4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * lustre/lod/lod_object.c
31 * This file contains implementations of methods for the OSD API
32 * for the Logical Object Device (LOD) layer, which provides a virtual
33 * local OSD object interface to the MDD layer, and abstracts the
34 * addressing of local (OSD) and remote (OSP) objects. The API is
35 * described in the file lustre/include/dt_object.h and in
36 * Documentation/osd-api.txt.
38 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
41 #define DEBUG_SUBSYSTEM S_MDS
43 #include <linux/random.h>
46 #include <obd_class.h>
47 #include <obd_support.h>
49 #include <lustre_fid.h>
50 #include <lustre_linkea.h>
51 #include <lustre_lmv.h>
52 #include <uapi/linux/lustre/lustre_param.h>
53 #include <lustre_swab.h>
54 #include <uapi/linux/lustre/lustre_ver.h>
55 #include <lprocfs_status.h>
56 #include <md_object.h>
58 #include "lod_internal.h"
60 static const char dot[] = ".";
61 static const char dotdot[] = "..";
64 * Implementation of dt_index_operations::dio_lookup
66 * Used with regular (non-striped) objects.
68 * \see dt_index_operations::dio_lookup() in the API description for details.
70 static int lod_lookup(const struct lu_env *env, struct dt_object *dt,
71 struct dt_rec *rec, const struct dt_key *key)
73 struct dt_object *next = dt_object_child(dt);
74 return next->do_index_ops->dio_lookup(env, next, rec, key);
78 * Implementation of dt_index_operations::dio_declare_insert.
80 * Used with regular (non-striped) objects.
82 * \see dt_index_operations::dio_declare_insert() in the API description
85 static int lod_declare_insert(const struct lu_env *env, struct dt_object *dt,
86 const struct dt_rec *rec,
87 const struct dt_key *key, struct thandle *th)
89 return lod_sub_declare_insert(env, dt_object_child(dt), rec, key, th);
93 * Implementation of dt_index_operations::dio_insert.
95 * Used with regular (non-striped) objects
97 * \see dt_index_operations::dio_insert() in the API description for details.
99 static int lod_insert(const struct lu_env *env, struct dt_object *dt,
100 const struct dt_rec *rec, const struct dt_key *key,
103 return lod_sub_insert(env, dt_object_child(dt), rec, key, th);
107 * Implementation of dt_index_operations::dio_declare_delete.
109 * Used with regular (non-striped) objects.
111 * \see dt_index_operations::dio_declare_delete() in the API description
114 static int lod_declare_delete(const struct lu_env *env, struct dt_object *dt,
115 const struct dt_key *key, struct thandle *th)
117 return lod_sub_declare_delete(env, dt_object_child(dt), key, th);
121 * Implementation of dt_index_operations::dio_delete.
123 * Used with regular (non-striped) objects.
125 * \see dt_index_operations::dio_delete() in the API description for details.
127 static int lod_delete(const struct lu_env *env, struct dt_object *dt,
128 const struct dt_key *key, struct thandle *th)
130 return lod_sub_delete(env, dt_object_child(dt), key, th);
134 * Implementation of dt_it_ops::init.
136 * Used with regular (non-striped) objects.
138 * \see dt_it_ops::init() in the API description for details.
140 static struct dt_it *lod_it_init(const struct lu_env *env,
141 struct dt_object *dt, __u32 attr)
143 struct dt_object *next = dt_object_child(dt);
144 struct lod_it *it = &lod_env_info(env)->lti_it;
145 struct dt_it *it_next;
147 it_next = next->do_index_ops->dio_it.init(env, next, attr);
151 /* currently we do not use more than one iterator per thread
152 * so we store it in thread info. if at some point we need
153 * more active iterators in a single thread, we can allocate
155 LASSERT(it->lit_obj == NULL);
157 it->lit_it = it_next;
160 return (struct dt_it *)it;
163 #define LOD_CHECK_IT(env, it) \
165 LASSERT((it)->lit_obj != NULL); \
166 LASSERT((it)->lit_it != NULL); \
170 * Implementation of dt_index_operations::dio_it.fini.
172 * Used with regular (non-striped) objects.
174 * \see dt_index_operations::dio_it.fini() in the API description for details.
176 static void lod_it_fini(const struct lu_env *env, struct dt_it *di)
178 struct lod_it *it = (struct lod_it *)di;
180 LOD_CHECK_IT(env, it);
181 it->lit_obj->do_index_ops->dio_it.fini(env, it->lit_it);
183 /* the iterator not in use any more */
189 * Implementation of dt_it_ops::get.
191 * Used with regular (non-striped) objects.
193 * \see dt_it_ops::get() in the API description for details.
195 static int lod_it_get(const struct lu_env *env, struct dt_it *di,
196 const struct dt_key *key)
198 const struct lod_it *it = (const struct lod_it *)di;
200 LOD_CHECK_IT(env, it);
201 return it->lit_obj->do_index_ops->dio_it.get(env, it->lit_it, key);
205 * Implementation of dt_it_ops::put.
207 * Used with regular (non-striped) objects.
209 * \see dt_it_ops::put() in the API description for details.
211 static void lod_it_put(const struct lu_env *env, struct dt_it *di)
213 struct lod_it *it = (struct lod_it *)di;
215 LOD_CHECK_IT(env, it);
216 return it->lit_obj->do_index_ops->dio_it.put(env, it->lit_it);
220 * Implementation of dt_it_ops::next.
222 * Used with regular (non-striped) objects
224 * \see dt_it_ops::next() in the API description for details.
226 static int lod_it_next(const struct lu_env *env, struct dt_it *di)
228 struct lod_it *it = (struct lod_it *)di;
230 LOD_CHECK_IT(env, it);
231 return it->lit_obj->do_index_ops->dio_it.next(env, it->lit_it);
235 * Implementation of dt_it_ops::key.
237 * Used with regular (non-striped) objects.
239 * \see dt_it_ops::key() in the API description for details.
241 static struct dt_key *lod_it_key(const struct lu_env *env,
242 const struct dt_it *di)
244 const struct lod_it *it = (const struct lod_it *)di;
246 LOD_CHECK_IT(env, it);
247 return it->lit_obj->do_index_ops->dio_it.key(env, it->lit_it);
251 * Implementation of dt_it_ops::key_size.
253 * Used with regular (non-striped) objects.
255 * \see dt_it_ops::key_size() in the API description for details.
257 static int lod_it_key_size(const struct lu_env *env, const struct dt_it *di)
259 struct lod_it *it = (struct lod_it *)di;
261 LOD_CHECK_IT(env, it);
262 return it->lit_obj->do_index_ops->dio_it.key_size(env, it->lit_it);
266 * Implementation of dt_it_ops::rec.
268 * Used with regular (non-striped) objects.
270 * \see dt_it_ops::rec() in the API description for details.
272 static int lod_it_rec(const struct lu_env *env, const struct dt_it *di,
273 struct dt_rec *rec, __u32 attr)
275 const struct lod_it *it = (const struct lod_it *)di;
277 LOD_CHECK_IT(env, it);
278 return it->lit_obj->do_index_ops->dio_it.rec(env, it->lit_it, rec,
283 * Implementation of dt_it_ops::rec_size.
285 * Used with regular (non-striped) objects.
287 * \see dt_it_ops::rec_size() in the API description for details.
289 static int lod_it_rec_size(const struct lu_env *env, const struct dt_it *di,
292 const struct lod_it *it = (const struct lod_it *)di;
294 LOD_CHECK_IT(env, it);
295 return it->lit_obj->do_index_ops->dio_it.rec_size(env, it->lit_it,
300 * Implementation of dt_it_ops::store.
302 * Used with regular (non-striped) objects.
304 * \see dt_it_ops::store() in the API description for details.
306 static __u64 lod_it_store(const struct lu_env *env, const struct dt_it *di)
308 const struct lod_it *it = (const struct lod_it *)di;
310 LOD_CHECK_IT(env, it);
311 return it->lit_obj->do_index_ops->dio_it.store(env, it->lit_it);
315 * Implementation of dt_it_ops::load.
317 * Used with regular (non-striped) objects.
319 * \see dt_it_ops::load() in the API description for details.
321 static int lod_it_load(const struct lu_env *env, const struct dt_it *di,
324 const struct lod_it *it = (const struct lod_it *)di;
326 LOD_CHECK_IT(env, it);
327 return it->lit_obj->do_index_ops->dio_it.load(env, it->lit_it, hash);
331 * Implementation of dt_it_ops::key_rec.
333 * Used with regular (non-striped) objects.
335 * \see dt_it_ops::rec() in the API description for details.
337 static int lod_it_key_rec(const struct lu_env *env, const struct dt_it *di,
340 const struct lod_it *it = (const struct lod_it *)di;
342 LOD_CHECK_IT(env, it);
343 return it->lit_obj->do_index_ops->dio_it.key_rec(env, it->lit_it,
347 static const struct dt_index_operations lod_index_ops = {
348 .dio_lookup = lod_lookup,
349 .dio_declare_insert = lod_declare_insert,
350 .dio_insert = lod_insert,
351 .dio_declare_delete = lod_declare_delete,
352 .dio_delete = lod_delete,
360 .key_size = lod_it_key_size,
362 .rec_size = lod_it_rec_size,
363 .store = lod_it_store,
365 .key_rec = lod_it_key_rec,
370 * Implementation of dt_index_operations::dio_lookup
372 * Used with striped directories.
374 * \see dt_index_operations::dio_lookup() in the API description for details.
376 static int lod_striped_lookup(const struct lu_env *env, struct dt_object *dt,
377 struct dt_rec *rec, const struct dt_key *key)
379 struct lod_object *lo = lod_dt_obj(dt);
380 struct dt_object *next;
381 const char *name = (const char *)key;
383 LASSERT(lo->ldo_dir_stripe_count > 0);
385 if (strcmp(name, dot) == 0) {
386 struct lu_fid *fid = (struct lu_fid *)rec;
388 *fid = *lod_object_fid(lo);
392 if (strcmp(name, dotdot) == 0) {
393 next = dt_object_child(dt);
397 index = __lmv_name_to_stripe_index(lo->ldo_dir_hash_type,
398 lo->ldo_dir_stripe_count,
399 lo->ldo_dir_migrate_hash,
400 lo->ldo_dir_migrate_offset,
401 name, strlen(name), true);
405 next = lo->ldo_stripe[index];
406 if (!next || !dt_object_exists(next))
410 return next->do_index_ops->dio_lookup(env, next, rec, key);
414 * Implementation of dt_it_ops::init.
416 * Used with striped objects. Internally just initializes the iterator
417 * on the first stripe.
419 * \see dt_it_ops::init() in the API description for details.
421 static struct dt_it *lod_striped_it_init(const struct lu_env *env,
422 struct dt_object *dt, __u32 attr)
424 struct lod_object *lo = lod_dt_obj(dt);
425 struct dt_object *next;
426 struct lod_it *it = &lod_env_info(env)->lti_it;
427 struct dt_it *it_next;
430 LASSERT(lo->ldo_dir_stripe_count > 0);
433 next = lo->ldo_stripe[index];
434 if (next && dt_object_exists(next))
436 } while (++index < lo->ldo_dir_stripe_count);
438 /* no valid stripe */
439 if (!next || !dt_object_exists(next))
440 return ERR_PTR(-ENODEV);
442 LASSERT(next->do_index_ops != NULL);
444 it_next = next->do_index_ops->dio_it.init(env, next, attr);
448 /* currently we do not use more than one iterator per thread
449 * so we store it in thread info. if at some point we need
450 * more active iterators in a single thread, we can allocate
452 LASSERT(it->lit_obj == NULL);
454 it->lit_stripe_index = index;
456 it->lit_it = it_next;
459 return (struct dt_it *)it;
462 #define LOD_CHECK_STRIPED_IT(env, it, lo) \
464 LASSERT((it)->lit_obj != NULL); \
465 LASSERT((it)->lit_it != NULL); \
466 LASSERT((lo)->ldo_dir_stripe_count > 0); \
467 LASSERT((it)->lit_stripe_index < (lo)->ldo_dir_stripe_count); \
471 * Implementation of dt_it_ops::fini.
473 * Used with striped objects.
475 * \see dt_it_ops::fini() in the API description for details.
477 static void lod_striped_it_fini(const struct lu_env *env, struct dt_it *di)
479 struct lod_it *it = (struct lod_it *)di;
480 struct lod_object *lo = lod_dt_obj(it->lit_obj);
481 struct dt_object *next;
483 /* If lit_it == NULL, then it means the sub_it has been finished,
484 * which only happens in failure cases, see lod_striped_it_next() */
485 if (it->lit_it != NULL) {
486 LOD_CHECK_STRIPED_IT(env, it, lo);
488 next = lo->ldo_stripe[it->lit_stripe_index];
490 LASSERT(next->do_index_ops != NULL);
491 next->do_index_ops->dio_it.fini(env, it->lit_it);
495 /* the iterator not in use any more */
498 it->lit_stripe_index = 0;
502 * Implementation of dt_it_ops::get.
504 * Right now it's not used widely, only to reset the iterator to the
505 * initial position. It should be possible to implement a full version
506 * which chooses a correct stripe to be able to position with any key.
508 * \see dt_it_ops::get() in the API description for details.
510 static int lod_striped_it_get(const struct lu_env *env, struct dt_it *di,
511 const struct dt_key *key)
513 const struct lod_it *it = (const struct lod_it *)di;
514 struct lod_object *lo = lod_dt_obj(it->lit_obj);
515 struct dt_object *next;
517 LOD_CHECK_STRIPED_IT(env, it, lo);
519 next = lo->ldo_stripe[it->lit_stripe_index];
520 LASSERT(next != NULL);
521 LASSERT(dt_object_exists(next));
522 LASSERT(next->do_index_ops != NULL);
524 return next->do_index_ops->dio_it.get(env, it->lit_it, key);
528 * Implementation of dt_it_ops::put.
530 * Used with striped objects.
532 * \see dt_it_ops::put() in the API description for details.
534 static void lod_striped_it_put(const struct lu_env *env, struct dt_it *di)
536 struct lod_it *it = (struct lod_it *)di;
537 struct lod_object *lo = lod_dt_obj(it->lit_obj);
538 struct dt_object *next;
541 * If lit_it == NULL, then it means the sub_it has been finished,
542 * which only happens in failure cases, see lod_striped_it_next()
547 LOD_CHECK_STRIPED_IT(env, it, lo);
549 next = lo->ldo_stripe[it->lit_stripe_index];
550 LASSERT(next != NULL);
551 LASSERT(next->do_index_ops != NULL);
553 return next->do_index_ops->dio_it.put(env, it->lit_it);
557 * Implementation of dt_it_ops::next.
559 * Used with striped objects. When the end of the current stripe is
560 * reached, the method takes the next stripe's iterator.
562 * \see dt_it_ops::next() in the API description for details.
564 static int lod_striped_it_next(const struct lu_env *env, struct dt_it *di)
566 struct lod_it *it = (struct lod_it *)di;
567 struct lod_object *lo = lod_dt_obj(it->lit_obj);
568 struct dt_object *next;
569 struct dt_it *it_next;
575 LOD_CHECK_STRIPED_IT(env, it, lo);
577 next = lo->ldo_stripe[it->lit_stripe_index];
578 LASSERT(next != NULL);
579 LASSERT(dt_object_exists(next));
580 LASSERT(next->do_index_ops != NULL);
582 rc = next->do_index_ops->dio_it.next(env, it->lit_it);
586 if (rc == 0 && it->lit_stripe_index == 0)
589 if (rc == 0 && it->lit_stripe_index > 0) {
590 struct lu_dirent *ent;
592 ent = (struct lu_dirent *)lod_env_info(env)->lti_key;
594 rc = next->do_index_ops->dio_it.rec(env, it->lit_it,
595 (struct dt_rec *)ent,
600 /* skip . and .. for slave stripe */
601 if ((strncmp(ent->lde_name, ".",
602 le16_to_cpu(ent->lde_namelen)) == 0 &&
603 le16_to_cpu(ent->lde_namelen) == 1) ||
604 (strncmp(ent->lde_name, "..",
605 le16_to_cpu(ent->lde_namelen)) == 0 &&
606 le16_to_cpu(ent->lde_namelen) == 2))
612 next->do_index_ops->dio_it.put(env, it->lit_it);
613 next->do_index_ops->dio_it.fini(env, it->lit_it);
616 /* go to next stripe */
617 index = it->lit_stripe_index;
618 while (++index < lo->ldo_dir_stripe_count) {
619 next = lo->ldo_stripe[index];
623 if (!dt_object_exists(next))
626 rc = next->do_ops->do_index_try(env, next,
627 &dt_directory_features);
631 LASSERT(next->do_index_ops != NULL);
633 it_next = next->do_index_ops->dio_it.init(env, next,
636 RETURN(PTR_ERR(it_next));
638 rc = next->do_index_ops->dio_it.get(env, it_next,
639 (const struct dt_key *)"");
641 RETURN(rc == 0 ? -EIO : rc);
643 it->lit_it = it_next;
644 it->lit_stripe_index = index;
653 * Implementation of dt_it_ops::key.
655 * Used with striped objects.
657 * \see dt_it_ops::key() in the API description for details.
659 static struct dt_key *lod_striped_it_key(const struct lu_env *env,
660 const struct dt_it *di)
662 const struct lod_it *it = (const struct lod_it *)di;
663 struct lod_object *lo = lod_dt_obj(it->lit_obj);
664 struct dt_object *next;
666 LOD_CHECK_STRIPED_IT(env, it, lo);
668 next = lo->ldo_stripe[it->lit_stripe_index];
669 LASSERT(next != NULL);
670 LASSERT(next->do_index_ops != NULL);
672 return next->do_index_ops->dio_it.key(env, it->lit_it);
676 * Implementation of dt_it_ops::key_size.
678 * Used with striped objects.
680 * \see dt_it_ops::size() in the API description for details.
682 static int lod_striped_it_key_size(const struct lu_env *env,
683 const struct dt_it *di)
685 struct lod_it *it = (struct lod_it *)di;
686 struct lod_object *lo = lod_dt_obj(it->lit_obj);
687 struct dt_object *next;
689 LOD_CHECK_STRIPED_IT(env, it, lo);
691 next = lo->ldo_stripe[it->lit_stripe_index];
692 LASSERT(next != NULL);
693 LASSERT(next->do_index_ops != NULL);
695 return next->do_index_ops->dio_it.key_size(env, it->lit_it);
699 * Implementation of dt_it_ops::rec.
701 * Used with striped objects.
703 * \see dt_it_ops::rec() in the API description for details.
705 static int lod_striped_it_rec(const struct lu_env *env, const struct dt_it *di,
706 struct dt_rec *rec, __u32 attr)
708 const struct lod_it *it = (const struct lod_it *)di;
709 struct lod_object *lo = lod_dt_obj(it->lit_obj);
710 struct dt_object *next;
712 LOD_CHECK_STRIPED_IT(env, it, lo);
714 next = lo->ldo_stripe[it->lit_stripe_index];
715 LASSERT(next != NULL);
716 LASSERT(next->do_index_ops != NULL);
718 return next->do_index_ops->dio_it.rec(env, it->lit_it, rec, attr);
722 * Implementation of dt_it_ops::rec_size.
724 * Used with striped objects.
726 * \see dt_it_ops::rec_size() in the API description for details.
728 static int lod_striped_it_rec_size(const struct lu_env *env,
729 const struct dt_it *di, __u32 attr)
731 struct lod_it *it = (struct lod_it *)di;
732 struct lod_object *lo = lod_dt_obj(it->lit_obj);
733 struct dt_object *next;
735 LOD_CHECK_STRIPED_IT(env, it, lo);
737 next = lo->ldo_stripe[it->lit_stripe_index];
738 LASSERT(next != NULL);
739 LASSERT(next->do_index_ops != NULL);
741 return next->do_index_ops->dio_it.rec_size(env, it->lit_it, attr);
745 * Implementation of dt_it_ops::store.
747 * Used with striped objects.
749 * \see dt_it_ops::store() in the API description for details.
751 static __u64 lod_striped_it_store(const struct lu_env *env,
752 const struct dt_it *di)
754 const struct lod_it *it = (const struct lod_it *)di;
755 struct lod_object *lo = lod_dt_obj(it->lit_obj);
756 struct dt_object *next;
758 LOD_CHECK_STRIPED_IT(env, it, lo);
760 next = lo->ldo_stripe[it->lit_stripe_index];
761 LASSERT(next != NULL);
762 LASSERT(next->do_index_ops != NULL);
764 return next->do_index_ops->dio_it.store(env, it->lit_it);
768 * Implementation of dt_it_ops::load.
770 * Used with striped objects.
772 * \see dt_it_ops::load() in the API description for details.
774 static int lod_striped_it_load(const struct lu_env *env,
775 const struct dt_it *di, __u64 hash)
777 const struct lod_it *it = (const struct lod_it *)di;
778 struct lod_object *lo = lod_dt_obj(it->lit_obj);
779 struct dt_object *next;
781 LOD_CHECK_STRIPED_IT(env, it, lo);
783 next = lo->ldo_stripe[it->lit_stripe_index];
784 LASSERT(next != NULL);
785 LASSERT(next->do_index_ops != NULL);
787 return next->do_index_ops->dio_it.load(env, it->lit_it, hash);
790 static const struct dt_index_operations lod_striped_index_ops = {
791 .dio_lookup = lod_striped_lookup,
792 .dio_declare_insert = lod_declare_insert,
793 .dio_insert = lod_insert,
794 .dio_declare_delete = lod_declare_delete,
795 .dio_delete = lod_delete,
797 .init = lod_striped_it_init,
798 .fini = lod_striped_it_fini,
799 .get = lod_striped_it_get,
800 .put = lod_striped_it_put,
801 .next = lod_striped_it_next,
802 .key = lod_striped_it_key,
803 .key_size = lod_striped_it_key_size,
804 .rec = lod_striped_it_rec,
805 .rec_size = lod_striped_it_rec_size,
806 .store = lod_striped_it_store,
807 .load = lod_striped_it_load,
812 * Append the FID for each shard of the striped directory after the
813 * given LMV EA header.
815 * To simplify striped directory and the consistency verification,
816 * we only store the LMV EA header on disk, for both master object
817 * and slave objects. When someone wants to know the whole LMV EA,
818 * such as client readdir(), we can build the entrie LMV EA on the
819 * MDT side (in RAM) via iterating the sub-directory entries that
820 * are contained in the master object of the stripe directory.
822 * For the master object of the striped directroy, the valid name
823 * for each shard is composed of the ${shard_FID}:${shard_idx}.
825 * There may be holes in the LMV EA if some shards' name entries
826 * are corrupted or lost.
828 * \param[in] env pointer to the thread context
829 * \param[in] lo pointer to the master object of the striped directory
830 * \param[in] buf pointer to the lu_buf which will hold the LMV EA
831 * \param[in] resize whether re-allocate the buffer if it is not big enough
833 * \retval positive size of the LMV EA
834 * \retval 0 for nothing to be loaded
835 * \retval negative error number on failure
837 int lod_load_lmv_shards(const struct lu_env *env, struct lod_object *lo,
838 struct lu_buf *buf, bool resize)
840 struct lu_dirent *ent =
841 (struct lu_dirent *)lod_env_info(env)->lti_key;
842 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
843 struct dt_object *obj = dt_object_child(&lo->ldo_obj);
844 struct lmv_mds_md_v1 *lmv1 = buf->lb_buf;
846 const struct dt_it_ops *iops;
848 __u32 magic = le32_to_cpu(lmv1->lmv_magic);
853 if (magic != LMV_MAGIC_V1)
856 stripes = le32_to_cpu(lmv1->lmv_stripe_count);
860 rc = lmv_mds_md_size(stripes, magic);
864 if (buf->lb_len < lmv1_size) {
873 lu_buf_alloc(buf, lmv1_size);
878 memcpy(buf->lb_buf, tbuf.lb_buf, tbuf.lb_len);
881 if (unlikely(!dt_try_as_dir(env, obj, true)))
884 memset(&lmv1->lmv_stripe_fids[0], 0, stripes * sizeof(struct lu_fid));
885 iops = &obj->do_index_ops->dio_it;
886 it = iops->init(env, obj, LUDA_64BITHASH);
890 rc = iops->load(env, it, 0);
892 rc = iops->next(env, it);
897 char name[FID_LEN + 2] = "";
902 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
908 fid_le_to_cpu(&fid, &ent->lde_fid);
909 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
910 if (ent->lde_name[0] == '.') {
911 if (ent->lde_namelen == 1)
914 if (ent->lde_namelen == 2 && ent->lde_name[1] == '.')
918 len = scnprintf(name, sizeof(name),
919 DFID":", PFID(&ent->lde_fid));
920 /* The ent->lde_name is composed of ${FID}:${index} */
921 if (ent->lde_namelen < len + 1 ||
922 memcmp(ent->lde_name, name, len) != 0) {
923 CDEBUG_LIMIT(lod->lod_lmv_failout ? D_ERROR : D_INFO,
924 "%s: invalid shard name %.*s with the FID "DFID" for the striped directory "DFID", %s\n",
925 lod2obd(lod)->obd_name, ent->lde_namelen,
926 ent->lde_name, PFID(&fid),
927 PFID(lu_object_fid(&obj->do_lu)),
928 lod->lod_lmv_failout ? "failout" : "skip");
930 if (lod->lod_lmv_failout)
938 if (ent->lde_name[len] < '0' ||
939 ent->lde_name[len] > '9') {
940 CDEBUG_LIMIT(lod->lod_lmv_failout ?
942 "%s: invalid shard name %.*s with the FID "DFID" for the striped directory "DFID", %s\n",
943 lod2obd(lod)->obd_name,
945 ent->lde_name, PFID(&fid),
946 PFID(lu_object_fid(&obj->do_lu)),
947 lod->lod_lmv_failout ?
950 if (lod->lod_lmv_failout)
956 index = index * 10 + ent->lde_name[len++] - '0';
957 } while (len < ent->lde_namelen);
959 if (len == ent->lde_namelen) {
960 /* Out of LMV EA range. */
961 if (index >= stripes) {
962 CERROR("%s: the shard %.*s for the striped "
963 "directory "DFID" is out of the known "
964 "LMV EA range [0 - %u], failout\n",
965 lod2obd(lod)->obd_name, ent->lde_namelen,
967 PFID(lu_object_fid(&obj->do_lu)),
973 /* The slot has been occupied. */
974 if (!fid_is_zero(&lmv1->lmv_stripe_fids[index]) &&
975 !CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME)) {
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 if (comp_end != LUSTRE_EOF &&
1140 comp_end & (LOV_MIN_STRIPE_SIZE - 1)) {
1141 CWARN("Component end %llu is not a multiple of min size %u\n",
1142 comp_end, LOV_MIN_STRIPE_SIZE);
1143 comp_end = round_up(comp_end, LOV_MIN_STRIPE_SIZE);
1145 /* check stripe size is multiplier of comp_end */
1146 if (comp_end != LUSTRE_EOF &&
1147 comp_end != comp->llc_extent.e_start &&
1148 comp_end % comp->llc_stripe_size) {
1149 /* fix that even for defined stripe size but warn
1150 * about the problem, that must not happen
1152 CWARN("Component end %llu is not aligned by the stripe size %u\n",
1153 comp_end, comp->llc_stripe_size);
1154 comp->llc_stripe_size = comp_end & ~(comp_end - 1);
1159 static inline void lod_adjust_stripe_info(struct lod_layout_component *comp,
1160 struct lov_desc *desc,
1163 if (!(comp->llc_pattern & LOV_PATTERN_MDT)) {
1164 if (append_stripes) {
1165 comp->llc_stripe_count = append_stripes;
1166 } else if (!comp->llc_stripe_count) {
1167 comp->llc_stripe_count =
1168 desc->ld_default_stripe_count;
1172 lod_adjust_stripe_size(comp, desc->ld_default_stripe_size);
1175 int lod_obj_for_each_stripe(const struct lu_env *env, struct lod_object *lo,
1177 struct lod_obj_stripe_cb_data *data)
1179 struct lod_layout_component *lod_comp;
1183 mutex_lock(&lo->ldo_layout_mutex);
1184 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1185 lod_comp = &lo->ldo_comp_entries[i];
1187 if (lod_comp->llc_magic == LOV_MAGIC_FOREIGN)
1190 if (lod_comp->llc_stripe == NULL)
1193 /* has stripe but not inited yet, this component has been
1194 * declared to be created, but hasn't created yet.
1196 if (!lod_comp_inited(lod_comp))
1199 if (data->locd_comp_skip_cb &&
1200 data->locd_comp_skip_cb(env, lo, i, data))
1203 if (data->locd_comp_cb) {
1204 rc = data->locd_comp_cb(env, lo, i, data);
1209 /* could used just to do sth about component, not each
1212 if (!data->locd_stripe_cb)
1215 LASSERT(lod_comp->llc_stripe_count > 0);
1216 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
1217 struct dt_object *dt = lod_comp->llc_stripe[j];
1221 rc = data->locd_stripe_cb(env, lo, dt, th, i, j, data);
1227 mutex_unlock(&lo->ldo_layout_mutex);
1232 lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
1233 struct dt_object *dt, struct thandle *th,
1234 int comp_idx, int stripe_idx,
1235 struct lod_obj_stripe_cb_data *data)
1237 if (data->locd_declare)
1238 return lod_sub_declare_attr_set(env, dt, data->locd_attr, th);
1240 if (data->locd_attr->la_valid & LA_LAYOUT_VERSION) {
1241 CDEBUG(D_LAYOUT, DFID": set layout version: %u, comp_idx: %d\n",
1242 PFID(lu_object_fid(&dt->do_lu)),
1243 data->locd_attr->la_layout_version, comp_idx);
1246 return lod_sub_attr_set(env, dt, data->locd_attr, th);
1250 * Implementation of dt_object_operations::do_declare_attr_set.
1252 * If the object is striped, then apply the changes to all the stripes.
1254 * \see dt_object_operations::do_declare_attr_set() in the API description
1257 static int lod_declare_attr_set(const struct lu_env *env,
1258 struct dt_object *dt,
1259 const struct lu_attr *attr,
1262 struct dt_object *next = dt_object_child(dt);
1263 struct lod_object *lo = lod_dt_obj(dt);
1268 * declare setattr on the local object
1270 rc = lod_sub_declare_attr_set(env, next, attr, th);
1274 /* osp_declare_attr_set() ignores all attributes other than
1275 * UID, GID, PROJID, and size, and osp_attr_set() ignores all
1276 * but UID, GID and PROJID. Declaration of size attr setting
1277 * happens through lod_declare_init_size(), and not through
1278 * this function. Therefore we need not load striping unless
1279 * ownership is changing. This should save memory and (we hope)
1280 * speed up rename().
1282 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1283 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1286 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1289 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID | LA_MODE |
1290 LA_ATIME | LA_MTIME | LA_CTIME |
1295 * load striping information, notice we don't do this when object
1296 * is being initialized as we don't need this information till
1297 * few specific cases like destroy, chown
1299 rc = lod_striping_load(env, lo);
1303 if (!lod_obj_is_striped(dt))
1307 * if object is striped declare changes on the stripes
1309 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1310 LASSERT(lo->ldo_stripe);
1311 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1312 if (lo->ldo_stripe[i] == NULL)
1314 if (!dt_object_exists(lo->ldo_stripe[i]))
1316 rc = lod_sub_declare_attr_set(env, lo->ldo_stripe[i],
1322 struct lod_obj_stripe_cb_data data = { { 0 } };
1324 data.locd_attr = attr;
1325 data.locd_declare = true;
1326 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1327 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1333 if (!dt_object_exists(next) || dt_object_remote(next) ||
1334 !S_ISREG(attr->la_mode))
1337 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1338 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
1342 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) ||
1343 CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1344 struct lod_thread_info *info = lod_env_info(env);
1345 struct lu_buf *buf = &info->lti_buf;
1347 buf->lb_buf = info->lti_ea_store;
1348 buf->lb_len = info->lti_ea_store_size;
1349 rc = lod_sub_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1350 LU_XATTR_REPLACE, th);
1357 * Implementation of dt_object_operations::do_attr_set.
1359 * If the object is striped, then apply the changes to all or subset of
1360 * the stripes depending on the object type and specific attributes.
1362 * \see dt_object_operations::do_attr_set() in the API description for details.
1364 static int lod_attr_set(const struct lu_env *env,
1365 struct dt_object *dt,
1366 const struct lu_attr *attr,
1369 struct dt_object *next = dt_object_child(dt);
1370 struct lod_object *lo = lod_dt_obj(dt);
1375 * apply changes to the local object
1377 rc = lod_sub_attr_set(env, next, attr, th);
1381 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1382 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1385 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1388 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE | LA_PROJID |
1389 LA_ATIME | LA_MTIME | LA_CTIME |
1394 /* FIXME: a tricky case in the code path of mdd_layout_change():
1395 * the in-memory striping information has been freed in lod_xattr_set()
1396 * due to layout change. It has to load stripe here again. It only
1397 * changes flags of layout so declare_attr_set() is still accurate */
1398 rc = lod_striping_load(env, lo);
1402 if (!lod_obj_is_striped(dt))
1406 * if object is striped, apply changes to all the stripes
1408 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1409 LASSERT(lo->ldo_stripe);
1410 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1411 if (unlikely(lo->ldo_stripe[i] == NULL))
1414 if ((dt_object_exists(lo->ldo_stripe[i]) == 0))
1417 rc = lod_sub_attr_set(env, lo->ldo_stripe[i], attr, th);
1422 struct lod_obj_stripe_cb_data data = { { 0 } };
1424 data.locd_attr = attr;
1425 data.locd_declare = false;
1426 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1427 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1433 if (!dt_object_exists(next) || dt_object_remote(next) ||
1434 !S_ISREG(attr->la_mode))
1437 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1438 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
1442 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE)) {
1443 struct lod_thread_info *info = lod_env_info(env);
1444 struct lu_buf *buf = &info->lti_buf;
1445 struct ost_id *oi = &info->lti_ostid;
1446 struct lu_fid *fid = &info->lti_fid;
1447 struct lov_mds_md_v1 *lmm;
1448 struct lov_ost_data_v1 *objs;
1451 rc = lod_get_lov_ea(env, lo);
1455 buf->lb_buf = info->lti_ea_store;
1456 buf->lb_len = info->lti_ea_store_size;
1457 lmm = info->lti_ea_store;
1458 magic = le32_to_cpu(lmm->lmm_magic);
1459 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1460 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1461 struct lov_comp_md_entry_v1 *lcme =
1462 &lcm->lcm_entries[0];
1464 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
1465 magic = le32_to_cpu(lmm->lmm_magic);
1468 if (magic == LOV_MAGIC_V1)
1469 objs = &(lmm->lmm_objects[0]);
1471 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1472 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1473 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1475 fid_to_ostid(fid, oi);
1476 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1478 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1479 LU_XATTR_REPLACE, th);
1480 } else if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1481 struct lod_thread_info *info = lod_env_info(env);
1482 struct lu_buf *buf = &info->lti_buf;
1483 struct lov_comp_md_v1 *lcm;
1484 struct lov_comp_md_entry_v1 *lcme;
1486 rc = lod_get_lov_ea(env, lo);
1490 buf->lb_buf = info->lti_ea_store;
1491 buf->lb_len = info->lti_ea_store_size;
1493 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1 &&
1494 le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_SEL)
1497 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1498 lcme = &lcm->lcm_entries[0];
1499 le64_add_cpu(&lcme->lcme_extent.e_start, 1);
1500 le64_add_cpu(&lcme->lcme_extent.e_end, -1);
1502 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1503 LU_XATTR_REPLACE, th);
1510 * Implementation of dt_object_operations::do_xattr_get.
1512 * If LOV EA is requested from the root object and it's not
1513 * found, then return default striping for the filesystem.
1515 * \see dt_object_operations::do_xattr_get() in the API description for details.
1517 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1518 struct lu_buf *buf, const char *name)
1520 struct lod_thread_info *info = lod_env_info(env);
1521 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1526 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1527 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1528 struct lmv_mds_md_v1 *lmv1;
1529 struct lmv_foreign_md *lfm;
1532 if (rc > (typeof(rc))sizeof(*lmv1))
1535 /* short (<= sizeof(struct lmv_mds_md_v1)) foreign LMV case */
1536 /* XXX empty foreign LMV is not allowed */
1537 if (rc <= offsetof(typeof(*lfm), lfm_value))
1538 RETURN(rc = rc > 0 ? -EINVAL : rc);
1540 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1541 BUILD_BUG_ON(sizeof(*lmv1) > sizeof(info->lti_key));
1543 /* lti_buf is large enough for *lmv1 or a short
1544 * (<= sizeof(struct lmv_mds_md_v1)) foreign LMV
1546 info->lti_buf.lb_buf = info->lti_key;
1547 info->lti_buf.lb_len = sizeof(*lmv1);
1548 rc = dt_xattr_get(env, dt_object_child(dt),
1549 &info->lti_buf, name);
1550 if (unlikely(rc <= offsetof(typeof(*lfm),
1552 RETURN(rc = rc > 0 ? -EINVAL : rc);
1554 lfm = info->lti_buf.lb_buf;
1555 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN)
1558 if (unlikely(rc != sizeof(*lmv1)))
1559 RETURN(rc = rc > 0 ? -EINVAL : rc);
1561 lmv1 = info->lti_buf.lb_buf;
1562 /* The on-disk LMV EA only contains header, but the
1563 * returned LMV EA size should contain the space for
1564 * the FIDs of all shards of the striped directory. */
1565 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1566 rc = lmv_mds_md_size(
1567 le32_to_cpu(lmv1->lmv_stripe_count),
1568 le32_to_cpu(lmv1->lmv_magic));
1571 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1574 if (rc != sizeof(*lmv1))
1575 RETURN(rc = rc > 0 ? -EINVAL : rc);
1577 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1581 RETURN(rc = rc1 != 0 ? rc1 : rc);
1584 if ((rc > 0) && buf->lb_buf && strcmp(name, XATTR_NAME_LOV) == 0) {
1585 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1587 if (lcm->lcm_magic == cpu_to_le32(LOV_MAGIC_SEL))
1588 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
1591 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1595 * XXX: Only used by lfsck
1597 * lod returns default striping on the real root of the device
1598 * this is like the root stores default striping for the whole
1599 * filesystem. historically we've been using a different approach
1600 * and store it in the config.
1602 dt_root_get(env, dev->lod_child, &info->lti_fid);
1603 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1605 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1606 struct lov_user_md *lum = buf->lb_buf;
1607 struct lov_desc *desc = &dev->lod_ost_descs.ltd_lov_desc;
1609 if (buf->lb_buf == NULL) {
1611 } else if (buf->lb_len >= sizeof(*lum)) {
1612 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1613 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1614 lmm_oi_set_id(&lum->lmm_oi, 0);
1615 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1616 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1617 lum->lmm_stripe_size = cpu_to_le32(
1618 desc->ld_default_stripe_size);
1619 lum->lmm_stripe_count = cpu_to_le16(
1620 desc->ld_default_stripe_count);
1621 lum->lmm_stripe_offset = cpu_to_le16(
1622 desc->ld_default_stripe_offset);
1635 * Checks that the magic of the stripe is sane.
1637 * \param[in] lod lod device
1638 * \param[in] lum a buffer storing LMV EA to verify
1640 * \retval 0 if the EA is sane
1641 * \retval negative otherwise
1643 static int lod_verify_md_striping(struct lod_device *lod,
1644 const struct lmv_user_md_v1 *lum)
1646 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1647 CERROR("%s: invalid lmv_user_md: magic = %x, "
1648 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1649 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1650 (int)le32_to_cpu(lum->lum_stripe_offset),
1651 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1659 * Initialize LMV EA for a slave.
1661 * Initialize slave's LMV EA from the master's LMV EA.
1663 * \param[in] master_lmv a buffer containing master's EA
1664 * \param[out] slave_lmv a buffer where slave's EA will be stored
1667 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1668 const struct lmv_mds_md_v1 *master_lmv)
1670 *slave_lmv = *master_lmv;
1671 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1677 * Generate LMV EA from the object passed as \a dt. The object must have
1678 * the stripes created and initialized.
1680 * \param[in] env execution environment
1681 * \param[in] dt object
1682 * \param[out] lmv_buf buffer storing generated LMV EA
1684 * \retval 0 on success
1685 * \retval negative if failed
1687 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1688 struct lu_buf *lmv_buf)
1690 struct lod_thread_info *info = lod_env_info(env);
1691 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1692 struct lod_object *lo = lod_dt_obj(dt);
1693 struct lmv_mds_md_v1 *lmm1;
1695 int type = LU_SEQ_RANGE_ANY;
1700 LASSERT(lo->ldo_dir_striped != 0);
1701 LASSERT(lo->ldo_dir_stripe_count > 0);
1702 stripe_count = lo->ldo_dir_stripe_count;
1703 /* Only store the LMV EA heahder on the disk. */
1704 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1705 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1709 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1712 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1713 memset(lmm1, 0, sizeof(*lmm1));
1714 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1715 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1716 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1717 lmm1->lmv_layout_version = cpu_to_le32(lo->ldo_dir_layout_version);
1718 if (lod_is_layout_changing(lo)) {
1719 lmm1->lmv_migrate_hash = cpu_to_le32(lo->ldo_dir_migrate_hash);
1720 lmm1->lmv_migrate_offset =
1721 cpu_to_le32(lo->ldo_dir_migrate_offset);
1723 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1728 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1729 lmv_buf->lb_buf = info->lti_ea_store;
1730 lmv_buf->lb_len = sizeof(*lmm1);
1736 * Create in-core represenation for a striped directory.
1738 * Parse the buffer containing LMV EA and instantiate LU objects
1739 * representing the stripe objects. The pointers to the objects are
1740 * stored in ldo_stripe field of \a lo. This function is used when
1741 * we need to access an already created object (i.e. load from a disk).
1743 * \param[in] env execution environment
1744 * \param[in] lo lod object
1745 * \param[in] buf buffer containing LMV EA
1747 * \retval 0 on success
1748 * \retval negative if failed
1750 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1751 const struct lu_buf *buf)
1753 struct lod_thread_info *info = lod_env_info(env);
1754 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1755 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1756 struct dt_object **stripe;
1757 union lmv_mds_md *lmm = buf->lb_buf;
1758 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1759 struct lu_fid *fid = &info->lti_fid;
1764 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1766 /* XXX may be useless as not called for foreign LMV ?? */
1767 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_FOREIGN)
1770 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1771 lo->ldo_dir_slave_stripe = 1;
1775 if (!lmv_is_sane(lmv1))
1778 LASSERT(lo->ldo_stripe == NULL);
1779 OBD_ALLOC_PTR_ARRAY(stripe, le32_to_cpu(lmv1->lmv_stripe_count));
1783 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1784 struct dt_device *tgt_dt;
1785 struct dt_object *dto;
1786 int type = LU_SEQ_RANGE_ANY;
1789 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1790 if (!fid_is_sane(fid)) {
1795 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1799 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1800 tgt_dt = lod->lod_child;
1802 struct lod_tgt_desc *tgt;
1804 tgt = LTD_TGT(ltd, idx);
1806 GOTO(out, rc = -ESTALE);
1807 tgt_dt = tgt->ltd_tgt;
1810 dto = dt_locate_at(env, tgt_dt, fid,
1811 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1814 GOTO(out, rc = PTR_ERR(dto));
1819 lo->ldo_stripe = stripe;
1820 lo->ldo_is_foreign = 0;
1821 lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1822 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1823 lo->ldo_dir_layout_version = le32_to_cpu(lmv1->lmv_layout_version);
1824 lo->ldo_dir_migrate_offset = le32_to_cpu(lmv1->lmv_migrate_offset);
1825 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv1->lmv_migrate_hash);
1826 lo->ldo_dir_hash_type = le32_to_cpu(lmv1->lmv_hash_type);
1828 lod_striping_free_nolock(env, lo);
1834 * Declare create a striped directory.
1836 * Declare creating a striped directory with a given stripe pattern on the
1837 * specified MDTs. A striped directory is represented as a regular directory
1838 * - an index listing all the stripes. The stripes point back to the master
1839 * object with ".." and LinkEA. The master object gets LMV EA which
1840 * identifies it as a striped directory. The function allocates FIDs
1843 * \param[in] env execution environment
1844 * \param[in] dt object
1845 * \param[in] attr attributes to initialize the objects with
1846 * \param[in] dof type of objects to be created
1847 * \param[in] th transaction handle
1849 * \retval 0 on success
1850 * \retval negative if failed
1852 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1853 struct dt_object *dt,
1854 struct lu_attr *attr,
1855 struct dt_object_format *dof,
1858 struct lod_thread_info *info = lod_env_info(env);
1859 struct lu_buf lmv_buf;
1860 struct lu_buf slave_lmv_buf;
1861 struct lmv_mds_md_v1 *lmm;
1862 struct lmv_mds_md_v1 *slave_lmm = NULL;
1863 struct dt_insert_rec *rec = &info->lti_dt_rec;
1864 struct lod_object *lo = lod_dt_obj(dt);
1869 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1872 lmm = lmv_buf.lb_buf;
1874 OBD_ALLOC_PTR(slave_lmm);
1875 if (slave_lmm == NULL)
1876 GOTO(out, rc = -ENOMEM);
1878 lod_prep_slave_lmv_md(slave_lmm, lmm);
1879 slave_lmv_buf.lb_buf = slave_lmm;
1880 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1882 if (!dt_try_as_dir(env, dt_object_child(dt), false))
1883 GOTO(out, rc = -EINVAL);
1885 rec->rec_type = S_IFDIR;
1886 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1887 struct dt_object *dto = lo->ldo_stripe[i];
1888 char *stripe_name = info->lti_key;
1889 struct lu_name *sname;
1890 struct linkea_data ldata = { NULL };
1891 struct lu_buf linkea_buf;
1893 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
1897 /* directory split skip create for existing stripes */
1898 if (!(lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) {
1899 rc = lod_sub_declare_create(env, dto, attr, NULL, dof,
1904 if (!dt_try_as_dir(env, dto, false))
1905 GOTO(out, rc = -EINVAL);
1907 rc = lod_sub_declare_ref_add(env, dto, th);
1911 rec->rec_fid = lu_object_fid(&dto->do_lu);
1912 rc = lod_sub_declare_insert(env, dto,
1913 (const struct dt_rec *)rec,
1914 (const struct dt_key *)dot,
1919 /* master stripe FID will be put to .. */
1920 rec->rec_fid = lu_object_fid(&dt->do_lu);
1921 rc = lod_sub_declare_insert(env, dto,
1922 (const struct dt_rec *)rec,
1923 (const struct dt_key *)dotdot,
1928 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1930 snprintf(stripe_name, sizeof(info->lti_key),
1932 PFID(lu_object_fid(&dto->do_lu)),
1935 snprintf(stripe_name, sizeof(info->lti_key),
1937 PFID(lu_object_fid(&dto->do_lu)), i);
1939 sname = lod_name_get(env, stripe_name,
1940 strlen(stripe_name));
1941 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1942 sname, lu_object_fid(&dt->do_lu));
1946 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1947 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1948 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
1949 XATTR_NAME_LINK, 0, th);
1953 rec->rec_fid = lu_object_fid(&dto->do_lu);
1954 rc = lod_sub_declare_insert(env, dt_object_child(dt),
1955 (const struct dt_rec *)rec,
1956 (const struct dt_key *)stripe_name, th);
1960 rc = lod_sub_declare_ref_add(env, dt_object_child(dt),
1966 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1967 cfs_fail_val != i) {
1968 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1970 slave_lmm->lmv_master_mdt_index =
1973 slave_lmm->lmv_master_mdt_index =
1975 rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
1976 XATTR_NAME_LMV, 0, th);
1982 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
1983 &lmv_buf, XATTR_NAME_LMV, 0, th);
1987 if (slave_lmm != NULL)
1988 OBD_FREE_PTR(slave_lmm);
1994 * Allocate a striping on a predefined set of MDTs.
1996 * Allocates new striping using the MDT index range provided by the data from
1997 * the lum_obejcts contained in the lmv_user_md passed to this method if
1998 * \a is_specific is true; or allocates new layout starting from MDT index in
1999 * lo->ldo_dir_stripe_offset. The exact order of MDTs is not important and
2000 * varies depending on MDT status. The number of stripes needed and stripe
2001 * offset are taken from the object. If that number cannot be met, then the
2002 * function returns an error and then it's the caller's responsibility to
2003 * release the stripes allocated. All the internal structures are protected,
2004 * but no concurrent allocation is allowed on the same objects.
2006 * \param[in] env execution environment for this thread
2007 * \param[in] lo LOD object
2008 * \param[out] stripes striping created
2009 * \param[out] mdt_indices MDT indices of striping created
2010 * \param[in] is_specific true if the MDTs are provided by lum; false if
2011 * only the starting MDT index is provided
2013 * \retval positive stripes allocated, including the first stripe allocated
2015 * \retval negative errno on failure
2017 static int lod_mdt_alloc_specific(const struct lu_env *env,
2018 struct lod_object *lo,
2019 struct dt_object **stripes,
2020 __u32 *mdt_indices, bool is_specific)
2022 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
2023 struct lu_tgt_descs *ltd = &lod->lod_mdt_descs;
2024 struct lu_tgt_desc *tgt = NULL;
2025 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2026 struct dt_device *tgt_dt = NULL;
2027 struct lu_fid fid = { 0 };
2028 struct dt_object *dto;
2030 u32 stripe_count = lo->ldo_dir_stripe_count;
2036 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
2037 if (!is_specific && stripe_count > 1)
2038 /* Set the start index for the 2nd stripe allocation */
2039 mdt_indices[1] = (mdt_indices[0] + 1) %
2040 (lod->lod_remote_mdt_count + 1);
2042 for (; stripe_idx < stripe_count; stripe_idx++) {
2043 /* Try to find next avaible target */
2044 idx = mdt_indices[stripe_idx];
2045 for (j = 0; j < lod->lod_remote_mdt_count;
2046 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
2047 bool already_allocated = false;
2050 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
2051 idx, lod->lod_remote_mdt_count + 1, stripe_idx);
2053 if (likely(!is_specific &&
2054 !CFS_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
2055 /* check whether the idx already exists
2056 * in current allocated array */
2057 for (k = 0; k < stripe_idx; k++) {
2058 if (mdt_indices[k] == idx) {
2059 already_allocated = true;
2064 if (already_allocated)
2068 /* Sigh, this index is not in the bitmap, let's check
2069 * next available target */
2070 if (!test_bit(idx, ltd->ltd_tgt_bitmap) &&
2071 idx != master_index)
2074 if (idx == master_index) {
2075 /* Allocate the FID locally */
2076 tgt_dt = lod->lod_child;
2077 rc = dt_fid_alloc(env, tgt_dt, &fid, NULL,
2084 /* check the status of the OSP */
2085 tgt = LTD_TGT(ltd, idx);
2089 tgt_dt = tgt->ltd_tgt;
2090 if (!tgt->ltd_active)
2091 /* this OSP doesn't feel well */
2094 rc = dt_fid_alloc(env, tgt_dt, &fid, NULL, NULL);
2101 /* Can not allocate more stripes */
2102 if (j == lod->lod_remote_mdt_count) {
2103 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
2104 lod2obd(lod)->obd_name, stripe_count,
2109 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
2110 idx, stripe_idx, PFID(&fid));
2111 mdt_indices[stripe_idx] = idx;
2112 /* Set the start index for next stripe allocation */
2113 if (!is_specific && stripe_idx < stripe_count - 1) {
2115 * for large dir test, put all other slaves on one
2116 * remote MDT, otherwise we may save too many local
2117 * slave locks which will exceed RS_MAX_LOCKS.
2119 if (unlikely(CFS_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)))
2121 mdt_indices[stripe_idx + 1] = (idx + 1) %
2122 (lod->lod_remote_mdt_count + 1);
2124 /* tgt_dt and fid must be ready after search avaible OSP
2125 * in the above loop */
2126 LASSERT(tgt_dt != NULL);
2127 LASSERT(fid_is_sane(&fid));
2129 /* fail a remote stripe FID allocation */
2130 if (stripe_idx && CFS_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_FID))
2133 dto = dt_locate_at(env, tgt_dt, &fid,
2134 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
2141 stripes[stripe_idx] = dto;
2147 for (j = 1; j < stripe_idx; j++) {
2148 LASSERT(stripes[j] != NULL);
2149 dt_object_put(env, stripes[j]);
2155 static int lod_prep_md_striped_create(const struct lu_env *env,
2156 struct dt_object *dt,
2157 struct lu_attr *attr,
2158 const struct lmv_user_md_v1 *lum,
2159 struct dt_object_format *dof,
2162 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
2163 struct lod_object *lo = lod_dt_obj(dt);
2164 struct dt_object **stripes;
2165 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2166 struct lu_fid fid = { 0 };
2173 /* The lum has been verifed in lod_verify_md_striping */
2174 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
2175 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
2177 stripe_count = lo->ldo_dir_stripe_count;
2179 OBD_ALLOC_PTR_ARRAY(stripes, stripe_count);
2183 /* Allocate the first stripe locally */
2184 rc = dt_fid_alloc(env, lod->lod_child, &fid, NULL, NULL);
2188 stripes[0] = dt_locate_at(env, lod->lod_child, &fid,
2189 dt->do_lu.lo_dev->ld_site->ls_top_dev, &conf);
2190 if (IS_ERR(stripes[0]))
2191 GOTO(out, rc = PTR_ERR(stripes[0]));
2193 if (lo->ldo_dir_stripe_offset == LMV_OFFSET_DEFAULT) {
2194 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
2195 rc = lod_mdt_alloc_qos(env, lo, stripes, 1, stripe_count);
2197 rc = lod_mdt_alloc_rr(env, lo, stripes, 1,
2201 bool is_specific = false;
2203 OBD_ALLOC_PTR_ARRAY(idx_array, stripe_count);
2205 GOTO(out, rc = -ENOMEM);
2207 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
2209 for (i = 0; i < stripe_count; i++)
2211 le32_to_cpu(lum->lum_objects[i].lum_mds);
2214 /* stripe 0 is local */
2216 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
2217 rc = lod_mdt_alloc_specific(env, lo, stripes, idx_array,
2219 OBD_FREE_PTR_ARRAY(idx_array, stripe_count);
2227 lo->ldo_dir_striped = 1;
2228 lo->ldo_stripe = stripes;
2229 lo->ldo_dir_stripe_count = rc;
2230 lo->ldo_dir_stripes_allocated = stripe_count;
2232 lo->ldo_dir_stripe_loaded = 1;
2234 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2236 lod_striping_free(env, lo);
2242 if (!IS_ERR_OR_NULL(stripes[0]))
2243 dt_object_put(env, stripes[0]);
2244 for (i = 1; i < stripe_count; i++)
2245 LASSERT(!stripes[i]);
2246 OBD_FREE_PTR_ARRAY(stripes, stripe_count);
2253 * Alloc cached foreign LOV
2255 * \param[in] lo object
2256 * \param[in] size size of foreign LOV
2258 * \retval 0 on success
2259 * \retval negative if failed
2261 int lod_alloc_foreign_lov(struct lod_object *lo, size_t size)
2263 OBD_ALLOC_LARGE(lo->ldo_foreign_lov, size);
2264 if (lo->ldo_foreign_lov == NULL)
2266 lo->ldo_foreign_lov_size = size;
2267 lo->ldo_is_foreign = 1;
2273 * Free cached foreign LOV
2275 * \param[in] lo object
2277 void lod_free_foreign_lov(struct lod_object *lo)
2279 if (lo->ldo_foreign_lov != NULL)
2280 OBD_FREE_LARGE(lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
2281 lo->ldo_foreign_lov = NULL;
2282 lo->ldo_foreign_lov_size = 0;
2283 lo->ldo_is_foreign = 0;
2288 * Alloc cached foreign LMV
2290 * \param[in] lo object
2291 * \param[in] size size of foreign LMV
2293 * \retval 0 on success
2294 * \retval negative if failed
2296 int lod_alloc_foreign_lmv(struct lod_object *lo, size_t size)
2298 OBD_ALLOC_LARGE(lo->ldo_foreign_lmv, size);
2299 if (lo->ldo_foreign_lmv == NULL)
2301 lo->ldo_foreign_lmv_size = size;
2302 lo->ldo_is_foreign = 1;
2307 static int lod_prep_md_replayed_create(const struct lu_env *env,
2308 struct dt_object *dt,
2309 struct lu_attr *attr,
2310 const struct lu_buf *lmv_buf,
2311 struct dt_object_format *dof,
2314 struct lod_object *lo = lod_dt_obj(dt);
2319 mutex_lock(&lo->ldo_layout_mutex);
2320 rc = lod_parse_dir_striping(env, lo, lmv_buf);
2322 lo->ldo_dir_stripe_loaded = 1;
2323 lo->ldo_dir_striped = 1;
2324 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2326 mutex_unlock(&lo->ldo_layout_mutex);
2333 * Free cached foreign LMV
2335 * \param[in] lo object
2337 void lod_free_foreign_lmv(struct lod_object *lo)
2339 if (lo->ldo_foreign_lmv != NULL)
2340 OBD_FREE_LARGE(lo->ldo_foreign_lmv, lo->ldo_foreign_lmv_size);
2341 lo->ldo_foreign_lmv = NULL;
2342 lo->ldo_foreign_lmv_size = 0;
2343 lo->ldo_is_foreign = 0;
2347 * Declare create striped md object.
2349 * The function declares intention to create a striped directory. This is a
2350 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2351 * is to verify pattern \a lum_buf is good. Check that function for the details.
2353 * \param[in] env execution environment
2354 * \param[in] dt object
2355 * \param[in] attr attributes to initialize the objects with
2356 * \param[in] lum_buf a pattern specifying the number of stripes and
2358 * \param[in] dof type of objects to be created
2359 * \param[in] th transaction handle
2361 * \retval 0 on success
2362 * \retval negative if failed
2365 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2366 struct dt_object *dt,
2367 struct lu_attr *attr,
2368 const struct lu_buf *lum_buf,
2369 struct dt_object_format *dof,
2372 struct lod_object *lo = lod_dt_obj(dt);
2373 struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
2377 LASSERT(lum != NULL);
2380 "lum magic=%x hash=%x count=%u offset=%d inherit=%u rr=%u\n",
2381 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_hash_type),
2382 le32_to_cpu(lum->lum_stripe_count),
2383 (int)le32_to_cpu(lum->lum_stripe_offset),
2384 lum->lum_max_inherit, lum->lum_max_inherit_rr);
2386 if (lo->ldo_dir_stripe_count == 0) {
2387 if (lo->ldo_is_foreign) {
2388 rc = lod_alloc_foreign_lmv(lo, lum_buf->lb_len);
2391 memcpy(lo->ldo_foreign_lmv, lum, lum_buf->lb_len);
2392 lo->ldo_dir_stripe_loaded = 1;
2397 /* client replay striped directory creation with LMV, this happens when
2398 * all involved MDTs were rebooted, or MDT recovery was aborted.
2400 if (le32_to_cpu(lum->lum_magic) == LMV_MAGIC_V1)
2401 rc = lod_prep_md_replayed_create(env, dt, attr, lum_buf, dof,
2404 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2406 /* failed to create striping, let's reset
2407 * config so that others don't get confused */
2408 lod_striping_free(env, lo);
2414 * Set or replace striped directory layout, and LFSCK may set layout on a plain
2415 * directory, so don't check stripe count.
2417 * \param[in] env execution environment
2418 * \param[in] dt target object
2419 * \param[in] lmv_buf LMV buf which contains source stripe FIDs
2420 * \param[in] fl set or replace
2421 * \param[in] th transaction handle
2423 * \retval 0 on success
2424 * \retval negative if failed
2426 static int lod_dir_layout_set(const struct lu_env *env,
2427 struct dt_object *dt,
2428 const struct lu_buf *lmv_buf,
2432 struct dt_object *next = dt_object_child(dt);
2433 struct lod_object *lo = lod_dt_obj(dt);
2434 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2435 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
2436 struct lmv_mds_md_v1 *slave_lmv;
2437 struct lu_buf slave_buf;
2443 if (!lmv_is_sane2(lmv))
2446 /* adjust hash for dir merge, which may not be set in user command */
2447 if (lmv_is_merging(lmv) &&
2448 !(lmv->lmv_migrate_hash & LMV_HASH_TYPE_MASK))
2449 lmv->lmv_merge_hash |=
2450 lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern &
2453 LMV_DEBUG(D_INFO, lmv, "set");
2455 rc = lod_sub_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV, fl, th);
2459 /* directory restripe may update stripe LMV directly */
2460 if (!lo->ldo_dir_stripe_count)
2463 lo->ldo_dir_hash_type = le32_to_cpu(lmv->lmv_hash_type);
2464 lo->ldo_dir_migrate_offset = le32_to_cpu(lmv->lmv_migrate_offset);
2465 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_migrate_hash);
2466 lo->ldo_dir_layout_version = le32_to_cpu(lmv->lmv_layout_version);
2468 OBD_ALLOC_PTR(slave_lmv);
2472 lod_prep_slave_lmv_md(slave_lmv, lmv);
2473 slave_buf.lb_buf = slave_lmv;
2474 slave_buf.lb_len = sizeof(*slave_lmv);
2476 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2477 if (!lo->ldo_stripe[i])
2480 if (!dt_object_exists(lo->ldo_stripe[i]))
2483 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], &slave_buf,
2484 XATTR_NAME_LMV, fl, th);
2489 OBD_FREE_PTR(slave_lmv);
2495 * Implementation of dt_object_operations::do_declare_xattr_set.
2497 * Used with regular (non-striped) objects. Basically it
2498 * initializes the striping information and applies the
2499 * change to all the stripes.
2501 * \see dt_object_operations::do_declare_xattr_set() in the API description
2504 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2505 struct dt_object *dt,
2506 const struct lu_buf *buf,
2507 const char *name, int fl,
2510 struct dt_object *next = dt_object_child(dt);
2511 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2512 struct lod_object *lo = lod_dt_obj(dt);
2517 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2518 struct lmv_user_md_v1 *lum;
2520 LASSERT(buf != NULL && buf->lb_buf != NULL);
2522 rc = lod_verify_md_striping(d, lum);
2525 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2526 rc = lod_verify_striping(env, d, lo, buf, false);
2531 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2535 /* Note: Do not set LinkEA on sub-stripes, otherwise
2536 * it will confuse the fid2path process(see mdt_path_current()).
2537 * The linkEA between master and sub-stripes is set in
2538 * lod_xattr_set_lmv(). */
2539 if (strcmp(name, XATTR_NAME_LINK) == 0)
2542 /* set xattr to each stripes, if needed */
2543 rc = lod_striping_load(env, lo);
2547 if (lo->ldo_dir_stripe_count == 0)
2550 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2551 if (!lo->ldo_stripe[i])
2554 if (!dt_object_exists(lo->ldo_stripe[i]))
2557 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2567 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2568 struct lod_object *lo,
2569 struct dt_object *dt, struct thandle *th,
2570 int comp_idx, int stripe_idx,
2571 struct lod_obj_stripe_cb_data *data)
2573 struct lod_thread_info *info = lod_env_info(env);
2574 struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
2575 struct filter_fid *ff = &info->lti_ff;
2576 struct lu_buf *buf = &info->lti_buf;
2580 buf->lb_len = sizeof(*ff);
2581 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2589 * locd_buf is set if it's called by dir migration, which doesn't check
2592 if (data->locd_buf) {
2593 memset(ff, 0, sizeof(*ff));
2594 ff->ff_parent = *(struct lu_fid *)data->locd_buf->lb_buf;
2596 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
2598 if (lu_fid_eq(lod_object_fid(lo), &ff->ff_parent) &&
2599 ff->ff_layout.ol_comp_id == comp->llc_id)
2602 memset(ff, 0, sizeof(*ff));
2603 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2606 /* rewrite filter_fid */
2607 ff->ff_parent.f_ver = stripe_idx;
2608 ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
2609 ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
2610 ff->ff_layout.ol_comp_id = comp->llc_id;
2611 ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
2612 ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
2613 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
2615 if (data->locd_declare)
2616 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2617 LU_XATTR_REPLACE, th);
2619 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2620 LU_XATTR_REPLACE, th);
2626 * Reset parent FID on OST object
2628 * Replace parent FID with @dt object FID, which is only called during migration
2629 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2630 * the FID is changed.
2632 * \param[in] env execution environment
2633 * \param[in] dt dt_object whose stripes's parent FID will be reset
2634 * \parem[in] th thandle
2635 * \param[in] declare if it is declare
2637 * \retval 0 if reset succeeds
2638 * \retval negative errno if reset fails
2640 static int lod_replace_parent_fid(const struct lu_env *env,
2641 struct dt_object *dt,
2642 const struct lu_buf *buf,
2643 struct thandle *th, bool declare)
2645 struct lod_object *lo = lod_dt_obj(dt);
2646 struct lod_thread_info *info = lod_env_info(env);
2647 struct filter_fid *ff;
2648 struct lod_obj_stripe_cb_data data = { { 0 } };
2652 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2654 /* set xattr to each stripes, if needed */
2655 rc = lod_striping_load(env, lo);
2659 if (!lod_obj_is_striped(dt))
2662 if (info->lti_ea_store_size < sizeof(*ff)) {
2663 rc = lod_ea_store_resize(info, sizeof(*ff));
2668 data.locd_declare = declare;
2669 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2670 data.locd_buf = buf;
2671 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2676 __u16 lod_comp_entry_stripe_count(struct lod_object *lo, int comp_idx,
2679 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2680 struct lod_layout_component *entry;
2681 enum lod_uses_hint flags = LOD_USES_ASSIGNED_STRIPE;
2686 entry = &lo->ldo_comp_entries[comp_idx];
2687 if (lod_comp_inited(entry))
2688 return entry->llc_stripe_count;
2689 if (entry->llc_stripe_count == LOV_ALL_STRIPES)
2690 return lod_get_stripe_count_plain(lod, lo,
2691 entry->llc_stripe_count,
2692 entry->llc_pattern &
2693 LOV_PATTERN_OVERSTRIPING,
2696 return lod_get_stripe_count(lod, lo, comp_idx, entry->llc_stripe_count,
2697 entry->llc_pattern & LOV_PATTERN_OVERSTRIPING,
2701 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2703 int magic, size = 0, i;
2704 struct lod_layout_component *comp_entries;
2706 bool is_composite, is_foreign = false;
2709 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2710 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2712 lo->ldo_def_striping->lds_def_striping_is_composite;
2714 comp_cnt = lo->ldo_comp_cnt;
2715 comp_entries = lo->ldo_comp_entries;
2716 is_composite = lo->ldo_is_composite;
2717 is_foreign = lo->ldo_is_foreign;
2721 return lo->ldo_foreign_lov_size;
2723 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2725 size = sizeof(struct lov_comp_md_v1) +
2726 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2727 LASSERT(size % sizeof(__u64) == 0);
2730 for (i = 0; i < comp_cnt; i++) {
2733 if (comp_entries[i].llc_magic == LOV_MAGIC_FOREIGN) {
2734 size += lov_foreign_md_size(comp_entries[i].llc_length);
2736 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 :
2738 stripe_count = lod_comp_entry_stripe_count(lo, i,
2740 if (!is_dir && is_composite)
2741 lod_comp_shrink_stripe_count(&comp_entries[i],
2744 size += lov_user_md_size(stripe_count, magic);
2746 LASSERT(size % sizeof(__u64) == 0);
2752 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2753 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2756 * \param[in] env execution environment
2757 * \param[in] dt dt_object to add components on
2758 * \param[in] buf buffer contains components to be added
2759 * \parem[in] th thandle
2761 * \retval 0 on success
2762 * \retval negative errno on failure
2764 static int lod_declare_layout_add(const struct lu_env *env,
2765 struct dt_object *dt,
2766 const struct lu_buf *buf,
2769 struct lod_thread_info *info = lod_env_info(env);
2770 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2771 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2772 struct dt_object *next = dt_object_child(dt);
2773 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
2774 struct lod_object *lo = lod_dt_obj(dt);
2775 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2777 int i, rc, array_cnt, old_array_cnt;
2780 LASSERT(lo->ldo_is_composite);
2782 if (lo->ldo_flr_state != LCM_FL_NONE)
2785 rc = lod_verify_striping(env, d, lo, buf, false);
2789 magic = comp_v1->lcm_magic;
2790 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2791 lustre_swab_lov_comp_md_v1(comp_v1);
2792 magic = comp_v1->lcm_magic;
2795 if (magic != LOV_USER_MAGIC_COMP_V1)
2798 mutex_lock(&lo->ldo_layout_mutex);
2800 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2801 OBD_ALLOC_PTR_ARRAY(comp_array, array_cnt);
2802 if (comp_array == NULL) {
2803 mutex_unlock(&lo->ldo_layout_mutex);
2808 memcpy(comp_array, lo->ldo_comp_entries,
2809 sizeof(*comp_array) * lo->ldo_comp_cnt);
2811 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2812 struct lov_user_md_v1 *v1;
2813 struct lu_extent *ext;
2815 v1 = (struct lov_user_md *)((char *)comp_v1 +
2816 comp_v1->lcm_entries[i].lcme_offset);
2817 ext = &comp_v1->lcm_entries[i].lcme_extent;
2819 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2820 lod_comp->llc_extent.e_start = ext->e_start;
2821 lod_comp->llc_extent.e_end = ext->e_end;
2822 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2823 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2825 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2826 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2828 * limit stripe count so that it's less than/equal to
2829 * extent_size / stripe_size.
2831 * Note: extension size reused llc_stripe_size field and
2832 * uninstantiated component could be defined with
2833 * extent_start == extent_end as extension component will
2836 if (!(lod_comp->llc_flags & LCME_FL_EXTENSION) &&
2837 (lod_comp_inited(lod_comp) ||
2838 lod_comp->llc_extent.e_start <
2839 lod_comp->llc_extent.e_end) &&
2840 lod_comp->llc_stripe_count != LOV_ALL_STRIPES &&
2841 ext->e_end != OBD_OBJECT_EOF &&
2842 (__u64)(lod_comp->llc_stripe_count *
2843 lod_comp->llc_stripe_size) >
2844 (ext->e_end - ext->e_start))
2845 lod_comp->llc_stripe_count =
2846 DIV_ROUND_UP(ext->e_end - ext->e_start,
2847 lod_comp->llc_stripe_size);
2848 lod_adjust_stripe_info(lod_comp, desc, 0);
2850 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2851 struct lov_user_md_v3 *v3 = (typeof(*v3) *) v1;
2853 if (v3->lmm_pool_name[0] != '\0' &&
2854 !lov_pool_is_ignored(v3->lmm_pool_name)) {
2855 rc = lod_set_pool(&lod_comp->llc_pool,
2863 old_array = lo->ldo_comp_entries;
2864 old_array_cnt = lo->ldo_comp_cnt;
2866 lo->ldo_comp_entries = comp_array;
2867 lo->ldo_comp_cnt = array_cnt;
2869 /* No need to increase layout generation here, it will be increased
2870 * later when generating component ID for the new components */
2872 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2873 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2874 XATTR_NAME_LOV, 0, th);
2876 lo->ldo_comp_entries = old_array;
2877 lo->ldo_comp_cnt = old_array_cnt;
2881 OBD_FREE_PTR_ARRAY(old_array, old_array_cnt);
2883 LASSERT(lo->ldo_mirror_count == 1);
2884 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2886 mutex_unlock(&lo->ldo_layout_mutex);
2891 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2892 lod_comp = &comp_array[i];
2893 if (lod_comp->llc_pool != NULL) {
2894 OBD_FREE(lod_comp->llc_pool,
2895 strlen(lod_comp->llc_pool) + 1);
2896 lod_comp->llc_pool = NULL;
2899 OBD_FREE_PTR_ARRAY(comp_array, array_cnt);
2900 mutex_unlock(&lo->ldo_layout_mutex);
2906 * lod_last_non_stale_mirror() - Check if a mirror is the last non-stale mirror.
2907 * @mirror_id: Mirror id to be checked.
2910 * This function checks if a mirror with specified @mirror_id is the last
2911 * non-stale mirror of a LOD object @lo.
2913 * Return: true or false.
2916 bool lod_last_non_stale_mirror(__u16 mirror_id, struct lod_object *lo)
2918 struct lod_layout_component *lod_comp;
2919 bool has_stale_flag;
2922 for (i = 0; i < lo->ldo_mirror_count; i++) {
2923 if (lo->ldo_mirrors[i].lme_id == mirror_id ||
2924 lo->ldo_mirrors[i].lme_stale)
2927 has_stale_flag = false;
2928 lod_foreach_mirror_comp(lod_comp, lo, i) {
2929 if (lod_comp->llc_flags & LCME_FL_STALE) {
2930 has_stale_flag = true;
2934 if (!has_stale_flag)
2942 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2943 * the '$field' can only be 'flags' now. The xattr value is binary
2944 * lov_comp_md_v1 which contains the component ID(s) and the value of
2945 * the field to be modified.
2946 * Please update allowed_lustre_lov macro if $field groks more values
2949 * \param[in] env execution environment
2950 * \param[in] dt dt_object to be modified
2951 * \param[in] op operation string, like "set.flags"
2952 * \param[in] buf buffer contains components to be set
2953 * \parem[in] th thandle
2955 * \retval 0 on success
2956 * \retval negative errno on failure
2958 static int lod_declare_layout_set(const struct lu_env *env,
2959 struct dt_object *dt,
2960 char *op, const struct lu_buf *buf,
2963 struct lod_layout_component *lod_comp;
2964 struct lod_thread_info *info = lod_env_info(env);
2965 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2966 struct lod_object *lo = lod_dt_obj(dt);
2967 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2970 bool changed = false;
2973 /* Please update allowed_lustre_lov macro if op
2974 * groks more values in the future
2976 if (strcmp(op, "set.flags") != 0) {
2977 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2978 lod2obd(d)->obd_name, op);
2982 magic = comp_v1->lcm_magic;
2983 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2984 lustre_swab_lov_comp_md_v1(comp_v1);
2985 magic = comp_v1->lcm_magic;
2988 if (magic != LOV_USER_MAGIC_COMP_V1)
2991 if (comp_v1->lcm_entry_count == 0) {
2992 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2993 lod2obd(d)->obd_name);
2997 mutex_lock(&lo->ldo_layout_mutex);
2998 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2999 __u32 id = comp_v1->lcm_entries[i].lcme_id;
3000 __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
3001 __u32 mirror_flag = flags & LCME_MIRROR_FLAGS;
3002 __u16 mirror_id = mirror_id_of(id);
3003 bool neg = flags & LCME_FL_NEG;
3005 if (flags & LCME_FL_INIT) {
3007 lod_striping_free_nolock(env, lo);
3008 mutex_unlock(&lo->ldo_layout_mutex);
3012 flags &= ~(LCME_MIRROR_FLAGS | LCME_FL_NEG);
3013 for (j = 0; j < lo->ldo_comp_cnt; j++) {
3014 lod_comp = &lo->ldo_comp_entries[j];
3016 /* lfs only put one flag in each entry */
3017 if ((flags && id != lod_comp->llc_id) ||
3018 (mirror_flag && mirror_id !=
3019 mirror_id_of(lod_comp->llc_id)))
3024 lod_comp->llc_flags &= ~flags;
3026 lod_comp->llc_flags &= ~mirror_flag;
3029 if ((flags & LCME_FL_STALE) &&
3030 lod_last_non_stale_mirror(mirror_id,
3033 &lo->ldo_layout_mutex);
3036 lod_comp->llc_flags |= flags;
3039 lod_comp->llc_flags |= mirror_flag;
3040 if (mirror_flag & LCME_FL_NOSYNC)
3041 lod_comp->llc_timestamp =
3042 ktime_get_real_seconds();
3048 mutex_unlock(&lo->ldo_layout_mutex);
3051 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
3052 lod2obd(d)->obd_name);
3056 lod_obj_inc_layout_gen(lo);
3058 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3059 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
3060 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3065 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
3066 * and the xattr value is a unique component ID or a special lcme_id.
3068 * \param[in] env execution environment
3069 * \param[in] dt dt_object to be operated on
3070 * \param[in] buf buffer contains component ID or lcme_id
3071 * \parem[in] th thandle
3073 * \retval 0 on success
3074 * \retval negative errno on failure
3076 static int lod_declare_layout_del(const struct lu_env *env,
3077 struct dt_object *dt,
3078 const struct lu_buf *buf,
3081 struct lod_thread_info *info = lod_env_info(env);
3082 struct dt_object *next = dt_object_child(dt);
3083 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3084 struct lod_object *lo = lod_dt_obj(dt);
3085 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3086 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3087 __u32 magic, id, flags, neg_flags = 0;
3091 LASSERT(lo->ldo_is_composite);
3093 if (lo->ldo_flr_state != LCM_FL_NONE)
3096 magic = comp_v1->lcm_magic;
3097 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
3098 lustre_swab_lov_comp_md_v1(comp_v1);
3099 magic = comp_v1->lcm_magic;
3102 if (magic != LOV_USER_MAGIC_COMP_V1)
3105 id = comp_v1->lcm_entries[0].lcme_id;
3106 flags = comp_v1->lcm_entries[0].lcme_flags;
3108 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
3109 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
3110 lod2obd(d)->obd_name, id, flags);
3114 if (id != LCME_ID_INVAL && flags != 0) {
3115 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
3116 lod2obd(d)->obd_name);
3120 if (id == LCME_ID_INVAL && !flags) {
3121 CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
3122 lod2obd(d)->obd_name);
3126 if (flags & LCME_FL_NEG) {
3127 neg_flags = flags & ~LCME_FL_NEG;
3131 mutex_lock(&lo->ldo_layout_mutex);
3133 left = lo->ldo_comp_cnt;
3135 mutex_unlock(&lo->ldo_layout_mutex);
3139 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3140 struct lod_layout_component *lod_comp;
3142 lod_comp = &lo->ldo_comp_entries[i];
3144 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
3146 else if (flags && !(flags & lod_comp->llc_flags))
3148 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
3151 if (left != (i + 1)) {
3152 CDEBUG(D_LAYOUT, "%s: this deletion will create "
3153 "a hole.\n", lod2obd(d)->obd_name);
3154 mutex_unlock(&lo->ldo_layout_mutex);
3159 /* Mark the component as deleted */
3160 lod_comp->llc_id = LCME_ID_INVAL;
3162 /* Not instantiated component */
3163 if (lod_comp->llc_stripe == NULL)
3166 LASSERT(lod_comp->llc_stripe_count > 0);
3167 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
3168 struct dt_object *obj = lod_comp->llc_stripe[j];
3172 rc = lod_sub_declare_destroy(env, obj, th);
3174 mutex_unlock(&lo->ldo_layout_mutex);
3180 LASSERTF(left >= 0, "left = %d\n", left);
3181 if (left == lo->ldo_comp_cnt) {
3182 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
3183 lod2obd(d)->obd_name, id);
3184 mutex_unlock(&lo->ldo_layout_mutex);
3188 mutex_unlock(&lo->ldo_layout_mutex);
3190 memset(attr, 0, sizeof(*attr));
3191 attr->la_valid = LA_SIZE;
3192 rc = lod_sub_declare_attr_set(env, next, attr, th);
3197 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3198 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
3199 XATTR_NAME_LOV, 0, th);
3201 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
3208 * Declare layout add/set/del operations issued by special xattr names:
3210 * XATTR_LUSTRE_LOV.add add component(s) to existing file
3211 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
3212 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
3214 * \param[in] env execution environment
3215 * \param[in] dt object
3216 * \param[in] name name of xattr
3217 * \param[in] buf lu_buf contains xattr value
3218 * \param[in] th transaction handle
3220 * \retval 0 on success
3221 * \retval negative if failed
3223 static int lod_declare_modify_layout(const struct lu_env *env,
3224 struct dt_object *dt,
3226 const struct lu_buf *buf,
3229 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3230 struct lod_object *lo = lod_dt_obj(dt);
3232 int rc, len = strlen(XATTR_LUSTRE_LOV);
3235 LASSERT(dt_object_exists(dt));
3237 if (strlen(name) <= len || name[len] != '.') {
3238 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
3239 lod2obd(d)->obd_name, name);
3244 rc = lod_striping_load(env, lo);
3248 /* the layout to be modified must be a composite layout */
3249 if (!lo->ldo_is_composite) {
3250 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
3251 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3252 GOTO(unlock, rc = -EINVAL);
3255 op = (char *)name + len;
3256 if (strcmp(op, "add") == 0) {
3257 rc = lod_declare_layout_add(env, dt, buf, th);
3258 } else if (strcmp(op, "del") == 0) {
3259 rc = lod_declare_layout_del(env, dt, buf, th);
3260 } else if (strncmp(op, "set", strlen("set")) == 0) {
3261 rc = lod_declare_layout_set(env, dt, op, buf, th);
3263 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
3264 lod2obd(d)->obd_name, name);
3265 GOTO(unlock, rc = -ENOTSUPP);
3269 lod_striping_free(env, lo);
3275 * Convert a plain file lov_mds_md to a composite layout.
3277 * \param[in,out] info the thread info::lti_ea_store buffer contains little
3278 * endian plain file layout
3280 * \retval 0 on success, <0 on failure
3282 static int lod_layout_convert(struct lod_thread_info *info)
3284 struct lov_mds_md *lmm = info->lti_ea_store;
3285 struct lov_mds_md *lmm_save;
3286 struct lov_comp_md_v1 *lcm;
3287 struct lov_comp_md_entry_v1 *lcme;
3293 /* realloc buffer to a composite layout which contains one component */
3294 blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3295 le32_to_cpu(lmm->lmm_magic));
3296 size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
3298 OBD_ALLOC_LARGE(lmm_save, blob_size);
3300 GOTO(out, rc = -ENOMEM);
3302 memcpy(lmm_save, lmm, blob_size);
3304 if (info->lti_ea_store_size < size) {
3305 rc = lod_ea_store_resize(info, size);
3310 lcm = info->lti_ea_store;
3311 memset(lcm, 0, sizeof(*lcm) + sizeof(*lcme));
3312 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
3313 lcm->lcm_size = cpu_to_le32(size);
3314 lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
3315 lmm_save->lmm_layout_gen));
3316 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
3317 lcm->lcm_entry_count = cpu_to_le16(1);
3319 lcme = &lcm->lcm_entries[0];
3320 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
3321 lcme->lcme_extent.e_start = 0;
3322 lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
3323 lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
3324 lcme->lcme_size = cpu_to_le32(blob_size);
3326 memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
3331 OBD_FREE_LARGE(lmm_save, blob_size);
3336 * Merge layouts to form a mirrored file.
3338 static int lod_declare_layout_merge(const struct lu_env *env,
3339 struct dt_object *dt,
3340 const struct lu_buf *mbuf,
3343 struct lod_thread_info *info = lod_env_info(env);
3344 struct lu_attr *layout_attr = &info->lti_layout_attr;
3345 struct lu_buf *buf = &info->lti_buf;
3346 struct lod_object *lo = lod_dt_obj(dt);
3347 struct lov_comp_md_v1 *lcm;
3348 struct lov_comp_md_v1 *cur_lcm;
3349 struct lov_comp_md_v1 *merge_lcm;
3350 struct lov_comp_md_entry_v1 *lcme;
3351 struct lov_mds_md_v1 *lmm;
3354 __u16 cur_entry_count;
3355 __u16 merge_entry_count;
3357 __u16 mirror_id = 0;
3364 merge_lcm = mbuf->lb_buf;
3365 if (mbuf->lb_len < sizeof(*merge_lcm))
3368 /* must be an existing layout from disk */
3369 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
3372 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
3374 /* do not allow to merge two mirrored files */
3375 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
3378 /* verify the target buffer */
3379 rc = lod_get_lov_ea(env, lo);
3381 RETURN(rc ? : -ENODATA);
3383 cur_lcm = info->lti_ea_store;
3384 switch (le32_to_cpu(cur_lcm->lcm_magic)) {
3387 rc = lod_layout_convert(info);
3389 case LOV_MAGIC_COMP_V1:
3399 /* info->lti_ea_store could be reallocated in lod_layout_convert() */
3400 cur_lcm = info->lti_ea_store;
3401 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
3403 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
3404 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
3405 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
3408 /* size of new layout */
3409 size = le32_to_cpu(cur_lcm->lcm_size) +
3410 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
3412 memset(buf, 0, sizeof(*buf));
3413 lu_buf_alloc(buf, size);
3414 if (buf->lb_buf == NULL)
3418 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
3420 offset = sizeof(*lcm) +
3421 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
3422 for (i = 0; i < cur_entry_count; i++) {
3423 struct lov_comp_md_entry_v1 *cur_lcme;
3425 lcme = &lcm->lcm_entries[i];
3426 cur_lcme = &cur_lcm->lcm_entries[i];
3428 lcme->lcme_offset = cpu_to_le32(offset);
3429 memcpy((char *)lcm + offset,
3430 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
3431 le32_to_cpu(lcme->lcme_size));
3433 offset += le32_to_cpu(lcme->lcme_size);
3435 if (mirror_count == 1 &&
3436 mirror_id_of(le32_to_cpu(lcme->lcme_id)) == 0) {
3437 /* Add mirror from a non-flr file, create new mirror ID.
3438 * Otherwise, keep existing mirror's component ID, used
3439 * for mirror extension.
3441 id = pflr_id(1, i + 1);
3442 lcme->lcme_id = cpu_to_le32(id);
3445 id = max(le32_to_cpu(lcme->lcme_id), id);
3448 mirror_id = mirror_id_of(id) + 1;
3450 /* check if first entry in new layout is DOM */
3451 lmm = (struct lov_mds_md_v1 *)((char *)merge_lcm +
3452 merge_lcm->lcm_entries[0].lcme_offset);
3453 merge_has_dom = lov_pattern(le32_to_cpu(lmm->lmm_pattern)) &
3456 for (i = 0; i < merge_entry_count; i++) {
3457 struct lov_comp_md_entry_v1 *merge_lcme;
3459 merge_lcme = &merge_lcm->lcm_entries[i];
3460 lcme = &lcm->lcm_entries[cur_entry_count + i];
3462 *lcme = *merge_lcme;
3463 lcme->lcme_offset = cpu_to_le32(offset);
3464 if (merge_has_dom && i == 0)
3465 lcme->lcme_flags |= cpu_to_le32(LCME_FL_STALE);
3467 id = pflr_id(mirror_id, i + 1);
3468 lcme->lcme_id = cpu_to_le32(id);
3470 memcpy((char *)lcm + offset,
3471 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
3472 le32_to_cpu(lcme->lcme_size));
3474 offset += le32_to_cpu(lcme->lcme_size);
3477 /* fixup layout information */
3478 lcm->lcm_size = cpu_to_le32(size);
3479 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
3480 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
3481 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NONE)
3482 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
3484 rc = lod_striping_reload(env, lo, buf, 0);
3488 lod_obj_inc_layout_gen(lo);
3489 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3491 /* transfer layout version to OST objects. */
3492 if (lo->ldo_mirror_count > 1) {
3493 struct lod_obj_stripe_cb_data data = { {0} };
3495 layout_attr->la_valid = LA_LAYOUT_VERSION;
3496 layout_attr->la_layout_version = 0;
3497 data.locd_attr = layout_attr;
3498 data.locd_declare = true;
3499 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
3500 rc = lod_obj_for_each_stripe(env, lo, th, &data);
3505 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
3506 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3514 * Split layouts, just set the LOVEA with the layout from mbuf.
3516 static int lod_declare_layout_split(const struct lu_env *env,
3517 struct dt_object *dt, const struct lu_buf *mbuf,
3520 struct lod_thread_info *info = lod_env_info(env);
3521 struct lu_attr *layout_attr = &info->lti_layout_attr;
3522 struct lod_object *lo = lod_dt_obj(dt);
3523 struct lov_comp_md_v1 *lcm = mbuf->lb_buf;
3527 rc = lod_striping_reload(env, lo, mbuf, LVF_ALL_STALE);
3531 lod_obj_inc_layout_gen(lo);
3532 /* fix on-disk layout gen */
3533 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3535 /* transfer layout version to OST objects. */
3536 if (lo->ldo_mirror_count > 1) {
3537 struct lod_obj_stripe_cb_data data = { {0} };
3539 layout_attr->la_valid = LA_LAYOUT_VERSION;
3540 layout_attr->la_layout_version = 0;
3541 data.locd_attr = layout_attr;
3542 data.locd_declare = true;
3543 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
3544 rc = lod_obj_for_each_stripe(env, lo, th, &data);
3549 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), mbuf,
3550 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3554 static int lod_layout_declare_or_purge_mirror(const struct lu_env *env,
3555 struct dt_object *dt, const struct lu_buf *buf,
3556 struct thandle *th, bool declare)
3558 struct lod_thread_info *info = lod_env_info(env);
3559 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3560 struct lod_object *lo = lod_dt_obj(dt);
3561 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3562 struct lov_comp_md_entry_v1 *entry;
3563 struct lov_mds_md_v1 *lmm;
3564 struct dt_object **sub_objs = NULL;
3565 int rc = 0, i, k, array_count = 0;
3570 * other ops (like lod_declare_destroy) could destroying sub objects
3573 mutex_lock(&lo->ldo_layout_mutex);
3576 /* prepare sub-objects array */
3577 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
3578 entry = &comp_v1->lcm_entries[i];
3580 if (!(entry->lcme_flags & LCME_FL_INIT))
3583 lmm = (struct lov_mds_md_v1 *)
3584 ((char *)comp_v1 + entry->lcme_offset);
3585 array_count += lmm->lmm_stripe_count;
3587 OBD_ALLOC_PTR_ARRAY(sub_objs, array_count);
3588 if (sub_objs == NULL) {
3589 mutex_unlock(&lo->ldo_layout_mutex);
3594 k = 0; /* sub_objs index */
3595 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
3596 struct lov_ost_data_v1 *objs;
3597 struct lu_object *o, *n;
3598 struct dt_object *dto;
3599 struct lu_device *nd;
3600 struct lov_mds_md_v3 *v3;
3604 entry = &comp_v1->lcm_entries[i];
3606 if (!(entry->lcme_flags & LCME_FL_INIT))
3609 lmm = (struct lov_mds_md_v1 *)
3610 ((char *)comp_v1 + entry->lcme_offset);
3611 v3 = (struct lov_mds_md_v3 *)lmm;
3612 if (lmm->lmm_magic == LOV_MAGIC_V3)
3613 objs = &v3->lmm_objects[0];
3615 objs = &lmm->lmm_objects[0];
3617 for (j = 0; j < lmm->lmm_stripe_count; j++) {
3618 idx = objs[j].l_ost_idx;
3619 rc = ostid_to_fid(&info->lti_fid, &objs[j].l_ost_oi,
3624 if (!fid_is_sane(&info->lti_fid)) {
3625 CERROR("%s: sub-object insane fid "DFID"\n",
3626 lod2obd(d)->obd_name,
3627 PFID(&info->lti_fid));
3628 GOTO(out, rc = -EINVAL);
3631 lod_getref(&d->lod_ost_descs);
3633 rc = validate_lod_and_idx(d, idx);
3635 lod_putref(d, &d->lod_ost_descs);
3639 nd = &OST_TGT(d, idx)->ltd_tgt->dd_lu_dev;
3640 lod_putref(d, &d->lod_ost_descs);
3642 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
3644 GOTO(out, rc = PTR_ERR(o));
3646 n = lu_object_locate(o->lo_header, nd->ld_type);
3648 lu_object_put(env, n);
3649 GOTO(out, rc = -ENOENT);
3652 dto = container_of(n, struct dt_object, do_lu);
3655 rc = lod_sub_declare_destroy(env, dto, th);
3656 dt_object_put(env, dto);
3661 * collect to-be-destroyed sub objects, the
3662 * reference would be released after actual
3668 } /* for each stripe */
3669 } /* for each component in the mirror */
3674 /* destroy the sub objects */
3675 for (; i < k; i++) {
3676 rc = lod_sub_destroy(env, sub_objs[i], th);
3679 dt_object_put(env, sub_objs[i]);
3683 * if a sub object destroy failed, we'd release sub objects
3684 * reference get from above sub_objs collection.
3687 dt_object_put(env, sub_objs[i]);
3689 OBD_FREE_PTR_ARRAY(sub_objs, array_count);
3691 mutex_unlock(&lo->ldo_layout_mutex);
3697 * Purge layouts, delete sub objects in the mirror stored in the vic_buf,
3698 * and set the LOVEA with the layout from mbuf.
3700 static int lod_declare_layout_purge(const struct lu_env *env,
3701 struct dt_object *dt, const struct lu_buf *buf,
3704 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3705 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3710 if (le32_to_cpu(comp_v1->lcm_magic) != LOV_MAGIC_COMP_V1) {
3711 CERROR("%s: invalid layout magic %#x != %#x\n",
3712 lod2obd(d)->obd_name, le32_to_cpu(comp_v1->lcm_magic),
3717 if (cpu_to_le32(LOV_MAGIC_COMP_V1) != LOV_MAGIC_COMP_V1)
3718 lustre_swab_lov_comp_md_v1(comp_v1);
3720 /* from now on, @buf contains cpu endian data */
3722 if (comp_v1->lcm_mirror_count != 0) {
3723 CERROR("%s: can only purge one mirror from "DFID"\n",
3724 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3728 /* delcare sub objects deletion in the mirror stored in @buf */
3729 rc = lod_layout_declare_or_purge_mirror(env, dt, buf, th, true);
3733 /* delete sub objects from the mirror stored in @buf */
3734 static int lod_layout_purge(const struct lu_env *env, struct dt_object *dt,
3735 const struct lu_buf *buf, struct thandle *th)
3740 rc = lod_layout_declare_or_purge_mirror(env, dt, buf, th, false);
3745 * Implementation of dt_object_operations::do_declare_xattr_set.
3747 * \see dt_object_operations::do_declare_xattr_set() in the API description
3750 * the extension to the API:
3751 * - declaring LOVEA requests striping creation
3752 * - LU_XATTR_REPLACE means layout swap
3754 static int lod_declare_xattr_set(const struct lu_env *env,
3755 struct dt_object *dt,
3756 const struct lu_buf *buf,
3757 const char *name, int fl,
3760 struct dt_object *next = dt_object_child(dt);
3761 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3766 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
3767 if ((S_ISREG(mode) || mode == 0) &&
3768 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE | LU_XATTR_SPLIT |
3770 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3771 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
3773 * this is a request to create object's striping.
3775 * allow to declare predefined striping on a new (!mode) object
3776 * which is supposed to be replay of regular file creation
3777 * (when LOV setting is declared)
3779 * LU_XATTR_REPLACE is set to indicate a layout swap
3781 if (dt_object_exists(dt)) {
3782 rc = dt_attr_get(env, next, attr);
3786 memset(attr, 0, sizeof(*attr));
3787 attr->la_valid = LA_TYPE | LA_MODE;
3788 attr->la_mode = S_IFREG;
3790 rc = lod_declare_striped_create(env, dt, attr, buf, th);
3791 } else if (fl & LU_XATTR_MERGE) {
3792 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3793 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3794 rc = lod_declare_layout_merge(env, dt, buf, th);
3795 } else if (fl & LU_XATTR_SPLIT) {
3796 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3797 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3798 rc = lod_declare_layout_split(env, dt, buf, th);
3799 } else if (fl & LU_XATTR_PURGE) {
3800 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3801 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3802 rc = lod_declare_layout_purge(env, dt, buf, th);
3803 } else if (S_ISREG(mode) &&
3804 strlen(name) >= sizeof(XATTR_LUSTRE_LOV) + 3 &&
3805 allowed_lustre_lov(name)) {
3807 * this is a request to modify object's striping.
3808 * add/set/del component(s).
3810 if (!dt_object_exists(dt))
3813 rc = lod_declare_modify_layout(env, dt, name, buf, th);
3814 } else if (S_ISDIR(mode)) {
3815 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
3816 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3817 rc = lod_replace_parent_fid(env, dt, buf, th, true);
3819 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
3826 * Apply xattr changes to the object.
3828 * Applies xattr changes to the object and the stripes if the latter exist.
3830 * \param[in] env execution environment
3831 * \param[in] dt object
3832 * \param[in] buf buffer pointing to the new value of xattr
3833 * \param[in] name name of xattr
3834 * \param[in] fl flags
3835 * \param[in] th transaction handle
3837 * \retval 0 on success
3838 * \retval negative if failed
3840 static int lod_xattr_set_internal(const struct lu_env *env,
3841 struct dt_object *dt,
3842 const struct lu_buf *buf,
3843 const char *name, int fl,
3846 struct dt_object *next = dt_object_child(dt);
3847 struct lod_object *lo = lod_dt_obj(dt);
3852 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3853 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3856 /* Note: Do not set LinkEA on sub-stripes, otherwise
3857 * it will confuse the fid2path process(see mdt_path_current()).
3858 * The linkEA between master and sub-stripes is set in
3859 * lod_xattr_set_lmv(). */
3860 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
3863 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3864 if (!lo->ldo_stripe[i])
3867 if (!dt_object_exists(lo->ldo_stripe[i]))
3870 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
3880 * Delete an extended attribute.
3882 * Deletes specified xattr from the object and the stripes if the latter exist.
3884 * \param[in] env execution environment
3885 * \param[in] dt object
3886 * \param[in] name name of xattr
3887 * \param[in] th transaction handle
3889 * \retval 0 on success
3890 * \retval negative if failed
3892 static int lod_xattr_del_internal(const struct lu_env *env,
3893 struct dt_object *dt,
3894 const char *name, struct thandle *th)
3896 struct dt_object *next = dt_object_child(dt);
3897 struct lod_object *lo = lod_dt_obj(dt);
3903 rc = lod_sub_xattr_del(env, next, name, th);
3904 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3907 if (lo->ldo_dir_stripe_count == 0)
3910 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3911 if (!lo->ldo_stripe[i])
3914 if (!dt_object_exists(lo->ldo_stripe[i]))
3917 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3926 * Set default striping on a directory.
3928 * Sets specified striping on a directory object unless it matches the default
3929 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3930 * EA. This striping will be used when regular file is being created in this
3933 * \param[in] env execution environment
3934 * \param[in] dt the striped object
3935 * \param[in] buf buffer with the striping
3936 * \param[in] name name of EA
3937 * \param[in] fl xattr flag (see OSD API description)
3938 * \param[in] th transaction handle
3940 * \retval 0 on success
3941 * \retval negative if failed
3943 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
3944 struct dt_object *dt,
3945 const struct lu_buf *buf,
3946 const char *name, int fl,
3949 struct lov_user_md_v1 *lum;
3950 struct lov_user_md_v3 *v3 = NULL;
3951 const char *pool_name = NULL;
3956 LASSERT(buf != NULL && buf->lb_buf != NULL);
3959 switch (lum->lmm_magic) {
3960 case LOV_USER_MAGIC_SPECIFIC:
3961 case LOV_USER_MAGIC_V3:
3963 if (lov_pool_is_reserved(v3->lmm_pool_name))
3964 memset(v3->lmm_pool_name, 0, sizeof(v3->lmm_pool_name));
3965 else if (v3->lmm_pool_name[0] != '\0')
3966 pool_name = v3->lmm_pool_name;
3968 case LOV_USER_MAGIC_V1:
3969 /* if { size, offset, count } = { 0, -1, 0 } and no pool
3970 * (i.e. all default values specified) then delete default
3971 * striping from dir. */
3973 "set default striping: sz %u # %u offset %d %s %s\n",
3974 (unsigned)lum->lmm_stripe_size,
3975 (unsigned)lum->lmm_stripe_count,
3976 (int)lum->lmm_stripe_offset,
3977 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
3979 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
3980 lum->lmm_stripe_count,
3981 lum->lmm_stripe_offset,
3984 case LOV_USER_MAGIC_COMP_V1:
3986 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lum;
3987 struct lov_comp_md_entry_v1 *lcme;
3990 comp_cnt = le16_to_cpu(lcm->lcm_entry_count);
3991 for (i = 0; i < comp_cnt; i++) {
3992 lcme = &lcm->lcm_entries[i];
3993 if (lcme->lcme_flags & cpu_to_le32(LCME_FL_EXTENSION)) {
3994 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
4003 CERROR("Invalid magic %x\n", lum->lmm_magic);
4008 rc = lod_xattr_del_internal(env, dt, name, th);
4012 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4018 static int lod_get_default_lov_striping(const struct lu_env *env,
4019 struct lod_object *lo,
4020 struct lod_default_striping *lds,
4021 struct dt_allocation_hint *ah);
4024 * Helper function to convert compound layout to compound layout with
4027 * Copy lcm_entries array of \a src to \a tgt. Replace lov_user_md_v1
4028 * components of \a src with lov_user_md_v3 using \a pool.
4030 * \param[in] src source layout
4031 * \param[in] pool pool to use in \a tgt
4032 * \param[out] tgt target layout
4034 static void embed_pool_to_comp_v1(const struct lov_comp_md_v1 *src,
4036 struct lov_comp_md_v1 *tgt)
4039 struct lov_user_md_v1 *lum;
4040 struct lov_user_md_v3 *lum3;
4041 struct lov_comp_md_entry_v1 *entry;
4045 entry = tgt->lcm_entries;
4047 for (i = 0; i < le16_to_cpu(src->lcm_entry_count); i++, entry++) {
4048 *entry = src->lcm_entries[i];
4049 offset = le32_to_cpu(src->lcm_entries[i].lcme_offset);
4050 entry->lcme_offset = cpu_to_le32(offset + shift);
4052 lum = (struct lov_user_md_v1 *)((char *)src + offset);
4053 lum3 = (struct lov_user_md_v3 *)((char *)tgt + offset + shift);
4054 *(struct lov_user_md_v1 *)lum3 = *lum;
4055 if (lum->lmm_pattern & cpu_to_le32(LOV_PATTERN_MDT)) {
4056 lum3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
4058 lum3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4059 entry->lcme_size = cpu_to_le32(sizeof(*lum3));
4060 strlcpy(lum3->lmm_pool_name, pool,
4061 sizeof(lum3->lmm_pool_name));
4062 shift += sizeof(*lum3) - sizeof(*lum);
4068 * Set default striping on a directory.
4070 * Sets specified striping on a directory object unless it matches the default
4071 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
4072 * EA. This striping will be used when regular file is being created in this
4074 * If current default striping includes a pool but specifed striping
4075 * does not - retain the pool if it exists.
4077 * \param[in] env execution environment
4078 * \param[in] dt the striped object
4079 * \param[in] buf buffer with the striping
4080 * \param[in] name name of EA
4081 * \param[in] fl xattr flag (see OSD API description)
4082 * \param[in] th transaction handle
4084 * \retval 0 on success
4085 * \retval negative if failed
4087 static int lod_xattr_set_default_lov_on_dir(const struct lu_env *env,
4088 struct dt_object *dt,
4089 const struct lu_buf *buf,
4090 const char *name, int fl,
4093 struct lod_default_striping *lds = lod_lds_buf_get(env);
4094 struct lov_user_md_v1 *v1 = buf->lb_buf;
4095 char pool[LOV_MAXPOOLNAME + 1];
4101 /* get existing striping config */
4102 rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds, NULL);
4106 memset(pool, 0, sizeof(pool));
4107 if (lds->lds_def_striping_set == 1)
4108 lod_layout_get_pool(lds->lds_def_comp_entries,
4109 lds->lds_def_comp_cnt, pool,
4112 is_del = LOVEA_DELETE_VALUES(v1->lmm_stripe_size,
4113 v1->lmm_stripe_count,
4114 v1->lmm_stripe_offset,
4117 /* Retain the pool name if it is not given */
4118 if (v1->lmm_magic == LOV_USER_MAGIC_V1 && pool[0] != '\0' &&
4120 struct lod_thread_info *info = lod_env_info(env);
4121 struct lov_user_md_v3 *v3 = info->lti_ea_store;
4123 memset(v3, 0, sizeof(*v3));
4124 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4125 v3->lmm_pattern = cpu_to_le32(v1->lmm_pattern);
4126 v3->lmm_stripe_count = cpu_to_le32(v1->lmm_stripe_count);
4127 v3->lmm_stripe_offset = cpu_to_le32(v1->lmm_stripe_offset);
4128 v3->lmm_stripe_size = cpu_to_le32(v1->lmm_stripe_size);
4130 strlcpy(v3->lmm_pool_name, pool, sizeof(v3->lmm_pool_name));
4132 info->lti_buf.lb_buf = v3;
4133 info->lti_buf.lb_len = sizeof(*v3);
4134 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4136 } else if (v1->lmm_magic == LOV_USER_MAGIC_COMP_V1 &&
4137 pool[0] != '\0' && !is_del) {
4139 * try to retain the pool from default layout if the
4140 * specified component layout does not provide pool
4143 struct lod_thread_info *info = lod_env_info(env);
4144 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
4145 struct lov_comp_md_v1 *comp_v1p;
4146 struct lov_user_md_v1 *lum;
4150 struct lov_comp_md_entry_v1 *entry;
4153 entry_count = le16_to_cpu(comp_v1->lcm_entry_count);
4154 size = sizeof(*comp_v1) +
4155 entry_count * sizeof(comp_v1->lcm_entries[0]);
4156 entry = comp_v1->lcm_entries;
4157 for (i = 0; i < entry_count; i++, entry++) {
4158 offset = le32_to_cpu(entry->lcme_offset);
4159 lum = (struct lov_user_md_v1 *)((char *)comp_v1 +
4161 if (le32_to_cpu(lum->lmm_magic) != LOV_USER_MAGIC_V1)
4162 /* the i-th component includes pool info */
4164 if (lum->lmm_pattern & cpu_to_le32(LOV_PATTERN_MDT))
4165 size += sizeof(struct lov_user_md_v1);
4167 size += sizeof(struct lov_user_md_v3);
4170 if (i == entry_count) {
4172 * re-compose the layout to include the pool for
4175 if (info->lti_ea_store_size < size)
4176 rc = lod_ea_store_resize(info, size);
4179 comp_v1p = info->lti_ea_store;
4180 *comp_v1p = *comp_v1;
4181 comp_v1p->lcm_size = cpu_to_le32(size);
4182 embed_pool_to_comp_v1(comp_v1, pool, comp_v1p);
4184 info->lti_buf.lb_buf = comp_v1p;
4185 info->lti_buf.lb_len = size;
4186 rc = lod_xattr_set_lov_on_dir(env, dt,
4191 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl,
4195 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl, th);
4198 if (lds->lds_def_striping_set == 1 && lds->lds_def_comp_entries != NULL)
4199 lod_free_def_comp_entries(lds);
4205 * Set default striping on a directory object.
4207 * Sets specified striping on a directory object unless it matches the default
4208 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
4209 * EA. This striping will be used when a new directory is being created in the
4212 * \param[in] env execution environment
4213 * \param[in] dt the striped object
4214 * \param[in] buf buffer with the striping
4215 * \param[in] name name of EA
4216 * \param[in] fl xattr flag (see OSD API description)
4217 * \param[in] th transaction handle
4219 * \retval 0 on success
4220 * \retval negative if failed
4222 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
4223 struct dt_object *dt,
4224 const struct lu_buf *buf,
4225 const char *name, int fl,
4228 struct lmv_user_md_v1 *lum;
4233 LASSERT(buf != NULL && buf->lb_buf != NULL);
4237 "set default stripe_count # %u stripe_offset %d hash %u\n",
4238 le32_to_cpu(lum->lum_stripe_count),
4239 (int)le32_to_cpu(lum->lum_stripe_offset),
4240 le32_to_cpu(lum->lum_hash_type));
4242 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
4243 le32_to_cpu(lum->lum_stripe_offset)) &&
4244 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
4245 rc = lod_xattr_del_internal(env, dt, name, th);
4249 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4258 * Turn directory into a striped directory.
4260 * During replay the client sends the striping created before MDT
4261 * failure, then the layer above LOD sends this defined striping
4262 * using ->do_xattr_set(), so LOD uses this method to replay creation
4263 * of the stripes. Notice the original information for the striping
4264 * (#stripes, FIDs, etc) was transferred in declare path.
4266 * \param[in] env execution environment
4267 * \param[in] dt the striped object
4268 * \param[in] buf buf lmv_user_md for create, or lmv_mds_md for replay
4269 * \param[in] name not used currently
4270 * \param[in] fl xattr flag (see OSD API description)
4271 * \param[in] th transaction handle
4273 * \retval 0 on success
4274 * \retval negative if failed
4276 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
4277 const struct lu_buf *buf, const char *name,
4278 int fl, struct thandle *th)
4280 struct lod_object *lo = lod_dt_obj(dt);
4281 struct lod_thread_info *info = lod_env_info(env);
4282 struct lu_attr *attr = &info->lti_attr;
4283 struct dt_object_format *dof = &info->lti_format;
4284 struct lu_buf lmv_buf;
4285 struct lu_buf slave_lmv_buf;
4286 struct lmv_user_md *lum = buf->lb_buf;
4287 struct lmv_mds_md_v1 *lmm;
4288 struct lmv_mds_md_v1 *slave_lmm = NULL;
4289 struct dt_insert_rec *rec = &info->lti_dt_rec;
4294 /* lum is used to know whether it's replay */
4296 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4299 /* The stripes are supposed to be allocated in declare phase,
4300 * if there are no stripes being allocated, it will skip */
4301 if (lo->ldo_dir_stripe_count == 0) {
4302 if (lo->ldo_is_foreign) {
4303 rc = lod_sub_xattr_set(env, dt_object_child(dt), buf,
4304 XATTR_NAME_LMV, fl, th);
4311 rc = dt_attr_get(env, dt_object_child(dt), attr);
4315 attr->la_valid &= LA_ATIME | LA_MTIME | LA_CTIME | LA_FLAGS |
4316 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
4317 dof->dof_type = DFT_DIR;
4319 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
4322 lmm = lmv_buf.lb_buf;
4324 OBD_ALLOC_PTR(slave_lmm);
4325 if (slave_lmm == NULL)
4328 lod_prep_slave_lmv_md(slave_lmm, lmm);
4329 slave_lmv_buf.lb_buf = slave_lmm;
4330 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
4332 rec->rec_type = S_IFDIR;
4333 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4334 struct dt_object *dto = lo->ldo_stripe[i];
4335 char *stripe_name = info->lti_key;
4336 struct lu_name *sname;
4337 struct linkea_data ldata = { NULL };
4338 struct lu_buf linkea_buf;
4339 bool stripe_created = false;
4341 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
4345 /* fail a remote stripe creation */
4346 if (i && CFS_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_CREATE))
4349 /* if it's replay by client request, and stripe exists on remote
4350 * MDT, it means mkdir was partially executed: stripe was
4351 * created on remote MDT successfully, but target not in last
4354 if (unlikely((le32_to_cpu(lum->lum_magic) == LMV_MAGIC_V1) &&
4355 dt_object_exists(dto) && dt_object_remote(dto)))
4356 stripe_created = true;
4358 /* don't create stripe if:
4359 * 1. it's source stripe of migrating directory
4360 * 2. it's existed stripe of splitting directory
4362 if ((lod_is_migrating(lo) && i >= lo->ldo_dir_migrate_offset) ||
4363 (lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) {
4364 if (!dt_object_exists(dto))
4365 GOTO(out, rc = -EINVAL);
4366 } else if (!stripe_created) {
4367 dt_write_lock(env, dto, DT_TGT_CHILD);
4368 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
4370 dt_write_unlock(env, dto);
4374 rc = lod_sub_ref_add(env, dto, th);
4375 dt_write_unlock(env, dto);
4379 rec->rec_fid = lu_object_fid(&dto->do_lu);
4380 rc = lod_sub_insert(env, dto,
4381 (const struct dt_rec *)rec,
4382 (const struct dt_key *)dot, th);
4387 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
4388 cfs_fail_val != i) {
4389 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
4391 slave_lmm->lmv_master_mdt_index =
4394 slave_lmm->lmv_master_mdt_index =
4397 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
4398 XATTR_NAME_LMV, 0, th);
4403 /* don't insert stripe if it's existed stripe of splitting
4404 * directory (this directory is striped).
4405 * NB, plain directory will insert itself as the first
4408 if (lod_is_splitting(lo) && lo->ldo_dir_split_offset > 1 &&
4409 lo->ldo_dir_split_offset > i)
4412 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
4414 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4415 PFID(lu_object_fid(&dto->do_lu)), i + 1);
4417 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4418 PFID(lu_object_fid(&dto->do_lu)), i);
4420 if (!stripe_created) {
4421 rec->rec_fid = lu_object_fid(&dt->do_lu);
4422 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
4423 (const struct dt_key *)dotdot, th);
4427 sname = lod_name_get(env, stripe_name,
4428 strlen(stripe_name));
4429 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
4430 sname, lu_object_fid(&dt->do_lu));
4434 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
4435 linkea_buf.lb_len = ldata.ld_leh->leh_len;
4436 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
4437 XATTR_NAME_LINK, 0, th);
4442 rec->rec_fid = lu_object_fid(&dto->do_lu);
4443 rc = lod_sub_insert(env, dt_object_child(dt),
4444 (const struct dt_rec *)rec,
4445 (const struct dt_key *)stripe_name, th);
4449 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
4454 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
4455 rc = lod_sub_xattr_set(env, dt_object_child(dt),
4456 &lmv_buf, XATTR_NAME_LMV, fl, th);
4458 if (slave_lmm != NULL)
4459 OBD_FREE_PTR(slave_lmm);
4465 * Helper function to declare/execute creation of a striped directory
4467 * Called in declare/create object path, prepare striping for a directory
4468 * and prepare defaults data striping for the objects to be created in
4469 * that directory. Notice the function calls "declaration" or "execution"
4470 * methods depending on \a declare param. This is a consequence of the
4471 * current approach while we don't have natural distributed transactions:
4472 * we basically execute non-local updates in the declare phase. So, the
4473 * arguments for the both phases are the same and this is the reason for
4474 * this function to exist.
4476 * \param[in] env execution environment
4477 * \param[in] dt object
4478 * \param[in] attr attributes the stripes will be created with
4479 * \param[in] lmu lmv_user_md if MDT indices are specified
4480 * \param[in] dof format of stripes (see OSD API description)
4481 * \param[in] th transaction handle
4482 * \param[in] declare where to call "declare" or "execute" methods
4484 * \retval 0 on success
4485 * \retval negative if failed
4487 static int lod_dir_striping_create_internal(const struct lu_env *env,
4488 struct dt_object *dt,
4489 struct lu_attr *attr,
4490 const struct lu_buf *lmu,
4491 struct dt_object_format *dof,
4495 struct lod_thread_info *info = lod_env_info(env);
4496 struct lod_object *lo = lod_dt_obj(dt);
4497 const struct lod_default_striping *lds = lo->ldo_def_striping;
4501 LASSERT(ergo(lds != NULL,
4502 lds->lds_def_striping_set ||
4503 lds->lds_dir_def_striping_set));
4506 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
4507 lo->ldo_dir_stripe_offset)) {
4509 /* mkdir by default LMV */
4510 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
4511 int stripe_count = lo->ldo_dir_stripe_count;
4513 if (info->lti_ea_store_size < sizeof(*v1)) {
4514 rc = lod_ea_store_resize(info, sizeof(*v1));
4517 v1 = info->lti_ea_store;
4520 memset(v1, 0, sizeof(*v1));
4521 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
4522 v1->lum_stripe_count = cpu_to_le32(stripe_count);
4523 v1->lum_stripe_offset =
4524 cpu_to_le32(lo->ldo_dir_stripe_offset);
4526 info->lti_buf.lb_buf = v1;
4527 info->lti_buf.lb_len = sizeof(*v1);
4528 lmu = &info->lti_buf;
4532 rc = lod_declare_xattr_set_lmv(env, dt, attr, lmu, dof,
4535 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
4539 } else if (lmu->lb_buf) {
4540 /* foreign LMV EA case */
4542 struct lmv_foreign_md *lfm = lmu->lb_buf;
4544 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN)
4545 rc = lod_declare_xattr_set_lmv(env, dt, attr,
4547 } else if (lo->ldo_is_foreign) {
4548 LASSERT(lo->ldo_foreign_lmv != NULL &&
4549 lo->ldo_foreign_lmv_size > 0);
4550 info->lti_buf.lb_buf = lo->ldo_foreign_lmv;
4551 info->lti_buf.lb_len = lo->ldo_foreign_lmv_size;
4552 lmu = &info->lti_buf;
4553 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
4558 /* Transfer default LMV striping from the parent */
4559 if (lds != NULL && lds->lds_dir_def_striping_set &&
4560 lds->lds_dir_def_max_inherit != LMV_INHERIT_END &&
4561 lds->lds_dir_def_max_inherit != LMV_INHERIT_NONE &&
4562 !(LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
4563 lds->lds_dir_def_stripe_offset) &&
4564 le32_to_cpu(lds->lds_dir_def_hash_type) !=
4565 LMV_HASH_TYPE_UNKNOWN)) {
4566 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
4568 if (info->lti_ea_store_size < sizeof(*v1)) {
4569 rc = lod_ea_store_resize(info, sizeof(*v1));
4572 v1 = info->lti_ea_store;
4575 memset(v1, 0, sizeof(*v1));
4576 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
4577 v1->lum_stripe_count =
4578 cpu_to_le32(lds->lds_dir_def_stripe_count);
4579 v1->lum_stripe_offset =
4580 cpu_to_le32(lds->lds_dir_def_stripe_offset);
4582 cpu_to_le32(lds->lds_dir_def_hash_type);
4583 v1->lum_max_inherit =
4584 lmv_inherit_next(lds->lds_dir_def_max_inherit);
4585 v1->lum_max_inherit_rr =
4586 lmv_inherit_rr_next(lds->lds_dir_def_max_inherit_rr);
4588 info->lti_buf.lb_buf = v1;
4589 info->lti_buf.lb_len = sizeof(*v1);
4591 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4592 XATTR_NAME_DEFAULT_LMV,
4595 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
4597 XATTR_NAME_DEFAULT_LMV, 0,
4603 /* Transfer default LOV striping from the parent */
4604 if (lds != NULL && lds->lds_def_striping_set &&
4605 lds->lds_def_comp_cnt != 0) {
4606 struct lov_mds_md *lmm;
4607 int lmm_size = lod_comp_md_size(lo, true);
4609 if (info->lti_ea_store_size < lmm_size) {
4610 rc = lod_ea_store_resize(info, lmm_size);
4614 lmm = info->lti_ea_store;
4616 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
4620 info->lti_buf.lb_buf = lmm;
4621 info->lti_buf.lb_len = lmm_size;
4624 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4625 XATTR_NAME_LOV, 0, th);
4627 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4628 XATTR_NAME_LOV, 0, th);
4633 /* ldo_def_striping is not allocated, clear after use, in case directory
4634 * layout is changed later.
4637 lo->ldo_def_striping = NULL;
4642 static int lod_declare_dir_striping_create(const struct lu_env *env,
4643 struct dt_object *dt,
4644 struct lu_attr *attr,
4646 struct dt_object_format *dof,
4649 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
4653 static int lod_dir_striping_create(const struct lu_env *env,
4654 struct dt_object *dt,
4655 struct lu_attr *attr,
4656 const struct lu_buf *lmu,
4657 struct dt_object_format *dof,
4660 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
4665 * Make LOV EA for striped object.
4667 * Generate striping information and store it in the LOV EA of the given
4668 * object. The caller must ensure nobody else is calling the function
4669 * against the object concurrently. The transaction must be started.
4670 * FLDB service must be running as well; it's used to map FID to the target,
4671 * which is stored in LOV EA.
4673 * \param[in] env execution environment for this thread
4674 * \param[in] lo LOD object
4675 * \param[in] th transaction handle
4677 * \retval 0 if LOV EA is stored successfully
4678 * \retval negative error number on failure
4680 static int lod_generate_and_set_lovea(const struct lu_env *env,
4681 struct lod_object *lo,
4684 struct lod_thread_info *info = lod_env_info(env);
4685 struct dt_object *next = dt_object_child(&lo->ldo_obj);
4686 struct lov_mds_md_v1 *lmm;
4692 if (lo->ldo_comp_cnt == 0 && !lo->ldo_is_foreign) {
4693 lod_striping_free_nolock(env, lo);
4694 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
4698 lmm_size = lod_comp_md_size(lo, false);
4699 if (info->lti_ea_store_size < lmm_size) {
4700 rc = lod_ea_store_resize(info, lmm_size);
4704 lmm = info->lti_ea_store;
4706 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
4710 info->lti_buf.lb_buf = lmm;
4711 info->lti_buf.lb_len = lmm_size;
4712 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
4713 XATTR_NAME_LOV, 0, th);
4717 static __u32 lod_gen_component_id(struct lod_object *lo,
4718 int mirror_id, int comp_idx);
4721 * Repeat an existing component
4723 * Creates a new layout by replicating an existing component. Uses striping
4724 * policy from previous component as a template for the striping for the new
4727 * New component starts with zero length, will be extended (or removed) before
4728 * returning layout to client.
4730 * NB: Reallocates layout components array (lo->ldo_comp_entries), invalidating
4731 * any pre-existing pointers to components. Handle with care.
4733 * \param[in] env execution environment for this thread
4734 * \param[in,out] lo object to update the layout of
4735 * \param[in] index index of component to copy
4737 * \retval 0 on success
4738 * \retval negative errno on error
4740 static int lod_layout_repeat_comp(const struct lu_env *env,
4741 struct lod_object *lo, int index)
4743 struct lod_layout_component *lod_comp;
4744 struct lod_layout_component *new_comp = NULL;
4745 struct lod_layout_component *comp_array;
4746 int rc = 0, i, new_cnt = lo->ldo_comp_cnt + 1;
4751 lod_comp = &lo->ldo_comp_entries[index];
4752 LASSERT(lod_comp_inited(lod_comp) && lod_comp->llc_id != LCME_ID_INVAL);
4754 CDEBUG(D_LAYOUT, "repeating component %d\n", index);
4756 OBD_ALLOC_PTR_ARRAY(comp_array, new_cnt);
4757 if (comp_array == NULL)
4758 GOTO(out, rc = -ENOMEM);
4760 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4761 memcpy(&comp_array[i + offset], &lo->ldo_comp_entries[i],
4762 sizeof(*comp_array));
4764 /* Duplicate this component in to the next slot */
4766 new_comp = &comp_array[i + 1];
4767 memcpy(&comp_array[i + 1], &lo->ldo_comp_entries[i],
4768 sizeof(*comp_array));
4769 /* We must now skip this new component when copying */
4774 /* Set up copied component */
4775 new_comp->llc_flags &= ~LCME_FL_INIT;
4776 new_comp->llc_stripe = NULL;
4777 new_comp->llc_stripes_allocated = 0;
4778 new_comp->llc_ost_indices = NULL;
4779 new_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4780 /* for uninstantiated components, layout gen stores default stripe
4782 new_comp->llc_layout_gen = lod_comp->llc_stripe_offset;
4783 /* This makes the repeated component zero-length, placed at the end of
4784 * the preceding component */
4785 new_comp->llc_extent.e_start = new_comp->llc_extent.e_end;
4786 new_comp->llc_timestamp = lod_comp->llc_timestamp;
4787 new_comp->llc_pool = NULL;
4789 rc = lod_set_pool(&new_comp->llc_pool, lod_comp->llc_pool);
4793 if (new_comp->llc_ostlist.op_array) {
4794 __u32 *op_array = NULL;
4796 OBD_ALLOC(op_array, new_comp->llc_ostlist.op_size);
4798 GOTO(out, rc = -ENOMEM);
4799 memcpy(op_array, &new_comp->llc_ostlist.op_array,
4800 new_comp->llc_ostlist.op_size);
4801 new_comp->llc_ostlist.op_array = op_array;
4804 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
4805 lo->ldo_comp_entries = comp_array;
4806 lo->ldo_comp_cnt = new_cnt;
4808 /* Generate an id for the new component */
4809 mirror_id = mirror_id_of(new_comp->llc_id);
4810 new_comp->llc_id = LCME_ID_INVAL;
4811 new_comp->llc_id = lod_gen_component_id(lo, mirror_id, index + 1);
4812 if (new_comp->llc_id == LCME_ID_INVAL)
4813 GOTO(out, rc = -ERANGE);
4818 OBD_FREE_PTR_ARRAY(comp_array, new_cnt);
4823 static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
4827 /* clear memory region that will be used for layout change */
4828 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
4829 info->lti_count = 0;
4831 if (info->lti_comp_size >= comp_cnt)
4834 if (info->lti_comp_size > 0) {
4835 OBD_FREE_PTR_ARRAY(info->lti_comp_idx, info->lti_comp_size);
4836 info->lti_comp_size = 0;
4839 OBD_ALLOC_PTR_ARRAY(info->lti_comp_idx, comp_cnt);
4840 if (!info->lti_comp_idx)
4843 info->lti_comp_size = comp_cnt;
4848 * Prepare new layout minus deleted components
4850 * Removes components marked for deletion (LCME_ID_INVAL) by copying to a new
4851 * layout and skipping those components. Removes stripe objects if any exist.
4854 * Reallocates layout components array (lo->ldo_comp_entries), invalidating
4855 * any pre-existing pointers to components.
4857 * Caller is responsible for updating mirror end (ldo_mirror[].lme_end).
4859 * \param[in] env execution environment for this thread
4860 * \param[in,out] lo object to update the layout of
4861 * \param[in] th transaction handle for this operation
4863 * \retval # of components deleted
4864 * \retval negative errno on error
4866 static int lod_layout_del_prep_layout(const struct lu_env *env,
4867 struct lod_object *lo,
4870 struct lod_layout_component *lod_comp;
4871 struct lod_thread_info *info = lod_env_info(env);
4872 int rc = 0, i, j, deleted = 0;
4876 LASSERT(lo->ldo_is_composite);
4877 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
4879 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
4883 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4884 lod_comp = &lo->ldo_comp_entries[i];
4886 if (lod_comp->llc_id != LCME_ID_INVAL) {
4887 /* Build array of things to keep */
4888 info->lti_comp_idx[info->lti_count++] = i;
4892 if (lod_comp->llc_magic == LOV_MAGIC_FOREIGN)
4895 lod_obj_set_pool(lo, i, NULL);
4896 if (lod_comp->llc_ostlist.op_array) {
4897 OBD_FREE(lod_comp->llc_ostlist.op_array,
4898 lod_comp->llc_ostlist.op_size);
4899 lod_comp->llc_ostlist.op_array = NULL;
4900 lod_comp->llc_ostlist.op_size = 0;
4904 CDEBUG(D_LAYOUT, "deleting comp %d, left %d\n", i,
4905 lo->ldo_comp_cnt - deleted);
4907 /* No striping info for this component */
4908 if (lod_comp->llc_stripe == NULL)
4911 LASSERT(lod_comp->llc_stripe_count > 0);
4912 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4913 struct dt_object *obj = lod_comp->llc_stripe[j];
4918 /* components which are not init have no sub objects
4920 if (lod_comp_inited(lod_comp)) {
4921 rc = lod_sub_destroy(env, obj, th);
4926 lu_object_put(env, &obj->do_lu);
4927 lod_comp->llc_stripe[j] = NULL;
4929 OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe,
4930 lod_comp->llc_stripes_allocated);
4931 lod_comp->llc_stripe = NULL;
4932 OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices,
4933 lod_comp->llc_stripes_allocated);
4934 lod_comp->llc_ost_indices = NULL;
4935 lod_comp->llc_stripes_allocated = 0;
4938 /* info->lti_count has the amount of left components */
4939 LASSERTF(info->lti_count >= 0 && info->lti_count < lo->ldo_comp_cnt,
4940 "left = %d, lo->ldo_comp_cnt %d\n", (int)info->lti_count,
4941 (int)lo->ldo_comp_cnt);
4943 if (info->lti_count > 0) {
4944 struct lod_layout_component *comp_array;
4946 OBD_ALLOC_PTR_ARRAY(comp_array, info->lti_count);
4947 if (comp_array == NULL)
4948 GOTO(out, rc = -ENOMEM);
4950 for (i = 0; i < info->lti_count; i++) {
4951 memcpy(&comp_array[i],
4952 &lo->ldo_comp_entries[info->lti_comp_idx[i]],
4953 sizeof(*comp_array));
4956 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
4957 lo->ldo_comp_entries = comp_array;
4958 lo->ldo_comp_cnt = info->lti_count;
4960 lod_free_comp_entries(lo);
4965 return rc ? rc : deleted;
4969 * Delete layout component(s)
4971 * This function sets up the layout data in the env and does the setattrs
4972 * required to write out the new layout. The layout itself is modified in
4973 * lod_layout_del_prep_layout.
4975 * \param[in] env execution environment for this thread
4976 * \param[in] dt object
4977 * \param[in] th transaction handle
4979 * \retval 0 on success
4980 * \retval negative error number on failure
4982 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
4985 struct lod_object *lo = lod_dt_obj(dt);
4986 struct dt_object *next = dt_object_child(dt);
4987 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4990 LASSERT(lo->ldo_mirror_count == 1);
4992 mutex_lock(&lo->ldo_layout_mutex);
4994 rc = lod_layout_del_prep_layout(env, lo, th);
4998 /* Only do this if we didn't delete all components */
4999 if (lo->ldo_comp_cnt > 0) {
5000 lo->ldo_mirrors[0].lme_end = lo->ldo_comp_cnt - 1;
5001 lod_obj_inc_layout_gen(lo);
5004 LASSERT(dt_object_exists(dt));
5005 rc = dt_attr_get(env, next, attr);
5009 if (attr->la_size > 0) {
5011 attr->la_valid = LA_SIZE;
5012 rc = lod_sub_attr_set(env, next, attr, th);
5017 rc = lod_generate_and_set_lovea(env, lo, th);
5021 lod_striping_free_nolock(env, lo);
5023 mutex_unlock(&lo->ldo_layout_mutex);
5030 * Implementation of dt_object_operations::do_xattr_set.
5032 * Sets specified extended attribute on the object. Three types of EAs are
5034 * LOV EA - stores striping for a regular file or default striping (when set
5036 * LMV EA - stores a marker for the striped directories
5037 * DMV EA - stores default directory striping
5039 * When striping is applied to a non-striped existing object (this is called
5040 * late striping), then LOD notices the caller wants to turn the object into a
5041 * striped one. The stripe objects are created and appropriate EA is set:
5042 * LOV EA storing all the stripes directly or LMV EA storing just a small header
5043 * with striping configuration.
5045 * \see dt_object_operations::do_xattr_set() in the API description for details.
5047 static int lod_xattr_set(const struct lu_env *env,
5048 struct dt_object *dt, const struct lu_buf *buf,
5049 const char *name, int fl, struct thandle *th)
5051 struct dt_object *next = dt_object_child(dt);
5052 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
5053 struct lod_object *lo = lod_dt_obj(dt);
5054 struct lod_obj_stripe_cb_data data = { {0} };
5059 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
5060 !strcmp(name, XATTR_NAME_LMV)) {
5062 case LU_XATTR_CREATE:
5063 rc = lod_dir_striping_create(env, dt, NULL, buf, NULL,
5067 case LU_XATTR_REPLACE:
5068 rc = lod_dir_layout_set(env, dt, buf, fl, th);
5075 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
5076 strcmp(name, XATTR_NAME_LOV) == 0) {
5077 rc = lod_xattr_set_default_lov_on_dir(env, dt, buf, name, fl,
5080 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
5081 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
5083 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
5086 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
5087 (strcmp(name, XATTR_NAME_LOV) == 0 ||
5088 strcmp(name, XATTR_LUSTRE_LOV) == 0 ||
5089 allowed_lustre_lov(name))) {
5090 /* in case of lov EA swap, just set it
5091 * if not, it is a replay so check striping match what we
5092 * already have during req replay, declare_xattr_set()
5093 * defines striping, then create() does the work */
5094 if (fl & LU_XATTR_REPLACE) {
5095 /* free stripes, then update disk */
5096 lod_striping_free(env, lod_dt_obj(dt));
5098 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
5099 } else if (fl & LU_XATTR_SPLIT) {
5100 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
5104 rc = lod_striping_reload(env, lo, buf, LVF_ALL_STALE);
5108 if (lo->ldo_mirror_count > 1 &&
5109 layout_attr->la_valid & LA_LAYOUT_VERSION) {
5111 layout_attr->la_layout_version =
5113 data.locd_attr = layout_attr;
5114 data.locd_declare = false;
5115 data.locd_stripe_cb =
5116 lod_obj_stripe_attr_set_cb;
5117 rc = lod_obj_for_each_stripe(env, lo, th,
5122 } else if (fl & LU_XATTR_PURGE) {
5123 rc = lod_layout_purge(env, dt, buf, th);
5124 } else if (dt_object_remote(dt)) {
5125 /* This only happens during migration, see
5126 * mdd_migrate_create(), in which Master MDT will
5127 * create a remote target object, and only set
5128 * (migrating) stripe EA on the remote object,
5129 * and does not need creating each stripes. */
5130 rc = lod_sub_xattr_set(env, next, buf, name,
5132 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
5133 /* delete component(s) */
5134 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
5135 rc = lod_layout_del(env, dt, th);
5138 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
5139 * it's going to create create file with specified
5140 * component(s), the striping must have not being
5141 * cached in this case;
5143 * Otherwise, it's going to add/change component(s) to
5144 * an existing file, the striping must have been cached
5147 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
5148 !strcmp(name, XATTR_NAME_LOV),
5149 !lod_dt_obj(dt)->ldo_comp_cached));
5151 rc = lod_striped_create(env, dt, NULL, NULL, th);
5155 if (fl & LU_XATTR_MERGE && lo->ldo_mirror_count > 1 &&
5156 layout_attr->la_valid & LA_LAYOUT_VERSION) {
5157 /* mirror merge exec phase */
5158 layout_attr->la_layout_version =
5160 data.locd_attr = layout_attr;
5161 data.locd_declare = false;
5162 data.locd_stripe_cb =
5163 lod_obj_stripe_attr_set_cb;
5164 rc = lod_obj_for_each_stripe(env, lo, th,
5171 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
5172 rc = lod_replace_parent_fid(env, dt, buf, th, false);
5177 /* then all other xattr */
5178 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
5184 * Implementation of dt_object_operations::do_declare_xattr_del.
5186 * \see dt_object_operations::do_declare_xattr_del() in the API description
5189 static int lod_declare_xattr_del(const struct lu_env *env,
5190 struct dt_object *dt, const char *name,
5193 struct lod_object *lo = lod_dt_obj(dt);
5194 struct dt_object *next = dt_object_child(dt);
5199 rc = lod_sub_declare_xattr_del(env, next, name, th);
5203 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
5206 /* NB: don't delete stripe LMV, because when we do this, normally we
5207 * will remove stripes, besides, if directory LMV is corrupt, this will
5208 * prevent deleting its LMV and fixing it (via LFSCK).
5210 if (!strcmp(name, XATTR_NAME_LMV))
5213 rc = lod_striping_load(env, lo);
5217 if (lo->ldo_dir_stripe_count == 0)
5220 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5221 struct dt_object *dto = lo->ldo_stripe[i];
5226 if (!dt_object_exists(dto))
5229 rc = lod_sub_declare_xattr_del(env, dto, name, th);
5238 * Implementation of dt_object_operations::do_xattr_del.
5240 * If EA storing a regular striping is being deleted, then release
5241 * all the references to the stripe objects in core.
5243 * \see dt_object_operations::do_xattr_del() in the API description for details.
5245 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
5246 const char *name, struct thandle *th)
5252 if (!strcmp(name, XATTR_NAME_LOV) || !strcmp(name, XATTR_NAME_LMV))
5253 lod_striping_free(env, lod_dt_obj(dt));
5255 rc = lod_xattr_del_internal(env, dt, name, th);
5261 * Implementation of dt_object_operations::do_xattr_list.
5263 * \see dt_object_operations::do_xattr_list() in the API description
5266 static int lod_xattr_list(const struct lu_env *env,
5267 struct dt_object *dt, const struct lu_buf *buf)
5269 return dt_xattr_list(env, dt_object_child(dt), buf);
5272 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
5274 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
5278 * Copy OST list from layout provided by user.
5280 * \param[in] lod_comp layout_component to be filled
5281 * \param[in] v3 LOV EA V3 user data
5283 * \retval 0 on success
5284 * \retval negative if failed
5286 int lod_comp_copy_ost_lists(struct lod_layout_component *lod_comp,
5287 struct lov_user_md_v3 *v3)
5293 if (v3->lmm_stripe_offset == LOV_OFFSET_DEFAULT)
5294 v3->lmm_stripe_offset = v3->lmm_objects[0].l_ost_idx;
5296 if (lod_comp->llc_ostlist.op_array) {
5297 if (lod_comp->llc_ostlist.op_size >=
5298 v3->lmm_stripe_count * sizeof(__u32)) {
5299 lod_comp->llc_ostlist.op_count =
5300 v3->lmm_stripe_count;
5303 OBD_FREE(lod_comp->llc_ostlist.op_array,
5304 lod_comp->llc_ostlist.op_size);
5307 /* copy ost list from lmm */
5308 lod_comp->llc_ostlist.op_count = v3->lmm_stripe_count;
5309 lod_comp->llc_ostlist.op_size = v3->lmm_stripe_count * sizeof(__u32);
5310 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
5311 lod_comp->llc_ostlist.op_size);
5312 if (!lod_comp->llc_ostlist.op_array)
5315 for (j = 0; j < v3->lmm_stripe_count; j++) {
5316 lod_comp->llc_ostlist.op_array[j] =
5317 v3->lmm_objects[j].l_ost_idx;
5325 * Get default striping.
5327 * \param[in] env execution environment
5328 * \param[in] lo object
5329 * \param[out] lds default striping
5331 * \retval 0 on success
5332 * \retval negative if failed
5334 static int lod_get_default_lov_striping(const struct lu_env *env,
5335 struct lod_object *lo,
5336 struct lod_default_striping *lds,
5337 struct dt_allocation_hint *dah)
5339 struct lod_thread_info *info = lod_env_info(env);
5340 struct lov_user_md_v1 *v1 = NULL;
5341 struct lov_user_md_v3 *v3 = NULL;
5342 struct lov_comp_md_v1 *lcm = NULL;
5344 int append_stripe_count = dah != NULL ? dah->dah_append_stripe_count : 0;
5345 const char *append_pool = (dah != NULL &&
5346 dah->dah_append_pool != NULL &&
5347 dah->dah_append_pool[0] != '\0') ?
5348 dah->dah_append_pool : NULL;
5349 __u16 entry_count = 1;
5350 __u16 mirror_count = 0;
5351 bool want_composite = false;
5356 lds->lds_def_striping_set = 0;
5358 rc = lod_get_lov_ea(env, lo);
5362 if (rc < (typeof(rc))sizeof(struct lov_user_md))
5365 magic = *(__u32 *)info->lti_ea_store;
5366 if (magic == __swab32(LOV_USER_MAGIC_V1)) {
5367 lustre_swab_lov_user_md_v1(info->lti_ea_store);
5368 } else if (magic == __swab32(LOV_USER_MAGIC_V3)) {
5369 lustre_swab_lov_user_md_v3(info->lti_ea_store);
5370 } else if (magic == __swab32(LOV_USER_MAGIC_SPECIFIC)) {
5371 v3 = (struct lov_user_md_v3 *)info->lti_ea_store;
5372 lustre_swab_lov_user_md_v3(v3);
5373 lustre_swab_lov_user_md_objects(v3->lmm_objects,
5374 v3->lmm_stripe_count);
5375 } else if (magic == __swab32(LOV_USER_MAGIC_COMP_V1) ||
5376 magic == __swab32(LOV_USER_MAGIC_SEL)) {
5377 lustre_swab_lov_comp_md_v1(info->lti_ea_store);
5383 case LOV_USER_MAGIC_SPECIFIC:
5384 v1 = info->lti_ea_store;
5386 case LOV_MAGIC_COMP_V1:
5388 lcm = info->lti_ea_store;
5389 entry_count = lcm->lcm_entry_count;
5390 if (entry_count == 0)
5393 mirror_count = lcm->lcm_mirror_count + 1;
5394 want_composite = true;
5400 if (append_stripe_count != 0 || append_pool != NULL) {
5403 want_composite = false;
5406 /* realloc default comp entries if necessary */
5407 rc = lod_def_striping_comp_resize(lds, entry_count);
5411 lds->lds_def_comp_cnt = entry_count;
5412 lds->lds_def_striping_is_composite = want_composite;
5413 lds->lds_def_mirror_cnt = mirror_count;
5415 for (i = 0; i < entry_count; i++) {
5416 struct lod_layout_component *llc = &lds->lds_def_comp_entries[i];
5420 * reset llc values, llc_stripes is always NULL in the
5421 * default striping template, llc_pool will be reset
5422 * later below using lod_set_pool().
5424 * XXX At this point llc_pool may point to valid (!)
5425 * kmalloced strings from previous RPCs.
5427 memset(llc, 0, offsetof(typeof(*llc), llc_pool));
5430 v1 = (struct lov_user_md *)((char *)lcm +
5431 lcm->lcm_entries[i].lcme_offset);
5433 if (want_composite) {
5434 llc->llc_extent = lcm->lcm_entries[i].lcme_extent;
5435 /* We only inherit certain flags from the layout */
5436 llc->llc_flags = lcm->lcm_entries[i].lcme_flags &
5437 LCME_TEMPLATE_FLAGS;
5441 CDEBUG(D_LAYOUT, DFID" magic = %#08x, pattern = %#x, stripe_count = %hu, stripe_size = %u, stripe_offset = %hu, append_pool = '%s', append_stripe_count = %d\n",
5442 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
5445 v1->lmm_stripe_count,
5446 v1->lmm_stripe_size,
5447 v1->lmm_stripe_offset,
5449 append_stripe_count);
5451 if (!lov_pattern_supported(v1->lmm_pattern) &&
5452 !(v1->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
5453 lod_free_def_comp_entries(lds);
5457 llc->llc_stripe_count = v1->lmm_stripe_count;
5458 llc->llc_stripe_size = v1->lmm_stripe_size;
5459 llc->llc_stripe_offset = v1->lmm_stripe_offset;
5460 llc->llc_pattern = v1->lmm_pattern;
5462 if (append_stripe_count != 0 || append_pool != NULL)
5463 llc->llc_pattern = LOV_PATTERN_RAID0;
5465 if (append_stripe_count != 0)
5466 llc->llc_stripe_count = append_stripe_count;
5469 if (append_pool != NULL) {
5471 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
5472 /* XXX: sanity check here */
5473 v3 = (struct lov_user_md_v3 *)v1;
5474 if (v3->lmm_pool_name[0] != '\0')
5475 pool = v3->lmm_pool_name;
5478 lod_set_pool(&llc->llc_pool, pool);
5480 if (v1->lmm_magic == LOV_USER_MAGIC_SPECIFIC &&
5481 append_stripe_count == 0 &&
5482 append_pool == NULL) {
5483 v3 = (struct lov_user_md_v3 *)v1;
5484 rc = lod_comp_copy_ost_lists(llc, v3);
5487 } else if (llc->llc_ostlist.op_array &&
5488 llc->llc_ostlist.op_count) {
5489 for (j = 0; j < llc->llc_ostlist.op_count; j++)
5490 llc->llc_ostlist.op_array[j] = -1;
5491 llc->llc_ostlist.op_count = 0;
5495 lds->lds_def_striping_set = 1;
5499 static inline void lod_lum2lds(struct lod_default_striping *lds,
5500 const struct lmv_user_md *lum)
5502 lds->lds_dir_def_stripe_count = le32_to_cpu(lum->lum_stripe_count);
5503 lds->lds_dir_def_stripe_offset = le32_to_cpu(lum->lum_stripe_offset);
5504 lds->lds_dir_def_hash_type = le32_to_cpu(lum->lum_hash_type);
5505 lds->lds_dir_def_max_inherit = lum->lum_max_inherit;
5506 lds->lds_dir_def_max_inherit_rr = lum->lum_max_inherit_rr;
5507 lds->lds_dir_def_striping_set = 1;
5511 * Get default directory striping.
5513 * \param[in] env execution environment
5514 * \param[in] lo object
5515 * \param[out] lds default striping
5517 * \retval 0 on success
5518 * \retval negative if failed
5520 static int lod_get_default_lmv_striping(const struct lu_env *env,
5521 struct lod_object *lo,
5522 struct lod_default_striping *lds)
5524 struct lmv_user_md *lmu;
5527 lds->lds_dir_def_striping_set = 0;
5529 rc = lod_get_default_lmv_ea(env, lo);
5533 if (rc >= (int)sizeof(*lmu)) {
5534 struct lod_thread_info *info = lod_env_info(env);
5536 lmu = info->lti_ea_store;
5537 lod_lum2lds(lds, lmu);
5544 * Get default striping in the object.
5546 * Get object default striping and default directory striping.
5548 * \param[in] env execution environment
5549 * \param[in] lo object
5550 * \param[out] lds default striping
5552 * \retval 0 on success
5553 * \retval negative if failed
5555 static int lod_get_default_striping(const struct lu_env *env,
5556 struct lod_object *lo,
5557 struct dt_allocation_hint *ah,
5558 struct lod_default_striping *lds)
5562 rc = lod_get_default_lov_striping(env, lo, lds, NULL);
5563 if (lds->lds_def_striping_set) {
5564 struct lod_thread_info *info = lod_env_info(env);
5565 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5567 rc = lod_verify_striping(env, d, lo, &info->lti_buf, false);
5569 lds->lds_def_striping_set = 0;
5572 if (ah->dah_eadata_is_dmv) {
5573 lod_lum2lds(lds, ah->dah_eadata);
5574 } else if (ah->dah_dmv_imp_inherit) {
5575 lds->lds_dir_def_striping_set = 0;
5577 rc1 = lod_get_default_lmv_striping(env, lo, lds);
5578 if (rc == 0 && rc1 < 0)
5586 * Apply default striping on object.
5588 * If object striping pattern is not set, set to the one in default striping.
5589 * The default striping is from parent or fs.
5591 * \param[in] lo new object
5592 * \param[in] lds default striping
5593 * \param[in] mode new object's mode
5595 static void lod_striping_from_default(struct lod_object *lo,
5596 const struct lod_default_striping *lds,
5599 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5602 if (lds->lds_def_striping_set && S_ISREG(mode)) {
5603 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
5605 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
5606 lds->lds_def_comp_cnt);
5610 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
5611 if (lds->lds_def_mirror_cnt > 1)
5612 lo->ldo_flr_state = LCM_FL_RDONLY;
5614 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5615 struct lod_layout_component *obj_comp =
5616 &lo->ldo_comp_entries[i];
5617 struct lod_layout_component *def_comp =
5618 &lds->lds_def_comp_entries[i];
5621 "inherit "DFID" file layout from default: flags=%#x size=%u nr=%u offset=%u pattern=%#x pool=%s\n",
5622 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
5623 def_comp->llc_flags,
5624 def_comp->llc_stripe_size,
5625 def_comp->llc_stripe_count,
5626 def_comp->llc_stripe_offset,
5627 def_comp->llc_pattern,
5628 def_comp->llc_pool ?: "");
5630 *obj_comp = *def_comp;
5631 if (def_comp->llc_pool != NULL) {
5632 /* pointer was copied from def_comp */
5633 obj_comp->llc_pool = NULL;
5634 lod_obj_set_pool(lo, i, def_comp->llc_pool);
5638 if (def_comp->llc_ostlist.op_array &&
5639 def_comp->llc_ostlist.op_count) {
5640 OBD_ALLOC(obj_comp->llc_ostlist.op_array,
5641 obj_comp->llc_ostlist.op_size);
5642 if (!obj_comp->llc_ostlist.op_array)
5644 memcpy(obj_comp->llc_ostlist.op_array,
5645 def_comp->llc_ostlist.op_array,
5646 obj_comp->llc_ostlist.op_size);
5647 } else if (def_comp->llc_ostlist.op_array) {
5648 obj_comp->llc_ostlist.op_array = NULL;
5652 * Don't initialize these fields for plain layout
5653 * (v1/v3) here, they are inherited in the order of
5654 * 'parent' -> 'fs default (root)' -> 'global default
5655 * values for stripe_count & stripe_size'.
5657 * see lod_ah_init().
5659 if (!lo->ldo_is_composite)
5662 lod_adjust_stripe_info(obj_comp, desc, 0);
5664 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
5665 if (lo->ldo_dir_stripe_count == 0)
5666 lo->ldo_dir_stripe_count =
5667 lds->lds_dir_def_stripe_count;
5668 if (lo->ldo_dir_stripe_offset == -1)
5669 lo->ldo_dir_stripe_offset =
5670 lds->lds_dir_def_stripe_offset;
5671 if (lo->ldo_dir_hash_type == LMV_HASH_TYPE_UNKNOWN)
5672 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
5675 "inherit "DFID" dir layout from default: count=%hu offset=%u hash_type=%x\n",
5676 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
5677 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
5678 lo->ldo_dir_hash_type);
5682 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root,
5683 const char *append_pool)
5685 struct lod_layout_component *lod_comp;
5687 if (lo->ldo_comp_cnt == 0)
5690 if (lo->ldo_is_composite)
5693 lod_comp = &lo->ldo_comp_entries[0];
5695 if (lod_comp->llc_stripe_count <= 0 ||
5696 lod_comp->llc_stripe_size <= 0)
5699 if (from_root && (lod_comp->llc_pool == NULL ||
5700 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
5703 if (append_pool && append_pool[0])
5710 * Implementation of dt_object_operations::do_ah_init.
5712 * This method is used to make a decision on the striping configuration for the
5713 * object being created. It can be taken from the \a parent object if it exists,
5714 * or filesystem's default. The resulting configuration (number of stripes,
5715 * stripe size/offset, pool name, hash_type, etc.) is stored in the object
5716 * itself and will be used by the methods like ->doo_declare_create().
5718 * \see dt_object_operations::do_ah_init() in the API description for details.
5720 static void lod_ah_init(const struct lu_env *env,
5721 struct dt_allocation_hint *ah,
5722 struct dt_object *parent,
5723 struct dt_object *child,
5726 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
5727 struct lod_thread_info *info = lod_env_info(env);
5728 struct lod_default_striping *lds = lod_lds_buf_get(env);
5729 struct dt_object *nextp = NULL;
5730 struct dt_object *nextc;
5731 struct lod_object *lp = NULL;
5732 struct lod_object *lc;
5733 struct lov_desc *desc;
5734 struct lod_layout_component *lod_comp;
5740 if (ah->dah_append_stripe_count == -1)
5741 ah->dah_append_stripe_count =
5742 d->lod_ost_descs.ltd_lov_desc.ld_tgt_count;
5744 if (likely(parent)) {
5745 nextp = dt_object_child(parent);
5746 lp = lod_dt_obj(parent);
5749 nextc = dt_object_child(child);
5750 lc = lod_dt_obj(child);
5752 LASSERT(!lod_obj_is_striped(child));
5753 /* default layout template may have been set on the regular file
5754 * when this is called from mdd_create_data() */
5755 if (S_ISREG(child_mode))
5756 lod_free_comp_entries(lc);
5758 if (!dt_object_exists(nextc))
5759 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
5761 if (S_ISDIR(child_mode)) {
5762 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
5764 /* other default values are 0 */
5765 lc->ldo_dir_stripe_offset = LMV_OFFSET_DEFAULT;
5767 /* no default striping configuration is needed for
5770 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5771 le32_to_cpu(lum1->lum_magic) == LMV_MAGIC_FOREIGN) {
5772 lc->ldo_is_foreign = true;
5773 /* keep stripe_count 0 and stripe_offset -1 */
5774 CDEBUG(D_INFO, "no default striping for foreign dir\n");
5778 if (likely(lp != NULL))
5779 lod_get_default_striping(env, lp, ah, lds);
5781 /* It should always honour the specified stripes */
5782 if (ah->dah_eadata && ah->dah_eadata_len &&
5783 !ah->dah_eadata_is_dmv &&
5784 (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
5785 le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC ||
5786 le32_to_cpu(lum1->lum_magic) == LMV_MAGIC_V1)) {
5787 lc->ldo_dir_stripe_count =
5788 le32_to_cpu(lum1->lum_stripe_count);
5789 lc->ldo_dir_stripe_offset =
5790 le32_to_cpu(lum1->lum_stripe_offset);
5791 lc->ldo_dir_hash_type =
5792 le32_to_cpu(lum1->lum_hash_type);
5794 "set dirstripe: count %hu, offset %d, hash %x\n",
5795 lc->ldo_dir_stripe_count,
5796 (int)lc->ldo_dir_stripe_offset,
5797 lc->ldo_dir_hash_type);
5799 if (d->lod_mdt_descs.ltd_lmv_desc.ld_active_tgt_count &&
5800 lc->ldo_dir_stripe_count < 2 &&
5801 lum1->lum_max_inherit != LMV_INHERIT_NONE) {
5802 /* when filesystem-wide default LMV is set, dirs
5803 * will be created on MDT by space usage, but if
5804 * dir is created with "lfs mkdir -c 1 ...", its
5805 * subdirs should be kept on the same MDT. To
5806 * guarantee this, set default LMV for such dir.
5808 lds->lds_dir_def_stripe_count =
5809 le32_to_cpu(lum1->lum_stripe_count);
5810 /* if "-1" stripe offset is set, save current
5811 * MDT index in default LMV.
5813 if (le32_to_cpu(lum1->lum_stripe_offset) ==
5815 lds->lds_dir_def_stripe_offset =
5816 lod2lu_dev(d)->ld_site->ld_seq_site->ss_node_id;
5818 lds->lds_dir_def_stripe_offset =
5819 le32_to_cpu(lum1->lum_stripe_offset);
5820 lds->lds_dir_def_hash_type =
5821 le32_to_cpu(lum1->lum_hash_type);
5822 lds->lds_dir_def_max_inherit =
5823 lum1->lum_max_inherit;
5824 /* it will be decreased by 1 later in setting */
5825 if (lum1->lum_max_inherit >= LMV_INHERIT_END &&
5826 lum1->lum_max_inherit < LMV_INHERIT_MAX)
5827 lds->lds_dir_def_max_inherit++;
5828 lds->lds_dir_def_max_inherit_rr =
5829 lum1->lum_max_inherit_rr;
5830 lds->lds_dir_def_striping_set = 1;
5831 /* don't inherit LOV from ROOT */
5832 if (lds->lds_def_striping_set &&
5833 fid_is_root(lod_object_fid(lp)))
5834 lds->lds_def_striping_set = 0;
5835 lc->ldo_def_striping = lds;
5836 } else if (lds->lds_def_striping_set &&
5837 !fid_is_root(lod_object_fid(lp))) {
5838 /* don't inherit default LMV for "lfs mkdir" */
5839 lds->lds_dir_def_striping_set = 0;
5840 lc->ldo_def_striping = lds;
5843 /* inherit default striping except ROOT */
5844 if ((lds->lds_def_striping_set ||
5845 lds->lds_dir_def_striping_set) &&
5846 !fid_is_root(lod_object_fid(lp)))
5847 lc->ldo_def_striping = lds;
5849 /* transfer defaults LMV to new directory */
5850 lod_striping_from_default(lc, lds, child_mode);
5852 /* set count 0 to create normal directory */
5853 if (lc->ldo_dir_stripe_count == 1)
5854 lc->ldo_dir_stripe_count = 0;
5856 /* do not save default LMV on server */
5857 if (ah->dah_dmv_imp_inherit) {
5858 lds->lds_dir_def_striping_set = 0;
5859 if (!lds->lds_def_striping_set)
5860 lc->ldo_def_striping = NULL;
5864 /* shrink the stripe count to max_mdt_stripecount if it is -1
5865 * and max_mdt_stripecount is not 0
5867 if (lc->ldo_dir_stripe_count == (__u16)(-1) &&
5868 d->lod_max_mdt_stripecount)
5869 lc->ldo_dir_stripe_count = d->lod_max_mdt_stripecount;
5871 /* shrink the stripe_count to the avaible MDT count */
5872 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
5873 !CFS_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)) {
5874 lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
5875 if (lc->ldo_dir_stripe_count == 1)
5876 lc->ldo_dir_stripe_count = 0;
5879 if (!lmv_is_known_hash_type(lc->ldo_dir_hash_type))
5880 lc->ldo_dir_hash_type =
5881 (lc->ldo_dir_hash_type & LMV_HASH_FLAG_KNOWN) |
5882 d->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
5884 /* make sure all fscrypt metadata stays on same mdt */
5885 if (child->do_lu.lo_header->loh_attr & LOHA_FSCRYPT_MD) {
5886 lc->ldo_dir_stripe_count = 0;
5887 lds->lds_dir_def_stripe_offset =
5888 lod2lu_dev(d)->ld_site->ld_seq_site->ss_node_id;
5889 lds->lds_dir_def_striping_set = 1;
5890 lc->ldo_def_striping = lds;
5893 CDEBUG(D_INFO, "final dir stripe_count=%hu offset=%d hash=%u\n",
5894 lc->ldo_dir_stripe_count,
5895 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
5900 /* child object regular file*/
5902 if (!lod_object_will_be_striped(S_ISREG(child_mode),
5903 lu_object_fid(&child->do_lu)))
5906 /* If object is going to be striped over OSTs, transfer default
5907 * striping information to the child, so that we can use it
5908 * during declaration and creation.
5910 * Try from the parent first.
5912 if (likely(lp != NULL)) {
5913 rc = lod_get_default_lov_striping(env, lp, lds, ah);
5914 if (rc == 0 && lds->lds_def_striping_set) {
5915 rc = lod_verify_striping(env, d, lp, &info->lti_buf,
5918 lod_striping_from_default(lc, lds, child_mode);
5922 /* Initialize lod_device::lod_md_root object reference */
5923 if (d->lod_md_root == NULL) {
5924 struct dt_object *root;
5925 struct lod_object *lroot;
5927 lu_root_fid(&info->lti_fid);
5928 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
5929 if (!IS_ERR(root)) {
5930 lroot = lod_dt_obj(root);
5932 spin_lock(&d->lod_lock);
5933 if (d->lod_md_root != NULL)
5934 dt_object_put(env, &d->lod_md_root->ldo_obj);
5935 d->lod_md_root = lroot;
5936 spin_unlock(&d->lod_lock);
5940 /* try inherit layout from the root object (fs default) when:
5941 * - parent does not have default layout; or
5942 * - parent has plain(v1/v3) default layout, and some attributes
5943 * are not specified in the default layout;
5945 if (d->lod_md_root != NULL &&
5946 lod_need_inherit_more(lc, true, ah->dah_append_pool)) {
5947 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds,
5949 if (rc || !lds->lds_def_striping_set)
5952 rc = lod_verify_striping(env, d, d->lod_md_root, &info->lti_buf,
5957 if (lc->ldo_comp_cnt == 0) {
5958 lod_striping_from_default(lc, lds, child_mode);
5959 } else if (!lds->lds_def_striping_is_composite) {
5960 struct lod_layout_component *def_comp;
5962 LASSERT(!lc->ldo_is_composite);
5963 lod_comp = &lc->ldo_comp_entries[0];
5964 def_comp = &lds->lds_def_comp_entries[0];
5966 if (lod_comp->llc_stripe_count <= 0)
5967 lod_comp->llc_stripe_count =
5968 def_comp->llc_stripe_count;
5969 if (lod_comp->llc_stripe_size <= 0)
5970 lod_comp->llc_stripe_size =
5971 def_comp->llc_stripe_size;
5972 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT &&
5973 (!lod_comp->llc_pool || !lod_comp->llc_pool[0]))
5974 lod_comp->llc_stripe_offset =
5975 def_comp->llc_stripe_offset;
5976 if (lod_comp->llc_pool == NULL)
5977 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
5982 * fs default striping may not be explicitly set, or historically set
5983 * in config log, use them.
5985 if (lod_need_inherit_more(lc, false, ah->dah_append_pool)) {
5986 if (lc->ldo_comp_cnt == 0) {
5987 rc = lod_alloc_comp_entries(lc, 0, 1);
5989 /* fail to allocate memory, will create a
5990 * non-striped file. */
5992 lc->ldo_is_composite = 0;
5993 lod_comp = &lc->ldo_comp_entries[0];
5994 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
5996 LASSERT(!lc->ldo_is_composite);
5997 lod_comp = &lc->ldo_comp_entries[0];
5998 desc = &d->lod_ost_descs.ltd_lov_desc;
5999 lod_adjust_stripe_info(lod_comp, desc,
6000 ah->dah_append_stripe_count);
6001 if (ah->dah_append_pool && ah->dah_append_pool[0])
6002 lod_obj_set_pool(lc, 0, ah->dah_append_pool);
6009 * Size initialization on late striping.
6011 * Propagate the size of a truncated object to a deferred striping.
6012 * This function handles a special case when truncate was done on a
6013 * non-striped object and now while the striping is being created
6014 * we can't lose that size, so we have to propagate it to the stripes
6017 * \param[in] env execution environment
6018 * \param[in] dt object
6019 * \param[in] th transaction handle
6021 * \retval 0 on success
6022 * \retval negative if failed
6024 static int lod_declare_init_size(const struct lu_env *env,
6025 struct dt_object *dt, struct thandle *th)
6027 struct dt_object *next = dt_object_child(dt);
6028 struct lod_object *lo = lod_dt_obj(dt);
6029 struct dt_object **objects = NULL;
6030 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
6031 uint64_t size, offs;
6032 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
6033 struct lu_extent size_ext;
6036 if (!lod_obj_is_striped(dt))
6039 rc = dt_attr_get(env, next, attr);
6040 LASSERT(attr->la_valid & LA_SIZE);
6044 size = attr->la_size;
6048 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
6049 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6050 struct lod_layout_component *lod_comp;
6051 struct lu_extent *extent;
6053 lod_comp = &lo->ldo_comp_entries[i];
6055 if (lod_comp->llc_stripe == NULL)
6058 extent = &lod_comp->llc_extent;
6059 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
6060 if (!lo->ldo_is_composite ||
6061 lu_extent_is_overlapped(extent, &size_ext)) {
6062 objects = lod_comp->llc_stripe;
6063 stripe_count = lod_comp->llc_stripe_count;
6064 stripe_size = lod_comp->llc_stripe_size;
6067 if (stripe_count == 0)
6070 LASSERT(objects != NULL && stripe_size != 0);
6071 do_div(size, stripe_size);
6072 stripe = do_div(size, stripe_count);
6073 LASSERT(objects[stripe] != NULL);
6075 size = size * stripe_size;
6076 offs = attr->la_size;
6077 size += do_div(offs, stripe_size);
6079 attr->la_valid = LA_SIZE;
6080 attr->la_size = size;
6082 rc = lod_sub_declare_attr_set(env, objects[stripe],
6091 * Declare creation of striped object.
6093 * The function declares creation stripes for a regular object. The function
6094 * also declares whether the stripes will be created with non-zero size if
6095 * previously size was set non-zero on the master object. If object \a dt is
6096 * not local, then only fully defined striping can be applied in \a lovea.
6097 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
6100 * \param[in] env execution environment
6101 * \param[in] dt object
6102 * \param[in] attr attributes the stripes will be created with
6103 * \param[in] lovea a buffer containing striping description
6104 * \param[in] th transaction handle
6106 * \retval 0 on success
6107 * \retval negative if failed
6109 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
6110 struct lu_attr *attr,
6111 const struct lu_buf *lovea, struct thandle *th)
6113 struct lod_thread_info *info = lod_env_info(env);
6114 struct dt_object *next = dt_object_child(dt);
6115 struct lod_object *lo = lod_dt_obj(dt);
6119 if (CFS_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
6120 GOTO(out, rc = -ENOMEM);
6122 if (!dt_object_remote(next)) {
6123 /* choose OST and generate appropriate objects */
6124 rc = lod_prepare_create(env, lo, attr, lovea, th);
6129 * declare storage for striping data
6131 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6133 /* LOD can not choose OST objects for remote objects, i.e.
6134 * stripes must be ready before that. Right now, it can only
6135 * happen during migrate, i.e. migrate process needs to create
6136 * remote regular file (mdd_migrate_create), then the migrate
6137 * process will provide stripeEA. */
6138 LASSERT(lovea != NULL);
6139 info->lti_buf = *lovea;
6142 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
6143 XATTR_NAME_LOV, 0, th);
6148 * if striping is created with local object's size > 0,
6149 * we have to propagate this size to specific object
6150 * the case is possible only when local object was created previously
6152 if (dt_object_exists(next))
6153 rc = lod_declare_init_size(env, dt, th);
6156 /* failed to create striping or to set initial size, let's reset
6157 * config so that others don't get confused */
6159 lod_striping_free(env, lo);
6165 * Whether subdirectories under \a dt should be created on MDTs by space QoS
6167 * If LMV_HASH_FLAG_SPACE is set on directory default layout, its subdirectories
6168 * should be created on MDT by space QoS.
6170 * \param[in] env execution environment
6171 * \param[in] dev lu device
6172 * \param[in] dt object
6174 * \retval 1 if directory should create subdir by space usage
6176 * \retval -ev if failed
6178 static inline int dt_object_qos_mkdir(const struct lu_env *env,
6179 struct lu_device *dev,
6180 struct dt_object *dt)
6182 struct lod_thread_info *info = lod_env_info(env);
6183 struct lu_object *obj;
6184 struct lod_object *lo;
6185 struct lmv_user_md *lmu;
6188 obj = lu_object_find_slice(env, dev, lu_object_fid(&dt->do_lu), NULL);
6190 return PTR_ERR(obj);
6192 lo = lu2lod_obj(obj);
6194 rc = lod_get_default_lmv_ea(env, lo);
6195 dt_object_put(env, dt);
6199 if (rc < (int)sizeof(*lmu))
6202 lmu = info->lti_ea_store;
6203 return le32_to_cpu(lmu->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
6207 * Implementation of dt_object_operations::do_declare_create.
6209 * The method declares creation of a new object. If the object will be striped,
6210 * then helper functions are called to find FIDs for the stripes, declare
6211 * creation of the stripes and declare initialization of the striping
6212 * information to be stored in the master object.
6214 * \see dt_object_operations::do_declare_create() in the API description
6217 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
6218 struct lu_attr *attr,
6219 struct dt_allocation_hint *hint,
6220 struct dt_object_format *dof, struct thandle *th)
6222 struct dt_object *next = dt_object_child(dt);
6223 struct lod_object *lo = lod_dt_obj(dt);
6232 * first of all, we declare creation of local object
6234 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
6239 * it's lod_ah_init() that has decided the object will be striped
6241 if (dof->dof_type == DFT_REGULAR) {
6242 /* callers don't want stripes */
6243 /* XXX: all tricky interactions with ->ah_make_hint() decided
6244 * to use striping, then ->declare_create() behaving differently
6245 * should be cleaned */
6246 if (dof->u.dof_reg.striped != 0)
6247 rc = lod_declare_striped_create(env, dt, attr,
6249 } else if (dof->dof_type == DFT_DIR) {
6250 struct seq_server_site *ss;
6251 struct lu_buf buf = { NULL };
6253 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
6255 /* If the parent has default stripeEA, and client
6256 * did not find it before sending create request,
6257 * then MDT will return -EREMOTE, and client will
6258 * retrieve the default stripeEA and re-create the
6261 * Note: if dah_eadata != NULL, it means creating the
6262 * striped directory with specified stripeEA, then it
6263 * should ignore the default stripeEA */
6264 if (hint != NULL && hint->dah_eadata == NULL) {
6265 if (CFS_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
6266 GOTO(out, rc = -EREMOTE);
6268 if (lo->ldo_dir_stripe_offset != LMV_OFFSET_DEFAULT &&
6269 lo->ldo_dir_stripe_offset != ss->ss_node_id) {
6270 struct lod_device *lod;
6271 struct lu_tgt_desc *mdt = NULL;
6272 bool found_mdt = false;
6274 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6275 lod_foreach_mdt(lod, mdt) {
6276 if (mdt->ltd_index ==
6277 lo->ldo_dir_stripe_offset) {
6283 /* If the MDT indicated by stripe_offset can be
6284 * found, then tell client to resend the create
6285 * request to the correct MDT, otherwise return
6286 * error to client */
6288 GOTO(out, rc = -EREMOTE);
6290 GOTO(out, rc = -EINVAL);
6292 } else if (hint && hint->dah_eadata) {
6293 buf.lb_buf = (void *)hint->dah_eadata;
6294 buf.lb_len = hint->dah_eadata_len;
6297 rc = lod_declare_dir_striping_create(env, dt, attr, &buf, dof,
6301 /* failed to create striping or to set initial size, let's reset
6302 * config so that others don't get confused */
6304 lod_striping_free(env, lo);
6309 * Generate component ID for new created component.
6311 * \param[in] lo LOD object
6312 * \param[in] comp_idx index of ldo_comp_entries
6314 * \retval component ID on success
6315 * \retval LCME_ID_INVAL on failure
6317 static __u32 lod_gen_component_id(struct lod_object *lo,
6318 int mirror_id, int comp_idx)
6320 struct lod_layout_component *lod_comp;
6321 __u32 id, start, end;
6324 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
6326 lod_obj_inc_layout_gen(lo);
6327 id = lo->ldo_layout_gen;
6328 if (likely(id <= SEQ_ID_MAX))
6329 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
6331 /* Layout generation wraps, need to check collisions. */
6332 start = id & SEQ_ID_MASK;
6335 for (id = start; id <= end; id++) {
6336 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6337 lod_comp = &lo->ldo_comp_entries[i];
6338 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
6341 /* Found the ununsed ID */
6342 if (i == lo->ldo_comp_cnt)
6343 RETURN(pflr_id(mirror_id, id));
6346 if (end == SEQ_ID_MAX) {
6347 end = min_t(__u32, start, SEQ_ID_MAX) - 1;
6352 RETURN(LCME_ID_INVAL);
6356 * Creation of a striped regular object.
6358 * The function is called to create the stripe objects for a regular
6359 * striped file. This can happen at the initial object creation or
6360 * when the caller asks LOD to do so using ->do_xattr_set() method
6361 * (so called late striping). Notice all the information are already
6362 * prepared in the form of the list of objects (ldo_stripe field).
6363 * This is done during declare phase.
6365 * \param[in] env execution environment
6366 * \param[in] dt object
6367 * \param[in] attr attributes the stripes will be created with
6368 * \param[in] dof format of stripes (see OSD API description)
6369 * \param[in] th transaction handle
6371 * \retval 0 on success
6372 * \retval negative if failed
6374 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
6375 struct lu_attr *attr, struct dt_object_format *dof,
6378 struct lod_layout_component *lod_comp;
6379 struct lod_object *lo = lod_dt_obj(dt);
6384 mutex_lock(&lo->ldo_layout_mutex);
6386 LASSERT((lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL) ||
6387 lo->ldo_is_foreign);
6389 mirror_id = 0; /* non-flr file's mirror_id is 0 */
6390 if (lo->ldo_mirror_count > 1) {
6391 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6392 lod_comp = &lo->ldo_comp_entries[i];
6393 if (lod_comp->llc_id != LCME_ID_INVAL &&
6394 mirror_id_of(lod_comp->llc_id) > mirror_id)
6395 mirror_id = mirror_id_of(lod_comp->llc_id);
6399 /* create all underlying objects */
6400 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6401 lod_comp = &lo->ldo_comp_entries[i];
6403 if (lod_comp->llc_id == LCME_ID_INVAL) {
6404 /* only the component of FLR layout with more than 1
6405 * mirror has mirror ID in its component ID.
6407 if (lod_comp->llc_extent.e_start == 0 &&
6408 lo->ldo_mirror_count > 1)
6411 lod_comp->llc_id = lod_gen_component_id(lo,
6413 if (lod_comp->llc_id == LCME_ID_INVAL)
6414 GOTO(out, rc = -ERANGE);
6417 if (lod_comp_inited(lod_comp))
6420 if (lod_comp->llc_magic == LOV_MAGIC_FOREIGN) {
6421 lod_comp_set_init(lod_comp);
6425 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
6426 lod_comp_set_init(lod_comp);
6428 if (lov_pattern(lod_comp->llc_pattern) & LOV_PATTERN_MDT)
6429 lod_comp_set_init(lod_comp);
6431 if (lod_comp->llc_stripe == NULL)
6434 LASSERT(lod_comp->llc_stripe_count);
6435 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6436 struct dt_object *object = lod_comp->llc_stripe[j];
6437 LASSERT(object != NULL);
6438 rc = lod_sub_create(env, object, attr, NULL, dof, th);
6442 lod_comp_set_init(lod_comp);
6445 rc = lod_fill_mirrors(lo);
6449 lo->ldo_comp_cached = 1;
6451 rc = lod_generate_and_set_lovea(env, lo, th);
6455 mutex_unlock(&lo->ldo_layout_mutex);
6460 lod_striping_free_nolock(env, lo);
6461 mutex_unlock(&lo->ldo_layout_mutex);
6466 static inline bool lod_obj_is_dom(struct dt_object *dt)
6468 struct lod_object *lo = lod_dt_obj(dt);
6470 if (!dt_object_exists(dt_object_child(dt)))
6473 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
6476 if (!lo->ldo_comp_cnt)
6479 return (lov_pattern(lo->ldo_comp_entries[0].llc_pattern) &
6484 * Implementation of dt_object_operations::do_create.
6486 * If any of preceeding methods (like ->do_declare_create(),
6487 * ->do_ah_init(), etc) chose to create a striped object,
6488 * then this method will create the master and the stripes.
6490 * \see dt_object_operations::do_create() in the API description for details.
6492 static int lod_create(const struct lu_env *env, struct dt_object *dt,
6493 struct lu_attr *attr, struct dt_allocation_hint *hint,
6494 struct dt_object_format *dof, struct thandle *th)
6499 /* create local object */
6500 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
6504 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
6505 (lod_obj_is_striped(dt) || lod_obj_is_dom(dt)) &&
6506 dof->u.dof_reg.striped != 0) {
6507 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
6508 rc = lod_striped_create(env, dt, attr, dof, th);
6515 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
6516 struct dt_object *dt, struct thandle *th,
6517 int comp_idx, int stripe_idx,
6518 struct lod_obj_stripe_cb_data *data)
6520 if (data->locd_declare)
6521 return lod_sub_declare_destroy(env, dt, th);
6523 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
6524 stripe_idx == cfs_fail_val)
6525 return lod_sub_destroy(env, dt, th);
6531 * Implementation of dt_object_operations::do_declare_destroy.
6533 * If the object is a striped directory, then the function declares reference
6534 * removal from the master object (this is an index) to the stripes and declares
6535 * destroy of all the stripes. In all the cases, it declares an intention to
6536 * destroy the object itself.
6538 * \see dt_object_operations::do_declare_destroy() in the API description
6541 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
6544 struct dt_object *next = dt_object_child(dt);
6545 struct lod_object *lo = lod_dt_obj(dt);
6546 struct lod_thread_info *info = lod_env_info(env);
6547 struct dt_object *stripe;
6548 char *stripe_name = info->lti_key;
6554 * load striping information, notice we don't do this when object
6555 * is being initialized as we don't need this information till
6556 * few specific cases like destroy, chown
6558 rc = lod_striping_load(env, lo);
6562 /* declare destroy for all underlying objects */
6563 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6564 rc = next->do_ops->do_index_try(env, next,
6565 &dt_directory_features);
6569 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6570 stripe = lo->ldo_stripe[i];
6574 rc = lod_sub_declare_ref_del(env, next, th);
6578 snprintf(stripe_name, sizeof(info->lti_key),
6580 PFID(lu_object_fid(&stripe->do_lu)), i);
6581 rc = lod_sub_declare_delete(env, next,
6582 (const struct dt_key *)stripe_name, th);
6589 * we declare destroy for the local object
6591 rc = lod_sub_declare_destroy(env, next, th);
6595 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
6596 CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
6599 if (!lod_obj_is_striped(dt))
6602 /* declare destroy all striped objects */
6603 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6604 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6605 stripe = lo->ldo_stripe[i];
6609 if (!dt_object_exists(stripe))
6612 rc = lod_sub_declare_ref_del(env, stripe, th);
6616 rc = lod_sub_declare_destroy(env, stripe, th);
6621 struct lod_obj_stripe_cb_data data = { { 0 } };
6623 data.locd_declare = true;
6624 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
6625 rc = lod_obj_for_each_stripe(env, lo, th, &data);
6632 * Implementation of dt_object_operations::do_destroy.
6634 * If the object is a striped directory, then the function removes references
6635 * from the master object (this is an index) to the stripes and destroys all
6636 * the stripes. In all the cases, the function destroys the object itself.
6638 * \see dt_object_operations::do_destroy() in the API description for details.
6640 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
6643 struct dt_object *next = dt_object_child(dt);
6644 struct lod_object *lo = lod_dt_obj(dt);
6645 struct lod_thread_info *info = lod_env_info(env);
6646 char *stripe_name = info->lti_key;
6647 struct dt_object *stripe;
6653 /* destroy sub-stripe of master object */
6654 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6655 rc = next->do_ops->do_index_try(env, next,
6656 &dt_directory_features);
6660 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6661 stripe = lo->ldo_stripe[i];
6665 rc = lod_sub_ref_del(env, next, th);
6669 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
6670 PFID(lu_object_fid(&stripe->do_lu)), i);
6672 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
6673 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
6674 PFID(lu_object_fid(&stripe->do_lu)));
6676 rc = lod_sub_delete(env, next,
6677 (const struct dt_key *)stripe_name, th);
6683 rc = lod_sub_destroy(env, next, th);
6687 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
6688 CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
6691 if (!lod_obj_is_striped(dt))
6694 /* destroy all striped objects */
6695 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6696 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6697 stripe = lo->ldo_stripe[i];
6701 if (!dt_object_exists(stripe))
6704 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
6705 i == cfs_fail_val) {
6706 dt_write_lock(env, stripe, DT_TGT_CHILD);
6707 rc = lod_sub_ref_del(env, stripe, th);
6708 dt_write_unlock(env, stripe);
6712 rc = lod_sub_destroy(env, stripe, th);
6718 struct lod_obj_stripe_cb_data data = { { 0 } };
6720 data.locd_declare = false;
6721 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
6722 rc = lod_obj_for_each_stripe(env, lo, th, &data);
6729 * Implementation of dt_object_operations::do_declare_ref_add.
6731 * \see dt_object_operations::do_declare_ref_add() in the API description
6734 static int lod_declare_ref_add(const struct lu_env *env,
6735 struct dt_object *dt, struct thandle *th)
6737 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
6741 * Implementation of dt_object_operations::do_ref_add.
6743 * \see dt_object_operations::do_ref_add() in the API description for details.
6745 static int lod_ref_add(const struct lu_env *env,
6746 struct dt_object *dt, struct thandle *th)
6748 return lod_sub_ref_add(env, dt_object_child(dt), th);
6752 * Implementation of dt_object_operations::do_declare_ref_del.
6754 * \see dt_object_operations::do_declare_ref_del() in the API description
6757 static int lod_declare_ref_del(const struct lu_env *env,
6758 struct dt_object *dt, struct thandle *th)
6760 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
6764 * Implementation of dt_object_operations::do_ref_del
6766 * \see dt_object_operations::do_ref_del() in the API description for details.
6768 static int lod_ref_del(const struct lu_env *env,
6769 struct dt_object *dt, struct thandle *th)
6771 return lod_sub_ref_del(env, dt_object_child(dt), th);
6775 * Implementation of dt_object_operations::do_object_sync.
6777 * \see dt_object_operations::do_object_sync() in the API description
6780 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
6781 __u64 start, __u64 end)
6783 return dt_object_sync(env, dt_object_child(dt), start, end);
6787 * Implementation of dt_object_operations::do_object_unlock.
6789 * Used to release LDLM lock(s).
6791 * \see dt_object_operations::do_object_unlock() in the API description
6794 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
6795 struct ldlm_enqueue_info *einfo,
6796 union ldlm_policy_data *policy)
6798 struct lod_object *lo = lod_dt_obj(dt);
6799 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
6800 int slave_locks_size;
6804 if (slave_locks == NULL)
6807 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
6808 /* Note: for remote lock for single stripe dir, MDT will cancel
6809 * the lock by lockh directly */
6810 LASSERT(!dt_object_remote(dt_object_child(dt)));
6812 /* locks were unlocked in MDT layer */
6813 for (i = 0; i < slave_locks->ha_count; i++)
6814 LASSERT(!lustre_handle_is_used(&slave_locks->ha_handles[i]));
6817 * NB, ha_count may not equal to ldo_dir_stripe_count, because dir
6818 * layout may change, e.g., shrink dir layout after migration.
6820 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6821 if (lo->ldo_stripe[i])
6822 dt_invalidate(env, lo->ldo_stripe[i]);
6825 slave_locks_size = offsetof(typeof(*slave_locks),
6826 ha_handles[slave_locks->ha_count]);
6827 OBD_FREE(slave_locks, slave_locks_size);
6828 einfo->ei_cbdata = NULL;
6834 * Implementation of dt_object_operations::do_object_lock.
6836 * Used to get LDLM lock on the non-striped and striped objects.
6838 * \see dt_object_operations::do_object_lock() in the API description
6841 static int lod_object_lock(const struct lu_env *env,
6842 struct dt_object *dt,
6843 struct lustre_handle *lh,
6844 struct ldlm_enqueue_info *einfo,
6845 union ldlm_policy_data *policy)
6847 struct lod_object *lo = lod_dt_obj(dt);
6848 int slave_locks_size;
6849 struct lustre_handle_array *slave_locks = NULL;
6854 /* remote object lock */
6855 if (!einfo->ei_enq_slave) {
6856 LASSERT(dt_object_remote(dt));
6857 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
6861 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
6864 rc = lod_striping_load(env, lo);
6869 if (lo->ldo_dir_stripe_count <= 1)
6872 slave_locks_size = offsetof(typeof(*slave_locks),
6873 ha_handles[lo->ldo_dir_stripe_count]);
6874 /* Freed in lod_object_unlock */
6875 OBD_ALLOC(slave_locks, slave_locks_size);
6878 slave_locks->ha_count = lo->ldo_dir_stripe_count;
6880 /* striped directory lock */
6881 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6882 struct lustre_handle lockh;
6883 struct ldlm_res_id *res_id;
6884 struct dt_object *stripe;
6886 stripe = lo->ldo_stripe[i];
6890 res_id = &lod_env_info(env)->lti_res_id;
6891 fid_build_reg_res_name(lu_object_fid(&stripe->do_lu), res_id);
6892 einfo->ei_res_id = res_id;
6894 if (dt_object_remote(stripe)) {
6895 set_bit(i, (void *)slave_locks->ha_map);
6896 rc = dt_object_lock(env, stripe, &lockh, einfo, policy);
6898 struct ldlm_namespace *ns = einfo->ei_namespace;
6899 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
6900 ldlm_completion_callback completion = einfo->ei_cb_cp;
6901 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
6903 LASSERT(ns != NULL);
6904 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS,
6905 policy, einfo->ei_mode,
6906 &dlmflags, blocking,
6908 NULL, 0, LVB_T_NONE,
6913 ldlm_lock_decref_and_cancel(
6914 &slave_locks->ha_handles[i],
6916 OBD_FREE(slave_locks, slave_locks_size);
6919 slave_locks->ha_handles[i] = lockh;
6921 einfo->ei_cbdata = slave_locks;
6927 * Implementation of dt_object_operations::do_invalidate.
6929 * \see dt_object_operations::do_invalidate() in the API description for details
6931 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
6933 return dt_invalidate(env, dt_object_child(dt));
6936 static int lod_declare_instantiate_components(const struct lu_env *env,
6937 struct lod_object *lo,
6941 struct lod_thread_info *info = lod_env_info(env);
6946 LASSERT(info->lti_count < lo->ldo_comp_cnt);
6948 for (i = 0; i < info->lti_count; i++) {
6949 rc = lod_qos_prep_create(env, lo, NULL, th,
6950 info->lti_comp_idx[i], reserve);
6956 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6957 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
6958 &info->lti_buf, XATTR_NAME_LOV, 0, th);
6965 * Check OSTs for an existing component for further extension
6967 * Checks if OSTs are still healthy and not out of space. Gets free space
6968 * on OSTs (relative to allocation watermark rmb_low) and compares to
6969 * the proposed new_end for this component.
6971 * Decides whether or not to extend a component on its current OSTs.
6973 * \param[in] env execution environment for this thread
6974 * \param[in] lo object we're checking
6975 * \param[in] index index of this component
6976 * \param[in] extension_size extension size for this component
6977 * \param[in] extent layout extent for requested operation
6978 * \param[in] comp_extent extension component extent
6979 * \param[in] write if this is write operation
6981 * \retval true - OK to extend on current OSTs
6982 * \retval false - do not extend on current OSTs
6984 static bool lod_sel_osts_allowed(const struct lu_env *env,
6985 struct lod_object *lo,
6986 int index, __u64 reserve,
6987 struct lu_extent *extent,
6988 struct lu_extent *comp_extent, int write)
6990 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[index];
6991 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6992 struct lod_thread_info *tinfo = lod_env_info(env);
6993 struct obd_statfs *sfs = &tinfo->lti_osfs;
6994 __u64 available = 0;
7000 LASSERT(lod_comp->llc_stripe_count != 0);
7002 lod_getref(&lod->lod_ost_descs);
7003 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
7004 int index = lod_comp->llc_ost_indices[i];
7005 struct lod_tgt_desc *ost = OST_TGT(lod, index);
7006 struct obd_statfs_info info = { 0 };
7007 int j, repeated = 0;
7011 /* Get the number of times this OST repeats in this component.
7012 * Note: inter-component repeats are not counted as this is
7013 * considered as a rare case: we try to not repeat OST in other
7014 * components if possible. */
7015 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
7016 if (index != lod_comp->llc_ost_indices[j])
7019 /* already handled */
7025 if (j < lod_comp->llc_stripe_count)
7028 if (!test_bit(index, lod->lod_ost_bitmap)) {
7029 CDEBUG(D_LAYOUT, "ost %d no longer present\n", index);
7034 rc = dt_statfs_info(env, ost->ltd_tgt, sfs, &info);
7036 CDEBUG(D_LAYOUT, "statfs failed for ost %d, error %d\n",
7042 if (sfs->os_state & OS_STATFS_ENOSPC ||
7043 sfs->os_state & OS_STATFS_READONLY ||
7044 sfs->os_state & OS_STATFS_DEGRADED) {
7045 CDEBUG(D_LAYOUT, "ost %d is not availble for SEL "
7046 "extension, state %u\n", index, sfs->os_state);
7052 available = sfs->os_bavail * sfs->os_bsize;
7053 /* 'available' is relative to the allocation threshold */
7054 available -= (__u64) info.os_reserved_mb_low << 20;
7056 CDEBUG(D_LAYOUT, "ost %d lowwm: %d highwm: %d, "
7057 "%llu %% blocks available, %llu %% blocks free\n",
7058 index, info.os_reserved_mb_low, info.os_reserved_mb_high,
7059 (100ull * sfs->os_bavail) / sfs->os_blocks,
7060 (100ull * sfs->os_bfree) / sfs->os_blocks);
7062 if (reserve * repeated > available) {
7064 CDEBUG(D_LAYOUT, "low space on ost %d, available %llu "
7065 "< extension size %llu repeated %d\n", index,
7066 available, reserve, repeated);
7070 lod_putref(lod, &lod->lod_ost_descs);
7076 * Adjust extents after component removal
7078 * When we remove an extension component, we move the start of the next
7079 * component to match the start of the extension component, so no space is left
7082 * \param[in] env execution environment for this thread
7083 * \param[in] lo object
7084 * \param[in] max_comp layout component
7085 * \param[in] index index of this component
7087 * \retval 0 on success
7088 * \retval negative errno on error
7090 static void lod_sel_adjust_extents(const struct lu_env *env,
7091 struct lod_object *lo,
7092 int max_comp, int index)
7094 struct lod_layout_component *lod_comp = NULL;
7095 struct lod_layout_component *next = NULL;
7096 struct lod_layout_component *prev = NULL;
7097 __u64 new_start = 0;
7101 /* Extension space component */
7102 lod_comp = &lo->ldo_comp_entries[index];
7103 next = &lo->ldo_comp_entries[index + 1];
7104 prev = &lo->ldo_comp_entries[index - 1];
7106 LASSERT(lod_comp != NULL && prev != NULL && next != NULL);
7107 LASSERT(lod_comp->llc_flags & LCME_FL_EXTENSION);
7109 /* Previous is being removed */
7110 if (prev && prev->llc_id == LCME_ID_INVAL)
7111 new_start = prev->llc_extent.e_start;
7113 new_start = lod_comp->llc_extent.e_start;
7115 for (i = index + 1; i < max_comp; i++) {
7116 lod_comp = &lo->ldo_comp_entries[i];
7118 start = lod_comp->llc_extent.e_start;
7119 lod_comp->llc_extent.e_start = new_start;
7121 /* We only move zero length extendable components */
7122 if (!(start == lod_comp->llc_extent.e_end))
7125 LASSERT(!(lod_comp->llc_flags & LCME_FL_INIT));
7127 lod_comp->llc_extent.e_end = new_start;
7131 /* Calculate the proposed 'new end' for a component we're extending */
7132 static __u64 lod_extension_new_end(__u64 extension_size, __u64 extent_end,
7133 __u32 stripe_size, __u64 component_end,
7134 __u64 extension_end)
7138 LASSERT(extension_size != 0 && stripe_size != 0);
7140 /* Round up to extension size */
7141 if (extent_end == OBD_OBJECT_EOF) {
7142 new_end = OBD_OBJECT_EOF;
7144 /* Add at least extension_size to the previous component_end,
7145 * covering the req layout extent */
7146 new_end = max(extent_end - component_end, extension_size);
7147 new_end = roundup(new_end, extension_size);
7148 new_end += component_end;
7150 /* Component end must be min stripe size aligned */
7151 if (new_end % stripe_size) {
7152 CDEBUG(D_LAYOUT, "new component end is not aligned "
7153 "by the stripe size %u: [%llu, %llu) ext size "
7154 "%llu new end %llu, aligning\n",
7155 stripe_size, component_end, extent_end,
7156 extension_size, new_end);
7157 new_end = roundup(new_end, stripe_size);
7161 if (new_end < extent_end)
7162 new_end = OBD_OBJECT_EOF;
7165 /* Don't extend past the end of the extension component */
7166 if (new_end > extension_end)
7167 new_end = extension_end;
7173 * Calculate the exact reservation (per-OST extension_size) on the OSTs being
7174 * instantiated. It needs to be calculated in advance and taken into account at
7175 * the instantiation time, because otherwise lod_statfs_and_check() may consider
7176 * an OST as OK, but SEL needs its extension_size to fit the free space and the
7177 * OST may turn out to be low-on-space, thus inappropriate OST may be used and
7180 * \param[in] lod_comp lod component we are checking
7182 * \retval size to reserved on each OST of lod_comp's stripe.
7184 static __u64 lod_sel_stripe_reserved(struct lod_layout_component *lod_comp)
7186 /* extension_size is file level, so we must divide by stripe count to
7187 * compare it to available space on a single OST */
7188 return lod_comp->llc_stripe_size * SEL_UNIT_SIZE /
7189 lod_comp->llc_stripe_count;
7192 /* As lod_sel_handler() could be re-entered for the same component several
7193 * times, this is the data for the next call. Fields could be changed to
7194 * component indexes when needed, (e.g. if there is no need to instantiate
7195 * all the previous components up to the current position) to tell the caller
7196 * where to start over from. */
7203 * Process extent updates for a particular layout component
7205 * Handle layout updates for a particular extension space component touched by
7206 * a layout update operation. Core function of self-extending PFL feature.
7208 * In general, this function processes exactly *one* stage of an extension
7209 * operation, modifying the layout accordingly, then returns to the caller.
7210 * The caller is responsible for restarting processing with the new layout,
7211 * which may repeatedly return to this function until the extension updates
7214 * This function does one of a few things to the layout:
7215 * 1. Extends the component before the current extension space component to
7216 * allow it to accomodate the requested operation (if space/policy permit that
7217 * component to continue on its current OSTs)
7219 * 2. If extension of the existing component fails, we do one of two things:
7220 * a. If there is a component after the extension space, we remove the
7221 * extension space component, move the start of the next component down
7222 * accordingly, then notify the caller to restart processing w/the new
7224 * b. If there is no following component, we try repeating the current
7225 * component, creating a new component using the current one as a
7226 * template (keeping its stripe properties but not specific striping),
7227 * and try assigning striping for this component. If there is sufficient
7228 * free space on the OSTs chosen for this component, it is instantiated
7229 * and i/o continues there.
7231 * If there is not sufficient space on the new OSTs, we remove this new
7232 * component & extend the current component.
7234 * Note further that uninited components followed by extension space can be zero
7235 * length meaning that we will try to extend them before initializing them, and
7236 * if that fails, they will be removed without initialization.
7238 * 3. If we extend to/beyond the end of an extension space component, that
7239 * component is exhausted (all of its range has been given to real components),
7240 * so we remove it and restart processing.
7242 * \param[in] env execution environment for this thread
7243 * \param[in,out] lo object to update the layout of
7244 * \param[in] extent layout extent for requested operation, update
7245 * layout to fit this operation
7246 * \param[in] th transaction handle for this operation
7247 * \param[in,out] max_comp the highest comp for the portion of the layout
7248 * we are operating on (For FLR, the chosen
7249 * replica). Updated because we may remove
7251 * \param[in] index index of the extension space component we're
7253 * \param[in] write if this is write op
7254 * \param[in,out] force if the extension is to be forced; set here
7255 to force it on the 2nd call for the same
7258 * \retval 0 on success
7259 * \retval negative errno on error
7261 static int lod_sel_handler(const struct lu_env *env,
7262 struct lod_object *lo,
7263 struct lu_extent *extent,
7264 struct thandle *th, int *max_comp,
7265 int index, int write,
7266 struct sel_data *sd)
7268 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7269 struct lod_thread_info *info = lod_env_info(env);
7270 struct lod_layout_component *lod_comp;
7271 struct lod_layout_component *prev;
7272 struct lod_layout_component *next = NULL;
7273 __u64 extension_size, reserve;
7280 /* First component cannot be extension space */
7282 CERROR("%s: "DFID" first component cannot be extension space\n",
7283 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
7287 lod_comp = &lo->ldo_comp_entries[index];
7288 prev = &lo->ldo_comp_entries[index - 1];
7289 if ((index + 1) < *max_comp)
7290 next = &lo->ldo_comp_entries[index + 1];
7292 /* extension size uses the stripe size field as KiB */
7293 extension_size = lod_comp->llc_stripe_size * SEL_UNIT_SIZE;
7295 CDEBUG(D_LAYOUT, "prev start %llu, extension start %llu, extension end"
7296 " %llu, extension size %llu\n", prev->llc_extent.e_start,
7297 lod_comp->llc_extent.e_start, lod_comp->llc_extent.e_end,
7300 /* Two extension space components cannot be adjacent & extension space
7301 * components cannot be init */
7302 if ((prev->llc_flags & LCME_FL_EXTENSION) ||
7303 !(ergo(next, !(next->llc_flags & LCME_FL_EXTENSION))) ||
7304 lod_comp_inited(lod_comp)) {
7305 CERROR("%s: "DFID" invalid extension space components\n",
7306 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
7310 reserve = lod_sel_stripe_reserved(lod_comp);
7312 if (!prev->llc_stripe) {
7313 CDEBUG(D_LAYOUT, "Previous component not inited\n");
7314 info->lti_count = 1;
7315 info->lti_comp_idx[0] = index - 1;
7316 rc = lod_declare_instantiate_components(env, lo, th, reserve);
7317 /* ENOSPC tells us we can't use this component. If there is
7318 * a next or we are repeating, we either spill over (next) or
7319 * extend the original comp (repeat). Otherwise, return the
7320 * error to the user. */
7321 if (rc == -ENOSPC && (next || sd->sd_repeat))
7327 if (sd->sd_force == 0 && rc == 0)
7328 rc = !lod_sel_osts_allowed(env, lo, index - 1, reserve, extent,
7329 &lod_comp->llc_extent, write);
7331 repeated = !!(sd->sd_repeat);
7335 /* Extend previous component */
7337 new_end = lod_extension_new_end(extension_size, extent->e_end,
7338 prev->llc_stripe_size,
7339 prev->llc_extent.e_end,
7340 lod_comp->llc_extent.e_end);
7342 CDEBUG(D_LAYOUT, "new end %llu\n", new_end);
7343 lod_comp->llc_extent.e_start = new_end;
7344 prev->llc_extent.e_end = new_end;
7346 if (prev->llc_extent.e_end == lod_comp->llc_extent.e_end) {
7347 CDEBUG(D_LAYOUT, "Extension component exhausted\n");
7348 lod_comp->llc_id = LCME_ID_INVAL;
7352 /* rc == 1, failed to extend current component */
7355 /* Normal 'spillover' case - Remove the extension
7356 * space component & bring down the start of the next
7358 lod_comp->llc_id = LCME_ID_INVAL;
7360 if (!(prev->llc_flags & LCME_FL_INIT)) {
7361 prev->llc_id = LCME_ID_INVAL;
7364 lod_sel_adjust_extents(env, lo, *max_comp, index);
7365 } else if (lod_comp_inited(prev)) {
7366 /* If there is no next, and the previous component is
7367 * INIT'ed, try repeating the previous component. */
7368 LASSERT(repeated == 0);
7369 rc = lod_layout_repeat_comp(env, lo, index - 1);
7373 /* The previous component is a repeated component.
7374 * Record this so we don't keep trying to repeat it. */
7377 /* If the previous component is not INIT'ed, this may
7378 * be a component we have just instantiated but failed
7379 * to extend. Or even a repeated component we failed
7380 * to prepare a striping for. Do not repeat but instead
7381 * remove the repeated component & force the extention
7382 * of the original one */
7385 prev->llc_id = LCME_ID_INVAL;
7392 rc = lod_layout_del_prep_layout(env, lo, NULL);
7395 LASSERTF(-rc == change,
7396 "number deleted %d != requested %d\n", -rc,
7399 *max_comp = *max_comp + change;
7401 /* lod_del_prep_layout reallocates ldo_comp_entries, so we must
7402 * refresh these pointers before using them */
7403 lod_comp = &lo->ldo_comp_entries[index];
7404 prev = &lo->ldo_comp_entries[index - 1];
7405 CDEBUG(D_LAYOUT, "After extent updates: prev start %llu, current start "
7406 "%llu, current end %llu max_comp %d ldo_comp_cnt %d\n",
7407 prev->llc_extent.e_start, lod_comp->llc_extent.e_start,
7408 lod_comp->llc_extent.e_end, *max_comp, lo->ldo_comp_cnt);
7410 /* Layout changed successfully */
7415 * Declare layout extent updates
7417 * Handles extensions. Identifies extension components touched by current
7418 * operation and passes them to processing function.
7420 * Restarts with updated layouts from the processing function until the current
7421 * operation no longer touches an extension space component.
7423 * \param[in] env execution environment for this thread
7424 * \param[in,out] lo object to update the layout of
7425 * \param[in] extent layout extent for requested operation, update layout to
7426 * fit this operation
7427 * \param[in] th transaction handle for this operation
7428 * \param[in] pick identifies chosen mirror for FLR layouts
7429 * \param[in] write if this is write op
7431 * \retval 1 on layout changed, 0 on no change
7432 * \retval negative errno on error
7434 static int lod_declare_update_extents(const struct lu_env *env,
7435 struct lod_object *lo, struct lu_extent *extent,
7436 struct thandle *th, int pick, int write)
7438 struct lod_thread_info *info = lod_env_info(env);
7439 struct lod_layout_component *lod_comp;
7440 bool layout_changed = false;
7441 struct sel_data sd = { 0 };
7449 /* This makes us work on the components of the chosen mirror */
7450 start_index = lo->ldo_mirrors[pick].lme_start;
7451 max_comp = lo->ldo_mirrors[pick].lme_end + 1;
7452 if (lo->ldo_flr_state == LCM_FL_NONE)
7453 LASSERT(start_index == 0 && max_comp == lo->ldo_comp_cnt);
7455 CDEBUG(D_LAYOUT, "extent->e_start %llu, extent->e_end %llu\n",
7456 extent->e_start, extent->e_end);
7457 for (i = start_index; i < max_comp; i++) {
7458 lod_comp = &lo->ldo_comp_entries[i];
7460 /* We've passed all components of interest */
7461 if (lod_comp->llc_extent.e_start >= extent->e_end)
7464 if (lod_comp->llc_flags & LCME_FL_EXTENSION) {
7465 layout_changed = true;
7466 rc = lod_sel_handler(env, lo, extent, th, &max_comp,
7471 /* Nothing has changed behind the prev one */
7477 /* We may have added or removed components. If so, we must update the
7478 * start & ends of all the mirrors after the current one, and the end
7479 * of the current mirror. */
7480 change = max_comp - 1 - lo->ldo_mirrors[pick].lme_end;
7482 lo->ldo_mirrors[pick].lme_end += change;
7483 for (i = pick + 1; i < lo->ldo_mirror_count; i++) {
7484 lo->ldo_mirrors[i].lme_start += change;
7485 lo->ldo_mirrors[i].lme_end += change;
7491 /* The amount of components has changed, adjust the lti_comp_idx */
7492 rc2 = lod_layout_data_init(info, lo->ldo_comp_cnt);
7494 return rc < 0 ? rc : rc2 < 0 ? rc2 : layout_changed;
7497 /* If striping is already instantiated or INIT'ed DOM? */
7498 static bool lod_is_instantiation_needed(struct lod_layout_component *comp)
7500 if (comp->llc_magic == LOV_MAGIC_FOREIGN)
7503 return !(((lov_pattern(comp->llc_pattern) & LOV_PATTERN_MDT) &&
7504 lod_comp_inited(comp)) || comp->llc_stripe);
7508 * Declare layout update for a non-FLR layout.
7510 * \param[in] env execution environment for this thread
7511 * \param[in,out] lo object to update the layout of
7512 * \param[in] layout layout intent for requested operation, "update" is
7513 * a process of reacting to this
7514 * \param[in] buf buffer containing lov ea (see comment on usage inline)
7515 * \param[in] th transaction handle for this operation
7517 * \retval 0 on success
7518 * \retval negative errno on error
7520 static int lod_declare_update_plain(const struct lu_env *env,
7521 struct lod_object *lo, struct layout_intent *layout,
7522 const struct lu_buf *buf, struct thandle *th)
7524 struct lod_thread_info *info = lod_env_info(env);
7525 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7526 struct lod_layout_component *lod_comp;
7527 struct lov_comp_md_v1 *comp_v1 = NULL;
7528 bool layout_changed = false;
7529 bool replay = false;
7533 LASSERT(lo->ldo_flr_state == LCM_FL_NONE);
7536 * In case the client is passing lovea, which only happens during
7537 * the replay of layout intent write RPC for now, we may need to
7538 * parse the lovea and apply new layout configuration.
7540 if (buf && buf->lb_len) {
7541 struct lov_user_md_v1 *v1 = buf->lb_buf;
7543 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
7544 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
7545 LOV_MAGIC_COMP_V1)) {
7546 CERROR("%s: the replay buffer of layout extend "
7547 "(magic %#x) does not contain expected "
7548 "composite layout.\n",
7549 lod2obd(d)->obd_name, v1->lmm_magic);
7550 GOTO(out, rc = -EINVAL);
7553 rc = lod_use_defined_striping(env, lo, buf);
7556 lo->ldo_comp_cached = 1;
7558 rc = lod_get_lov_ea(env, lo);
7561 /* old on-disk EA is stored in info->lti_buf */
7562 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
7564 layout_changed = true;
7566 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
7570 /* non replay path */
7571 rc = lod_striping_load(env, lo);
7576 /* Make sure defined layout covers the requested write range. */
7577 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
7578 if (lo->ldo_comp_cnt > 1 &&
7579 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
7580 lod_comp->llc_extent.e_end < layout->li_extent.e_end) {
7581 CDEBUG_LIMIT(replay ? D_ERROR : D_LAYOUT,
7582 "%s: the defined layout [0, %#llx) does not "
7583 "covers the write range "DEXT"\n",
7584 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
7585 PEXT(&layout->li_extent));
7586 GOTO(out, rc = -EINVAL);
7589 CDEBUG(D_LAYOUT, "%s: "DFID": update components "DEXT"\n",
7590 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
7591 PEXT(&layout->li_extent));
7594 rc = lod_declare_update_extents(env, lo, &layout->li_extent,
7595 th, 0, layout->li_opc == LAYOUT_INTENT_WRITE);
7599 layout_changed = true;
7603 * Iterate ld->ldo_comp_entries, find the component whose extent under
7604 * the write range and not instantianted.
7606 for (i = 0; i < lo->ldo_comp_cnt; i++) {
7607 lod_comp = &lo->ldo_comp_entries[i];
7609 if (lod_comp->llc_extent.e_start >= layout->li_extent.e_end)
7613 /* If striping is instantiated or INIT'ed DOM skip */
7614 if (!lod_is_instantiation_needed(lod_comp))
7618 * In replay path, lod_comp is the EA passed by
7619 * client replay buffer, comp_v1 is the pre-recovery
7620 * on-disk EA, we'd sift out those components which
7621 * were init-ed in the on-disk EA.
7623 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
7628 * this component hasn't instantiated in normal path, or during
7629 * replay it needs replay the instantiation.
7632 /* A released component is being extended */
7633 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
7634 GOTO(out, rc = -EINVAL);
7636 LASSERT(info->lti_comp_idx != NULL);
7637 info->lti_comp_idx[info->lti_count++] = i;
7638 layout_changed = true;
7641 if (!layout_changed)
7644 lod_obj_inc_layout_gen(lo);
7645 rc = lod_declare_instantiate_components(env, lo, th, 0);
7649 lod_striping_free(env, lo);
7653 static inline int lod_comp_index(struct lod_object *lo,
7654 struct lod_layout_component *lod_comp)
7656 LASSERT(lod_comp >= lo->ldo_comp_entries &&
7657 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
7659 return lod_comp - lo->ldo_comp_entries;
7663 * Stale other mirrors by writing extent.
7665 static int lod_stale_components(const struct lu_env *env, struct lod_object *lo,
7666 int primary, struct lu_extent *extent,
7669 struct lod_layout_component *pri_comp, *lod_comp;
7670 struct lu_extent pri_extent;
7675 /* The writing extent decides which components in the primary
7676 * are affected... */
7677 CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
7680 lod_foreach_mirror_comp(pri_comp, lo, primary) {
7681 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
7684 CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
7685 lod_comp_index(lo, pri_comp),
7686 PEXT(&pri_comp->llc_extent));
7688 pri_extent.e_start = pri_comp->llc_extent.e_start;
7689 pri_extent.e_end = pri_comp->llc_extent.e_end;
7691 for (i = 0; i < lo->ldo_mirror_count; i++) {
7694 rc = lod_declare_update_extents(env, lo, &pri_extent,
7696 /* if update_extents changed the layout, it may have
7697 * reallocated the component array, so start over to
7698 * avoid using stale pointers */
7704 /* ... and then stale other components that are
7705 * overlapping with primary components */
7706 lod_foreach_mirror_comp(lod_comp, lo, i) {
7707 if (!lu_extent_is_overlapped(
7709 &lod_comp->llc_extent))
7712 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
7713 i, lod_comp_index(lo, lod_comp));
7715 lod_comp->llc_flags |= LCME_FL_STALE;
7716 lo->ldo_mirrors[i].lme_stale = 1;
7717 if (lod_is_hsm(lod_comp))
7718 lod_comp->llc_foreign_flags |= HS_DIRTY;
7727 * check an OST's availability
7728 * \param[in] env execution environment
7729 * \param[in] lo lod object
7730 * \param[in] dt dt object
7731 * \param[in] index mirror index
7733 * \retval negative if failed
7734 * \retval 1 if \a dt is available
7735 * \retval 0 if \a dt is not available
7737 static inline int lod_check_ost_avail(const struct lu_env *env,
7738 struct lod_object *lo,
7739 struct dt_object *dt, int index)
7741 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7742 struct lod_tgt_desc *ost;
7744 int type = LU_SEQ_RANGE_OST;
7747 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &idx, &type);
7749 CERROR("%s: can't locate "DFID":rc = %d\n",
7750 lod2obd(lod)->obd_name, PFID(lu_object_fid(&dt->do_lu)),
7755 ost = OST_TGT(lod, idx);
7756 if (ost->ltd_active == 0) {
7757 CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail\n",
7758 PFID(lod_object_fid(lo)), index, idx);
7766 * Pick primary mirror for write
7767 * \param[in] env execution environment
7768 * \param[in] lo object
7769 * \param[in] extent write range
7771 static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo,
7772 struct lu_extent *extent)
7774 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7775 unsigned int seq = 0;
7776 struct lod_layout_component *lod_comp;
7778 int picked = -1, second_pick = -1, third_pick = -1;
7781 if (CFS_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
7782 get_random_bytes(&seq, sizeof(seq));
7783 seq %= lo->ldo_mirror_count;
7787 * Pick a mirror as the primary, and check the availability of OSTs.
7789 * This algo can be revised later after knowing the topology of
7792 lod_qos_statfs_update(env, lod, &lod->lod_ost_descs);
7794 rc = lod_fill_mirrors(lo);
7798 for (i = 0; i < lo->ldo_mirror_count; i++) {
7799 bool ost_avail = true;
7800 int index = (i + seq) % lo->ldo_mirror_count;
7802 if (lo->ldo_mirrors[index].lme_stale) {
7803 CDEBUG(D_LAYOUT, DFID": mirror %d stale\n",
7804 PFID(lod_object_fid(lo)), index);
7808 /* 2nd pick is for the primary mirror containing unavail OST */
7809 if (lo->ldo_mirrors[index].lme_prefer && second_pick < 0)
7810 second_pick = index;
7812 /* 3rd pick is for non-primary mirror containing unavail OST */
7813 if (second_pick < 0 && third_pick < 0)
7817 * we found a non-primary 1st pick, we'd like to find a
7818 * potential pirmary mirror.
7820 if (picked >= 0 && !lo->ldo_mirrors[index].lme_prefer)
7823 /* check the availability of OSTs */
7824 lod_foreach_mirror_comp(lod_comp, lo, index) {
7825 if (!lod_comp_inited(lod_comp) || !lod_comp->llc_stripe)
7828 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
7829 struct dt_object *dt = lod_comp->llc_stripe[j];
7831 rc = lod_check_ost_avail(env, lo, dt, index);
7838 } /* for all dt object in one component */
7841 } /* for all components in a mirror */
7844 * the OSTs where allocated objects locates in the components
7845 * of the mirror are available.
7850 /* this mirror has all OSTs available */
7854 * primary with all OSTs are available, this is the perfect
7857 if (lo->ldo_mirrors[index].lme_prefer)
7859 } /* for all mirrors */
7861 /* failed to pick a sound mirror, lower our expectation */
7863 picked = second_pick;
7865 picked = third_pick;
7872 static int lod_prepare_resync_mirror(const struct lu_env *env,
7873 struct lod_object *lo,
7876 struct lod_thread_info *info = lod_env_info(env);
7877 struct lod_layout_component *lod_comp;
7878 bool neg = !!(MIRROR_ID_NEG & mirror_id);
7881 mirror_id &= ~MIRROR_ID_NEG;
7883 for (i = 0; i < lo->ldo_mirror_count; i++) {
7884 if ((!neg && lo->ldo_mirrors[i].lme_id != mirror_id) ||
7885 (neg && lo->ldo_mirrors[i].lme_id == mirror_id))
7888 lod_foreach_mirror_comp(lod_comp, lo, i) {
7889 if (lod_comp_inited(lod_comp))
7892 info->lti_comp_idx[info->lti_count++] =
7893 lod_comp_index(lo, lod_comp);
7901 * figure out the components should be instantiated for resync.
7903 static int lod_prepare_resync(const struct lu_env *env, struct lod_object *lo,
7904 struct lu_extent *extent)
7906 struct lod_thread_info *info = lod_env_info(env);
7907 struct lod_layout_component *lod_comp;
7908 unsigned int need_sync = 0;
7912 DFID": instantiate all stale components in "DEXT"\n",
7913 PFID(lod_object_fid(lo)), PEXT(extent));
7916 * instantiate all components within this extent, even non-stale
7919 for (i = 0; i < lo->ldo_mirror_count; i++) {
7920 if (!lo->ldo_mirrors[i].lme_stale)
7923 lod_foreach_mirror_comp(lod_comp, lo, i) {
7924 if (!lu_extent_is_overlapped(extent,
7925 &lod_comp->llc_extent))
7930 if (lod_comp_inited(lod_comp))
7933 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
7934 i, lod_comp_index(lo, lod_comp));
7935 info->lti_comp_idx[info->lti_count++] =
7936 lod_comp_index(lo, lod_comp);
7940 return need_sync ? 0 : -EALREADY;
7943 static struct lod_layout_component *
7944 lod_locate_comp_hsm(struct lod_object *lo, int *hsm_mirror_id)
7946 struct lod_layout_component *lod_comp = NULL;
7949 if (!lo->ldo_is_composite)
7952 for (i = 0; i < lo->ldo_mirror_count; i++) {
7954 * FIXME: In the current design, there is only one HSM
7955 * mirror component in range [0, EOF] for a FLR file. This
7956 * should be fixed to support multiple HSM mirror components
7957 * with different HSM backend types and partial file ranges
7960 if (lo->ldo_mirrors[i].lme_hsm) {
7966 start_idx = lo->ldo_mirrors[i].lme_start;
7967 end_idx = lo->ldo_mirrors[i].lme_end;
7968 LASSERT(start_idx == end_idx);
7969 lod_comp = &lo->ldo_comp_entries[start_idx];
7970 LASSERT(lo->ldo_is_composite && lod_is_hsm(lod_comp) &&
7971 lod_comp->llc_extent.e_start == 0 &&
7972 lod_comp->llc_extent.e_end == LUSTRE_EOF);
7980 static int lod_declare_pccro_set(const struct lu_env *env,
7981 struct dt_object *dt, struct thandle *th)
7983 struct lod_thread_info *info = lod_env_info(env);
7984 struct lu_buf *buf = &info->lti_buf;
7985 struct lod_object *lo = lod_dt_obj(dt);
7986 struct lod_layout_component *lod_comp;
7987 struct lod_layout_component *comp_array;
7988 struct lod_mirror_entry *mirror_array;
7998 rc = lod_striping_load(env, lo);
8002 if (lo->ldo_flr_state & LCM_FL_PCC_RDONLY)
8005 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
8009 lod_comp = lod_locate_comp_hsm(lo, &hsm_mirror_id);
8011 if (lod_comp->llc_foreign_flags & HS_PCCRO) {
8012 CDEBUG(D_LAYOUT, "bad HSM flags: %#x\n",
8013 lod_comp->llc_foreign_flags);
8017 lod_obj_inc_layout_gen(lo);
8018 lod_comp->llc_foreign_flags |= HS_PCCRO;
8019 lod_comp->llc_foreign_flags &= ~HS_DIRTY;
8020 lod_comp->llc_flags &= ~LCME_FL_STALE;
8021 lo->ldo_mirrors[hsm_mirror_id].lme_stale = 0;
8022 lo->ldo_flr_state |= LCM_FL_PCC_RDONLY;
8023 buf->lb_len = lod_comp_md_size(lo, false);
8024 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
8025 buf, XATTR_NAME_LOV, 0, th);
8030 * Create an new composite layout with only one HSM component.
8031 * Field @lhm_archive_uuid is used to be the identifier within HSM
8032 * backend for the archive copy. In the PCC case with a POSIX archive,
8033 * This can just be the original inode FID. This is important because
8034 * the inode FID may change due to layout swaps or migration to a new
8035 * MDT, and we do not want that to cause problems with finding the copy
8038 mirror_cnt = lo->ldo_mirror_count + 1;
8039 if (!lo->ldo_is_composite) {
8040 LASSERT(lo->ldo_mirror_count == 0);
8044 OBD_ALLOC_PTR_ARRAY(mirror_array, mirror_cnt);
8045 if (mirror_array == NULL)
8048 new_cnt = lo->ldo_comp_cnt + 1;
8049 OBD_ALLOC_PTR_ARRAY(comp_array, new_cnt);
8050 if (comp_array == NULL) {
8051 OBD_FREE_PTR_ARRAY(mirror_array, mirror_cnt);
8056 for (i = 0; i < lo->ldo_comp_cnt; i++) {
8057 lod_comp = &lo->ldo_comp_entries[i];
8060 * Add mirror from a non-flr file, create new mirror ID.
8061 * Otherwise, keep existing mirror's component ID, used
8062 * for mirror extension.
8064 if (lo->ldo_mirror_count == 0 &&
8065 mirror_id_of(lod_comp->llc_id) == 0)
8066 lod_comp->llc_id = pflr_id(1, i + 1);
8068 if (lod_comp->llc_id != LCME_ID_INVAL &&
8069 mirror_id_of(lod_comp->llc_id) > mirror_id)
8070 mirror_id = mirror_id_of(lod_comp->llc_id);
8072 if (!lo->ldo_is_composite) {
8073 lod_comp->llc_extent.e_start = 0;
8074 lod_comp->llc_extent.e_end = LUSTRE_EOF;
8075 lod_comp_set_init(lod_comp);
8079 memcpy(comp_array, lo->ldo_comp_entries,
8080 sizeof(*comp_array) * lo->ldo_comp_cnt);
8082 lod_comp = &comp_array[new_cnt - 1];
8083 lod_comp->llc_magic = LOV_MAGIC_FOREIGN;
8084 lod_comp->llc_extent.e_start = 0;
8085 lod_comp->llc_extent.e_end = LUSTRE_EOF;
8086 lod_comp->llc_length = sizeof(struct lov_hsm_base);
8087 lod_comp->llc_type = LU_FOREIGN_TYPE_PCCRO;
8088 lod_comp->llc_foreign_flags = HS_EXISTS | HS_ARCHIVED | HS_PCCRO;
8089 memset(&lod_comp->llc_hsm, 0, sizeof(lod_comp->llc_hsm));
8091 if (lo->ldo_mirrors)
8092 OBD_FREE_PTR_ARRAY(lo->ldo_mirrors, lo->ldo_mirror_count);
8093 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
8096 * The @ldo_mirror will be refilled by lod_fill_mirrors() when
8097 * call lod_striped_create() for layout change.
8099 lo->ldo_mirrors = mirror_array;
8100 lo->ldo_mirror_count = mirror_cnt;
8101 lo->ldo_comp_entries = comp_array;
8102 lo->ldo_comp_cnt = new_cnt;
8103 lo->ldo_is_composite = 1;
8106 lod_comp->llc_id = LCME_ID_INVAL;
8107 lod_comp->llc_id = lod_gen_component_id(lo, mirror_id, new_cnt - 1);
8109 if (lo->ldo_flr_state == LCM_FL_NONE)
8110 lo->ldo_flr_state = LCM_FL_RDONLY;
8111 lo->ldo_flr_state |= LCM_FL_PCC_RDONLY;
8112 buf->lb_len = lod_comp_md_size(lo, false);
8113 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
8114 buf, XATTR_NAME_LOV, 0, th);
8116 lod_striping_free(env, lo);
8122 * TODO: When clear LCM_FL_PCC_RDONLY flag from the layouts, it means the file
8123 * is going to be modified. Currently it needs two RPCs: first one is to clear
8124 * LCM_FL_PCC_RDONLY flag; the second one is to pick primary mirror and mark
8125 * the file as LCM_FL_WRITE_PENDING.
8126 * These two RPCs can be combined in one RPC call.
8128 static int lod_declare_pccro_clear(const struct lu_env *env,
8129 struct dt_object *dt, struct thandle *th)
8131 struct lod_thread_info *info = lod_env_info(env);
8132 struct lod_object *lo = lod_dt_obj(dt);
8133 struct lod_layout_component *lod_comp;
8138 rc = lod_striping_load(env, lo);
8142 if (!(lo->ldo_flr_state & LCM_FL_PCC_RDONLY))
8145 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
8149 lod_comp = lod_locate_comp_hsm(lo, NULL);
8150 if (lod_comp == NULL) {
8151 CDEBUG(D_LAYOUT, "Not found any HSM component\n");
8152 GOTO(out, rc = -EINVAL);
8155 lod_comp->llc_foreign_flags &= ~HS_PCCRO;
8156 lo->ldo_flr_state &= ~LCM_FL_PCC_RDONLY;
8157 lod_obj_inc_layout_gen(lo);
8158 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
8159 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
8160 &info->lti_buf, XATTR_NAME_LOV, 0, th);
8163 lod_striping_free(env, lo);
8168 static int lod_declare_update_pccro(const struct lu_env *env,
8169 struct dt_object *dt,
8170 struct md_layout_change *mlc,
8173 struct layout_intent *intent = mlc->mlc_intent;
8176 switch (intent->li_opc) {
8177 case LAYOUT_INTENT_PCCRO_SET:
8178 rc = lod_declare_pccro_set(env, dt, th);
8180 case LAYOUT_INTENT_PCCRO_CLEAR:
8181 rc = lod_declare_pccro_clear(env, dt, th);
8191 static int lod_declare_update_rdonly(const struct lu_env *env,
8192 struct lod_object *lo, struct md_layout_change *mlc,
8195 struct lod_thread_info *info = lod_env_info(env);
8196 struct lu_attr *layout_attr = &info->lti_layout_attr;
8197 struct lod_layout_component *lod_comp;
8198 struct lu_extent extent = { 0 };
8202 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
8203 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
8204 mlc->mlc_opc == MD_LAYOUT_RESYNC);
8205 LASSERT(lo->ldo_mirror_count > 0);
8207 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
8208 struct layout_intent *layout = mlc->mlc_intent;
8209 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
8212 extent = layout->li_extent;
8213 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
8214 PFID(lod_object_fid(lo)), PEXT(&extent));
8216 picked = lod_primary_pick(env, lo, &extent);
8220 CDEBUG(D_LAYOUT, DFID": picked mirror id %u as primary\n",
8221 PFID(lod_object_fid(lo)),
8222 lo->ldo_mirrors[picked].lme_id);
8224 /* Update extents of primary before staling */
8225 rc = lod_declare_update_extents(env, lo, &extent, th, picked,
8230 if (layout->li_opc == LAYOUT_INTENT_TRUNC) {
8232 * trunc transfers [0, size) in the intent extent, we'd
8233 * stale components overlapping [size, eof).
8235 extent.e_start = extent.e_end;
8236 extent.e_end = OBD_OBJECT_EOF;
8239 /* stale overlapping components from other mirrors */
8240 rc = lod_stale_components(env, lo, picked, &extent, th);
8244 /* restore truncate intent extent */
8245 if (layout->li_opc == LAYOUT_INTENT_TRUNC)
8246 extent.e_end = extent.e_start;
8248 /* instantiate components for the picked mirror, start from 0 */
8251 lod_foreach_mirror_comp(lod_comp, lo, picked) {
8252 if (!lu_extent_is_overlapped(&extent,
8253 &lod_comp->llc_extent))
8256 if (!lod_is_instantiation_needed(lod_comp))
8259 info->lti_comp_idx[info->lti_count++] =
8260 lod_comp_index(lo, lod_comp);
8263 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
8264 } else { /* MD_LAYOUT_RESYNC */
8268 * could contain multiple non-stale mirrors, so we need to
8269 * prep uninited all components assuming any non-stale mirror
8270 * could be picked as the primary mirror.
8272 if (mlc->mlc_mirror_id == 0) {
8274 for (i = 0; i < lo->ldo_mirror_count; i++) {
8275 if (lo->ldo_mirrors[i].lme_stale)
8278 lod_foreach_mirror_comp(lod_comp, lo, i) {
8279 if (!lod_comp_inited(lod_comp))
8283 lod_comp->llc_extent.e_end)
8285 lod_comp->llc_extent.e_end;
8288 rc = lod_prepare_resync(env, lo, &extent);
8292 /* mirror write, try to init its all components */
8293 rc = lod_prepare_resync_mirror(env, lo,
8294 mlc->mlc_mirror_id);
8299 /* change the file state to SYNC_PENDING */
8300 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
8303 /* Reset the layout version once it's becoming too large.
8304 * This way it can make sure that the layout version is
8305 * monotonously increased in this writing era. */
8306 lod_obj_inc_layout_gen(lo);
8308 rc = lod_declare_instantiate_components(env, lo, th, 0);
8312 layout_attr->la_valid = LA_LAYOUT_VERSION;
8313 layout_attr->la_layout_version = 0;
8314 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
8315 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
8316 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
8322 lod_striping_free(env, lo);
8326 static int lod_declare_update_write_pending(const struct lu_env *env,
8327 struct lod_object *lo, struct md_layout_change *mlc,
8330 struct lod_thread_info *info = lod_env_info(env);
8331 struct lu_attr *layout_attr = &info->lti_layout_attr;
8332 struct lod_layout_component *lod_comp;
8333 struct lu_extent extent = { 0 };
8339 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
8340 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
8341 mlc->mlc_opc == MD_LAYOUT_RESYNC);
8343 /* look for the first preferred mirror */
8344 for (i = 0; i < lo->ldo_mirror_count; i++) {
8345 if (lo->ldo_mirrors[i].lme_stale)
8347 if (lo->ldo_mirrors[i].lme_prefer == 0)
8349 if (lo->ldo_mirrors[i].lme_hsm)
8356 /* no primary, use any in-sync */
8357 for (i = 0; i < lo->ldo_mirror_count; i++) {
8358 if (lo->ldo_mirrors[i].lme_stale)
8364 CERROR(DFID ": doesn't have a primary mirror\n",
8365 PFID(lod_object_fid(lo)));
8366 GOTO(out, rc = -ENODATA);
8370 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
8371 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
8373 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
8375 /* for LAYOUT_WRITE opc, it has to do the following operations:
8376 * 1. stale overlapping componets from stale mirrors;
8377 * 2. instantiate components of the primary mirror;
8378 * 3. transfter layout version to all objects of the primary;
8380 * for LAYOUT_RESYNC opc, it will do:
8381 * 1. instantiate components of all stale mirrors;
8382 * 2. transfer layout version to all objects to close write era. */
8384 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
8385 struct layout_intent *layout = mlc->mlc_intent;
8386 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
8388 LASSERT(mlc->mlc_intent != NULL);
8390 extent = mlc->mlc_intent->li_extent;
8392 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
8393 PFID(lod_object_fid(lo)), PEXT(&extent));
8395 /* 1. Update extents of primary before staling */
8396 rc = lod_declare_update_extents(env, lo, &extent, th, primary,
8401 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC) {
8403 * trunc transfers [0, size) in the intent extent, we'd
8404 * stale components overlapping [size, eof).
8406 extent.e_start = extent.e_end;
8407 extent.e_end = OBD_OBJECT_EOF;
8410 /* 2. stale overlapping components */
8411 rc = lod_stale_components(env, lo, primary, &extent, th);
8415 /* 3. find the components which need instantiating.
8416 * instantiate [0, mlc->mlc_intent->e_end) */
8418 /* restore truncate intent extent */
8419 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC)
8420 extent.e_end = extent.e_start;
8423 lod_foreach_mirror_comp(lod_comp, lo, primary) {
8424 if (!lu_extent_is_overlapped(&extent,
8425 &lod_comp->llc_extent))
8428 if (!lod_is_instantiation_needed(lod_comp))
8431 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
8432 primary, lod_comp_index(lo, lod_comp));
8433 info->lti_comp_idx[info->lti_count++] =
8434 lod_comp_index(lo, lod_comp);
8436 } else { /* MD_LAYOUT_RESYNC */
8437 if (mlc->mlc_mirror_id == 0) {
8439 lod_foreach_mirror_comp(lod_comp, lo, primary) {
8440 if (!lod_comp_inited(lod_comp))
8443 extent.e_end = lod_comp->llc_extent.e_end;
8446 rc = lod_prepare_resync(env, lo, &extent);
8450 /* mirror write, try to init its all components */
8451 rc = lod_prepare_resync_mirror(env, lo,
8452 mlc->mlc_mirror_id);
8457 /* change the file state to SYNC_PENDING */
8458 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
8461 rc = lod_declare_instantiate_components(env, lo, th, 0);
8465 lod_obj_inc_layout_gen(lo);
8467 /* 3. transfer layout version to OST objects.
8468 * transfer new layout version to OST objects so that stale writes
8469 * can be denied. It also ends an era of writing by setting
8470 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
8471 * send write RPC; only resync RPCs could do it. */
8472 layout_attr->la_valid = LA_LAYOUT_VERSION;
8473 layout_attr->la_layout_version = 0;
8474 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
8475 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
8476 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
8481 lod_striping_free(env, lo);
8485 static int lod_declare_update_sync_pending(const struct lu_env *env,
8486 struct lod_object *lo, struct md_layout_change *mlc,
8489 struct lod_thread_info *info = lod_env_info(env);
8490 struct lu_attr *layout_attr = &info->lti_layout_attr;
8491 unsigned sync_components = 0;
8492 unsigned resync_components = 0;
8497 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
8498 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
8499 mlc->mlc_opc == MD_LAYOUT_WRITE);
8501 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
8502 PFID(lod_object_fid(lo)), mlc->mlc_opc);
8504 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
8505 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
8506 PFID(lod_object_fid(lo)));
8508 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
8509 return lod_declare_update_write_pending(env, lo, mlc, th);
8512 /* MD_LAYOUT_RESYNC_DONE */
8514 for (i = 0; i < lo->ldo_comp_cnt; i++) {
8515 struct lod_layout_component *lod_comp;
8518 lod_comp = &lo->ldo_comp_entries[i];
8520 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
8525 for (j = 0; j < mlc->mlc_resync_count; j++) {
8526 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
8529 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
8530 lod_comp->llc_flags &= ~LCME_FL_STALE;
8531 resync_components++;
8537 for (i = 0; i < mlc->mlc_resync_count; i++) {
8538 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
8541 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
8542 "or already synced\n", PFID(lod_object_fid(lo)),
8543 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
8544 GOTO(out, rc = -EINVAL);
8547 if (!sync_components || (mlc->mlc_resync_count && !resync_components)) {
8548 CDEBUG(D_LAYOUT, DFID": no mirror in sync\n",
8549 PFID(lod_object_fid(lo)));
8551 /* tend to return an error code here to prevent
8552 * the MDT from setting SoM attribute */
8553 GOTO(out, rc = -EINVAL);
8556 CDEBUG(D_LAYOUT, DFID": synced %u resynced %u/%zu components\n",
8557 PFID(lod_object_fid(lo)),
8558 sync_components, resync_components, mlc->mlc_resync_count);
8560 lo->ldo_flr_state = LCM_FL_RDONLY;
8561 lod_obj_inc_layout_gen(lo);
8563 layout_attr->la_valid = LA_LAYOUT_VERSION;
8564 layout_attr->la_layout_version = 0;
8565 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
8569 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
8570 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
8571 &info->lti_buf, XATTR_NAME_LOV, 0, th);
8576 lod_striping_free(env, lo);
8580 typedef int (*mlc_handler)(const struct lu_env *env, struct dt_object *dt,
8581 const struct md_layout_change *mlc,
8582 struct thandle *th);
8585 * Attach stripes after target's for migrating directory. NB, we
8586 * only need to declare this, the actual work is done inside
8587 * lod_xattr_set_lmv().
8589 * \param[in] env execution environment
8590 * \param[in] dt target object
8591 * \param[in] mlc layout change data
8592 * \param[in] th transaction handle
8594 * \retval 0 on success
8595 * \retval negative if failed
8597 static int lod_dir_declare_layout_attach(const struct lu_env *env,
8598 struct dt_object *dt,
8599 const struct md_layout_change *mlc,
8602 struct lod_thread_info *info = lod_env_info(env);
8603 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
8604 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
8605 struct lod_object *lo = lod_dt_obj(dt);
8606 struct dt_object *next = dt_object_child(dt);
8607 struct dt_object_format *dof = &info->lti_format;
8608 struct lmv_mds_md_v1 *lmv = mlc->mlc_buf.lb_buf;
8609 struct dt_object **stripes;
8610 __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
8611 struct lu_fid *fid = &info->lti_fid;
8612 struct lod_tgt_desc *tgt;
8613 struct dt_object *dto;
8614 struct dt_device *tgt_dt;
8615 int type = LU_SEQ_RANGE_ANY;
8616 struct dt_insert_rec *rec = &info->lti_dt_rec;
8617 char *stripe_name = info->lti_key;
8618 struct lu_name *sname;
8619 struct linkea_data ldata = { NULL };
8620 struct lu_buf linkea_buf;
8627 if (!lmv_is_sane(lmv))
8630 if (!dt_try_as_dir(env, dt, false))
8633 dof->dof_type = DFT_DIR;
8635 OBD_ALLOC_PTR_ARRAY(stripes, (lo->ldo_dir_stripe_count + stripe_count));
8639 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
8640 stripes[i] = lo->ldo_stripe[i];
8642 rec->rec_type = S_IFDIR;
8644 for (i = 0; i < stripe_count; i++) {
8646 &lmv->lmv_stripe_fids[i]);
8647 if (!fid_is_sane(fid))
8650 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
8654 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
8655 tgt_dt = lod->lod_child;
8657 tgt = LTD_TGT(ltd, idx);
8659 GOTO(out, rc = -ESTALE);
8660 tgt_dt = tgt->ltd_tgt;
8663 dto = dt_locate_at(env, tgt_dt, fid,
8664 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
8667 GOTO(out, rc = PTR_ERR(dto));
8669 stripes[i + lo->ldo_dir_stripe_count] = dto;
8671 if (!dt_try_as_dir(env, dto, true))
8672 GOTO(out, rc = -ENOTDIR);
8674 rc = lod_sub_declare_ref_add(env, dto, th);
8678 rec->rec_fid = lu_object_fid(&dto->do_lu);
8679 rc = lod_sub_declare_insert(env, dto,
8680 (const struct dt_rec *)rec,
8681 (const struct dt_key *)dot, th);
8685 rc = lod_sub_declare_insert(env, dto,
8686 (const struct dt_rec *)rec,
8687 (const struct dt_key *)dotdot, th);
8691 rc = lod_sub_declare_xattr_set(env, dto, &mlc->mlc_buf,
8692 XATTR_NAME_LMV, 0, th);
8696 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
8697 PFID(lu_object_fid(&dto->do_lu)),
8698 i + lo->ldo_dir_stripe_count);
8700 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
8701 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
8702 sname, lu_object_fid(&dt->do_lu));
8706 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
8707 linkea_buf.lb_len = ldata.ld_leh->leh_len;
8708 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
8709 XATTR_NAME_LINK, 0, th);
8713 rc = lod_sub_declare_insert(env, next,
8714 (const struct dt_rec *)rec,
8715 (const struct dt_key *)stripe_name,
8720 rc = lod_sub_declare_ref_add(env, next, th);
8726 OBD_FREE_PTR_ARRAY(lo->ldo_stripe,
8727 lo->ldo_dir_stripes_allocated);
8728 lo->ldo_stripe = stripes;
8729 lo->ldo_is_foreign = 0;
8730 lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count;
8731 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type);
8732 lo->ldo_dir_stripe_count += stripe_count;
8733 lo->ldo_dir_layout_version++;
8734 lo->ldo_dir_stripes_allocated += stripe_count;
8736 /* plain directory split creates target as a plain directory, while
8737 * after source attached as the first stripe, it becomes a striped
8738 * directory, set correct do_index_ops, otherwise it can't be unlinked.
8740 dt->do_index_ops = &lod_striped_index_ops;
8744 i = lo->ldo_dir_stripe_count;
8745 while (i < lo->ldo_dir_stripe_count + stripe_count && stripes[i])
8746 dt_object_put(env, stripes[i++]);
8748 OBD_FREE_PTR_ARRAY(stripes, stripe_count + lo->ldo_dir_stripe_count);
8752 static int lod_dir_declare_layout_detach(const struct lu_env *env,
8753 struct dt_object *dt,
8754 const struct md_layout_change *unused,
8757 struct lod_thread_info *info = lod_env_info(env);
8758 struct lod_object *lo = lod_dt_obj(dt);
8759 struct dt_object *next = dt_object_child(dt);
8760 char *stripe_name = info->lti_key;
8761 struct dt_object *dto;
8765 if (!dt_try_as_dir(env, dt, true))
8768 if (!lo->ldo_dir_stripe_count)
8769 return lod_sub_declare_delete(env, next,
8770 (const struct dt_key *)dotdot, th);
8772 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8773 dto = lo->ldo_stripe[i];
8777 if (!dt_try_as_dir(env, dto, true))
8780 rc = lod_sub_declare_delete(env, dto,
8781 (const struct dt_key *)dotdot, th);
8785 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8786 PFID(lu_object_fid(&dto->do_lu)), i);
8788 rc = lod_sub_declare_delete(env, next,
8789 (const struct dt_key *)stripe_name, th);
8793 rc = lod_sub_declare_ref_del(env, next, th);
8801 static int dt_dir_is_empty(const struct lu_env *env,
8802 struct dt_object *obj)
8805 const struct dt_it_ops *iops;
8810 if (!dt_try_as_dir(env, obj, true))
8813 iops = &obj->do_index_ops->dio_it;
8814 it = iops->init(env, obj, LUDA_64BITHASH);
8816 RETURN(PTR_ERR(it));
8818 rc = iops->get(env, it, (const struct dt_key *)"");
8822 for (rc = 0, i = 0; rc == 0 && i < 3; ++i)
8823 rc = iops->next(env, it);
8829 /* Huh? Index contains no zero key? */
8834 iops->fini(env, it);
8839 static int lod_dir_declare_layout_shrink(const struct lu_env *env,
8840 struct dt_object *dt,
8841 const struct md_layout_change *mlc,
8844 struct lod_thread_info *info = lod_env_info(env);
8845 struct lod_object *lo = lod_dt_obj(dt);
8846 struct dt_object *next = dt_object_child(dt);
8847 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
8848 char *stripe_name = info->lti_key;
8849 struct lu_buf *lmv_buf = &info->lti_buf;
8850 __u32 final_stripe_count;
8851 struct dt_object *dto;
8857 if (!dt_try_as_dir(env, dt, true))
8860 /* shouldn't be called on plain directory */
8861 LASSERT(lo->ldo_dir_stripe_count);
8863 lmv_buf->lb_buf = &info->lti_lmv.lmv_md_v1;
8864 lmv_buf->lb_len = sizeof(info->lti_lmv.lmv_md_v1);
8866 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
8867 LASSERT(final_stripe_count &&
8868 final_stripe_count < lo->ldo_dir_stripe_count);
8870 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8871 dto = lo->ldo_stripe[i];
8875 if (i < final_stripe_count) {
8876 rc = lod_sub_declare_xattr_set(env, dto, lmv_buf,
8878 LU_XATTR_REPLACE, th);
8885 rc = dt_dir_is_empty(env, dto);
8889 rc = lod_sub_declare_ref_del(env, dto, th);
8893 rc = lod_sub_declare_destroy(env, dto, th);
8897 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8898 PFID(lu_object_fid(&dto->do_lu)), i);
8900 rc = lod_sub_declare_delete(env, next,
8901 (const struct dt_key *)stripe_name, th);
8905 rc = lod_sub_declare_ref_del(env, next, th);
8910 rc = lod_sub_declare_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
8911 LU_XATTR_REPLACE, th);
8916 * Allocate stripes for split directory.
8918 * \param[in] env execution environment
8919 * \param[in] dt target object
8920 * \param[in] mlc layout change data
8921 * \param[in] th transaction handle
8923 * \retval 0 on success
8924 * \retval negative if failed
8926 static int lod_dir_declare_layout_split(const struct lu_env *env,
8927 struct dt_object *dt,
8928 const struct md_layout_change *mlc,
8931 struct lod_thread_info *info = lod_env_info(env);
8932 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
8933 struct lod_object *lo = lod_dt_obj(dt);
8934 struct dt_object_format *dof = &info->lti_format;
8935 struct lmv_user_md_v1 *lum = mlc->mlc_spec->u.sp_ea.eadata;
8936 struct dt_object **stripes;
8944 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
8945 LASSERT(le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT);
8947 saved_count = lo->ldo_dir_stripes_allocated;
8948 stripe_count = le32_to_cpu(lum->lum_stripe_count);
8949 if (stripe_count <= saved_count)
8952 dof->dof_type = DFT_DIR;
8954 OBD_ALLOC(stripes, sizeof(*stripes) * stripe_count);
8958 for (i = 0; i < lo->ldo_dir_stripes_allocated; i++)
8959 stripes[i] = lo->ldo_stripe[i];
8961 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
8962 rc = lod_mdt_alloc_qos(env, lo, stripes, saved_count, stripe_count);
8964 rc = lod_mdt_alloc_rr(env, lo, stripes, saved_count,
8967 OBD_FREE(stripes, sizeof(*stripes) * stripe_count);
8971 LASSERT(rc > saved_count);
8972 OBD_FREE(lo->ldo_stripe,
8973 sizeof(*stripes) * lo->ldo_dir_stripes_allocated);
8974 lo->ldo_stripe = stripes;
8975 lo->ldo_is_foreign = 0;
8976 lo->ldo_dir_striped = 1;
8977 lo->ldo_dir_stripe_count = rc;
8978 lo->ldo_dir_stripes_allocated = stripe_count;
8979 lo->ldo_dir_split_hash = lo->ldo_dir_hash_type;
8980 lo->ldo_dir_hash_type = le32_to_cpu(lum->lum_hash_type);
8981 if (!lmv_is_known_hash_type(lo->ldo_dir_hash_type))
8982 lo->ldo_dir_hash_type =
8983 lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
8984 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_SPLIT | LMV_HASH_FLAG_MIGRATION;
8985 lo->ldo_dir_split_offset = saved_count;
8986 lo->ldo_dir_layout_version++;
8987 lo->ldo_dir_stripe_loaded = 1;
8989 rc = lod_dir_declare_create_stripes(env, dt, mlc->mlc_attr, dof, th);
8991 lod_striping_free(env, lo);
8997 * detach all stripes from dir master object, NB, stripes are not destroyed, but
8998 * deleted from it's parent namespace, this function is called in two places:
8999 * 1. mdd_migrate_mdt() detach stripes from source, and attach them to
9001 * 2. mdd_dir_layout_update() detach stripe before turning 1-stripe directory to
9002 * a plain directory.
9004 * \param[in] env execution environment
9005 * \param[in] dt target object
9006 * \param[in] mlc layout change data
9007 * \param[in] th transaction handle
9009 * \retval 0 on success
9010 * \retval negative if failed
9012 static int lod_dir_layout_detach(const struct lu_env *env,
9013 struct dt_object *dt,
9014 const struct md_layout_change *mlc,
9017 struct lod_thread_info *info = lod_env_info(env);
9018 struct lod_object *lo = lod_dt_obj(dt);
9019 struct dt_object *next = dt_object_child(dt);
9020 char *stripe_name = info->lti_key;
9021 struct dt_object *dto;
9027 if (!lo->ldo_dir_stripe_count) {
9028 /* plain directory delete .. */
9029 rc = lod_sub_delete(env, next,
9030 (const struct dt_key *)dotdot, th);
9034 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
9035 dto = lo->ldo_stripe[i];
9039 rc = lod_sub_delete(env, dto,
9040 (const struct dt_key *)dotdot, th);
9044 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
9045 PFID(lu_object_fid(&dto->do_lu)), i);
9047 rc = lod_sub_delete(env, next,
9048 (const struct dt_key *)stripe_name, th);
9052 rc = lod_sub_ref_del(env, next, th);
9057 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
9058 dto = lo->ldo_stripe[i];
9060 dt_object_put(env, dto);
9062 OBD_FREE_PTR_ARRAY(lo->ldo_stripe, lo->ldo_dir_stripes_allocated);
9063 lo->ldo_stripe = NULL;
9064 lo->ldo_dir_stripes_allocated = 0;
9065 lo->ldo_dir_stripe_count = 0;
9066 dt->do_index_ops = &lod_index_ops;
9071 static int lod_dir_layout_shrink(const struct lu_env *env,
9072 struct dt_object *dt,
9073 const struct md_layout_change *mlc,
9076 struct lod_thread_info *info = lod_env_info(env);
9077 struct lod_object *lo = lod_dt_obj(dt);
9078 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
9079 struct dt_object *next = dt_object_child(dt);
9080 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
9081 __u32 final_stripe_count;
9082 char *stripe_name = info->lti_key;
9083 struct dt_object *dto;
9084 struct lu_buf *lmv_buf = &info->lti_buf;
9085 struct lmv_mds_md_v1 *lmv = &info->lti_lmv.lmv_md_v1;
9087 int type = LU_SEQ_RANGE_ANY;
9093 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
9095 lmv_buf->lb_buf = lmv;
9096 lmv_buf->lb_len = sizeof(*lmv);
9097 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
9098 lmv->lmv_stripe_count = cpu_to_le32(final_stripe_count);
9099 lmv->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type) &
9100 cpu_to_le32(LMV_HASH_TYPE_MASK |
9101 LMV_HASH_FLAG_FIXED);
9102 lmv->lmv_layout_version =
9103 cpu_to_le32(lo->ldo_dir_layout_version + 1);
9104 lmv->lmv_migrate_offset = 0;
9105 lmv->lmv_migrate_hash = 0;
9107 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
9108 dto = lo->ldo_stripe[i];
9112 if (i < final_stripe_count) {
9113 rc = lod_fld_lookup(env, lod,
9114 lu_object_fid(&dto->do_lu),
9119 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
9120 rc = lod_sub_xattr_set(env, dto, lmv_buf,
9122 LU_XATTR_REPLACE, th);
9129 dt_write_lock(env, dto, DT_TGT_CHILD);
9130 rc = lod_sub_ref_del(env, dto, th);
9131 dt_write_unlock(env, dto);
9135 rc = lod_sub_destroy(env, dto, th);
9139 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
9140 PFID(lu_object_fid(&dto->do_lu)), i);
9142 rc = lod_sub_delete(env, next,
9143 (const struct dt_key *)stripe_name, th);
9147 rc = lod_sub_ref_del(env, next, th);
9152 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &mdtidx,
9157 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
9158 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
9159 rc = lod_sub_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
9160 LU_XATTR_REPLACE, th);
9164 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
9165 dto = lo->ldo_stripe[i];
9167 dt_object_put(env, dto);
9169 lo->ldo_dir_stripe_count = final_stripe_count;
9174 static mlc_handler dir_mlc_declare_ops[MD_LAYOUT_MAX] = {
9175 [MD_LAYOUT_ATTACH] = lod_dir_declare_layout_attach,
9176 [MD_LAYOUT_DETACH] = lod_dir_declare_layout_detach,
9177 [MD_LAYOUT_SHRINK] = lod_dir_declare_layout_shrink,
9178 [MD_LAYOUT_SPLIT] = lod_dir_declare_layout_split,
9181 static mlc_handler dir_mlc_ops[MD_LAYOUT_MAX] = {
9182 [MD_LAYOUT_DETACH] = lod_dir_layout_detach,
9183 [MD_LAYOUT_SHRINK] = lod_dir_layout_shrink,
9186 static int lod_declare_layout_change(const struct lu_env *env,
9187 struct dt_object *dt, struct md_layout_change *mlc,
9190 struct lod_thread_info *info = lod_env_info(env);
9191 struct lod_object *lo = lod_dt_obj(dt);
9196 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
9197 LASSERT(dir_mlc_declare_ops[mlc->mlc_opc]);
9198 rc = dir_mlc_declare_ops[mlc->mlc_opc](env, dt, mlc, th);
9202 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
9203 dt_object_remote(dt_object_child(dt)))
9206 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
9207 struct layout_intent *intent = mlc->mlc_intent;
9209 if (intent->li_opc == LAYOUT_INTENT_PCCRO_SET ||
9210 intent->li_opc == LAYOUT_INTENT_PCCRO_CLEAR) {
9211 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
9214 rc = lod_declare_update_pccro(env, dt, mlc, th);
9219 rc = lod_striping_load(env, lo);
9223 LASSERT(lo->ldo_comp_cnt > 0);
9225 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
9229 switch (lo->ldo_flr_state) {
9231 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
9235 rc = lod_declare_update_rdonly(env, lo, mlc, th);
9237 case LCM_FL_WRITE_PENDING:
9238 rc = lod_declare_update_write_pending(env, lo, mlc, th);
9240 case LCM_FL_SYNC_PENDING:
9241 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
9252 * Instantiate layout component objects which covers the intent write offset.
9254 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
9255 struct md_layout_change *mlc, struct thandle *th)
9257 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
9258 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
9259 struct lod_object *lo = lod_dt_obj(dt);
9264 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
9265 LASSERT(dir_mlc_ops[mlc->mlc_opc]);
9266 rc = dir_mlc_ops[mlc->mlc_opc](env, dt, mlc, th);
9270 rc = lod_striped_create(env, dt, attr, NULL, th);
9271 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
9272 layout_attr->la_layout_version |= lo->ldo_layout_gen;
9273 rc = lod_attr_set(env, dt, layout_attr, th);
9279 const struct dt_object_operations lod_obj_ops = {
9280 .do_read_lock = lod_read_lock,
9281 .do_write_lock = lod_write_lock,
9282 .do_read_unlock = lod_read_unlock,
9283 .do_write_unlock = lod_write_unlock,
9284 .do_write_locked = lod_write_locked,
9285 .do_attr_get = lod_attr_get,
9286 .do_declare_attr_set = lod_declare_attr_set,
9287 .do_attr_set = lod_attr_set,
9288 .do_xattr_get = lod_xattr_get,
9289 .do_declare_xattr_set = lod_declare_xattr_set,
9290 .do_xattr_set = lod_xattr_set,
9291 .do_declare_xattr_del = lod_declare_xattr_del,
9292 .do_xattr_del = lod_xattr_del,
9293 .do_xattr_list = lod_xattr_list,
9294 .do_ah_init = lod_ah_init,
9295 .do_declare_create = lod_declare_create,
9296 .do_create = lod_create,
9297 .do_declare_destroy = lod_declare_destroy,
9298 .do_destroy = lod_destroy,
9299 .do_index_try = lod_index_try,
9300 .do_declare_ref_add = lod_declare_ref_add,
9301 .do_ref_add = lod_ref_add,
9302 .do_declare_ref_del = lod_declare_ref_del,
9303 .do_ref_del = lod_ref_del,
9304 .do_object_sync = lod_object_sync,
9305 .do_object_lock = lod_object_lock,
9306 .do_object_unlock = lod_object_unlock,
9307 .do_invalidate = lod_invalidate,
9308 .do_declare_layout_change = lod_declare_layout_change,
9309 .do_layout_change = lod_layout_change,
9313 * Implementation of dt_body_operations::dbo_read.
9315 * \see dt_body_operations::dbo_read() in the API description for details.
9317 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
9318 struct lu_buf *buf, loff_t *pos)
9320 struct dt_object *next = dt_object_child(dt);
9322 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
9323 S_ISLNK(dt->do_lu.lo_header->loh_attr));
9324 return next->do_body_ops->dbo_read(env, next, buf, pos);
9328 * Implementation of dt_body_operations::dbo_declare_write.
9330 * \see dt_body_operations::dbo_declare_write() in the API description
9333 static ssize_t lod_declare_write(const struct lu_env *env,
9334 struct dt_object *dt,
9335 const struct lu_buf *buf, loff_t pos,
9338 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
9342 * Implementation of dt_body_operations::dbo_write.
9344 * \see dt_body_operations::dbo_write() in the API description for details.
9346 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
9347 const struct lu_buf *buf, loff_t *pos,
9350 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
9351 S_ISLNK(dt->do_lu.lo_header->loh_attr));
9352 return lod_sub_write(env, dt_object_child(dt), buf, pos, th);
9355 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
9356 __u64 start, __u64 end, struct thandle *th)
9358 if (dt_object_remote(dt))
9361 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
9364 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
9365 __u64 start, __u64 end, struct thandle *th)
9367 if (dt_object_remote(dt))
9370 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
9371 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
9375 * different type of files use the same body_ops because object may be created
9376 * in OUT, where there is no chance to set correct body_ops for each type, so
9377 * body_ops themselves will check file type inside, see lod_read/write/punch for
9380 static const struct dt_body_operations lod_body_ops = {
9381 .dbo_read = lod_read,
9382 .dbo_declare_write = lod_declare_write,
9383 .dbo_write = lod_write,
9384 .dbo_declare_punch = lod_declare_punch,
9385 .dbo_punch = lod_punch,
9389 * Implementation of lu_object_operations::loo_object_init.
9391 * The function determines the type and the index of the target device using
9392 * sequence of the object's FID. Then passes control down to the
9393 * corresponding device:
9394 * OSD for the local objects, OSP for remote
9396 * \see lu_object_operations::loo_object_init() in the API description
9399 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
9400 const struct lu_object_conf *conf)
9402 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
9403 struct lu_device *cdev = NULL;
9404 struct lu_object *cobj;
9405 struct lod_tgt_descs *ltd = NULL;
9406 struct lod_tgt_desc *tgt;
9408 int type = LU_SEQ_RANGE_ANY;
9412 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
9416 if (type == LU_SEQ_RANGE_MDT &&
9417 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
9418 cdev = &lod->lod_child->dd_lu_dev;
9419 } else if (type == LU_SEQ_RANGE_MDT) {
9420 ltd = &lod->lod_mdt_descs;
9422 } else if (type == LU_SEQ_RANGE_OST) {
9423 ltd = &lod->lod_ost_descs;
9430 if (ltd->ltd_tgts_size > idx &&
9431 test_bit(idx, ltd->ltd_tgt_bitmap)) {
9432 tgt = LTD_TGT(ltd, idx);
9434 LASSERT(tgt != NULL);
9435 LASSERT(tgt->ltd_tgt != NULL);
9437 cdev = &(tgt->ltd_tgt->dd_lu_dev);
9439 lod_putref(lod, ltd);
9442 if (unlikely(cdev == NULL))
9445 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
9446 if (unlikely(cobj == NULL))
9449 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
9451 lu_object_add(lo, cobj);
9458 * Release resources associated with striping.
9460 * If the object is striped (regular or directory), then release
9461 * the stripe objects references and free the ldo_stripe array.
9463 * \param[in] env execution environment
9464 * \param[in] lo object
9466 void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo)
9468 struct lod_layout_component *lod_comp;
9469 __u32 obj_attr = lo->ldo_obj.do_lu.lo_header->loh_attr;
9472 if (unlikely(lo->ldo_is_foreign)) {
9473 if (S_ISREG(obj_attr)) {
9474 lod_free_foreign_lov(lo);
9475 lo->ldo_comp_cached = 0;
9476 } else if (S_ISDIR(obj_attr)) {
9477 lod_free_foreign_lmv(lo);
9478 lo->ldo_dir_stripe_loaded = 0;
9480 } else if (lo->ldo_stripe != NULL) {
9481 LASSERT(lo->ldo_comp_entries == NULL);
9482 LASSERT(lo->ldo_dir_stripes_allocated > 0);
9484 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
9485 if (lo->ldo_stripe[i])
9486 dt_object_put(env, lo->ldo_stripe[i]);
9489 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
9490 OBD_FREE(lo->ldo_stripe, j);
9491 lo->ldo_stripe = NULL;
9492 lo->ldo_dir_stripes_allocated = 0;
9493 lo->ldo_dir_stripe_loaded = 0;
9494 lo->ldo_dir_stripe_count = 0;
9495 lo->ldo_obj.do_index_ops = NULL;
9496 } else if (lo->ldo_comp_entries != NULL) {
9497 for (i = 0; i < lo->ldo_comp_cnt; i++) {
9498 /* free lod_layout_component::llc_stripe array */
9499 lod_comp = &lo->ldo_comp_entries[i];
9501 /* HSM layout component */
9502 if (lod_comp->llc_magic == LOV_MAGIC_FOREIGN)
9504 if (lod_comp->llc_stripe == NULL)
9506 LASSERT(lod_comp->llc_stripes_allocated != 0);
9507 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
9508 if (lod_comp->llc_stripe[j] != NULL)
9510 &lod_comp->llc_stripe[j]->do_lu);
9512 OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe,
9513 lod_comp->llc_stripes_allocated);
9514 lod_comp->llc_stripe = NULL;
9515 OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices,
9516 lod_comp->llc_stripes_allocated);
9517 lod_comp->llc_ost_indices = NULL;
9518 lod_comp->llc_stripes_allocated = 0;
9520 lod_free_comp_entries(lo);
9521 lo->ldo_comp_cached = 0;
9525 void lod_striping_free(const struct lu_env *env, struct lod_object *lo)
9527 mutex_lock(&lo->ldo_layout_mutex);
9528 lod_striping_free_nolock(env, lo);
9529 mutex_unlock(&lo->ldo_layout_mutex);
9533 * Implementation of lu_object_operations::loo_object_free.
9535 * \see lu_object_operations::loo_object_free() in the API description
9538 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
9540 struct lod_object *lo = lu2lod_obj(o);
9542 /* release all underlying object pinned */
9543 lod_striping_free(env, lo);
9545 /* lo doesn't contain a lu_object_header, so we don't need call_rcu */
9546 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
9550 * Implementation of lu_object_operations::loo_object_release.
9552 * \see lu_object_operations::loo_object_release() in the API description
9555 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
9557 /* XXX: shouldn't we release everything here in case if object
9558 * creation failed before? */
9562 * Implementation of lu_object_operations::loo_object_print.
9564 * \see lu_object_operations::loo_object_print() in the API description
9567 static int lod_object_print(const struct lu_env *env, void *cookie,
9568 lu_printer_t p, const struct lu_object *l)
9570 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
9572 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
9575 const struct lu_object_operations lod_lu_obj_ops = {
9576 .loo_object_init = lod_object_init,
9577 .loo_object_free = lod_object_free,
9578 .loo_object_release = lod_object_release,
9579 .loo_object_print = lod_object_print,