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 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 struct dt_index_operations lod_striped_index_ops = {
791 .dio_lookup = lod_striped_lookup,
792 .dio_declare_insert = lod_declare_insert,
793 .dio_insert = lod_insert,
794 .dio_declare_delete = lod_declare_delete,
795 .dio_delete = lod_delete,
797 .init = lod_striped_it_init,
798 .fini = lod_striped_it_fini,
799 .get = lod_striped_it_get,
800 .put = lod_striped_it_put,
801 .next = lod_striped_it_next,
802 .key = lod_striped_it_key,
803 .key_size = lod_striped_it_key_size,
804 .rec = lod_striped_it_rec,
805 .rec_size = lod_striped_it_rec_size,
806 .store = lod_striped_it_store,
807 .load = lod_striped_it_load,
812 * Append the FID for each shard of the striped directory after the
813 * given LMV EA header.
815 * To simplify striped directory and the consistency verification,
816 * we only store the LMV EA header on disk, for both master object
817 * and slave objects. When someone wants to know the whole LMV EA,
818 * such as client readdir(), we can build the entrie LMV EA on the
819 * MDT side (in RAM) via iterating the sub-directory entries that
820 * are contained in the master object of the stripe directory.
822 * For the master object of the striped directroy, the valid name
823 * for each shard is composed of the ${shard_FID}:${shard_idx}.
825 * There may be holes in the LMV EA if some shards' name entries
826 * are corrupted or lost.
828 * \param[in] env pointer to the thread context
829 * \param[in] lo pointer to the master object of the striped directory
830 * \param[in] buf pointer to the lu_buf which will hold the LMV EA
831 * \param[in] resize whether re-allocate the buffer if it is not big enough
833 * \retval positive size of the LMV EA
834 * \retval 0 for nothing to be loaded
835 * \retval negative error number on failure
837 int lod_load_lmv_shards(const struct lu_env *env, struct lod_object *lo,
838 struct lu_buf *buf, bool resize)
840 struct lu_dirent *ent =
841 (struct lu_dirent *)lod_env_info(env)->lti_key;
842 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
843 struct dt_object *obj = dt_object_child(&lo->ldo_obj);
844 struct lmv_mds_md_v1 *lmv1 = buf->lb_buf;
846 const struct dt_it_ops *iops;
848 __u32 magic = le32_to_cpu(lmv1->lmv_magic);
853 if (magic != LMV_MAGIC_V1)
856 stripes = le32_to_cpu(lmv1->lmv_stripe_count);
860 rc = lmv_mds_md_size(stripes, magic);
864 if (buf->lb_len < lmv1_size) {
873 lu_buf_alloc(buf, lmv1_size);
878 memcpy(buf->lb_buf, tbuf.lb_buf, tbuf.lb_len);
881 if (unlikely(!dt_try_as_dir(env, obj)))
884 memset(&lmv1->lmv_stripe_fids[0], 0, stripes * sizeof(struct lu_fid));
885 iops = &obj->do_index_ops->dio_it;
886 it = iops->init(env, obj, LUDA_64BITHASH);
890 rc = iops->load(env, it, 0);
892 rc = iops->next(env, it);
897 char name[FID_LEN + 2] = "";
902 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
908 fid_le_to_cpu(&fid, &ent->lde_fid);
909 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
910 if (ent->lde_name[0] == '.') {
911 if (ent->lde_namelen == 1)
914 if (ent->lde_namelen == 2 && ent->lde_name[1] == '.')
918 len = scnprintf(name, sizeof(name),
919 DFID":", PFID(&ent->lde_fid));
920 /* The ent->lde_name is composed of ${FID}:${index} */
921 if (ent->lde_namelen < len + 1 ||
922 memcmp(ent->lde_name, name, len) != 0) {
923 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
924 "%s: invalid shard name %.*s with the FID "DFID
925 " for the striped directory "DFID", %s\n",
926 lod2obd(lod)->obd_name, ent->lde_namelen,
927 ent->lde_name, PFID(&fid),
928 PFID(lu_object_fid(&obj->do_lu)),
929 lod->lod_lmv_failout ? "failout" : "skip");
931 if (lod->lod_lmv_failout)
939 if (ent->lde_name[len] < '0' ||
940 ent->lde_name[len] > '9') {
941 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
942 "%s: invalid shard name %.*s with the "
943 "FID "DFID" for the striped directory "
945 lod2obd(lod)->obd_name, ent->lde_namelen,
946 ent->lde_name, PFID(&fid),
947 PFID(lu_object_fid(&obj->do_lu)),
948 lod->lod_lmv_failout ?
951 if (lod->lod_lmv_failout)
957 index = index * 10 + ent->lde_name[len++] - '0';
958 } while (len < ent->lde_namelen);
960 if (len == ent->lde_namelen) {
961 /* Out of LMV EA range. */
962 if (index >= stripes) {
963 CERROR("%s: the shard %.*s for the striped "
964 "directory "DFID" is out of the known "
965 "LMV EA range [0 - %u], failout\n",
966 lod2obd(lod)->obd_name, ent->lde_namelen,
968 PFID(lu_object_fid(&obj->do_lu)),
974 /* The slot has been occupied. */
975 if (!fid_is_zero(&lmv1->lmv_stripe_fids[index])) {
979 &lmv1->lmv_stripe_fids[index]);
980 CERROR("%s: both the shard "DFID" and "DFID
981 " for the striped directory "DFID
982 " claim the same LMV EA slot at the "
983 "index %d, failout\n",
984 lod2obd(lod)->obd_name,
985 PFID(&fid0), PFID(&fid),
986 PFID(lu_object_fid(&obj->do_lu)), index);
991 /* stored as LE mode */
992 lmv1->lmv_stripe_fids[index] = ent->lde_fid;
995 rc = iops->next(env, it);
1000 iops->fini(env, it);
1002 RETURN(rc > 0 ? lmv_mds_md_size(stripes, magic) : rc);
1006 * Implementation of dt_object_operations::do_index_try.
1008 * \see dt_object_operations::do_index_try() in the API description for details.
1010 static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
1011 const struct dt_index_features *feat)
1013 struct lod_object *lo = lod_dt_obj(dt);
1014 struct dt_object *next = dt_object_child(dt);
1018 LASSERT(next->do_ops);
1019 LASSERT(next->do_ops->do_index_try);
1021 rc = lod_striping_load(env, lo);
1025 rc = next->do_ops->do_index_try(env, next, feat);
1029 if (lo->ldo_dir_stripe_count > 0) {
1032 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1033 if (!lo->ldo_stripe[i])
1035 if (!dt_object_exists(lo->ldo_stripe[i]))
1037 rc = lo->ldo_stripe[i]->do_ops->do_index_try(env,
1038 lo->ldo_stripe[i], feat);
1042 dt->do_index_ops = &lod_striped_index_ops;
1044 dt->do_index_ops = &lod_index_ops;
1051 * Implementation of dt_object_operations::do_read_lock.
1053 * \see dt_object_operations::do_read_lock() in the API description for details.
1055 static void lod_read_lock(const struct lu_env *env, struct dt_object *dt,
1058 dt_read_lock(env, dt_object_child(dt), role);
1062 * Implementation of dt_object_operations::do_write_lock.
1064 * \see dt_object_operations::do_write_lock() in the API description for
1067 static void lod_write_lock(const struct lu_env *env, struct dt_object *dt,
1070 dt_write_lock(env, dt_object_child(dt), role);
1074 * Implementation of dt_object_operations::do_read_unlock.
1076 * \see dt_object_operations::do_read_unlock() in the API description for
1079 static void lod_read_unlock(const struct lu_env *env, struct dt_object *dt)
1081 dt_read_unlock(env, dt_object_child(dt));
1085 * Implementation of dt_object_operations::do_write_unlock.
1087 * \see dt_object_operations::do_write_unlock() in the API description for
1090 static void lod_write_unlock(const struct lu_env *env, struct dt_object *dt)
1092 dt_write_unlock(env, dt_object_child(dt));
1096 * Implementation of dt_object_operations::do_write_locked.
1098 * \see dt_object_operations::do_write_locked() in the API description for
1101 static int lod_write_locked(const struct lu_env *env, struct dt_object *dt)
1103 return dt_write_locked(env, dt_object_child(dt));
1107 * Implementation of dt_object_operations::do_attr_get.
1109 * \see dt_object_operations::do_attr_get() in the API description for details.
1111 static int lod_attr_get(const struct lu_env *env,
1112 struct dt_object *dt,
1113 struct lu_attr *attr)
1115 /* Note: for striped directory, client will merge attributes
1116 * from all of the sub-stripes see lmv_merge_attr(), and there
1117 * no MDD logic depend on directory nlink/size/time, so we can
1118 * always use master inode nlink and size for now. */
1119 return dt_attr_get(env, dt_object_child(dt), attr);
1122 void lod_adjust_stripe_size(struct lod_layout_component *comp,
1123 __u32 def_stripe_size)
1125 __u64 comp_end = comp->llc_extent.e_end;
1127 /* Choose stripe size if not set. Note that default stripe size can't
1128 * be used as is, because it must be multiplier of given component end.
1129 * - first check if default stripe size can be used
1130 * - if not than select the lowest set bit from component end and use
1131 * that value as stripe size
1133 if (!comp->llc_stripe_size) {
1134 if (comp_end == LUSTRE_EOF || !(comp_end % def_stripe_size))
1135 comp->llc_stripe_size = def_stripe_size;
1137 comp->llc_stripe_size = comp_end & ~(comp_end - 1);
1139 /* check stripe size is multiplier of comp_end */
1140 if (comp_end != LUSTRE_EOF &&
1141 comp_end % comp->llc_stripe_size) {
1142 /* fix that even for defined stripe size but warn
1143 * about the problem, that must not happen
1145 CWARN("Component end %llu is not aligned by the stripe size %u\n",
1146 comp_end, comp->llc_stripe_size);
1148 comp->llc_stripe_size = comp_end & ~(comp_end - 1);
1153 static inline void lod_adjust_stripe_info(struct lod_layout_component *comp,
1154 struct lov_desc *desc,
1157 if (comp->llc_pattern != LOV_PATTERN_MDT) {
1158 if (append_stripes) {
1159 comp->llc_stripe_count = append_stripes;
1160 } else if (!comp->llc_stripe_count) {
1161 comp->llc_stripe_count =
1162 desc->ld_default_stripe_count;
1166 lod_adjust_stripe_size(comp, desc->ld_default_stripe_size);
1169 int lod_obj_for_each_stripe(const struct lu_env *env, struct lod_object *lo,
1171 struct lod_obj_stripe_cb_data *data)
1173 struct lod_layout_component *lod_comp;
1177 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1178 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1179 lod_comp = &lo->ldo_comp_entries[i];
1181 if (lod_comp->llc_stripe == NULL)
1184 /* has stripe but not inited yet, this component has been
1185 * declared to be created, but hasn't created yet.
1187 if (!lod_comp_inited(lod_comp))
1190 if (data->locd_comp_skip_cb &&
1191 data->locd_comp_skip_cb(env, lo, i, data))
1194 if (data->locd_comp_cb) {
1195 rc = data->locd_comp_cb(env, lo, i, data);
1200 /* could used just to do sth about component, not each
1203 if (!data->locd_stripe_cb)
1206 LASSERT(lod_comp->llc_stripe_count > 0);
1207 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
1208 struct dt_object *dt = lod_comp->llc_stripe[j];
1212 rc = data->locd_stripe_cb(env, lo, dt, th, i, j, data);
1220 static bool lod_obj_attr_set_comp_skip_cb(const struct lu_env *env,
1221 struct lod_object *lo, int comp_idx,
1222 struct lod_obj_stripe_cb_data *data)
1224 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[comp_idx];
1225 bool skipped = false;
1227 if (!(data->locd_attr->la_valid & LA_LAYOUT_VERSION))
1230 switch (lo->ldo_flr_state) {
1231 case LCM_FL_WRITE_PENDING: {
1234 /* skip stale components */
1235 if (lod_comp->llc_flags & LCME_FL_STALE) {
1240 /* skip valid and overlapping components, therefore any
1241 * attempts to write overlapped components will never succeed
1242 * because client will get EINPROGRESS. */
1243 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1247 if (lo->ldo_comp_entries[i].llc_flags & LCME_FL_STALE)
1250 if (lu_extent_is_overlapped(&lod_comp->llc_extent,
1251 &lo->ldo_comp_entries[i].llc_extent)) {
1259 LASSERTF(0, "impossible: %d\n", lo->ldo_flr_state);
1260 case LCM_FL_SYNC_PENDING:
1264 CDEBUG(D_LAYOUT, DFID": %s to set component %x to version: %u\n",
1265 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
1266 skipped ? "skipped" : "chose", lod_comp->llc_id,
1267 data->locd_attr->la_layout_version);
1273 lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
1274 struct dt_object *dt, struct thandle *th,
1275 int comp_idx, int stripe_idx,
1276 struct lod_obj_stripe_cb_data *data)
1278 if (data->locd_declare)
1279 return lod_sub_declare_attr_set(env, dt, data->locd_attr, th);
1281 if (data->locd_attr->la_valid & LA_LAYOUT_VERSION) {
1282 CDEBUG(D_LAYOUT, DFID": set layout version: %u, comp_idx: %d\n",
1283 PFID(lu_object_fid(&dt->do_lu)),
1284 data->locd_attr->la_layout_version, comp_idx);
1287 return lod_sub_attr_set(env, dt, data->locd_attr, th);
1291 * Implementation of dt_object_operations::do_declare_attr_set.
1293 * If the object is striped, then apply the changes to all the stripes.
1295 * \see dt_object_operations::do_declare_attr_set() in the API description
1298 static int lod_declare_attr_set(const struct lu_env *env,
1299 struct dt_object *dt,
1300 const struct lu_attr *attr,
1303 struct dt_object *next = dt_object_child(dt);
1304 struct lod_object *lo = lod_dt_obj(dt);
1309 * declare setattr on the local object
1311 rc = lod_sub_declare_attr_set(env, next, attr, th);
1315 /* osp_declare_attr_set() ignores all attributes other than
1316 * UID, GID, PROJID, and size, and osp_attr_set() ignores all
1317 * but UID, GID and PROJID. Declaration of size attr setting
1318 * happens through lod_declare_init_size(), and not through
1319 * this function. Therefore we need not load striping unless
1320 * ownership is changing. This should save memory and (we hope)
1321 * speed up rename().
1323 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1324 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1327 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1330 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID | LA_MODE |
1331 LA_ATIME | LA_MTIME | LA_CTIME |
1336 * load striping information, notice we don't do this when object
1337 * is being initialized as we don't need this information till
1338 * few specific cases like destroy, chown
1340 rc = lod_striping_load(env, lo);
1344 if (!lod_obj_is_striped(dt))
1348 * if object is striped declare changes on the stripes
1350 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1351 LASSERT(lo->ldo_stripe);
1352 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1353 if (lo->ldo_stripe[i] == NULL)
1355 if (!dt_object_exists(lo->ldo_stripe[i]))
1357 rc = lod_sub_declare_attr_set(env, lo->ldo_stripe[i],
1363 struct lod_obj_stripe_cb_data data = { { 0 } };
1365 data.locd_attr = attr;
1366 data.locd_declare = true;
1367 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1368 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1374 if (!dt_object_exists(next) || dt_object_remote(next) ||
1375 !S_ISREG(attr->la_mode))
1378 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1379 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
1383 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) ||
1384 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1385 struct lod_thread_info *info = lod_env_info(env);
1386 struct lu_buf *buf = &info->lti_buf;
1388 buf->lb_buf = info->lti_ea_store;
1389 buf->lb_len = info->lti_ea_store_size;
1390 rc = lod_sub_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1391 LU_XATTR_REPLACE, th);
1398 * Implementation of dt_object_operations::do_attr_set.
1400 * If the object is striped, then apply the changes to all or subset of
1401 * the stripes depending on the object type and specific attributes.
1403 * \see dt_object_operations::do_attr_set() in the API description for details.
1405 static int lod_attr_set(const struct lu_env *env,
1406 struct dt_object *dt,
1407 const struct lu_attr *attr,
1410 struct dt_object *next = dt_object_child(dt);
1411 struct lod_object *lo = lod_dt_obj(dt);
1416 * apply changes to the local object
1418 rc = lod_sub_attr_set(env, next, attr, th);
1422 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1423 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1426 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1429 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE | LA_PROJID |
1430 LA_ATIME | LA_MTIME | LA_CTIME |
1435 /* FIXME: a tricky case in the code path of mdd_layout_change():
1436 * the in-memory striping information has been freed in lod_xattr_set()
1437 * due to layout change. It has to load stripe here again. It only
1438 * changes flags of layout so declare_attr_set() is still accurate */
1439 rc = lod_striping_load(env, lo);
1443 if (!lod_obj_is_striped(dt))
1447 * if object is striped, apply changes to all the stripes
1449 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1450 LASSERT(lo->ldo_stripe);
1451 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1452 if (unlikely(lo->ldo_stripe[i] == NULL))
1455 if ((dt_object_exists(lo->ldo_stripe[i]) == 0))
1458 rc = lod_sub_attr_set(env, lo->ldo_stripe[i], attr, th);
1463 struct lod_obj_stripe_cb_data data = { { 0 } };
1465 data.locd_attr = attr;
1466 data.locd_declare = false;
1467 data.locd_comp_skip_cb = lod_obj_attr_set_comp_skip_cb;
1468 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1469 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1475 if (!dt_object_exists(next) || dt_object_remote(next) ||
1476 !S_ISREG(attr->la_mode))
1479 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1480 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
1484 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE)) {
1485 struct lod_thread_info *info = lod_env_info(env);
1486 struct lu_buf *buf = &info->lti_buf;
1487 struct ost_id *oi = &info->lti_ostid;
1488 struct lu_fid *fid = &info->lti_fid;
1489 struct lov_mds_md_v1 *lmm;
1490 struct lov_ost_data_v1 *objs;
1493 rc = lod_get_lov_ea(env, lo);
1497 buf->lb_buf = info->lti_ea_store;
1498 buf->lb_len = info->lti_ea_store_size;
1499 lmm = info->lti_ea_store;
1500 magic = le32_to_cpu(lmm->lmm_magic);
1501 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1502 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1503 struct lov_comp_md_entry_v1 *lcme =
1504 &lcm->lcm_entries[0];
1506 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
1507 magic = le32_to_cpu(lmm->lmm_magic);
1510 if (magic == LOV_MAGIC_V1)
1511 objs = &(lmm->lmm_objects[0]);
1513 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1514 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1515 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1517 fid_to_ostid(fid, oi);
1518 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1520 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1521 LU_XATTR_REPLACE, th);
1522 } else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1523 struct lod_thread_info *info = lod_env_info(env);
1524 struct lu_buf *buf = &info->lti_buf;
1525 struct lov_comp_md_v1 *lcm;
1526 struct lov_comp_md_entry_v1 *lcme;
1528 rc = lod_get_lov_ea(env, lo);
1532 buf->lb_buf = info->lti_ea_store;
1533 buf->lb_len = info->lti_ea_store_size;
1535 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1 &&
1536 le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_SEL)
1539 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1540 lcme = &lcm->lcm_entries[0];
1541 le64_add_cpu(&lcme->lcme_extent.e_start, 1);
1542 le64_add_cpu(&lcme->lcme_extent.e_end, -1);
1544 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1545 LU_XATTR_REPLACE, th);
1552 * Implementation of dt_object_operations::do_xattr_get.
1554 * If LOV EA is requested from the root object and it's not
1555 * found, then return default striping for the filesystem.
1557 * \see dt_object_operations::do_xattr_get() in the API description for details.
1559 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1560 struct lu_buf *buf, const char *name)
1562 struct lod_thread_info *info = lod_env_info(env);
1563 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1568 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1569 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1570 struct lmv_mds_md_v1 *lmv1;
1571 struct lmv_foreign_md *lfm;
1574 if (rc > (typeof(rc))sizeof(*lmv1))
1577 /* short (<= sizeof(struct lmv_mds_md_v1)) foreign LMV case */
1578 /* XXX empty foreign LMV is not allowed */
1579 if (rc <= offsetof(typeof(*lfm), lfm_value))
1580 RETURN(rc = rc > 0 ? -EINVAL : rc);
1582 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1583 BUILD_BUG_ON(sizeof(*lmv1) > sizeof(info->lti_key));
1585 /* lti_buf is large enough for *lmv1 or a short
1586 * (<= sizeof(struct lmv_mds_md_v1)) foreign LMV
1588 info->lti_buf.lb_buf = info->lti_key;
1589 info->lti_buf.lb_len = sizeof(*lmv1);
1590 rc = dt_xattr_get(env, dt_object_child(dt),
1591 &info->lti_buf, name);
1592 if (unlikely(rc <= offsetof(typeof(*lfm),
1594 RETURN(rc = rc > 0 ? -EINVAL : rc);
1596 lfm = info->lti_buf.lb_buf;
1597 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN)
1600 if (unlikely(rc != sizeof(*lmv1)))
1601 RETURN(rc = rc > 0 ? -EINVAL : rc);
1603 lmv1 = info->lti_buf.lb_buf;
1604 /* The on-disk LMV EA only contains header, but the
1605 * returned LMV EA size should contain the space for
1606 * the FIDs of all shards of the striped directory. */
1607 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1608 rc = lmv_mds_md_size(
1609 le32_to_cpu(lmv1->lmv_stripe_count),
1610 le32_to_cpu(lmv1->lmv_magic));
1613 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1616 if (rc != sizeof(*lmv1))
1617 RETURN(rc = rc > 0 ? -EINVAL : rc);
1619 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1623 RETURN(rc = rc1 != 0 ? rc1 : rc);
1626 if ((rc > 0) && buf->lb_buf && strcmp(name, XATTR_NAME_LOV) == 0) {
1627 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1629 if (lcm->lcm_magic == cpu_to_le32(LOV_MAGIC_SEL))
1630 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
1633 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1637 * XXX: Only used by lfsck
1639 * lod returns default striping on the real root of the device
1640 * this is like the root stores default striping for the whole
1641 * filesystem. historically we've been using a different approach
1642 * and store it in the config.
1644 dt_root_get(env, dev->lod_child, &info->lti_fid);
1645 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1647 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1648 struct lov_user_md *lum = buf->lb_buf;
1649 struct lov_desc *desc = &dev->lod_ost_descs.ltd_lov_desc;
1651 if (buf->lb_buf == NULL) {
1653 } else if (buf->lb_len >= sizeof(*lum)) {
1654 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1655 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1656 lmm_oi_set_id(&lum->lmm_oi, 0);
1657 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1658 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1659 lum->lmm_stripe_size = cpu_to_le32(
1660 desc->ld_default_stripe_size);
1661 lum->lmm_stripe_count = cpu_to_le16(
1662 desc->ld_default_stripe_count);
1663 lum->lmm_stripe_offset = cpu_to_le16(
1664 desc->ld_default_stripe_offset);
1677 * Checks that the magic of the stripe is sane.
1679 * \param[in] lod lod device
1680 * \param[in] lum a buffer storing LMV EA to verify
1682 * \retval 0 if the EA is sane
1683 * \retval negative otherwise
1685 static int lod_verify_md_striping(struct lod_device *lod,
1686 const struct lmv_user_md_v1 *lum)
1688 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1689 CERROR("%s: invalid lmv_user_md: magic = %x, "
1690 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1691 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1692 (int)le32_to_cpu(lum->lum_stripe_offset),
1693 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1701 * Initialize LMV EA for a slave.
1703 * Initialize slave's LMV EA from the master's LMV EA.
1705 * \param[in] master_lmv a buffer containing master's EA
1706 * \param[out] slave_lmv a buffer where slave's EA will be stored
1709 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1710 const struct lmv_mds_md_v1 *master_lmv)
1712 *slave_lmv = *master_lmv;
1713 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1719 * Generate LMV EA from the object passed as \a dt. The object must have
1720 * the stripes created and initialized.
1722 * \param[in] env execution environment
1723 * \param[in] dt object
1724 * \param[out] lmv_buf buffer storing generated LMV EA
1726 * \retval 0 on success
1727 * \retval negative if failed
1729 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1730 struct lu_buf *lmv_buf)
1732 struct lod_thread_info *info = lod_env_info(env);
1733 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1734 struct lod_object *lo = lod_dt_obj(dt);
1735 struct lmv_mds_md_v1 *lmm1;
1737 int type = LU_SEQ_RANGE_ANY;
1742 LASSERT(lo->ldo_dir_striped != 0);
1743 LASSERT(lo->ldo_dir_stripe_count > 0);
1744 stripe_count = lo->ldo_dir_stripe_count;
1745 /* Only store the LMV EA heahder on the disk. */
1746 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1747 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1751 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1754 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1755 memset(lmm1, 0, sizeof(*lmm1));
1756 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1757 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1758 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1759 lmm1->lmv_layout_version = cpu_to_le32(lo->ldo_dir_layout_version);
1760 if (lod_is_layout_changing(lo)) {
1761 lmm1->lmv_migrate_hash = cpu_to_le32(lo->ldo_dir_migrate_hash);
1762 lmm1->lmv_migrate_offset =
1763 cpu_to_le32(lo->ldo_dir_migrate_offset);
1765 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1770 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1771 lmv_buf->lb_buf = info->lti_ea_store;
1772 lmv_buf->lb_len = sizeof(*lmm1);
1778 * Create in-core represenation for a striped directory.
1780 * Parse the buffer containing LMV EA and instantiate LU objects
1781 * representing the stripe objects. The pointers to the objects are
1782 * stored in ldo_stripe field of \a lo. This function is used when
1783 * we need to access an already created object (i.e. load from a disk).
1785 * \param[in] env execution environment
1786 * \param[in] lo lod object
1787 * \param[in] buf buffer containing LMV EA
1789 * \retval 0 on success
1790 * \retval negative if failed
1792 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1793 const struct lu_buf *buf)
1795 struct lod_thread_info *info = lod_env_info(env);
1796 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1797 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1798 struct dt_object **stripe;
1799 union lmv_mds_md *lmm = buf->lb_buf;
1800 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1801 struct lu_fid *fid = &info->lti_fid;
1806 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1808 /* XXX may be useless as not called for foreign LMV ?? */
1809 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_FOREIGN)
1812 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1813 lo->ldo_dir_slave_stripe = 1;
1817 if (!lmv_is_sane(lmv1))
1820 LASSERT(lo->ldo_stripe == NULL);
1821 OBD_ALLOC_PTR_ARRAY(stripe, le32_to_cpu(lmv1->lmv_stripe_count));
1825 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1826 struct dt_device *tgt_dt;
1827 struct dt_object *dto;
1828 int type = LU_SEQ_RANGE_ANY;
1831 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1832 if (!fid_is_sane(fid)) {
1837 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1841 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1842 tgt_dt = lod->lod_child;
1844 struct lod_tgt_desc *tgt;
1846 tgt = LTD_TGT(ltd, idx);
1848 GOTO(out, rc = -ESTALE);
1849 tgt_dt = tgt->ltd_tgt;
1852 dto = dt_locate_at(env, tgt_dt, fid,
1853 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1856 GOTO(out, rc = PTR_ERR(dto));
1861 lo->ldo_stripe = stripe;
1862 lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1863 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1864 lo->ldo_dir_layout_version = le32_to_cpu(lmv1->lmv_layout_version);
1865 lo->ldo_dir_migrate_offset = le32_to_cpu(lmv1->lmv_migrate_offset);
1866 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv1->lmv_migrate_hash);
1867 lo->ldo_dir_hash_type = le32_to_cpu(lmv1->lmv_hash_type);
1869 lod_striping_free_nolock(env, lo);
1875 * Declare create a striped directory.
1877 * Declare creating a striped directory with a given stripe pattern on the
1878 * specified MDTs. A striped directory is represented as a regular directory
1879 * - an index listing all the stripes. The stripes point back to the master
1880 * object with ".." and LinkEA. The master object gets LMV EA which
1881 * identifies it as a striped directory. The function allocates FIDs
1884 * \param[in] env execution environment
1885 * \param[in] dt object
1886 * \param[in] attr attributes to initialize the objects with
1887 * \param[in] dof type of objects to be created
1888 * \param[in] th transaction handle
1890 * \retval 0 on success
1891 * \retval negative if failed
1893 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1894 struct dt_object *dt,
1895 struct lu_attr *attr,
1896 struct dt_object_format *dof,
1899 struct lod_thread_info *info = lod_env_info(env);
1900 struct lu_buf lmv_buf;
1901 struct lu_buf slave_lmv_buf;
1902 struct lmv_mds_md_v1 *lmm;
1903 struct lmv_mds_md_v1 *slave_lmm = NULL;
1904 struct dt_insert_rec *rec = &info->lti_dt_rec;
1905 struct lod_object *lo = lod_dt_obj(dt);
1910 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1913 lmm = lmv_buf.lb_buf;
1915 OBD_ALLOC_PTR(slave_lmm);
1916 if (slave_lmm == NULL)
1917 GOTO(out, rc = -ENOMEM);
1919 lod_prep_slave_lmv_md(slave_lmm, lmm);
1920 slave_lmv_buf.lb_buf = slave_lmm;
1921 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1923 if (!dt_try_as_dir(env, dt_object_child(dt)))
1924 GOTO(out, rc = -EINVAL);
1926 rec->rec_type = S_IFDIR;
1927 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1928 struct dt_object *dto = lo->ldo_stripe[i];
1929 char *stripe_name = info->lti_key;
1930 struct lu_name *sname;
1931 struct linkea_data ldata = { NULL };
1932 struct lu_buf linkea_buf;
1934 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
1938 /* directory split skip create for existing stripes */
1939 if (!(lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) {
1940 rc = lod_sub_declare_create(env, dto, attr, NULL, dof,
1945 if (!dt_try_as_dir(env, dto))
1946 GOTO(out, rc = -EINVAL);
1948 rc = lod_sub_declare_ref_add(env, dto, th);
1952 rec->rec_fid = lu_object_fid(&dto->do_lu);
1953 rc = lod_sub_declare_insert(env, dto,
1954 (const struct dt_rec *)rec,
1955 (const struct dt_key *)dot,
1960 /* master stripe FID will be put to .. */
1961 rec->rec_fid = lu_object_fid(&dt->do_lu);
1962 rc = lod_sub_declare_insert(env, dto,
1963 (const struct dt_rec *)rec,
1964 (const struct dt_key *)dotdot,
1969 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1971 snprintf(stripe_name, sizeof(info->lti_key),
1973 PFID(lu_object_fid(&dto->do_lu)),
1976 snprintf(stripe_name, sizeof(info->lti_key),
1978 PFID(lu_object_fid(&dto->do_lu)), i);
1980 sname = lod_name_get(env, stripe_name,
1981 strlen(stripe_name));
1982 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1983 sname, lu_object_fid(&dt->do_lu));
1987 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1988 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1989 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
1990 XATTR_NAME_LINK, 0, th);
1994 rec->rec_fid = lu_object_fid(&dto->do_lu);
1995 rc = lod_sub_declare_insert(env, dt_object_child(dt),
1996 (const struct dt_rec *)rec,
1997 (const struct dt_key *)stripe_name, th);
2001 rc = lod_sub_declare_ref_add(env, dt_object_child(dt),
2007 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
2008 cfs_fail_val != i) {
2009 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
2011 slave_lmm->lmv_master_mdt_index =
2014 slave_lmm->lmv_master_mdt_index =
2016 rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
2017 XATTR_NAME_LMV, 0, th);
2023 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
2024 &lmv_buf, XATTR_NAME_LMV, 0, th);
2028 if (slave_lmm != NULL)
2029 OBD_FREE_PTR(slave_lmm);
2035 * Allocate a striping on a predefined set of MDTs.
2037 * Allocates new striping using the MDT index range provided by the data from
2038 * the lum_obejcts contained in the lmv_user_md passed to this method if
2039 * \a is_specific is true; or allocates new layout starting from MDT index in
2040 * lo->ldo_dir_stripe_offset. The exact order of MDTs is not important and
2041 * varies depending on MDT status. The number of stripes needed and stripe
2042 * offset are taken from the object. If that number cannot be met, then the
2043 * function returns an error and then it's the caller's responsibility to
2044 * release the stripes allocated. All the internal structures are protected,
2045 * but no concurrent allocation is allowed on the same objects.
2047 * \param[in] env execution environment for this thread
2048 * \param[in] lo LOD object
2049 * \param[out] stripes striping created
2050 * \param[out] mdt_indices MDT indices of striping created
2051 * \param[in] is_specific true if the MDTs are provided by lum; false if
2052 * only the starting MDT index is provided
2054 * \retval positive stripes allocated, including the first stripe allocated
2056 * \retval negative errno on failure
2058 static int lod_mdt_alloc_specific(const struct lu_env *env,
2059 struct lod_object *lo,
2060 struct dt_object **stripes,
2061 __u32 *mdt_indices, bool is_specific)
2063 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
2064 struct lu_tgt_descs *ltd = &lod->lod_mdt_descs;
2065 struct lu_tgt_desc *tgt = NULL;
2066 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2067 struct dt_device *tgt_dt = NULL;
2068 struct lu_fid fid = { 0 };
2069 struct dt_object *dto;
2071 u32 stripe_count = lo->ldo_dir_stripe_count;
2077 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
2078 if (stripe_count > 1)
2079 /* Set the start index for the 2nd stripe allocation */
2080 mdt_indices[1] = (mdt_indices[0] + 1) %
2081 (lod->lod_remote_mdt_count + 1);
2083 for (; stripe_idx < stripe_count; stripe_idx++) {
2084 /* Try to find next avaible target */
2085 idx = mdt_indices[stripe_idx];
2086 for (j = 0; j < lod->lod_remote_mdt_count;
2087 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
2088 bool already_allocated = false;
2091 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
2092 idx, lod->lod_remote_mdt_count + 1, stripe_idx);
2094 if (likely(!is_specific &&
2095 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
2096 /* check whether the idx already exists
2097 * in current allocated array */
2098 for (k = 0; k < stripe_idx; k++) {
2099 if (mdt_indices[k] == idx) {
2100 already_allocated = true;
2105 if (already_allocated)
2109 /* Sigh, this index is not in the bitmap, let's check
2110 * next available target */
2111 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) &&
2112 idx != master_index)
2115 if (idx == master_index) {
2116 /* Allocate the FID locally */
2117 tgt_dt = lod->lod_child;
2118 rc = dt_fid_alloc(env, tgt_dt, &fid, NULL,
2125 /* check the status of the OSP */
2126 tgt = LTD_TGT(ltd, idx);
2130 tgt_dt = tgt->ltd_tgt;
2131 if (!tgt->ltd_active)
2132 /* this OSP doesn't feel well */
2135 rc = dt_fid_alloc(env, tgt_dt, &fid, NULL, NULL);
2142 /* Can not allocate more stripes */
2143 if (j == lod->lod_remote_mdt_count) {
2144 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
2145 lod2obd(lod)->obd_name, stripe_count,
2150 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
2151 idx, stripe_idx, PFID(&fid));
2152 mdt_indices[stripe_idx] = idx;
2153 /* Set the start index for next stripe allocation */
2154 if (!is_specific && stripe_idx < stripe_count - 1) {
2156 * for large dir test, put all other slaves on one
2157 * remote MDT, otherwise we may save too many local
2158 * slave locks which will exceed RS_MAX_LOCKS.
2160 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)))
2162 mdt_indices[stripe_idx + 1] = (idx + 1) %
2163 (lod->lod_remote_mdt_count + 1);
2165 /* tgt_dt and fid must be ready after search avaible OSP
2166 * in the above loop */
2167 LASSERT(tgt_dt != NULL);
2168 LASSERT(fid_is_sane(&fid));
2170 /* fail a remote stripe FID allocation */
2171 if (stripe_idx && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_FID))
2174 dto = dt_locate_at(env, tgt_dt, &fid,
2175 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
2182 stripes[stripe_idx] = dto;
2188 for (j = 1; j < stripe_idx; j++) {
2189 LASSERT(stripes[j] != NULL);
2190 dt_object_put(env, stripes[j]);
2196 static int lod_prep_md_striped_create(const struct lu_env *env,
2197 struct dt_object *dt,
2198 struct lu_attr *attr,
2199 const struct lmv_user_md_v1 *lum,
2200 struct dt_object_format *dof,
2203 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
2204 struct lod_object *lo = lod_dt_obj(dt);
2205 struct dt_object **stripes;
2206 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2207 struct lu_fid fid = { 0 };
2214 /* The lum has been verifed in lod_verify_md_striping */
2215 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
2216 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
2218 stripe_count = lo->ldo_dir_stripe_count;
2220 OBD_ALLOC_PTR_ARRAY(stripes, stripe_count);
2224 /* Allocate the first stripe locally */
2225 rc = dt_fid_alloc(env, lod->lod_child, &fid, NULL, NULL);
2229 stripes[0] = dt_locate_at(env, lod->lod_child, &fid,
2230 dt->do_lu.lo_dev->ld_site->ls_top_dev, &conf);
2231 if (IS_ERR(stripes[0]))
2232 GOTO(out, rc = PTR_ERR(stripes[0]));
2234 if (lo->ldo_dir_stripe_offset == LMV_OFFSET_DEFAULT) {
2235 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
2236 rc = lod_mdt_alloc_qos(env, lo, stripes, 1, stripe_count);
2238 rc = lod_mdt_alloc_rr(env, lo, stripes, 1,
2242 bool is_specific = false;
2244 OBD_ALLOC_PTR_ARRAY(idx_array, stripe_count);
2246 GOTO(out, rc = -ENOMEM);
2248 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
2250 for (i = 0; i < stripe_count; i++)
2252 le32_to_cpu(lum->lum_objects[i].lum_mds);
2255 /* stripe 0 is local */
2257 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
2258 rc = lod_mdt_alloc_specific(env, lo, stripes, idx_array,
2260 OBD_FREE_PTR_ARRAY(idx_array, stripe_count);
2268 lo->ldo_dir_striped = 1;
2269 lo->ldo_stripe = stripes;
2270 lo->ldo_dir_stripe_count = rc;
2271 lo->ldo_dir_stripes_allocated = stripe_count;
2273 lo->ldo_dir_stripe_loaded = 1;
2275 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2277 lod_striping_free(env, lo);
2283 if (!IS_ERR_OR_NULL(stripes[0]))
2284 dt_object_put(env, stripes[0]);
2285 for (i = 1; i < stripe_count; i++)
2286 LASSERT(!stripes[i]);
2287 OBD_FREE_PTR_ARRAY(stripes, stripe_count);
2294 * Alloc cached foreign LMV
2296 * \param[in] lo object
2297 * \param[in] size size of foreign LMV
2299 * \retval 0 on success
2300 * \retval negative if failed
2302 int lod_alloc_foreign_lmv(struct lod_object *lo, size_t size)
2304 OBD_ALLOC_LARGE(lo->ldo_foreign_lmv, size);
2305 if (lo->ldo_foreign_lmv == NULL)
2307 lo->ldo_foreign_lmv_size = size;
2308 lo->ldo_dir_is_foreign = 1;
2314 * Declare create striped md object.
2316 * The function declares intention to create a striped directory. This is a
2317 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2318 * is to verify pattern \a lum_buf is good. Check that function for the details.
2320 * \param[in] env execution environment
2321 * \param[in] dt object
2322 * \param[in] attr attributes to initialize the objects with
2323 * \param[in] lum_buf a pattern specifying the number of stripes and
2325 * \param[in] dof type of objects to be created
2326 * \param[in] th transaction handle
2328 * \retval 0 on success
2329 * \retval negative if failed
2332 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2333 struct dt_object *dt,
2334 struct lu_attr *attr,
2335 const struct lu_buf *lum_buf,
2336 struct dt_object_format *dof,
2339 struct lod_object *lo = lod_dt_obj(dt);
2340 struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
2344 LASSERT(lum != NULL);
2346 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
2347 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
2348 (int)le32_to_cpu(lum->lum_stripe_offset));
2350 if (lo->ldo_dir_stripe_count == 0) {
2351 if (lo->ldo_dir_is_foreign) {
2352 rc = lod_alloc_foreign_lmv(lo, lum_buf->lb_len);
2355 memcpy(lo->ldo_foreign_lmv, lum, lum_buf->lb_len);
2356 lo->ldo_dir_stripe_loaded = 1;
2361 /* prepare dir striped objects */
2362 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2364 /* failed to create striping, let's reset
2365 * config so that others don't get confused */
2366 lod_striping_free(env, lo);
2374 * Set or replace striped directory layout, and LFSCK may set layout on a plain
2375 * directory, so don't check stripe count.
2377 * \param[in] env execution environment
2378 * \param[in] dt target object
2379 * \param[in] buf LMV buf which contains source stripe fids
2380 * \param[in] fl set or replace
2381 * \param[in] th transaction handle
2383 * \retval 0 on success
2384 * \retval negative if failed
2386 static int lod_dir_layout_set(const struct lu_env *env,
2387 struct dt_object *dt,
2388 const struct lu_buf *buf,
2392 struct dt_object *next = dt_object_child(dt);
2393 struct lod_object *lo = lod_dt_obj(dt);
2394 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2395 struct lmv_mds_md_v1 *lmv = buf->lb_buf;
2396 struct lmv_mds_md_v1 *slave_lmv;
2397 struct lu_buf slave_buf;
2403 if (!lmv_is_sane2(lmv))
2406 /* adjust hash for dir merge, which may not be set in user command */
2407 if (lmv_is_merging(lmv) && !lmv->lmv_migrate_hash)
2408 lmv->lmv_merge_hash =
2409 lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
2411 LMV_DEBUG(D_INFO, lmv, "set");
2413 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LMV, fl, th);
2417 /* directory restripe may update stripe LMV directly */
2418 if (!lo->ldo_dir_stripe_count)
2421 lo->ldo_dir_hash_type = le32_to_cpu(lmv->lmv_hash_type);
2422 lo->ldo_dir_migrate_offset = le32_to_cpu(lmv->lmv_migrate_offset);
2423 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_migrate_hash);
2424 lo->ldo_dir_layout_version = le32_to_cpu(lmv->lmv_layout_version);
2426 OBD_ALLOC_PTR(slave_lmv);
2430 lod_prep_slave_lmv_md(slave_lmv, lmv);
2431 slave_buf.lb_buf = slave_lmv;
2432 slave_buf.lb_len = sizeof(*slave_lmv);
2434 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2435 if (!lo->ldo_stripe[i])
2438 if (!dt_object_exists(lo->ldo_stripe[i]))
2441 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], &slave_buf,
2442 XATTR_NAME_LMV, fl, th);
2447 OBD_FREE_PTR(slave_lmv);
2453 * Implementation of dt_object_operations::do_declare_xattr_set.
2455 * Used with regular (non-striped) objects. Basically it
2456 * initializes the striping information and applies the
2457 * change to all the stripes.
2459 * \see dt_object_operations::do_declare_xattr_set() in the API description
2462 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2463 struct dt_object *dt,
2464 const struct lu_buf *buf,
2465 const char *name, int fl,
2468 struct dt_object *next = dt_object_child(dt);
2469 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2470 struct lod_object *lo = lod_dt_obj(dt);
2475 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2476 struct lmv_user_md_v1 *lum;
2478 LASSERT(buf != NULL && buf->lb_buf != NULL);
2480 rc = lod_verify_md_striping(d, lum);
2483 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2484 rc = lod_verify_striping(env, d, lo, buf, false);
2489 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2493 /* Note: Do not set LinkEA on sub-stripes, otherwise
2494 * it will confuse the fid2path process(see mdt_path_current()).
2495 * The linkEA between master and sub-stripes is set in
2496 * lod_xattr_set_lmv(). */
2497 if (strcmp(name, XATTR_NAME_LINK) == 0)
2500 /* set xattr to each stripes, if needed */
2501 rc = lod_striping_load(env, lo);
2505 if (lo->ldo_dir_stripe_count == 0)
2508 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2509 if (!lo->ldo_stripe[i])
2512 if (!dt_object_exists(lo->ldo_stripe[i]))
2515 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2525 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2526 struct lod_object *lo,
2527 struct dt_object *dt, struct thandle *th,
2528 int comp_idx, int stripe_idx,
2529 struct lod_obj_stripe_cb_data *data)
2531 struct lod_thread_info *info = lod_env_info(env);
2532 struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
2533 struct filter_fid *ff = &info->lti_ff;
2534 struct lu_buf *buf = &info->lti_buf;
2538 buf->lb_len = sizeof(*ff);
2539 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2547 * locd_buf is set if it's called by dir migration, which doesn't check
2550 if (data->locd_buf) {
2551 memset(ff, 0, sizeof(*ff));
2552 ff->ff_parent = *(struct lu_fid *)data->locd_buf->lb_buf;
2554 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
2556 if (lu_fid_eq(lod_object_fid(lo), &ff->ff_parent) &&
2557 ff->ff_layout.ol_comp_id == comp->llc_id)
2560 memset(ff, 0, sizeof(*ff));
2561 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2564 /* rewrite filter_fid */
2565 ff->ff_parent.f_ver = stripe_idx;
2566 ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
2567 ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
2568 ff->ff_layout.ol_comp_id = comp->llc_id;
2569 ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
2570 ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
2571 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
2573 if (data->locd_declare)
2574 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2575 LU_XATTR_REPLACE, th);
2577 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2578 LU_XATTR_REPLACE, th);
2584 * Reset parent FID on OST object
2586 * Replace parent FID with @dt object FID, which is only called during migration
2587 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2588 * the FID is changed.
2590 * \param[in] env execution environment
2591 * \param[in] dt dt_object whose stripes's parent FID will be reset
2592 * \parem[in] th thandle
2593 * \param[in] declare if it is declare
2595 * \retval 0 if reset succeeds
2596 * \retval negative errno if reset fails
2598 static int lod_replace_parent_fid(const struct lu_env *env,
2599 struct dt_object *dt,
2600 const struct lu_buf *buf,
2601 struct thandle *th, bool declare)
2603 struct lod_object *lo = lod_dt_obj(dt);
2604 struct lod_thread_info *info = lod_env_info(env);
2605 struct filter_fid *ff;
2606 struct lod_obj_stripe_cb_data data = { { 0 } };
2610 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2612 /* set xattr to each stripes, if needed */
2613 rc = lod_striping_load(env, lo);
2617 if (!lod_obj_is_striped(dt))
2620 if (info->lti_ea_store_size < sizeof(*ff)) {
2621 rc = lod_ea_store_resize(info, sizeof(*ff));
2626 data.locd_declare = declare;
2627 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2628 data.locd_buf = buf;
2629 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2634 inline __u16 lod_comp_entry_stripe_count(struct lod_object *lo,
2635 struct lod_layout_component *entry,
2638 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2642 else if (lod_comp_inited(entry))
2643 return entry->llc_stripe_count;
2644 else if ((__u16)-1 == entry->llc_stripe_count)
2645 return lod->lod_ost_count;
2647 return lod_get_stripe_count(lod, lo,
2648 entry->llc_stripe_count, false);
2651 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2653 int magic, size = 0, i;
2654 struct lod_layout_component *comp_entries;
2656 bool is_composite, is_foreign = false;
2659 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2660 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2662 lo->ldo_def_striping->lds_def_striping_is_composite;
2664 comp_cnt = lo->ldo_comp_cnt;
2665 comp_entries = lo->ldo_comp_entries;
2666 is_composite = lo->ldo_is_composite;
2667 is_foreign = lo->ldo_is_foreign;
2671 return lo->ldo_foreign_lov_size;
2673 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2675 size = sizeof(struct lov_comp_md_v1) +
2676 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2677 LASSERT(size % sizeof(__u64) == 0);
2680 for (i = 0; i < comp_cnt; i++) {
2683 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2684 stripe_count = lod_comp_entry_stripe_count(lo, &comp_entries[i],
2686 if (!is_dir && is_composite)
2687 lod_comp_shrink_stripe_count(&comp_entries[i],
2690 size += lov_user_md_size(stripe_count, magic);
2691 LASSERT(size % sizeof(__u64) == 0);
2697 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2698 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2701 * \param[in] env execution environment
2702 * \param[in] dt dt_object to add components on
2703 * \param[in] buf buffer contains components to be added
2704 * \parem[in] th thandle
2706 * \retval 0 on success
2707 * \retval negative errno on failure
2709 static int lod_declare_layout_add(const struct lu_env *env,
2710 struct dt_object *dt,
2711 const struct lu_buf *buf,
2714 struct lod_thread_info *info = lod_env_info(env);
2715 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2716 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2717 struct dt_object *next = dt_object_child(dt);
2718 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
2719 struct lod_object *lo = lod_dt_obj(dt);
2720 struct lov_user_md_v3 *v3;
2721 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2723 int i, rc, array_cnt, old_array_cnt;
2726 LASSERT(lo->ldo_is_composite);
2728 if (lo->ldo_flr_state != LCM_FL_NONE)
2731 rc = lod_verify_striping(env, d, lo, buf, false);
2735 magic = comp_v1->lcm_magic;
2736 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2737 lustre_swab_lov_comp_md_v1(comp_v1);
2738 magic = comp_v1->lcm_magic;
2741 if (magic != LOV_USER_MAGIC_COMP_V1)
2744 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2745 OBD_ALLOC_PTR_ARRAY(comp_array, array_cnt);
2746 if (comp_array == NULL)
2749 memcpy(comp_array, lo->ldo_comp_entries,
2750 sizeof(*comp_array) * lo->ldo_comp_cnt);
2752 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2753 struct lov_user_md_v1 *v1;
2754 struct lu_extent *ext;
2756 v1 = (struct lov_user_md *)((char *)comp_v1 +
2757 comp_v1->lcm_entries[i].lcme_offset);
2758 ext = &comp_v1->lcm_entries[i].lcme_extent;
2760 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2761 lod_comp->llc_extent.e_start = ext->e_start;
2762 lod_comp->llc_extent.e_end = ext->e_end;
2763 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2764 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2766 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2767 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2768 lod_adjust_stripe_info(lod_comp, desc, 0);
2770 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2771 v3 = (struct lov_user_md_v3 *) v1;
2772 if (v3->lmm_pool_name[0] != '\0') {
2773 rc = lod_set_pool(&lod_comp->llc_pool,
2781 old_array = lo->ldo_comp_entries;
2782 old_array_cnt = lo->ldo_comp_cnt;
2784 lo->ldo_comp_entries = comp_array;
2785 lo->ldo_comp_cnt = array_cnt;
2787 /* No need to increase layout generation here, it will be increased
2788 * later when generating component ID for the new components */
2790 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2791 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2792 XATTR_NAME_LOV, 0, th);
2794 lo->ldo_comp_entries = old_array;
2795 lo->ldo_comp_cnt = old_array_cnt;
2799 OBD_FREE_PTR_ARRAY(old_array, old_array_cnt);
2801 LASSERT(lo->ldo_mirror_count == 1);
2802 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2807 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2808 lod_comp = &comp_array[i];
2809 if (lod_comp->llc_pool != NULL) {
2810 OBD_FREE(lod_comp->llc_pool,
2811 strlen(lod_comp->llc_pool) + 1);
2812 lod_comp->llc_pool = NULL;
2815 OBD_FREE_PTR_ARRAY(comp_array, array_cnt);
2820 * lod_last_non_stale_mirror() - Check if a mirror is the last non-stale mirror.
2821 * @mirror_id: Mirror id to be checked.
2824 * This function checks if a mirror with specified @mirror_id is the last
2825 * non-stale mirror of a LOD object @lo.
2827 * Return: true or false.
2830 bool lod_last_non_stale_mirror(__u16 mirror_id, struct lod_object *lo)
2832 struct lod_layout_component *lod_comp;
2833 bool has_stale_flag;
2836 for (i = 0; i < lo->ldo_mirror_count; i++) {
2837 if (lo->ldo_mirrors[i].lme_id == mirror_id ||
2838 lo->ldo_mirrors[i].lme_stale)
2841 has_stale_flag = false;
2842 lod_foreach_mirror_comp(lod_comp, lo, i) {
2843 if (lod_comp->llc_flags & LCME_FL_STALE) {
2844 has_stale_flag = true;
2848 if (!has_stale_flag)
2856 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2857 * the '$field' can only be 'flags' now. The xattr value is binary
2858 * lov_comp_md_v1 which contains the component ID(s) and the value of
2859 * the field to be modified.
2860 * Please update allowed_lustre_lov macro if $field groks more values
2863 * \param[in] env execution environment
2864 * \param[in] dt dt_object to be modified
2865 * \param[in] op operation string, like "set.flags"
2866 * \param[in] buf buffer contains components to be set
2867 * \parem[in] th thandle
2869 * \retval 0 on success
2870 * \retval negative errno on failure
2872 static int lod_declare_layout_set(const struct lu_env *env,
2873 struct dt_object *dt,
2874 char *op, const struct lu_buf *buf,
2877 struct lod_layout_component *lod_comp;
2878 struct lod_thread_info *info = lod_env_info(env);
2879 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2880 struct lod_object *lo = lod_dt_obj(dt);
2881 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2884 bool changed = false;
2887 /* Please update allowed_lustre_lov macro if op
2888 * groks more values in the future
2890 if (strcmp(op, "set.flags") != 0) {
2891 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2892 lod2obd(d)->obd_name, op);
2896 magic = comp_v1->lcm_magic;
2897 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2898 lustre_swab_lov_comp_md_v1(comp_v1);
2899 magic = comp_v1->lcm_magic;
2902 if (magic != LOV_USER_MAGIC_COMP_V1)
2905 if (comp_v1->lcm_entry_count == 0) {
2906 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2907 lod2obd(d)->obd_name);
2911 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2912 __u32 id = comp_v1->lcm_entries[i].lcme_id;
2913 __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
2914 __u32 mirror_flag = flags & LCME_MIRROR_FLAGS;
2915 __u16 mirror_id = mirror_id_of(id);
2916 bool neg = flags & LCME_FL_NEG;
2918 if (flags & LCME_FL_INIT) {
2920 lod_striping_free(env, lo);
2924 flags &= ~(LCME_MIRROR_FLAGS | LCME_FL_NEG);
2925 for (j = 0; j < lo->ldo_comp_cnt; j++) {
2926 lod_comp = &lo->ldo_comp_entries[j];
2928 /* lfs only put one flag in each entry */
2929 if ((flags && id != lod_comp->llc_id) ||
2930 (mirror_flag && mirror_id !=
2931 mirror_id_of(lod_comp->llc_id)))
2936 lod_comp->llc_flags &= ~flags;
2938 lod_comp->llc_flags &= ~mirror_flag;
2941 if ((flags & LCME_FL_STALE) &&
2942 lod_last_non_stale_mirror(mirror_id,
2945 lod_comp->llc_flags |= flags;
2948 lod_comp->llc_flags |= mirror_flag;
2949 if (mirror_flag & LCME_FL_NOSYNC)
2950 lod_comp->llc_timestamp =
2951 ktime_get_real_seconds();
2959 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
2960 lod2obd(d)->obd_name);
2964 lod_obj_inc_layout_gen(lo);
2966 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2967 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
2968 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
2973 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
2974 * and the xattr value is a unique component ID or a special lcme_id.
2976 * \param[in] env execution environment
2977 * \param[in] dt dt_object to be operated on
2978 * \param[in] buf buffer contains component ID or lcme_id
2979 * \parem[in] th thandle
2981 * \retval 0 on success
2982 * \retval negative errno on failure
2984 static int lod_declare_layout_del(const struct lu_env *env,
2985 struct dt_object *dt,
2986 const struct lu_buf *buf,
2989 struct lod_thread_info *info = lod_env_info(env);
2990 struct dt_object *next = dt_object_child(dt);
2991 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2992 struct lod_object *lo = lod_dt_obj(dt);
2993 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2994 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2995 __u32 magic, id, flags, neg_flags = 0;
2999 LASSERT(lo->ldo_is_composite);
3001 if (lo->ldo_flr_state != LCM_FL_NONE)
3004 magic = comp_v1->lcm_magic;
3005 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
3006 lustre_swab_lov_comp_md_v1(comp_v1);
3007 magic = comp_v1->lcm_magic;
3010 if (magic != LOV_USER_MAGIC_COMP_V1)
3013 id = comp_v1->lcm_entries[0].lcme_id;
3014 flags = comp_v1->lcm_entries[0].lcme_flags;
3016 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
3017 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
3018 lod2obd(d)->obd_name, id, flags);
3022 if (id != LCME_ID_INVAL && flags != 0) {
3023 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
3024 lod2obd(d)->obd_name);
3028 if (id == LCME_ID_INVAL && !flags) {
3029 CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
3030 lod2obd(d)->obd_name);
3034 if (flags & LCME_FL_NEG) {
3035 neg_flags = flags & ~LCME_FL_NEG;
3039 left = lo->ldo_comp_cnt;
3043 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3044 struct lod_layout_component *lod_comp;
3046 lod_comp = &lo->ldo_comp_entries[i];
3048 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
3050 else if (flags && !(flags & lod_comp->llc_flags))
3052 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
3055 if (left != (i + 1)) {
3056 CDEBUG(D_LAYOUT, "%s: this deletion will create "
3057 "a hole.\n", lod2obd(d)->obd_name);
3062 /* Mark the component as deleted */
3063 lod_comp->llc_id = LCME_ID_INVAL;
3065 /* Not instantiated component */
3066 if (lod_comp->llc_stripe == NULL)
3069 LASSERT(lod_comp->llc_stripe_count > 0);
3070 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
3071 struct dt_object *obj = lod_comp->llc_stripe[j];
3075 rc = lod_sub_declare_destroy(env, obj, th);
3081 LASSERTF(left >= 0, "left = %d\n", left);
3082 if (left == lo->ldo_comp_cnt) {
3083 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
3084 lod2obd(d)->obd_name, id);
3088 memset(attr, 0, sizeof(*attr));
3089 attr->la_valid = LA_SIZE;
3090 rc = lod_sub_declare_attr_set(env, next, attr, th);
3095 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3096 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
3097 XATTR_NAME_LOV, 0, th);
3099 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
3106 * Declare layout add/set/del operations issued by special xattr names:
3108 * XATTR_LUSTRE_LOV.add add component(s) to existing file
3109 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
3110 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
3112 * \param[in] env execution environment
3113 * \param[in] dt object
3114 * \param[in] name name of xattr
3115 * \param[in] buf lu_buf contains xattr value
3116 * \param[in] th transaction handle
3118 * \retval 0 on success
3119 * \retval negative if failed
3121 static int lod_declare_modify_layout(const struct lu_env *env,
3122 struct dt_object *dt,
3124 const struct lu_buf *buf,
3127 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3128 struct lod_object *lo = lod_dt_obj(dt);
3130 int rc, len = strlen(XATTR_LUSTRE_LOV);
3133 LASSERT(dt_object_exists(dt));
3135 if (strlen(name) <= len || name[len] != '.') {
3136 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
3137 lod2obd(d)->obd_name, name);
3142 rc = lod_striping_load(env, lo);
3146 /* the layout to be modified must be a composite layout */
3147 if (!lo->ldo_is_composite) {
3148 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
3149 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3150 GOTO(unlock, rc = -EINVAL);
3153 op = (char *)name + len;
3154 if (strcmp(op, "add") == 0) {
3155 rc = lod_declare_layout_add(env, dt, buf, th);
3156 } else if (strcmp(op, "del") == 0) {
3157 rc = lod_declare_layout_del(env, dt, buf, th);
3158 } else if (strncmp(op, "set", strlen("set")) == 0) {
3159 rc = lod_declare_layout_set(env, dt, op, buf, th);
3161 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
3162 lod2obd(d)->obd_name, name);
3163 GOTO(unlock, rc = -ENOTSUPP);
3167 lod_striping_free(env, lo);
3173 * Convert a plain file lov_mds_md to a composite layout.
3175 * \param[in,out] info the thread info::lti_ea_store buffer contains little
3176 * endian plain file layout
3178 * \retval 0 on success, <0 on failure
3180 static int lod_layout_convert(struct lod_thread_info *info)
3182 struct lov_mds_md *lmm = info->lti_ea_store;
3183 struct lov_mds_md *lmm_save;
3184 struct lov_comp_md_v1 *lcm;
3185 struct lov_comp_md_entry_v1 *lcme;
3191 /* realloc buffer to a composite layout which contains one component */
3192 blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3193 le32_to_cpu(lmm->lmm_magic));
3194 size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
3196 OBD_ALLOC_LARGE(lmm_save, blob_size);
3198 GOTO(out, rc = -ENOMEM);
3200 memcpy(lmm_save, lmm, blob_size);
3202 if (info->lti_ea_store_size < size) {
3203 rc = lod_ea_store_resize(info, size);
3208 lcm = info->lti_ea_store;
3209 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
3210 lcm->lcm_size = cpu_to_le32(size);
3211 lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
3212 lmm_save->lmm_layout_gen));
3213 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
3214 lcm->lcm_entry_count = cpu_to_le16(1);
3215 lcm->lcm_mirror_count = 0;
3217 lcme = &lcm->lcm_entries[0];
3218 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
3219 lcme->lcme_extent.e_start = 0;
3220 lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
3221 lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
3222 lcme->lcme_size = cpu_to_le32(blob_size);
3224 memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
3229 OBD_FREE_LARGE(lmm_save, blob_size);
3234 * Merge layouts to form a mirrored file.
3236 static int lod_declare_layout_merge(const struct lu_env *env,
3237 struct dt_object *dt, const struct lu_buf *mbuf,
3240 struct lod_thread_info *info = lod_env_info(env);
3241 struct lu_buf *buf = &info->lti_buf;
3242 struct lod_object *lo = lod_dt_obj(dt);
3243 struct lov_comp_md_v1 *lcm;
3244 struct lov_comp_md_v1 *cur_lcm;
3245 struct lov_comp_md_v1 *merge_lcm;
3246 struct lov_comp_md_entry_v1 *lcme;
3247 struct lov_mds_md_v1 *lmm;
3250 __u16 cur_entry_count;
3251 __u16 merge_entry_count;
3253 __u16 mirror_id = 0;
3260 merge_lcm = mbuf->lb_buf;
3261 if (mbuf->lb_len < sizeof(*merge_lcm))
3264 /* must be an existing layout from disk */
3265 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
3268 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
3270 /* do not allow to merge two mirrored files */
3271 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
3274 /* verify the target buffer */
3275 rc = lod_get_lov_ea(env, lo);
3277 RETURN(rc ? : -ENODATA);
3279 cur_lcm = info->lti_ea_store;
3280 switch (le32_to_cpu(cur_lcm->lcm_magic)) {
3283 rc = lod_layout_convert(info);
3285 case LOV_MAGIC_COMP_V1:
3295 /* info->lti_ea_store could be reallocated in lod_layout_convert() */
3296 cur_lcm = info->lti_ea_store;
3297 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
3299 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
3300 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
3301 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
3304 /* size of new layout */
3305 size = le32_to_cpu(cur_lcm->lcm_size) +
3306 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
3308 memset(buf, 0, sizeof(*buf));
3309 lu_buf_alloc(buf, size);
3310 if (buf->lb_buf == NULL)
3314 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
3316 offset = sizeof(*lcm) +
3317 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
3318 for (i = 0; i < cur_entry_count; i++) {
3319 struct lov_comp_md_entry_v1 *cur_lcme;
3321 lcme = &lcm->lcm_entries[i];
3322 cur_lcme = &cur_lcm->lcm_entries[i];
3324 lcme->lcme_offset = cpu_to_le32(offset);
3325 memcpy((char *)lcm + offset,
3326 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
3327 le32_to_cpu(lcme->lcme_size));
3329 offset += le32_to_cpu(lcme->lcme_size);
3331 if (mirror_count == 1 &&
3332 mirror_id_of(le32_to_cpu(lcme->lcme_id)) == 0) {
3333 /* Add mirror from a non-flr file, create new mirror ID.
3334 * Otherwise, keep existing mirror's component ID, used
3335 * for mirror extension.
3337 id = pflr_id(1, i + 1);
3338 lcme->lcme_id = cpu_to_le32(id);
3341 id = max(le32_to_cpu(lcme->lcme_id), id);
3344 mirror_id = mirror_id_of(id) + 1;
3346 /* check if first entry in new layout is DOM */
3347 lmm = (struct lov_mds_md_v1 *)((char *)merge_lcm +
3348 merge_lcm->lcm_entries[0].lcme_offset);
3349 merge_has_dom = lov_pattern(le32_to_cpu(lmm->lmm_pattern)) ==
3352 for (i = 0; i < merge_entry_count; i++) {
3353 struct lov_comp_md_entry_v1 *merge_lcme;
3355 merge_lcme = &merge_lcm->lcm_entries[i];
3356 lcme = &lcm->lcm_entries[cur_entry_count + i];
3358 *lcme = *merge_lcme;
3359 lcme->lcme_offset = cpu_to_le32(offset);
3360 if (merge_has_dom && i == 0)
3361 lcme->lcme_flags |= cpu_to_le32(LCME_FL_STALE);
3363 id = pflr_id(mirror_id, i + 1);
3364 lcme->lcme_id = cpu_to_le32(id);
3366 memcpy((char *)lcm + offset,
3367 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
3368 le32_to_cpu(lcme->lcme_size));
3370 offset += le32_to_cpu(lcme->lcme_size);
3373 /* fixup layout information */
3374 lod_obj_inc_layout_gen(lo);
3375 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3376 lcm->lcm_size = cpu_to_le32(size);
3377 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
3378 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
3379 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NONE)
3380 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
3382 rc = lod_striping_reload(env, lo, buf);
3386 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
3387 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3395 * Split layouts, just set the LOVEA with the layout from mbuf.
3397 static int lod_declare_layout_split(const struct lu_env *env,
3398 struct dt_object *dt, const struct lu_buf *mbuf,
3401 struct lod_object *lo = lod_dt_obj(dt);
3402 struct lov_comp_md_v1 *lcm = mbuf->lb_buf;
3406 lod_obj_inc_layout_gen(lo);
3407 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3409 rc = lod_striping_reload(env, lo, mbuf);
3413 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), mbuf,
3414 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3419 * Implementation of dt_object_operations::do_declare_xattr_set.
3421 * \see dt_object_operations::do_declare_xattr_set() in the API description
3424 * the extension to the API:
3425 * - declaring LOVEA requests striping creation
3426 * - LU_XATTR_REPLACE means layout swap
3428 static int lod_declare_xattr_set(const struct lu_env *env,
3429 struct dt_object *dt,
3430 const struct lu_buf *buf,
3431 const char *name, int fl,
3434 struct dt_object *next = dt_object_child(dt);
3435 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3440 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
3441 if ((S_ISREG(mode) || mode == 0) &&
3442 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE | LU_XATTR_SPLIT)) &&
3443 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3444 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
3446 * this is a request to create object's striping.
3448 * allow to declare predefined striping on a new (!mode) object
3449 * which is supposed to be replay of regular file creation
3450 * (when LOV setting is declared)
3452 * LU_XATTR_REPLACE is set to indicate a layout swap
3454 if (dt_object_exists(dt)) {
3455 rc = dt_attr_get(env, next, attr);
3459 memset(attr, 0, sizeof(*attr));
3460 attr->la_valid = LA_TYPE | LA_MODE;
3461 attr->la_mode = S_IFREG;
3463 rc = lod_declare_striped_create(env, dt, attr, buf, th);
3464 } else if (fl & LU_XATTR_MERGE) {
3465 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3466 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3467 rc = lod_declare_layout_merge(env, dt, buf, th);
3468 } else if (fl & LU_XATTR_SPLIT) {
3469 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3470 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3471 rc = lod_declare_layout_split(env, dt, buf, th);
3472 } else if (S_ISREG(mode) &&
3473 strlen(name) >= sizeof(XATTR_LUSTRE_LOV) + 3 &&
3474 allowed_lustre_lov(name)) {
3476 * this is a request to modify object's striping.
3477 * add/set/del component(s).
3479 if (!dt_object_exists(dt))
3482 rc = lod_declare_modify_layout(env, dt, name, buf, th);
3483 } else if (S_ISDIR(mode)) {
3484 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
3485 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3486 rc = lod_replace_parent_fid(env, dt, buf, th, true);
3488 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
3495 * Apply xattr changes to the object.
3497 * Applies xattr changes to the object and the stripes if the latter exist.
3499 * \param[in] env execution environment
3500 * \param[in] dt object
3501 * \param[in] buf buffer pointing to the new value of xattr
3502 * \param[in] name name of xattr
3503 * \param[in] fl flags
3504 * \param[in] th transaction handle
3506 * \retval 0 on success
3507 * \retval negative if failed
3509 static int lod_xattr_set_internal(const struct lu_env *env,
3510 struct dt_object *dt,
3511 const struct lu_buf *buf,
3512 const char *name, int fl,
3515 struct dt_object *next = dt_object_child(dt);
3516 struct lod_object *lo = lod_dt_obj(dt);
3521 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3522 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3525 /* Note: Do not set LinkEA on sub-stripes, otherwise
3526 * it will confuse the fid2path process(see mdt_path_current()).
3527 * The linkEA between master and sub-stripes is set in
3528 * lod_xattr_set_lmv(). */
3529 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
3532 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3533 if (!lo->ldo_stripe[i])
3536 if (!dt_object_exists(lo->ldo_stripe[i]))
3539 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
3549 * Delete an extended attribute.
3551 * Deletes specified xattr from the object and the stripes if the latter exist.
3553 * \param[in] env execution environment
3554 * \param[in] dt object
3555 * \param[in] name name of xattr
3556 * \param[in] th transaction handle
3558 * \retval 0 on success
3559 * \retval negative if failed
3561 static int lod_xattr_del_internal(const struct lu_env *env,
3562 struct dt_object *dt,
3563 const char *name, struct thandle *th)
3565 struct dt_object *next = dt_object_child(dt);
3566 struct lod_object *lo = lod_dt_obj(dt);
3571 rc = lod_sub_xattr_del(env, next, name, th);
3572 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3575 if (lo->ldo_dir_stripe_count == 0)
3578 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3579 LASSERT(lo->ldo_stripe[i]);
3581 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3590 * Set default striping on a directory.
3592 * Sets specified striping on a directory object unless it matches the default
3593 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3594 * EA. This striping will be used when regular file is being created in this
3597 * \param[in] env execution environment
3598 * \param[in] dt the striped object
3599 * \param[in] buf buffer with the striping
3600 * \param[in] name name of EA
3601 * \param[in] fl xattr flag (see OSD API description)
3602 * \param[in] th transaction handle
3604 * \retval 0 on success
3605 * \retval negative if failed
3607 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
3608 struct dt_object *dt,
3609 const struct lu_buf *buf,
3610 const char *name, int fl,
3613 struct lov_user_md_v1 *lum;
3614 struct lov_user_md_v3 *v3 = NULL;
3615 const char *pool_name = NULL;
3620 LASSERT(buf != NULL && buf->lb_buf != NULL);
3623 switch (lum->lmm_magic) {
3624 case LOV_USER_MAGIC_SPECIFIC:
3625 case LOV_USER_MAGIC_V3:
3627 if (v3->lmm_pool_name[0] != '\0')
3628 pool_name = v3->lmm_pool_name;
3630 case LOV_USER_MAGIC_V1:
3631 /* if { size, offset, count } = { 0, -1, 0 } and no pool
3632 * (i.e. all default values specified) then delete default
3633 * striping from dir. */
3635 "set default striping: sz %u # %u offset %d %s %s\n",
3636 (unsigned)lum->lmm_stripe_size,
3637 (unsigned)lum->lmm_stripe_count,
3638 (int)lum->lmm_stripe_offset,
3639 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
3641 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
3642 lum->lmm_stripe_count,
3643 lum->lmm_stripe_offset,
3646 case LOV_USER_MAGIC_COMP_V1:
3648 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lum;
3649 struct lov_comp_md_entry_v1 *lcme;
3652 comp_cnt = le16_to_cpu(lcm->lcm_entry_count);
3653 for (i = 0; i < comp_cnt; i++) {
3654 lcme = &lcm->lcm_entries[i];
3655 if (lcme->lcme_flags & cpu_to_le32(LCME_FL_EXTENSION)) {
3656 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
3665 CERROR("Invalid magic %x\n", lum->lmm_magic);
3670 rc = lod_xattr_del_internal(env, dt, name, th);
3674 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3681 * Set default striping on a directory object.
3683 * Sets specified striping on a directory object unless it matches the default
3684 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3685 * EA. This striping will be used when a new directory is being created in the
3688 * \param[in] env execution environment
3689 * \param[in] dt the striped object
3690 * \param[in] buf buffer with the striping
3691 * \param[in] name name of EA
3692 * \param[in] fl xattr flag (see OSD API description)
3693 * \param[in] th transaction handle
3695 * \retval 0 on success
3696 * \retval negative if failed
3698 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
3699 struct dt_object *dt,
3700 const struct lu_buf *buf,
3701 const char *name, int fl,
3704 struct lmv_user_md_v1 *lum;
3709 LASSERT(buf != NULL && buf->lb_buf != NULL);
3713 "set default stripe_count # %u stripe_offset %d hash %u\n",
3714 le32_to_cpu(lum->lum_stripe_count),
3715 (int)le32_to_cpu(lum->lum_stripe_offset),
3716 le32_to_cpu(lum->lum_hash_type));
3718 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
3719 le32_to_cpu(lum->lum_stripe_offset)) &&
3720 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
3721 rc = lod_xattr_del_internal(env, dt, name, th);
3725 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3734 * Turn directory into a striped directory.
3736 * During replay the client sends the striping created before MDT
3737 * failure, then the layer above LOD sends this defined striping
3738 * using ->do_xattr_set(), so LOD uses this method to replay creation
3739 * of the stripes. Notice the original information for the striping
3740 * (#stripes, FIDs, etc) was transferred in declare path.
3742 * \param[in] env execution environment
3743 * \param[in] dt the striped object
3744 * \param[in] buf not used currently
3745 * \param[in] name not used currently
3746 * \param[in] fl xattr flag (see OSD API description)
3747 * \param[in] th transaction handle
3749 * \retval 0 on success
3750 * \retval negative if failed
3752 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
3753 const struct lu_buf *buf, const char *name,
3754 int fl, struct thandle *th)
3756 struct lod_object *lo = lod_dt_obj(dt);
3757 struct lod_thread_info *info = lod_env_info(env);
3758 struct lu_attr *attr = &info->lti_attr;
3759 struct dt_object_format *dof = &info->lti_format;
3760 struct lu_buf lmv_buf;
3761 struct lu_buf slave_lmv_buf;
3762 struct lmv_mds_md_v1 *lmm;
3763 struct lmv_mds_md_v1 *slave_lmm = NULL;
3764 struct dt_insert_rec *rec = &info->lti_dt_rec;
3769 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3772 /* The stripes are supposed to be allocated in declare phase,
3773 * if there are no stripes being allocated, it will skip */
3774 if (lo->ldo_dir_stripe_count == 0) {
3775 if (lo->ldo_dir_is_foreign) {
3776 rc = lod_sub_xattr_set(env, dt_object_child(dt), buf,
3777 XATTR_NAME_LMV, fl, th);
3784 rc = dt_attr_get(env, dt_object_child(dt), attr);
3788 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME |
3789 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
3790 dof->dof_type = DFT_DIR;
3792 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
3795 lmm = lmv_buf.lb_buf;
3797 OBD_ALLOC_PTR(slave_lmm);
3798 if (slave_lmm == NULL)
3801 lod_prep_slave_lmv_md(slave_lmm, lmm);
3802 slave_lmv_buf.lb_buf = slave_lmm;
3803 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
3805 rec->rec_type = S_IFDIR;
3806 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3807 struct dt_object *dto = lo->ldo_stripe[i];
3808 char *stripe_name = info->lti_key;
3809 struct lu_name *sname;
3810 struct linkea_data ldata = { NULL };
3811 struct lu_buf linkea_buf;
3813 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
3817 /* fail a remote stripe creation */
3818 if (i && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_CREATE))
3821 /* don't create stripe if:
3822 * 1. it's source stripe of migrating directory
3823 * 2. it's existed stripe of splitting directory
3825 if ((lod_is_migrating(lo) && i >= lo->ldo_dir_migrate_offset) ||
3826 (lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) {
3827 if (!dt_object_exists(dto))
3828 GOTO(out, rc = -EINVAL);
3830 dt_write_lock(env, dto, DT_TGT_CHILD);
3831 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
3833 dt_write_unlock(env, dto);
3837 rc = lod_sub_ref_add(env, dto, th);
3838 dt_write_unlock(env, dto);
3842 rec->rec_fid = lu_object_fid(&dto->do_lu);
3843 rc = lod_sub_insert(env, dto,
3844 (const struct dt_rec *)rec,
3845 (const struct dt_key *)dot, th);
3850 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
3851 cfs_fail_val != i) {
3852 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
3854 slave_lmm->lmv_master_mdt_index =
3857 slave_lmm->lmv_master_mdt_index =
3860 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
3861 XATTR_NAME_LMV, 0, th);
3866 /* don't insert stripe if it's existed stripe of splitting
3867 * directory (this directory is striped).
3868 * NB, plain directory will insert itself as the first
3871 if (lod_is_splitting(lo) && lo->ldo_dir_split_offset > 1 &&
3872 lo->ldo_dir_split_offset > i)
3875 rec->rec_fid = lu_object_fid(&dt->do_lu);
3876 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
3877 (const struct dt_key *)dotdot, th);
3881 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
3883 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3884 PFID(lu_object_fid(&dto->do_lu)), i + 1);
3886 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3887 PFID(lu_object_fid(&dto->do_lu)), i);
3889 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
3890 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
3891 sname, lu_object_fid(&dt->do_lu));
3895 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
3896 linkea_buf.lb_len = ldata.ld_leh->leh_len;
3897 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
3898 XATTR_NAME_LINK, 0, th);
3902 rec->rec_fid = lu_object_fid(&dto->do_lu);
3903 rc = lod_sub_insert(env, dt_object_child(dt),
3904 (const struct dt_rec *)rec,
3905 (const struct dt_key *)stripe_name, th);
3909 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
3914 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
3915 rc = lod_sub_xattr_set(env, dt_object_child(dt),
3916 &lmv_buf, XATTR_NAME_LMV, fl, th);
3918 if (slave_lmm != NULL)
3919 OBD_FREE_PTR(slave_lmm);
3925 * Helper function to declare/execute creation of a striped directory
3927 * Called in declare/create object path, prepare striping for a directory
3928 * and prepare defaults data striping for the objects to be created in
3929 * that directory. Notice the function calls "declaration" or "execution"
3930 * methods depending on \a declare param. This is a consequence of the
3931 * current approach while we don't have natural distributed transactions:
3932 * we basically execute non-local updates in the declare phase. So, the
3933 * arguments for the both phases are the same and this is the reason for
3934 * this function to exist.
3936 * \param[in] env execution environment
3937 * \param[in] dt object
3938 * \param[in] attr attributes the stripes will be created with
3939 * \param[in] lmu lmv_user_md if MDT indices are specified
3940 * \param[in] dof format of stripes (see OSD API description)
3941 * \param[in] th transaction handle
3942 * \param[in] declare where to call "declare" or "execute" methods
3944 * \retval 0 on success
3945 * \retval negative if failed
3947 static int lod_dir_striping_create_internal(const struct lu_env *env,
3948 struct dt_object *dt,
3949 struct lu_attr *attr,
3950 const struct lu_buf *lmu,
3951 struct dt_object_format *dof,
3955 struct lod_thread_info *info = lod_env_info(env);
3956 struct lod_object *lo = lod_dt_obj(dt);
3957 const struct lod_default_striping *lds = lo->ldo_def_striping;
3961 LASSERT(ergo(lds != NULL,
3962 lds->lds_def_striping_set ||
3963 lds->lds_dir_def_striping_set));
3965 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
3966 lo->ldo_dir_stripe_offset)) {
3968 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3969 int stripe_count = lo->ldo_dir_stripe_count;
3971 if (info->lti_ea_store_size < sizeof(*v1)) {
3972 rc = lod_ea_store_resize(info, sizeof(*v1));
3975 v1 = info->lti_ea_store;
3978 memset(v1, 0, sizeof(*v1));
3979 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3980 v1->lum_stripe_count = cpu_to_le32(stripe_count);
3981 v1->lum_stripe_offset =
3982 cpu_to_le32(lo->ldo_dir_stripe_offset);
3984 info->lti_buf.lb_buf = v1;
3985 info->lti_buf.lb_len = sizeof(*v1);
3986 lmu = &info->lti_buf;
3990 rc = lod_declare_xattr_set_lmv(env, dt, attr, lmu, dof,
3993 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
3998 /* foreign LMV EA case */
4000 struct lmv_foreign_md *lfm = lmu->lb_buf;
4002 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
4003 rc = lod_declare_xattr_set_lmv(env, dt, attr,
4007 if (lo->ldo_dir_is_foreign) {
4008 LASSERT(lo->ldo_foreign_lmv != NULL &&
4009 lo->ldo_foreign_lmv_size > 0);
4010 info->lti_buf.lb_buf = lo->ldo_foreign_lmv;
4011 info->lti_buf.lb_len = lo->ldo_foreign_lmv_size;
4012 lmu = &info->lti_buf;
4013 rc = lod_xattr_set_lmv(env, dt, lmu,
4014 XATTR_NAME_LMV, 0, th);
4019 /* Transfer default LMV striping from the parent */
4020 if (lds != NULL && lds->lds_dir_def_striping_set &&
4021 !(LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
4022 lds->lds_dir_def_stripe_offset) &&
4023 le32_to_cpu(lds->lds_dir_def_hash_type) !=
4024 LMV_HASH_TYPE_UNKNOWN)) {
4025 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
4027 if (info->lti_ea_store_size < sizeof(*v1)) {
4028 rc = lod_ea_store_resize(info, sizeof(*v1));
4031 v1 = info->lti_ea_store;
4034 memset(v1, 0, sizeof(*v1));
4035 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
4036 v1->lum_stripe_count =
4037 cpu_to_le32(lds->lds_dir_def_stripe_count);
4038 v1->lum_stripe_offset =
4039 cpu_to_le32(lds->lds_dir_def_stripe_offset);
4041 cpu_to_le32(lds->lds_dir_def_hash_type);
4043 info->lti_buf.lb_buf = v1;
4044 info->lti_buf.lb_len = sizeof(*v1);
4046 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4047 XATTR_NAME_DEFAULT_LMV,
4050 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
4052 XATTR_NAME_DEFAULT_LMV, 0,
4058 /* Transfer default LOV striping from the parent */
4059 if (lds != NULL && lds->lds_def_striping_set &&
4060 lds->lds_def_comp_cnt != 0) {
4061 struct lov_mds_md *lmm;
4062 int lmm_size = lod_comp_md_size(lo, true);
4064 if (info->lti_ea_store_size < lmm_size) {
4065 rc = lod_ea_store_resize(info, lmm_size);
4069 lmm = info->lti_ea_store;
4071 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
4075 info->lti_buf.lb_buf = lmm;
4076 info->lti_buf.lb_len = lmm_size;
4079 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4080 XATTR_NAME_LOV, 0, th);
4082 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4083 XATTR_NAME_LOV, 0, th);
4091 static int lod_declare_dir_striping_create(const struct lu_env *env,
4092 struct dt_object *dt,
4093 struct lu_attr *attr,
4095 struct dt_object_format *dof,
4098 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
4102 static int lod_dir_striping_create(const struct lu_env *env,
4103 struct dt_object *dt,
4104 struct lu_attr *attr,
4105 struct dt_object_format *dof,
4108 return lod_dir_striping_create_internal(env, dt, attr, NULL, dof, th,
4113 * Make LOV EA for striped object.
4115 * Generate striping information and store it in the LOV EA of the given
4116 * object. The caller must ensure nobody else is calling the function
4117 * against the object concurrently. The transaction must be started.
4118 * FLDB service must be running as well; it's used to map FID to the target,
4119 * which is stored in LOV EA.
4121 * \param[in] env execution environment for this thread
4122 * \param[in] lo LOD object
4123 * \param[in] th transaction handle
4125 * \retval 0 if LOV EA is stored successfully
4126 * \retval negative error number on failure
4128 static int lod_generate_and_set_lovea(const struct lu_env *env,
4129 struct lod_object *lo,
4132 struct lod_thread_info *info = lod_env_info(env);
4133 struct dt_object *next = dt_object_child(&lo->ldo_obj);
4134 struct lov_mds_md_v1 *lmm;
4140 if (lo->ldo_comp_cnt == 0 && !lo->ldo_is_foreign) {
4141 lod_striping_free(env, lo);
4142 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
4146 lmm_size = lod_comp_md_size(lo, false);
4147 if (info->lti_ea_store_size < lmm_size) {
4148 rc = lod_ea_store_resize(info, lmm_size);
4152 lmm = info->lti_ea_store;
4154 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
4158 info->lti_buf.lb_buf = lmm;
4159 info->lti_buf.lb_len = lmm_size;
4160 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
4161 XATTR_NAME_LOV, 0, th);
4165 static __u32 lod_gen_component_id(struct lod_object *lo,
4166 int mirror_id, int comp_idx);
4169 * Repeat an existing component
4171 * Creates a new layout by replicating an existing component. Uses striping
4172 * policy from previous component as a template for the striping for the new
4175 * New component starts with zero length, will be extended (or removed) before
4176 * returning layout to client.
4178 * NB: Reallocates layout components array (lo->ldo_comp_entries), invalidating
4179 * any pre-existing pointers to components. Handle with care.
4181 * \param[in] env execution environment for this thread
4182 * \param[in,out] lo object to update the layout of
4183 * \param[in] index index of component to copy
4185 * \retval 0 on success
4186 * \retval negative errno on error
4188 static int lod_layout_repeat_comp(const struct lu_env *env,
4189 struct lod_object *lo, int index)
4191 struct lod_layout_component *lod_comp;
4192 struct lod_layout_component *new_comp = NULL;
4193 struct lod_layout_component *comp_array;
4194 int rc = 0, i, new_cnt = lo->ldo_comp_cnt + 1;
4199 lod_comp = &lo->ldo_comp_entries[index];
4200 LASSERT(lod_comp_inited(lod_comp) && lod_comp->llc_id != LCME_ID_INVAL);
4202 CDEBUG(D_LAYOUT, "repeating component %d\n", index);
4204 OBD_ALLOC_PTR_ARRAY(comp_array, new_cnt);
4205 if (comp_array == NULL)
4206 GOTO(out, rc = -ENOMEM);
4208 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4209 memcpy(&comp_array[i + offset], &lo->ldo_comp_entries[i],
4210 sizeof(*comp_array));
4212 /* Duplicate this component in to the next slot */
4214 new_comp = &comp_array[i + 1];
4215 memcpy(&comp_array[i + 1], &lo->ldo_comp_entries[i],
4216 sizeof(*comp_array));
4217 /* We must now skip this new component when copying */
4222 /* Set up copied component */
4223 new_comp->llc_flags &= ~LCME_FL_INIT;
4224 new_comp->llc_stripe = NULL;
4225 new_comp->llc_stripes_allocated = 0;
4226 new_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4227 /* for uninstantiated components, layout gen stores default stripe
4229 new_comp->llc_layout_gen = lod_comp->llc_stripe_offset;
4230 /* This makes the repeated component zero-length, placed at the end of
4231 * the preceding component */
4232 new_comp->llc_extent.e_start = new_comp->llc_extent.e_end;
4233 new_comp->llc_timestamp = lod_comp->llc_timestamp;
4234 new_comp->llc_pool = NULL;
4236 rc = lod_set_pool(&new_comp->llc_pool, lod_comp->llc_pool);
4240 if (new_comp->llc_ostlist.op_array) {
4241 __u32 *op_array = NULL;
4243 OBD_ALLOC(op_array, new_comp->llc_ostlist.op_size);
4245 GOTO(out, rc = -ENOMEM);
4246 memcpy(op_array, &new_comp->llc_ostlist.op_array,
4247 new_comp->llc_ostlist.op_size);
4248 new_comp->llc_ostlist.op_array = op_array;
4251 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
4252 lo->ldo_comp_entries = comp_array;
4253 lo->ldo_comp_cnt = new_cnt;
4255 /* Generate an id for the new component */
4256 mirror_id = mirror_id_of(new_comp->llc_id);
4257 new_comp->llc_id = LCME_ID_INVAL;
4258 new_comp->llc_id = lod_gen_component_id(lo, mirror_id, index + 1);
4259 if (new_comp->llc_id == LCME_ID_INVAL)
4260 GOTO(out, rc = -ERANGE);
4265 OBD_FREE_PTR_ARRAY(comp_array, new_cnt);
4270 static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
4274 /* clear memory region that will be used for layout change */
4275 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
4276 info->lti_count = 0;
4278 if (info->lti_comp_size >= comp_cnt)
4281 if (info->lti_comp_size > 0) {
4282 OBD_FREE_PTR_ARRAY(info->lti_comp_idx, info->lti_comp_size);
4283 info->lti_comp_size = 0;
4286 OBD_ALLOC_PTR_ARRAY(info->lti_comp_idx, comp_cnt);
4287 if (!info->lti_comp_idx)
4290 info->lti_comp_size = comp_cnt;
4295 * Prepare new layout minus deleted components
4297 * Removes components marked for deletion (LCME_ID_INVAL) by copying to a new
4298 * layout and skipping those components. Removes stripe objects if any exist.
4301 * Reallocates layout components array (lo->ldo_comp_entries), invalidating
4302 * any pre-existing pointers to components.
4304 * Caller is responsible for updating mirror end (ldo_mirror[].lme_end).
4306 * \param[in] env execution environment for this thread
4307 * \param[in,out] lo object to update the layout of
4308 * \param[in] th transaction handle for this operation
4310 * \retval # of components deleted
4311 * \retval negative errno on error
4313 static int lod_layout_del_prep_layout(const struct lu_env *env,
4314 struct lod_object *lo,
4317 struct lod_layout_component *lod_comp;
4318 struct lod_thread_info *info = lod_env_info(env);
4319 int rc = 0, i, j, deleted = 0;
4323 LASSERT(lo->ldo_is_composite);
4324 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
4326 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
4330 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4331 lod_comp = &lo->ldo_comp_entries[i];
4333 if (lod_comp->llc_id != LCME_ID_INVAL) {
4334 /* Build array of things to keep */
4335 info->lti_comp_idx[info->lti_count++] = i;
4339 lod_obj_set_pool(lo, i, NULL);
4340 if (lod_comp->llc_ostlist.op_array) {
4341 OBD_FREE(lod_comp->llc_ostlist.op_array,
4342 lod_comp->llc_ostlist.op_size);
4343 lod_comp->llc_ostlist.op_array = NULL;
4344 lod_comp->llc_ostlist.op_size = 0;
4348 CDEBUG(D_LAYOUT, "deleting comp %d, left %d\n", i,
4349 lo->ldo_comp_cnt - deleted);
4351 /* No striping info for this component */
4352 if (lod_comp->llc_stripe == NULL)
4355 LASSERT(lod_comp->llc_stripe_count > 0);
4356 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4357 struct dt_object *obj = lod_comp->llc_stripe[j];
4362 /* components which are not init have no sub objects
4364 if (lod_comp_inited(lod_comp)) {
4365 rc = lod_sub_destroy(env, obj, th);
4370 lu_object_put(env, &obj->do_lu);
4371 lod_comp->llc_stripe[j] = NULL;
4373 OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe,
4374 lod_comp->llc_stripes_allocated);
4375 lod_comp->llc_stripe = NULL;
4376 OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices,
4377 lod_comp->llc_stripes_allocated);
4378 lod_comp->llc_ost_indices = NULL;
4379 lod_comp->llc_stripes_allocated = 0;
4382 /* info->lti_count has the amount of left components */
4383 LASSERTF(info->lti_count >= 0 && info->lti_count < lo->ldo_comp_cnt,
4384 "left = %d, lo->ldo_comp_cnt %d\n", (int)info->lti_count,
4385 (int)lo->ldo_comp_cnt);
4387 if (info->lti_count > 0) {
4388 struct lod_layout_component *comp_array;
4390 OBD_ALLOC_PTR_ARRAY(comp_array, info->lti_count);
4391 if (comp_array == NULL)
4392 GOTO(out, rc = -ENOMEM);
4394 for (i = 0; i < info->lti_count; i++) {
4395 memcpy(&comp_array[i],
4396 &lo->ldo_comp_entries[info->lti_comp_idx[i]],
4397 sizeof(*comp_array));
4400 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
4401 lo->ldo_comp_entries = comp_array;
4402 lo->ldo_comp_cnt = info->lti_count;
4404 lod_free_comp_entries(lo);
4409 return rc ? rc : deleted;
4413 * Delete layout component(s)
4415 * This function sets up the layout data in the env and does the setattrs
4416 * required to write out the new layout. The layout itself is modified in
4417 * lod_layout_del_prep_layout.
4419 * \param[in] env execution environment for this thread
4420 * \param[in] dt object
4421 * \param[in] th transaction handle
4423 * \retval 0 on success
4424 * \retval negative error number on failure
4426 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
4429 struct lod_object *lo = lod_dt_obj(dt);
4430 struct dt_object *next = dt_object_child(dt);
4431 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4434 LASSERT(lo->ldo_mirror_count == 1);
4436 rc = lod_layout_del_prep_layout(env, lo, th);
4440 /* Only do this if we didn't delete all components */
4441 if (lo->ldo_comp_cnt > 0) {
4442 lo->ldo_mirrors[0].lme_end = lo->ldo_comp_cnt - 1;
4443 lod_obj_inc_layout_gen(lo);
4446 LASSERT(dt_object_exists(dt));
4447 rc = dt_attr_get(env, next, attr);
4451 if (attr->la_size > 0) {
4453 attr->la_valid = LA_SIZE;
4454 rc = lod_sub_attr_set(env, next, attr, th);
4459 rc = lod_generate_and_set_lovea(env, lo, th);
4463 lod_striping_free(env, lo);
4468 static int lod_get_default_lov_striping(const struct lu_env *env,
4469 struct lod_object *lo,
4470 struct lod_default_striping *lds,
4471 struct dt_allocation_hint *ah);
4473 * Implementation of dt_object_operations::do_xattr_set.
4475 * Sets specified extended attribute on the object. Three types of EAs are
4477 * LOV EA - stores striping for a regular file or default striping (when set
4479 * LMV EA - stores a marker for the striped directories
4480 * DMV EA - stores default directory striping
4482 * When striping is applied to a non-striped existing object (this is called
4483 * late striping), then LOD notices the caller wants to turn the object into a
4484 * striped one. The stripe objects are created and appropriate EA is set:
4485 * LOV EA storing all the stripes directly or LMV EA storing just a small header
4486 * with striping configuration.
4488 * \see dt_object_operations::do_xattr_set() in the API description for details.
4490 static int lod_xattr_set(const struct lu_env *env,
4491 struct dt_object *dt, const struct lu_buf *buf,
4492 const char *name, int fl, struct thandle *th)
4494 struct dt_object *next = dt_object_child(dt);
4499 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4500 !strcmp(name, XATTR_NAME_LMV)) {
4502 case LU_XATTR_CREATE:
4503 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
4506 case LU_XATTR_REPLACE:
4507 rc = lod_dir_layout_set(env, dt, buf, fl, th);
4514 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4515 strcmp(name, XATTR_NAME_LOV) == 0) {
4516 struct lod_default_striping *lds = lod_lds_buf_get(env);
4517 struct lov_user_md_v1 *v1 = buf->lb_buf;
4518 char pool[LOV_MAXPOOLNAME + 1];
4521 /* get existing striping config */
4522 rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds,
4527 memset(pool, 0, sizeof(pool));
4528 if (lds->lds_def_striping_set == 1)
4529 lod_layout_get_pool(lds->lds_def_comp_entries,
4530 lds->lds_def_comp_cnt, pool,
4533 is_del = LOVEA_DELETE_VALUES(v1->lmm_stripe_size,
4534 v1->lmm_stripe_count,
4535 v1->lmm_stripe_offset,
4538 /* Retain the pool name if it is not given */
4539 if (v1->lmm_magic == LOV_USER_MAGIC_V1 && pool[0] != '\0' &&
4541 struct lod_thread_info *info = lod_env_info(env);
4542 struct lov_user_md_v3 *v3 = info->lti_ea_store;
4544 memset(v3, 0, sizeof(*v3));
4545 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4546 v3->lmm_pattern = cpu_to_le32(v1->lmm_pattern);
4547 v3->lmm_stripe_count =
4548 cpu_to_le32(v1->lmm_stripe_count);
4549 v3->lmm_stripe_offset =
4550 cpu_to_le32(v1->lmm_stripe_offset);
4551 v3->lmm_stripe_size = cpu_to_le32(v1->lmm_stripe_size);
4553 strlcpy(v3->lmm_pool_name, pool,
4554 sizeof(v3->lmm_pool_name));
4556 info->lti_buf.lb_buf = v3;
4557 info->lti_buf.lb_len = sizeof(*v3);
4558 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4561 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name,
4565 if (lds->lds_def_striping_set == 1 &&
4566 lds->lds_def_comp_entries != NULL)
4567 lod_free_def_comp_entries(lds);
4570 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4571 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
4573 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
4576 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
4577 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4578 strcmp(name, XATTR_LUSTRE_LOV) == 0 ||
4579 allowed_lustre_lov(name))) {
4580 /* in case of lov EA swap, just set it
4581 * if not, it is a replay so check striping match what we
4582 * already have during req replay, declare_xattr_set()
4583 * defines striping, then create() does the work */
4584 if (fl & LU_XATTR_REPLACE) {
4585 /* free stripes, then update disk */
4586 lod_striping_free(env, lod_dt_obj(dt));
4588 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
4589 } else if (dt_object_remote(dt)) {
4590 /* This only happens during migration, see
4591 * mdd_migrate_create(), in which Master MDT will
4592 * create a remote target object, and only set
4593 * (migrating) stripe EA on the remote object,
4594 * and does not need creating each stripes. */
4595 rc = lod_sub_xattr_set(env, next, buf, name,
4597 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
4598 /* delete component(s) */
4599 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
4600 rc = lod_layout_del(env, dt, th);
4603 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
4604 * it's going to create create file with specified
4605 * component(s), the striping must have not being
4606 * cached in this case;
4608 * Otherwise, it's going to add/change component(s) to
4609 * an existing file, the striping must have been cached
4612 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
4613 !strcmp(name, XATTR_NAME_LOV),
4614 !lod_dt_obj(dt)->ldo_comp_cached));
4616 rc = lod_striped_create(env, dt, NULL, NULL, th);
4619 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4620 rc = lod_replace_parent_fid(env, dt, buf, th, false);
4625 /* then all other xattr */
4626 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4632 * Implementation of dt_object_operations::do_declare_xattr_del.
4634 * \see dt_object_operations::do_declare_xattr_del() in the API description
4637 static int lod_declare_xattr_del(const struct lu_env *env,
4638 struct dt_object *dt, const char *name,
4641 struct lod_object *lo = lod_dt_obj(dt);
4642 struct dt_object *next = dt_object_child(dt);
4647 rc = lod_sub_declare_xattr_del(env, next, name, th);
4651 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4654 /* NB: don't delete stripe LMV, because when we do this, normally we
4655 * will remove stripes, besides, if directory LMV is corrupt, this will
4656 * prevent deleting its LMV and fixing it (via LFSCK).
4658 if (!strcmp(name, XATTR_NAME_LMV))
4661 rc = lod_striping_load(env, lo);
4665 if (lo->ldo_dir_stripe_count == 0)
4668 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4669 struct dt_object *dto = lo->ldo_stripe[i];
4674 rc = lod_sub_declare_xattr_del(env, dto, name, th);
4683 * Implementation of dt_object_operations::do_xattr_del.
4685 * If EA storing a regular striping is being deleted, then release
4686 * all the references to the stripe objects in core.
4688 * \see dt_object_operations::do_xattr_del() in the API description for details.
4690 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
4691 const char *name, struct thandle *th)
4693 struct dt_object *next = dt_object_child(dt);
4694 struct lod_object *lo = lod_dt_obj(dt);
4699 if (!strcmp(name, XATTR_NAME_LOV) || !strcmp(name, XATTR_NAME_LMV))
4700 lod_striping_free(env, lod_dt_obj(dt));
4702 rc = lod_sub_xattr_del(env, next, name, th);
4703 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
4706 if (!strcmp(name, XATTR_NAME_LMV))
4709 if (lo->ldo_dir_stripe_count == 0)
4712 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4713 struct dt_object *dto = lo->ldo_stripe[i];
4718 rc = lod_sub_xattr_del(env, dto, name, th);
4727 * Implementation of dt_object_operations::do_xattr_list.
4729 * \see dt_object_operations::do_xattr_list() in the API description
4732 static int lod_xattr_list(const struct lu_env *env,
4733 struct dt_object *dt, const struct lu_buf *buf)
4735 return dt_xattr_list(env, dt_object_child(dt), buf);
4738 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
4740 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
4744 * Copy OST list from layout provided by user.
4746 * \param[in] lod_comp layout_component to be filled
4747 * \param[in] v3 LOV EA V3 user data
4749 * \retval 0 on success
4750 * \retval negative if failed
4752 int lod_comp_copy_ost_lists(struct lod_layout_component *lod_comp,
4753 struct lov_user_md_v3 *v3)
4759 if (v3->lmm_stripe_offset == LOV_OFFSET_DEFAULT)
4760 v3->lmm_stripe_offset = v3->lmm_objects[0].l_ost_idx;
4762 if (lod_comp->llc_ostlist.op_array) {
4763 if (lod_comp->llc_ostlist.op_size >=
4764 v3->lmm_stripe_count * sizeof(__u32)) {
4765 lod_comp->llc_ostlist.op_count =
4766 v3->lmm_stripe_count;
4769 OBD_FREE(lod_comp->llc_ostlist.op_array,
4770 lod_comp->llc_ostlist.op_size);
4773 /* copy ost list from lmm */
4774 lod_comp->llc_ostlist.op_count = v3->lmm_stripe_count;
4775 lod_comp->llc_ostlist.op_size = v3->lmm_stripe_count * sizeof(__u32);
4776 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
4777 lod_comp->llc_ostlist.op_size);
4778 if (!lod_comp->llc_ostlist.op_array)
4781 for (j = 0; j < v3->lmm_stripe_count; j++) {
4782 lod_comp->llc_ostlist.op_array[j] =
4783 v3->lmm_objects[j].l_ost_idx;
4791 * Get default striping.
4793 * \param[in] env execution environment
4794 * \param[in] lo object
4795 * \param[out] lds default striping
4797 * \retval 0 on success
4798 * \retval negative if failed
4800 static int lod_get_default_lov_striping(const struct lu_env *env,
4801 struct lod_object *lo,
4802 struct lod_default_striping *lds,
4803 struct dt_allocation_hint *ah)
4805 struct lod_thread_info *info = lod_env_info(env);
4806 struct lov_user_md_v1 *v1 = NULL;
4807 struct lov_user_md_v3 *v3 = NULL;
4808 struct lov_comp_md_v1 *comp_v1 = NULL;
4816 rc = lod_get_lov_ea(env, lo);
4820 if (rc < (typeof(rc))sizeof(struct lov_user_md))
4823 v1 = info->lti_ea_store;
4824 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
4825 lustre_swab_lov_user_md_v1(v1);
4826 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
4827 v3 = (struct lov_user_md_v3 *)v1;
4828 lustre_swab_lov_user_md_v3(v3);
4829 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_SPECIFIC)) {
4830 v3 = (struct lov_user_md_v3 *)v1;
4831 lustre_swab_lov_user_md_v3(v3);
4832 lustre_swab_lov_user_md_objects(v3->lmm_objects,
4833 v3->lmm_stripe_count);
4834 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_COMP_V1) ||
4835 v1->lmm_magic == __swab32(LOV_USER_MAGIC_SEL)) {
4836 comp_v1 = (struct lov_comp_md_v1 *)v1;
4837 lustre_swab_lov_comp_md_v1(comp_v1);
4840 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1 &&
4841 v1->lmm_magic != LOV_MAGIC_COMP_V1 &&
4842 v1->lmm_magic != LOV_MAGIC_SEL &&
4843 v1->lmm_magic != LOV_USER_MAGIC_SPECIFIC)
4846 if ((v1->lmm_magic == LOV_MAGIC_COMP_V1 ||
4847 v1->lmm_magic == LOV_MAGIC_SEL) &&
4848 !(ah && ah->dah_append_stripes)) {
4849 comp_v1 = (struct lov_comp_md_v1 *)v1;
4850 comp_cnt = comp_v1->lcm_entry_count;
4853 mirror_cnt = comp_v1->lcm_mirror_count + 1;
4861 /* realloc default comp entries if necessary */
4862 rc = lod_def_striping_comp_resize(lds, comp_cnt);
4866 lds->lds_def_comp_cnt = comp_cnt;
4867 lds->lds_def_striping_is_composite = composite;
4868 lds->lds_def_mirror_cnt = mirror_cnt;
4870 for (i = 0; i < comp_cnt; i++) {
4871 struct lod_layout_component *lod_comp;
4874 lod_comp = &lds->lds_def_comp_entries[i];
4876 * reset lod_comp values, llc_stripes is always NULL in
4877 * the default striping template, llc_pool will be reset
4880 memset(lod_comp, 0, offsetof(typeof(*lod_comp), llc_pool));
4883 v1 = (struct lov_user_md *)((char *)comp_v1 +
4884 comp_v1->lcm_entries[i].lcme_offset);
4885 lod_comp->llc_extent =
4886 comp_v1->lcm_entries[i].lcme_extent;
4887 /* We only inherit certain flags from the layout */
4888 lod_comp->llc_flags =
4889 comp_v1->lcm_entries[i].lcme_flags &
4890 LCME_TEMPLATE_FLAGS;
4893 if (!lov_pattern_supported(v1->lmm_pattern) &&
4894 !(v1->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
4895 lod_free_def_comp_entries(lds);
4899 CDEBUG(D_LAYOUT, DFID" stripe_count=%d stripe_size=%d stripe_offset=%d append_stripes=%d\n",
4900 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
4901 (int)v1->lmm_stripe_count, (int)v1->lmm_stripe_size,
4902 (int)v1->lmm_stripe_offset,
4903 ah ? ah->dah_append_stripes : 0);
4905 if (ah && ah->dah_append_stripes)
4906 lod_comp->llc_stripe_count = ah->dah_append_stripes;
4908 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
4909 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
4910 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
4911 lod_comp->llc_pattern = v1->lmm_pattern;
4914 if (ah && ah->dah_append_pool && ah->dah_append_pool[0]) {
4915 pool = ah->dah_append_pool;
4916 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
4917 /* XXX: sanity check here */
4918 v3 = (struct lov_user_md_v3 *) v1;
4919 if (v3->lmm_pool_name[0] != '\0')
4920 pool = v3->lmm_pool_name;
4922 lod_set_def_pool(lds, i, pool);
4923 if (v1->lmm_magic == LOV_USER_MAGIC_SPECIFIC) {
4924 v3 = (struct lov_user_md_v3 *)v1;
4925 rc = lod_comp_copy_ost_lists(lod_comp, v3);
4928 } else if (lod_comp->llc_ostlist.op_array &&
4929 lod_comp->llc_ostlist.op_count) {
4930 for (j = 0; j < lod_comp->llc_ostlist.op_count; j++)
4931 lod_comp->llc_ostlist.op_array[j] = -1;
4932 lod_comp->llc_ostlist.op_count = 0;
4936 lds->lds_def_striping_set = 1;
4941 * Get default directory striping.
4943 * \param[in] env execution environment
4944 * \param[in] lo object
4945 * \param[out] lds default striping
4947 * \retval 0 on success
4948 * \retval negative if failed
4950 static int lod_get_default_lmv_striping(const struct lu_env *env,
4951 struct lod_object *lo,
4952 struct lod_default_striping *lds)
4954 struct lmv_user_md *lmu;
4957 lds->lds_dir_def_striping_set = 0;
4959 rc = lod_get_default_lmv_ea(env, lo);
4963 if (rc >= (int)sizeof(*lmu)) {
4964 struct lod_thread_info *info = lod_env_info(env);
4966 lmu = info->lti_ea_store;
4968 lds->lds_dir_def_stripe_count =
4969 le32_to_cpu(lmu->lum_stripe_count);
4970 lds->lds_dir_def_stripe_offset =
4971 le32_to_cpu(lmu->lum_stripe_offset);
4972 lds->lds_dir_def_hash_type =
4973 le32_to_cpu(lmu->lum_hash_type);
4974 lds->lds_dir_def_striping_set = 1;
4981 * Get default striping in the object.
4983 * Get object default striping and default directory striping.
4985 * \param[in] env execution environment
4986 * \param[in] lo object
4987 * \param[out] lds default striping
4989 * \retval 0 on success
4990 * \retval negative if failed
4992 static int lod_get_default_striping(const struct lu_env *env,
4993 struct lod_object *lo,
4994 struct lod_default_striping *lds)
4998 rc = lod_get_default_lov_striping(env, lo, lds, NULL);
4999 rc1 = lod_get_default_lmv_striping(env, lo, lds);
5000 if (rc == 0 && rc1 < 0)
5007 * Apply default striping on object.
5009 * If object striping pattern is not set, set to the one in default striping.
5010 * The default striping is from parent or fs.
5012 * \param[in] lo new object
5013 * \param[in] lds default striping
5014 * \param[in] mode new object's mode
5016 static void lod_striping_from_default(struct lod_object *lo,
5017 const struct lod_default_striping *lds,
5020 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5023 if (lds->lds_def_striping_set && S_ISREG(mode)) {
5024 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
5026 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
5027 lds->lds_def_comp_cnt);
5031 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
5032 if (lds->lds_def_mirror_cnt > 1)
5033 lo->ldo_flr_state = LCM_FL_RDONLY;
5035 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5036 struct lod_layout_component *obj_comp =
5037 &lo->ldo_comp_entries[i];
5038 struct lod_layout_component *def_comp =
5039 &lds->lds_def_comp_entries[i];
5041 CDEBUG(D_LAYOUT, "Inherit from default: flags=%#x "
5042 "size=%hu nr=%u offset=%u pattern=%#x pool=%s\n",
5043 def_comp->llc_flags,
5044 def_comp->llc_stripe_size,
5045 def_comp->llc_stripe_count,
5046 def_comp->llc_stripe_offset,
5047 def_comp->llc_pattern,
5048 def_comp->llc_pool ?: "");
5050 *obj_comp = *def_comp;
5051 if (def_comp->llc_pool != NULL) {
5052 /* pointer was copied from def_comp */
5053 obj_comp->llc_pool = NULL;
5054 lod_obj_set_pool(lo, i, def_comp->llc_pool);
5058 if (def_comp->llc_ostlist.op_array &&
5059 def_comp->llc_ostlist.op_count) {
5060 OBD_ALLOC(obj_comp->llc_ostlist.op_array,
5061 obj_comp->llc_ostlist.op_size);
5062 if (!obj_comp->llc_ostlist.op_array)
5064 memcpy(obj_comp->llc_ostlist.op_array,
5065 def_comp->llc_ostlist.op_array,
5066 obj_comp->llc_ostlist.op_size);
5067 } else if (def_comp->llc_ostlist.op_array) {
5068 obj_comp->llc_ostlist.op_array = NULL;
5072 * Don't initialize these fields for plain layout
5073 * (v1/v3) here, they are inherited in the order of
5074 * 'parent' -> 'fs default (root)' -> 'global default
5075 * values for stripe_count & stripe_size'.
5077 * see lod_ah_init().
5079 if (!lo->ldo_is_composite)
5082 lod_adjust_stripe_info(obj_comp, desc, 0);
5084 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
5085 if (lo->ldo_dir_stripe_count == 0)
5086 lo->ldo_dir_stripe_count =
5087 lds->lds_dir_def_stripe_count;
5088 if (lo->ldo_dir_stripe_offset == -1)
5089 lo->ldo_dir_stripe_offset =
5090 lds->lds_dir_def_stripe_offset;
5091 if (lo->ldo_dir_hash_type == 0)
5092 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
5094 CDEBUG(D_LAYOUT, "striping from default dir: count:%hu, "
5095 "offset:%u, hash_type:%u\n",
5096 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
5097 lo->ldo_dir_hash_type);
5101 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root,
5104 struct lod_layout_component *lod_comp;
5106 if (lo->ldo_comp_cnt == 0)
5109 if (lo->ldo_is_composite)
5112 lod_comp = &lo->ldo_comp_entries[0];
5114 if (lod_comp->llc_stripe_count <= 0 ||
5115 lod_comp->llc_stripe_size <= 0)
5118 if (from_root && (lod_comp->llc_pool == NULL ||
5119 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
5122 if (append_pool && append_pool[0])
5129 * Implementation of dt_object_operations::do_ah_init.
5131 * This method is used to make a decision on the striping configuration for the
5132 * object being created. It can be taken from the \a parent object if it exists,
5133 * or filesystem's default. The resulting configuration (number of stripes,
5134 * stripe size/offset, pool name, etc) is stored in the object itself and will
5135 * be used by the methods like ->doo_declare_create().
5137 * \see dt_object_operations::do_ah_init() in the API description for details.
5139 static void lod_ah_init(const struct lu_env *env,
5140 struct dt_allocation_hint *ah,
5141 struct dt_object *parent,
5142 struct dt_object *child,
5145 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
5146 struct lod_thread_info *info = lod_env_info(env);
5147 struct lod_default_striping *lds = lod_lds_buf_get(env);
5148 struct dt_object *nextp = NULL;
5149 struct dt_object *nextc;
5150 struct lod_object *lp = NULL;
5151 struct lod_object *lc;
5152 struct lov_desc *desc;
5153 struct lod_layout_component *lod_comp;
5159 if (ah->dah_append_stripes == -1)
5160 ah->dah_append_stripes =
5161 d->lod_ost_descs.ltd_lov_desc.ld_tgt_count;
5163 if (likely(parent)) {
5164 nextp = dt_object_child(parent);
5165 lp = lod_dt_obj(parent);
5168 nextc = dt_object_child(child);
5169 lc = lod_dt_obj(child);
5171 LASSERT(!lod_obj_is_striped(child));
5172 /* default layout template may have been set on the regular file
5173 * when this is called from mdd_create_data() */
5174 if (S_ISREG(child_mode))
5175 lod_free_comp_entries(lc);
5177 if (!dt_object_exists(nextc))
5178 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
5180 if (S_ISDIR(child_mode)) {
5181 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
5183 /* other default values are 0 */
5184 lc->ldo_dir_stripe_offset = -1;
5186 /* no default striping configuration is needed for
5189 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5190 le32_to_cpu(lum1->lum_magic) == LMV_MAGIC_FOREIGN) {
5191 lc->ldo_dir_is_foreign = true;
5192 /* keep stripe_count 0 and stripe_offset -1 */
5193 CDEBUG(D_INFO, "no default striping for foreign dir\n");
5198 * If parent object is not root directory,
5199 * then get default striping from parent object.
5201 if (likely(lp != NULL)) {
5202 lod_get_default_striping(env, lp, lds);
5204 /* inherit default striping except ROOT */
5205 if ((lds->lds_def_striping_set ||
5206 lds->lds_dir_def_striping_set) &&
5207 !fid_is_root(lod_object_fid(lp)))
5208 lc->ldo_def_striping = lds;
5211 /* It should always honour the specified stripes */
5212 /* Note: old client (< 2.7)might also do lfs mkdir, whose EA
5213 * will have old magic. In this case, we should ignore the
5214 * stripe count and try to create dir by default stripe.
5216 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5217 (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
5218 le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC)) {
5219 lc->ldo_dir_stripe_count =
5220 le32_to_cpu(lum1->lum_stripe_count);
5221 lc->ldo_dir_stripe_offset =
5222 le32_to_cpu(lum1->lum_stripe_offset);
5223 lc->ldo_dir_hash_type =
5224 le32_to_cpu(lum1->lum_hash_type);
5226 "set dirstripe: count %hu, offset %d, hash %u\n",
5227 lc->ldo_dir_stripe_count,
5228 (int)lc->ldo_dir_stripe_offset,
5229 lc->ldo_dir_hash_type);
5231 /* transfer defaults LMV to new directory */
5232 lod_striping_from_default(lc, lds, child_mode);
5234 /* set count 0 to create normal directory */
5235 if (lc->ldo_dir_stripe_count == 1)
5236 lc->ldo_dir_stripe_count = 0;
5239 /* shrink the stripe_count to the avaible MDT count */
5240 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
5241 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)) {
5242 lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
5243 if (lc->ldo_dir_stripe_count == 1)
5244 lc->ldo_dir_stripe_count = 0;
5247 if (!(lc->ldo_dir_hash_type & LMV_HASH_TYPE_MASK))
5248 lc->ldo_dir_hash_type |=
5249 d->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
5251 CDEBUG(D_INFO, "final dir stripe [%hu %d %u]\n",
5252 lc->ldo_dir_stripe_count,
5253 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
5258 /* child object regular file*/
5260 if (!lod_object_will_be_striped(S_ISREG(child_mode),
5261 lu_object_fid(&child->do_lu)))
5264 /* If object is going to be striped over OSTs, transfer default
5265 * striping information to the child, so that we can use it
5266 * during declaration and creation.
5268 * Try from the parent first.
5270 if (likely(lp != NULL)) {
5271 rc = lod_get_default_lov_striping(env, lp, lds, ah);
5273 lod_striping_from_default(lc, lds, child_mode);
5276 /* Initialize lod_device::lod_md_root object reference */
5277 if (d->lod_md_root == NULL) {
5278 struct dt_object *root;
5279 struct lod_object *lroot;
5281 lu_root_fid(&info->lti_fid);
5282 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
5283 if (!IS_ERR(root)) {
5284 lroot = lod_dt_obj(root);
5286 spin_lock(&d->lod_lock);
5287 if (d->lod_md_root != NULL)
5288 dt_object_put(env, &d->lod_md_root->ldo_obj);
5289 d->lod_md_root = lroot;
5290 spin_unlock(&d->lod_lock);
5294 /* try inherit layout from the root object (fs default) when:
5295 * - parent does not have default layout; or
5296 * - parent has plain(v1/v3) default layout, and some attributes
5297 * are not specified in the default layout;
5299 if (d->lod_md_root != NULL &&
5300 lod_need_inherit_more(lc, true, ah->dah_append_pool)) {
5301 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds,
5305 if (lc->ldo_comp_cnt == 0) {
5306 lod_striping_from_default(lc, lds, child_mode);
5307 } else if (!lds->lds_def_striping_is_composite) {
5308 struct lod_layout_component *def_comp;
5310 LASSERT(!lc->ldo_is_composite);
5311 lod_comp = &lc->ldo_comp_entries[0];
5312 def_comp = &lds->lds_def_comp_entries[0];
5314 if (lod_comp->llc_stripe_count <= 0)
5315 lod_comp->llc_stripe_count =
5316 def_comp->llc_stripe_count;
5317 if (lod_comp->llc_stripe_size <= 0)
5318 lod_comp->llc_stripe_size =
5319 def_comp->llc_stripe_size;
5320 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT &&
5321 (!lod_comp->llc_pool || !lod_comp->llc_pool[0]))
5322 lod_comp->llc_stripe_offset =
5323 def_comp->llc_stripe_offset;
5324 if (lod_comp->llc_pool == NULL)
5325 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
5330 * fs default striping may not be explicitly set, or historically set
5331 * in config log, use them.
5333 if (lod_need_inherit_more(lc, false, ah->dah_append_pool)) {
5334 if (lc->ldo_comp_cnt == 0) {
5335 rc = lod_alloc_comp_entries(lc, 0, 1);
5337 /* fail to allocate memory, will create a
5338 * non-striped file. */
5340 lc->ldo_is_composite = 0;
5341 lod_comp = &lc->ldo_comp_entries[0];
5342 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
5344 LASSERT(!lc->ldo_is_composite);
5345 lod_comp = &lc->ldo_comp_entries[0];
5346 desc = &d->lod_ost_descs.ltd_lov_desc;
5347 lod_adjust_stripe_info(lod_comp, desc, ah->dah_append_stripes);
5348 if (ah->dah_append_pool && ah->dah_append_pool[0])
5349 lod_obj_set_pool(lc, 0, ah->dah_append_pool);
5356 * Size initialization on late striping.
5358 * Propagate the size of a truncated object to a deferred striping.
5359 * This function handles a special case when truncate was done on a
5360 * non-striped object and now while the striping is being created
5361 * we can't lose that size, so we have to propagate it to the stripes
5364 * \param[in] env execution environment
5365 * \param[in] dt object
5366 * \param[in] th transaction handle
5368 * \retval 0 on success
5369 * \retval negative if failed
5371 static int lod_declare_init_size(const struct lu_env *env,
5372 struct dt_object *dt, struct thandle *th)
5374 struct dt_object *next = dt_object_child(dt);
5375 struct lod_object *lo = lod_dt_obj(dt);
5376 struct dt_object **objects = NULL;
5377 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
5378 uint64_t size, offs;
5379 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
5380 struct lu_extent size_ext;
5383 if (!lod_obj_is_striped(dt))
5386 rc = dt_attr_get(env, next, attr);
5387 LASSERT(attr->la_valid & LA_SIZE);
5391 size = attr->la_size;
5395 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
5396 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5397 struct lod_layout_component *lod_comp;
5398 struct lu_extent *extent;
5400 lod_comp = &lo->ldo_comp_entries[i];
5402 if (lod_comp->llc_stripe == NULL)
5405 extent = &lod_comp->llc_extent;
5406 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
5407 if (!lo->ldo_is_composite ||
5408 lu_extent_is_overlapped(extent, &size_ext)) {
5409 objects = lod_comp->llc_stripe;
5410 stripe_count = lod_comp->llc_stripe_count;
5411 stripe_size = lod_comp->llc_stripe_size;
5414 if (stripe_count == 0)
5417 LASSERT(objects != NULL && stripe_size != 0);
5418 do_div(size, stripe_size);
5419 stripe = do_div(size, stripe_count);
5420 LASSERT(objects[stripe] != NULL);
5422 size = size * stripe_size;
5423 offs = attr->la_size;
5424 size += do_div(offs, stripe_size);
5426 attr->la_valid = LA_SIZE;
5427 attr->la_size = size;
5429 rc = lod_sub_declare_attr_set(env, objects[stripe],
5438 * Declare creation of striped object.
5440 * The function declares creation stripes for a regular object. The function
5441 * also declares whether the stripes will be created with non-zero size if
5442 * previously size was set non-zero on the master object. If object \a dt is
5443 * not local, then only fully defined striping can be applied in \a lovea.
5444 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
5447 * \param[in] env execution environment
5448 * \param[in] dt object
5449 * \param[in] attr attributes the stripes will be created with
5450 * \param[in] lovea a buffer containing striping description
5451 * \param[in] th transaction handle
5453 * \retval 0 on success
5454 * \retval negative if failed
5456 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
5457 struct lu_attr *attr,
5458 const struct lu_buf *lovea, struct thandle *th)
5460 struct lod_thread_info *info = lod_env_info(env);
5461 struct dt_object *next = dt_object_child(dt);
5462 struct lod_object *lo = lod_dt_obj(dt);
5466 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
5467 GOTO(out, rc = -ENOMEM);
5469 if (!dt_object_remote(next)) {
5470 /* choose OST and generate appropriate objects */
5471 rc = lod_prepare_create(env, lo, attr, lovea, th);
5476 * declare storage for striping data
5478 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
5480 /* LOD can not choose OST objects for remote objects, i.e.
5481 * stripes must be ready before that. Right now, it can only
5482 * happen during migrate, i.e. migrate process needs to create
5483 * remote regular file (mdd_migrate_create), then the migrate
5484 * process will provide stripeEA. */
5485 LASSERT(lovea != NULL);
5486 info->lti_buf = *lovea;
5489 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
5490 XATTR_NAME_LOV, 0, th);
5495 * if striping is created with local object's size > 0,
5496 * we have to propagate this size to specific object
5497 * the case is possible only when local object was created previously
5499 if (dt_object_exists(next))
5500 rc = lod_declare_init_size(env, dt, th);
5503 /* failed to create striping or to set initial size, let's reset
5504 * config so that others don't get confused */
5506 lod_striping_free(env, lo);
5512 * Whether subdirectories under \a dt should be created on MDTs by space QoS
5514 * If LMV_HASH_FLAG_SPACE is set on directory default layout, its subdirectories
5515 * should be created on MDT by space QoS.
5517 * \param[in] env execution environment
5518 * \param[in] dev lu device
5519 * \param[in] dt object
5521 * \retval 1 if directory should create subdir by space usage
5523 * \retval -ev if failed
5525 static inline int dt_object_qos_mkdir(const struct lu_env *env,
5526 struct lu_device *dev,
5527 struct dt_object *dt)
5529 struct lod_thread_info *info = lod_env_info(env);
5530 struct lu_object *obj;
5531 struct lod_object *lo;
5532 struct lmv_user_md *lmu;
5535 obj = lu_object_find_slice(env, dev, lu_object_fid(&dt->do_lu), NULL);
5537 return PTR_ERR(obj);
5539 lo = lu2lod_obj(obj);
5541 rc = lod_get_default_lmv_ea(env, lo);
5542 dt_object_put(env, dt);
5546 if (rc < (int)sizeof(*lmu))
5549 lmu = info->lti_ea_store;
5550 return le32_to_cpu(lmu->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
5554 * Implementation of dt_object_operations::do_declare_create.
5556 * The method declares creation of a new object. If the object will be striped,
5557 * then helper functions are called to find FIDs for the stripes, declare
5558 * creation of the stripes and declare initialization of the striping
5559 * information to be stored in the master object.
5561 * \see dt_object_operations::do_declare_create() in the API description
5564 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
5565 struct lu_attr *attr,
5566 struct dt_allocation_hint *hint,
5567 struct dt_object_format *dof, struct thandle *th)
5569 struct dt_object *next = dt_object_child(dt);
5570 struct lod_object *lo = lod_dt_obj(dt);
5579 * first of all, we declare creation of local object
5581 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
5586 * it's lod_ah_init() that has decided the object will be striped
5588 if (dof->dof_type == DFT_REGULAR) {
5589 /* callers don't want stripes */
5590 /* XXX: all tricky interactions with ->ah_make_hint() decided
5591 * to use striping, then ->declare_create() behaving differently
5592 * should be cleaned */
5593 if (dof->u.dof_reg.striped != 0)
5594 rc = lod_declare_striped_create(env, dt, attr,
5596 } else if (dof->dof_type == DFT_DIR) {
5597 struct seq_server_site *ss;
5598 struct lu_buf buf = { NULL };
5599 struct lu_buf *lmu = NULL;
5601 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
5603 /* If the parent has default stripeEA, and client
5604 * did not find it before sending create request,
5605 * then MDT will return -EREMOTE, and client will
5606 * retrieve the default stripeEA and re-create the
5609 * Note: if dah_eadata != NULL, it means creating the
5610 * striped directory with specified stripeEA, then it
5611 * should ignore the default stripeEA */
5612 if (hint != NULL && hint->dah_eadata == NULL) {
5613 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
5614 GOTO(out, rc = -EREMOTE);
5616 if (lo->ldo_dir_stripe_offset == LMV_OFFSET_DEFAULT) {
5617 struct lod_default_striping *lds;
5619 lds = lo->ldo_def_striping;
5621 * child and parent should be on the same MDT,
5622 * but if parent has default LMV, and the start
5623 * MDT offset is -1, it's allowed. This check
5624 * is not necessary after 2.12.22 because client
5625 * follows this already, but old client may not.
5627 if (hint->dah_parent &&
5628 dt_object_remote(hint->dah_parent) && lds &&
5629 lds->lds_dir_def_stripe_offset !=
5631 GOTO(out, rc = -EREMOTE);
5632 } else if (lo->ldo_dir_stripe_offset !=
5634 struct lod_device *lod;
5635 struct lu_tgt_desc *mdt = NULL;
5636 bool found_mdt = false;
5638 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5639 lod_foreach_mdt(lod, mdt) {
5640 if (mdt->ltd_index ==
5641 lo->ldo_dir_stripe_offset) {
5647 /* If the MDT indicated by stripe_offset can be
5648 * found, then tell client to resend the create
5649 * request to the correct MDT, otherwise return
5650 * error to client */
5652 GOTO(out, rc = -EREMOTE);
5654 GOTO(out, rc = -EINVAL);
5656 } else if (hint && hint->dah_eadata) {
5658 lmu->lb_buf = (void *)hint->dah_eadata;
5659 lmu->lb_len = hint->dah_eadata_len;
5662 rc = lod_declare_dir_striping_create(env, dt, attr, lmu, dof,
5666 /* failed to create striping or to set initial size, let's reset
5667 * config so that others don't get confused */
5669 lod_striping_free(env, lo);
5674 * Generate component ID for new created component.
5676 * \param[in] lo LOD object
5677 * \param[in] comp_idx index of ldo_comp_entries
5679 * \retval component ID on success
5680 * \retval LCME_ID_INVAL on failure
5682 static __u32 lod_gen_component_id(struct lod_object *lo,
5683 int mirror_id, int comp_idx)
5685 struct lod_layout_component *lod_comp;
5686 __u32 id, start, end;
5689 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
5691 lod_obj_inc_layout_gen(lo);
5692 id = lo->ldo_layout_gen;
5693 if (likely(id <= SEQ_ID_MAX))
5694 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
5696 /* Layout generation wraps, need to check collisions. */
5697 start = id & SEQ_ID_MASK;
5700 for (id = start; id <= end; id++) {
5701 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5702 lod_comp = &lo->ldo_comp_entries[i];
5703 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
5706 /* Found the ununsed ID */
5707 if (i == lo->ldo_comp_cnt)
5708 RETURN(pflr_id(mirror_id, id));
5710 if (end == LCME_ID_MAX) {
5712 end = min(lo->ldo_layout_gen & LCME_ID_MASK,
5713 (__u32)(LCME_ID_MAX - 1));
5717 RETURN(LCME_ID_INVAL);
5721 * Creation of a striped regular object.
5723 * The function is called to create the stripe objects for a regular
5724 * striped file. This can happen at the initial object creation or
5725 * when the caller asks LOD to do so using ->do_xattr_set() method
5726 * (so called late striping). Notice all the information are already
5727 * prepared in the form of the list of objects (ldo_stripe field).
5728 * This is done during declare phase.
5730 * \param[in] env execution environment
5731 * \param[in] dt object
5732 * \param[in] attr attributes the stripes will be created with
5733 * \param[in] dof format of stripes (see OSD API description)
5734 * \param[in] th transaction handle
5736 * \retval 0 on success
5737 * \retval negative if failed
5739 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
5740 struct lu_attr *attr, struct dt_object_format *dof,
5743 struct lod_layout_component *lod_comp;
5744 struct lod_object *lo = lod_dt_obj(dt);
5749 LASSERT((lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL) ||
5750 lo->ldo_is_foreign);
5752 mirror_id = 0; /* non-flr file's mirror_id is 0 */
5753 if (lo->ldo_mirror_count > 1) {
5754 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5755 lod_comp = &lo->ldo_comp_entries[i];
5756 if (lod_comp->llc_id != LCME_ID_INVAL &&
5757 mirror_id_of(lod_comp->llc_id) > mirror_id)
5758 mirror_id = mirror_id_of(lod_comp->llc_id);
5762 /* create all underlying objects */
5763 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5764 lod_comp = &lo->ldo_comp_entries[i];
5766 if (lod_comp->llc_id == LCME_ID_INVAL) {
5767 /* only the component of FLR layout with more than 1
5768 * mirror has mirror ID in its component ID.
5770 if (lod_comp->llc_extent.e_start == 0 &&
5771 lo->ldo_mirror_count > 1)
5774 lod_comp->llc_id = lod_gen_component_id(lo,
5776 if (lod_comp->llc_id == LCME_ID_INVAL)
5777 GOTO(out, rc = -ERANGE);
5780 if (lod_comp_inited(lod_comp))
5783 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
5784 lod_comp_set_init(lod_comp);
5786 if (lov_pattern(lod_comp->llc_pattern) == LOV_PATTERN_MDT)
5787 lod_comp_set_init(lod_comp);
5789 if (lod_comp->llc_stripe == NULL)
5792 LASSERT(lod_comp->llc_stripe_count);
5793 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
5794 struct dt_object *object = lod_comp->llc_stripe[j];
5795 LASSERT(object != NULL);
5796 rc = lod_sub_create(env, object, attr, NULL, dof, th);
5800 lod_comp_set_init(lod_comp);
5803 rc = lod_fill_mirrors(lo);
5807 rc = lod_generate_and_set_lovea(env, lo, th);
5811 lo->ldo_comp_cached = 1;
5815 lod_striping_free(env, lo);
5819 static inline bool lod_obj_is_dom(struct dt_object *dt)
5821 struct lod_object *lo = lod_dt_obj(dt);
5823 if (!dt_object_exists(dt_object_child(dt)))
5826 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
5829 if (!lo->ldo_comp_cnt)
5832 return (lov_pattern(lo->ldo_comp_entries[0].llc_pattern) ==
5837 * Implementation of dt_object_operations::do_create.
5839 * If any of preceeding methods (like ->do_declare_create(),
5840 * ->do_ah_init(), etc) chose to create a striped object,
5841 * then this method will create the master and the stripes.
5843 * \see dt_object_operations::do_create() in the API description for details.
5845 static int lod_create(const struct lu_env *env, struct dt_object *dt,
5846 struct lu_attr *attr, struct dt_allocation_hint *hint,
5847 struct dt_object_format *dof, struct thandle *th)
5852 /* create local object */
5853 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
5857 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
5858 (lod_obj_is_striped(dt) || lod_obj_is_dom(dt)) &&
5859 dof->u.dof_reg.striped != 0) {
5860 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
5861 rc = lod_striped_create(env, dt, attr, dof, th);
5868 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
5869 struct dt_object *dt, struct thandle *th,
5870 int comp_idx, int stripe_idx,
5871 struct lod_obj_stripe_cb_data *data)
5873 if (data->locd_declare)
5874 return lod_sub_declare_destroy(env, dt, th);
5875 else if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
5876 stripe_idx == cfs_fail_val)
5877 return lod_sub_destroy(env, dt, th);
5883 * Implementation of dt_object_operations::do_declare_destroy.
5885 * If the object is a striped directory, then the function declares reference
5886 * removal from the master object (this is an index) to the stripes and declares
5887 * destroy of all the stripes. In all the cases, it declares an intention to
5888 * destroy the object itself.
5890 * \see dt_object_operations::do_declare_destroy() in the API description
5893 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
5896 struct dt_object *next = dt_object_child(dt);
5897 struct lod_object *lo = lod_dt_obj(dt);
5898 struct lod_thread_info *info = lod_env_info(env);
5899 struct dt_object *stripe;
5900 char *stripe_name = info->lti_key;
5906 * load striping information, notice we don't do this when object
5907 * is being initialized as we don't need this information till
5908 * few specific cases like destroy, chown
5910 rc = lod_striping_load(env, lo);
5914 /* declare destroy for all underlying objects */
5915 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5916 rc = next->do_ops->do_index_try(env, next,
5917 &dt_directory_features);
5921 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5922 stripe = lo->ldo_stripe[i];
5926 rc = lod_sub_declare_ref_del(env, next, th);
5930 snprintf(stripe_name, sizeof(info->lti_key),
5932 PFID(lu_object_fid(&stripe->do_lu)), i);
5933 rc = lod_sub_declare_delete(env, next,
5934 (const struct dt_key *)stripe_name, th);
5941 * we declare destroy for the local object
5943 rc = lod_sub_declare_destroy(env, next, th);
5947 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
5948 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
5951 if (!lod_obj_is_striped(dt))
5954 /* declare destroy all striped objects */
5955 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5956 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5957 stripe = lo->ldo_stripe[i];
5961 if (!dt_object_exists(stripe))
5964 rc = lod_sub_declare_ref_del(env, stripe, th);
5968 rc = lod_sub_declare_destroy(env, stripe, th);
5973 struct lod_obj_stripe_cb_data data = { { 0 } };
5975 data.locd_declare = true;
5976 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
5977 rc = lod_obj_for_each_stripe(env, lo, th, &data);
5984 * Implementation of dt_object_operations::do_destroy.
5986 * If the object is a striped directory, then the function removes references
5987 * from the master object (this is an index) to the stripes and destroys all
5988 * the stripes. In all the cases, the function destroys the object itself.
5990 * \see dt_object_operations::do_destroy() in the API description for details.
5992 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
5995 struct dt_object *next = dt_object_child(dt);
5996 struct lod_object *lo = lod_dt_obj(dt);
5997 struct lod_thread_info *info = lod_env_info(env);
5998 char *stripe_name = info->lti_key;
5999 struct dt_object *stripe;
6005 /* destroy sub-stripe of master object */
6006 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6007 rc = next->do_ops->do_index_try(env, next,
6008 &dt_directory_features);
6012 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6013 stripe = lo->ldo_stripe[i];
6017 rc = lod_sub_ref_del(env, next, th);
6021 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
6022 PFID(lu_object_fid(&stripe->do_lu)), i);
6024 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
6025 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
6026 PFID(lu_object_fid(&stripe->do_lu)));
6028 rc = lod_sub_delete(env, next,
6029 (const struct dt_key *)stripe_name, th);
6035 rc = lod_sub_destroy(env, next, th);
6039 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
6040 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
6043 if (!lod_obj_is_striped(dt))
6046 /* destroy all striped objects */
6047 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6048 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6049 stripe = lo->ldo_stripe[i];
6053 if (!dt_object_exists(stripe))
6056 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
6057 i == cfs_fail_val) {
6058 dt_write_lock(env, stripe, DT_TGT_CHILD);
6059 rc = lod_sub_ref_del(env, stripe, th);
6060 dt_write_unlock(env, stripe);
6064 rc = lod_sub_destroy(env, stripe, th);
6070 struct lod_obj_stripe_cb_data data = { { 0 } };
6072 data.locd_declare = false;
6073 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
6074 rc = lod_obj_for_each_stripe(env, lo, th, &data);
6081 * Implementation of dt_object_operations::do_declare_ref_add.
6083 * \see dt_object_operations::do_declare_ref_add() in the API description
6086 static int lod_declare_ref_add(const struct lu_env *env,
6087 struct dt_object *dt, struct thandle *th)
6089 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
6093 * Implementation of dt_object_operations::do_ref_add.
6095 * \see dt_object_operations::do_ref_add() in the API description for details.
6097 static int lod_ref_add(const struct lu_env *env,
6098 struct dt_object *dt, struct thandle *th)
6100 return lod_sub_ref_add(env, dt_object_child(dt), th);
6104 * Implementation of dt_object_operations::do_declare_ref_del.
6106 * \see dt_object_operations::do_declare_ref_del() in the API description
6109 static int lod_declare_ref_del(const struct lu_env *env,
6110 struct dt_object *dt, struct thandle *th)
6112 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
6116 * Implementation of dt_object_operations::do_ref_del
6118 * \see dt_object_operations::do_ref_del() in the API description for details.
6120 static int lod_ref_del(const struct lu_env *env,
6121 struct dt_object *dt, struct thandle *th)
6123 return lod_sub_ref_del(env, dt_object_child(dt), th);
6127 * Implementation of dt_object_operations::do_object_sync.
6129 * \see dt_object_operations::do_object_sync() in the API description
6132 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
6133 __u64 start, __u64 end)
6135 return dt_object_sync(env, dt_object_child(dt), start, end);
6139 * Implementation of dt_object_operations::do_object_unlock.
6141 * Used to release LDLM lock(s).
6143 * \see dt_object_operations::do_object_unlock() in the API description
6146 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
6147 struct ldlm_enqueue_info *einfo,
6148 union ldlm_policy_data *policy)
6150 struct lod_object *lo = lod_dt_obj(dt);
6151 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
6152 int slave_locks_size;
6156 if (slave_locks == NULL)
6159 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
6160 /* Note: for remote lock for single stripe dir, MDT will cancel
6161 * the lock by lockh directly */
6162 LASSERT(!dt_object_remote(dt_object_child(dt)));
6164 /* locks were unlocked in MDT layer */
6165 for (i = 0; i < slave_locks->ha_count; i++)
6166 LASSERT(!lustre_handle_is_used(&slave_locks->ha_handles[i]));
6169 * NB, ha_count may not equal to ldo_dir_stripe_count, because dir
6170 * layout may change, e.g., shrink dir layout after migration.
6172 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6173 if (lo->ldo_stripe[i])
6174 dt_invalidate(env, lo->ldo_stripe[i]);
6177 slave_locks_size = offsetof(typeof(*slave_locks),
6178 ha_handles[slave_locks->ha_count]);
6179 OBD_FREE(slave_locks, slave_locks_size);
6180 einfo->ei_cbdata = NULL;
6186 * Implementation of dt_object_operations::do_object_lock.
6188 * Used to get LDLM lock on the non-striped and striped objects.
6190 * \see dt_object_operations::do_object_lock() in the API description
6193 static int lod_object_lock(const struct lu_env *env,
6194 struct dt_object *dt,
6195 struct lustre_handle *lh,
6196 struct ldlm_enqueue_info *einfo,
6197 union ldlm_policy_data *policy)
6199 struct lod_object *lo = lod_dt_obj(dt);
6200 int slave_locks_size;
6201 struct lustre_handle_array *slave_locks = NULL;
6206 /* remote object lock */
6207 if (!einfo->ei_enq_slave) {
6208 LASSERT(dt_object_remote(dt));
6209 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
6213 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
6216 rc = lod_striping_load(env, lo);
6221 if (lo->ldo_dir_stripe_count <= 1)
6224 slave_locks_size = offsetof(typeof(*slave_locks),
6225 ha_handles[lo->ldo_dir_stripe_count]);
6226 /* Freed in lod_object_unlock */
6227 OBD_ALLOC(slave_locks, slave_locks_size);
6230 slave_locks->ha_count = lo->ldo_dir_stripe_count;
6232 /* striped directory lock */
6233 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6234 struct lustre_handle lockh;
6235 struct ldlm_res_id *res_id;
6236 struct dt_object *stripe;
6238 stripe = lo->ldo_stripe[i];
6242 res_id = &lod_env_info(env)->lti_res_id;
6243 fid_build_reg_res_name(lu_object_fid(&stripe->do_lu), res_id);
6244 einfo->ei_res_id = res_id;
6246 if (dt_object_remote(stripe)) {
6247 set_bit(i, (void *)slave_locks->ha_map);
6248 rc = dt_object_lock(env, stripe, &lockh, einfo, policy);
6250 struct ldlm_namespace *ns = einfo->ei_namespace;
6251 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
6252 ldlm_completion_callback completion = einfo->ei_cb_cp;
6253 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
6255 if (einfo->ei_mode == LCK_PW ||
6256 einfo->ei_mode == LCK_EX)
6257 dlmflags |= LDLM_FL_COS_INCOMPAT;
6259 LASSERT(ns != NULL);
6260 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS,
6261 policy, einfo->ei_mode,
6262 &dlmflags, blocking,
6264 NULL, 0, LVB_T_NONE,
6269 ldlm_lock_decref_and_cancel(
6270 &slave_locks->ha_handles[i],
6272 OBD_FREE(slave_locks, slave_locks_size);
6275 slave_locks->ha_handles[i] = lockh;
6277 einfo->ei_cbdata = slave_locks;
6283 * Implementation of dt_object_operations::do_invalidate.
6285 * \see dt_object_operations::do_invalidate() in the API description for details
6287 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
6289 return dt_invalidate(env, dt_object_child(dt));
6292 static int lod_declare_instantiate_components(const struct lu_env *env,
6293 struct lod_object *lo, struct thandle *th)
6295 struct lod_thread_info *info = lod_env_info(env);
6300 LASSERT(info->lti_count < lo->ldo_comp_cnt);
6302 for (i = 0; i < info->lti_count; i++) {
6303 rc = lod_qos_prep_create(env, lo, NULL, th,
6304 info->lti_comp_idx[i]);
6310 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6311 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
6312 &info->lti_buf, XATTR_NAME_LOV, 0, th);
6319 * Check OSTs for an existing component for further extension
6321 * Checks if OSTs are still healthy and not out of space. Gets free space
6322 * on OSTs (relative to allocation watermark rmb_low) and compares to
6323 * the proposed new_end for this component.
6325 * Decides whether or not to extend a component on its current OSTs.
6327 * \param[in] env execution environment for this thread
6328 * \param[in] lo object we're checking
6329 * \param[in] index index of this component
6330 * \param[in] extension_size extension size for this component
6331 * \param[in] extent layout extent for requested operation
6332 * \param[in] comp_extent extension component extent
6333 * \param[in] write if this is write operation
6335 * \retval true - OK to extend on current OSTs
6336 * \retval false - do not extend on current OSTs
6338 static bool lod_sel_osts_allowed(const struct lu_env *env,
6339 struct lod_object *lo,
6340 int index, __u64 extension_size,
6341 struct lu_extent *extent,
6342 struct lu_extent *comp_extent, int write)
6344 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[index];
6345 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6346 struct obd_statfs *sfs = &lod_env_info(env)->lti_osfs;
6347 __u64 available = 0;
6354 LASSERT(lod_comp->llc_stripe_count != 0);
6357 (extent->e_start == 0 && extent->e_end == OBD_OBJECT_EOF)) {
6358 /* truncate or append */
6359 size = extension_size;
6361 /* In case of write op, check the real write extent,
6362 * it may be larger than the extension_size */
6363 size = roundup(min(extent->e_end, comp_extent->e_end) -
6364 max(extent->e_start, comp_extent->e_start),
6367 /* extension_size is file level, so we must divide by stripe count to
6368 * compare it to available space on a single OST */
6369 size /= lod_comp->llc_stripe_count;
6371 lod_getref(&lod->lod_ost_descs);
6372 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
6373 int index = lod_comp->llc_ost_indices[i];
6374 struct lod_tgt_desc *ost = OST_TGT(lod, index);
6375 struct obd_statfs_info info = { 0 };
6376 int j, repeated = 0;
6380 /* Get the number of times this OST repeats in this component.
6381 * Note: inter-component repeats are not counted as this is
6382 * considered as a rare case: we try to not repeat OST in other
6383 * components if possible. */
6384 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6385 if (index != lod_comp->llc_ost_indices[j])
6388 /* already handled */
6394 if (j < lod_comp->llc_stripe_count)
6397 if (!cfs_bitmap_check(lod->lod_ost_bitmap, index)) {
6398 CDEBUG(D_LAYOUT, "ost %d no longer present\n", index);
6403 rc = dt_statfs_info(env, ost->ltd_tgt, sfs, &info);
6405 CDEBUG(D_LAYOUT, "statfs failed for ost %d, error %d\n",
6411 if (sfs->os_state & OS_STATFS_ENOSPC ||
6412 sfs->os_state & OS_STATFS_READONLY ||
6413 sfs->os_state & OS_STATFS_DEGRADED) {
6414 CDEBUG(D_LAYOUT, "ost %d is not availble for SEL "
6415 "extension, state %u\n", index, sfs->os_state);
6421 available = sfs->os_bavail * sfs->os_bsize;
6422 /* 'available' is relative to the allocation threshold */
6423 available -= (__u64) info.os_reserved_mb_low << 20;
6425 CDEBUG(D_LAYOUT, "ost %d lowwm: %d highwm: %d, "
6426 "%llu %% blocks available, %llu %% blocks free\n",
6427 index, info.os_reserved_mb_low, info.os_reserved_mb_high,
6428 (100ull * sfs->os_bavail) / sfs->os_blocks,
6429 (100ull * sfs->os_bfree) / sfs->os_blocks);
6431 if (size * repeated > available) {
6433 CDEBUG(D_LAYOUT, "low space on ost %d, available %llu "
6434 "< extension size %llu\n", index, available,
6439 lod_putref(lod, &lod->lod_ost_descs);
6445 * Adjust extents after component removal
6447 * When we remove an extension component, we move the start of the next
6448 * component to match the start of the extension component, so no space is left
6451 * \param[in] env execution environment for this thread
6452 * \param[in] lo object
6453 * \param[in] max_comp layout component
6454 * \param[in] index index of this component
6456 * \retval 0 on success
6457 * \retval negative errno on error
6459 static void lod_sel_adjust_extents(const struct lu_env *env,
6460 struct lod_object *lo,
6461 int max_comp, int index)
6463 struct lod_layout_component *lod_comp = NULL;
6464 struct lod_layout_component *next = NULL;
6465 struct lod_layout_component *prev = NULL;
6466 __u64 new_start = 0;
6470 /* Extension space component */
6471 lod_comp = &lo->ldo_comp_entries[index];
6472 next = &lo->ldo_comp_entries[index + 1];
6473 prev = &lo->ldo_comp_entries[index - 1];
6475 LASSERT(lod_comp != NULL && prev != NULL && next != NULL);
6476 LASSERT(lod_comp->llc_flags & LCME_FL_EXTENSION);
6478 /* Previous is being removed */
6479 if (prev && prev->llc_id == LCME_ID_INVAL)
6480 new_start = prev->llc_extent.e_start;
6482 new_start = lod_comp->llc_extent.e_start;
6484 for (i = index + 1; i < max_comp; i++) {
6485 lod_comp = &lo->ldo_comp_entries[i];
6487 start = lod_comp->llc_extent.e_start;
6488 lod_comp->llc_extent.e_start = new_start;
6490 /* We only move zero length extendable components */
6491 if (!(start == lod_comp->llc_extent.e_end))
6494 LASSERT(!(lod_comp->llc_flags & LCME_FL_INIT));
6496 lod_comp->llc_extent.e_end = new_start;
6500 /* Calculate the proposed 'new end' for a component we're extending */
6501 static __u64 lod_extension_new_end(__u64 extension_size, __u64 extent_end,
6502 __u32 stripe_size, __u64 component_end,
6503 __u64 extension_end)
6507 LASSERT(extension_size != 0 && stripe_size != 0);
6509 /* Round up to extension size */
6510 if (extent_end == OBD_OBJECT_EOF) {
6511 new_end = OBD_OBJECT_EOF;
6513 /* Add at least extension_size to the previous component_end,
6514 * covering the req layout extent */
6515 new_end = max(extent_end - component_end, extension_size);
6516 new_end = roundup(new_end, extension_size);
6517 new_end += component_end;
6519 /* Component end must be min stripe size aligned */
6520 if (new_end % stripe_size) {
6521 CDEBUG(D_LAYOUT, "new component end is not aligned "
6522 "by the stripe size %u: [%llu, %llu) ext size "
6523 "%llu new end %llu, aligning\n",
6524 stripe_size, component_end, extent_end,
6525 extension_size, new_end);
6526 new_end = roundup(new_end, stripe_size);
6530 if (new_end < extent_end)
6531 new_end = OBD_OBJECT_EOF;
6534 /* Don't extend past the end of the extension component */
6535 if (new_end > extension_end)
6536 new_end = extension_end;
6541 /* As lod_sel_handler() could be re-entered for the same component several
6542 * times, this is the data for the next call. Fields could be changed to
6543 * component indexes when needed, (e.g. if there is no need to instantiate
6544 * all the previous components up to the current position) to tell the caller
6545 * where to start over from. */
6552 * Process extent updates for a particular layout component
6554 * Handle layout updates for a particular extension space component touched by
6555 * a layout update operation. Core function of self-extending PFL feature.
6557 * In general, this function processes exactly *one* stage of an extension
6558 * operation, modifying the layout accordingly, then returns to the caller.
6559 * The caller is responsible for restarting processing with the new layout,
6560 * which may repeatedly return to this function until the extension updates
6563 * This function does one of a few things to the layout:
6564 * 1. Extends the component before the current extension space component to
6565 * allow it to accomodate the requested operation (if space/policy permit that
6566 * component to continue on its current OSTs)
6568 * 2. If extension of the existing component fails, we do one of two things:
6569 * a. If there is a component after the extension space, we remove the
6570 * extension space component, move the start of the next component down
6571 * accordingly, then notify the caller to restart processing w/the new
6573 * b. If there is no following component, we try repeating the current
6574 * component, creating a new component using the current one as a
6575 * template (keeping its stripe properties but not specific striping),
6576 * and try assigning striping for this component. If there is sufficient
6577 * free space on the OSTs chosen for this component, it is instantiated
6578 * and i/o continues there.
6580 * If there is not sufficient space on the new OSTs, we remove this new
6581 * component & extend the current component.
6583 * Note further that uninited components followed by extension space can be zero
6584 * length meaning that we will try to extend them before initializing them, and
6585 * if that fails, they will be removed without initialization.
6587 * 3. If we extend to/beyond the end of an extension space component, that
6588 * component is exhausted (all of its range has been given to real components),
6589 * so we remove it and restart processing.
6591 * \param[in] env execution environment for this thread
6592 * \param[in,out] lo object to update the layout of
6593 * \param[in] extent layout extent for requested operation, update
6594 * layout to fit this operation
6595 * \param[in] th transaction handle for this operation
6596 * \param[in,out] max_comp the highest comp for the portion of the layout
6597 * we are operating on (For FLR, the chosen
6598 * replica). Updated because we may remove
6600 * \param[in] index index of the extension space component we're
6602 * \param[in] write if this is write op
6603 * \param[in,out] force if the extension is to be forced; set here
6604 to force it on the 2nd call for the same
6607 * \retval 0 on success
6608 * \retval negative errno on error
6610 static int lod_sel_handler(const struct lu_env *env,
6611 struct lod_object *lo,
6612 struct lu_extent *extent,
6613 struct thandle *th, int *max_comp,
6614 int index, int write,
6615 struct sel_data *sd)
6617 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6618 struct lod_thread_info *info = lod_env_info(env);
6619 struct lod_layout_component *lod_comp;
6620 struct lod_layout_component *prev;
6621 struct lod_layout_component *next = NULL;
6622 __u64 extension_size;
6629 /* First component cannot be extension space */
6631 CERROR("%s: "DFID" first component cannot be extension space\n",
6632 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
6636 lod_comp = &lo->ldo_comp_entries[index];
6637 prev = &lo->ldo_comp_entries[index - 1];
6638 if ((index + 1) < *max_comp)
6639 next = &lo->ldo_comp_entries[index + 1];
6641 /* extension size uses the stripe size field as KiB */
6642 extension_size = lod_comp->llc_stripe_size * SEL_UNIT_SIZE;
6644 CDEBUG(D_LAYOUT, "prev start %llu, extension start %llu, extension end"
6645 " %llu, extension size %llu\n", prev->llc_extent.e_start,
6646 lod_comp->llc_extent.e_start, lod_comp->llc_extent.e_end,
6649 /* Two extension space components cannot be adjacent & extension space
6650 * components cannot be init */
6651 if ((prev->llc_flags & LCME_FL_EXTENSION) ||
6652 !(ergo(next, !(next->llc_flags & LCME_FL_EXTENSION))) ||
6653 lod_comp_inited(lod_comp)) {
6654 CERROR("%s: "DFID" invalid extension space components\n",
6655 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
6659 if (!prev->llc_stripe) {
6660 CDEBUG(D_LAYOUT, "Previous component not inited\n");
6661 info->lti_count = 1;
6662 info->lti_comp_idx[0] = index - 1;
6663 rc = lod_declare_instantiate_components(env, lo, th);
6664 /* ENOSPC tells us we can't use this component. If there is
6665 * a next or we are repeating, we either spill over (next) or
6666 * extend the original comp (repeat). Otherwise, return the
6667 * error to the user. */
6668 if (rc == -ENOSPC && (next || sd->sd_repeat))
6674 if (sd->sd_force == 0 && rc == 0)
6675 rc = !lod_sel_osts_allowed(env, lo, index - 1,
6676 extension_size, extent,
6677 &lod_comp->llc_extent, write);
6679 repeated = !!(sd->sd_repeat);
6683 /* Extend previous component */
6685 new_end = lod_extension_new_end(extension_size, extent->e_end,
6686 prev->llc_stripe_size,
6687 prev->llc_extent.e_end,
6688 lod_comp->llc_extent.e_end);
6690 CDEBUG(D_LAYOUT, "new end %llu\n", new_end);
6691 lod_comp->llc_extent.e_start = new_end;
6692 prev->llc_extent.e_end = new_end;
6694 if (prev->llc_extent.e_end == lod_comp->llc_extent.e_end) {
6695 CDEBUG(D_LAYOUT, "Extension component exhausted\n");
6696 lod_comp->llc_id = LCME_ID_INVAL;
6700 /* rc == 1, failed to extend current component */
6703 /* Normal 'spillover' case - Remove the extension
6704 * space component & bring down the start of the next
6706 lod_comp->llc_id = LCME_ID_INVAL;
6708 if (!(prev->llc_flags & LCME_FL_INIT)) {
6709 prev->llc_id = LCME_ID_INVAL;
6712 lod_sel_adjust_extents(env, lo, *max_comp, index);
6713 } else if (lod_comp_inited(prev)) {
6714 /* If there is no next, and the previous component is
6715 * INIT'ed, try repeating the previous component. */
6716 LASSERT(repeated == 0);
6717 rc = lod_layout_repeat_comp(env, lo, index - 1);
6721 /* The previous component is a repeated component.
6722 * Record this so we don't keep trying to repeat it. */
6725 /* If the previous component is not INIT'ed, this may
6726 * be a component we have just instantiated but failed
6727 * to extend. Or even a repeated component we failed
6728 * to prepare a striping for. Do not repeat but instead
6729 * remove the repeated component & force the extention
6730 * of the original one */
6733 prev->llc_id = LCME_ID_INVAL;
6740 rc = lod_layout_del_prep_layout(env, lo, NULL);
6743 LASSERTF(-rc == change,
6744 "number deleted %d != requested %d\n", -rc,
6747 *max_comp = *max_comp + change;
6749 /* lod_del_prep_layout reallocates ldo_comp_entries, so we must
6750 * refresh these pointers before using them */
6751 lod_comp = &lo->ldo_comp_entries[index];
6752 prev = &lo->ldo_comp_entries[index - 1];
6753 CDEBUG(D_LAYOUT, "After extent updates: prev start %llu, current start "
6754 "%llu, current end %llu max_comp %d ldo_comp_cnt %d\n",
6755 prev->llc_extent.e_start, lod_comp->llc_extent.e_start,
6756 lod_comp->llc_extent.e_end, *max_comp, lo->ldo_comp_cnt);
6758 /* Layout changed successfully */
6763 * Declare layout extent updates
6765 * Handles extensions. Identifies extension components touched by current
6766 * operation and passes them to processing function.
6768 * Restarts with updated layouts from the processing function until the current
6769 * operation no longer touches an extension space component.
6771 * \param[in] env execution environment for this thread
6772 * \param[in,out] lo object to update the layout of
6773 * \param[in] extent layout extent for requested operation, update layout to
6774 * fit this operation
6775 * \param[in] th transaction handle for this operation
6776 * \param[in] pick identifies chosen mirror for FLR layouts
6777 * \param[in] write if this is write op
6779 * \retval 1 on layout changed, 0 on no change
6780 * \retval negative errno on error
6782 static int lod_declare_update_extents(const struct lu_env *env,
6783 struct lod_object *lo, struct lu_extent *extent,
6784 struct thandle *th, int pick, int write)
6786 struct lod_thread_info *info = lod_env_info(env);
6787 struct lod_layout_component *lod_comp;
6788 bool layout_changed = false;
6789 struct sel_data sd = { 0 };
6797 /* This makes us work on the components of the chosen mirror */
6798 start_index = lo->ldo_mirrors[pick].lme_start;
6799 max_comp = lo->ldo_mirrors[pick].lme_end + 1;
6800 if (lo->ldo_flr_state == LCM_FL_NONE)
6801 LASSERT(start_index == 0 && max_comp == lo->ldo_comp_cnt);
6803 CDEBUG(D_LAYOUT, "extent->e_start %llu, extent->e_end %llu\n",
6804 extent->e_start, extent->e_end);
6805 for (i = start_index; i < max_comp; i++) {
6806 lod_comp = &lo->ldo_comp_entries[i];
6808 /* We've passed all components of interest */
6809 if (lod_comp->llc_extent.e_start >= extent->e_end)
6812 if (lod_comp->llc_flags & LCME_FL_EXTENSION) {
6813 layout_changed = true;
6814 rc = lod_sel_handler(env, lo, extent, th, &max_comp,
6819 /* Nothing has changed behind the prev one */
6825 /* We may have added or removed components. If so, we must update the
6826 * start & ends of all the mirrors after the current one, and the end
6827 * of the current mirror. */
6828 change = max_comp - 1 - lo->ldo_mirrors[pick].lme_end;
6830 lo->ldo_mirrors[pick].lme_end += change;
6831 for (i = pick + 1; i < lo->ldo_mirror_count; i++) {
6832 lo->ldo_mirrors[i].lme_start += change;
6833 lo->ldo_mirrors[i].lme_end += change;
6839 /* The amount of components has changed, adjust the lti_comp_idx */
6840 rc2 = lod_layout_data_init(info, lo->ldo_comp_cnt);
6842 return rc < 0 ? rc : rc2 < 0 ? rc2 : layout_changed;
6845 /* If striping is already instantiated or INIT'ed DOM? */
6846 static bool lod_is_instantiation_needed(struct lod_layout_component *comp)
6848 return !(((lov_pattern(comp->llc_pattern) == LOV_PATTERN_MDT) &&
6849 lod_comp_inited(comp)) || comp->llc_stripe);
6853 * Declare layout update for a non-FLR layout.
6855 * \param[in] env execution environment for this thread
6856 * \param[in,out] lo object to update the layout of
6857 * \param[in] layout layout intent for requested operation, "update" is
6858 * a process of reacting to this
6859 * \param[in] buf buffer containing lov ea (see comment on usage inline)
6860 * \param[in] th transaction handle for this operation
6862 * \retval 0 on success
6863 * \retval negative errno on error
6865 static int lod_declare_update_plain(const struct lu_env *env,
6866 struct lod_object *lo, struct layout_intent *layout,
6867 const struct lu_buf *buf, struct thandle *th)
6869 struct lod_thread_info *info = lod_env_info(env);
6870 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6871 struct lod_layout_component *lod_comp;
6872 struct lov_comp_md_v1 *comp_v1 = NULL;
6873 bool layout_changed = false;
6874 bool replay = false;
6878 LASSERT(lo->ldo_flr_state == LCM_FL_NONE);
6881 * In case the client is passing lovea, which only happens during
6882 * the replay of layout intent write RPC for now, we may need to
6883 * parse the lovea and apply new layout configuration.
6885 if (buf && buf->lb_len) {
6886 struct lov_user_md_v1 *v1 = buf->lb_buf;
6888 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
6889 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
6890 LOV_MAGIC_COMP_V1)) {
6891 CERROR("%s: the replay buffer of layout extend "
6892 "(magic %#x) does not contain expected "
6893 "composite layout.\n",
6894 lod2obd(d)->obd_name, v1->lmm_magic);
6895 GOTO(out, rc = -EINVAL);
6898 rc = lod_use_defined_striping(env, lo, buf);
6901 lo->ldo_comp_cached = 1;
6903 rc = lod_get_lov_ea(env, lo);
6906 /* old on-disk EA is stored in info->lti_buf */
6907 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
6909 layout_changed = true;
6911 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
6915 /* non replay path */
6916 rc = lod_striping_load(env, lo);
6921 /* Make sure defined layout covers the requested write range. */
6922 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
6923 if (lo->ldo_comp_cnt > 1 &&
6924 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
6925 lod_comp->llc_extent.e_end < layout->li_extent.e_end) {
6926 CDEBUG_LIMIT(replay ? D_ERROR : D_LAYOUT,
6927 "%s: the defined layout [0, %#llx) does not "
6928 "covers the write range "DEXT"\n",
6929 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
6930 PEXT(&layout->li_extent));
6931 GOTO(out, rc = -EINVAL);
6934 CDEBUG(D_LAYOUT, "%s: "DFID": update components "DEXT"\n",
6935 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
6936 PEXT(&layout->li_extent));
6939 rc = lod_declare_update_extents(env, lo, &layout->li_extent,
6940 th, 0, layout->li_opc == LAYOUT_INTENT_WRITE);
6944 layout_changed = true;
6948 * Iterate ld->ldo_comp_entries, find the component whose extent under
6949 * the write range and not instantianted.
6951 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6952 lod_comp = &lo->ldo_comp_entries[i];
6954 if (lod_comp->llc_extent.e_start >= layout->li_extent.e_end)
6958 /* If striping is instantiated or INIT'ed DOM skip */
6959 if (!lod_is_instantiation_needed(lod_comp))
6963 * In replay path, lod_comp is the EA passed by
6964 * client replay buffer, comp_v1 is the pre-recovery
6965 * on-disk EA, we'd sift out those components which
6966 * were init-ed in the on-disk EA.
6968 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
6973 * this component hasn't instantiated in normal path, or during
6974 * replay it needs replay the instantiation.
6977 /* A released component is being extended */
6978 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
6979 GOTO(out, rc = -EINVAL);
6981 LASSERT(info->lti_comp_idx != NULL);
6982 info->lti_comp_idx[info->lti_count++] = i;
6983 layout_changed = true;
6986 if (!layout_changed)
6989 lod_obj_inc_layout_gen(lo);
6990 rc = lod_declare_instantiate_components(env, lo, th);
6994 lod_striping_free(env, lo);
6998 static inline int lod_comp_index(struct lod_object *lo,
6999 struct lod_layout_component *lod_comp)
7001 LASSERT(lod_comp >= lo->ldo_comp_entries &&
7002 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
7004 return lod_comp - lo->ldo_comp_entries;
7008 * Stale other mirrors by writing extent.
7010 static int lod_stale_components(const struct lu_env *env, struct lod_object *lo,
7011 int primary, struct lu_extent *extent,
7014 struct lod_layout_component *pri_comp, *lod_comp;
7015 struct lu_extent pri_extent;
7020 /* The writing extent decides which components in the primary
7021 * are affected... */
7022 CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
7025 lod_foreach_mirror_comp(pri_comp, lo, primary) {
7026 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
7029 CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
7030 lod_comp_index(lo, pri_comp),
7031 PEXT(&pri_comp->llc_extent));
7033 pri_extent.e_start = pri_comp->llc_extent.e_start;
7034 pri_extent.e_end = pri_comp->llc_extent.e_end;
7036 for (i = 0; i < lo->ldo_mirror_count; i++) {
7039 rc = lod_declare_update_extents(env, lo, &pri_extent,
7041 /* if update_extents changed the layout, it may have
7042 * reallocated the component array, so start over to
7043 * avoid using stale pointers */
7049 /* ... and then stale other components that are
7050 * overlapping with primary components */
7051 lod_foreach_mirror_comp(lod_comp, lo, i) {
7052 if (!lu_extent_is_overlapped(
7054 &lod_comp->llc_extent))
7057 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
7058 i, lod_comp_index(lo, lod_comp));
7060 lod_comp->llc_flags |= LCME_FL_STALE;
7061 lo->ldo_mirrors[i].lme_stale = 1;
7070 * check an OST's availability
7071 * \param[in] env execution environment
7072 * \param[in] lo lod object
7073 * \param[in] dt dt object
7074 * \param[in] index mirror index
7076 * \retval negative if failed
7077 * \retval 1 if \a dt is available
7078 * \retval 0 if \a dt is not available
7080 static inline int lod_check_ost_avail(const struct lu_env *env,
7081 struct lod_object *lo,
7082 struct dt_object *dt, int index)
7084 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7085 struct lod_tgt_desc *ost;
7087 int type = LU_SEQ_RANGE_OST;
7090 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &idx, &type);
7092 CERROR("%s: can't locate "DFID":rc = %d\n",
7093 lod2obd(lod)->obd_name, PFID(lu_object_fid(&dt->do_lu)),
7098 ost = OST_TGT(lod, idx);
7099 if (ost->ltd_statfs.os_state &
7100 (OS_STATFS_READONLY | OS_STATFS_ENOSPC | OS_STATFS_ENOINO |
7101 OS_STATFS_NOPRECREATE) ||
7102 ost->ltd_active == 0) {
7103 CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail, rc = %d\n",
7104 PFID(lod_object_fid(lo)), index, idx, rc);
7112 * Pick primary mirror for write
7113 * \param[in] env execution environment
7114 * \param[in] lo object
7115 * \param[in] extent write range
7117 static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo,
7118 struct lu_extent *extent)
7120 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7121 unsigned int seq = 0;
7122 struct lod_layout_component *lod_comp;
7124 int picked = -1, second_pick = -1, third_pick = -1;
7127 if (OBD_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
7128 get_random_bytes(&seq, sizeof(seq));
7129 seq %= lo->ldo_mirror_count;
7133 * Pick a mirror as the primary, and check the availability of OSTs.
7135 * This algo can be revised later after knowing the topology of
7138 lod_qos_statfs_update(env, lod, &lod->lod_ost_descs);
7139 for (i = 0; i < lo->ldo_mirror_count; i++) {
7140 bool ost_avail = true;
7141 int index = (i + seq) % lo->ldo_mirror_count;
7143 if (lo->ldo_mirrors[index].lme_stale) {
7144 CDEBUG(D_LAYOUT, DFID": mirror %d stale\n",
7145 PFID(lod_object_fid(lo)), index);
7149 /* 2nd pick is for the primary mirror containing unavail OST */
7150 if (lo->ldo_mirrors[index].lme_primary && second_pick < 0)
7151 second_pick = index;
7153 /* 3rd pick is for non-primary mirror containing unavail OST */
7154 if (second_pick < 0 && third_pick < 0)
7158 * we found a non-primary 1st pick, we'd like to find a
7159 * potential pirmary mirror.
7161 if (picked >= 0 && !lo->ldo_mirrors[index].lme_primary)
7164 /* check the availability of OSTs */
7165 lod_foreach_mirror_comp(lod_comp, lo, index) {
7166 if (!lod_comp_inited(lod_comp) || !lod_comp->llc_stripe)
7169 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
7170 struct dt_object *dt = lod_comp->llc_stripe[j];
7172 rc = lod_check_ost_avail(env, lo, dt, index);
7179 } /* for all dt object in one component */
7182 } /* for all components in a mirror */
7185 * the OSTs where allocated objects locates in the components
7186 * of the mirror are available.
7191 /* this mirror has all OSTs available */
7195 * primary with all OSTs are available, this is the perfect
7198 if (lo->ldo_mirrors[index].lme_primary)
7200 } /* for all mirrors */
7202 /* failed to pick a sound mirror, lower our expectation */
7204 picked = second_pick;
7206 picked = third_pick;
7213 static int lod_prepare_resync_mirror(const struct lu_env *env,
7214 struct lod_object *lo,
7217 struct lod_thread_info *info = lod_env_info(env);
7218 struct lod_layout_component *lod_comp;
7219 bool neg = !!(MIRROR_ID_NEG & mirror_id);
7222 mirror_id &= ~MIRROR_ID_NEG;
7224 for (i = 0; i < lo->ldo_mirror_count; i++) {
7225 if ((!neg && lo->ldo_mirrors[i].lme_id != mirror_id) ||
7226 (neg && lo->ldo_mirrors[i].lme_id == mirror_id))
7229 lod_foreach_mirror_comp(lod_comp, lo, i) {
7230 if (lod_comp_inited(lod_comp))
7233 info->lti_comp_idx[info->lti_count++] =
7234 lod_comp_index(lo, lod_comp);
7242 * figure out the components should be instantiated for resync.
7244 static int lod_prepare_resync(const struct lu_env *env, struct lod_object *lo,
7245 struct lu_extent *extent)
7247 struct lod_thread_info *info = lod_env_info(env);
7248 struct lod_layout_component *lod_comp;
7249 unsigned int need_sync = 0;
7253 DFID": instantiate all stale components in "DEXT"\n",
7254 PFID(lod_object_fid(lo)), PEXT(extent));
7257 * instantiate all components within this extent, even non-stale
7260 for (i = 0; i < lo->ldo_mirror_count; i++) {
7261 if (!lo->ldo_mirrors[i].lme_stale)
7264 lod_foreach_mirror_comp(lod_comp, lo, i) {
7265 if (!lu_extent_is_overlapped(extent,
7266 &lod_comp->llc_extent))
7271 if (lod_comp_inited(lod_comp))
7274 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
7275 i, lod_comp_index(lo, lod_comp));
7276 info->lti_comp_idx[info->lti_count++] =
7277 lod_comp_index(lo, lod_comp);
7281 return need_sync ? 0 : -EALREADY;
7284 static int lod_declare_update_rdonly(const struct lu_env *env,
7285 struct lod_object *lo, struct md_layout_change *mlc,
7288 struct lod_thread_info *info = lod_env_info(env);
7289 struct lu_attr *layout_attr = &info->lti_layout_attr;
7290 struct lod_layout_component *lod_comp;
7291 struct lu_extent extent = { 0 };
7295 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
7296 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
7297 mlc->mlc_opc == MD_LAYOUT_RESYNC);
7298 LASSERT(lo->ldo_mirror_count > 0);
7300 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7301 struct layout_intent *layout = mlc->mlc_intent;
7302 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
7305 extent = layout->li_extent;
7306 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
7307 PFID(lod_object_fid(lo)), PEXT(&extent));
7309 picked = lod_primary_pick(env, lo, &extent);
7313 CDEBUG(D_LAYOUT, DFID": picked mirror id %u as primary\n",
7314 PFID(lod_object_fid(lo)),
7315 lo->ldo_mirrors[picked].lme_id);
7317 /* Update extents of primary before staling */
7318 rc = lod_declare_update_extents(env, lo, &extent, th, picked,
7323 if (layout->li_opc == LAYOUT_INTENT_TRUNC) {
7325 * trunc transfers [0, size) in the intent extent, we'd
7326 * stale components overlapping [size, eof).
7328 extent.e_start = extent.e_end;
7329 extent.e_end = OBD_OBJECT_EOF;
7332 /* stale overlapping components from other mirrors */
7333 rc = lod_stale_components(env, lo, picked, &extent, th);
7337 /* restore truncate intent extent */
7338 if (layout->li_opc == LAYOUT_INTENT_TRUNC)
7339 extent.e_end = extent.e_start;
7341 /* instantiate components for the picked mirror, start from 0 */
7344 lod_foreach_mirror_comp(lod_comp, lo, picked) {
7345 if (!lu_extent_is_overlapped(&extent,
7346 &lod_comp->llc_extent))
7349 if (!lod_is_instantiation_needed(lod_comp))
7352 info->lti_comp_idx[info->lti_count++] =
7353 lod_comp_index(lo, lod_comp);
7356 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
7357 } else { /* MD_LAYOUT_RESYNC */
7361 * could contain multiple non-stale mirrors, so we need to
7362 * prep uninited all components assuming any non-stale mirror
7363 * could be picked as the primary mirror.
7365 if (mlc->mlc_mirror_id == 0) {
7367 for (i = 0; i < lo->ldo_mirror_count; i++) {
7368 if (lo->ldo_mirrors[i].lme_stale)
7371 lod_foreach_mirror_comp(lod_comp, lo, i) {
7372 if (!lod_comp_inited(lod_comp))
7376 lod_comp->llc_extent.e_end)
7378 lod_comp->llc_extent.e_end;
7381 rc = lod_prepare_resync(env, lo, &extent);
7385 /* mirror write, try to init its all components */
7386 rc = lod_prepare_resync_mirror(env, lo,
7387 mlc->mlc_mirror_id);
7392 /* change the file state to SYNC_PENDING */
7393 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
7396 /* Reset the layout version once it's becoming too large.
7397 * This way it can make sure that the layout version is
7398 * monotonously increased in this writing era. */
7399 lod_obj_inc_layout_gen(lo);
7400 if (lo->ldo_layout_gen > (LCME_ID_MAX >> 1)) {
7401 __u32 layout_version;
7403 get_random_bytes(&layout_version, sizeof(layout_version));
7404 lo->ldo_layout_gen = layout_version & 0xffff;
7407 rc = lod_declare_instantiate_components(env, lo, th);
7411 layout_attr->la_valid = LA_LAYOUT_VERSION;
7412 layout_attr->la_layout_version = 0; /* set current version */
7413 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
7414 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
7415 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
7421 lod_striping_free(env, lo);
7425 static int lod_declare_update_write_pending(const struct lu_env *env,
7426 struct lod_object *lo, struct md_layout_change *mlc,
7429 struct lod_thread_info *info = lod_env_info(env);
7430 struct lu_attr *layout_attr = &info->lti_layout_attr;
7431 struct lod_layout_component *lod_comp;
7432 struct lu_extent extent = { 0 };
7438 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
7439 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
7440 mlc->mlc_opc == MD_LAYOUT_RESYNC);
7442 /* look for the primary mirror */
7443 for (i = 0; i < lo->ldo_mirror_count; i++) {
7444 if (lo->ldo_mirrors[i].lme_stale)
7447 LASSERTF(primary < 0, DFID " has multiple primary: %u / %u\n",
7448 PFID(lod_object_fid(lo)),
7449 lo->ldo_mirrors[i].lme_id,
7450 lo->ldo_mirrors[primary].lme_id);
7455 CERROR(DFID ": doesn't have a primary mirror\n",
7456 PFID(lod_object_fid(lo)));
7457 GOTO(out, rc = -ENODATA);
7460 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
7461 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
7463 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
7465 /* for LAYOUT_WRITE opc, it has to do the following operations:
7466 * 1. stale overlapping componets from stale mirrors;
7467 * 2. instantiate components of the primary mirror;
7468 * 3. transfter layout version to all objects of the primary;
7470 * for LAYOUT_RESYNC opc, it will do:
7471 * 1. instantiate components of all stale mirrors;
7472 * 2. transfer layout version to all objects to close write era. */
7474 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7475 struct layout_intent *layout = mlc->mlc_intent;
7476 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
7478 LASSERT(mlc->mlc_intent != NULL);
7480 extent = mlc->mlc_intent->li_extent;
7482 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
7483 PFID(lod_object_fid(lo)), PEXT(&extent));
7485 /* 1. Update extents of primary before staling */
7486 rc = lod_declare_update_extents(env, lo, &extent, th, primary,
7491 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC) {
7493 * trunc transfers [0, size) in the intent extent, we'd
7494 * stale components overlapping [size, eof).
7496 extent.e_start = extent.e_end;
7497 extent.e_end = OBD_OBJECT_EOF;
7500 /* 2. stale overlapping components */
7501 rc = lod_stale_components(env, lo, primary, &extent, th);
7505 /* 3. find the components which need instantiating.
7506 * instantiate [0, mlc->mlc_intent->e_end) */
7508 /* restore truncate intent extent */
7509 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC)
7510 extent.e_end = extent.e_start;
7513 lod_foreach_mirror_comp(lod_comp, lo, primary) {
7514 if (!lu_extent_is_overlapped(&extent,
7515 &lod_comp->llc_extent))
7518 if (!lod_is_instantiation_needed(lod_comp))
7521 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
7522 primary, lod_comp_index(lo, lod_comp));
7523 info->lti_comp_idx[info->lti_count++] =
7524 lod_comp_index(lo, lod_comp);
7526 } else { /* MD_LAYOUT_RESYNC */
7527 if (mlc->mlc_mirror_id == 0) {
7529 lod_foreach_mirror_comp(lod_comp, lo, primary) {
7530 if (!lod_comp_inited(lod_comp))
7533 extent.e_end = lod_comp->llc_extent.e_end;
7536 rc = lod_prepare_resync(env, lo, &extent);
7540 /* mirror write, try to init its all components */
7541 rc = lod_prepare_resync_mirror(env, lo,
7542 mlc->mlc_mirror_id);
7547 /* change the file state to SYNC_PENDING */
7548 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
7551 rc = lod_declare_instantiate_components(env, lo, th);
7555 /* 3. transfer layout version to OST objects.
7556 * transfer new layout version to OST objects so that stale writes
7557 * can be denied. It also ends an era of writing by setting
7558 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
7559 * send write RPC; only resync RPCs could do it. */
7560 layout_attr->la_valid = LA_LAYOUT_VERSION;
7561 layout_attr->la_layout_version = 0; /* set current version */
7562 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
7563 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
7564 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
7568 lod_obj_inc_layout_gen(lo);
7571 lod_striping_free(env, lo);
7575 static int lod_declare_update_sync_pending(const struct lu_env *env,
7576 struct lod_object *lo, struct md_layout_change *mlc,
7579 struct lod_thread_info *info = lod_env_info(env);
7580 unsigned sync_components = 0;
7581 unsigned resync_components = 0;
7586 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
7587 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
7588 mlc->mlc_opc == MD_LAYOUT_WRITE);
7590 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
7591 PFID(lod_object_fid(lo)), mlc->mlc_opc);
7593 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7594 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
7595 PFID(lod_object_fid(lo)));
7597 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
7598 return lod_declare_update_write_pending(env, lo, mlc, th);
7601 /* MD_LAYOUT_RESYNC_DONE */
7603 for (i = 0; i < lo->ldo_comp_cnt; i++) {
7604 struct lod_layout_component *lod_comp;
7607 lod_comp = &lo->ldo_comp_entries[i];
7609 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
7614 for (j = 0; j < mlc->mlc_resync_count; j++) {
7615 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
7618 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
7619 lod_comp->llc_flags &= ~LCME_FL_STALE;
7620 resync_components++;
7626 for (i = 0; i < mlc->mlc_resync_count; i++) {
7627 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
7630 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
7631 "or already synced\n", PFID(lod_object_fid(lo)),
7632 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
7633 GOTO(out, rc = -EINVAL);
7636 if (!sync_components || (mlc->mlc_resync_count && !resync_components)) {
7637 CDEBUG(D_LAYOUT, DFID": no mirror in sync\n",
7638 PFID(lod_object_fid(lo)));
7640 /* tend to return an error code here to prevent
7641 * the MDT from setting SoM attribute */
7642 GOTO(out, rc = -EINVAL);
7645 CDEBUG(D_LAYOUT, DFID": synced %u resynced %u/%zu components\n",
7646 PFID(lod_object_fid(lo)),
7647 sync_components, resync_components, mlc->mlc_resync_count);
7649 lo->ldo_flr_state = LCM_FL_RDONLY;
7650 lod_obj_inc_layout_gen(lo);
7652 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
7653 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
7654 &info->lti_buf, XATTR_NAME_LOV, 0, th);
7659 lod_striping_free(env, lo);
7663 typedef int (*mlc_handler)(const struct lu_env *env, struct dt_object *dt,
7664 const struct md_layout_change *mlc,
7665 struct thandle *th);
7668 * Attach stripes after target's for migrating directory. NB, we
7669 * only need to declare this, the actual work is done inside
7670 * lod_xattr_set_lmv().
7672 * \param[in] env execution environment
7673 * \param[in] dt target object
7674 * \param[in] mlc layout change data
7675 * \param[in] th transaction handle
7677 * \retval 0 on success
7678 * \retval negative if failed
7680 static int lod_dir_declare_layout_attach(const struct lu_env *env,
7681 struct dt_object *dt,
7682 const struct md_layout_change *mlc,
7685 struct lod_thread_info *info = lod_env_info(env);
7686 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
7687 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
7688 struct lod_object *lo = lod_dt_obj(dt);
7689 struct dt_object *next = dt_object_child(dt);
7690 struct dt_object_format *dof = &info->lti_format;
7691 struct lmv_mds_md_v1 *lmv = mlc->mlc_buf.lb_buf;
7692 struct dt_object **stripes;
7693 __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
7694 struct lu_fid *fid = &info->lti_fid;
7695 struct lod_tgt_desc *tgt;
7696 struct dt_object *dto;
7697 struct dt_device *tgt_dt;
7698 int type = LU_SEQ_RANGE_ANY;
7699 struct dt_insert_rec *rec = &info->lti_dt_rec;
7700 char *stripe_name = info->lti_key;
7701 struct lu_name *sname;
7702 struct linkea_data ldata = { NULL };
7703 struct lu_buf linkea_buf;
7710 if (!lmv_is_sane(lmv))
7713 if (!dt_try_as_dir(env, dt))
7716 dof->dof_type = DFT_DIR;
7718 OBD_ALLOC_PTR_ARRAY(stripes, (lo->ldo_dir_stripe_count + stripe_count));
7722 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
7723 stripes[i] = lo->ldo_stripe[i];
7725 rec->rec_type = S_IFDIR;
7727 for (i = 0; i < stripe_count; i++) {
7729 &lmv->lmv_stripe_fids[i]);
7730 if (!fid_is_sane(fid))
7733 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
7737 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
7738 tgt_dt = lod->lod_child;
7740 tgt = LTD_TGT(ltd, idx);
7742 GOTO(out, rc = -ESTALE);
7743 tgt_dt = tgt->ltd_tgt;
7746 dto = dt_locate_at(env, tgt_dt, fid,
7747 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
7750 GOTO(out, rc = PTR_ERR(dto));
7752 stripes[i + lo->ldo_dir_stripe_count] = dto;
7754 if (!dt_try_as_dir(env, dto))
7755 GOTO(out, rc = -ENOTDIR);
7757 rc = lod_sub_declare_ref_add(env, dto, th);
7761 rec->rec_fid = lu_object_fid(&dto->do_lu);
7762 rc = lod_sub_declare_insert(env, dto,
7763 (const struct dt_rec *)rec,
7764 (const struct dt_key *)dot, th);
7768 rc = lod_sub_declare_insert(env, dto,
7769 (const struct dt_rec *)rec,
7770 (const struct dt_key *)dotdot, th);
7774 rc = lod_sub_declare_xattr_set(env, dto, &mlc->mlc_buf,
7775 XATTR_NAME_LMV, 0, th);
7779 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
7780 PFID(lu_object_fid(&dto->do_lu)),
7781 i + lo->ldo_dir_stripe_count);
7783 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
7784 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
7785 sname, lu_object_fid(&dt->do_lu));
7789 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
7790 linkea_buf.lb_len = ldata.ld_leh->leh_len;
7791 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
7792 XATTR_NAME_LINK, 0, th);
7796 rc = lod_sub_declare_insert(env, next,
7797 (const struct dt_rec *)rec,
7798 (const struct dt_key *)stripe_name,
7803 rc = lod_sub_declare_ref_add(env, next, th);
7809 OBD_FREE_PTR_ARRAY(lo->ldo_stripe,
7810 lo->ldo_dir_stripes_allocated);
7811 lo->ldo_stripe = stripes;
7812 lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count;
7813 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type);
7814 lo->ldo_dir_stripe_count += stripe_count;
7815 lo->ldo_dir_stripes_allocated += stripe_count;
7817 /* plain directory split creates target as a plain directory, while
7818 * after source attached as the first stripe, it becomes a striped
7819 * directory, set correct do_index_ops, otherwise it can't be unlinked.
7821 dt->do_index_ops = &lod_striped_index_ops;
7825 i = lo->ldo_dir_stripe_count;
7826 while (i < lo->ldo_dir_stripe_count + stripe_count && stripes[i])
7827 dt_object_put(env, stripes[i++]);
7829 OBD_FREE_PTR_ARRAY(stripes, stripe_count + lo->ldo_dir_stripe_count);
7833 static int lod_dir_declare_layout_detach(const struct lu_env *env,
7834 struct dt_object *dt,
7835 const struct md_layout_change *unused,
7838 struct lod_thread_info *info = lod_env_info(env);
7839 struct lod_object *lo = lod_dt_obj(dt);
7840 struct dt_object *next = dt_object_child(dt);
7841 char *stripe_name = info->lti_key;
7842 struct dt_object *dto;
7846 if (!dt_try_as_dir(env, dt))
7849 if (!lo->ldo_dir_stripe_count)
7850 return lod_sub_declare_delete(env, next,
7851 (const struct dt_key *)dotdot, th);
7853 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
7854 dto = lo->ldo_stripe[i];
7858 if (!dt_try_as_dir(env, dto))
7861 rc = lod_sub_declare_delete(env, dto,
7862 (const struct dt_key *)dotdot, th);
7866 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
7867 PFID(lu_object_fid(&dto->do_lu)), i);
7869 rc = lod_sub_declare_delete(env, next,
7870 (const struct dt_key *)stripe_name, th);
7874 rc = lod_sub_declare_ref_del(env, next, th);
7882 static int dt_dir_is_empty(const struct lu_env *env,
7883 struct dt_object *obj)
7886 const struct dt_it_ops *iops;
7891 if (!dt_try_as_dir(env, obj))
7894 iops = &obj->do_index_ops->dio_it;
7895 it = iops->init(env, obj, LUDA_64BITHASH);
7897 RETURN(PTR_ERR(it));
7899 rc = iops->get(env, it, (const struct dt_key *)"");
7903 for (rc = 0, i = 0; rc == 0 && i < 3; ++i)
7904 rc = iops->next(env, it);
7910 /* Huh? Index contains no zero key? */
7915 iops->fini(env, it);
7920 static int lod_dir_declare_layout_shrink(const struct lu_env *env,
7921 struct dt_object *dt,
7922 const struct md_layout_change *mlc,
7925 struct lod_thread_info *info = lod_env_info(env);
7926 struct lod_object *lo = lod_dt_obj(dt);
7927 struct dt_object *next = dt_object_child(dt);
7928 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
7929 __u32 final_stripe_count;
7930 char *stripe_name = info->lti_key;
7931 struct lu_buf *lmv_buf = &info->lti_buf;
7932 struct dt_object *dto;
7938 if (!dt_try_as_dir(env, dt))
7941 /* shouldn't be called on plain directory */
7942 LASSERT(lo->ldo_dir_stripe_count);
7944 lmv_buf->lb_buf = &info->lti_lmv.lmv_md_v1;
7945 lmv_buf->lb_len = sizeof(info->lti_lmv.lmv_md_v1);
7947 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
7948 LASSERT(final_stripe_count &&
7949 final_stripe_count < lo->ldo_dir_stripe_count);
7951 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
7952 dto = lo->ldo_stripe[i];
7956 if (i < final_stripe_count) {
7957 if (final_stripe_count == 1)
7960 rc = lod_sub_declare_xattr_set(env, dto, lmv_buf,
7962 LU_XATTR_REPLACE, th);
7969 rc = dt_dir_is_empty(env, dto);
7973 rc = lod_sub_declare_ref_del(env, dto, th);
7977 rc = lod_sub_declare_destroy(env, dto, th);
7981 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
7982 PFID(lu_object_fid(&dto->do_lu)), i);
7984 rc = lod_sub_declare_delete(env, next,
7985 (const struct dt_key *)stripe_name, th);
7989 rc = lod_sub_declare_ref_del(env, next, th);
7994 rc = lod_sub_declare_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
7995 LU_XATTR_REPLACE, th);
8000 * Allocate stripes for split directory.
8002 * \param[in] env execution environment
8003 * \param[in] dt target object
8004 * \param[in] mlc layout change data
8005 * \param[in] th transaction handle
8007 * \retval 0 on success
8008 * \retval negative if failed
8010 static int lod_dir_declare_layout_split(const struct lu_env *env,
8011 struct dt_object *dt,
8012 const struct md_layout_change *mlc,
8015 struct lod_thread_info *info = lod_env_info(env);
8016 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
8017 struct lod_object *lo = lod_dt_obj(dt);
8018 struct dt_object_format *dof = &info->lti_format;
8019 struct lmv_user_md_v1 *lum = mlc->mlc_spec->u.sp_ea.eadata;
8020 struct dt_object **stripes;
8028 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
8029 LASSERT(le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT);
8031 saved_count = lo->ldo_dir_stripes_allocated;
8032 stripe_count = le32_to_cpu(lum->lum_stripe_count);
8033 if (stripe_count <= saved_count)
8036 dof->dof_type = DFT_DIR;
8038 OBD_ALLOC(stripes, sizeof(*stripes) * stripe_count);
8042 for (i = 0; i < lo->ldo_dir_stripes_allocated; i++)
8043 stripes[i] = lo->ldo_stripe[i];
8045 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
8046 rc = lod_mdt_alloc_qos(env, lo, stripes, saved_count, stripe_count);
8048 rc = lod_mdt_alloc_rr(env, lo, stripes, saved_count,
8051 OBD_FREE(stripes, sizeof(*stripes) * stripe_count);
8055 LASSERT(rc > saved_count);
8056 OBD_FREE(lo->ldo_stripe,
8057 sizeof(*stripes) * lo->ldo_dir_stripes_allocated);
8058 lo->ldo_stripe = stripes;
8059 lo->ldo_dir_striped = 1;
8060 lo->ldo_dir_stripe_count = rc;
8061 lo->ldo_dir_stripes_allocated = stripe_count;
8062 lo->ldo_dir_split_hash = lo->ldo_dir_hash_type;
8063 lo->ldo_dir_hash_type = le32_to_cpu(lum->lum_hash_type);
8064 if (!lmv_is_known_hash_type(lo->ldo_dir_hash_type))
8065 lo->ldo_dir_hash_type =
8066 lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
8067 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_SPLIT | LMV_HASH_FLAG_MIGRATION;
8068 lo->ldo_dir_split_offset = saved_count;
8069 lo->ldo_dir_layout_version++;
8070 lo->ldo_dir_stripe_loaded = 1;
8072 rc = lod_dir_declare_create_stripes(env, dt, mlc->mlc_attr, dof, th);
8074 lod_striping_free(env, lo);
8080 * detach all stripes from dir master object, NB, stripes are not destroyed, but
8081 * deleted from it's parent namespace, this function is called in two places:
8082 * 1. mdd_migrate_mdt() detach stripes from source, and attach them to
8084 * 2. mdd_dir_layout_update() detach stripe before turning 1-stripe directory to
8085 * a plain directory.
8087 * \param[in] env execution environment
8088 * \param[in] dt target object
8089 * \param[in] mlc layout change data
8090 * \param[in] th transaction handle
8092 * \retval 0 on success
8093 * \retval negative if failed
8095 static int lod_dir_layout_detach(const struct lu_env *env,
8096 struct dt_object *dt,
8097 const struct md_layout_change *mlc,
8100 struct lod_thread_info *info = lod_env_info(env);
8101 struct lod_object *lo = lod_dt_obj(dt);
8102 struct dt_object *next = dt_object_child(dt);
8103 char *stripe_name = info->lti_key;
8104 struct dt_object *dto;
8110 if (!lo->ldo_dir_stripe_count) {
8111 /* plain directory delete .. */
8112 rc = lod_sub_delete(env, next,
8113 (const struct dt_key *)dotdot, th);
8117 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8118 dto = lo->ldo_stripe[i];
8122 rc = lod_sub_delete(env, dto,
8123 (const struct dt_key *)dotdot, th);
8127 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8128 PFID(lu_object_fid(&dto->do_lu)), i);
8130 rc = lod_sub_delete(env, next,
8131 (const struct dt_key *)stripe_name, th);
8135 rc = lod_sub_ref_del(env, next, th);
8140 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8141 dto = lo->ldo_stripe[i];
8143 dt_object_put(env, dto);
8145 OBD_FREE_PTR_ARRAY(lo->ldo_stripe, lo->ldo_dir_stripes_allocated);
8146 lo->ldo_stripe = NULL;
8147 lo->ldo_dir_stripes_allocated = 0;
8148 lo->ldo_dir_stripe_count = 0;
8153 static int lod_dir_layout_shrink(const struct lu_env *env,
8154 struct dt_object *dt,
8155 const struct md_layout_change *mlc,
8158 struct lod_thread_info *info = lod_env_info(env);
8159 struct lod_object *lo = lod_dt_obj(dt);
8160 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
8161 struct dt_object *next = dt_object_child(dt);
8162 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
8163 __u32 final_stripe_count;
8164 char *stripe_name = info->lti_key;
8165 struct dt_object *dto;
8166 struct lu_buf *lmv_buf = &info->lti_buf;
8167 struct lmv_mds_md_v1 *lmv = &info->lti_lmv.lmv_md_v1;
8169 int type = LU_SEQ_RANGE_ANY;
8175 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
8177 lmv_buf->lb_buf = lmv;
8178 lmv_buf->lb_len = sizeof(*lmv);
8179 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
8180 lmv->lmv_stripe_count = cpu_to_le32(final_stripe_count);
8181 lmv->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type) &
8182 cpu_to_le32(LMV_HASH_TYPE_MASK);
8183 lmv->lmv_layout_version =
8184 cpu_to_le32(lo->ldo_dir_layout_version + 1);
8186 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8187 dto = lo->ldo_stripe[i];
8191 if (i < final_stripe_count) {
8192 /* if only one stripe left, no need to update
8193 * LMV because this stripe will replace master
8194 * object and act as a plain directory.
8196 if (final_stripe_count == 1)
8200 rc = lod_fld_lookup(env, lod,
8201 lu_object_fid(&dto->do_lu),
8206 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
8207 rc = lod_sub_xattr_set(env, dto, lmv_buf,
8209 LU_XATTR_REPLACE, th);
8216 dt_write_lock(env, dto, DT_TGT_CHILD);
8217 rc = lod_sub_ref_del(env, dto, th);
8218 dt_write_unlock(env, dto);
8222 rc = lod_sub_destroy(env, dto, th);
8226 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8227 PFID(lu_object_fid(&dto->do_lu)), i);
8229 rc = lod_sub_delete(env, next,
8230 (const struct dt_key *)stripe_name, th);
8234 rc = lod_sub_ref_del(env, next, th);
8239 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &mdtidx,
8244 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
8245 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
8246 rc = lod_sub_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
8247 LU_XATTR_REPLACE, th);
8251 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
8252 dto = lo->ldo_stripe[i];
8254 dt_object_put(env, dto);
8256 lo->ldo_dir_stripe_count = final_stripe_count;
8261 static mlc_handler dir_mlc_declare_ops[MD_LAYOUT_MAX] = {
8262 [MD_LAYOUT_ATTACH] = lod_dir_declare_layout_attach,
8263 [MD_LAYOUT_DETACH] = lod_dir_declare_layout_detach,
8264 [MD_LAYOUT_SHRINK] = lod_dir_declare_layout_shrink,
8265 [MD_LAYOUT_SPLIT] = lod_dir_declare_layout_split,
8268 static mlc_handler dir_mlc_ops[MD_LAYOUT_MAX] = {
8269 [MD_LAYOUT_DETACH] = lod_dir_layout_detach,
8270 [MD_LAYOUT_SHRINK] = lod_dir_layout_shrink,
8273 static int lod_declare_layout_change(const struct lu_env *env,
8274 struct dt_object *dt, struct md_layout_change *mlc,
8277 struct lod_thread_info *info = lod_env_info(env);
8278 struct lod_object *lo = lod_dt_obj(dt);
8283 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
8284 LASSERT(dir_mlc_declare_ops[mlc->mlc_opc]);
8285 rc = dir_mlc_declare_ops[mlc->mlc_opc](env, dt, mlc, th);
8289 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
8290 dt_object_remote(dt_object_child(dt)))
8293 rc = lod_striping_load(env, lo);
8297 LASSERT(lo->ldo_comp_cnt > 0);
8299 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
8303 switch (lo->ldo_flr_state) {
8305 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
8309 rc = lod_declare_update_rdonly(env, lo, mlc, th);
8311 case LCM_FL_WRITE_PENDING:
8312 rc = lod_declare_update_write_pending(env, lo, mlc, th);
8314 case LCM_FL_SYNC_PENDING:
8315 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
8326 * Instantiate layout component objects which covers the intent write offset.
8328 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
8329 struct md_layout_change *mlc, struct thandle *th)
8331 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
8332 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
8333 struct lod_object *lo = lod_dt_obj(dt);
8338 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
8339 LASSERT(dir_mlc_ops[mlc->mlc_opc]);
8340 rc = dir_mlc_ops[mlc->mlc_opc](env, dt, mlc, th);
8344 rc = lod_striped_create(env, dt, attr, NULL, th);
8345 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
8346 layout_attr->la_layout_version |= lo->ldo_layout_gen;
8347 rc = lod_attr_set(env, dt, layout_attr, th);
8353 struct dt_object_operations lod_obj_ops = {
8354 .do_read_lock = lod_read_lock,
8355 .do_write_lock = lod_write_lock,
8356 .do_read_unlock = lod_read_unlock,
8357 .do_write_unlock = lod_write_unlock,
8358 .do_write_locked = lod_write_locked,
8359 .do_attr_get = lod_attr_get,
8360 .do_declare_attr_set = lod_declare_attr_set,
8361 .do_attr_set = lod_attr_set,
8362 .do_xattr_get = lod_xattr_get,
8363 .do_declare_xattr_set = lod_declare_xattr_set,
8364 .do_xattr_set = lod_xattr_set,
8365 .do_declare_xattr_del = lod_declare_xattr_del,
8366 .do_xattr_del = lod_xattr_del,
8367 .do_xattr_list = lod_xattr_list,
8368 .do_ah_init = lod_ah_init,
8369 .do_declare_create = lod_declare_create,
8370 .do_create = lod_create,
8371 .do_declare_destroy = lod_declare_destroy,
8372 .do_destroy = lod_destroy,
8373 .do_index_try = lod_index_try,
8374 .do_declare_ref_add = lod_declare_ref_add,
8375 .do_ref_add = lod_ref_add,
8376 .do_declare_ref_del = lod_declare_ref_del,
8377 .do_ref_del = lod_ref_del,
8378 .do_object_sync = lod_object_sync,
8379 .do_object_lock = lod_object_lock,
8380 .do_object_unlock = lod_object_unlock,
8381 .do_invalidate = lod_invalidate,
8382 .do_declare_layout_change = lod_declare_layout_change,
8383 .do_layout_change = lod_layout_change,
8387 * Implementation of dt_body_operations::dbo_read.
8389 * \see dt_body_operations::dbo_read() in the API description for details.
8391 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
8392 struct lu_buf *buf, loff_t *pos)
8394 struct dt_object *next = dt_object_child(dt);
8396 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
8397 S_ISLNK(dt->do_lu.lo_header->loh_attr));
8398 return next->do_body_ops->dbo_read(env, next, buf, pos);
8402 * Implementation of dt_body_operations::dbo_declare_write.
8404 * \see dt_body_operations::dbo_declare_write() in the API description
8407 static ssize_t lod_declare_write(const struct lu_env *env,
8408 struct dt_object *dt,
8409 const struct lu_buf *buf, loff_t pos,
8412 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
8416 * Implementation of dt_body_operations::dbo_write.
8418 * \see dt_body_operations::dbo_write() in the API description for details.
8420 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
8421 const struct lu_buf *buf, loff_t *pos,
8424 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
8425 S_ISLNK(dt->do_lu.lo_header->loh_attr));
8426 return lod_sub_write(env, dt_object_child(dt), buf, pos, th);
8429 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
8430 __u64 start, __u64 end, struct thandle *th)
8432 if (dt_object_remote(dt))
8435 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
8438 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
8439 __u64 start, __u64 end, struct thandle *th)
8441 if (dt_object_remote(dt))
8444 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
8445 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
8449 * different type of files use the same body_ops because object may be created
8450 * in OUT, where there is no chance to set correct body_ops for each type, so
8451 * body_ops themselves will check file type inside, see lod_read/write/punch for
8454 const struct dt_body_operations lod_body_ops = {
8455 .dbo_read = lod_read,
8456 .dbo_declare_write = lod_declare_write,
8457 .dbo_write = lod_write,
8458 .dbo_declare_punch = lod_declare_punch,
8459 .dbo_punch = lod_punch,
8463 * Implementation of lu_object_operations::loo_object_init.
8465 * The function determines the type and the index of the target device using
8466 * sequence of the object's FID. Then passes control down to the
8467 * corresponding device:
8468 * OSD for the local objects, OSP for remote
8470 * \see lu_object_operations::loo_object_init() in the API description
8473 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
8474 const struct lu_object_conf *conf)
8476 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
8477 struct lu_device *cdev = NULL;
8478 struct lu_object *cobj;
8479 struct lod_tgt_descs *ltd = NULL;
8480 struct lod_tgt_desc *tgt;
8482 int type = LU_SEQ_RANGE_ANY;
8486 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
8490 if (type == LU_SEQ_RANGE_MDT &&
8491 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
8492 cdev = &lod->lod_child->dd_lu_dev;
8493 } else if (type == LU_SEQ_RANGE_MDT) {
8494 ltd = &lod->lod_mdt_descs;
8496 } else if (type == LU_SEQ_RANGE_OST) {
8497 ltd = &lod->lod_ost_descs;
8504 if (ltd->ltd_tgts_size > idx &&
8505 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
8506 tgt = LTD_TGT(ltd, idx);
8508 LASSERT(tgt != NULL);
8509 LASSERT(tgt->ltd_tgt != NULL);
8511 cdev = &(tgt->ltd_tgt->dd_lu_dev);
8513 lod_putref(lod, ltd);
8516 if (unlikely(cdev == NULL))
8519 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
8520 if (unlikely(cobj == NULL))
8523 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
8525 lu_object_add(lo, cobj);
8532 * Alloc cached foreign LOV
8534 * \param[in] lo object
8535 * \param[in] size size of foreign LOV
8537 * \retval 0 on success
8538 * \retval negative if failed
8540 int lod_alloc_foreign_lov(struct lod_object *lo, size_t size)
8542 OBD_ALLOC_LARGE(lo->ldo_foreign_lov, size);
8543 if (lo->ldo_foreign_lov == NULL)
8545 lo->ldo_foreign_lov_size = size;
8546 lo->ldo_is_foreign = 1;
8552 * Free cached foreign LOV
8554 * \param[in] lo object
8556 void lod_free_foreign_lov(struct lod_object *lo)
8558 if (lo->ldo_foreign_lov != NULL)
8559 OBD_FREE_LARGE(lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
8560 lo->ldo_foreign_lov = NULL;
8561 lo->ldo_foreign_lov_size = 0;
8562 lo->ldo_is_foreign = 0;
8567 * Free cached foreign LMV
8569 * \param[in] lo object
8571 void lod_free_foreign_lmv(struct lod_object *lo)
8573 if (lo->ldo_foreign_lmv != NULL)
8574 OBD_FREE_LARGE(lo->ldo_foreign_lmv, lo->ldo_foreign_lmv_size);
8575 lo->ldo_foreign_lmv = NULL;
8576 lo->ldo_foreign_lmv_size = 0;
8577 lo->ldo_dir_is_foreign = 0;
8582 * Release resources associated with striping.
8584 * If the object is striped (regular or directory), then release
8585 * the stripe objects references and free the ldo_stripe array.
8587 * \param[in] env execution environment
8588 * \param[in] lo object
8590 void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo)
8592 struct lod_layout_component *lod_comp;
8595 if (unlikely(lo->ldo_is_foreign)) {
8596 lod_free_foreign_lov(lo);
8597 lo->ldo_comp_cached = 0;
8598 } else if (unlikely(lo->ldo_dir_is_foreign)) {
8599 lod_free_foreign_lmv(lo);
8600 lo->ldo_dir_stripe_loaded = 0;
8601 } else if (lo->ldo_stripe != NULL) {
8602 LASSERT(lo->ldo_comp_entries == NULL);
8603 LASSERT(lo->ldo_dir_stripes_allocated > 0);
8605 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8606 if (lo->ldo_stripe[i])
8607 dt_object_put(env, lo->ldo_stripe[i]);
8610 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
8611 OBD_FREE(lo->ldo_stripe, j);
8612 lo->ldo_stripe = NULL;
8613 lo->ldo_dir_stripes_allocated = 0;
8614 lo->ldo_dir_stripe_loaded = 0;
8615 lo->ldo_dir_stripe_count = 0;
8616 } else if (lo->ldo_comp_entries != NULL) {
8617 for (i = 0; i < lo->ldo_comp_cnt; i++) {
8618 /* free lod_layout_component::llc_stripe array */
8619 lod_comp = &lo->ldo_comp_entries[i];
8621 if (lod_comp->llc_stripe == NULL)
8623 LASSERT(lod_comp->llc_stripes_allocated != 0);
8624 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
8625 if (lod_comp->llc_stripe[j] != NULL)
8627 &lod_comp->llc_stripe[j]->do_lu);
8629 OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe,
8630 lod_comp->llc_stripes_allocated);
8631 lod_comp->llc_stripe = NULL;
8632 OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices,
8633 lod_comp->llc_stripes_allocated);
8634 lod_comp->llc_ost_indices = NULL;
8635 lod_comp->llc_stripes_allocated = 0;
8637 lod_free_comp_entries(lo);
8638 lo->ldo_comp_cached = 0;
8642 void lod_striping_free(const struct lu_env *env, struct lod_object *lo)
8644 mutex_lock(&lo->ldo_layout_mutex);
8645 lod_striping_free_nolock(env, lo);
8646 mutex_unlock(&lo->ldo_layout_mutex);
8650 * Implementation of lu_object_operations::loo_object_free.
8652 * \see lu_object_operations::loo_object_free() in the API description
8655 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
8657 struct lod_object *lo = lu2lod_obj(o);
8659 /* release all underlying object pinned */
8660 lod_striping_free(env, lo);
8662 /* lo doesn't contain a lu_object_header, so we don't need call_rcu */
8663 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
8667 * Implementation of lu_object_operations::loo_object_release.
8669 * \see lu_object_operations::loo_object_release() in the API description
8672 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
8674 /* XXX: shouldn't we release everything here in case if object
8675 * creation failed before? */
8679 * Implementation of lu_object_operations::loo_object_print.
8681 * \see lu_object_operations::loo_object_print() in the API description
8684 static int lod_object_print(const struct lu_env *env, void *cookie,
8685 lu_printer_t p, const struct lu_object *l)
8687 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
8689 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
8692 struct lu_object_operations lod_lu_obj_ops = {
8693 .loo_object_init = lod_object_init,
8694 .loo_object_free = lod_object_free,
8695 .loo_object_release = lod_object_release,
8696 .loo_object_print = lod_object_print,