \retval 0 on success
2310 * \retval negative if failed
2312 int lod_alloc_foreign_lov(struct lod_object *lo, size_t size)
2314 OBD_ALLOC_LARGE(lo->ldo_foreign_lov, size);
2315 if (lo->ldo_foreign_lov == NULL)
2317 lo->ldo_foreign_lov_size = size;
2318 lo->ldo_is_foreign = 1;
2324 * Free cached foreign LOV
2326 * \param[in] lo object
2328 void lod_free_foreign_lov(struct lod_object *lo)
2330 if (lo->ldo_foreign_lov != NULL)
2331 OBD_FREE_LARGE(lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
2332 lo->ldo_foreign_lov = NULL;
2333 lo->ldo_foreign_lov_size = 0;
2334 lo->ldo_is_foreign = 0;
2339 * Alloc cached foreign LMV
2341 * \param[in] lo object
2342 * \param[in] size size of foreign LMV
2344 * \retval 0 on success
2345 * \retval negative if failed
2347 int lod_alloc_foreign_lmv(struct lod_object *lo, size_t size)
2349 OBD_ALLOC_LARGE(lo->ldo_foreign_lmv, size);
2350 if (lo->ldo_foreign_lmv == NULL)
2352 lo->ldo_foreign_lmv_size = size;
2353 lo->ldo_is_foreign = 1;
2360 * Free cached foreign LMV
2362 * \param[in] lo object
2364 void lod_free_foreign_lmv(struct lod_object *lo)
2366 if (lo->ldo_foreign_lmv != NULL)
2367 OBD_FREE_LARGE(lo->ldo_foreign_lmv, lo->ldo_foreign_lmv_size);
2368 lo->ldo_foreign_lmv = NULL;
2369 lo->ldo_foreign_lmv_size = 0;
2370 lo->ldo_is_foreign = 0;
2374 * Declare create striped md object.
2376 * The function declares intention to create a striped directory. This is a
2377 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2378 * is to verify pattern \a lum_buf is good. Check that function for the details.
2380 * \param[in] env execution environment
2381 * \param[in] dt object
2382 * \param[in] attr attributes to initialize the objects with
2383 * \param[in] lum_buf a pattern specifying the number of stripes and
2385 * \param[in] dof type of objects to be created
2386 * \param[in] th transaction handle
2388 * \retval 0 on success
2389 * \retval negative if failed
2392 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2393 struct dt_object *dt,
2394 struct lu_attr *attr,
2395 const struct lu_buf *lum_buf,
2396 struct dt_object_format *dof,
2399 struct lod_object *lo = lod_dt_obj(dt);
2400 struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
2404 LASSERT(lum != NULL);
2407 "lum magic=%x hash=%x count=%u offset=%d inherit=%u rr=%u\n",
2408 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_hash_type),
2409 le32_to_cpu(lum->lum_stripe_count),
2410 (int)le32_to_cpu(lum->lum_stripe_offset),
2411 lum->lum_max_inherit, lum->lum_max_inherit_rr);
2413 if (lo->ldo_dir_stripe_count == 0) {
2414 if (lo->ldo_is_foreign) {
2415 rc = lod_alloc_foreign_lmv(lo, lum_buf->lb_len);
2418 memcpy(lo->ldo_foreign_lmv, lum, lum_buf->lb_len);
2419 lo->ldo_dir_stripe_loaded = 1;
2424 /* prepare dir striped objects */
2425 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2427 /* failed to create striping, let's reset
2428 * config so that others don't get confused */
2429 lod_striping_free(env, lo);
2437 * Set or replace striped directory layout, and LFSCK may set layout on a plain
2438 * directory, so don't check stripe count.
2440 * \param[in] env execution environment
2441 * \param[in] dt target object
2442 * \param[in] lmv_buf LMV buf which contains source stripe FIDs
2443 * \param[in] fl set or replace
2444 * \param[in] th transaction handle
2446 * \retval 0 on success
2447 * \retval negative if failed
2449 static int lod_dir_layout_set(const struct lu_env *env,
2450 struct dt_object *dt,
2451 const struct lu_buf *lmv_buf,
2455 struct dt_object *next = dt_object_child(dt);
2456 struct lod_object *lo = lod_dt_obj(dt);
2457 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2458 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
2459 struct lmv_mds_md_v1 *slave_lmv;
2460 struct lu_buf slave_buf;
2466 if (!lmv_is_sane2(lmv))
2469 /* adjust hash for dir merge, which may not be set in user command */
2470 if (lmv_is_merging(lmv) &&
2471 !(lmv->lmv_migrate_hash & LMV_HASH_TYPE_MASK))
2472 lmv->lmv_merge_hash |=
2473 lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern &
2476 LMV_DEBUG(D_INFO, lmv, "set");
2478 rc = lod_sub_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV, fl, th);
2482 /* directory restripe may update stripe LMV directly */
2483 if (!lo->ldo_dir_stripe_count)
2486 lo->ldo_dir_hash_type = le32_to_cpu(lmv->lmv_hash_type);
2487 lo->ldo_dir_migrate_offset = le32_to_cpu(lmv->lmv_migrate_offset);
2488 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_migrate_hash);
2489 lo->ldo_dir_layout_version = le32_to_cpu(lmv->lmv_layout_version);
2491 OBD_ALLOC_PTR(slave_lmv);
2495 lod_prep_slave_lmv_md(slave_lmv, lmv);
2496 slave_buf.lb_buf = slave_lmv;
2497 slave_buf.lb_len = sizeof(*slave_lmv);
2499 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2500 if (!lo->ldo_stripe[i])
2503 if (!dt_object_exists(lo->ldo_stripe[i]))
2506 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], &slave_buf,
2507 XATTR_NAME_LMV, fl, th);
2512 OBD_FREE_PTR(slave_lmv);
2518 * Implementation of dt_object_operations::do_declare_xattr_set.
2520 * Used with regular (non-striped) objects. Basically it
2521 * initializes the striping information and applies the
2522 * change to all the stripes.
2524 * \see dt_object_operations::do_declare_xattr_set() in the API description
2527 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2528 struct dt_object *dt,
2529 const struct lu_buf *buf,
2530 const char *name, int fl,
2533 struct dt_object *next = dt_object_child(dt);
2534 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2535 struct lod_object *lo = lod_dt_obj(dt);
2540 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2541 struct lmv_user_md_v1 *lum;
2543 LASSERT(buf != NULL && buf->lb_buf != NULL);
2545 rc = lod_verify_md_striping(d, lum);
2548 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2549 rc = lod_verify_striping(env, d, lo, buf, false);
2554 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2558 /* Note: Do not set LinkEA on sub-stripes, otherwise
2559 * it will confuse the fid2path process(see mdt_path_current()).
2560 * The linkEA between master and sub-stripes is set in
2561 * lod_xattr_set_lmv(). */
2562 if (strcmp(name, XATTR_NAME_LINK) == 0)
2565 /* set xattr to each stripes, if needed */
2566 rc = lod_striping_load(env, lo);
2570 if (lo->ldo_dir_stripe_count == 0)
2573 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2574 if (!lo->ldo_stripe[i])
2577 if (!dt_object_exists(lo->ldo_stripe[i]))
2580 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2590 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2591 struct lod_object *lo,
2592 struct dt_object *dt, struct thandle *th,
2593 int comp_idx, int stripe_idx,
2594 struct lod_obj_stripe_cb_data *data)
2596 struct lod_thread_info *info = lod_env_info(env);
2597 struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
2598 struct filter_fid *ff = &info->lti_ff;
2599 struct lu_buf *buf = &info->lti_buf;
2603 buf->lb_len = sizeof(*ff);
2604 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2612 * locd_buf is set if it's called by dir migration, which doesn't check
2615 if (data->locd_buf) {
2616 memset(ff, 0, sizeof(*ff));
2617 ff->ff_parent = *(struct lu_fid *)data->locd_buf->lb_buf;
2619 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
2621 if (lu_fid_eq(lod_object_fid(lo), &ff->ff_parent) &&
2622 ff->ff_layout.ol_comp_id == comp->llc_id)
2625 memset(ff, 0, sizeof(*ff));
2626 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2629 /* rewrite filter_fid */
2630 ff->ff_parent.f_ver = stripe_idx;
2631 ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
2632 ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
2633 ff->ff_layout.ol_comp_id = comp->llc_id;
2634 ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
2635 ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
2636 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
2638 if (data->locd_declare)
2639 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2640 LU_XATTR_REPLACE, th);
2642 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2643 LU_XATTR_REPLACE, th);
2649 * Reset parent FID on OST object
2651 * Replace parent FID with @dt object FID, which is only called during migration
2652 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2653 * the FID is changed.
2655 * \param[in] env execution environment
2656 * \param[in] dt dt_object whose stripes's parent FID will be reset
2657 * \parem[in] th thandle
2658 * \param[in] declare if it is declare
2660 * \retval 0 if reset succeeds
2661 * \retval negative errno if reset fails
2663 static int lod_replace_parent_fid(const struct lu_env *env,
2664 struct dt_object *dt,
2665 const struct lu_buf *buf,
2666 struct thandle *th, bool declare)
2668 struct lod_object *lo = lod_dt_obj(dt);
2669 struct lod_thread_info *info = lod_env_info(env);
2670 struct filter_fid *ff;
2671 struct lod_obj_stripe_cb_data data = { { 0 } };
2675 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2677 /* set xattr to each stripes, if needed */
2678 rc = lod_striping_load(env, lo);
2682 if (!lod_obj_is_striped(dt))
2685 if (info->lti_ea_store_size < sizeof(*ff)) {
2686 rc = lod_ea_store_resize(info, sizeof(*ff));
2691 data.locd_declare = declare;
2692 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2693 data.locd_buf = buf;
2694 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2699 __u16 lod_comp_entry_stripe_count(struct lod_object *lo,
2700 int comp_idx, bool is_dir)
2702 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2703 struct lod_layout_component *entry;
2708 entry = &lo->ldo_comp_entries[comp_idx];
2709 if (lod_comp_inited(entry))
2710 return entry->llc_stripe_count;
2711 else if ((__u16)-1 == entry->llc_stripe_count)
2712 return lod->lod_ost_count;
2714 return lod_get_stripe_count(lod, lo, comp_idx,
2715 entry->llc_stripe_count,
2716 entry->llc_pattern &
2717 LOV_PATTERN_OVERSTRIPING);
2720 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2722 int magic, size = 0, i;
2723 struct lod_layout_component *comp_entries;
2725 bool is_composite, is_foreign = false;
2728 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2729 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2731 lo->ldo_def_striping->lds_def_striping_is_composite;
2733 comp_cnt = lo->ldo_comp_cnt;
2734 comp_entries = lo->ldo_comp_entries;
2735 is_composite = lo->ldo_is_composite;
2736 is_foreign = lo->ldo_is_foreign;
2740 return lo->ldo_foreign_lov_size;
2742 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2744 size = sizeof(struct lov_comp_md_v1) +
2745 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2746 LASSERT(size % sizeof(__u64) == 0);
2749 for (i = 0; i < comp_cnt; i++) {
2752 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2753 stripe_count = lod_comp_entry_stripe_count(lo, i, is_dir);
2754 if (!is_dir && is_composite)
2755 lod_comp_shrink_stripe_count(&comp_entries[i],
2758 size += lov_user_md_size(stripe_count, magic);
2759 LASSERT(size % sizeof(__u64) == 0);
2765 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2766 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2769 * \param[in] env execution environment
2770 * \param[in] dt dt_object to add components on
2771 * \param[in] buf buffer contains components to be added
2772 * \parem[in] th thandle
2774 * \retval 0 on success
2775 * \retval negative errno on failure
2777 static int lod_declare_layout_add(const struct lu_env *env,
2778 struct dt_object *dt,
2779 const struct lu_buf *buf,
2782 struct lod_thread_info *info = lod_env_info(env);
2783 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2784 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2785 struct dt_object *next = dt_object_child(dt);
2786 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
2787 struct lod_object *lo = lod_dt_obj(dt);
2788 struct lov_user_md_v3 *v3;
2789 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2791 int i, rc, array_cnt, old_array_cnt;
2794 LASSERT(lo->ldo_is_composite);
2796 if (lo->ldo_flr_state != LCM_FL_NONE)
2799 rc = lod_verify_striping(env, d, lo, buf, false);
2803 magic = comp_v1->lcm_magic;
2804 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2805 lustre_swab_lov_comp_md_v1(comp_v1);
2806 magic = comp_v1->lcm_magic;
2809 if (magic != LOV_USER_MAGIC_COMP_V1)
2812 mutex_lock(&lo->ldo_layout_mutex);
2814 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2815 OBD_ALLOC_PTR_ARRAY(comp_array, array_cnt);
2816 if (comp_array == NULL) {
2817 mutex_unlock(&lo->ldo_layout_mutex);
2822 memcpy(comp_array, lo->ldo_comp_entries,
2823 sizeof(*comp_array) * lo->ldo_comp_cnt);
2825 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2826 struct lov_user_md_v1 *v1;
2827 struct lu_extent *ext;
2829 v1 = (struct lov_user_md *)((char *)comp_v1 +
2830 comp_v1->lcm_entries[i].lcme_offset);
2831 ext = &comp_v1->lcm_entries[i].lcme_extent;
2833 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2834 lod_comp->llc_extent.e_start = ext->e_start;
2835 lod_comp->llc_extent.e_end = ext->e_end;
2836 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2837 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2839 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2840 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2841 lod_adjust_stripe_info(lod_comp, desc, 0);
2843 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2844 v3 = (struct lov_user_md_v3 *) v1;
2845 if (v3->lmm_pool_name[0] != '\0') {
2846 rc = lod_set_pool(&lod_comp->llc_pool,
2854 old_array = lo->ldo_comp_entries;
2855 old_array_cnt = lo->ldo_comp_cnt;
2857 lo->ldo_comp_entries = comp_array;
2858 lo->ldo_comp_cnt = array_cnt;
2860 /* No need to increase layout generation here, it will be increased
2861 * later when generating component ID for the new components */
2863 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2864 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2865 XATTR_NAME_LOV, 0, th);
2867 lo->ldo_comp_entries = old_array;
2868 lo->ldo_comp_cnt = old_array_cnt;
2872 OBD_FREE_PTR_ARRAY(old_array, old_array_cnt);
2874 LASSERT(lo->ldo_mirror_count == 1);
2875 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2877 mutex_unlock(&lo->ldo_layout_mutex);
2882 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2883 lod_comp = &comp_array[i];
2884 if (lod_comp->llc_pool != NULL) {
2885 OBD_FREE(lod_comp->llc_pool,
2886 strlen(lod_comp->llc_pool) + 1);
2887 lod_comp->llc_pool = NULL;
2890 OBD_FREE_PTR_ARRAY(comp_array, array_cnt);
2891 mutex_unlock(&lo->ldo_layout_mutex);
2897 * lod_last_non_stale_mirror() - Check if a mirror is the last non-stale mirror.
2898 * @mirror_id: Mirror id to be checked.
2901 * This function checks if a mirror with specified @mirror_id is the last
2902 * non-stale mirror of a LOD object @lo.
2904 * Return: true or false.
2907 bool lod_last_non_stale_mirror(__u16 mirror_id, struct lod_object *lo)
2909 struct lod_layout_component *lod_comp;
2910 bool has_stale_flag;
2913 for (i = 0; i < lo->ldo_mirror_count; i++) {
2914 if (lo->ldo_mirrors[i].lme_id == mirror_id ||
2915 lo->ldo_mirrors[i].lme_stale)
2918 has_stale_flag = false;
2919 lod_foreach_mirror_comp(lod_comp, lo, i) {
2920 if (lod_comp->llc_flags & LCME_FL_STALE) {
2921 has_stale_flag = true;
2925 if (!has_stale_flag)
2933 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2934 * the '$field' can only be 'flags' now. The xattr value is binary
2935 * lov_comp_md_v1 which contains the component ID(s) and the value of
2936 * the field to be modified.
2937 * Please update allowed_lustre_lov macro if $field groks more values
2940 * \param[in] env execution environment
2941 * \param[in] dt dt_object to be modified
2942 * \param[in] op operation string, like "set.flags"
2943 * \param[in] buf buffer contains components to be set
2944 * \parem[in] th thandle
2946 * \retval 0 on success
2947 * \retval negative errno on failure
2949 static int lod_declare_layout_set(const struct lu_env *env,
2950 struct dt_object *dt,
2951 char *op, const struct lu_buf *buf,
2954 struct lod_layout_component *lod_comp;
2955 struct lod_thread_info *info = lod_env_info(env);
2956 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2957 struct lod_object *lo = lod_dt_obj(dt);
2958 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2961 bool changed = false;
2964 /* Please update allowed_lustre_lov macro if op
2965 * groks more values in the future
2967 if (strcmp(op, "set.flags") != 0) {
2968 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2969 lod2obd(d)->obd_name, op);
2973 magic = comp_v1->lcm_magic;
2974 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2975 lustre_swab_lov_comp_md_v1(comp_v1);
2976 magic = comp_v1->lcm_magic;
2979 if (magic != LOV_USER_MAGIC_COMP_V1)
2982 if (comp_v1->lcm_entry_count == 0) {
2983 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2984 lod2obd(d)->obd_name);
2988 mutex_lock(&lo->ldo_layout_mutex);
2989 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2990 __u32 id = comp_v1->lcm_entries[i].lcme_id;
2991 __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
2992 __u32 mirror_flag = flags & LCME_MIRROR_FLAGS;
2993 __u16 mirror_id = mirror_id_of(id);
2994 bool neg = flags & LCME_FL_NEG;
2996 if (flags & LCME_FL_INIT) {
2998 lod_striping_free_nolock(env, lo);
2999 mutex_unlock(&lo->ldo_layout_mutex);
3003 flags &= ~(LCME_MIRROR_FLAGS | LCME_FL_NEG);
3004 for (j = 0; j < lo->ldo_comp_cnt; j++) {
3005 lod_comp = &lo->ldo_comp_entries[j];
3007 /* lfs only put one flag in each entry */
3008 if ((flags && id != lod_comp->llc_id) ||
3009 (mirror_flag && mirror_id !=
3010 mirror_id_of(lod_comp->llc_id)))
3015 lod_comp->llc_flags &= ~flags;
3017 lod_comp->llc_flags &= ~mirror_flag;
3020 if ((flags & LCME_FL_STALE) &&
3021 lod_last_non_stale_mirror(mirror_id,
3024 &lo->ldo_layout_mutex);
3027 lod_comp->llc_flags |= flags;
3030 lod_comp->llc_flags |= mirror_flag;
3031 if (mirror_flag & LCME_FL_NOSYNC)
3032 lod_comp->llc_timestamp =
3033 ktime_get_real_seconds();
3039 mutex_unlock(&lo->ldo_layout_mutex);
3042 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
3043 lod2obd(d)->obd_name);
3047 lod_obj_inc_layout_gen(lo);
3049 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3050 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
3051 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3056 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
3057 * and the xattr value is a unique component ID or a special lcme_id.
3059 * \param[in] env execution environment
3060 * \param[in] dt dt_object to be operated on
3061 * \param[in] buf buffer contains component ID or lcme_id
3062 * \parem[in] th thandle
3064 * \retval 0 on success
3065 * \retval negative errno on failure
3067 static int lod_declare_layout_del(const struct lu_env *env,
3068 struct dt_object *dt,
3069 const struct lu_buf *buf,
3072 struct lod_thread_info *info = lod_env_info(env);
3073 struct dt_object *next = dt_object_child(dt);
3074 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3075 struct lod_object *lo = lod_dt_obj(dt);
3076 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3077 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3078 __u32 magic, id, flags, neg_flags = 0;
3082 LASSERT(lo->ldo_is_composite);
3084 if (lo->ldo_flr_state != LCM_FL_NONE)
3087 magic = comp_v1->lcm_magic;
3088 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
3089 lustre_swab_lov_comp_md_v1(comp_v1);
3090 magic = comp_v1->lcm_magic;
3093 if (magic != LOV_USER_MAGIC_COMP_V1)
3096 id = comp_v1->lcm_entries[0].lcme_id;
3097 flags = comp_v1->lcm_entries[0].lcme_flags;
3099 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
3100 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
3101 lod2obd(d)->obd_name, id, flags);
3105 if (id != LCME_ID_INVAL && flags != 0) {
3106 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
3107 lod2obd(d)->obd_name);
3111 if (id == LCME_ID_INVAL && !flags) {
3112 CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
3113 lod2obd(d)->obd_name);
3117 if (flags & LCME_FL_NEG) {
3118 neg_flags = flags & ~LCME_FL_NEG;
3122 mutex_lock(&lo->ldo_layout_mutex);
3124 left = lo->ldo_comp_cnt;
3126 mutex_unlock(&lo->ldo_layout_mutex);
3130 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3131 struct lod_layout_component *lod_comp;
3133 lod_comp = &lo->ldo_comp_entries[i];
3135 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
3137 else if (flags && !(flags & lod_comp->llc_flags))
3139 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
3142 if (left != (i + 1)) {
3143 CDEBUG(D_LAYOUT, "%s: this deletion will create "
3144 "a hole.\n", lod2obd(d)->obd_name);
3145 mutex_unlock(&lo->ldo_layout_mutex);
3150 /* Mark the component as deleted */
3151 lod_comp->llc_id = LCME_ID_INVAL;
3153 /* Not instantiated component */
3154 if (lod_comp->llc_stripe == NULL)
3157 LASSERT(lod_comp->llc_stripe_count > 0);
3158 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
3159 struct dt_object *obj = lod_comp->llc_stripe[j];
3163 rc = lod_sub_declare_destroy(env, obj, th);
3165 mutex_unlock(&lo->ldo_layout_mutex);
3171 LASSERTF(left >= 0, "left = %d\n", left);
3172 if (left == lo->ldo_comp_cnt) {
3173 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
3174 lod2obd(d)->obd_name, id);
3175 mutex_unlock(&lo->ldo_layout_mutex);
3179 mutex_unlock(&lo->ldo_layout_mutex);
3181 memset(attr, 0, sizeof(*attr));
3182 attr->la_valid = LA_SIZE;
3183 rc = lod_sub_declare_attr_set(env, next, attr, th);
3188 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3189 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
3190 XATTR_NAME_LOV, 0, th);
3192 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
3199 * Declare layout add/set/del operations issued by special xattr names:
3201 * XATTR_LUSTRE_LOV.add add component(s) to existing file
3202 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
3203 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
3205 * \param[in] env execution environment
3206 * \param[in] dt object
3207 * \param[in] name name of xattr
3208 * \param[in] buf lu_buf contains xattr value
3209 * \param[in] th transaction handle
3211 * \retval 0 on success
3212 * \retval negative if failed
3214 static int lod_declare_modify_layout(const struct lu_env *env,
3215 struct dt_object *dt,
3217 const struct lu_buf *buf,
3220 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3221 struct lod_object *lo = lod_dt_obj(dt);
3223 int rc, len = strlen(XATTR_LUSTRE_LOV);
3226 LASSERT(dt_object_exists(dt));
3228 if (strlen(name) <= len || name[len] != '.') {
3229 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
3230 lod2obd(d)->obd_name, name);
3235 rc = lod_striping_load(env, lo);
3239 /* the layout to be modified must be a composite layout */
3240 if (!lo->ldo_is_composite) {
3241 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
3242 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3243 GOTO(unlock, rc = -EINVAL);
3246 op = (char *)name + len;
3247 if (strcmp(op, "add") == 0) {
3248 rc = lod_declare_layout_add(env, dt, buf, th);
3249 } else if (strcmp(op, "del") == 0) {
3250 rc = lod_declare_layout_del(env, dt, buf, th);
3251 } else if (strncmp(op, "set", strlen("set")) == 0) {
3252 rc = lod_declare_layout_set(env, dt, op, buf, th);
3254 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
3255 lod2obd(d)->obd_name, name);
3256 GOTO(unlock, rc = -ENOTSUPP);
3260 lod_striping_free(env, lo);
3266 * Convert a plain file lov_mds_md to a composite layout.
3268 * \param[in,out] info the thread info::lti_ea_store buffer contains little
3269 * endian plain file layout
3271 * \retval 0 on success, <0 on failure
3273 static int lod_layout_convert(struct lod_thread_info *info)
3275 struct lov_mds_md *lmm = info->lti_ea_store;
3276 struct lov_mds_md *lmm_save;
3277 struct lov_comp_md_v1 *lcm;
3278 struct lov_comp_md_entry_v1 *lcme;
3284 /* realloc buffer to a composite layout which contains one component */
3285 blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3286 le32_to_cpu(lmm->lmm_magic));
3287 size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
3289 OBD_ALLOC_LARGE(lmm_save, blob_size);
3291 GOTO(out, rc = -ENOMEM);
3293 memcpy(lmm_save, lmm, blob_size);
3295 if (info->lti_ea_store_size < size) {
3296 rc = lod_ea_store_resize(info, size);
3301 lcm = info->lti_ea_store;
3302 memset(lcm, 0, sizeof(*lcm) + sizeof(*lcme));
3303 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
3304 lcm->lcm_size = cpu_to_le32(size);
3305 lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
3306 lmm_save->lmm_layout_gen));
3307 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
3308 lcm->lcm_entry_count = cpu_to_le16(1);
3310 lcme = &lcm->lcm_entries[0];
3311 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
3312 lcme->lcme_extent.e_start = 0;
3313 lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
3314 lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
3315 lcme->lcme_size = cpu_to_le32(blob_size);
3317 memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
3322 OBD_FREE_LARGE(lmm_save, blob_size);
3327 * Merge layouts to form a mirrored file.
3329 static int lod_declare_layout_merge(const struct lu_env *env,
3330 struct dt_object *dt, const struct lu_buf *mbuf,
3333 struct lod_thread_info *info = lod_env_info(env);
3334 struct lu_attr *layout_attr = &info->lti_layout_attr;
3335 struct lu_buf *buf = &info->lti_buf;
3336 struct lod_object *lo = lod_dt_obj(dt);
3337 struct lov_comp_md_v1 *lcm;
3338 struct lov_comp_md_v1 *cur_lcm;
3339 struct lov_comp_md_v1 *merge_lcm;
3340 struct lov_comp_md_entry_v1 *lcme;
3341 struct lov_mds_md_v1 *lmm;
3344 __u16 cur_entry_count;
3345 __u16 merge_entry_count;
3347 __u16 mirror_id = 0;
3354 merge_lcm = mbuf->lb_buf;
3355 if (mbuf->lb_len < sizeof(*merge_lcm))
3358 /* must be an existing layout from disk */
3359 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
3362 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
3364 /* do not allow to merge two mirrored files */
3365 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
3368 /* verify the target buffer */
3369 rc = lod_get_lov_ea(env, lo);
3371 RETURN(rc ? : -ENODATA);
3373 cur_lcm = info->lti_ea_store;
3374 switch (le32_to_cpu(cur_lcm->lcm_magic)) {
3377 rc = lod_layout_convert(info);
3379 case LOV_MAGIC_COMP_V1:
3389 /* info->lti_ea_store could be reallocated in lod_layout_convert() */
3390 cur_lcm = info->lti_ea_store;
3391 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
3393 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
3394 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
3395 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
3398 /* size of new layout */
3399 size = le32_to_cpu(cur_lcm->lcm_size) +
3400 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
3402 memset(buf, 0, sizeof(*buf));
3403 lu_buf_alloc(buf, size);
3404 if (buf->lb_buf == NULL)
3408 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
3410 offset = sizeof(*lcm) +
3411 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
3412 for (i = 0; i < cur_entry_count; i++) {
3413 struct lov_comp_md_entry_v1 *cur_lcme;
3415 lcme = &lcm->lcm_entries[i];
3416 cur_lcme = &cur_lcm->lcm_entries[i];
3418 lcme->lcme_offset = cpu_to_le32(offset);
3419 memcpy((char *)lcm + offset,
3420 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
3421 le32_to_cpu(lcme->lcme_size));
3423 offset += le32_to_cpu(lcme->lcme_size);
3425 if (mirror_count == 1 &&
3426 mirror_id_of(le32_to_cpu(lcme->lcme_id)) == 0) {
3427 /* Add mirror from a non-flr file, create new mirror ID.
3428 * Otherwise, keep existing mirror's component ID, used
3429 * for mirror extension.
3431 id = pflr_id(1, i + 1);
3432 lcme->lcme_id = cpu_to_le32(id);
3435 id = max(le32_to_cpu(lcme->lcme_id), id);
3438 mirror_id = mirror_id_of(id) + 1;
3440 /* check if first entry in new layout is DOM */
3441 lmm = (struct lov_mds_md_v1 *)((char *)merge_lcm +
3442 merge_lcm->lcm_entries[0].lcme_offset);
3443 merge_has_dom = lov_pattern(le32_to_cpu(lmm->lmm_pattern)) ==
3446 for (i = 0; i < merge_entry_count; i++) {
3447 struct lov_comp_md_entry_v1 *merge_lcme;
3449 merge_lcme = &merge_lcm->lcm_entries[i];
3450 lcme = &lcm->lcm_entries[cur_entry_count + i];
3452 *lcme = *merge_lcme;
3453 lcme->lcme_offset = cpu_to_le32(offset);
3454 if (merge_has_dom && i == 0)
3455 lcme->lcme_flags |= cpu_to_le32(LCME_FL_STALE);
3457 id = pflr_id(mirror_id, i + 1);
3458 lcme->lcme_id = cpu_to_le32(id);
3460 memcpy((char *)lcm + offset,
3461 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
3462 le32_to_cpu(lcme->lcme_size));
3464 offset += le32_to_cpu(lcme->lcme_size);
3467 /* fixup layout information */
3468 lcm->lcm_size = cpu_to_le32(size);
3469 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
3470 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
3471 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NONE)
3472 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
3474 rc = lod_striping_reload(env, lo, buf, 0);
3478 lod_obj_inc_layout_gen(lo);
3479 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3481 /* transfer layout version to OST objects. */
3482 if (lo->ldo_mirror_count > 1) {
3483 struct lod_obj_stripe_cb_data data = { {0} };
3485 layout_attr->la_valid = LA_LAYOUT_VERSION;
3486 layout_attr->la_layout_version = 0;
3487 data.locd_attr = layout_attr;
3488 data.locd_declare = true;
3489 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
3490 rc = lod_obj_for_each_stripe(env, lo, th, &data);
3495 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
3496 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3504 * Split layouts, just set the LOVEA with the layout from mbuf.
3506 static int lod_declare_layout_split(const struct lu_env *env,
3507 struct dt_object *dt, const struct lu_buf *mbuf,
3510 struct lod_thread_info *info = lod_env_info(env);
3511 struct lu_attr *layout_attr = &info->lti_layout_attr;
3512 struct lod_object *lo = lod_dt_obj(dt);
3513 struct lov_comp_md_v1 *lcm = mbuf->lb_buf;
3517 rc = lod_striping_reload(env, lo, mbuf, LVF_ALL_STALE);
3521 lod_obj_inc_layout_gen(lo);
3522 /* fix on-disk layout gen */
3523 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3526 /* transfer layout version to OST objects. */
3527 if (lo->ldo_mirror_count > 1) {
3528 struct lod_obj_stripe_cb_data data = { {0} };
3530 layout_attr->la_valid = LA_LAYOUT_VERSION;
3531 layout_attr->la_layout_version = 0;
3532 data.locd_attr = layout_attr;
3533 data.locd_declare = true;
3534 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
3535 rc = lod_obj_for_each_stripe(env, lo, th, &data);
3540 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), mbuf,
3541 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3545 static int lod_layout_declare_or_purge_mirror(const struct lu_env *env,
3546 struct dt_object *dt, const struct lu_buf *buf,
3547 struct thandle *th, bool declare)
3549 struct lod_thread_info *info = lod_env_info(env);
3550 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3551 struct lod_object *lo = lod_dt_obj(dt);
3552 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3553 struct lov_comp_md_entry_v1 *entry;
3554 struct lov_mds_md_v1 *lmm;
3555 struct dt_object **sub_objs = NULL;
3556 int rc = 0, i, k, array_count = 0;
3561 * other ops (like lod_declare_destroy) could destroying sub objects
3564 mutex_lock(&lo->ldo_layout_mutex);
3567 /* prepare sub-objects array */
3568 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
3569 entry = &comp_v1->lcm_entries[i];
3571 if (!(entry->lcme_flags & LCME_FL_INIT))
3574 lmm = (struct lov_mds_md_v1 *)
3575 ((char *)comp_v1 + entry->lcme_offset);
3576 array_count += lmm->lmm_stripe_count;
3578 OBD_ALLOC_PTR_ARRAY(sub_objs, array_count);
3579 if (sub_objs == NULL) {
3580 mutex_unlock(&lo->ldo_layout_mutex);
3585 k = 0; /* sub_objs index */
3586 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
3587 struct lov_ost_data_v1 *objs;
3588 struct lu_object *o, *n;
3589 struct dt_object *dto;
3590 struct lu_device *nd;
3591 struct lov_mds_md_v3 *v3;
3595 entry = &comp_v1->lcm_entries[i];
3597 if (!(entry->lcme_flags & LCME_FL_INIT))
3600 lmm = (struct lov_mds_md_v1 *)
3601 ((char *)comp_v1 + entry->lcme_offset);
3602 v3 = (struct lov_mds_md_v3 *)lmm;
3603 if (lmm->lmm_magic == LOV_MAGIC_V3)
3604 objs = &v3->lmm_objects[0];
3606 objs = &lmm->lmm_objects[0];
3608 for (j = 0; j < lmm->lmm_stripe_count; j++) {
3609 idx = objs[j].l_ost_idx;
3610 rc = ostid_to_fid(&info->lti_fid, &objs[j].l_ost_oi,
3615 if (!fid_is_sane(&info->lti_fid)) {
3616 CERROR("%s: sub-object insane fid "DFID"\n",
3617 lod2obd(d)->obd_name,
3618 PFID(&info->lti_fid));
3619 GOTO(out, rc = -EINVAL);
3622 lod_getref(&d->lod_ost_descs);
3624 rc = validate_lod_and_idx(d, idx);
3626 lod_putref(d, &d->lod_ost_descs);
3630 nd = &OST_TGT(d, idx)->ltd_tgt->dd_lu_dev;
3631 lod_putref(d, &d->lod_ost_descs);
3633 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
3635 GOTO(out, rc = PTR_ERR(o));
3637 n = lu_object_locate(o->lo_header, nd->ld_type);
3639 lu_object_put(env, n);
3640 GOTO(out, rc = -ENOENT);
3643 dto = container_of(n, struct dt_object, do_lu);
3646 rc = lod_sub_declare_destroy(env, dto, th);
3647 dt_object_put(env, dto);
3652 * collect to-be-destroyed sub objects, the
3653 * reference would be released after actual
3659 } /* for each stripe */
3660 } /* for each component in the mirror */
3665 /* destroy the sub objects */
3666 for (; i < k; i++) {
3667 rc = lod_sub_destroy(env, sub_objs[i], th);
3670 dt_object_put(env, sub_objs[i]);
3674 * if a sub object destroy failed, we'd release sub objects
3675 * reference get from above sub_objs collection.
3678 dt_object_put(env, sub_objs[i]);
3680 OBD_FREE_PTR_ARRAY(sub_objs, array_count);
3682 mutex_unlock(&lo->ldo_layout_mutex);
3688 * Purge layouts, delete sub objects in the mirror stored in the vic_buf,
3689 * and set the LOVEA with the layout from mbuf.
3691 static int lod_declare_layout_purge(const struct lu_env *env,
3692 struct dt_object *dt, const struct lu_buf *buf,
3695 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3696 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3701 if (le32_to_cpu(comp_v1->lcm_magic) != LOV_MAGIC_COMP_V1) {
3702 CERROR("%s: invalid layout magic %#x != %#x\n",
3703 lod2obd(d)->obd_name, le32_to_cpu(comp_v1->lcm_magic),
3708 if (cpu_to_le32(LOV_MAGIC_COMP_V1) != LOV_MAGIC_COMP_V1)
3709 lustre_swab_lov_comp_md_v1(comp_v1);
3711 /* from now on, @buf contains cpu endian data */
3713 if (comp_v1->lcm_mirror_count != 0) {
3714 CERROR("%s: can only purge one mirror from "DFID"\n",
3715 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3719 /* delcare sub objects deletion in the mirror stored in @buf */
3720 rc = lod_layout_declare_or_purge_mirror(env, dt, buf, th, true);
3724 /* delete sub objects from the mirror stored in @buf */
3725 static int lod_layout_purge(const struct lu_env *env, struct dt_object *dt,
3726 const struct lu_buf *buf, struct thandle *th)
3731 rc = lod_layout_declare_or_purge_mirror(env, dt, buf, th, false);
3736 * Implementation of dt_object_operations::do_declare_xattr_set.
3738 * \see dt_object_operations::do_declare_xattr_set() in the API description
3741 * the extension to the API:
3742 * - declaring LOVEA requests striping creation
3743 * - LU_XATTR_REPLACE means layout swap
3745 static int lod_declare_xattr_set(const struct lu_env *env,
3746 struct dt_object *dt,
3747 const struct lu_buf *buf,
3748 const char *name, int fl,
3751 struct dt_object *next = dt_object_child(dt);
3752 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3757 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
3758 if ((S_ISREG(mode) || mode == 0) &&
3759 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE | LU_XATTR_SPLIT |
3761 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3762 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
3764 * this is a request to create object's striping.
3766 * allow to declare predefined striping on a new (!mode) object
3767 * which is supposed to be replay of regular file creation
3768 * (when LOV setting is declared)
3770 * LU_XATTR_REPLACE is set to indicate a layout swap
3772 if (dt_object_exists(dt)) {
3773 rc = dt_attr_get(env, next, attr);
3777 memset(attr, 0, sizeof(*attr));
3778 attr->la_valid = LA_TYPE | LA_MODE;
3779 attr->la_mode = S_IFREG;
3781 rc = lod_declare_striped_create(env, dt, attr, buf, th);
3782 } else if (fl & LU_XATTR_MERGE) {
3783 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3784 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3785 rc = lod_declare_layout_merge(env, dt, buf, th);
3786 } else if (fl & LU_XATTR_SPLIT) {
3787 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3788 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3789 rc = lod_declare_layout_split(env, dt, buf, th);
3790 } else if (fl & LU_XATTR_PURGE) {
3791 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3792 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3793 rc = lod_declare_layout_purge(env, dt, buf, th);
3794 } else if (S_ISREG(mode) &&
3795 strlen(name) >= sizeof(XATTR_LUSTRE_LOV) + 3 &&
3796 allowed_lustre_lov(name)) {
3798 * this is a request to modify object's striping.
3799 * add/set/del component(s).
3801 if (!dt_object_exists(dt))
3804 rc = lod_declare_modify_layout(env, dt, name, buf, th);
3805 } else if (S_ISDIR(mode)) {
3806 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
3807 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3808 rc = lod_replace_parent_fid(env, dt, buf, th, true);
3810 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
3817 * Apply xattr changes to the object.
3819 * Applies xattr changes to the object and the stripes if the latter exist.
3821 * \param[in] env execution environment
3822 * \param[in] dt object
3823 * \param[in] buf buffer pointing to the new value of xattr
3824 * \param[in] name name of xattr
3825 * \param[in] fl flags
3826 * \param[in] th transaction handle
3828 * \retval 0 on success
3829 * \retval negative if failed
3831 static int lod_xattr_set_internal(const struct lu_env *env,
3832 struct dt_object *dt,
3833 const struct lu_buf *buf,
3834 const char *name, int fl,
3837 struct dt_object *next = dt_object_child(dt);
3838 struct lod_object *lo = lod_dt_obj(dt);
3843 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3844 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3847 /* Note: Do not set LinkEA on sub-stripes, otherwise
3848 * it will confuse the fid2path process(see mdt_path_current()).
3849 * The linkEA between master and sub-stripes is set in
3850 * lod_xattr_set_lmv(). */
3851 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
3854 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3855 if (!lo->ldo_stripe[i])
3858 if (!dt_object_exists(lo->ldo_stripe[i]))
3861 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
3871 * Delete an extended attribute.
3873 * Deletes specified xattr from the object and the stripes if the latter exist.
3875 * \param[in] env execution environment
3876 * \param[in] dt object
3877 * \param[in] name name of xattr
3878 * \param[in] th transaction handle
3880 * \retval 0 on success
3881 * \retval negative if failed
3883 static int lod_xattr_del_internal(const struct lu_env *env,
3884 struct dt_object *dt,
3885 const char *name, struct thandle *th)
3887 struct dt_object *next = dt_object_child(dt);
3888 struct lod_object *lo = lod_dt_obj(dt);
3894 rc = lod_sub_xattr_del(env, next, name, th);
3895 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3898 if (lo->ldo_dir_stripe_count == 0)
3901 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3902 if (!lo->ldo_stripe[i])
3905 if (!dt_object_exists(lo->ldo_stripe[i]))
3908 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3917 * Set default striping on a directory.
3919 * Sets specified striping on a directory object unless it matches the default
3920 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3921 * EA. This striping will be used when regular file is being created in this
3924 * \param[in] env execution environment
3925 * \param[in] dt the striped object
3926 * \param[in] buf buffer with the striping
3927 * \param[in] name name of EA
3928 * \param[in] fl xattr flag (see OSD API description)
3929 * \param[in] th transaction handle
3931 * \retval 0 on success
3932 * \retval negative if failed
3934 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
3935 struct dt_object *dt,
3936 const struct lu_buf *buf,
3937 const char *name, int fl,
3940 struct lov_user_md_v1 *lum;
3941 struct lov_user_md_v3 *v3 = NULL;
3942 const char *pool_name = NULL;
3947 LASSERT(buf != NULL && buf->lb_buf != NULL);
3950 switch (lum->lmm_magic) {
3951 case LOV_USER_MAGIC_SPECIFIC:
3952 case LOV_USER_MAGIC_V3:
3954 if (v3->lmm_pool_name[0] != '\0')
3955 pool_name = v3->lmm_pool_name;
3957 case LOV_USER_MAGIC_V1:
3958 /* if { size, offset, count } = { 0, -1, 0 } and no pool
3959 * (i.e. all default values specified) then delete default
3960 * striping from dir. */
3962 "set default striping: sz %u # %u offset %d %s %s\n",
3963 (unsigned)lum->lmm_stripe_size,
3964 (unsigned)lum->lmm_stripe_count,
3965 (int)lum->lmm_stripe_offset,
3966 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
3968 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
3969 lum->lmm_stripe_count,
3970 lum->lmm_stripe_offset,
3973 case LOV_USER_MAGIC_COMP_V1:
3975 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lum;
3976 struct lov_comp_md_entry_v1 *lcme;
3979 comp_cnt = le16_to_cpu(lcm->lcm_entry_count);
3980 for (i = 0; i < comp_cnt; i++) {
3981 lcme = &lcm->lcm_entries[i];
3982 if (lcme->lcme_flags & cpu_to_le32(LCME_FL_EXTENSION)) {
3983 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
3992 CERROR("Invalid magic %x\n", lum->lmm_magic);
3997 rc = lod_xattr_del_internal(env, dt, name, th);
4001 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4007 static int lod_get_default_lov_striping(const struct lu_env *env,
4008 struct lod_object *lo,
4009 struct lod_default_striping *lds,
4010 struct dt_allocation_hint *ah);
4013 * Helper function to convert compound layout to compound layout with
4016 * Copy lcm_entries array of \a src to \a tgt. Replace lov_user_md_v1
4017 * components of \a src with lov_user_md_v3 using \a pool.
4019 * \param[in] src source layout
4020 * \param[in] pool pool to use in \a tgt
4021 * \param[out] tgt target layout
4023 static void embed_pool_to_comp_v1(const struct lov_comp_md_v1 *src,
4025 struct lov_comp_md_v1 *tgt)
4028 struct lov_user_md_v1 *lum;
4029 struct lov_user_md_v3 *lum3;
4030 struct lov_comp_md_entry_v1 *entry;
4034 entry = tgt->lcm_entries;
4036 for (i = 0; i < le16_to_cpu(src->lcm_entry_count); i++, entry++) {
4037 *entry = src->lcm_entries[i];
4038 offset = le32_to_cpu(src->lcm_entries[i].lcme_offset);
4039 entry->lcme_offset = cpu_to_le32(offset + shift);
4041 lum = (struct lov_user_md_v1 *)((char *)src + offset);
4042 lum3 = (struct lov_user_md_v3 *)((char *)tgt + offset + shift);
4043 *(struct lov_user_md_v1 *)lum3 = *lum;
4044 if (lum->lmm_pattern == cpu_to_le32(LOV_PATTERN_MDT)) {
4045 lum3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
4047 lum3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4048 entry->lcme_size = cpu_to_le32(sizeof(*lum3));
4049 strlcpy(lum3->lmm_pool_name, pool,
4050 sizeof(lum3->lmm_pool_name));
4051 shift += sizeof(*lum3) - sizeof(*lum);
4057 * Set default striping on a directory.
4059 * Sets specified striping on a directory object unless it matches the default
4060 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
4061 * EA. This striping will be used when regular file is being created in this
4063 * If current default striping includes a pool but specifed striping
4064 * does not - retain the pool if it exists.
4066 * \param[in] env execution environment
4067 * \param[in] dt the striped object
4068 * \param[in] buf buffer with the striping
4069 * \param[in] name name of EA
4070 * \param[in] fl xattr flag (see OSD API description)
4071 * \param[in] th transaction handle
4073 * \retval 0 on success
4074 * \retval negative if failed
4076 static int lod_xattr_set_default_lov_on_dir(const struct lu_env *env,
4077 struct dt_object *dt,
4078 const struct lu_buf *buf,
4079 const char *name, int fl,
4082 struct lod_default_striping *lds = lod_lds_buf_get(env);
4083 struct lov_user_md_v1 *v1 = buf->lb_buf;
4084 char pool[LOV_MAXPOOLNAME + 1];
4090 /* get existing striping config */
4091 rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds, NULL);
4095 memset(pool, 0, sizeof(pool));
4096 if (lds->lds_def_striping_set == 1)
4097 lod_layout_get_pool(lds->lds_def_comp_entries,
4098 lds->lds_def_comp_cnt, pool,
4101 is_del = LOVEA_DELETE_VALUES(v1->lmm_stripe_size,
4102 v1->lmm_stripe_count,
4103 v1->lmm_stripe_offset,
4106 /* Retain the pool name if it is not given */
4107 if (v1->lmm_magic == LOV_USER_MAGIC_V1 && pool[0] != '\0' &&
4109 struct lod_thread_info *info = lod_env_info(env);
4110 struct lov_user_md_v3 *v3 = info->lti_ea_store;
4112 memset(v3, 0, sizeof(*v3));
4113 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4114 v3->lmm_pattern = cpu_to_le32(v1->lmm_pattern);
4115 v3->lmm_stripe_count = cpu_to_le32(v1->lmm_stripe_count);
4116 v3->lmm_stripe_offset = cpu_to_le32(v1->lmm_stripe_offset);
4117 v3->lmm_stripe_size = cpu_to_le32(v1->lmm_stripe_size);
4119 strlcpy(v3->lmm_pool_name, pool, sizeof(v3->lmm_pool_name));
4121 info->lti_buf.lb_buf = v3;
4122 info->lti_buf.lb_len = sizeof(*v3);
4123 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4125 } else if (v1->lmm_magic == LOV_USER_MAGIC_COMP_V1 &&
4126 pool[0] != '\0' && !is_del) {
4128 * try to retain the pool from default layout if the
4129 * specified component layout does not provide pool
4132 struct lod_thread_info *info = lod_env_info(env);
4133 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
4134 struct lov_comp_md_v1 *comp_v1p;
4135 struct lov_user_md_v1 *lum;
4139 struct lov_comp_md_entry_v1 *entry;
4142 entry_count = le16_to_cpu(comp_v1->lcm_entry_count);
4143 size = sizeof(*comp_v1) +
4144 entry_count * sizeof(comp_v1->lcm_entries[0]);
4145 entry = comp_v1->lcm_entries;
4146 for (i = 0; i < entry_count; i++, entry++) {
4147 offset = le32_to_cpu(entry->lcme_offset);
4148 lum = (struct lov_user_md_v1 *)((char *)comp_v1 +
4150 if (le32_to_cpu(lum->lmm_magic) != LOV_USER_MAGIC_V1)
4151 /* the i-th component includes pool info */
4153 if (lum->lmm_pattern == cpu_to_le32(LOV_PATTERN_MDT))
4154 size += sizeof(struct lov_user_md_v1);
4156 size += sizeof(struct lov_user_md_v3);
4159 if (i == entry_count) {
4161 * re-compose the layout to include the pool for
4164 if (info->lti_ea_store_size < size)
4165 rc = lod_ea_store_resize(info, size);
4168 comp_v1p = info->lti_ea_store;
4169 *comp_v1p = *comp_v1;
4170 comp_v1p->lcm_size = cpu_to_le32(size);
4171 embed_pool_to_comp_v1(comp_v1, pool, comp_v1p);
4173 info->lti_buf.lb_buf = comp_v1p;
4174 info->lti_buf.lb_len = size;
4175 rc = lod_xattr_set_lov_on_dir(env, dt,
4180 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl,
4184 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl, th);
4187 if (lds->lds_def_striping_set == 1 && lds->lds_def_comp_entries != NULL)
4188 lod_free_def_comp_entries(lds);
4194 * Set default striping on a directory object.
4196 * Sets specified striping on a directory object unless it matches the default
4197 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
4198 * EA. This striping will be used when a new directory is being created in the
4201 * \param[in] env execution environment
4202 * \param[in] dt the striped object
4203 * \param[in] buf buffer with the striping
4204 * \param[in] name name of EA
4205 * \param[in] fl xattr flag (see OSD API description)
4206 * \param[in] th transaction handle
4208 * \retval 0 on success
4209 * \retval negative if failed
4211 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
4212 struct dt_object *dt,
4213 const struct lu_buf *buf,
4214 const char *name, int fl,
4217 struct lmv_user_md_v1 *lum;
4222 LASSERT(buf != NULL && buf->lb_buf != NULL);
4226 "set default stripe_count # %u stripe_offset %d hash %u\n",
4227 le32_to_cpu(lum->lum_stripe_count),
4228 (int)le32_to_cpu(lum->lum_stripe_offset),
4229 le32_to_cpu(lum->lum_hash_type));
4231 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
4232 le32_to_cpu(lum->lum_stripe_offset)) &&
4233 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
4234 rc = lod_xattr_del_internal(env, dt, name, th);
4238 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4247 * Turn directory into a striped directory.
4249 * During replay the client sends the striping created before MDT
4250 * failure, then the layer above LOD sends this defined striping
4251 * using ->do_xattr_set(), so LOD uses this method to replay creation
4252 * of the stripes. Notice the original information for the striping
4253 * (#stripes, FIDs, etc) was transferred in declare path.
4255 * \param[in] env execution environment
4256 * \param[in] dt the striped object
4257 * \param[in] buf not used currently
4258 * \param[in] name not used currently
4259 * \param[in] fl xattr flag (see OSD API description)
4260 * \param[in] th transaction handle
4262 * \retval 0 on success
4263 * \retval negative if failed
4265 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
4266 const struct lu_buf *buf, const char *name,
4267 int fl, struct thandle *th)
4269 struct lod_object *lo = lod_dt_obj(dt);
4270 struct lod_thread_info *info = lod_env_info(env);
4271 struct lu_attr *attr = &info->lti_attr;
4272 struct dt_object_format *dof = &info->lti_format;
4273 struct lu_buf lmv_buf;
4274 struct lu_buf slave_lmv_buf;
4275 struct lmv_mds_md_v1 *lmm;
4276 struct lmv_mds_md_v1 *slave_lmm = NULL;
4277 struct dt_insert_rec *rec = &info->lti_dt_rec;
4282 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4285 /* The stripes are supposed to be allocated in declare phase,
4286 * if there are no stripes being allocated, it will skip */
4287 if (lo->ldo_dir_stripe_count == 0) {
4288 if (lo->ldo_is_foreign) {
4289 rc = lod_sub_xattr_set(env, dt_object_child(dt), buf,
4290 XATTR_NAME_LMV, fl, th);
4297 rc = dt_attr_get(env, dt_object_child(dt), attr);
4301 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME | LA_FLAGS |
4302 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
4303 dof->dof_type = DFT_DIR;
4305 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
4308 lmm = lmv_buf.lb_buf;
4310 OBD_ALLOC_PTR(slave_lmm);
4311 if (slave_lmm == NULL)
4314 lod_prep_slave_lmv_md(slave_lmm, lmm);
4315 slave_lmv_buf.lb_buf = slave_lmm;
4316 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
4318 rec->rec_type = S_IFDIR;
4319 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4320 struct dt_object *dto = lo->ldo_stripe[i];
4321 char *stripe_name = info->lti_key;
4322 struct lu_name *sname;
4323 struct linkea_data ldata = { NULL };
4324 struct lu_buf linkea_buf;
4326 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
4330 /* fail a remote stripe creation */
4331 if (i && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_CREATE))
4334 /* don't create stripe if:
4335 * 1. it's source stripe of migrating directory
4336 * 2. it's existed stripe of splitting directory
4338 if ((lod_is_migrating(lo) && i >= lo->ldo_dir_migrate_offset) ||
4339 (lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) {
4340 if (!dt_object_exists(dto))
4341 GOTO(out, rc = -EINVAL);
4343 dt_write_lock(env, dto, DT_TGT_CHILD);
4344 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
4346 dt_write_unlock(env, dto);
4350 rc = lod_sub_ref_add(env, dto, th);
4351 dt_write_unlock(env, dto);
4355 rec->rec_fid = lu_object_fid(&dto->do_lu);
4356 rc = lod_sub_insert(env, dto,
4357 (const struct dt_rec *)rec,
4358 (const struct dt_key *)dot, th);
4363 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
4364 cfs_fail_val != i) {
4365 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
4367 slave_lmm->lmv_master_mdt_index =
4370 slave_lmm->lmv_master_mdt_index =
4373 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
4374 XATTR_NAME_LMV, 0, th);
4379 /* don't insert stripe if it's existed stripe of splitting
4380 * directory (this directory is striped).
4381 * NB, plain directory will insert itself as the first
4384 if (lod_is_splitting(lo) && lo->ldo_dir_split_offset > 1 &&
4385 lo->ldo_dir_split_offset > i)
4388 rec->rec_fid = lu_object_fid(&dt->do_lu);
4389 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
4390 (const struct dt_key *)dotdot, th);
4394 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
4396 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4397 PFID(lu_object_fid(&dto->do_lu)), i + 1);
4399 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4400 PFID(lu_object_fid(&dto->do_lu)), i);
4402 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
4403 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
4404 sname, lu_object_fid(&dt->do_lu));
4408 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
4409 linkea_buf.lb_len = ldata.ld_leh->leh_len;
4410 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
4411 XATTR_NAME_LINK, 0, th);
4415 rec->rec_fid = lu_object_fid(&dto->do_lu);
4416 rc = lod_sub_insert(env, dt_object_child(dt),
4417 (const struct dt_rec *)rec,
4418 (const struct dt_key *)stripe_name, th);
4422 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
4427 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
4428 rc = lod_sub_xattr_set(env, dt_object_child(dt),
4429 &lmv_buf, XATTR_NAME_LMV, fl, th);
4431 if (slave_lmm != NULL)
4432 OBD_FREE_PTR(slave_lmm);
4438 * Helper function to declare/execute creation of a striped directory
4440 * Called in declare/create object path, prepare striping for a directory
4441 * and prepare defaults data striping for the objects to be created in
4442 * that directory. Notice the function calls "declaration" or "execution"
4443 * methods depending on \a declare param. This is a consequence of the
4444 * current approach while we don't have natural distributed transactions:
4445 * we basically execute non-local updates in the declare phase. So, the
4446 * arguments for the both phases are the same and this is the reason for
4447 * this function to exist.
4449 * \param[in] env execution environment
4450 * \param[in] dt object
4451 * \param[in] attr attributes the stripes will be created with
4452 * \param[in] lmu lmv_user_md if MDT indices are specified
4453 * \param[in] dof format of stripes (see OSD API description)
4454 * \param[in] th transaction handle
4455 * \param[in] declare where to call "declare" or "execute" methods
4457 * \retval 0 on success
4458 * \retval negative if failed
4460 static int lod_dir_striping_create_internal(const struct lu_env *env,
4461 struct dt_object *dt,
4462 struct lu_attr *attr,
4463 const struct lu_buf *lmu,
4464 struct dt_object_format *dof,
4468 struct lod_thread_info *info = lod_env_info(env);
4469 struct lod_object *lo = lod_dt_obj(dt);
4470 const struct lod_default_striping *lds = lo->ldo_def_striping;
4474 LASSERT(ergo(lds != NULL,
4475 lds->lds_def_striping_set ||
4476 lds->lds_dir_def_striping_set));
4478 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
4479 lo->ldo_dir_stripe_offset)) {
4481 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
4482 int stripe_count = lo->ldo_dir_stripe_count;
4484 if (info->lti_ea_store_size < sizeof(*v1)) {
4485 rc = lod_ea_store_resize(info, sizeof(*v1));
4488 v1 = info->lti_ea_store;
4491 memset(v1, 0, sizeof(*v1));
4492 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
4493 v1->lum_stripe_count = cpu_to_le32(stripe_count);
4494 v1->lum_stripe_offset =
4495 cpu_to_le32(lo->ldo_dir_stripe_offset);
4497 info->lti_buf.lb_buf = v1;
4498 info->lti_buf.lb_len = sizeof(*v1);
4499 lmu = &info->lti_buf;
4503 rc = lod_declare_xattr_set_lmv(env, dt, attr, lmu, dof,
4506 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
4511 /* foreign LMV EA case */
4513 struct lmv_foreign_md *lfm = lmu->lb_buf;
4515 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
4516 rc = lod_declare_xattr_set_lmv(env, dt, attr,
4520 if (lo->ldo_is_foreign) {
4521 LASSERT(lo->ldo_foreign_lmv != NULL &&
4522 lo->ldo_foreign_lmv_size > 0);
4523 info->lti_buf.lb_buf = lo->ldo_foreign_lmv;
4524 info->lti_buf.lb_len = lo->ldo_foreign_lmv_size;
4525 lmu = &info->lti_buf;
4526 rc = lod_xattr_set_lmv(env, dt, lmu,
4527 XATTR_NAME_LMV, 0, th);
4532 /* Transfer default LMV striping from the parent */
4533 if (lds != NULL && lds->lds_dir_def_striping_set &&
4534 lds->lds_dir_def_max_inherit != LMV_INHERIT_END &&
4535 lds->lds_dir_def_max_inherit != LMV_INHERIT_NONE &&
4536 !(LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
4537 lds->lds_dir_def_stripe_offset) &&
4538 le32_to_cpu(lds->lds_dir_def_hash_type) !=
4539 LMV_HASH_TYPE_UNKNOWN)) {
4540 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
4542 if (info->lti_ea_store_size < sizeof(*v1)) {
4543 rc = lod_ea_store_resize(info, sizeof(*v1));
4546 v1 = info->lti_ea_store;
4549 memset(v1, 0, sizeof(*v1));
4550 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
4551 v1->lum_stripe_count =
4552 cpu_to_le32(lds->lds_dir_def_stripe_count);
4553 v1->lum_stripe_offset =
4554 cpu_to_le32(lds->lds_dir_def_stripe_offset);
4556 cpu_to_le32(lds->lds_dir_def_hash_type);
4557 v1->lum_max_inherit =
4558 lmv_inherit_next(lds->lds_dir_def_max_inherit);
4559 v1->lum_max_inherit_rr =
4560 lmv_inherit_rr_next(lds->lds_dir_def_max_inherit_rr);
4562 info->lti_buf.lb_buf = v1;
4563 info->lti_buf.lb_len = sizeof(*v1);
4565 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4566 XATTR_NAME_DEFAULT_LMV,
4569 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
4571 XATTR_NAME_DEFAULT_LMV, 0,
4577 /* Transfer default LOV striping from the parent */
4578 if (lds != NULL && lds->lds_def_striping_set &&
4579 lds->lds_def_comp_cnt != 0) {
4580 struct lov_mds_md *lmm;
4581 int lmm_size = lod_comp_md_size(lo, true);
4583 if (info->lti_ea_store_size < lmm_size) {
4584 rc = lod_ea_store_resize(info, lmm_size);
4588 lmm = info->lti_ea_store;
4590 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
4594 info->lti_buf.lb_buf = lmm;
4595 info->lti_buf.lb_len = lmm_size;
4598 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4599 XATTR_NAME_LOV, 0, th);
4601 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4602 XATTR_NAME_LOV, 0, th);
4607 /* ldo_def_striping is not allocated, clear after use, in case directory
4608 * layout is changed later.
4611 lo->ldo_def_striping = NULL;
4616 static int lod_declare_dir_striping_create(const struct lu_env *env,
4617 struct dt_object *dt,
4618 struct lu_attr *attr,
4620 struct dt_object_format *dof,
4623 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
4627 static int lod_dir_striping_create(const struct lu_env *env,
4628 struct dt_object *dt,
4629 struct lu_attr *attr,
4630 struct dt_object_format *dof,
4633 return lod_dir_striping_create_internal(env, dt, attr, NULL, dof, th,
4638 * Make LOV EA for striped object.
4640 * Generate striping information and store it in the LOV EA of the given
4641 * object. The caller must ensure nobody else is calling the function
4642 * against the object concurrently. The transaction must be started.
4643 * FLDB service must be running as well; it's used to map FID to the target,
4644 * which is stored in LOV EA.
4646 * \param[in] env execution environment for this thread
4647 * \param[in] lo LOD object
4648 * \param[in] th transaction handle
4650 * \retval 0 if LOV EA is stored successfully
4651 * \retval negative error number on failure
4653 static int lod_generate_and_set_lovea(const struct lu_env *env,
4654 struct lod_object *lo,
4657 struct lod_thread_info *info = lod_env_info(env);
4658 struct dt_object *next = dt_object_child(&lo->ldo_obj);
4659 struct lov_mds_md_v1 *lmm;
4665 if (lo->ldo_comp_cnt == 0 && !lo->ldo_is_foreign) {
4666 lod_striping_free_nolock(env, lo);
4667 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
4671 lmm_size = lod_comp_md_size(lo, false);
4672 if (info->lti_ea_store_size < lmm_size) {
4673 rc = lod_ea_store_resize(info, lmm_size);
4677 lmm = info->lti_ea_store;
4679 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
4683 info->lti_buf.lb_buf = lmm;
4684 info->lti_buf.lb_len = lmm_size;
4685 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
4686 XATTR_NAME_LOV, 0, th);
4690 static __u32 lod_gen_component_id(struct lod_object *lo,
4691 int mirror_id, int comp_idx);
4694 * Repeat an existing component
4696 * Creates a new layout by replicating an existing component. Uses striping
4697 * policy from previous component as a template for the striping for the new
4700 * New component starts with zero length, will be extended (or removed) before
4701 * returning layout to client.
4703 * NB: Reallocates layout components array (lo->ldo_comp_entries), invalidating
4704 * any pre-existing pointers to components. Handle with care.
4706 * \param[in] env execution environment for this thread
4707 * \param[in,out] lo object to update the layout of
4708 * \param[in] index index of component to copy
4710 * \retval 0 on success
4711 * \retval negative errno on error
4713 static int lod_layout_repeat_comp(const struct lu_env *env,
4714 struct lod_object *lo, int index)
4716 struct lod_layout_component *lod_comp;
4717 struct lod_layout_component *new_comp = NULL;
4718 struct lod_layout_component *comp_array;
4719 int rc = 0, i, new_cnt = lo->ldo_comp_cnt + 1;
4724 lod_comp = &lo->ldo_comp_entries[index];
4725 LASSERT(lod_comp_inited(lod_comp) && lod_comp->llc_id != LCME_ID_INVAL);
4727 CDEBUG(D_LAYOUT, "repeating component %d\n", index);
4729 OBD_ALLOC_PTR_ARRAY(comp_array, new_cnt);
4730 if (comp_array == NULL)
4731 GOTO(out, rc = -ENOMEM);
4733 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4734 memcpy(&comp_array[i + offset], &lo->ldo_comp_entries[i],
4735 sizeof(*comp_array));
4737 /* Duplicate this component in to the next slot */
4739 new_comp = &comp_array[i + 1];
4740 memcpy(&comp_array[i + 1], &lo->ldo_comp_entries[i],
4741 sizeof(*comp_array));
4742 /* We must now skip this new component when copying */
4747 /* Set up copied component */
4748 new_comp->llc_flags &= ~LCME_FL_INIT;
4749 new_comp->llc_stripe = NULL;
4750 new_comp->llc_stripes_allocated = 0;
4751 new_comp->llc_ost_indices = NULL;
4752 new_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4753 /* for uninstantiated components, layout gen stores default stripe
4755 new_comp->llc_layout_gen = lod_comp->llc_stripe_offset;
4756 /* This makes the repeated component zero-length, placed at the end of
4757 * the preceding component */
4758 new_comp->llc_extent.e_start = new_comp->llc_extent.e_end;
4759 new_comp->llc_timestamp = lod_comp->llc_timestamp;
4760 new_comp->llc_pool = NULL;
4762 rc = lod_set_pool(&new_comp->llc_pool, lod_comp->llc_pool);
4766 if (new_comp->llc_ostlist.op_array) {
4767 __u32 *op_array = NULL;
4769 OBD_ALLOC(op_array, new_comp->llc_ostlist.op_size);
4771 GOTO(out, rc = -ENOMEM);
4772 memcpy(op_array, &new_comp->llc_ostlist.op_array,
4773 new_comp->llc_ostlist.op_size);
4774 new_comp->llc_ostlist.op_array = op_array;
4777 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
4778 lo->ldo_comp_entries = comp_array;
4779 lo->ldo_comp_cnt = new_cnt;
4781 /* Generate an id for the new component */
4782 mirror_id = mirror_id_of(new_comp->llc_id);
4783 new_comp->llc_id = LCME_ID_INVAL;
4784 new_comp->llc_id = lod_gen_component_id(lo, mirror_id, index + 1);
4785 if (new_comp->llc_id == LCME_ID_INVAL)
4786 GOTO(out, rc = -ERANGE);
4791 OBD_FREE_PTR_ARRAY(comp_array, new_cnt);
4796 static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
4800 /* clear memory region that will be used for layout change */
4801 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
4802 info->lti_count = 0;
4804 if (info->lti_comp_size >= comp_cnt)
4807 if (info->lti_comp_size > 0) {
4808 OBD_FREE_PTR_ARRAY(info->lti_comp_idx, info->lti_comp_size);
4809 info->lti_comp_size = 0;
4812 OBD_ALLOC_PTR_ARRAY(info->lti_comp_idx, comp_cnt);
4813 if (!info->lti_comp_idx)
4816 info->lti_comp_size = comp_cnt;
4821 * Prepare new layout minus deleted components
4823 * Removes components marked for deletion (LCME_ID_INVAL) by copying to a new
4824 * layout and skipping those components. Removes stripe objects if any exist.
4827 * Reallocates layout components array (lo->ldo_comp_entries), invalidating
4828 * any pre-existing pointers to components.
4830 * Caller is responsible for updating mirror end (ldo_mirror[].lme_end).
4832 * \param[in] env execution environment for this thread
4833 * \param[in,out] lo object to update the layout of
4834 * \param[in] th transaction handle for this operation
4836 * \retval # of components deleted
4837 * \retval negative errno on error
4839 static int lod_layout_del_prep_layout(const struct lu_env *env,
4840 struct lod_object *lo,
4843 struct lod_layout_component *lod_comp;
4844 struct lod_thread_info *info = lod_env_info(env);
4845 int rc = 0, i, j, deleted = 0;
4849 LASSERT(lo->ldo_is_composite);
4850 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
4852 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
4856 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4857 lod_comp = &lo->ldo_comp_entries[i];
4859 if (lod_comp->llc_id != LCME_ID_INVAL) {
4860 /* Build array of things to keep */
4861 info->lti_comp_idx[info->lti_count++] = i;
4865 lod_obj_set_pool(lo, i, NULL);
4866 if (lod_comp->llc_ostlist.op_array) {
4867 OBD_FREE(lod_comp->llc_ostlist.op_array,
4868 lod_comp->llc_ostlist.op_size);
4869 lod_comp->llc_ostlist.op_array = NULL;
4870 lod_comp->llc_ostlist.op_size = 0;
4874 CDEBUG(D_LAYOUT, "deleting comp %d, left %d\n", i,
4875 lo->ldo_comp_cnt - deleted);
4877 /* No striping info for this component */
4878 if (lod_comp->llc_stripe == NULL)
4881 LASSERT(lod_comp->llc_stripe_count > 0);
4882 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4883 struct dt_object *obj = lod_comp->llc_stripe[j];
4888 /* components which are not init have no sub objects
4890 if (lod_comp_inited(lod_comp)) {
4891 rc = lod_sub_destroy(env, obj, th);
4896 lu_object_put(env, &obj->do_lu);
4897 lod_comp->llc_stripe[j] = NULL;
4899 OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe,
4900 lod_comp->llc_stripes_allocated);
4901 lod_comp->llc_stripe = NULL;
4902 OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices,
4903 lod_comp->llc_stripes_allocated);
4904 lod_comp->llc_ost_indices = NULL;
4905 lod_comp->llc_stripes_allocated = 0;
4908 /* info->lti_count has the amount of left components */
4909 LASSERTF(info->lti_count >= 0 && info->lti_count < lo->ldo_comp_cnt,
4910 "left = %d, lo->ldo_comp_cnt %d\n", (int)info->lti_count,
4911 (int)lo->ldo_comp_cnt);
4913 if (info->lti_count > 0) {
4914 struct lod_layout_component *comp_array;
4916 OBD_ALLOC_PTR_ARRAY(comp_array, info->lti_count);
4917 if (comp_array == NULL)
4918 GOTO(out, rc = -ENOMEM);
4920 for (i = 0; i < info->lti_count; i++) {
4921 memcpy(&comp_array[i],
4922 &lo->ldo_comp_entries[info->lti_comp_idx[i]],
4923 sizeof(*comp_array));
4926 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
4927 lo->ldo_comp_entries = comp_array;
4928 lo->ldo_comp_cnt = info->lti_count;
4930 lod_free_comp_entries(lo);
4935 return rc ? rc : deleted;
4939 * Delete layout component(s)
4941 * This function sets up the layout data in the env and does the setattrs
4942 * required to write out the new layout. The layout itself is modified in
4943 * lod_layout_del_prep_layout.
4945 * \param[in] env execution environment for this thread
4946 * \param[in] dt object
4947 * \param[in] th transaction handle
4949 * \retval 0 on success
4950 * \retval negative error number on failure
4952 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
4955 struct lod_object *lo = lod_dt_obj(dt);
4956 struct dt_object *next = dt_object_child(dt);
4957 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4960 LASSERT(lo->ldo_mirror_count == 1);
4962 mutex_lock(&lo->ldo_layout_mutex);
4964 rc = lod_layout_del_prep_layout(env, lo, th);
4968 /* Only do this if we didn't delete all components */
4969 if (lo->ldo_comp_cnt > 0) {
4970 lo->ldo_mirrors[0].lme_end = lo->ldo_comp_cnt - 1;
4971 lod_obj_inc_layout_gen(lo);
4974 LASSERT(dt_object_exists(dt));
4975 rc = dt_attr_get(env, next, attr);
4979 if (attr->la_size > 0) {
4981 attr->la_valid = LA_SIZE;
4982 rc = lod_sub_attr_set(env, next, attr, th);
4987 rc = lod_generate_and_set_lovea(env, lo, th);
4991 lod_striping_free_nolock(env, lo);
4993 mutex_unlock(&lo->ldo_layout_mutex);
5000 * Implementation of dt_object_operations::do_xattr_set.
5002 * Sets specified extended attribute on the object. Three types of EAs are
5004 * LOV EA - stores striping for a regular file or default striping (when set
5006 * LMV EA - stores a marker for the striped directories
5007 * DMV EA - stores default directory striping
5009 * When striping is applied to a non-striped existing object (this is called
5010 * late striping), then LOD notices the caller wants to turn the object into a
5011 * striped one. The stripe objects are created and appropriate EA is set:
5012 * LOV EA storing all the stripes directly or LMV EA storing just a small header
5013 * with striping configuration.
5015 * \see dt_object_operations::do_xattr_set() in the API description for details.
5017 static int lod_xattr_set(const struct lu_env *env,
5018 struct dt_object *dt, const struct lu_buf *buf,
5019 const char *name, int fl, struct thandle *th)
5021 struct dt_object *next = dt_object_child(dt);
5022 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
5023 struct lod_object *lo = lod_dt_obj(dt);
5024 struct lod_obj_stripe_cb_data data = { {0} };
5029 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
5030 !strcmp(name, XATTR_NAME_LMV)) {
5032 case LU_XATTR_CREATE:
5033 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
5036 case LU_XATTR_REPLACE:
5037 rc = lod_dir_layout_set(env, dt, buf, fl, th);
5044 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
5045 strcmp(name, XATTR_NAME_LOV) == 0) {
5046 rc = lod_xattr_set_default_lov_on_dir(env, dt, buf, name, fl,
5049 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
5050 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
5052 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
5055 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
5056 (strcmp(name, XATTR_NAME_LOV) == 0 ||
5057 strcmp(name, XATTR_LUSTRE_LOV) == 0 ||
5058 allowed_lustre_lov(name))) {
5059 /* in case of lov EA swap, just set it
5060 * if not, it is a replay so check striping match what we
5061 * already have during req replay, declare_xattr_set()
5062 * defines striping, then create() does the work */
5063 if (fl & LU_XATTR_REPLACE) {
5064 /* free stripes, then update disk */
5065 lod_striping_free(env, lod_dt_obj(dt));
5067 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
5068 } else if (fl & LU_XATTR_SPLIT) {
5069 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
5073 rc = lod_striping_reload(env, lo, buf, LVF_ALL_STALE);
5077 if (lo->ldo_mirror_count > 1 &&
5078 layout_attr->la_valid & LA_LAYOUT_VERSION) {
5080 layout_attr->la_layout_version =
5082 data.locd_attr = layout_attr;
5083 data.locd_declare = false;
5084 data.locd_stripe_cb =
5085 lod_obj_stripe_attr_set_cb;
5086 rc = lod_obj_for_each_stripe(env, lo, th,
5091 } else if (fl & LU_XATTR_PURGE) {
5092 rc = lod_layout_purge(env, dt, buf, th);
5093 } else if (dt_object_remote(dt)) {
5094 /* This only happens during migration, see
5095 * mdd_migrate_create(), in which Master MDT will
5096 * create a remote target object, and only set
5097 * (migrating) stripe EA on the remote object,
5098 * and does not need creating each stripes. */
5099 rc = lod_sub_xattr_set(env, next, buf, name,
5101 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
5102 /* delete component(s) */
5103 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
5104 rc = lod_layout_del(env, dt, th);
5107 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
5108 * it's going to create create file with specified
5109 * component(s), the striping must have not being
5110 * cached in this case;
5112 * Otherwise, it's going to add/change component(s) to
5113 * an existing file, the striping must have been cached
5116 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
5117 !strcmp(name, XATTR_NAME_LOV),
5118 !lod_dt_obj(dt)->ldo_comp_cached));
5120 rc = lod_striped_create(env, dt, NULL, NULL, th);
5124 if (fl & LU_XATTR_MERGE && lo->ldo_mirror_count > 1 &&
5125 layout_attr->la_valid & LA_LAYOUT_VERSION) {
5126 /* mirror merge exec phase */
5127 layout_attr->la_layout_version =
5129 data.locd_attr = layout_attr;
5130 data.locd_declare = false;
5131 data.locd_stripe_cb =
5132 lod_obj_stripe_attr_set_cb;
5133 rc = lod_obj_for_each_stripe(env, lo, th,
5140 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
5141 rc = lod_replace_parent_fid(env, dt, buf, th, false);
5146 /* then all other xattr */
5147 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
5153 * Implementation of dt_object_operations::do_declare_xattr_del.
5155 * \see dt_object_operations::do_declare_xattr_del() in the API description
5158 static int lod_declare_xattr_del(const struct lu_env *env,
5159 struct dt_object *dt, const char *name,
5162 struct lod_object *lo = lod_dt_obj(dt);
5163 struct dt_object *next = dt_object_child(dt);
5168 rc = lod_sub_declare_xattr_del(env, next, name, th);
5172 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
5175 /* NB: don't delete stripe LMV, because when we do this, normally we
5176 * will remove stripes, besides, if directory LMV is corrupt, this will
5177 * prevent deleting its LMV and fixing it (via LFSCK).
5179 if (!strcmp(name, XATTR_NAME_LMV))
5182 rc = lod_striping_load(env, lo);
5186 if (lo->ldo_dir_stripe_count == 0)
5189 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5190 struct dt_object *dto = lo->ldo_stripe[i];
5195 if (!dt_object_exists(dto))
5198 rc = lod_sub_declare_xattr_del(env, dto, name, th);
5207 * Implementation of dt_object_operations::do_xattr_del.
5209 * If EA storing a regular striping is being deleted, then release
5210 * all the references to the stripe objects in core.
5212 * \see dt_object_operations::do_xattr_del() in the API description for details.
5214 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
5215 const char *name, struct thandle *th)
5221 if (!strcmp(name, XATTR_NAME_LOV) || !strcmp(name, XATTR_NAME_LMV))
5222 lod_striping_free(env, lod_dt_obj(dt));
5224 rc = lod_xattr_del_internal(env, dt, name, th);
5230 * Implementation of dt_object_operations::do_xattr_list.
5232 * \see dt_object_operations::do_xattr_list() in the API description
5235 static int lod_xattr_list(const struct lu_env *env,
5236 struct dt_object *dt, const struct lu_buf *buf)
5238 return dt_xattr_list(env, dt_object_child(dt), buf);
5241 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
5243 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
5247 * Copy OST list from layout provided by user.
5249 * \param[in] lod_comp layout_component to be filled
5250 * \param[in] v3 LOV EA V3 user data
5252 * \retval 0 on success
5253 * \retval negative if failed
5255 int lod_comp_copy_ost_lists(struct lod_layout_component *lod_comp,
5256 struct lov_user_md_v3 *v3)
5262 if (v3->lmm_stripe_offset == LOV_OFFSET_DEFAULT)
5263 v3->lmm_stripe_offset = v3->lmm_objects[0].l_ost_idx;
5265 if (lod_comp->llc_ostlist.op_array) {
5266 if (lod_comp->llc_ostlist.op_size >=
5267 v3->lmm_stripe_count * sizeof(__u32)) {
5268 lod_comp->llc_ostlist.op_count =
5269 v3->lmm_stripe_count;
5272 OBD_FREE(lod_comp->llc_ostlist.op_array,
5273 lod_comp->llc_ostlist.op_size);
5276 /* copy ost list from lmm */
5277 lod_comp->llc_ostlist.op_count = v3->lmm_stripe_count;
5278 lod_comp->llc_ostlist.op_size = v3->lmm_stripe_count * sizeof(__u32);
5279 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
5280 lod_comp->llc_ostlist.op_size);
5281 if (!lod_comp->llc_ostlist.op_array)
5284 for (j = 0; j < v3->lmm_stripe_count; j++) {
5285 lod_comp->llc_ostlist.op_array[j] =
5286 v3->lmm_objects[j].l_ost_idx;
5294 * Get default striping.
5296 * \param[in] env execution environment
5297 * \param[in] lo object
5298 * \param[out] lds default striping
5300 * \retval 0 on success
5301 * \retval negative if failed
5303 static int lod_get_default_lov_striping(const struct lu_env *env,
5304 struct lod_object *lo,
5305 struct lod_default_striping *lds,
5306 struct dt_allocation_hint *dah)
5308 struct lod_thread_info *info = lod_env_info(env);
5309 struct lov_user_md_v1 *v1 = NULL;
5310 struct lov_user_md_v3 *v3 = NULL;
5311 struct lov_comp_md_v1 *lcm = NULL;
5313 int append_stripe_count = dah != NULL ? dah->dah_append_stripe_count : 0;
5314 const char *append_pool = (dah != NULL &&
5315 dah->dah_append_pool != NULL &&
5316 dah->dah_append_pool[0] != '\0') ?
5317 dah->dah_append_pool : NULL;
5318 __u16 entry_count = 1;
5319 __u16 mirror_count = 0;
5320 bool want_composite = false;
5325 lds->lds_def_striping_set = 0;
5327 rc = lod_get_lov_ea(env, lo);
5331 if (rc < (typeof(rc))sizeof(struct lov_user_md))
5334 magic = *(__u32 *)info->lti_ea_store;
5335 if (magic == __swab32(LOV_USER_MAGIC_V1)) {
5336 lustre_swab_lov_user_md_v1(info->lti_ea_store);
5337 } else if (magic == __swab32(LOV_USER_MAGIC_V3)) {
5338 lustre_swab_lov_user_md_v3(info->lti_ea_store);
5339 } else if (magic == __swab32(LOV_USER_MAGIC_SPECIFIC)) {
5340 v3 = (struct lov_user_md_v3 *)info->lti_ea_store;
5341 lustre_swab_lov_user_md_v3(v3);
5342 lustre_swab_lov_user_md_objects(v3->lmm_objects,
5343 v3->lmm_stripe_count);
5344 } else if (magic == __swab32(LOV_USER_MAGIC_COMP_V1) ||
5345 magic == __swab32(LOV_USER_MAGIC_SEL)) {
5346 lustre_swab_lov_comp_md_v1(info->lti_ea_store);
5352 case LOV_USER_MAGIC_SPECIFIC:
5353 v1 = info->lti_ea_store;
5355 case LOV_MAGIC_COMP_V1:
5357 lcm = info->lti_ea_store;
5358 entry_count = lcm->lcm_entry_count;
5359 if (entry_count == 0)
5362 mirror_count = lcm->lcm_mirror_count + 1;
5363 want_composite = true;
5369 if (append_stripe_count != 0 || append_pool != NULL) {
5372 want_composite = false;
5375 /* realloc default comp entries if necessary */
5376 rc = lod_def_striping_comp_resize(lds, entry_count);
5380 lds->lds_def_comp_cnt = entry_count;
5381 lds->lds_def_striping_is_composite = want_composite;
5382 lds->lds_def_mirror_cnt = mirror_count;
5384 for (i = 0; i < entry_count; i++) {
5385 struct lod_layout_component *llc = &lds->lds_def_comp_entries[i];
5389 * reset llc values, llc_stripes is always NULL in the
5390 * default striping template, llc_pool will be reset
5391 * later below using lod_set_pool().
5393 * XXX At this point llc_pool may point to valid (!)
5394 * kmalloced strings from previous RPCs.
5396 memset(llc, 0, offsetof(typeof(*llc), llc_pool));
5399 v1 = (struct lov_user_md *)((char *)lcm +
5400 lcm->lcm_entries[i].lcme_offset);
5402 if (want_composite) {
5403 llc->llc_extent = lcm->lcm_entries[i].lcme_extent;
5404 /* We only inherit certain flags from the layout */
5405 llc->llc_flags = lcm->lcm_entries[i].lcme_flags &
5406 LCME_TEMPLATE_FLAGS;
5410 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",
5411 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
5414 v1->lmm_stripe_count,
5415 v1->lmm_stripe_size,
5416 v1->lmm_stripe_offset,
5418 append_stripe_count);
5420 if (!lov_pattern_supported(v1->lmm_pattern) &&
5421 !(v1->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
5422 lod_free_def_comp_entries(lds);
5426 llc->llc_stripe_count = v1->lmm_stripe_count;
5427 llc->llc_stripe_size = v1->lmm_stripe_size;
5428 llc->llc_stripe_offset = v1->lmm_stripe_offset;
5429 llc->llc_pattern = v1->lmm_pattern;
5431 if (append_stripe_count != 0 || append_pool != NULL)
5432 llc->llc_pattern = LOV_PATTERN_RAID0;
5434 if (append_stripe_count != 0)
5435 llc->llc_stripe_count = append_stripe_count;
5438 if (append_pool != NULL) {
5440 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
5441 /* XXX: sanity check here */
5442 v3 = (struct lov_user_md_v3 *)v1;
5443 if (v3->lmm_pool_name[0] != '\0')
5444 pool = v3->lmm_pool_name;
5447 lod_set_pool(&llc->llc_pool, pool);
5449 if (append_stripe_count != 0 || append_pool != NULL) {
5450 /* Ignore specific striping for append. */
5451 } else if (v1->lmm_magic == LOV_USER_MAGIC_SPECIFIC) {
5452 v3 = (struct lov_user_md_v3 *)v1;
5453 rc = lod_comp_copy_ost_lists(llc, v3);
5456 } else if (llc->llc_ostlist.op_array &&
5457 llc->llc_ostlist.op_count) {
5458 for (j = 0; j < llc->llc_ostlist.op_count; j++)
5459 llc->llc_ostlist.op_array[j] = -1;
5460 llc->llc_ostlist.op_count = 0;
5464 lds->lds_def_striping_set = 1;
5469 * Get default directory striping.
5471 * \param[in] env execution environment
5472 * \param[in] lo object
5473 * \param[out] lds default striping
5475 * \retval 0 on success
5476 * \retval negative if failed
5478 static int lod_get_default_lmv_striping(const struct lu_env *env,
5479 struct lod_object *lo,
5480 struct lod_default_striping *lds)
5482 struct lmv_user_md *lmu;
5485 lds->lds_dir_def_striping_set = 0;
5487 rc = lod_get_default_lmv_ea(env, lo);
5491 if (rc >= (int)sizeof(*lmu)) {
5492 struct lod_thread_info *info = lod_env_info(env);
5494 lmu = info->lti_ea_store;
5496 lds->lds_dir_def_stripe_count =
5497 le32_to_cpu(lmu->lum_stripe_count);
5498 lds->lds_dir_def_stripe_offset =
5499 le32_to_cpu(lmu->lum_stripe_offset);
5500 lds->lds_dir_def_hash_type =
5501 le32_to_cpu(lmu->lum_hash_type);
5502 lds->lds_dir_def_max_inherit = lmu->lum_max_inherit;
5503 lds->lds_dir_def_max_inherit_rr = lmu->lum_max_inherit_rr;
5504 lds->lds_dir_def_striping_set = 1;
5511 * Get default striping in the object.
5513 * Get object default striping and default directory striping.
5515 * \param[in] env execution environment
5516 * \param[in] lo object
5517 * \param[out] lds default striping
5519 * \retval 0 on success
5520 * \retval negative if failed
5522 static int lod_get_default_striping(const struct lu_env *env,
5523 struct lod_object *lo,
5524 struct lod_default_striping *lds)
5528 rc = lod_get_default_lov_striping(env, lo, lds, NULL);
5529 if (lds->lds_def_striping_set) {
5530 struct lod_thread_info *info = lod_env_info(env);
5531 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5533 rc = lod_verify_striping(env, d, lo, &info->lti_buf, false);
5535 lds->lds_def_striping_set = 0;
5538 rc1 = lod_get_default_lmv_striping(env, lo, lds);
5539 if (rc == 0 && rc1 < 0)
5546 * Apply default striping on object.
5548 * If object striping pattern is not set, set to the one in default striping.
5549 * The default striping is from parent or fs.
5551 * \param[in] lo new object
5552 * \param[in] lds default striping
5553 * \param[in] mode new object's mode
5555 static void lod_striping_from_default(struct lod_object *lo,
5556 const struct lod_default_striping *lds,
5559 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5562 if (lds->lds_def_striping_set && S_ISREG(mode)) {
5563 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
5565 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
5566 lds->lds_def_comp_cnt);
5570 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
5571 if (lds->lds_def_mirror_cnt > 1)
5572 lo->ldo_flr_state = LCM_FL_RDONLY;
5574 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5575 struct lod_layout_component *obj_comp =
5576 &lo->ldo_comp_entries[i];
5577 struct lod_layout_component *def_comp =
5578 &lds->lds_def_comp_entries[i];
5581 "inherit "DFID" file layout from default: flags=%#x size=%hu nr=%u offset=%u pattern=%#x pool=%s\n",
5582 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
5583 def_comp->llc_flags,
5584 def_comp->llc_stripe_size,
5585 def_comp->llc_stripe_count,
5586 def_comp->llc_stripe_offset,
5587 def_comp->llc_pattern,
5588 def_comp->llc_pool ?: "");
5590 *obj_comp = *def_comp;
5591 if (def_comp->llc_pool != NULL) {
5592 /* pointer was copied from def_comp */
5593 obj_comp->llc_pool = NULL;
5594 lod_obj_set_pool(lo, i, def_comp->llc_pool);
5598 if (def_comp->llc_ostlist.op_array &&
5599 def_comp->llc_ostlist.op_count) {
5600 OBD_ALLOC(obj_comp->llc_ostlist.op_array,
5601 obj_comp->llc_ostlist.op_size);
5602 if (!obj_comp->llc_ostlist.op_array)
5604 memcpy(obj_comp->llc_ostlist.op_array,
5605 def_comp->llc_ostlist.op_array,
5606 obj_comp->llc_ostlist.op_size);
5607 } else if (def_comp->llc_ostlist.op_array) {
5608 obj_comp->llc_ostlist.op_array = NULL;
5612 * Don't initialize these fields for plain layout
5613 * (v1/v3) here, they are inherited in the order of
5614 * 'parent' -> 'fs default (root)' -> 'global default
5615 * values for stripe_count & stripe_size'.
5617 * see lod_ah_init().
5619 if (!lo->ldo_is_composite)
5622 lod_adjust_stripe_info(obj_comp, desc, 0);
5624 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
5625 if (lo->ldo_dir_stripe_count == 0)
5626 lo->ldo_dir_stripe_count =
5627 lds->lds_dir_def_stripe_count;
5628 if (lo->ldo_dir_stripe_offset == -1)
5629 lo->ldo_dir_stripe_offset =
5630 lds->lds_dir_def_stripe_offset;
5631 if (lo->ldo_dir_hash_type == LMV_HASH_TYPE_UNKNOWN)
5632 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
5635 "inherit "DFID" dir layout from default: count=%hu offset=%u hash_type=%x\n",
5636 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
5637 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
5638 lo->ldo_dir_hash_type);
5642 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root,
5643 const char *append_pool)
5645 struct lod_layout_component *lod_comp;
5647 if (lo->ldo_comp_cnt == 0)
5650 if (lo->ldo_is_composite)
5653 lod_comp = &lo->ldo_comp_entries[0];
5655 if (lod_comp->llc_stripe_count <= 0 ||
5656 lod_comp->llc_stripe_size <= 0)
5659 if (from_root && (lod_comp->llc_pool == NULL ||
5660 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
5663 if (append_pool && append_pool[0])
5670 * Implementation of dt_object_operations::do_ah_init.
5672 * This method is used to make a decision on the striping configuration for the
5673 * object being created. It can be taken from the \a parent object if it exists,
5674 * or filesystem's default. The resulting configuration (number of stripes,
5675 * stripe size/offset, pool name, hash_type, etc.) is stored in the object
5676 * itself and will be used by the methods like ->doo_declare_create().
5678 * \see dt_object_operations::do_ah_init() in the API description for details.
5680 static void lod_ah_init(const struct lu_env *env,
5681 struct dt_allocation_hint *ah,
5682 struct dt_object *parent,
5683 struct dt_object *child,
5686 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
5687 struct lod_thread_info *info = lod_env_info(env);
5688 struct lod_default_striping *lds = lod_lds_buf_get(env);
5689 struct dt_object *nextp = NULL;
5690 struct dt_object *nextc;
5691 struct lod_object *lp = NULL;
5692 struct lod_object *lc;
5693 struct lov_desc *desc;
5694 struct lod_layout_component *lod_comp;
5700 if (ah->dah_append_stripe_count == -1)
5701 ah->dah_append_stripe_count =
5702 d->lod_ost_descs.ltd_lov_desc.ld_tgt_count;
5704 if (likely(parent)) {
5705 nextp = dt_object_child(parent);
5706 lp = lod_dt_obj(parent);
5709 nextc = dt_object_child(child);
5710 lc = lod_dt_obj(child);
5712 LASSERT(!lod_obj_is_striped(child));
5713 /* default layout template may have been set on the regular file
5714 * when this is called from mdd_create_data() */
5715 if (S_ISREG(child_mode))
5716 lod_free_comp_entries(lc);
5718 if (!dt_object_exists(nextc))
5719 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
5721 if (S_ISDIR(child_mode)) {
5722 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
5724 /* other default values are 0 */
5725 lc->ldo_dir_stripe_offset = -1;
5727 /* no default striping configuration is needed for
5730 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5731 le32_to_cpu(lum1->lum_magic) == LMV_MAGIC_FOREIGN) {
5732 lc->ldo_is_foreign = true;
5733 /* keep stripe_count 0 and stripe_offset -1 */
5734 CDEBUG(D_INFO, "no default striping for foreign dir\n");
5738 if (likely(lp != NULL))
5739 lod_get_default_striping(env, lp, lds);
5741 /* It should always honour the specified stripes */
5742 /* Note: old client (< 2.7)might also do lfs mkdir, whose EA
5743 * will have old magic. In this case, we should ignore the
5744 * stripe count and try to create dir by default stripe.
5746 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5747 (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
5748 le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC)) {
5749 lc->ldo_dir_stripe_count =
5750 le32_to_cpu(lum1->lum_stripe_count);
5751 lc->ldo_dir_stripe_offset =
5752 le32_to_cpu(lum1->lum_stripe_offset);
5753 lc->ldo_dir_hash_type =
5754 le32_to_cpu(lum1->lum_hash_type);
5756 "set dirstripe: count %hu, offset %d, hash %x\n",
5757 lc->ldo_dir_stripe_count,
5758 (int)lc->ldo_dir_stripe_offset,
5759 lc->ldo_dir_hash_type);
5761 if (d->lod_mdt_descs.ltd_lmv_desc.ld_active_tgt_count &&
5762 lc->ldo_dir_stripe_count < 2 &&
5763 lum1->lum_max_inherit != LMV_INHERIT_NONE) {
5764 /* when filesystem-wide default LMV is set, dirs
5765 * will be created on MDT by space usage, but if
5766 * dir is created with "lfs mkdir -c 1 ...", its
5767 * subdirs should be kept on the same MDT. To
5768 * guarantee this, set default LMV for such dir.
5770 lds->lds_dir_def_stripe_count =
5771 le32_to_cpu(lum1->lum_stripe_count);
5772 /* if "-1" stripe offset is set, save current
5773 * MDT index in default LMV.
5775 if (le32_to_cpu(lum1->lum_stripe_offset) ==
5777 lds->lds_dir_def_stripe_offset =
5778 lod2lu_dev(d)->ld_site->ld_seq_site->ss_node_id;
5780 lds->lds_dir_def_stripe_offset =
5781 le32_to_cpu(lum1->lum_stripe_offset);
5782 lds->lds_dir_def_hash_type =
5783 le32_to_cpu(lum1->lum_hash_type);
5784 lds->lds_dir_def_max_inherit =
5785 lum1->lum_max_inherit;
5786 /* it will be decreased by 1 later in setting */
5787 if (lum1->lum_max_inherit >= LMV_INHERIT_END &&
5788 lum1->lum_max_inherit < LMV_INHERIT_MAX)
5789 lds->lds_dir_def_max_inherit++;
5790 lds->lds_dir_def_max_inherit_rr =
5791 lum1->lum_max_inherit_rr;
5792 lds->lds_dir_def_striping_set = 1;
5793 /* don't inherit LOV from ROOT */
5794 if (lds->lds_def_striping_set &&
5795 fid_is_root(lod_object_fid(lp)))
5796 lds->lds_def_striping_set = 0;
5797 lc->ldo_def_striping = lds;
5798 } else if (lds->lds_def_striping_set &&
5799 !fid_is_root(lod_object_fid(lp))) {
5800 /* don't inherit default LMV for "lfs mkdir" */
5801 lds->lds_dir_def_striping_set = 0;
5802 lc->ldo_def_striping = lds;
5805 /* inherit default striping except ROOT */
5806 if ((lds->lds_def_striping_set ||
5807 lds->lds_dir_def_striping_set) &&
5808 !fid_is_root(lod_object_fid(lp)))
5809 lc->ldo_def_striping = lds;
5811 /* transfer defaults LMV to new directory */
5812 lod_striping_from_default(lc, lds, child_mode);
5814 /* set count 0 to create normal directory */
5815 if (lc->ldo_dir_stripe_count == 1)
5816 lc->ldo_dir_stripe_count = 0;
5819 /* shrink the stripe count to max_mdt_stripecount if it is -1
5820 * and max_mdt_stripecount is not 0
5822 if (lc->ldo_dir_stripe_count == (__u16)(-1) &&
5823 d->lod_max_mdt_stripecount)
5824 lc->ldo_dir_stripe_count = d->lod_max_mdt_stripecount;
5826 /* shrink the stripe_count to the avaible MDT count */
5827 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
5828 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)) {
5829 lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
5830 if (lc->ldo_dir_stripe_count == 1)
5831 lc->ldo_dir_stripe_count = 0;
5834 if (!lmv_is_known_hash_type(lc->ldo_dir_hash_type))
5835 lc->ldo_dir_hash_type =
5836 (lc->ldo_dir_hash_type & LMV_HASH_FLAG_KNOWN) |
5837 d->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
5839 CDEBUG(D_INFO, "final dir stripe_count=%hu offset=%d hash=%u\n",
5840 lc->ldo_dir_stripe_count,
5841 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
5846 /* child object regular file*/
5848 if (!lod_object_will_be_striped(S_ISREG(child_mode),
5849 lu_object_fid(&child->do_lu)))
5852 /* If object is going to be striped over OSTs, transfer default
5853 * striping information to the child, so that we can use it
5854 * during declaration and creation.
5856 * Try from the parent first.
5858 if (likely(lp != NULL)) {
5859 rc = lod_get_default_lov_striping(env, lp, lds, ah);
5860 if (rc == 0 && lds->lds_def_striping_set) {
5861 rc = lod_verify_striping(env, d, lp, &info->lti_buf,
5864 lod_striping_from_default(lc, lds, child_mode);
5868 /* Initialize lod_device::lod_md_root object reference */
5869 if (d->lod_md_root == NULL) {
5870 struct dt_object *root;
5871 struct lod_object *lroot;
5873 lu_root_fid(&info->lti_fid);
5874 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
5875 if (!IS_ERR(root)) {
5876 lroot = lod_dt_obj(root);
5878 spin_lock(&d->lod_lock);
5879 if (d->lod_md_root != NULL)
5880 dt_object_put(env, &d->lod_md_root->ldo_obj);
5881 d->lod_md_root = lroot;
5882 spin_unlock(&d->lod_lock);
5886 /* try inherit layout from the root object (fs default) when:
5887 * - parent does not have default layout; or
5888 * - parent has plain(v1/v3) default layout, and some attributes
5889 * are not specified in the default layout;
5891 if (d->lod_md_root != NULL &&
5892 lod_need_inherit_more(lc, true, ah->dah_append_pool)) {
5893 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds,
5895 if (rc || !lds->lds_def_striping_set)
5898 rc = lod_verify_striping(env, d, d->lod_md_root, &info->lti_buf,
5903 if (lc->ldo_comp_cnt == 0) {
5904 lod_striping_from_default(lc, lds, child_mode);
5905 } else if (!lds->lds_def_striping_is_composite) {
5906 struct lod_layout_component *def_comp;
5908 LASSERT(!lc->ldo_is_composite);
5909 lod_comp = &lc->ldo_comp_entries[0];
5910 def_comp = &lds->lds_def_comp_entries[0];
5912 if (lod_comp->llc_stripe_count <= 0)
5913 lod_comp->llc_stripe_count =
5914 def_comp->llc_stripe_count;
5915 if (lod_comp->llc_stripe_size <= 0)
5916 lod_comp->llc_stripe_size =
5917 def_comp->llc_stripe_size;
5918 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT &&
5919 (!lod_comp->llc_pool || !lod_comp->llc_pool[0]))
5920 lod_comp->llc_stripe_offset =
5921 def_comp->llc_stripe_offset;
5922 if (lod_comp->llc_pool == NULL)
5923 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
5928 * fs default striping may not be explicitly set, or historically set
5929 * in config log, use them.
5931 if (lod_need_inherit_more(lc, false, ah->dah_append_pool)) {
5932 if (lc->ldo_comp_cnt == 0) {
5933 rc = lod_alloc_comp_entries(lc, 0, 1);
5935 /* fail to allocate memory, will create a
5936 * non-striped file. */
5938 lc->ldo_is_composite = 0;
5939 lod_comp = &lc->ldo_comp_entries[0];
5940 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
5942 LASSERT(!lc->ldo_is_composite);
5943 lod_comp = &lc->ldo_comp_entries[0];
5944 desc = &d->lod_ost_descs.ltd_lov_desc;
5945 lod_adjust_stripe_info(lod_comp, desc,
5946 ah->dah_append_stripe_count);
5947 if (ah->dah_append_pool && ah->dah_append_pool[0])
5948 lod_obj_set_pool(lc, 0, ah->dah_append_pool);
5955 * Size initialization on late striping.
5957 * Propagate the size of a truncated object to a deferred striping.
5958 * This function handles a special case when truncate was done on a
5959 * non-striped object and now while the striping is being created
5960 * we can't lose that size, so we have to propagate it to the stripes
5963 * \param[in] env execution environment
5964 * \param[in] dt object
5965 * \param[in] th transaction handle
5967 * \retval 0 on success
5968 * \retval negative if failed
5970 static int lod_declare_init_size(const struct lu_env *env,
5971 struct dt_object *dt, struct thandle *th)
5973 struct dt_object *next = dt_object_child(dt);
5974 struct lod_object *lo = lod_dt_obj(dt);
5975 struct dt_object **objects = NULL;
5976 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
5977 uint64_t size, offs;
5978 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
5979 struct lu_extent size_ext;
5982 if (!lod_obj_is_striped(dt))
5985 rc = dt_attr_get(env, next, attr);
5986 LASSERT(attr->la_valid & LA_SIZE);
5990 size = attr->la_size;
5994 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
5995 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5996 struct lod_layout_component *lod_comp;
5997 struct lu_extent *extent;
5999 lod_comp = &lo->ldo_comp_entries[i];
6001 if (lod_comp->llc_stripe == NULL)
6004 extent = &lod_comp->llc_extent;
6005 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
6006 if (!lo->ldo_is_composite ||
6007 lu_extent_is_overlapped(extent, &size_ext)) {
6008 objects = lod_comp->llc_stripe;
6009 stripe_count = lod_comp->llc_stripe_count;
6010 stripe_size = lod_comp->llc_stripe_size;
6013 if (stripe_count == 0)
6016 LASSERT(objects != NULL && stripe_size != 0);
6017 do_div(size, stripe_size);
6018 stripe = do_div(size, stripe_count);
6019 LASSERT(objects[stripe] != NULL);
6021 size = size * stripe_size;
6022 offs = attr->la_size;
6023 size += do_div(offs, stripe_size);
6025 attr->la_valid = LA_SIZE;
6026 attr->la_size = size;
6028 rc = lod_sub_declare_attr_set(env, objects[stripe],
6037 * Declare creation of striped object.
6039 * The function declares creation stripes for a regular object. The function
6040 * also declares whether the stripes will be created with non-zero size if
6041 * previously size was set non-zero on the master object. If object \a dt is
6042 * not local, then only fully defined striping can be applied in \a lovea.
6043 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
6046 * \param[in] env execution environment
6047 * \param[in] dt object
6048 * \param[in] attr attributes the stripes will be created with
6049 * \param[in] lovea a buffer containing striping description
6050 * \param[in] th transaction handle
6052 * \retval 0 on success
6053 * \retval negative if failed
6055 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
6056 struct lu_attr *attr,
6057 const struct lu_buf *lovea, struct thandle *th)
6059 struct lod_thread_info *info = lod_env_info(env);
6060 struct dt_object *next = dt_object_child(dt);
6061 struct lod_object *lo = lod_dt_obj(dt);
6065 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
6066 GOTO(out, rc = -ENOMEM);
6068 if (!dt_object_remote(next)) {
6069 /* choose OST and generate appropriate objects */
6070 rc = lod_prepare_create(env, lo, attr, lovea, th);
6075 * declare storage for striping data
6077 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6079 /* LOD can not choose OST objects for remote objects, i.e.
6080 * stripes must be ready before that. Right now, it can only
6081 * happen during migrate, i.e. migrate process needs to create
6082 * remote regular file (mdd_migrate_create), then the migrate
6083 * process will provide stripeEA. */
6084 LASSERT(lovea != NULL);
6085 info->lti_buf = *lovea;
6088 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
6089 XATTR_NAME_LOV, 0, th);
6094 * if striping is created with local object's size > 0,
6095 * we have to propagate this size to specific object
6096 * the case is possible only when local object was created previously
6098 if (dt_object_exists(next))
6099 rc = lod_declare_init_size(env, dt, th);
6102 /* failed to create striping or to set initial size, let's reset
6103 * config so that others don't get confused */
6105 lod_striping_free(env, lo);
6111 * Whether subdirectories under \a dt should be created on MDTs by space QoS
6113 * If LMV_HASH_FLAG_SPACE is set on directory default layout, its subdirectories
6114 * should be created on MDT by space QoS.
6116 * \param[in] env execution environment
6117 * \param[in] dev lu device
6118 * \param[in] dt object
6120 * \retval 1 if directory should create subdir by space usage
6122 * \retval -ev if failed
6124 static inline int dt_object_qos_mkdir(const struct lu_env *env,
6125 struct lu_device *dev,
6126 struct dt_object *dt)
6128 struct lod_thread_info *info = lod_env_info(env);
6129 struct lu_object *obj;
6130 struct lod_object *lo;
6131 struct lmv_user_md *lmu;
6134 obj = lu_object_find_slice(env, dev, lu_object_fid(&dt->do_lu), NULL);
6136 return PTR_ERR(obj);
6138 lo = lu2lod_obj(obj);
6140 rc = lod_get_default_lmv_ea(env, lo);
6141 dt_object_put(env, dt);
6145 if (rc < (int)sizeof(*lmu))
6148 lmu = info->lti_ea_store;
6149 return le32_to_cpu(lmu->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
6153 * Implementation of dt_object_operations::do_declare_create.
6155 * The method declares creation of a new object. If the object will be striped,
6156 * then helper functions are called to find FIDs for the stripes, declare
6157 * creation of the stripes and declare initialization of the striping
6158 * information to be stored in the master object.
6160 * \see dt_object_operations::do_declare_create() in the API description
6163 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
6164 struct lu_attr *attr,
6165 struct dt_allocation_hint *hint,
6166 struct dt_object_format *dof, struct thandle *th)
6168 struct dt_object *next = dt_object_child(dt);
6169 struct lod_object *lo = lod_dt_obj(dt);
6178 * first of all, we declare creation of local object
6180 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
6185 * it's lod_ah_init() that has decided the object will be striped
6187 if (dof->dof_type == DFT_REGULAR) {
6188 /* callers don't want stripes */
6189 /* XXX: all tricky interactions with ->ah_make_hint() decided
6190 * to use striping, then ->declare_create() behaving differently
6191 * should be cleaned */
6192 if (dof->u.dof_reg.striped != 0)
6193 rc = lod_declare_striped_create(env, dt, attr,
6195 } else if (dof->dof_type == DFT_DIR) {
6196 struct seq_server_site *ss;
6197 struct lu_buf buf = { NULL };
6198 struct lu_buf *lmu = NULL;
6200 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
6202 /* If the parent has default stripeEA, and client
6203 * did not find it before sending create request,
6204 * then MDT will return -EREMOTE, and client will
6205 * retrieve the default stripeEA and re-create the
6208 * Note: if dah_eadata != NULL, it means creating the
6209 * striped directory with specified stripeEA, then it
6210 * should ignore the default stripeEA */
6211 if (hint != NULL && hint->dah_eadata == NULL) {
6212 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
6213 GOTO(out, rc = -EREMOTE);
6215 if (lo->ldo_dir_stripe_offset != LMV_OFFSET_DEFAULT &&
6216 lo->ldo_dir_stripe_offset != ss->ss_node_id) {
6217 struct lod_device *lod;
6218 struct lu_tgt_desc *mdt = NULL;
6219 bool found_mdt = false;
6221 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6222 lod_foreach_mdt(lod, mdt) {
6223 if (mdt->ltd_index ==
6224 lo->ldo_dir_stripe_offset) {
6230 /* If the MDT indicated by stripe_offset can be
6231 * found, then tell client to resend the create
6232 * request to the correct MDT, otherwise return
6233 * error to client */
6235 GOTO(out, rc = -EREMOTE);
6237 GOTO(out, rc = -EINVAL);
6239 } else if (hint && hint->dah_eadata) {
6241 lmu->lb_buf = (void *)hint->dah_eadata;
6242 lmu->lb_len = hint->dah_eadata_len;
6245 rc = lod_declare_dir_striping_create(env, dt, attr, lmu, dof,
6249 /* failed to create striping or to set initial size, let's reset
6250 * config so that others don't get confused */
6252 lod_striping_free(env, lo);
6257 * Generate component ID for new created component.
6259 * \param[in] lo LOD object
6260 * \param[in] comp_idx index of ldo_comp_entries
6262 * \retval component ID on success
6263 * \retval LCME_ID_INVAL on failure
6265 static __u32 lod_gen_component_id(struct lod_object *lo,
6266 int mirror_id, int comp_idx)
6268 struct lod_layout_component *lod_comp;
6269 __u32 id, start, end;
6272 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
6274 lod_obj_inc_layout_gen(lo);
6275 id = lo->ldo_layout_gen;
6276 if (likely(id <= SEQ_ID_MAX))
6277 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
6279 /* Layout generation wraps, need to check collisions. */
6280 start = id & SEQ_ID_MASK;
6283 for (id = start; id <= end; id++) {
6284 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6285 lod_comp = &lo->ldo_comp_entries[i];
6286 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
6289 /* Found the ununsed ID */
6290 if (i == lo->ldo_comp_cnt)
6291 RETURN(pflr_id(mirror_id, id));
6293 if (end == LCME_ID_MAX) {
6295 end = min(lo->ldo_layout_gen & LCME_ID_MASK,
6296 (__u32)(LCME_ID_MAX - 1));
6300 RETURN(LCME_ID_INVAL);
6304 * Creation of a striped regular object.
6306 * The function is called to create the stripe objects for a regular
6307 * striped file. This can happen at the initial object creation or
6308 * when the caller asks LOD to do so using ->do_xattr_set() method
6309 * (so called late striping). Notice all the information are already
6310 * prepared in the form of the list of objects (ldo_stripe field).
6311 * This is done during declare phase.
6313 * \param[in] env execution environment
6314 * \param[in] dt object
6315 * \param[in] attr attributes the stripes will be created with
6316 * \param[in] dof format of stripes (see OSD API description)
6317 * \param[in] th transaction handle
6319 * \retval 0 on success
6320 * \retval negative if failed
6322 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
6323 struct lu_attr *attr, struct dt_object_format *dof,
6326 struct lod_layout_component *lod_comp;
6327 struct lod_object *lo = lod_dt_obj(dt);
6332 mutex_lock(&lo->ldo_layout_mutex);
6334 LASSERT((lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL) ||
6335 lo->ldo_is_foreign);
6337 mirror_id = 0; /* non-flr file's mirror_id is 0 */
6338 if (lo->ldo_mirror_count > 1) {
6339 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6340 lod_comp = &lo->ldo_comp_entries[i];
6341 if (lod_comp->llc_id != LCME_ID_INVAL &&
6342 mirror_id_of(lod_comp->llc_id) > mirror_id)
6343 mirror_id = mirror_id_of(lod_comp->llc_id);
6347 /* create all underlying objects */
6348 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6349 lod_comp = &lo->ldo_comp_entries[i];
6351 if (lod_comp->llc_id == LCME_ID_INVAL) {
6352 /* only the component of FLR layout with more than 1
6353 * mirror has mirror ID in its component ID.
6355 if (lod_comp->llc_extent.e_start == 0 &&
6356 lo->ldo_mirror_count > 1)
6359 lod_comp->llc_id = lod_gen_component_id(lo,
6361 if (lod_comp->llc_id == LCME_ID_INVAL)
6362 GOTO(out, rc = -ERANGE);
6365 if (lod_comp_inited(lod_comp))
6368 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
6369 lod_comp_set_init(lod_comp);
6371 if (lov_pattern(lod_comp->llc_pattern) == LOV_PATTERN_MDT)
6372 lod_comp_set_init(lod_comp);
6374 if (lod_comp->llc_stripe == NULL)
6377 LASSERT(lod_comp->llc_stripe_count);
6378 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6379 struct dt_object *object = lod_comp->llc_stripe[j];
6380 LASSERT(object != NULL);
6381 rc = lod_sub_create(env, object, attr, NULL, dof, th);
6385 lod_comp_set_init(lod_comp);
6388 rc = lod_fill_mirrors(lo);
6392 lo->ldo_comp_cached = 1;
6394 rc = lod_generate_and_set_lovea(env, lo, th);
6398 mutex_unlock(&lo->ldo_layout_mutex);
6403 lod_striping_free_nolock(env, lo);
6404 mutex_unlock(&lo->ldo_layout_mutex);
6409 static inline bool lod_obj_is_dom(struct dt_object *dt)
6411 struct lod_object *lo = lod_dt_obj(dt);
6413 if (!dt_object_exists(dt_object_child(dt)))
6416 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
6419 if (!lo->ldo_comp_cnt)
6422 return (lov_pattern(lo->ldo_comp_entries[0].llc_pattern) ==
6427 * Implementation of dt_object_operations::do_create.
6429 * If any of preceeding methods (like ->do_declare_create(),
6430 * ->do_ah_init(), etc) chose to create a striped object,
6431 * then this method will create the master and the stripes.
6433 * \see dt_object_operations::do_create() in the API description for details.
6435 static int lod_create(const struct lu_env *env, struct dt_object *dt,
6436 struct lu_attr *attr, struct dt_allocation_hint *hint,
6437 struct dt_object_format *dof, struct thandle *th)
6442 /* create local object */
6443 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
6447 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
6448 (lod_obj_is_striped(dt) || lod_obj_is_dom(dt)) &&
6449 dof->u.dof_reg.striped != 0) {
6450 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
6451 rc = lod_striped_create(env, dt, attr, dof, th);
6458 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
6459 struct dt_object *dt, struct thandle *th,
6460 int comp_idx, int stripe_idx,
6461 struct lod_obj_stripe_cb_data *data)
6463 if (data->locd_declare)
6464 return lod_sub_declare_destroy(env, dt, th);
6466 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
6467 stripe_idx == cfs_fail_val)
6468 return lod_sub_destroy(env, dt, th);
6474 * Implementation of dt_object_operations::do_declare_destroy.
6476 * If the object is a striped directory, then the function declares reference
6477 * removal from the master object (this is an index) to the stripes and declares
6478 * destroy of all the stripes. In all the cases, it declares an intention to
6479 * destroy the object itself.
6481 * \see dt_object_operations::do_declare_destroy() in the API description
6484 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
6487 struct dt_object *next = dt_object_child(dt);
6488 struct lod_object *lo = lod_dt_obj(dt);
6489 struct lod_thread_info *info = lod_env_info(env);
6490 struct dt_object *stripe;
6491 char *stripe_name = info->lti_key;
6497 * load striping information, notice we don't do this when object
6498 * is being initialized as we don't need this information till
6499 * few specific cases like destroy, chown
6501 rc = lod_striping_load(env, lo);
6505 /* declare destroy for all underlying objects */
6506 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6507 rc = next->do_ops->do_index_try(env, next,
6508 &dt_directory_features);
6512 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6513 stripe = lo->ldo_stripe[i];
6517 rc = lod_sub_declare_ref_del(env, next, th);
6521 snprintf(stripe_name, sizeof(info->lti_key),
6523 PFID(lu_object_fid(&stripe->do_lu)), i);
6524 rc = lod_sub_declare_delete(env, next,
6525 (const struct dt_key *)stripe_name, th);
6532 * we declare destroy for the local object
6534 rc = lod_sub_declare_destroy(env, next, th);
6538 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
6539 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
6542 if (!lod_obj_is_striped(dt))
6545 /* declare destroy all striped objects */
6546 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6547 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6548 stripe = lo->ldo_stripe[i];
6552 if (!dt_object_exists(stripe))
6555 rc = lod_sub_declare_ref_del(env, stripe, th);
6559 rc = lod_sub_declare_destroy(env, stripe, th);
6564 struct lod_obj_stripe_cb_data data = { { 0 } };
6566 data.locd_declare = true;
6567 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
6568 rc = lod_obj_for_each_stripe(env, lo, th, &data);
6575 * Implementation of dt_object_operations::do_destroy.
6577 * If the object is a striped directory, then the function removes references
6578 * from the master object (this is an index) to the stripes and destroys all
6579 * the stripes. In all the cases, the function destroys the object itself.
6581 * \see dt_object_operations::do_destroy() in the API description for details.
6583 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
6586 struct dt_object *next = dt_object_child(dt);
6587 struct lod_object *lo = lod_dt_obj(dt);
6588 struct lod_thread_info *info = lod_env_info(env);
6589 char *stripe_name = info->lti_key;
6590 struct dt_object *stripe;
6596 /* destroy sub-stripe of master object */
6597 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6598 rc = next->do_ops->do_index_try(env, next,
6599 &dt_directory_features);
6603 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6604 stripe = lo->ldo_stripe[i];
6608 rc = lod_sub_ref_del(env, next, th);
6612 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
6613 PFID(lu_object_fid(&stripe->do_lu)), i);
6615 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
6616 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
6617 PFID(lu_object_fid(&stripe->do_lu)));
6619 rc = lod_sub_delete(env, next,
6620 (const struct dt_key *)stripe_name, th);
6626 rc = lod_sub_destroy(env, next, th);
6630 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
6631 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
6634 if (!lod_obj_is_striped(dt))
6637 /* destroy all striped objects */
6638 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6639 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6640 stripe = lo->ldo_stripe[i];
6644 if (!dt_object_exists(stripe))
6647 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
6648 i == cfs_fail_val) {
6649 dt_write_lock(env, stripe, DT_TGT_CHILD);
6650 rc = lod_sub_ref_del(env, stripe, th);
6651 dt_write_unlock(env, stripe);
6655 rc = lod_sub_destroy(env, stripe, th);
6661 struct lod_obj_stripe_cb_data data = { { 0 } };
6663 data.locd_declare = false;
6664 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
6665 rc = lod_obj_for_each_stripe(env, lo, th, &data);
6672 * Implementation of dt_object_operations::do_declare_ref_add.
6674 * \see dt_object_operations::do_declare_ref_add() in the API description
6677 static int lod_declare_ref_add(const struct lu_env *env,
6678 struct dt_object *dt, struct thandle *th)
6680 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
6684 * Implementation of dt_object_operations::do_ref_add.
6686 * \see dt_object_operations::do_ref_add() in the API description for details.
6688 static int lod_ref_add(const struct lu_env *env,
6689 struct dt_object *dt, struct thandle *th)
6691 return lod_sub_ref_add(env, dt_object_child(dt), th);
6695 * Implementation of dt_object_operations::do_declare_ref_del.
6697 * \see dt_object_operations::do_declare_ref_del() in the API description
6700 static int lod_declare_ref_del(const struct lu_env *env,
6701 struct dt_object *dt, struct thandle *th)
6703 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
6707 * Implementation of dt_object_operations::do_ref_del
6709 * \see dt_object_operations::do_ref_del() in the API description for details.
6711 static int lod_ref_del(const struct lu_env *env,
6712 struct dt_object *dt, struct thandle *th)
6714 return lod_sub_ref_del(env, dt_object_child(dt), th);
6718 * Implementation of dt_object_operations::do_object_sync.
6720 * \see dt_object_operations::do_object_sync() in the API description
6723 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
6724 __u64 start, __u64 end)
6726 return dt_object_sync(env, dt_object_child(dt), start, end);
6730 * Implementation of dt_object_operations::do_object_unlock.
6732 * Used to release LDLM lock(s).
6734 * \see dt_object_operations::do_object_unlock() in the API description
6737 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
6738 struct ldlm_enqueue_info *einfo,
6739 union ldlm_policy_data *policy)
6741 struct lod_object *lo = lod_dt_obj(dt);
6742 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
6743 int slave_locks_size;
6747 if (slave_locks == NULL)
6750 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
6751 /* Note: for remote lock for single stripe dir, MDT will cancel
6752 * the lock by lockh directly */
6753 LASSERT(!dt_object_remote(dt_object_child(dt)));
6755 /* locks were unlocked in MDT layer */
6756 for (i = 0; i < slave_locks->ha_count; i++)
6757 LASSERT(!lustre_handle_is_used(&slave_locks->ha_handles[i]));
6760 * NB, ha_count may not equal to ldo_dir_stripe_count, because dir
6761 * layout may change, e.g., shrink dir layout after migration.
6763 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6764 if (lo->ldo_stripe[i])
6765 dt_invalidate(env, lo->ldo_stripe[i]);
6768 slave_locks_size = offsetof(typeof(*slave_locks),
6769 ha_handles[slave_locks->ha_count]);
6770 OBD_FREE(slave_locks, slave_locks_size);
6771 einfo->ei_cbdata = NULL;
6777 * Implementation of dt_object_operations::do_object_lock.
6779 * Used to get LDLM lock on the non-striped and striped objects.
6781 * \see dt_object_operations::do_object_lock() in the API description
6784 static int lod_object_lock(const struct lu_env *env,
6785 struct dt_object *dt,
6786 struct lustre_handle *lh,
6787 struct ldlm_enqueue_info *einfo,
6788 union ldlm_policy_data *policy)
6790 struct lod_object *lo = lod_dt_obj(dt);
6791 int slave_locks_size;
6792 struct lustre_handle_array *slave_locks = NULL;
6797 /* remote object lock */
6798 if (!einfo->ei_enq_slave) {
6799 LASSERT(dt_object_remote(dt));
6800 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
6804 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
6807 rc = lod_striping_load(env, lo);
6812 if (lo->ldo_dir_stripe_count <= 1)
6815 slave_locks_size = offsetof(typeof(*slave_locks),
6816 ha_handles[lo->ldo_dir_stripe_count]);
6817 /* Freed in lod_object_unlock */
6818 OBD_ALLOC(slave_locks, slave_locks_size);
6821 slave_locks->ha_count = lo->ldo_dir_stripe_count;
6823 /* striped directory lock */
6824 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6825 struct lustre_handle lockh;
6826 struct ldlm_res_id *res_id;
6827 struct dt_object *stripe;
6829 stripe = lo->ldo_stripe[i];
6833 res_id = &lod_env_info(env)->lti_res_id;
6834 fid_build_reg_res_name(lu_object_fid(&stripe->do_lu), res_id);
6835 einfo->ei_res_id = res_id;
6837 if (dt_object_remote(stripe)) {
6838 set_bit(i, (void *)slave_locks->ha_map);
6839 rc = dt_object_lock(env, stripe, &lockh, einfo, policy);
6841 struct ldlm_namespace *ns = einfo->ei_namespace;
6842 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
6843 ldlm_completion_callback completion = einfo->ei_cb_cp;
6844 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
6846 if (einfo->ei_mode == LCK_PW ||
6847 einfo->ei_mode == LCK_EX)
6848 dlmflags |= LDLM_FL_COS_INCOMPAT;
6850 LASSERT(ns != NULL);
6851 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS,
6852 policy, einfo->ei_mode,
6853 &dlmflags, blocking,
6855 NULL, 0, LVB_T_NONE,
6860 ldlm_lock_decref_and_cancel(
6861 &slave_locks->ha_handles[i],
6863 OBD_FREE(slave_locks, slave_locks_size);
6866 slave_locks->ha_handles[i] = lockh;
6868 einfo->ei_cbdata = slave_locks;
6874 * Implementation of dt_object_operations::do_invalidate.
6876 * \see dt_object_operations::do_invalidate() in the API description for details
6878 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
6880 return dt_invalidate(env, dt_object_child(dt));
6883 static int lod_declare_instantiate_components(const struct lu_env *env,
6884 struct lod_object *lo,
6888 struct lod_thread_info *info = lod_env_info(env);
6893 LASSERT(info->lti_count < lo->ldo_comp_cnt);
6895 for (i = 0; i < info->lti_count; i++) {
6896 rc = lod_qos_prep_create(env, lo, NULL, th,
6897 info->lti_comp_idx[i], reserve);
6903 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6904 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
6905 &info->lti_buf, XATTR_NAME_LOV, 0, th);
6912 * Check OSTs for an existing component for further extension
6914 * Checks if OSTs are still healthy and not out of space. Gets free space
6915 * on OSTs (relative to allocation watermark rmb_low) and compares to
6916 * the proposed new_end for this component.
6918 * Decides whether or not to extend a component on its current OSTs.
6920 * \param[in] env execution environment for this thread
6921 * \param[in] lo object we're checking
6922 * \param[in] index index of this component
6923 * \param[in] extension_size extension size for this component
6924 * \param[in] extent layout extent for requested operation
6925 * \param[in] comp_extent extension component extent
6926 * \param[in] write if this is write operation
6928 * \retval true - OK to extend on current OSTs
6929 * \retval false - do not extend on current OSTs
6931 static bool lod_sel_osts_allowed(const struct lu_env *env,
6932 struct lod_object *lo,
6933 int index, __u64 reserve,
6934 struct lu_extent *extent,
6935 struct lu_extent *comp_extent, int write)
6937 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[index];
6938 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6939 struct lod_thread_info *tinfo = lod_env_info(env);
6940 struct obd_statfs *sfs = &tinfo->lti_osfs;
6941 __u64 available = 0;
6947 LASSERT(lod_comp->llc_stripe_count != 0);
6949 lod_getref(&lod->lod_ost_descs);
6950 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
6951 int index = lod_comp->llc_ost_indices[i];
6952 struct lod_tgt_desc *ost = OST_TGT(lod, index);
6953 struct obd_statfs_info info = { 0 };
6954 int j, repeated = 0;
6958 /* Get the number of times this OST repeats in this component.
6959 * Note: inter-component repeats are not counted as this is
6960 * considered as a rare case: we try to not repeat OST in other
6961 * components if possible. */
6962 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6963 if (index != lod_comp->llc_ost_indices[j])
6966 /* already handled */
6972 if (j < lod_comp->llc_stripe_count)
6975 if (!test_bit(index, lod->lod_ost_bitmap)) {
6976 CDEBUG(D_LAYOUT, "ost %d no longer present\n", index);
6981 rc = dt_statfs_info(env, ost->ltd_tgt, sfs, &info);
6983 CDEBUG(D_LAYOUT, "statfs failed for ost %d, error %d\n",
6989 if (sfs->os_state & OS_STATFS_ENOSPC ||
6990 sfs->os_state & OS_STATFS_READONLY ||
6991 sfs->os_state & OS_STATFS_DEGRADED) {
6992 CDEBUG(D_LAYOUT, "ost %d is not availble for SEL "
6993 "extension, state %u\n", index, sfs->os_state);
6999 available = sfs->os_bavail * sfs->os_bsize;
7000 /* 'available' is relative to the allocation threshold */
7001 available -= (__u64) info.os_reserved_mb_low << 20;
7003 CDEBUG(D_LAYOUT, "ost %d lowwm: %d highwm: %d, "
7004 "%llu %% blocks available, %llu %% blocks free\n",
7005 index, info.os_reserved_mb_low, info.os_reserved_mb_high,
7006 (100ull * sfs->os_bavail) / sfs->os_blocks,
7007 (100ull * sfs->os_bfree) / sfs->os_blocks);
7009 if (reserve * repeated > available) {
7011 CDEBUG(D_LAYOUT, "low space on ost %d, available %llu "
7012 "< extension size %llu repeated %d\n", index,
7013 available, reserve, repeated);
7017 lod_putref(lod, &lod->lod_ost_descs);
7023 * Adjust extents after component removal
7025 * When we remove an extension component, we move the start of the next
7026 * component to match the start of the extension component, so no space is left
7029 * \param[in] env execution environment for this thread
7030 * \param[in] lo object
7031 * \param[in] max_comp layout component
7032 * \param[in] index index of this component
7034 * \retval 0 on success
7035 * \retval negative errno on error
7037 static void lod_sel_adjust_extents(const struct lu_env *env,
7038 struct lod_object *lo,
7039 int max_comp, int index)
7041 struct lod_layout_component *lod_comp = NULL;
7042 struct lod_layout_component *next = NULL;
7043 struct lod_layout_component *prev = NULL;
7044 __u64 new_start = 0;
7048 /* Extension space component */
7049 lod_comp = &lo->ldo_comp_entries[index];
7050 next = &lo->ldo_comp_entries[index + 1];
7051 prev = &lo->ldo_comp_entries[index - 1];
7053 LASSERT(lod_comp != NULL && prev != NULL && next != NULL);
7054 LASSERT(lod_comp->llc_flags & LCME_FL_EXTENSION);
7056 /* Previous is being removed */
7057 if (prev && prev->llc_id == LCME_ID_INVAL)
7058 new_start = prev->llc_extent.e_start;
7060 new_start = lod_comp->llc_extent.e_start;
7062 for (i = index + 1; i < max_comp; i++) {
7063 lod_comp = &lo->ldo_comp_entries[i];
7065 start = lod_comp->llc_extent.e_start;
7066 lod_comp->llc_extent.e_start = new_start;
7068 /* We only move zero length extendable components */
7069 if (!(start == lod_comp->llc_extent.e_end))
7072 LASSERT(!(lod_comp->llc_flags & LCME_FL_INIT));
7074 lod_comp->llc_extent.e_end = new_start;
7078 /* Calculate the proposed 'new end' for a component we're extending */
7079 static __u64 lod_extension_new_end(__u64 extension_size, __u64 extent_end,
7080 __u32 stripe_size, __u64 component_end,
7081 __u64 extension_end)
7085 LASSERT(extension_size != 0 && stripe_size != 0);
7087 /* Round up to extension size */
7088 if (extent_end == OBD_OBJECT_EOF) {
7089 new_end = OBD_OBJECT_EOF;
7091 /* Add at least extension_size to the previous component_end,
7092 * covering the req layout extent */
7093 new_end = max(extent_end - component_end, extension_size);
7094 new_end = roundup(new_end, extension_size);
7095 new_end += component_end;
7097 /* Component end must be min stripe size aligned */
7098 if (new_end % stripe_size) {
7099 CDEBUG(D_LAYOUT, "new component end is not aligned "
7100 "by the stripe size %u: [%llu, %llu) ext size "
7101 "%llu new end %llu, aligning\n",
7102 stripe_size, component_end, extent_end,
7103 extension_size, new_end);
7104 new_end = roundup(new_end, stripe_size);
7108 if (new_end < extent_end)
7109 new_end = OBD_OBJECT_EOF;
7112 /* Don't extend past the end of the extension component */
7113 if (new_end > extension_end)
7114 new_end = extension_end;
7120 * Calculate the exact reservation (per-OST extension_size) on the OSTs being
7121 * instantiated. It needs to be calculated in advance and taken into account at
7122 * the instantiation time, because otherwise lod_statfs_and_check() may consider
7123 * an OST as OK, but SEL needs its extension_size to fit the free space and the
7124 * OST may turn out to be low-on-space, thus inappropriate OST may be used and
7127 * \param[in] lod_comp lod component we are checking
7129 * \retval size to reserved on each OST of lod_comp's stripe.
7131 static __u64 lod_sel_stripe_reserved(struct lod_layout_component *lod_comp)
7133 /* extension_size is file level, so we must divide by stripe count to
7134 * compare it to available space on a single OST */
7135 return lod_comp->llc_stripe_size * SEL_UNIT_SIZE /
7136 lod_comp->llc_stripe_count;
7139 /* As lod_sel_handler() could be re-entered for the same component several
7140 * times, this is the data for the next call. Fields could be changed to
7141 * component indexes when needed, (e.g. if there is no need to instantiate
7142 * all the previous components up to the current position) to tell the caller
7143 * where to start over from. */
7150 * Process extent updates for a particular layout component
7152 * Handle layout updates for a particular extension space component touched by
7153 * a layout update operation. Core function of self-extending PFL feature.
7155 * In general, this function processes exactly *one* stage of an extension
7156 * operation, modifying the layout accordingly, then returns to the caller.
7157 * The caller is responsible for restarting processing with the new layout,
7158 * which may repeatedly return to this function until the extension updates
7161 * This function does one of a few things to the layout:
7162 * 1. Extends the component before the current extension space component to
7163 * allow it to accomodate the requested operation (if space/policy permit that
7164 * component to continue on its current OSTs)
7166 * 2. If extension of the existing component fails, we do one of two things:
7167 * a. If there is a component after the extension space, we remove the
7168 * extension space component, move the start of the next component down
7169 * accordingly, then notify the caller to restart processing w/the new
7171 * b. If there is no following component, we try repeating the current
7172 * component, creating a new component using the current one as a
7173 * template (keeping its stripe properties but not specific striping),
7174 * and try assigning striping for this component. If there is sufficient
7175 * free space on the OSTs chosen for this component, it is instantiated
7176 * and i/o continues there.
7178 * If there is not sufficient space on the new OSTs, we remove this new
7179 * component & extend the current component.
7181 * Note further that uninited components followed by extension space can be zero
7182 * length meaning that we will try to extend them before initializing them, and
7183 * if that fails, they will be removed without initialization.
7185 * 3. If we extend to/beyond the end of an extension space component, that
7186 * component is exhausted (all of its range has been given to real components),
7187 * so we remove it and restart processing.
7189 * \param[in] env execution environment for this thread
7190 * \param[in,out] lo object to update the layout of
7191 * \param[in] extent layout extent for requested operation, update
7192 * layout to fit this operation
7193 * \param[in] th transaction handle for this operation
7194 * \param[in,out] max_comp the highest comp for the portion of the layout
7195 * we are operating on (For FLR, the chosen
7196 * replica). Updated because we may remove
7198 * \param[in] index index of the extension space component we're
7200 * \param[in] write if this is write op
7201 * \param[in,out] force if the extension is to be forced; set here
7202 to force it on the 2nd call for the same
7205 * \retval 0 on success
7206 * \retval negative errno on error
7208 static int lod_sel_handler(const struct lu_env *env,
7209 struct lod_object *lo,
7210 struct lu_extent *extent,
7211 struct thandle *th, int *max_comp,
7212 int index, int write,
7213 struct sel_data *sd)
7215 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7216 struct lod_thread_info *info = lod_env_info(env);
7217 struct lod_layout_component *lod_comp;
7218 struct lod_layout_component *prev;
7219 struct lod_layout_component *next = NULL;
7220 __u64 extension_size, reserve;
7227 /* First component cannot be extension space */
7229 CERROR("%s: "DFID" first component cannot be extension space\n",
7230 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
7234 lod_comp = &lo->ldo_comp_entries[index];
7235 prev = &lo->ldo_comp_entries[index - 1];
7236 if ((index + 1) < *max_comp)
7237 next = &lo->ldo_comp_entries[index + 1];
7239 /* extension size uses the stripe size field as KiB */
7240 extension_size = lod_comp->llc_stripe_size * SEL_UNIT_SIZE;
7242 CDEBUG(D_LAYOUT, "prev start %llu, extension start %llu, extension end"
7243 " %llu, extension size %llu\n", prev->llc_extent.e_start,
7244 lod_comp->llc_extent.e_start, lod_comp->llc_extent.e_end,
7247 /* Two extension space components cannot be adjacent & extension space
7248 * components cannot be init */
7249 if ((prev->llc_flags & LCME_FL_EXTENSION) ||
7250 !(ergo(next, !(next->llc_flags & LCME_FL_EXTENSION))) ||
7251 lod_comp_inited(lod_comp)) {
7252 CERROR("%s: "DFID" invalid extension space components\n",
7253 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
7257 reserve = lod_sel_stripe_reserved(lod_comp);
7259 if (!prev->llc_stripe) {
7260 CDEBUG(D_LAYOUT, "Previous component not inited\n");
7261 info->lti_count = 1;
7262 info->lti_comp_idx[0] = index - 1;
7263 rc = lod_declare_instantiate_components(env, lo, th, reserve);
7264 /* ENOSPC tells us we can't use this component. If there is
7265 * a next or we are repeating, we either spill over (next) or
7266 * extend the original comp (repeat). Otherwise, return the
7267 * error to the user. */
7268 if (rc == -ENOSPC && (next || sd->sd_repeat))
7274 if (sd->sd_force == 0 && rc == 0)
7275 rc = !lod_sel_osts_allowed(env, lo, index - 1, reserve, extent,
7276 &lod_comp->llc_extent, write);
7278 repeated = !!(sd->sd_repeat);
7282 /* Extend previous component */
7284 new_end = lod_extension_new_end(extension_size, extent->e_end,
7285 prev->llc_stripe_size,
7286 prev->llc_extent.e_end,
7287 lod_comp->llc_extent.e_end);
7289 CDEBUG(D_LAYOUT, "new end %llu\n", new_end);
7290 lod_comp->llc_extent.e_start = new_end;
7291 prev->llc_extent.e_end = new_end;
7293 if (prev->llc_extent.e_end == lod_comp->llc_extent.e_end) {
7294 CDEBUG(D_LAYOUT, "Extension component exhausted\n");
7295 lod_comp->llc_id = LCME_ID_INVAL;
7299 /* rc == 1, failed to extend current component */
7302 /* Normal 'spillover' case - Remove the extension
7303 * space component & bring down the start of the next
7305 lod_comp->llc_id = LCME_ID_INVAL;
7307 if (!(prev->llc_flags & LCME_FL_INIT)) {
7308 prev->llc_id = LCME_ID_INVAL;
7311 lod_sel_adjust_extents(env, lo, *max_comp, index);
7312 } else if (lod_comp_inited(prev)) {
7313 /* If there is no next, and the previous component is
7314 * INIT'ed, try repeating the previous component. */
7315 LASSERT(repeated == 0);
7316 rc = lod_layout_repeat_comp(env, lo, index - 1);
7320 /* The previous component is a repeated component.
7321 * Record this so we don't keep trying to repeat it. */
7324 /* If the previous component is not INIT'ed, this may
7325 * be a component we have just instantiated but failed
7326 * to extend. Or even a repeated component we failed
7327 * to prepare a striping for. Do not repeat but instead
7328 * remove the repeated component & force the extention
7329 * of the original one */
7332 prev->llc_id = LCME_ID_INVAL;
7339 rc = lod_layout_del_prep_layout(env, lo, NULL);
7342 LASSERTF(-rc == change,
7343 "number deleted %d != requested %d\n", -rc,
7346 *max_comp = *max_comp + change;
7348 /* lod_del_prep_layout reallocates ldo_comp_entries, so we must
7349 * refresh these pointers before using them */
7350 lod_comp = &lo->ldo_comp_entries[index];
7351 prev = &lo->ldo_comp_entries[index - 1];
7352 CDEBUG(D_LAYOUT, "After extent updates: prev start %llu, current start "
7353 "%llu, current end %llu max_comp %d ldo_comp_cnt %d\n",
7354 prev->llc_extent.e_start, lod_comp->llc_extent.e_start,
7355 lod_comp->llc_extent.e_end, *max_comp, lo->ldo_comp_cnt);
7357 /* Layout changed successfully */
7362 * Declare layout extent updates
7364 * Handles extensions. Identifies extension components touched by current
7365 * operation and passes them to processing function.
7367 * Restarts with updated layouts from the processing function until the current
7368 * operation no longer touches an extension space component.
7370 * \param[in] env execution environment for this thread
7371 * \param[in,out] lo object to update the layout of
7372 * \param[in] extent layout extent for requested operation, update layout to
7373 * fit this operation
7374 * \param[in] th transaction handle for this operation
7375 * \param[in] pick identifies chosen mirror for FLR layouts
7376 * \param[in] write if this is write op
7378 * \retval 1 on layout changed, 0 on no change
7379 * \retval negative errno on error
7381 static int lod_declare_update_extents(const struct lu_env *env,
7382 struct lod_object *lo, struct lu_extent *extent,
7383 struct thandle *th, int pick, int write)
7385 struct lod_thread_info *info = lod_env_info(env);
7386 struct lod_layout_component *lod_comp;
7387 bool layout_changed = false;
7388 struct sel_data sd = { 0 };
7396 /* This makes us work on the components of the chosen mirror */
7397 start_index = lo->ldo_mirrors[pick].lme_start;
7398 max_comp = lo->ldo_mirrors[pick].lme_end + 1;
7399 if (lo->ldo_flr_state == LCM_FL_NONE)
7400 LASSERT(start_index == 0 && max_comp == lo->ldo_comp_cnt);
7402 CDEBUG(D_LAYOUT, "extent->e_start %llu, extent->e_end %llu\n",
7403 extent->e_start, extent->e_end);
7404 for (i = start_index; i < max_comp; i++) {
7405 lod_comp = &lo->ldo_comp_entries[i];
7407 /* We've passed all components of interest */
7408 if (lod_comp->llc_extent.e_start >= extent->e_end)
7411 if (lod_comp->llc_flags & LCME_FL_EXTENSION) {
7412 layout_changed = true;
7413 rc = lod_sel_handler(env, lo, extent, th, &max_comp,
7418 /* Nothing has changed behind the prev one */
7424 /* We may have added or removed components. If so, we must update the
7425 * start & ends of all the mirrors after the current one, and the end
7426 * of the current mirror. */
7427 change = max_comp - 1 - lo->ldo_mirrors[pick].lme_end;
7429 lo->ldo_mirrors[pick].lme_end += change;
7430 for (i = pick + 1; i < lo->ldo_mirror_count; i++) {
7431 lo->ldo_mirrors[i].lme_start += change;
7432 lo->ldo_mirrors[i].lme_end += change;
7438 /* The amount of components has changed, adjust the lti_comp_idx */
7439 rc2 = lod_layout_data_init(info, lo->ldo_comp_cnt);
7441 return rc < 0 ? rc : rc2 < 0 ? rc2 : layout_changed;
7444 /* If striping is already instantiated or INIT'ed DOM? */
7445 static bool lod_is_instantiation_needed(struct lod_layout_component *comp)
7447 return !(((lov_pattern(comp->llc_pattern) == LOV_PATTERN_MDT) &&
7448 lod_comp_inited(comp)) || comp->llc_stripe);
7452 * Declare layout update for a non-FLR layout.
7454 * \param[in] env execution environment for this thread
7455 * \param[in,out] lo object to update the layout of
7456 * \param[in] layout layout intent for requested operation, "update" is
7457 * a process of reacting to this
7458 * \param[in] buf buffer containing lov ea (see comment on usage inline)
7459 * \param[in] th transaction handle for this operation
7461 * \retval 0 on success
7462 * \retval negative errno on error
7464 static int lod_declare_update_plain(const struct lu_env *env,
7465 struct lod_object *lo, struct layout_intent *layout,
7466 const struct lu_buf *buf, struct thandle *th)
7468 struct lod_thread_info *info = lod_env_info(env);
7469 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7470 struct lod_layout_component *lod_comp;
7471 struct lov_comp_md_v1 *comp_v1 = NULL;
7472 bool layout_changed = false;
7473 bool replay = false;
7477 LASSERT(lo->ldo_flr_state == LCM_FL_NONE);
7480 * In case the client is passing lovea, which only happens during
7481 * the replay of layout intent write RPC for now, we may need to
7482 * parse the lovea and apply new layout configuration.
7484 if (buf && buf->lb_len) {
7485 struct lov_user_md_v1 *v1 = buf->lb_buf;
7487 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
7488 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
7489 LOV_MAGIC_COMP_V1)) {
7490 CERROR("%s: the replay buffer of layout extend "
7491 "(magic %#x) does not contain expected "
7492 "composite layout.\n",
7493 lod2obd(d)->obd_name, v1->lmm_magic);
7494 GOTO(out, rc = -EINVAL);
7497 rc = lod_use_defined_striping(env, lo, buf);
7500 lo->ldo_comp_cached = 1;
7502 rc = lod_get_lov_ea(env, lo);
7505 /* old on-disk EA is stored in info->lti_buf */
7506 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
7508 layout_changed = true;
7510 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
7514 /* non replay path */
7515 rc = lod_striping_load(env, lo);
7520 /* Make sure defined layout covers the requested write range. */
7521 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
7522 if (lo->ldo_comp_cnt > 1 &&
7523 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
7524 lod_comp->llc_extent.e_end < layout->li_extent.e_end) {
7525 CDEBUG_LIMIT(replay ? D_ERROR : D_LAYOUT,
7526 "%s: the defined layout [0, %#llx) does not "
7527 "covers the write range "DEXT"\n",
7528 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
7529 PEXT(&layout->li_extent));
7530 GOTO(out, rc = -EINVAL);
7533 CDEBUG(D_LAYOUT, "%s: "DFID": update components "DEXT"\n",
7534 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
7535 PEXT(&layout->li_extent));
7538 rc = lod_declare_update_extents(env, lo, &layout->li_extent,
7539 th, 0, layout->li_opc == LAYOUT_INTENT_WRITE);
7543 layout_changed = true;
7547 * Iterate ld->ldo_comp_entries, find the component whose extent under
7548 * the write range and not instantianted.
7550 for (i = 0; i < lo->ldo_comp_cnt; i++) {
7551 lod_comp = &lo->ldo_comp_entries[i];
7553 if (lod_comp->llc_extent.e_start >= layout->li_extent.e_end)
7557 /* If striping is instantiated or INIT'ed DOM skip */
7558 if (!lod_is_instantiation_needed(lod_comp))
7562 * In replay path, lod_comp is the EA passed by
7563 * client replay buffer, comp_v1 is the pre-recovery
7564 * on-disk EA, we'd sift out those components which
7565 * were init-ed in the on-disk EA.
7567 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
7572 * this component hasn't instantiated in normal path, or during
7573 * replay it needs replay the instantiation.
7576 /* A released component is being extended */
7577 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
7578 GOTO(out, rc = -EINVAL);
7580 LASSERT(info->lti_comp_idx != NULL);
7581 info->lti_comp_idx[info->lti_count++] = i;
7582 layout_changed = true;
7585 if (!layout_changed)
7588 lod_obj_inc_layout_gen(lo);
7589 rc = lod_declare_instantiate_components(env, lo, th, 0);
7593 lod_striping_free(env, lo);
7597 static inline int lod_comp_index(struct lod_object *lo,
7598 struct lod_layout_component *lod_comp)
7600 LASSERT(lod_comp >= lo->ldo_comp_entries &&
7601 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
7603 return lod_comp - lo->ldo_comp_entries;
7607 * Stale other mirrors by writing extent.
7609 static int lod_stale_components(const struct lu_env *env, struct lod_object *lo,
7610 int primary, struct lu_extent *extent,
7613 struct lod_layout_component *pri_comp, *lod_comp;
7614 struct lu_extent pri_extent;
7619 /* The writing extent decides which components in the primary
7620 * are affected... */
7621 CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
7624 lod_foreach_mirror_comp(pri_comp, lo, primary) {
7625 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
7628 CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
7629 lod_comp_index(lo, pri_comp),
7630 PEXT(&pri_comp->llc_extent));
7632 pri_extent.e_start = pri_comp->llc_extent.e_start;
7633 pri_extent.e_end = pri_comp->llc_extent.e_end;
7635 for (i = 0; i < lo->ldo_mirror_count; i++) {
7638 rc = lod_declare_update_extents(env, lo, &pri_extent,
7640 /* if update_extents changed the layout, it may have
7641 * reallocated the component array, so start over to
7642 * avoid using stale pointers */
7648 /* ... and then stale other components that are
7649 * overlapping with primary components */
7650 lod_foreach_mirror_comp(lod_comp, lo, i) {
7651 if (!lu_extent_is_overlapped(
7653 &lod_comp->llc_extent))
7656 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
7657 i, lod_comp_index(lo, lod_comp));
7659 lod_comp->llc_flags |= LCME_FL_STALE;
7660 lo->ldo_mirrors[i].lme_stale = 1;
7669 * check an OST's availability
7670 * \param[in] env execution environment
7671 * \param[in] lo lod object
7672 * \param[in] dt dt object
7673 * \param[in] index mirror index
7675 * \retval negative if failed
7676 * \retval 1 if \a dt is available
7677 * \retval 0 if \a dt is not available
7679 static inline int lod_check_ost_avail(const struct lu_env *env,
7680 struct lod_object *lo,
7681 struct dt_object *dt, int index)
7683 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7684 struct lod_tgt_desc *ost;
7686 int type = LU_SEQ_RANGE_OST;
7689 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &idx, &type);
7691 CERROR("%s: can't locate "DFID":rc = %d\n",
7692 lod2obd(lod)->obd_name, PFID(lu_object_fid(&dt->do_lu)),
7697 ost = OST_TGT(lod, idx);
7698 if (ost->ltd_statfs.os_state &
7699 (OS_STATFS_READONLY | OS_STATFS_ENOSPC | OS_STATFS_ENOINO |
7700 OS_STATFS_NOPRECREATE) ||
7701 ost->ltd_active == 0) {
7702 CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail, rc = %d\n",
7703 PFID(lod_object_fid(lo)), index, idx, rc);
7711 * Pick primary mirror for write
7712 * \param[in] env execution environment
7713 * \param[in] lo object
7714 * \param[in] extent write range
7716 static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo,
7717 struct lu_extent *extent)
7719 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7720 unsigned int seq = 0;
7721 struct lod_layout_component *lod_comp;
7723 int picked = -1, second_pick = -1, third_pick = -1;
7726 if (OBD_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
7727 get_random_bytes(&seq, sizeof(seq));
7728 seq %= lo->ldo_mirror_count;
7732 * Pick a mirror as the primary, and check the availability of OSTs.
7734 * This algo can be revised later after knowing the topology of
7737 lod_qos_statfs_update(env, lod, &lod->lod_ost_descs);
7739 rc = lod_fill_mirrors(lo);
7743 for (i = 0; i < lo->ldo_mirror_count; i++) {
7744 bool ost_avail = true;
7745 int index = (i + seq) % lo->ldo_mirror_count;
7747 if (lo->ldo_mirrors[index].lme_stale) {
7748 CDEBUG(D_LAYOUT, DFID": mirror %d stale\n",
7749 PFID(lod_object_fid(lo)), index);
7753 /* 2nd pick is for the primary mirror containing unavail OST */
7754 if (lo->ldo_mirrors[index].lme_prefer && second_pick < 0)
7755 second_pick = index;
7757 /* 3rd pick is for non-primary mirror containing unavail OST */
7758 if (second_pick < 0 && third_pick < 0)
7762 * we found a non-primary 1st pick, we'd like to find a
7763 * potential pirmary mirror.
7765 if (picked >= 0 && !lo->ldo_mirrors[index].lme_prefer)
7768 /* check the availability of OSTs */
7769 lod_foreach_mirror_comp(lod_comp, lo, index) {
7770 if (!lod_comp_inited(lod_comp) || !lod_comp->llc_stripe)
7773 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
7774 struct dt_object *dt = lod_comp->llc_stripe[j];
7776 rc = lod_check_ost_avail(env, lo, dt, index);
7783 } /* for all dt object in one component */
7786 } /* for all components in a mirror */
7789 * the OSTs where allocated objects locates in the components
7790 * of the mirror are available.
7795 /* this mirror has all OSTs available */
7799 * primary with all OSTs are available, this is the perfect
7802 if (lo->ldo_mirrors[index].lme_prefer)
7804 } /* for all mirrors */
7806 /* failed to pick a sound mirror, lower our expectation */
7808 picked = second_pick;
7810 picked = third_pick;
7817 static int lod_prepare_resync_mirror(const struct lu_env *env,
7818 struct lod_object *lo,
7821 struct lod_thread_info *info = lod_env_info(env);
7822 struct lod_layout_component *lod_comp;
7823 bool neg = !!(MIRROR_ID_NEG & mirror_id);
7826 mirror_id &= ~MIRROR_ID_NEG;
7828 for (i = 0; i < lo->ldo_mirror_count; i++) {
7829 if ((!neg && lo->ldo_mirrors[i].lme_id != mirror_id) ||
7830 (neg && lo->ldo_mirrors[i].lme_id == mirror_id))
7833 lod_foreach_mirror_comp(lod_comp, lo, i) {
7834 if (lod_comp_inited(lod_comp))
7837 info->lti_comp_idx[info->lti_count++] =
7838 lod_comp_index(lo, lod_comp);
7846 * figure out the components should be instantiated for resync.
7848 static int lod_prepare_resync(const struct lu_env *env, struct lod_object *lo,
7849 struct lu_extent *extent)
7851 struct lod_thread_info *info = lod_env_info(env);
7852 struct lod_layout_component *lod_comp;
7853 unsigned int need_sync = 0;
7857 DFID": instantiate all stale components in "DEXT"\n",
7858 PFID(lod_object_fid(lo)), PEXT(extent));
7861 * instantiate all components within this extent, even non-stale
7864 for (i = 0; i < lo->ldo_mirror_count; i++) {
7865 if (!lo->ldo_mirrors[i].lme_stale)
7868 lod_foreach_mirror_comp(lod_comp, lo, i) {
7869 if (!lu_extent_is_overlapped(extent,
7870 &lod_comp->llc_extent))
7875 if (lod_comp_inited(lod_comp))
7878 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
7879 i, lod_comp_index(lo, lod_comp));
7880 info->lti_comp_idx[info->lti_count++] =
7881 lod_comp_index(lo, lod_comp);
7885 return need_sync ? 0 : -EALREADY;
7888 static int lod_declare_update_rdonly(const struct lu_env *env,
7889 struct lod_object *lo, struct md_layout_change *mlc,
7892 struct lod_thread_info *info = lod_env_info(env);
7893 struct lu_attr *layout_attr = &info->lti_layout_attr;
7894 struct lod_layout_component *lod_comp;
7895 struct lu_extent extent = { 0 };
7899 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
7900 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
7901 mlc->mlc_opc == MD_LAYOUT_RESYNC);
7902 LASSERT(lo->ldo_mirror_count > 0);
7904 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7905 struct layout_intent *layout = mlc->mlc_intent;
7906 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
7909 extent = layout->li_extent;
7910 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
7911 PFID(lod_object_fid(lo)), PEXT(&extent));
7913 picked = lod_primary_pick(env, lo, &extent);
7917 CDEBUG(D_LAYOUT, DFID": picked mirror id %u as primary\n",
7918 PFID(lod_object_fid(lo)),
7919 lo->ldo_mirrors[picked].lme_id);
7921 /* Update extents of primary before staling */
7922 rc = lod_declare_update_extents(env, lo, &extent, th, picked,
7927 if (layout->li_opc == LAYOUT_INTENT_TRUNC) {
7929 * trunc transfers [0, size) in the intent extent, we'd
7930 * stale components overlapping [size, eof).
7932 extent.e_start = extent.e_end;
7933 extent.e_end = OBD_OBJECT_EOF;
7936 /* stale overlapping components from other mirrors */
7937 rc = lod_stale_components(env, lo, picked, &extent, th);
7941 /* restore truncate intent extent */
7942 if (layout->li_opc == LAYOUT_INTENT_TRUNC)
7943 extent.e_end = extent.e_start;
7945 /* instantiate components for the picked mirror, start from 0 */
7948 lod_foreach_mirror_comp(lod_comp, lo, picked) {
7949 if (!lu_extent_is_overlapped(&extent,
7950 &lod_comp->llc_extent))
7953 if (!lod_is_instantiation_needed(lod_comp))
7956 info->lti_comp_idx[info->lti_count++] =
7957 lod_comp_index(lo, lod_comp);
7960 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
7961 } else { /* MD_LAYOUT_RESYNC */
7965 * could contain multiple non-stale mirrors, so we need to
7966 * prep uninited all components assuming any non-stale mirror
7967 * could be picked as the primary mirror.
7969 if (mlc->mlc_mirror_id == 0) {
7971 for (i = 0; i < lo->ldo_mirror_count; i++) {
7972 if (lo->ldo_mirrors[i].lme_stale)
7975 lod_foreach_mirror_comp(lod_comp, lo, i) {
7976 if (!lod_comp_inited(lod_comp))
7980 lod_comp->llc_extent.e_end)
7982 lod_comp->llc_extent.e_end;
7985 rc = lod_prepare_resync(env, lo, &extent);
7989 /* mirror write, try to init its all components */
7990 rc = lod_prepare_resync_mirror(env, lo,
7991 mlc->mlc_mirror_id);
7996 /* change the file state to SYNC_PENDING */
7997 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
8000 /* Reset the layout version once it's becoming too large.
8001 * This way it can make sure that the layout version is
8002 * monotonously increased in this writing era. */
8003 lod_obj_inc_layout_gen(lo);
8004 if (lo->ldo_layout_gen > (LCME_ID_MAX >> 1)) {
8005 __u32 layout_version;
8007 get_random_bytes(&layout_version, sizeof(layout_version));
8008 lo->ldo_layout_gen = layout_version & 0xffff;
8011 rc = lod_declare_instantiate_components(env, lo, th, 0);
8015 layout_attr->la_valid = LA_LAYOUT_VERSION;
8016 layout_attr->la_layout_version = 0; /* set current version */
8017 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
8018 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
8019 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
8025 lod_striping_free(env, lo);
8029 static int lod_declare_update_write_pending(const struct lu_env *env,
8030 struct lod_object *lo, struct md_layout_change *mlc,
8033 struct lod_thread_info *info = lod_env_info(env);
8034 struct lu_attr *layout_attr = &info->lti_layout_attr;
8035 struct lod_layout_component *lod_comp;
8036 struct lu_extent extent = { 0 };
8042 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
8043 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
8044 mlc->mlc_opc == MD_LAYOUT_RESYNC);
8046 /* look for the first preferred mirror */
8047 for (i = 0; i < lo->ldo_mirror_count; i++) {
8048 if (lo->ldo_mirrors[i].lme_stale)
8050 if (lo->ldo_mirrors[i].lme_prefer == 0)
8057 /* no primary, use any in-sync */
8058 for (i = 0; i < lo->ldo_mirror_count; i++) {
8059 if (lo->ldo_mirrors[i].lme_stale)
8065 CERROR(DFID ": doesn't have a primary mirror\n",
8066 PFID(lod_object_fid(lo)));
8067 GOTO(out, rc = -ENODATA);
8071 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
8072 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
8074 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
8076 /* for LAYOUT_WRITE opc, it has to do the following operations:
8077 * 1. stale overlapping componets from stale mirrors;
8078 * 2. instantiate components of the primary mirror;
8079 * 3. transfter layout version to all objects of the primary;
8081 * for LAYOUT_RESYNC opc, it will do:
8082 * 1. instantiate components of all stale mirrors;
8083 * 2. transfer layout version to all objects to close write era. */
8085 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
8086 struct layout_intent *layout = mlc->mlc_intent;
8087 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
8089 LASSERT(mlc->mlc_intent != NULL);
8091 extent = mlc->mlc_intent->li_extent;
8093 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
8094 PFID(lod_object_fid(lo)), PEXT(&extent));
8096 /* 1. Update extents of primary before staling */
8097 rc = lod_declare_update_extents(env, lo, &extent, th, primary,
8102 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC) {
8104 * trunc transfers [0, size) in the intent extent, we'd
8105 * stale components overlapping [size, eof).
8107 extent.e_start = extent.e_end;
8108 extent.e_end = OBD_OBJECT_EOF;
8111 /* 2. stale overlapping components */
8112 rc = lod_stale_components(env, lo, primary, &extent, th);
8116 /* 3. find the components which need instantiating.
8117 * instantiate [0, mlc->mlc_intent->e_end) */
8119 /* restore truncate intent extent */
8120 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC)
8121 extent.e_end = extent.e_start;
8124 lod_foreach_mirror_comp(lod_comp, lo, primary) {
8125 if (!lu_extent_is_overlapped(&extent,
8126 &lod_comp->llc_extent))
8129 if (!lod_is_instantiation_needed(lod_comp))
8132 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
8133 primary, lod_comp_index(lo, lod_comp));
8134 info->lti_comp_idx[info->lti_count++] =
8135 lod_comp_index(lo, lod_comp);
8137 } else { /* MD_LAYOUT_RESYNC */
8138 if (mlc->mlc_mirror_id == 0) {
8140 lod_foreach_mirror_comp(lod_comp, lo, primary) {
8141 if (!lod_comp_inited(lod_comp))
8144 extent.e_end = lod_comp->llc_extent.e_end;
8147 rc = lod_prepare_resync(env, lo, &extent);
8151 /* mirror write, try to init its all components */
8152 rc = lod_prepare_resync_mirror(env, lo,
8153 mlc->mlc_mirror_id);
8158 /* change the file state to SYNC_PENDING */
8159 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
8162 rc = lod_declare_instantiate_components(env, lo, th, 0);
8166 /* 3. transfer layout version to OST objects.
8167 * transfer new layout version to OST objects so that stale writes
8168 * can be denied. It also ends an era of writing by setting
8169 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
8170 * send write RPC; only resync RPCs could do it. */
8171 layout_attr->la_valid = LA_LAYOUT_VERSION;
8172 layout_attr->la_layout_version = 0; /* set current version */
8173 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
8174 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
8175 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
8179 lod_obj_inc_layout_gen(lo);
8182 lod_striping_free(env, lo);
8186 static int lod_declare_update_sync_pending(const struct lu_env *env,
8187 struct lod_object *lo, struct md_layout_change *mlc,
8190 struct lod_thread_info *info = lod_env_info(env);
8191 struct lu_attr *layout_attr = &info->lti_layout_attr;
8192 unsigned sync_components = 0;
8193 unsigned resync_components = 0;
8198 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
8199 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
8200 mlc->mlc_opc == MD_LAYOUT_WRITE);
8202 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
8203 PFID(lod_object_fid(lo)), mlc->mlc_opc);
8205 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
8206 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
8207 PFID(lod_object_fid(lo)));
8209 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
8210 return lod_declare_update_write_pending(env, lo, mlc, th);
8213 /* MD_LAYOUT_RESYNC_DONE */
8215 for (i = 0; i < lo->ldo_comp_cnt; i++) {
8216 struct lod_layout_component *lod_comp;
8219 lod_comp = &lo->ldo_comp_entries[i];
8221 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
8226 for (j = 0; j < mlc->mlc_resync_count; j++) {
8227 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
8230 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
8231 lod_comp->llc_flags &= ~LCME_FL_STALE;
8232 resync_components++;
8238 for (i = 0; i < mlc->mlc_resync_count; i++) {
8239 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
8242 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
8243 "or already synced\n", PFID(lod_object_fid(lo)),
8244 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
8245 GOTO(out, rc = -EINVAL);
8248 if (!sync_components || (mlc->mlc_resync_count && !resync_components)) {
8249 CDEBUG(D_LAYOUT, DFID": no mirror in sync\n",
8250 PFID(lod_object_fid(lo)));
8252 /* tend to return an error code here to prevent
8253 * the MDT from setting SoM attribute */
8254 GOTO(out, rc = -EINVAL);
8257 CDEBUG(D_LAYOUT, DFID": synced %u resynced %u/%zu components\n",
8258 PFID(lod_object_fid(lo)),
8259 sync_components, resync_components, mlc->mlc_resync_count);
8261 lo->ldo_flr_state = LCM_FL_RDONLY;
8262 lod_obj_inc_layout_gen(lo);
8264 layout_attr->la_valid = LA_LAYOUT_VERSION;
8265 layout_attr->la_layout_version = 0; /* set current version */
8266 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
8270 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
8271 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
8272 &info->lti_buf, XATTR_NAME_LOV, 0, th);
8277 lod_striping_free(env, lo);
8281 typedef int (*mlc_handler)(const struct lu_env *env, struct dt_object *dt,
8282 const struct md_layout_change *mlc,
8283 struct thandle *th);
8286 * Attach stripes after target's for migrating directory. NB, we
8287 * only need to declare this, the actual work is done inside
8288 * lod_xattr_set_lmv().
8290 * \param[in] env execution environment
8291 * \param[in] dt target object
8292 * \param[in] mlc layout change data
8293 * \param[in] th transaction handle
8295 * \retval 0 on success
8296 * \retval negative if failed
8298 static int lod_dir_declare_layout_attach(const struct lu_env *env,
8299 struct dt_object *dt,
8300 const struct md_layout_change *mlc,
8303 struct lod_thread_info *info = lod_env_info(env);
8304 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
8305 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
8306 struct lod_object *lo = lod_dt_obj(dt);
8307 struct dt_object *next = dt_object_child(dt);
8308 struct dt_object_format *dof = &info->lti_format;
8309 struct lmv_mds_md_v1 *lmv = mlc->mlc_buf.lb_buf;
8310 struct dt_object **stripes;
8311 __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
8312 struct lu_fid *fid = &info->lti_fid;
8313 struct lod_tgt_desc *tgt;
8314 struct dt_object *dto;
8315 struct dt_device *tgt_dt;
8316 int type = LU_SEQ_RANGE_ANY;
8317 struct dt_insert_rec *rec = &info->lti_dt_rec;
8318 char *stripe_name = info->lti_key;
8319 struct lu_name *sname;
8320 struct linkea_data ldata = { NULL };
8321 struct lu_buf linkea_buf;
8328 if (!lmv_is_sane(lmv))
8331 if (!dt_try_as_dir(env, dt, false))
8334 dof->dof_type = DFT_DIR;
8336 OBD_ALLOC_PTR_ARRAY(stripes, (lo->ldo_dir_stripe_count + stripe_count));
8340 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
8341 stripes[i] = lo->ldo_stripe[i];
8343 rec->rec_type = S_IFDIR;
8345 for (i = 0; i < stripe_count; i++) {
8347 &lmv->lmv_stripe_fids[i]);
8348 if (!fid_is_sane(fid))
8351 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
8355 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
8356 tgt_dt = lod->lod_child;
8358 tgt = LTD_TGT(ltd, idx);
8360 GOTO(out, rc = -ESTALE);
8361 tgt_dt = tgt->ltd_tgt;
8364 dto = dt_locate_at(env, tgt_dt, fid,
8365 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
8368 GOTO(out, rc = PTR_ERR(dto));
8370 stripes[i + lo->ldo_dir_stripe_count] = dto;
8372 if (!dt_try_as_dir(env, dto, true))
8373 GOTO(out, rc = -ENOTDIR);
8375 rc = lod_sub_declare_ref_add(env, dto, th);
8379 rec->rec_fid = lu_object_fid(&dto->do_lu);
8380 rc = lod_sub_declare_insert(env, dto,
8381 (const struct dt_rec *)rec,
8382 (const struct dt_key *)dot, th);
8386 rc = lod_sub_declare_insert(env, dto,
8387 (const struct dt_rec *)rec,
8388 (const struct dt_key *)dotdot, th);
8392 rc = lod_sub_declare_xattr_set(env, dto, &mlc->mlc_buf,
8393 XATTR_NAME_LMV, 0, th);
8397 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
8398 PFID(lu_object_fid(&dto->do_lu)),
8399 i + lo->ldo_dir_stripe_count);
8401 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
8402 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
8403 sname, lu_object_fid(&dt->do_lu));
8407 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
8408 linkea_buf.lb_len = ldata.ld_leh->leh_len;
8409 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
8410 XATTR_NAME_LINK, 0, th);
8414 rc = lod_sub_declare_insert(env, next,
8415 (const struct dt_rec *)rec,
8416 (const struct dt_key *)stripe_name,
8421 rc = lod_sub_declare_ref_add(env, next, th);
8427 OBD_FREE_PTR_ARRAY(lo->ldo_stripe,
8428 lo->ldo_dir_stripes_allocated);
8429 lo->ldo_stripe = stripes;
8430 lo->ldo_is_foreign = 0;
8431 lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count;
8432 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type);
8433 lo->ldo_dir_stripe_count += stripe_count;
8434 lo->ldo_dir_stripes_allocated += stripe_count;
8436 /* plain directory split creates target as a plain directory, while
8437 * after source attached as the first stripe, it becomes a striped
8438 * directory, set correct do_index_ops, otherwise it can't be unlinked.
8440 dt->do_index_ops = &lod_striped_index_ops;
8444 i = lo->ldo_dir_stripe_count;
8445 while (i < lo->ldo_dir_stripe_count + stripe_count && stripes[i])
8446 dt_object_put(env, stripes[i++]);
8448 OBD_FREE_PTR_ARRAY(stripes, stripe_count + lo->ldo_dir_stripe_count);
8452 static int lod_dir_declare_layout_detach(const struct lu_env *env,
8453 struct dt_object *dt,
8454 const struct md_layout_change *unused,
8457 struct lod_thread_info *info = lod_env_info(env);
8458 struct lod_object *lo = lod_dt_obj(dt);
8459 struct dt_object *next = dt_object_child(dt);
8460 char *stripe_name = info->lti_key;
8461 struct dt_object *dto;
8465 if (!dt_try_as_dir(env, dt, true))
8468 if (!lo->ldo_dir_stripe_count)
8469 return lod_sub_declare_delete(env, next,
8470 (const struct dt_key *)dotdot, th);
8472 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8473 dto = lo->ldo_stripe[i];
8477 if (!dt_try_as_dir(env, dto, true))
8480 rc = lod_sub_declare_delete(env, dto,
8481 (const struct dt_key *)dotdot, th);
8485 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8486 PFID(lu_object_fid(&dto->do_lu)), i);
8488 rc = lod_sub_declare_delete(env, next,
8489 (const struct dt_key *)stripe_name, th);
8493 rc = lod_sub_declare_ref_del(env, next, th);
8501 static int dt_dir_is_empty(const struct lu_env *env,
8502 struct dt_object *obj)
8505 const struct dt_it_ops *iops;
8510 if (!dt_try_as_dir(env, obj, true))
8513 iops = &obj->do_index_ops->dio_it;
8514 it = iops->init(env, obj, LUDA_64BITHASH);
8516 RETURN(PTR_ERR(it));
8518 rc = iops->get(env, it, (const struct dt_key *)"");
8522 for (rc = 0, i = 0; rc == 0 && i < 3; ++i)
8523 rc = iops->next(env, it);
8529 /* Huh? Index contains no zero key? */
8534 iops->fini(env, it);
8539 static int lod_dir_declare_layout_shrink(const struct lu_env *env,
8540 struct dt_object *dt,
8541 const struct md_layout_change *mlc,
8544 struct lod_thread_info *info = lod_env_info(env);
8545 struct lod_object *lo = lod_dt_obj(dt);
8546 struct dt_object *next = dt_object_child(dt);
8547 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
8548 char *stripe_name = info->lti_key;
8549 struct lu_buf *lmv_buf = &info->lti_buf;
8550 __u32 final_stripe_count;
8551 struct dt_object *dto;
8557 if (!dt_try_as_dir(env, dt, true))
8560 /* shouldn't be called on plain directory */
8561 LASSERT(lo->ldo_dir_stripe_count);
8563 lmv_buf->lb_buf = &info->lti_lmv.lmv_md_v1;
8564 lmv_buf->lb_len = sizeof(info->lti_lmv.lmv_md_v1);
8566 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
8567 LASSERT(final_stripe_count &&
8568 final_stripe_count < lo->ldo_dir_stripe_count);
8570 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8571 dto = lo->ldo_stripe[i];
8575 if (i < final_stripe_count) {
8576 rc = lod_sub_declare_xattr_set(env, dto, lmv_buf,
8578 LU_XATTR_REPLACE, th);
8585 rc = dt_dir_is_empty(env, dto);
8589 rc = lod_sub_declare_ref_del(env, dto, th);
8593 rc = lod_sub_declare_destroy(env, dto, th);
8597 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8598 PFID(lu_object_fid(&dto->do_lu)), i);
8600 rc = lod_sub_declare_delete(env, next,
8601 (const struct dt_key *)stripe_name, th);
8605 rc = lod_sub_declare_ref_del(env, next, th);
8610 rc = lod_sub_declare_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
8611 LU_XATTR_REPLACE, th);
8616 * Allocate stripes for split directory.
8618 * \param[in] env execution environment
8619 * \param[in] dt target object
8620 * \param[in] mlc layout change data
8621 * \param[in] th transaction handle
8623 * \retval 0 on success
8624 * \retval negative if failed
8626 static int lod_dir_declare_layout_split(const struct lu_env *env,
8627 struct dt_object *dt,
8628 const struct md_layout_change *mlc,
8631 struct lod_thread_info *info = lod_env_info(env);
8632 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
8633 struct lod_object *lo = lod_dt_obj(dt);
8634 struct dt_object_format *dof = &info->lti_format;
8635 struct lmv_user_md_v1 *lum = mlc->mlc_spec->u.sp_ea.eadata;
8636 struct dt_object **stripes;
8644 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
8645 LASSERT(le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT);
8647 saved_count = lo->ldo_dir_stripes_allocated;
8648 stripe_count = le32_to_cpu(lum->lum_stripe_count);
8649 if (stripe_count <= saved_count)
8652 dof->dof_type = DFT_DIR;
8654 OBD_ALLOC(stripes, sizeof(*stripes) * stripe_count);
8658 for (i = 0; i < lo->ldo_dir_stripes_allocated; i++)
8659 stripes[i] = lo->ldo_stripe[i];
8661 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
8662 rc = lod_mdt_alloc_qos(env, lo, stripes, saved_count, stripe_count);
8664 rc = lod_mdt_alloc_rr(env, lo, stripes, saved_count,
8667 OBD_FREE(stripes, sizeof(*stripes) * stripe_count);
8671 LASSERT(rc > saved_count);
8672 OBD_FREE(lo->ldo_stripe,
8673 sizeof(*stripes) * lo->ldo_dir_stripes_allocated);
8674 lo->ldo_stripe = stripes;
8675 lo->ldo_is_foreign = 0;
8676 lo->ldo_dir_striped = 1;
8677 lo->ldo_dir_stripe_count = rc;
8678 lo->ldo_dir_stripes_allocated = stripe_count;
8679 lo->ldo_dir_split_hash = lo->ldo_dir_hash_type;
8680 lo->ldo_dir_hash_type = le32_to_cpu(lum->lum_hash_type);
8681 if (!lmv_is_known_hash_type(lo->ldo_dir_hash_type))
8682 lo->ldo_dir_hash_type =
8683 lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
8684 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_SPLIT | LMV_HASH_FLAG_MIGRATION;
8685 lo->ldo_dir_split_offset = saved_count;
8686 lo->ldo_dir_layout_version++;
8687 lo->ldo_dir_stripe_loaded = 1;
8689 rc = lod_dir_declare_create_stripes(env, dt, mlc->mlc_attr, dof, th);
8691 lod_striping_free(env, lo);
8697 * detach all stripes from dir master object, NB, stripes are not destroyed, but
8698 * deleted from it's parent namespace, this function is called in two places:
8699 * 1. mdd_migrate_mdt() detach stripes from source, and attach them to
8701 * 2. mdd_dir_layout_update() detach stripe before turning 1-stripe directory to
8702 * a plain directory.
8704 * \param[in] env execution environment
8705 * \param[in] dt target object
8706 * \param[in] mlc layout change data
8707 * \param[in] th transaction handle
8709 * \retval 0 on success
8710 * \retval negative if failed
8712 static int lod_dir_layout_detach(const struct lu_env *env,
8713 struct dt_object *dt,
8714 const struct md_layout_change *mlc,
8717 struct lod_thread_info *info = lod_env_info(env);
8718 struct lod_object *lo = lod_dt_obj(dt);
8719 struct dt_object *next = dt_object_child(dt);
8720 char *stripe_name = info->lti_key;
8721 struct dt_object *dto;
8727 if (!lo->ldo_dir_stripe_count) {
8728 /* plain directory delete .. */
8729 rc = lod_sub_delete(env, next,
8730 (const struct dt_key *)dotdot, th);
8734 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8735 dto = lo->ldo_stripe[i];
8739 rc = lod_sub_delete(env, dto,
8740 (const struct dt_key *)dotdot, th);
8744 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8745 PFID(lu_object_fid(&dto->do_lu)), i);
8747 rc = lod_sub_delete(env, next,
8748 (const struct dt_key *)stripe_name, th);
8752 rc = lod_sub_ref_del(env, next, th);
8757 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8758 dto = lo->ldo_stripe[i];
8760 dt_object_put(env, dto);
8762 OBD_FREE_PTR_ARRAY(lo->ldo_stripe, lo->ldo_dir_stripes_allocated);
8763 lo->ldo_stripe = NULL;
8764 lo->ldo_dir_stripes_allocated = 0;
8765 lo->ldo_dir_stripe_count = 0;
8766 dt->do_index_ops = &lod_index_ops;
8771 static int lod_dir_layout_shrink(const struct lu_env *env,
8772 struct dt_object *dt,
8773 const struct md_layout_change *mlc,
8776 struct lod_thread_info *info = lod_env_info(env);
8777 struct lod_object *lo = lod_dt_obj(dt);
8778 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
8779 struct dt_object *next = dt_object_child(dt);
8780 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
8781 __u32 final_stripe_count;
8782 char *stripe_name = info->lti_key;
8783 struct dt_object *dto;
8784 struct lu_buf *lmv_buf = &info->lti_buf;
8785 struct lmv_mds_md_v1 *lmv = &info->lti_lmv.lmv_md_v1;
8787 int type = LU_SEQ_RANGE_ANY;
8793 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
8795 lmv_buf->lb_buf = lmv;
8796 lmv_buf->lb_len = sizeof(*lmv);
8797 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
8798 lmv->lmv_stripe_count = cpu_to_le32(final_stripe_count);
8799 lmv->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type) &
8800 cpu_to_le32(LMV_HASH_TYPE_MASK |
8801 LMV_HASH_FLAG_FIXED);
8802 lmv->lmv_layout_version =
8803 cpu_to_le32(lo->ldo_dir_layout_version + 1);
8804 lmv->lmv_migrate_offset = 0;
8805 lmv->lmv_migrate_hash = 0;
8807 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8808 dto = lo->ldo_stripe[i];
8812 if (i < final_stripe_count) {
8813 rc = lod_fld_lookup(env, lod,
8814 lu_object_fid(&dto->do_lu),
8819 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
8820 rc = lod_sub_xattr_set(env, dto, lmv_buf,
8822 LU_XATTR_REPLACE, th);
8829 dt_write_lock(env, dto, DT_TGT_CHILD);
8830 rc = lod_sub_ref_del(env, dto, th);
8831 dt_write_unlock(env, dto);
8835 rc = lod_sub_destroy(env, dto, th);
8839 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8840 PFID(lu_object_fid(&dto->do_lu)), i);
8842 rc = lod_sub_delete(env, next,
8843 (const struct dt_key *)stripe_name, th);
8847 rc = lod_sub_ref_del(env, next, th);
8852 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &mdtidx,
8857 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
8858 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
8859 rc = lod_sub_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
8860 LU_XATTR_REPLACE, th);
8864 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
8865 dto = lo->ldo_stripe[i];
8867 dt_object_put(env, dto);
8869 lo->ldo_dir_stripe_count = final_stripe_count;
8874 static mlc_handler dir_mlc_declare_ops[MD_LAYOUT_MAX] = {
8875 [MD_LAYOUT_ATTACH] = lod_dir_declare_layout_attach,
8876 [MD_LAYOUT_DETACH] = lod_dir_declare_layout_detach,
8877 [MD_LAYOUT_SHRINK] = lod_dir_declare_layout_shrink,
8878 [MD_LAYOUT_SPLIT] = lod_dir_declare_layout_split,
8881 static mlc_handler dir_mlc_ops[MD_LAYOUT_MAX] = {
8882 [MD_LAYOUT_DETACH] = lod_dir_layout_detach,
8883 [MD_LAYOUT_SHRINK] = lod_dir_layout_shrink,
8886 static int lod_declare_layout_change(const struct lu_env *env,
8887 struct dt_object *dt, struct md_layout_change *mlc,
8890 struct lod_thread_info *info = lod_env_info(env);
8891 struct lod_object *lo = lod_dt_obj(dt);
8896 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
8897 LASSERT(dir_mlc_declare_ops[mlc->mlc_opc]);
8898 rc = dir_mlc_declare_ops[mlc->mlc_opc](env, dt, mlc, th);
8902 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
8903 dt_object_remote(dt_object_child(dt)))
8906 rc = lod_striping_load(env, lo);
8910 LASSERT(lo->ldo_comp_cnt > 0);
8912 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
8916 switch (lo->ldo_flr_state) {
8918 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
8922 rc = lod_declare_update_rdonly(env, lo, mlc, th);
8924 case LCM_FL_WRITE_PENDING:
8925 rc = lod_declare_update_write_pending(env, lo, mlc, th);
8927 case LCM_FL_SYNC_PENDING:
8928 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
8939 * Instantiate layout component objects which covers the intent write offset.
8941 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
8942 struct md_layout_change *mlc, struct thandle *th)
8944 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
8945 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
8946 struct lod_object *lo = lod_dt_obj(dt);
8951 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
8952 LASSERT(dir_mlc_ops[mlc->mlc_opc]);
8953 rc = dir_mlc_ops[mlc->mlc_opc](env, dt, mlc, th);
8957 rc = lod_striped_create(env, dt, attr, NULL, th);
8958 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
8959 layout_attr->la_layout_version |= lo->ldo_layout_gen;
8960 rc = lod_attr_set(env, dt, layout_attr, th);
8966 const struct dt_object_operations lod_obj_ops = {
8967 .do_read_lock = lod_read_lock,
8968 .do_write_lock = lod_write_lock,
8969 .do_read_unlock = lod_read_unlock,
8970 .do_write_unlock = lod_write_unlock,
8971 .do_write_locked = lod_write_locked,
8972 .do_attr_get = lod_attr_get,
8973 .do_declare_attr_set = lod_declare_attr_set,
8974 .do_attr_set = lod_attr_set,
8975 .do_xattr_get = lod_xattr_get,
8976 .do_declare_xattr_set = lod_declare_xattr_set,
8977 .do_xattr_set = lod_xattr_set,
8978 .do_declare_xattr_del = lod_declare_xattr_del,
8979 .do_xattr_del = lod_xattr_del,
8980 .do_xattr_list = lod_xattr_list,
8981 .do_ah_init = lod_ah_init,
8982 .do_declare_create = lod_declare_create,
8983 .do_create = lod_create,
8984 .do_declare_destroy = lod_declare_destroy,
8985 .do_destroy = lod_destroy,
8986 .do_index_try = lod_index_try,
8987 .do_declare_ref_add = lod_declare_ref_add,
8988 .do_ref_add = lod_ref_add,
8989 .do_declare_ref_del = lod_declare_ref_del,
8990 .do_ref_del = lod_ref_del,
8991 .do_object_sync = lod_object_sync,
8992 .do_object_lock = lod_object_lock,
8993 .do_object_unlock = lod_object_unlock,
8994 .do_invalidate = lod_invalidate,
8995 .do_declare_layout_change = lod_declare_layout_change,
8996 .do_layout_change = lod_layout_change,
9000 * Implementation of dt_body_operations::dbo_read.
9002 * \see dt_body_operations::dbo_read() in the API description for details.
9004 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
9005 struct lu_buf *buf, loff_t *pos)
9007 struct dt_object *next = dt_object_child(dt);
9009 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
9010 S_ISLNK(dt->do_lu.lo_header->loh_attr));
9011 return next->do_body_ops->dbo_read(env, next, buf, pos);
9015 * Implementation of dt_body_operations::dbo_declare_write.
9017 * \see dt_body_operations::dbo_declare_write() in the API description
9020 static ssize_t lod_declare_write(const struct lu_env *env,
9021 struct dt_object *dt,
9022 const struct lu_buf *buf, loff_t pos,
9025 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
9029 * Implementation of dt_body_operations::dbo_write.
9031 * \see dt_body_operations::dbo_write() in the API description for details.
9033 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
9034 const struct lu_buf *buf, loff_t *pos,
9037 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
9038 S_ISLNK(dt->do_lu.lo_header->loh_attr));
9039 return lod_sub_write(env, dt_object_child(dt), buf, pos, th);
9042 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
9043 __u64 start, __u64 end, struct thandle *th)
9045 if (dt_object_remote(dt))
9048 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
9051 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
9052 __u64 start, __u64 end, struct thandle *th)
9054 if (dt_object_remote(dt))
9057 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
9058 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
9062 * different type of files use the same body_ops because object may be created
9063 * in OUT, where there is no chance to set correct body_ops for each type, so
9064 * body_ops themselves will check file type inside, see lod_read/write/punch for
9067 static const struct dt_body_operations lod_body_ops = {
9068 .dbo_read = lod_read,
9069 .dbo_declare_write = lod_declare_write,
9070 .dbo_write = lod_write,
9071 .dbo_declare_punch = lod_declare_punch,
9072 .dbo_punch = lod_punch,
9076 * Implementation of lu_object_operations::loo_object_init.
9078 * The function determines the type and the index of the target device using
9079 * sequence of the object's FID. Then passes control down to the
9080 * corresponding device:
9081 * OSD for the local objects, OSP for remote
9083 * \see lu_object_operations::loo_object_init() in the API description
9086 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
9087 const struct lu_object_conf *conf)
9089 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
9090 struct lu_device *cdev = NULL;
9091 struct lu_object *cobj;
9092 struct lod_tgt_descs *ltd = NULL;
9093 struct lod_tgt_desc *tgt;
9095 int type = LU_SEQ_RANGE_ANY;
9099 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
9103 if (type == LU_SEQ_RANGE_MDT &&
9104 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
9105 cdev = &lod->lod_child->dd_lu_dev;
9106 } else if (type == LU_SEQ_RANGE_MDT) {
9107 ltd = &lod->lod_mdt_descs;
9109 } else if (type == LU_SEQ_RANGE_OST) {
9110 ltd = &lod->lod_ost_descs;
9117 if (ltd->ltd_tgts_size > idx &&
9118 test_bit(idx, ltd->ltd_tgt_bitmap)) {
9119 tgt = LTD_TGT(ltd, idx);
9121 LASSERT(tgt != NULL);
9122 LASSERT(tgt->ltd_tgt != NULL);
9124 cdev = &(tgt->ltd_tgt->dd_lu_dev);
9126 lod_putref(lod, ltd);
9129 if (unlikely(cdev == NULL))
9132 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
9133 if (unlikely(cobj == NULL))
9136 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
9138 lu_object_add(lo, cobj);
9145 * Release resources associated with striping.
9147 * If the object is striped (regular or directory), then release
9148 * the stripe objects references and free the ldo_stripe array.
9150 * \param[in] env execution environment
9151 * \param[in] lo object
9153 void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo)
9155 struct lod_layout_component *lod_comp;
9156 __u32 obj_attr = lo->ldo_obj.do_lu.lo_header->loh_attr;
9159 if (unlikely(lo->ldo_is_foreign)) {
9160 if (S_ISREG(obj_attr)) {
9161 lod_free_foreign_lov(lo);
9162 lo->ldo_comp_cached = 0;
9163 } else if (S_ISDIR(obj_attr)) {
9164 lod_free_foreign_lmv(lo);
9165 lo->ldo_dir_stripe_loaded = 0;
9167 } else if (lo->ldo_stripe != NULL) {
9168 LASSERT(lo->ldo_comp_entries == NULL);
9169 LASSERT(lo->ldo_dir_stripes_allocated > 0);
9171 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
9172 if (lo->ldo_stripe[i])
9173 dt_object_put(env, lo->ldo_stripe[i]);
9176 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
9177 OBD_FREE(lo->ldo_stripe, j);
9178 lo->ldo_stripe = NULL;
9179 lo->ldo_dir_stripes_allocated = 0;
9180 lo->ldo_dir_stripe_loaded = 0;
9181 lo->ldo_dir_stripe_count = 0;
9182 lo->ldo_obj.do_index_ops = NULL;
9183 } else if (lo->ldo_comp_entries != NULL) {
9184 for (i = 0; i < lo->ldo_comp_cnt; i++) {
9185 /* free lod_layout_component::llc_stripe array */
9186 lod_comp = &lo->ldo_comp_entries[i];
9188 if (lod_comp->llc_stripe == NULL)
9190 LASSERT(lod_comp->llc_stripes_allocated != 0);
9191 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
9192 if (lod_comp->llc_stripe[j] != NULL)
9194 &lod_comp->llc_stripe[j]->do_lu);
9196 OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe,
9197 lod_comp->llc_stripes_allocated);
9198 lod_comp->llc_stripe = NULL;
9199 OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices,
9200 lod_comp->llc_stripes_allocated);
9201 lod_comp->llc_ost_indices = NULL;
9202 lod_comp->llc_stripes_allocated = 0;
9204 lod_free_comp_entries(lo);
9205 lo->ldo_comp_cached = 0;
9209 void lod_striping_free(const struct lu_env *env, struct lod_object *lo)
9211 mutex_lock(&lo->ldo_layout_mutex);
9212 lod_striping_free_nolock(env, lo);
9213 mutex_unlock(&lo->ldo_layout_mutex);
9217 * Implementation of lu_object_operations::loo_object_free.
9219 * \see lu_object_operations::loo_object_free() in the API description
9222 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
9224 struct lod_object *lo = lu2lod_obj(o);
9226 /* release all underlying object pinned */
9227 lod_striping_free(env, lo);
9229 /* lo doesn't contain a lu_object_header, so we don't need call_rcu */
9230 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
9234 * Implementation of lu_object_operations::loo_object_release.
9236 * \see lu_object_operations::loo_object_release() in the API description
9239 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
9241 /* XXX: shouldn't we release everything here in case if object
9242 * creation failed before? */
9246 * Implementation of lu_object_operations::loo_object_print.
9248 * \see lu_object_operations::loo_object_print() in the API description
9251 static int lod_object_print(const struct lu_env *env, void *cookie,
9252 lu_printer_t p, const struct lu_object *l)
9254 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
9256 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
9259 const struct lu_object_operations lod_lu_obj_ops = {
9260 .loo_object_init = lod_object_init,
9261 .loo_object_free = lod_object_free,
9262 .loo_object_release = lod_object_release,
9263 .loo_object_print = lod_object_print,