4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * lustre/lod/lod_object.c
31 * This file contains implementations of methods for the OSD API
32 * for the Logical Object Device (LOD) layer, which provides a virtual
33 * local OSD object interface to the MDD layer, and abstracts the
34 * addressing of local (OSD) and remote (OSP) objects. The API is
35 * described in the file lustre/include/dt_object.h and in
36 * Documentation/osd-api.txt.
38 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
41 #define DEBUG_SUBSYSTEM S_MDS
43 #include <linux/random.h>
46 #include <obd_class.h>
47 #include <obd_support.h>
49 #include <lustre_fid.h>
50 #include <lustre_linkea.h>
51 #include <lustre_lmv.h>
52 #include <uapi/linux/lustre/lustre_param.h>
53 #include <lustre_swab.h>
54 #include <uapi/linux/lustre/lustre_ver.h>
55 #include <lprocfs_status.h>
56 #include <md_object.h>
58 #include "lod_internal.h"
60 static const char dot[] = ".";
61 static const char dotdot[] = "..";
64 * Implementation of dt_index_operations::dio_lookup
66 * Used with regular (non-striped) objects.
68 * \see dt_index_operations::dio_lookup() in the API description for details.
70 static int lod_lookup(const struct lu_env *env, struct dt_object *dt,
71 struct dt_rec *rec, const struct dt_key *key)
73 struct dt_object *next = dt_object_child(dt);
74 return next->do_index_ops->dio_lookup(env, next, rec, key);
78 * Implementation of dt_index_operations::dio_declare_insert.
80 * Used with regular (non-striped) objects.
82 * \see dt_index_operations::dio_declare_insert() in the API description
85 static int lod_declare_insert(const struct lu_env *env, struct dt_object *dt,
86 const struct dt_rec *rec,
87 const struct dt_key *key, struct thandle *th)
89 return lod_sub_declare_insert(env, dt_object_child(dt), rec, key, th);
93 * Implementation of dt_index_operations::dio_insert.
95 * Used with regular (non-striped) objects
97 * \see dt_index_operations::dio_insert() in the API description for details.
99 static int lod_insert(const struct lu_env *env, struct dt_object *dt,
100 const struct dt_rec *rec, const struct dt_key *key,
103 return lod_sub_insert(env, dt_object_child(dt), rec, key, th);
107 * Implementation of dt_index_operations::dio_declare_delete.
109 * Used with regular (non-striped) objects.
111 * \see dt_index_operations::dio_declare_delete() in the API description
114 static int lod_declare_delete(const struct lu_env *env, struct dt_object *dt,
115 const struct dt_key *key, struct thandle *th)
117 return lod_sub_declare_delete(env, dt_object_child(dt), key, th);
121 * Implementation of dt_index_operations::dio_delete.
123 * Used with regular (non-striped) objects.
125 * \see dt_index_operations::dio_delete() in the API description for details.
127 static int lod_delete(const struct lu_env *env, struct dt_object *dt,
128 const struct dt_key *key, struct thandle *th)
130 return lod_sub_delete(env, dt_object_child(dt), key, th);
134 * Implementation of dt_it_ops::init.
136 * Used with regular (non-striped) objects.
138 * \see dt_it_ops::init() in the API description for details.
140 static struct dt_it *lod_it_init(const struct lu_env *env,
141 struct dt_object *dt, __u32 attr)
143 struct dt_object *next = dt_object_child(dt);
144 struct lod_it *it = &lod_env_info(env)->lti_it;
145 struct dt_it *it_next;
147 it_next = next->do_index_ops->dio_it.init(env, next, attr);
151 /* currently we do not use more than one iterator per thread
152 * so we store it in thread info. if at some point we need
153 * more active iterators in a single thread, we can allocate
155 LASSERT(it->lit_obj == NULL);
157 it->lit_it = it_next;
160 return (struct dt_it *)it;
163 #define LOD_CHECK_IT(env, it) \
165 LASSERT((it)->lit_obj != NULL); \
166 LASSERT((it)->lit_it != NULL); \
170 * Implementation of dt_index_operations::dio_it.fini.
172 * Used with regular (non-striped) objects.
174 * \see dt_index_operations::dio_it.fini() in the API description for details.
176 static void lod_it_fini(const struct lu_env *env, struct dt_it *di)
178 struct lod_it *it = (struct lod_it *)di;
180 LOD_CHECK_IT(env, it);
181 it->lit_obj->do_index_ops->dio_it.fini(env, it->lit_it);
183 /* the iterator not in use any more */
189 * Implementation of dt_it_ops::get.
191 * Used with regular (non-striped) objects.
193 * \see dt_it_ops::get() in the API description for details.
195 static int lod_it_get(const struct lu_env *env, struct dt_it *di,
196 const struct dt_key *key)
198 const struct lod_it *it = (const struct lod_it *)di;
200 LOD_CHECK_IT(env, it);
201 return it->lit_obj->do_index_ops->dio_it.get(env, it->lit_it, key);
205 * Implementation of dt_it_ops::put.
207 * Used with regular (non-striped) objects.
209 * \see dt_it_ops::put() in the API description for details.
211 static void lod_it_put(const struct lu_env *env, struct dt_it *di)
213 struct lod_it *it = (struct lod_it *)di;
215 LOD_CHECK_IT(env, it);
216 return it->lit_obj->do_index_ops->dio_it.put(env, it->lit_it);
220 * Implementation of dt_it_ops::next.
222 * Used with regular (non-striped) objects
224 * \see dt_it_ops::next() in the API description for details.
226 static int lod_it_next(const struct lu_env *env, struct dt_it *di)
228 struct lod_it *it = (struct lod_it *)di;
230 LOD_CHECK_IT(env, it);
231 return it->lit_obj->do_index_ops->dio_it.next(env, it->lit_it);
235 * Implementation of dt_it_ops::key.
237 * Used with regular (non-striped) objects.
239 * \see dt_it_ops::key() in the API description for details.
241 static struct dt_key *lod_it_key(const struct lu_env *env,
242 const struct dt_it *di)
244 const struct lod_it *it = (const struct lod_it *)di;
246 LOD_CHECK_IT(env, it);
247 return it->lit_obj->do_index_ops->dio_it.key(env, it->lit_it);
251 * Implementation of dt_it_ops::key_size.
253 * Used with regular (non-striped) objects.
255 * \see dt_it_ops::key_size() in the API description for details.
257 static int lod_it_key_size(const struct lu_env *env, const struct dt_it *di)
259 struct lod_it *it = (struct lod_it *)di;
261 LOD_CHECK_IT(env, it);
262 return it->lit_obj->do_index_ops->dio_it.key_size(env, it->lit_it);
266 * Implementation of dt_it_ops::rec.
268 * Used with regular (non-striped) objects.
270 * \see dt_it_ops::rec() in the API description for details.
272 static int lod_it_rec(const struct lu_env *env, const struct dt_it *di,
273 struct dt_rec *rec, __u32 attr)
275 const struct lod_it *it = (const struct lod_it *)di;
277 LOD_CHECK_IT(env, it);
278 return it->lit_obj->do_index_ops->dio_it.rec(env, it->lit_it, rec,
283 * Implementation of dt_it_ops::rec_size.
285 * Used with regular (non-striped) objects.
287 * \see dt_it_ops::rec_size() in the API description for details.
289 static int lod_it_rec_size(const struct lu_env *env, const struct dt_it *di,
292 const struct lod_it *it = (const struct lod_it *)di;
294 LOD_CHECK_IT(env, it);
295 return it->lit_obj->do_index_ops->dio_it.rec_size(env, it->lit_it,
300 * Implementation of dt_it_ops::store.
302 * Used with regular (non-striped) objects.
304 * \see dt_it_ops::store() in the API description for details.
306 static __u64 lod_it_store(const struct lu_env *env, const struct dt_it *di)
308 const struct lod_it *it = (const struct lod_it *)di;
310 LOD_CHECK_IT(env, it);
311 return it->lit_obj->do_index_ops->dio_it.store(env, it->lit_it);
315 * Implementation of dt_it_ops::load.
317 * Used with regular (non-striped) objects.
319 * \see dt_it_ops::load() in the API description for details.
321 static int lod_it_load(const struct lu_env *env, const struct dt_it *di,
324 const struct lod_it *it = (const struct lod_it *)di;
326 LOD_CHECK_IT(env, it);
327 return it->lit_obj->do_index_ops->dio_it.load(env, it->lit_it, hash);
331 * Implementation of dt_it_ops::key_rec.
333 * Used with regular (non-striped) objects.
335 * \see dt_it_ops::rec() in the API description for details.
337 static int lod_it_key_rec(const struct lu_env *env, const struct dt_it *di,
340 const struct lod_it *it = (const struct lod_it *)di;
342 LOD_CHECK_IT(env, it);
343 return it->lit_obj->do_index_ops->dio_it.key_rec(env, it->lit_it,
347 static struct dt_index_operations lod_index_ops = {
348 .dio_lookup = lod_lookup,
349 .dio_declare_insert = lod_declare_insert,
350 .dio_insert = lod_insert,
351 .dio_declare_delete = lod_declare_delete,
352 .dio_delete = lod_delete,
360 .key_size = lod_it_key_size,
362 .rec_size = lod_it_rec_size,
363 .store = lod_it_store,
365 .key_rec = lod_it_key_rec,
370 * Implementation of dt_index_operations::dio_lookup
372 * Used with striped directories.
374 * \see dt_index_operations::dio_lookup() in the API description for details.
376 static int lod_striped_lookup(const struct lu_env *env, struct dt_object *dt,
377 struct dt_rec *rec, const struct dt_key *key)
379 struct lod_object *lo = lod_dt_obj(dt);
380 struct dt_object *next;
381 const char *name = (const char *)key;
383 LASSERT(lo->ldo_dir_stripe_count > 0);
385 if (strcmp(name, dot) == 0) {
386 struct lu_fid *fid = (struct lu_fid *)rec;
388 *fid = *lod_object_fid(lo);
392 if (strcmp(name, dotdot) == 0) {
393 next = dt_object_child(dt);
397 index = __lmv_name_to_stripe_index(lo->ldo_dir_hash_type,
398 lo->ldo_dir_stripe_count,
399 lo->ldo_dir_migrate_hash,
400 lo->ldo_dir_migrate_offset,
401 name, strlen(name), true);
405 next = lo->ldo_stripe[index];
406 if (!next || !dt_object_exists(next))
410 return next->do_index_ops->dio_lookup(env, next, rec, key);
414 * Implementation of dt_it_ops::init.
416 * Used with striped objects. Internally just initializes the iterator
417 * on the first stripe.
419 * \see dt_it_ops::init() in the API description for details.
421 static struct dt_it *lod_striped_it_init(const struct lu_env *env,
422 struct dt_object *dt, __u32 attr)
424 struct lod_object *lo = lod_dt_obj(dt);
425 struct dt_object *next;
426 struct lod_it *it = &lod_env_info(env)->lti_it;
427 struct dt_it *it_next;
430 LASSERT(lo->ldo_dir_stripe_count > 0);
433 next = lo->ldo_stripe[index];
434 if (next && dt_object_exists(next))
436 } while (++index < lo->ldo_dir_stripe_count);
438 /* no valid stripe */
439 if (!next || !dt_object_exists(next))
440 return ERR_PTR(-ENODEV);
442 LASSERT(next->do_index_ops != NULL);
444 it_next = next->do_index_ops->dio_it.init(env, next, attr);
448 /* currently we do not use more than one iterator per thread
449 * so we store it in thread info. if at some point we need
450 * more active iterators in a single thread, we can allocate
452 LASSERT(it->lit_obj == NULL);
454 it->lit_stripe_index = index;
456 it->lit_it = it_next;
459 return (struct dt_it *)it;
462 #define LOD_CHECK_STRIPED_IT(env, it, lo) \
464 LASSERT((it)->lit_obj != NULL); \
465 LASSERT((it)->lit_it != NULL); \
466 LASSERT((lo)->ldo_dir_stripe_count > 0); \
467 LASSERT((it)->lit_stripe_index < (lo)->ldo_dir_stripe_count); \
471 * Implementation of dt_it_ops::fini.
473 * Used with striped objects.
475 * \see dt_it_ops::fini() in the API description for details.
477 static void lod_striped_it_fini(const struct lu_env *env, struct dt_it *di)
479 struct lod_it *it = (struct lod_it *)di;
480 struct lod_object *lo = lod_dt_obj(it->lit_obj);
481 struct dt_object *next;
483 /* If lit_it == NULL, then it means the sub_it has been finished,
484 * which only happens in failure cases, see lod_striped_it_next() */
485 if (it->lit_it != NULL) {
486 LOD_CHECK_STRIPED_IT(env, it, lo);
488 next = lo->ldo_stripe[it->lit_stripe_index];
490 LASSERT(next->do_index_ops != NULL);
491 next->do_index_ops->dio_it.fini(env, it->lit_it);
495 /* the iterator not in use any more */
498 it->lit_stripe_index = 0;
502 * Implementation of dt_it_ops::get.
504 * Right now it's not used widely, only to reset the iterator to the
505 * initial position. It should be possible to implement a full version
506 * which chooses a correct stripe to be able to position with any key.
508 * \see dt_it_ops::get() in the API description for details.
510 static int lod_striped_it_get(const struct lu_env *env, struct dt_it *di,
511 const struct dt_key *key)
513 const struct lod_it *it = (const struct lod_it *)di;
514 struct lod_object *lo = lod_dt_obj(it->lit_obj);
515 struct dt_object *next;
517 LOD_CHECK_STRIPED_IT(env, it, lo);
519 next = lo->ldo_stripe[it->lit_stripe_index];
520 LASSERT(next != NULL);
521 LASSERT(dt_object_exists(next));
522 LASSERT(next->do_index_ops != NULL);
524 return next->do_index_ops->dio_it.get(env, it->lit_it, key);
528 * Implementation of dt_it_ops::put.
530 * Used with striped objects.
532 * \see dt_it_ops::put() in the API description for details.
534 static void lod_striped_it_put(const struct lu_env *env, struct dt_it *di)
536 struct lod_it *it = (struct lod_it *)di;
537 struct lod_object *lo = lod_dt_obj(it->lit_obj);
538 struct dt_object *next;
541 * If lit_it == NULL, then it means the sub_it has been finished,
542 * which only happens in failure cases, see lod_striped_it_next()
547 LOD_CHECK_STRIPED_IT(env, it, lo);
549 next = lo->ldo_stripe[it->lit_stripe_index];
550 LASSERT(next != NULL);
551 LASSERT(next->do_index_ops != NULL);
553 return next->do_index_ops->dio_it.put(env, it->lit_it);
557 * Implementation of dt_it_ops::next.
559 * Used with striped objects. When the end of the current stripe is
560 * reached, the method takes the next stripe's iterator.
562 * \see dt_it_ops::next() in the API description for details.
564 static int lod_striped_it_next(const struct lu_env *env, struct dt_it *di)
566 struct lod_it *it = (struct lod_it *)di;
567 struct lod_object *lo = lod_dt_obj(it->lit_obj);
568 struct dt_object *next;
569 struct dt_it *it_next;
575 LOD_CHECK_STRIPED_IT(env, it, lo);
577 next = lo->ldo_stripe[it->lit_stripe_index];
578 LASSERT(next != NULL);
579 LASSERT(dt_object_exists(next));
580 LASSERT(next->do_index_ops != NULL);
582 rc = next->do_index_ops->dio_it.next(env, it->lit_it);
586 if (rc == 0 && it->lit_stripe_index == 0)
589 if (rc == 0 && it->lit_stripe_index > 0) {
590 struct lu_dirent *ent;
592 ent = (struct lu_dirent *)lod_env_info(env)->lti_key;
594 rc = next->do_index_ops->dio_it.rec(env, it->lit_it,
595 (struct dt_rec *)ent,
600 /* skip . and .. for slave stripe */
601 if ((strncmp(ent->lde_name, ".",
602 le16_to_cpu(ent->lde_namelen)) == 0 &&
603 le16_to_cpu(ent->lde_namelen) == 1) ||
604 (strncmp(ent->lde_name, "..",
605 le16_to_cpu(ent->lde_namelen)) == 0 &&
606 le16_to_cpu(ent->lde_namelen) == 2))
612 next->do_index_ops->dio_it.put(env, it->lit_it);
613 next->do_index_ops->dio_it.fini(env, it->lit_it);
616 /* go to next stripe */
617 index = it->lit_stripe_index;
618 while (++index < lo->ldo_dir_stripe_count) {
619 next = lo->ldo_stripe[index];
623 if (!dt_object_exists(next))
626 rc = next->do_ops->do_index_try(env, next,
627 &dt_directory_features);
631 LASSERT(next->do_index_ops != NULL);
633 it_next = next->do_index_ops->dio_it.init(env, next,
636 RETURN(PTR_ERR(it_next));
638 rc = next->do_index_ops->dio_it.get(env, it_next,
639 (const struct dt_key *)"");
641 RETURN(rc == 0 ? -EIO : rc);
643 it->lit_it = it_next;
644 it->lit_stripe_index = index;
653 * Implementation of dt_it_ops::key.
655 * Used with striped objects.
657 * \see dt_it_ops::key() in the API description for details.
659 static struct dt_key *lod_striped_it_key(const struct lu_env *env,
660 const struct dt_it *di)
662 const struct lod_it *it = (const struct lod_it *)di;
663 struct lod_object *lo = lod_dt_obj(it->lit_obj);
664 struct dt_object *next;
666 LOD_CHECK_STRIPED_IT(env, it, lo);
668 next = lo->ldo_stripe[it->lit_stripe_index];
669 LASSERT(next != NULL);
670 LASSERT(next->do_index_ops != NULL);
672 return next->do_index_ops->dio_it.key(env, it->lit_it);
676 * Implementation of dt_it_ops::key_size.
678 * Used with striped objects.
680 * \see dt_it_ops::size() in the API description for details.
682 static int lod_striped_it_key_size(const struct lu_env *env,
683 const struct dt_it *di)
685 struct lod_it *it = (struct lod_it *)di;
686 struct lod_object *lo = lod_dt_obj(it->lit_obj);
687 struct dt_object *next;
689 LOD_CHECK_STRIPED_IT(env, it, lo);
691 next = lo->ldo_stripe[it->lit_stripe_index];
692 LASSERT(next != NULL);
693 LASSERT(next->do_index_ops != NULL);
695 return next->do_index_ops->dio_it.key_size(env, it->lit_it);
699 * Implementation of dt_it_ops::rec.
701 * Used with striped objects.
703 * \see dt_it_ops::rec() in the API description for details.
705 static int lod_striped_it_rec(const struct lu_env *env, const struct dt_it *di,
706 struct dt_rec *rec, __u32 attr)
708 const struct lod_it *it = (const struct lod_it *)di;
709 struct lod_object *lo = lod_dt_obj(it->lit_obj);
710 struct dt_object *next;
712 LOD_CHECK_STRIPED_IT(env, it, lo);
714 next = lo->ldo_stripe[it->lit_stripe_index];
715 LASSERT(next != NULL);
716 LASSERT(next->do_index_ops != NULL);
718 return next->do_index_ops->dio_it.rec(env, it->lit_it, rec, attr);
722 * Implementation of dt_it_ops::rec_size.
724 * Used with striped objects.
726 * \see dt_it_ops::rec_size() in the API description for details.
728 static int lod_striped_it_rec_size(const struct lu_env *env,
729 const struct dt_it *di, __u32 attr)
731 struct lod_it *it = (struct lod_it *)di;
732 struct lod_object *lo = lod_dt_obj(it->lit_obj);
733 struct dt_object *next;
735 LOD_CHECK_STRIPED_IT(env, it, lo);
737 next = lo->ldo_stripe[it->lit_stripe_index];
738 LASSERT(next != NULL);
739 LASSERT(next->do_index_ops != NULL);
741 return next->do_index_ops->dio_it.rec_size(env, it->lit_it, attr);
745 * Implementation of dt_it_ops::store.
747 * Used with striped objects.
749 * \see dt_it_ops::store() in the API description for details.
751 static __u64 lod_striped_it_store(const struct lu_env *env,
752 const struct dt_it *di)
754 const struct lod_it *it = (const struct lod_it *)di;
755 struct lod_object *lo = lod_dt_obj(it->lit_obj);
756 struct dt_object *next;
758 LOD_CHECK_STRIPED_IT(env, it, lo);
760 next = lo->ldo_stripe[it->lit_stripe_index];
761 LASSERT(next != NULL);
762 LASSERT(next->do_index_ops != NULL);
764 return next->do_index_ops->dio_it.store(env, it->lit_it);
768 * Implementation of dt_it_ops::load.
770 * Used with striped objects.
772 * \see dt_it_ops::load() in the API description for details.
774 static int lod_striped_it_load(const struct lu_env *env,
775 const struct dt_it *di, __u64 hash)
777 const struct lod_it *it = (const struct lod_it *)di;
778 struct lod_object *lo = lod_dt_obj(it->lit_obj);
779 struct dt_object *next;
781 LOD_CHECK_STRIPED_IT(env, it, lo);
783 next = lo->ldo_stripe[it->lit_stripe_index];
784 LASSERT(next != NULL);
785 LASSERT(next->do_index_ops != NULL);
787 return next->do_index_ops->dio_it.load(env, it->lit_it, hash);
790 static struct dt_index_operations lod_striped_index_ops = {
791 .dio_lookup = lod_striped_lookup,
792 .dio_declare_insert = lod_declare_insert,
793 .dio_insert = lod_insert,
794 .dio_declare_delete = lod_declare_delete,
795 .dio_delete = lod_delete,
797 .init = lod_striped_it_init,
798 .fini = lod_striped_it_fini,
799 .get = lod_striped_it_get,
800 .put = lod_striped_it_put,
801 .next = lod_striped_it_next,
802 .key = lod_striped_it_key,
803 .key_size = lod_striped_it_key_size,
804 .rec = lod_striped_it_rec,
805 .rec_size = lod_striped_it_rec_size,
806 .store = lod_striped_it_store,
807 .load = lod_striped_it_load,
812 * Append the FID for each shard of the striped directory after the
813 * given LMV EA header.
815 * To simplify striped directory and the consistency verification,
816 * we only store the LMV EA header on disk, for both master object
817 * and slave objects. When someone wants to know the whole LMV EA,
818 * such as client readdir(), we can build the entrie LMV EA on the
819 * MDT side (in RAM) via iterating the sub-directory entries that
820 * are contained in the master object of the stripe directory.
822 * For the master object of the striped directroy, the valid name
823 * for each shard is composed of the ${shard_FID}:${shard_idx}.
825 * There may be holes in the LMV EA if some shards' name entries
826 * are corrupted or lost.
828 * \param[in] env pointer to the thread context
829 * \param[in] lo pointer to the master object of the striped directory
830 * \param[in] buf pointer to the lu_buf which will hold the LMV EA
831 * \param[in] resize whether re-allocate the buffer if it is not big enough
833 * \retval positive size of the LMV EA
834 * \retval 0 for nothing to be loaded
835 * \retval negative error number on failure
837 int lod_load_lmv_shards(const struct lu_env *env, struct lod_object *lo,
838 struct lu_buf *buf, bool resize)
840 struct lu_dirent *ent =
841 (struct lu_dirent *)lod_env_info(env)->lti_key;
842 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
843 struct dt_object *obj = dt_object_child(&lo->ldo_obj);
844 struct lmv_mds_md_v1 *lmv1 = buf->lb_buf;
846 const struct dt_it_ops *iops;
848 __u32 magic = le32_to_cpu(lmv1->lmv_magic);
853 if (magic != LMV_MAGIC_V1)
856 stripes = le32_to_cpu(lmv1->lmv_stripe_count);
860 rc = lmv_mds_md_size(stripes, magic);
864 if (buf->lb_len < lmv1_size) {
873 lu_buf_alloc(buf, lmv1_size);
878 memcpy(buf->lb_buf, tbuf.lb_buf, tbuf.lb_len);
881 if (unlikely(!dt_try_as_dir(env, obj)))
884 memset(&lmv1->lmv_stripe_fids[0], 0, stripes * sizeof(struct lu_fid));
885 iops = &obj->do_index_ops->dio_it;
886 it = iops->init(env, obj, LUDA_64BITHASH);
890 rc = iops->load(env, it, 0);
892 rc = iops->next(env, it);
897 char name[FID_LEN + 2] = "";
902 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
908 fid_le_to_cpu(&fid, &ent->lde_fid);
909 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
910 if (ent->lde_name[0] == '.') {
911 if (ent->lde_namelen == 1)
914 if (ent->lde_namelen == 2 && ent->lde_name[1] == '.')
918 len = scnprintf(name, sizeof(name),
919 DFID":", PFID(&ent->lde_fid));
920 /* The ent->lde_name is composed of ${FID}:${index} */
921 if (ent->lde_namelen < len + 1 ||
922 memcmp(ent->lde_name, name, len) != 0) {
923 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
924 "%s: invalid shard name %.*s with the FID "DFID
925 " for the striped directory "DFID", %s\n",
926 lod2obd(lod)->obd_name, ent->lde_namelen,
927 ent->lde_name, PFID(&fid),
928 PFID(lu_object_fid(&obj->do_lu)),
929 lod->lod_lmv_failout ? "failout" : "skip");
931 if (lod->lod_lmv_failout)
939 if (ent->lde_name[len] < '0' ||
940 ent->lde_name[len] > '9') {
941 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
942 "%s: invalid shard name %.*s with the "
943 "FID "DFID" for the striped directory "
945 lod2obd(lod)->obd_name, ent->lde_namelen,
946 ent->lde_name, PFID(&fid),
947 PFID(lu_object_fid(&obj->do_lu)),
948 lod->lod_lmv_failout ?
951 if (lod->lod_lmv_failout)
957 index = index * 10 + ent->lde_name[len++] - '0';
958 } while (len < ent->lde_namelen);
960 if (len == ent->lde_namelen) {
961 /* Out of LMV EA range. */
962 if (index >= stripes) {
963 CERROR("%s: the shard %.*s for the striped "
964 "directory "DFID" is out of the known "
965 "LMV EA range [0 - %u], failout\n",
966 lod2obd(lod)->obd_name, ent->lde_namelen,
968 PFID(lu_object_fid(&obj->do_lu)),
974 /* The slot has been occupied. */
975 if (!fid_is_zero(&lmv1->lmv_stripe_fids[index])) {
979 &lmv1->lmv_stripe_fids[index]);
980 CERROR("%s: both the shard "DFID" and "DFID
981 " for the striped directory "DFID
982 " claim the same LMV EA slot at the "
983 "index %d, failout\n",
984 lod2obd(lod)->obd_name,
985 PFID(&fid0), PFID(&fid),
986 PFID(lu_object_fid(&obj->do_lu)), index);
991 /* stored as LE mode */
992 lmv1->lmv_stripe_fids[index] = ent->lde_fid;
995 rc = iops->next(env, it);
1000 iops->fini(env, it);
1002 RETURN(rc > 0 ? lmv_mds_md_size(stripes, magic) : rc);
1006 * Implementation of dt_object_operations::do_index_try.
1008 * \see dt_object_operations::do_index_try() in the API description for details.
1010 static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
1011 const struct dt_index_features *feat)
1013 struct lod_object *lo = lod_dt_obj(dt);
1014 struct dt_object *next = dt_object_child(dt);
1018 LASSERT(next->do_ops);
1019 LASSERT(next->do_ops->do_index_try);
1021 rc = lod_striping_load(env, lo);
1025 rc = next->do_ops->do_index_try(env, next, feat);
1029 if (lo->ldo_dir_stripe_count > 0) {
1032 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1033 if (!lo->ldo_stripe[i])
1035 if (!dt_object_exists(lo->ldo_stripe[i]))
1037 rc = lo->ldo_stripe[i]->do_ops->do_index_try(env,
1038 lo->ldo_stripe[i], feat);
1042 dt->do_index_ops = &lod_striped_index_ops;
1044 dt->do_index_ops = &lod_index_ops;
1051 * Implementation of dt_object_operations::do_read_lock.
1053 * \see dt_object_operations::do_read_lock() in the API description for details.
1055 static void lod_read_lock(const struct lu_env *env, struct dt_object *dt,
1058 dt_read_lock(env, dt_object_child(dt), role);
1062 * Implementation of dt_object_operations::do_write_lock.
1064 * \see dt_object_operations::do_write_lock() in the API description for
1067 static void lod_write_lock(const struct lu_env *env, struct dt_object *dt,
1070 dt_write_lock(env, dt_object_child(dt), role);
1074 * Implementation of dt_object_operations::do_read_unlock.
1076 * \see dt_object_operations::do_read_unlock() in the API description for
1079 static void lod_read_unlock(const struct lu_env *env, struct dt_object *dt)
1081 dt_read_unlock(env, dt_object_child(dt));
1085 * Implementation of dt_object_operations::do_write_unlock.
1087 * \see dt_object_operations::do_write_unlock() in the API description for
1090 static void lod_write_unlock(const struct lu_env *env, struct dt_object *dt)
1092 dt_write_unlock(env, dt_object_child(dt));
1096 * Implementation of dt_object_operations::do_write_locked.
1098 * \see dt_object_operations::do_write_locked() in the API description for
1101 static int lod_write_locked(const struct lu_env *env, struct dt_object *dt)
1103 return dt_write_locked(env, dt_object_child(dt));
1107 * Implementation of dt_object_operations::do_attr_get.
1109 * \see dt_object_operations::do_attr_get() in the API description for details.
1111 static int lod_attr_get(const struct lu_env *env,
1112 struct dt_object *dt,
1113 struct lu_attr *attr)
1115 /* Note: for striped directory, client will merge attributes
1116 * from all of the sub-stripes see lmv_merge_attr(), and there
1117 * no MDD logic depend on directory nlink/size/time, so we can
1118 * always use master inode nlink and size for now. */
1119 return dt_attr_get(env, dt_object_child(dt), attr);
1122 void lod_adjust_stripe_size(struct lod_layout_component *comp,
1123 __u32 def_stripe_size)
1125 __u64 comp_end = comp->llc_extent.e_end;
1127 /* Choose stripe size if not set. Note that default stripe size can't
1128 * be used as is, because it must be multiplier of given component end.
1129 * - first check if default stripe size can be used
1130 * - if not than select the lowest set bit from component end and use
1131 * that value as stripe size
1133 if (!comp->llc_stripe_size) {
1134 if (comp_end == LUSTRE_EOF || !(comp_end % def_stripe_size))
1135 comp->llc_stripe_size = def_stripe_size;
1137 comp->llc_stripe_size = comp_end & ~(comp_end - 1);
1139 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_stripe_size) {
1148 /* fix that even for defined stripe size but warn
1149 * about the problem, that must not happen
1151 CWARN("Component end %llu is not aligned by the stripe size %u\n",
1152 comp_end, comp->llc_stripe_size);
1153 comp->llc_stripe_size = comp_end & ~(comp_end - 1);
1158 static inline void lod_adjust_stripe_info(struct lod_layout_component *comp,
1159 struct lov_desc *desc,
1162 if (comp->llc_pattern != LOV_PATTERN_MDT) {
1163 if (append_stripes) {
1164 comp->llc_stripe_count = append_stripes;
1165 } else if (!comp->llc_stripe_count) {
1166 comp->llc_stripe_count =
1167 desc->ld_default_stripe_count;
1171 lod_adjust_stripe_size(comp, desc->ld_default_stripe_size);
1174 int lod_obj_for_each_stripe(const struct lu_env *env, struct lod_object *lo,
1176 struct lod_obj_stripe_cb_data *data)
1178 struct lod_layout_component *lod_comp;
1182 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1183 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1184 lod_comp = &lo->ldo_comp_entries[i];
1186 if (lod_comp->llc_stripe == NULL)
1189 /* has stripe but not inited yet, this component has been
1190 * declared to be created, but hasn't created yet.
1192 if (!lod_comp_inited(lod_comp))
1195 if (data->locd_comp_skip_cb &&
1196 data->locd_comp_skip_cb(env, lo, i, data))
1199 if (data->locd_comp_cb) {
1200 rc = data->locd_comp_cb(env, lo, i, data);
1205 /* could used just to do sth about component, not each
1208 if (!data->locd_stripe_cb)
1211 LASSERT(lod_comp->llc_stripe_count > 0);
1212 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
1213 struct dt_object *dt = lod_comp->llc_stripe[j];
1217 rc = data->locd_stripe_cb(env, lo, dt, th, i, j, data);
1225 static bool lod_obj_attr_set_comp_skip_cb(const struct lu_env *env,
1226 struct lod_object *lo, int comp_idx,
1227 struct lod_obj_stripe_cb_data *data)
1229 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[comp_idx];
1230 bool skipped = false;
1232 if (!(data->locd_attr->la_valid & LA_LAYOUT_VERSION))
1235 switch (lo->ldo_flr_state) {
1236 case LCM_FL_WRITE_PENDING: {
1239 /* skip stale components */
1240 if (lod_comp->llc_flags & LCME_FL_STALE) {
1245 /* skip valid and overlapping components, therefore any
1246 * attempts to write overlapped components will never succeed
1247 * because client will get EINPROGRESS. */
1248 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1252 if (lo->ldo_comp_entries[i].llc_flags & LCME_FL_STALE)
1255 if (lu_extent_is_overlapped(&lod_comp->llc_extent,
1256 &lo->ldo_comp_entries[i].llc_extent)) {
1264 LASSERTF(0, "impossible: %d\n", lo->ldo_flr_state);
1265 case LCM_FL_SYNC_PENDING:
1269 CDEBUG(D_LAYOUT, DFID": %s to set component %x to version: %u\n",
1270 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
1271 skipped ? "skipped" : "chose", lod_comp->llc_id,
1272 data->locd_attr->la_layout_version);
1278 lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
1279 struct dt_object *dt, struct thandle *th,
1280 int comp_idx, int stripe_idx,
1281 struct lod_obj_stripe_cb_data *data)
1283 if (data->locd_declare)
1284 return lod_sub_declare_attr_set(env, dt, data->locd_attr, th);
1286 if (data->locd_attr->la_valid & LA_LAYOUT_VERSION) {
1287 CDEBUG(D_LAYOUT, DFID": set layout version: %u, comp_idx: %d\n",
1288 PFID(lu_object_fid(&dt->do_lu)),
1289 data->locd_attr->la_layout_version, comp_idx);
1292 return lod_sub_attr_set(env, dt, data->locd_attr, th);
1296 * Implementation of dt_object_operations::do_declare_attr_set.
1298 * If the object is striped, then apply the changes to all the stripes.
1300 * \see dt_object_operations::do_declare_attr_set() in the API description
1303 static int lod_declare_attr_set(const struct lu_env *env,
1304 struct dt_object *dt,
1305 const struct lu_attr *attr,
1308 struct dt_object *next = dt_object_child(dt);
1309 struct lod_object *lo = lod_dt_obj(dt);
1314 * declare setattr on the local object
1316 rc = lod_sub_declare_attr_set(env, next, attr, th);
1320 /* osp_declare_attr_set() ignores all attributes other than
1321 * UID, GID, PROJID, and size, and osp_attr_set() ignores all
1322 * but UID, GID and PROJID. Declaration of size attr setting
1323 * happens through lod_declare_init_size(), and not through
1324 * this function. Therefore we need not load striping unless
1325 * ownership is changing. This should save memory and (we hope)
1326 * speed up rename().
1328 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1329 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1332 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1335 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID | LA_MODE |
1336 LA_ATIME | LA_MTIME | LA_CTIME |
1341 * load striping information, notice we don't do this when object
1342 * is being initialized as we don't need this information till
1343 * few specific cases like destroy, chown
1345 rc = lod_striping_load(env, lo);
1349 if (!lod_obj_is_striped(dt))
1353 * if object is striped declare changes on the stripes
1355 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1356 LASSERT(lo->ldo_stripe);
1357 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1358 if (lo->ldo_stripe[i] == NULL)
1360 if (!dt_object_exists(lo->ldo_stripe[i]))
1362 rc = lod_sub_declare_attr_set(env, lo->ldo_stripe[i],
1368 struct lod_obj_stripe_cb_data data = { { 0 } };
1370 data.locd_attr = attr;
1371 data.locd_declare = true;
1372 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1373 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1379 if (!dt_object_exists(next) || dt_object_remote(next) ||
1380 !S_ISREG(attr->la_mode))
1383 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1384 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
1388 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) ||
1389 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1390 struct lod_thread_info *info = lod_env_info(env);
1391 struct lu_buf *buf = &info->lti_buf;
1393 buf->lb_buf = info->lti_ea_store;
1394 buf->lb_len = info->lti_ea_store_size;
1395 rc = lod_sub_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1396 LU_XATTR_REPLACE, th);
1403 * Implementation of dt_object_operations::do_attr_set.
1405 * If the object is striped, then apply the changes to all or subset of
1406 * the stripes depending on the object type and specific attributes.
1408 * \see dt_object_operations::do_attr_set() in the API description for details.
1410 static int lod_attr_set(const struct lu_env *env,
1411 struct dt_object *dt,
1412 const struct lu_attr *attr,
1415 struct dt_object *next = dt_object_child(dt);
1416 struct lod_object *lo = lod_dt_obj(dt);
1421 * apply changes to the local object
1423 rc = lod_sub_attr_set(env, next, attr, th);
1427 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1428 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1431 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1434 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE | LA_PROJID |
1435 LA_ATIME | LA_MTIME | LA_CTIME |
1440 /* FIXME: a tricky case in the code path of mdd_layout_change():
1441 * the in-memory striping information has been freed in lod_xattr_set()
1442 * due to layout change. It has to load stripe here again. It only
1443 * changes flags of layout so declare_attr_set() is still accurate */
1444 rc = lod_striping_load(env, lo);
1448 if (!lod_obj_is_striped(dt))
1452 * if object is striped, apply changes to all the stripes
1454 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1455 LASSERT(lo->ldo_stripe);
1456 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1457 if (unlikely(lo->ldo_stripe[i] == NULL))
1460 if ((dt_object_exists(lo->ldo_stripe[i]) == 0))
1463 rc = lod_sub_attr_set(env, lo->ldo_stripe[i], attr, th);
1468 struct lod_obj_stripe_cb_data data = { { 0 } };
1470 data.locd_attr = attr;
1471 data.locd_declare = false;
1472 data.locd_comp_skip_cb = lod_obj_attr_set_comp_skip_cb;
1473 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1474 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1480 if (!dt_object_exists(next) || dt_object_remote(next) ||
1481 !S_ISREG(attr->la_mode))
1484 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1485 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
1489 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE)) {
1490 struct lod_thread_info *info = lod_env_info(env);
1491 struct lu_buf *buf = &info->lti_buf;
1492 struct ost_id *oi = &info->lti_ostid;
1493 struct lu_fid *fid = &info->lti_fid;
1494 struct lov_mds_md_v1 *lmm;
1495 struct lov_ost_data_v1 *objs;
1498 rc = lod_get_lov_ea(env, lo);
1502 buf->lb_buf = info->lti_ea_store;
1503 buf->lb_len = info->lti_ea_store_size;
1504 lmm = info->lti_ea_store;
1505 magic = le32_to_cpu(lmm->lmm_magic);
1506 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1507 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1508 struct lov_comp_md_entry_v1 *lcme =
1509 &lcm->lcm_entries[0];
1511 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
1512 magic = le32_to_cpu(lmm->lmm_magic);
1515 if (magic == LOV_MAGIC_V1)
1516 objs = &(lmm->lmm_objects[0]);
1518 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1519 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1520 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1522 fid_to_ostid(fid, oi);
1523 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1525 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1526 LU_XATTR_REPLACE, th);
1527 } else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1528 struct lod_thread_info *info = lod_env_info(env);
1529 struct lu_buf *buf = &info->lti_buf;
1530 struct lov_comp_md_v1 *lcm;
1531 struct lov_comp_md_entry_v1 *lcme;
1533 rc = lod_get_lov_ea(env, lo);
1537 buf->lb_buf = info->lti_ea_store;
1538 buf->lb_len = info->lti_ea_store_size;
1540 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1 &&
1541 le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_SEL)
1544 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1545 lcme = &lcm->lcm_entries[0];
1546 le64_add_cpu(&lcme->lcme_extent.e_start, 1);
1547 le64_add_cpu(&lcme->lcme_extent.e_end, -1);
1549 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1550 LU_XATTR_REPLACE, th);
1557 * Implementation of dt_object_operations::do_xattr_get.
1559 * If LOV EA is requested from the root object and it's not
1560 * found, then return default striping for the filesystem.
1562 * \see dt_object_operations::do_xattr_get() in the API description for details.
1564 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1565 struct lu_buf *buf, const char *name)
1567 struct lod_thread_info *info = lod_env_info(env);
1568 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1573 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1574 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1575 struct lmv_mds_md_v1 *lmv1;
1576 struct lmv_foreign_md *lfm;
1579 if (rc > (typeof(rc))sizeof(*lmv1))
1582 /* short (<= sizeof(struct lmv_mds_md_v1)) foreign LMV case */
1583 /* XXX empty foreign LMV is not allowed */
1584 if (rc <= offsetof(typeof(*lfm), lfm_value))
1585 RETURN(rc = rc > 0 ? -EINVAL : rc);
1587 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1588 BUILD_BUG_ON(sizeof(*lmv1) > sizeof(info->lti_key));
1590 /* lti_buf is large enough for *lmv1 or a short
1591 * (<= sizeof(struct lmv_mds_md_v1)) foreign LMV
1593 info->lti_buf.lb_buf = info->lti_key;
1594 info->lti_buf.lb_len = sizeof(*lmv1);
1595 rc = dt_xattr_get(env, dt_object_child(dt),
1596 &info->lti_buf, name);
1597 if (unlikely(rc <= offsetof(typeof(*lfm),
1599 RETURN(rc = rc > 0 ? -EINVAL : rc);
1601 lfm = info->lti_buf.lb_buf;
1602 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN)
1605 if (unlikely(rc != sizeof(*lmv1)))
1606 RETURN(rc = rc > 0 ? -EINVAL : rc);
1608 lmv1 = info->lti_buf.lb_buf;
1609 /* The on-disk LMV EA only contains header, but the
1610 * returned LMV EA size should contain the space for
1611 * the FIDs of all shards of the striped directory. */
1612 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1613 rc = lmv_mds_md_size(
1614 le32_to_cpu(lmv1->lmv_stripe_count),
1615 le32_to_cpu(lmv1->lmv_magic));
1618 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1621 if (rc != sizeof(*lmv1))
1622 RETURN(rc = rc > 0 ? -EINVAL : rc);
1624 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1628 RETURN(rc = rc1 != 0 ? rc1 : rc);
1631 if ((rc > 0) && buf->lb_buf && strcmp(name, XATTR_NAME_LOV) == 0) {
1632 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1634 if (lcm->lcm_magic == cpu_to_le32(LOV_MAGIC_SEL))
1635 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
1638 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1642 * XXX: Only used by lfsck
1644 * lod returns default striping on the real root of the device
1645 * this is like the root stores default striping for the whole
1646 * filesystem. historically we've been using a different approach
1647 * and store it in the config.
1649 dt_root_get(env, dev->lod_child, &info->lti_fid);
1650 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1652 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1653 struct lov_user_md *lum = buf->lb_buf;
1654 struct lov_desc *desc = &dev->lod_ost_descs.ltd_lov_desc;
1656 if (buf->lb_buf == NULL) {
1658 } else if (buf->lb_len >= sizeof(*lum)) {
1659 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1660 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1661 lmm_oi_set_id(&lum->lmm_oi, 0);
1662 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1663 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1664 lum->lmm_stripe_size = cpu_to_le32(
1665 desc->ld_default_stripe_size);
1666 lum->lmm_stripe_count = cpu_to_le16(
1667 desc->ld_default_stripe_count);
1668 lum->lmm_stripe_offset = cpu_to_le16(
1669 desc->ld_default_stripe_offset);
1682 * Checks that the magic of the stripe is sane.
1684 * \param[in] lod lod device
1685 * \param[in] lum a buffer storing LMV EA to verify
1687 * \retval 0 if the EA is sane
1688 * \retval negative otherwise
1690 static int lod_verify_md_striping(struct lod_device *lod,
1691 const struct lmv_user_md_v1 *lum)
1693 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1694 CERROR("%s: invalid lmv_user_md: magic = %x, "
1695 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1696 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1697 (int)le32_to_cpu(lum->lum_stripe_offset),
1698 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1706 * Initialize LMV EA for a slave.
1708 * Initialize slave's LMV EA from the master's LMV EA.
1710 * \param[in] master_lmv a buffer containing master's EA
1711 * \param[out] slave_lmv a buffer where slave's EA will be stored
1714 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1715 const struct lmv_mds_md_v1 *master_lmv)
1717 *slave_lmv = *master_lmv;
1718 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1724 * Generate LMV EA from the object passed as \a dt. The object must have
1725 * the stripes created and initialized.
1727 * \param[in] env execution environment
1728 * \param[in] dt object
1729 * \param[out] lmv_buf buffer storing generated LMV EA
1731 * \retval 0 on success
1732 * \retval negative if failed
1734 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1735 struct lu_buf *lmv_buf)
1737 struct lod_thread_info *info = lod_env_info(env);
1738 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1739 struct lod_object *lo = lod_dt_obj(dt);
1740 struct lmv_mds_md_v1 *lmm1;
1742 int type = LU_SEQ_RANGE_ANY;
1747 LASSERT(lo->ldo_dir_striped != 0);
1748 LASSERT(lo->ldo_dir_stripe_count > 0);
1749 stripe_count = lo->ldo_dir_stripe_count;
1750 /* Only store the LMV EA heahder on the disk. */
1751 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1752 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1756 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1759 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1760 memset(lmm1, 0, sizeof(*lmm1));
1761 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1762 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1763 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1764 lmm1->lmv_layout_version = cpu_to_le32(lo->ldo_dir_layout_version);
1765 if (lod_is_layout_changing(lo)) {
1766 lmm1->lmv_migrate_hash = cpu_to_le32(lo->ldo_dir_migrate_hash);
1767 lmm1->lmv_migrate_offset =
1768 cpu_to_le32(lo->ldo_dir_migrate_offset);
1770 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1775 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1776 lmv_buf->lb_buf = info->lti_ea_store;
1777 lmv_buf->lb_len = sizeof(*lmm1);
1783 * Create in-core represenation for a striped directory.
1785 * Parse the buffer containing LMV EA and instantiate LU objects
1786 * representing the stripe objects. The pointers to the objects are
1787 * stored in ldo_stripe field of \a lo. This function is used when
1788 * we need to access an already created object (i.e. load from a disk).
1790 * \param[in] env execution environment
1791 * \param[in] lo lod object
1792 * \param[in] buf buffer containing LMV EA
1794 * \retval 0 on success
1795 * \retval negative if failed
1797 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1798 const struct lu_buf *buf)
1800 struct lod_thread_info *info = lod_env_info(env);
1801 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1802 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1803 struct dt_object **stripe;
1804 union lmv_mds_md *lmm = buf->lb_buf;
1805 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1806 struct lu_fid *fid = &info->lti_fid;
1811 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1813 /* XXX may be useless as not called for foreign LMV ?? */
1814 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_FOREIGN)
1817 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1818 lo->ldo_dir_slave_stripe = 1;
1822 if (!lmv_is_sane(lmv1))
1825 LASSERT(lo->ldo_stripe == NULL);
1826 OBD_ALLOC_PTR_ARRAY(stripe, le32_to_cpu(lmv1->lmv_stripe_count));
1830 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1831 struct dt_device *tgt_dt;
1832 struct dt_object *dto;
1833 int type = LU_SEQ_RANGE_ANY;
1836 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1837 if (!fid_is_sane(fid)) {
1842 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1846 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1847 tgt_dt = lod->lod_child;
1849 struct lod_tgt_desc *tgt;
1851 tgt = LTD_TGT(ltd, idx);
1853 GOTO(out, rc = -ESTALE);
1854 tgt_dt = tgt->ltd_tgt;
1857 dto = dt_locate_at(env, tgt_dt, fid,
1858 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1861 GOTO(out, rc = PTR_ERR(dto));
1866 lo->ldo_stripe = stripe;
1867 lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1868 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1869 lo->ldo_dir_layout_version = le32_to_cpu(lmv1->lmv_layout_version);
1870 lo->ldo_dir_migrate_offset = le32_to_cpu(lmv1->lmv_migrate_offset);
1871 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv1->lmv_migrate_hash);
1872 lo->ldo_dir_hash_type = le32_to_cpu(lmv1->lmv_hash_type);
1874 lod_striping_free_nolock(env, lo);
1880 * Declare create a striped directory.
1882 * Declare creating a striped directory with a given stripe pattern on the
1883 * specified MDTs. A striped directory is represented as a regular directory
1884 * - an index listing all the stripes. The stripes point back to the master
1885 * object with ".." and LinkEA. The master object gets LMV EA which
1886 * identifies it as a striped directory. The function allocates FIDs
1889 * \param[in] env execution environment
1890 * \param[in] dt object
1891 * \param[in] attr attributes to initialize the objects with
1892 * \param[in] dof type of objects to be created
1893 * \param[in] th transaction handle
1895 * \retval 0 on success
1896 * \retval negative if failed
1898 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1899 struct dt_object *dt,
1900 struct lu_attr *attr,
1901 struct dt_object_format *dof,
1904 struct lod_thread_info *info = lod_env_info(env);
1905 struct lu_buf lmv_buf;
1906 struct lu_buf slave_lmv_buf;
1907 struct lmv_mds_md_v1 *lmm;
1908 struct lmv_mds_md_v1 *slave_lmm = NULL;
1909 struct dt_insert_rec *rec = &info->lti_dt_rec;
1910 struct lod_object *lo = lod_dt_obj(dt);
1915 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1918 lmm = lmv_buf.lb_buf;
1920 OBD_ALLOC_PTR(slave_lmm);
1921 if (slave_lmm == NULL)
1922 GOTO(out, rc = -ENOMEM);
1924 lod_prep_slave_lmv_md(slave_lmm, lmm);
1925 slave_lmv_buf.lb_buf = slave_lmm;
1926 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1928 if (!dt_try_as_dir(env, dt_object_child(dt)))
1929 GOTO(out, rc = -EINVAL);
1931 rec->rec_type = S_IFDIR;
1932 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1933 struct dt_object *dto = lo->ldo_stripe[i];
1934 char *stripe_name = info->lti_key;
1935 struct lu_name *sname;
1936 struct linkea_data ldata = { NULL };
1937 struct lu_buf linkea_buf;
1939 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
1943 /* directory split skip create for existing stripes */
1944 if (!(lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) {
1945 rc = lod_sub_declare_create(env, dto, attr, NULL, dof,
1950 if (!dt_try_as_dir(env, dto))
1951 GOTO(out, rc = -EINVAL);
1953 rc = lod_sub_declare_ref_add(env, dto, th);
1957 rec->rec_fid = lu_object_fid(&dto->do_lu);
1958 rc = lod_sub_declare_insert(env, dto,
1959 (const struct dt_rec *)rec,
1960 (const struct dt_key *)dot,
1965 /* master stripe FID will be put to .. */
1966 rec->rec_fid = lu_object_fid(&dt->do_lu);
1967 rc = lod_sub_declare_insert(env, dto,
1968 (const struct dt_rec *)rec,
1969 (const struct dt_key *)dotdot,
1974 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1976 snprintf(stripe_name, sizeof(info->lti_key),
1978 PFID(lu_object_fid(&dto->do_lu)),
1981 snprintf(stripe_name, sizeof(info->lti_key),
1983 PFID(lu_object_fid(&dto->do_lu)), i);
1985 sname = lod_name_get(env, stripe_name,
1986 strlen(stripe_name));
1987 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1988 sname, lu_object_fid(&dt->do_lu));
1992 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1993 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1994 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
1995 XATTR_NAME_LINK, 0, th);
1999 rec->rec_fid = lu_object_fid(&dto->do_lu);
2000 rc = lod_sub_declare_insert(env, dt_object_child(dt),
2001 (const struct dt_rec *)rec,
2002 (const struct dt_key *)stripe_name, th);
2006 rc = lod_sub_declare_ref_add(env, dt_object_child(dt),
2012 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
2013 cfs_fail_val != i) {
2014 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
2016 slave_lmm->lmv_master_mdt_index =
2019 slave_lmm->lmv_master_mdt_index =
2021 rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
2022 XATTR_NAME_LMV, 0, th);
2028 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
2029 &lmv_buf, XATTR_NAME_LMV, 0, th);
2033 if (slave_lmm != NULL)
2034 OBD_FREE_PTR(slave_lmm);
2040 * Allocate a striping on a predefined set of MDTs.
2042 * Allocates new striping using the MDT index range provided by the data from
2043 * the lum_obejcts contained in the lmv_user_md passed to this method if
2044 * \a is_specific is true; or allocates new layout starting from MDT index in
2045 * lo->ldo_dir_stripe_offset. The exact order of MDTs is not important and
2046 * varies depending on MDT status. The number of stripes needed and stripe
2047 * offset are taken from the object. If that number cannot be met, then the
2048 * function returns an error and then it's the caller's responsibility to
2049 * release the stripes allocated. All the internal structures are protected,
2050 * but no concurrent allocation is allowed on the same objects.
2052 * \param[in] env execution environment for this thread
2053 * \param[in] lo LOD object
2054 * \param[out] stripes striping created
2055 * \param[out] mdt_indices MDT indices of striping created
2056 * \param[in] is_specific true if the MDTs are provided by lum; false if
2057 * only the starting MDT index is provided
2059 * \retval positive stripes allocated, including the first stripe allocated
2061 * \retval negative errno on failure
2063 static int lod_mdt_alloc_specific(const struct lu_env *env,
2064 struct lod_object *lo,
2065 struct dt_object **stripes,
2066 __u32 *mdt_indices, bool is_specific)
2068 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
2069 struct lu_tgt_descs *ltd = &lod->lod_mdt_descs;
2070 struct lu_tgt_desc *tgt = NULL;
2071 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2072 struct dt_device *tgt_dt = NULL;
2073 struct lu_fid fid = { 0 };
2074 struct dt_object *dto;
2076 u32 stripe_count = lo->ldo_dir_stripe_count;
2082 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
2083 if (!is_specific && stripe_count > 1)
2084 /* Set the start index for the 2nd stripe allocation */
2085 mdt_indices[1] = (mdt_indices[0] + 1) %
2086 (lod->lod_remote_mdt_count + 1);
2088 for (; stripe_idx < stripe_count; stripe_idx++) {
2089 /* Try to find next avaible target */
2090 idx = mdt_indices[stripe_idx];
2091 for (j = 0; j < lod->lod_remote_mdt_count;
2092 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
2093 bool already_allocated = false;
2096 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
2097 idx, lod->lod_remote_mdt_count + 1, stripe_idx);
2099 if (likely(!is_specific &&
2100 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
2101 /* check whether the idx already exists
2102 * in current allocated array */
2103 for (k = 0; k < stripe_idx; k++) {
2104 if (mdt_indices[k] == idx) {
2105 already_allocated = true;
2110 if (already_allocated)
2114 /* Sigh, this index is not in the bitmap, let's check
2115 * next available target */
2116 if (!test_bit(idx, ltd->ltd_tgt_bitmap) &&
2117 idx != master_index)
2120 if (idx == master_index) {
2121 /* Allocate the FID locally */
2122 tgt_dt = lod->lod_child;
2123 rc = dt_fid_alloc(env, tgt_dt, &fid, NULL,
2130 /* check the status of the OSP */
2131 tgt = LTD_TGT(ltd, idx);
2135 tgt_dt = tgt->ltd_tgt;
2136 if (!tgt->ltd_active)
2137 /* this OSP doesn't feel well */
2140 rc = dt_fid_alloc(env, tgt_dt, &fid, NULL, NULL);
2147 /* Can not allocate more stripes */
2148 if (j == lod->lod_remote_mdt_count) {
2149 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
2150 lod2obd(lod)->obd_name, stripe_count,
2155 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
2156 idx, stripe_idx, PFID(&fid));
2157 mdt_indices[stripe_idx] = idx;
2158 /* Set the start index for next stripe allocation */
2159 if (!is_specific && stripe_idx < stripe_count - 1) {
2161 * for large dir test, put all other slaves on one
2162 * remote MDT, otherwise we may save too many local
2163 * slave locks which will exceed RS_MAX_LOCKS.
2165 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)))
2167 mdt_indices[stripe_idx + 1] = (idx + 1) %
2168 (lod->lod_remote_mdt_count + 1);
2170 /* tgt_dt and fid must be ready after search avaible OSP
2171 * in the above loop */
2172 LASSERT(tgt_dt != NULL);
2173 LASSERT(fid_is_sane(&fid));
2175 /* fail a remote stripe FID allocation */
2176 if (stripe_idx && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_FID))
2179 dto = dt_locate_at(env, tgt_dt, &fid,
2180 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
2187 stripes[stripe_idx] = dto;
2193 for (j = 1; j < stripe_idx; j++) {
2194 LASSERT(stripes[j] != NULL);
2195 dt_object_put(env, stripes[j]);
2201 static int lod_prep_md_striped_create(const struct lu_env *env,
2202 struct dt_object *dt,
2203 struct lu_attr *attr,
2204 const struct lmv_user_md_v1 *lum,
2205 struct dt_object_format *dof,
2208 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
2209 struct lod_object *lo = lod_dt_obj(dt);
2210 struct dt_object **stripes;
2211 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2212 struct lu_fid fid = { 0 };
2219 /* The lum has been verifed in lod_verify_md_striping */
2220 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
2221 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
2223 stripe_count = lo->ldo_dir_stripe_count;
2225 OBD_ALLOC_PTR_ARRAY(stripes, stripe_count);
2229 /* Allocate the first stripe locally */
2230 rc = dt_fid_alloc(env, lod->lod_child, &fid, NULL, NULL);
2234 stripes[0] = dt_locate_at(env, lod->lod_child, &fid,
2235 dt->do_lu.lo_dev->ld_site->ls_top_dev, &conf);
2236 if (IS_ERR(stripes[0]))
2237 GOTO(out, rc = PTR_ERR(stripes[0]));
2239 if (lo->ldo_dir_stripe_offset == LMV_OFFSET_DEFAULT) {
2240 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
2241 rc = lod_mdt_alloc_qos(env, lo, stripes, 1, stripe_count);
2243 rc = lod_mdt_alloc_rr(env, lo, stripes, 1,
2247 bool is_specific = false;
2249 OBD_ALLOC_PTR_ARRAY(idx_array, stripe_count);
2251 GOTO(out, rc = -ENOMEM);
2253 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
2255 for (i = 0; i < stripe_count; i++)
2257 le32_to_cpu(lum->lum_objects[i].lum_mds);
2260 /* stripe 0 is local */
2262 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
2263 rc = lod_mdt_alloc_specific(env, lo, stripes, idx_array,
2265 OBD_FREE_PTR_ARRAY(idx_array, stripe_count);
2273 lo->ldo_dir_striped = 1;
2274 lo->ldo_stripe = stripes;
2275 lo->ldo_dir_stripe_count = rc;
2276 lo->ldo_dir_stripes_allocated = stripe_count;
2278 lo->ldo_dir_stripe_loaded = 1;
2280 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2282 lod_striping_free(env, lo);
2288 if (!IS_ERR_OR_NULL(stripes[0]))
2289 dt_object_put(env, stripes[0]);
2290 for (i = 1; i < stripe_count; i++)
2291 LASSERT(!stripes[i]);
2292 OBD_FREE_PTR_ARRAY(stripes, stripe_count);
2299 * Alloc cached foreign LMV
2301 * \param[in] lo object
2302 * \param[in] size size of foreign LMV
2304 * \retval 0 on success
2305 * \retval negative if failed
2307 int lod_alloc_foreign_lmv(struct lod_object *lo, size_t size)
2309 OBD_ALLOC_LARGE(lo->ldo_foreign_lmv, size);
2310 if (lo->ldo_foreign_lmv == NULL)
2312 lo->ldo_foreign_lmv_size = size;
2313 lo->ldo_dir_is_foreign = 1;
2319 * Declare create striped md object.
2321 * The function declares intention to create a striped directory. This is a
2322 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2323 * is to verify pattern \a lum_buf is good. Check that function for the details.
2325 * \param[in] env execution environment
2326 * \param[in] dt object
2327 * \param[in] attr attributes to initialize the objects with
2328 * \param[in] lum_buf a pattern specifying the number of stripes and
2330 * \param[in] dof type of objects to be created
2331 * \param[in] th transaction handle
2333 * \retval 0 on success
2334 * \retval negative if failed
2337 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2338 struct dt_object *dt,
2339 struct lu_attr *attr,
2340 const struct lu_buf *lum_buf,
2341 struct dt_object_format *dof,
2344 struct lod_object *lo = lod_dt_obj(dt);
2345 struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
2349 LASSERT(lum != NULL);
2351 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
2352 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
2353 (int)le32_to_cpu(lum->lum_stripe_offset));
2355 if (lo->ldo_dir_stripe_count == 0) {
2356 if (lo->ldo_dir_is_foreign) {
2357 rc = lod_alloc_foreign_lmv(lo, lum_buf->lb_len);
2360 memcpy(lo->ldo_foreign_lmv, lum, lum_buf->lb_len);
2361 lo->ldo_dir_stripe_loaded = 1;
2366 /* prepare dir striped objects */
2367 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2369 /* failed to create striping, let's reset
2370 * config so that others don't get confused */
2371 lod_striping_free(env, lo);
2379 * Set or replace striped directory layout, and LFSCK may set layout on a plain
2380 * directory, so don't check stripe count.
2382 * \param[in] env execution environment
2383 * \param[in] dt target object
2384 * \param[in] buf LMV buf which contains source stripe fids
2385 * \param[in] fl set or replace
2386 * \param[in] th transaction handle
2388 * \retval 0 on success
2389 * \retval negative if failed
2391 static int lod_dir_layout_set(const struct lu_env *env,
2392 struct dt_object *dt,
2393 const struct lu_buf *buf,
2397 struct dt_object *next = dt_object_child(dt);
2398 struct lod_object *lo = lod_dt_obj(dt);
2399 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2400 struct lmv_mds_md_v1 *lmv = buf->lb_buf;
2401 struct lmv_mds_md_v1 *slave_lmv;
2402 struct lu_buf slave_buf;
2408 if (!lmv_is_sane2(lmv))
2411 /* adjust hash for dir merge, which may not be set in user command */
2412 if (lmv_is_merging(lmv) && !lmv->lmv_migrate_hash)
2413 lmv->lmv_merge_hash =
2414 lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
2416 LMV_DEBUG(D_INFO, lmv, "set");
2418 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LMV, fl, th);
2422 /* directory restripe may update stripe LMV directly */
2423 if (!lo->ldo_dir_stripe_count)
2426 lo->ldo_dir_hash_type = le32_to_cpu(lmv->lmv_hash_type);
2427 lo->ldo_dir_migrate_offset = le32_to_cpu(lmv->lmv_migrate_offset);
2428 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_migrate_hash);
2429 lo->ldo_dir_layout_version = le32_to_cpu(lmv->lmv_layout_version);
2431 OBD_ALLOC_PTR(slave_lmv);
2435 lod_prep_slave_lmv_md(slave_lmv, lmv);
2436 slave_buf.lb_buf = slave_lmv;
2437 slave_buf.lb_len = sizeof(*slave_lmv);
2439 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2440 if (!lo->ldo_stripe[i])
2443 if (!dt_object_exists(lo->ldo_stripe[i]))
2446 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], &slave_buf,
2447 XATTR_NAME_LMV, fl, th);
2452 OBD_FREE_PTR(slave_lmv);
2458 * Implementation of dt_object_operations::do_declare_xattr_set.
2460 * Used with regular (non-striped) objects. Basically it
2461 * initializes the striping information and applies the
2462 * change to all the stripes.
2464 * \see dt_object_operations::do_declare_xattr_set() in the API description
2467 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2468 struct dt_object *dt,
2469 const struct lu_buf *buf,
2470 const char *name, int fl,
2473 struct dt_object *next = dt_object_child(dt);
2474 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2475 struct lod_object *lo = lod_dt_obj(dt);
2480 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2481 struct lmv_user_md_v1 *lum;
2483 LASSERT(buf != NULL && buf->lb_buf != NULL);
2485 rc = lod_verify_md_striping(d, lum);
2488 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2489 rc = lod_verify_striping(env, d, lo, buf, false);
2494 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2498 /* Note: Do not set LinkEA on sub-stripes, otherwise
2499 * it will confuse the fid2path process(see mdt_path_current()).
2500 * The linkEA between master and sub-stripes is set in
2501 * lod_xattr_set_lmv(). */
2502 if (strcmp(name, XATTR_NAME_LINK) == 0)
2505 /* set xattr to each stripes, if needed */
2506 rc = lod_striping_load(env, lo);
2510 if (lo->ldo_dir_stripe_count == 0)
2513 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2514 if (!lo->ldo_stripe[i])
2517 if (!dt_object_exists(lo->ldo_stripe[i]))
2520 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2530 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2531 struct lod_object *lo,
2532 struct dt_object *dt, struct thandle *th,
2533 int comp_idx, int stripe_idx,
2534 struct lod_obj_stripe_cb_data *data)
2536 struct lod_thread_info *info = lod_env_info(env);
2537 struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
2538 struct filter_fid *ff = &info->lti_ff;
2539 struct lu_buf *buf = &info->lti_buf;
2543 buf->lb_len = sizeof(*ff);
2544 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2552 * locd_buf is set if it's called by dir migration, which doesn't check
2555 if (data->locd_buf) {
2556 memset(ff, 0, sizeof(*ff));
2557 ff->ff_parent = *(struct lu_fid *)data->locd_buf->lb_buf;
2559 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
2561 if (lu_fid_eq(lod_object_fid(lo), &ff->ff_parent) &&
2562 ff->ff_layout.ol_comp_id == comp->llc_id)
2565 memset(ff, 0, sizeof(*ff));
2566 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2569 /* rewrite filter_fid */
2570 ff->ff_parent.f_ver = stripe_idx;
2571 ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
2572 ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
2573 ff->ff_layout.ol_comp_id = comp->llc_id;
2574 ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
2575 ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
2576 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
2578 if (data->locd_declare)
2579 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2580 LU_XATTR_REPLACE, th);
2582 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2583 LU_XATTR_REPLACE, th);
2589 * Reset parent FID on OST object
2591 * Replace parent FID with @dt object FID, which is only called during migration
2592 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2593 * the FID is changed.
2595 * \param[in] env execution environment
2596 * \param[in] dt dt_object whose stripes's parent FID will be reset
2597 * \parem[in] th thandle
2598 * \param[in] declare if it is declare
2600 * \retval 0 if reset succeeds
2601 * \retval negative errno if reset fails
2603 static int lod_replace_parent_fid(const struct lu_env *env,
2604 struct dt_object *dt,
2605 const struct lu_buf *buf,
2606 struct thandle *th, bool declare)
2608 struct lod_object *lo = lod_dt_obj(dt);
2609 struct lod_thread_info *info = lod_env_info(env);
2610 struct filter_fid *ff;
2611 struct lod_obj_stripe_cb_data data = { { 0 } };
2615 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2617 /* set xattr to each stripes, if needed */
2618 rc = lod_striping_load(env, lo);
2622 if (!lod_obj_is_striped(dt))
2625 if (info->lti_ea_store_size < sizeof(*ff)) {
2626 rc = lod_ea_store_resize(info, sizeof(*ff));
2631 data.locd_declare = declare;
2632 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2633 data.locd_buf = buf;
2634 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2639 __u16 lod_comp_entry_stripe_count(struct lod_object *lo,
2640 int comp_idx, bool is_dir)
2642 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2643 struct lod_layout_component *entry;
2648 entry = &lo->ldo_comp_entries[comp_idx];
2649 if (lod_comp_inited(entry))
2650 return entry->llc_stripe_count;
2651 else if ((__u16)-1 == entry->llc_stripe_count)
2652 return lod->lod_ost_count;
2654 return lod_get_stripe_count(lod, lo, comp_idx,
2655 entry->llc_stripe_count, false);
2658 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2660 int magic, size = 0, i;
2661 struct lod_layout_component *comp_entries;
2663 bool is_composite, is_foreign = false;
2666 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2667 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2669 lo->ldo_def_striping->lds_def_striping_is_composite;
2671 comp_cnt = lo->ldo_comp_cnt;
2672 comp_entries = lo->ldo_comp_entries;
2673 is_composite = lo->ldo_is_composite;
2674 is_foreign = lo->ldo_is_foreign;
2678 return lo->ldo_foreign_lov_size;
2680 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2682 size = sizeof(struct lov_comp_md_v1) +
2683 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2684 LASSERT(size % sizeof(__u64) == 0);
2687 for (i = 0; i < comp_cnt; i++) {
2690 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2691 stripe_count = lod_comp_entry_stripe_count(lo, i, is_dir);
2692 if (!is_dir && is_composite)
2693 lod_comp_shrink_stripe_count(&comp_entries[i],
2696 size += lov_user_md_size(stripe_count, magic);
2697 LASSERT(size % sizeof(__u64) == 0);
2703 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2704 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2707 * \param[in] env execution environment
2708 * \param[in] dt dt_object to add components on
2709 * \param[in] buf buffer contains components to be added
2710 * \parem[in] th thandle
2712 * \retval 0 on success
2713 * \retval negative errno on failure
2715 static int lod_declare_layout_add(const struct lu_env *env,
2716 struct dt_object *dt,
2717 const struct lu_buf *buf,
2720 struct lod_thread_info *info = lod_env_info(env);
2721 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2722 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2723 struct dt_object *next = dt_object_child(dt);
2724 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
2725 struct lod_object *lo = lod_dt_obj(dt);
2726 struct lov_user_md_v3 *v3;
2727 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2729 int i, rc, array_cnt, old_array_cnt;
2732 LASSERT(lo->ldo_is_composite);
2734 if (lo->ldo_flr_state != LCM_FL_NONE)
2737 rc = lod_verify_striping(env, d, lo, buf, false);
2741 magic = comp_v1->lcm_magic;
2742 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2743 lustre_swab_lov_comp_md_v1(comp_v1);
2744 magic = comp_v1->lcm_magic;
2747 if (magic != LOV_USER_MAGIC_COMP_V1)
2750 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2751 OBD_ALLOC_PTR_ARRAY(comp_array, array_cnt);
2752 if (comp_array == NULL)
2755 memcpy(comp_array, lo->ldo_comp_entries,
2756 sizeof(*comp_array) * lo->ldo_comp_cnt);
2758 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2759 struct lov_user_md_v1 *v1;
2760 struct lu_extent *ext;
2762 v1 = (struct lov_user_md *)((char *)comp_v1 +
2763 comp_v1->lcm_entries[i].lcme_offset);
2764 ext = &comp_v1->lcm_entries[i].lcme_extent;
2766 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2767 lod_comp->llc_extent.e_start = ext->e_start;
2768 lod_comp->llc_extent.e_end = ext->e_end;
2769 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2770 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2772 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2773 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2774 lod_adjust_stripe_info(lod_comp, desc, 0);
2776 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2777 v3 = (struct lov_user_md_v3 *) v1;
2778 if (v3->lmm_pool_name[0] != '\0') {
2779 rc = lod_set_pool(&lod_comp->llc_pool,
2787 old_array = lo->ldo_comp_entries;
2788 old_array_cnt = lo->ldo_comp_cnt;
2790 lo->ldo_comp_entries = comp_array;
2791 lo->ldo_comp_cnt = array_cnt;
2793 /* No need to increase layout generation here, it will be increased
2794 * later when generating component ID for the new components */
2796 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2797 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2798 XATTR_NAME_LOV, 0, th);
2800 lo->ldo_comp_entries = old_array;
2801 lo->ldo_comp_cnt = old_array_cnt;
2805 OBD_FREE_PTR_ARRAY(old_array, old_array_cnt);
2807 LASSERT(lo->ldo_mirror_count == 1);
2808 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2813 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2814 lod_comp = &comp_array[i];
2815 if (lod_comp->llc_pool != NULL) {
2816 OBD_FREE(lod_comp->llc_pool,
2817 strlen(lod_comp->llc_pool) + 1);
2818 lod_comp->llc_pool = NULL;
2821 OBD_FREE_PTR_ARRAY(comp_array, array_cnt);
2826 * lod_last_non_stale_mirror() - Check if a mirror is the last non-stale mirror.
2827 * @mirror_id: Mirror id to be checked.
2830 * This function checks if a mirror with specified @mirror_id is the last
2831 * non-stale mirror of a LOD object @lo.
2833 * Return: true or false.
2836 bool lod_last_non_stale_mirror(__u16 mirror_id, struct lod_object *lo)
2838 struct lod_layout_component *lod_comp;
2839 bool has_stale_flag;
2842 for (i = 0; i < lo->ldo_mirror_count; i++) {
2843 if (lo->ldo_mirrors[i].lme_id == mirror_id ||
2844 lo->ldo_mirrors[i].lme_stale)
2847 has_stale_flag = false;
2848 lod_foreach_mirror_comp(lod_comp, lo, i) {
2849 if (lod_comp->llc_flags & LCME_FL_STALE) {
2850 has_stale_flag = true;
2854 if (!has_stale_flag)
2862 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2863 * the '$field' can only be 'flags' now. The xattr value is binary
2864 * lov_comp_md_v1 which contains the component ID(s) and the value of
2865 * the field to be modified.
2866 * Please update allowed_lustre_lov macro if $field groks more values
2869 * \param[in] env execution environment
2870 * \param[in] dt dt_object to be modified
2871 * \param[in] op operation string, like "set.flags"
2872 * \param[in] buf buffer contains components to be set
2873 * \parem[in] th thandle
2875 * \retval 0 on success
2876 * \retval negative errno on failure
2878 static int lod_declare_layout_set(const struct lu_env *env,
2879 struct dt_object *dt,
2880 char *op, const struct lu_buf *buf,
2883 struct lod_layout_component *lod_comp;
2884 struct lod_thread_info *info = lod_env_info(env);
2885 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2886 struct lod_object *lo = lod_dt_obj(dt);
2887 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2890 bool changed = false;
2893 /* Please update allowed_lustre_lov macro if op
2894 * groks more values in the future
2896 if (strcmp(op, "set.flags") != 0) {
2897 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2898 lod2obd(d)->obd_name, op);
2902 magic = comp_v1->lcm_magic;
2903 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2904 lustre_swab_lov_comp_md_v1(comp_v1);
2905 magic = comp_v1->lcm_magic;
2908 if (magic != LOV_USER_MAGIC_COMP_V1)
2911 if (comp_v1->lcm_entry_count == 0) {
2912 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2913 lod2obd(d)->obd_name);
2917 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2918 __u32 id = comp_v1->lcm_entries[i].lcme_id;
2919 __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
2920 __u32 mirror_flag = flags & LCME_MIRROR_FLAGS;
2921 __u16 mirror_id = mirror_id_of(id);
2922 bool neg = flags & LCME_FL_NEG;
2924 if (flags & LCME_FL_INIT) {
2926 lod_striping_free(env, lo);
2930 flags &= ~(LCME_MIRROR_FLAGS | LCME_FL_NEG);
2931 for (j = 0; j < lo->ldo_comp_cnt; j++) {
2932 lod_comp = &lo->ldo_comp_entries[j];
2934 /* lfs only put one flag in each entry */
2935 if ((flags && id != lod_comp->llc_id) ||
2936 (mirror_flag && mirror_id !=
2937 mirror_id_of(lod_comp->llc_id)))
2942 lod_comp->llc_flags &= ~flags;
2944 lod_comp->llc_flags &= ~mirror_flag;
2947 if ((flags & LCME_FL_STALE) &&
2948 lod_last_non_stale_mirror(mirror_id,
2951 lod_comp->llc_flags |= flags;
2954 lod_comp->llc_flags |= mirror_flag;
2955 if (mirror_flag & LCME_FL_NOSYNC)
2956 lod_comp->llc_timestamp =
2957 ktime_get_real_seconds();
2965 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
2966 lod2obd(d)->obd_name);
2970 lod_obj_inc_layout_gen(lo);
2972 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2973 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
2974 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
2979 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
2980 * and the xattr value is a unique component ID or a special lcme_id.
2982 * \param[in] env execution environment
2983 * \param[in] dt dt_object to be operated on
2984 * \param[in] buf buffer contains component ID or lcme_id
2985 * \parem[in] th thandle
2987 * \retval 0 on success
2988 * \retval negative errno on failure
2990 static int lod_declare_layout_del(const struct lu_env *env,
2991 struct dt_object *dt,
2992 const struct lu_buf *buf,
2995 struct lod_thread_info *info = lod_env_info(env);
2996 struct dt_object *next = dt_object_child(dt);
2997 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2998 struct lod_object *lo = lod_dt_obj(dt);
2999 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3000 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3001 __u32 magic, id, flags, neg_flags = 0;
3005 LASSERT(lo->ldo_is_composite);
3007 if (lo->ldo_flr_state != LCM_FL_NONE)
3010 magic = comp_v1->lcm_magic;
3011 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
3012 lustre_swab_lov_comp_md_v1(comp_v1);
3013 magic = comp_v1->lcm_magic;
3016 if (magic != LOV_USER_MAGIC_COMP_V1)
3019 id = comp_v1->lcm_entries[0].lcme_id;
3020 flags = comp_v1->lcm_entries[0].lcme_flags;
3022 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
3023 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
3024 lod2obd(d)->obd_name, id, flags);
3028 if (id != LCME_ID_INVAL && flags != 0) {
3029 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
3030 lod2obd(d)->obd_name);
3034 if (id == LCME_ID_INVAL && !flags) {
3035 CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
3036 lod2obd(d)->obd_name);
3040 if (flags & LCME_FL_NEG) {
3041 neg_flags = flags & ~LCME_FL_NEG;
3045 left = lo->ldo_comp_cnt;
3049 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3050 struct lod_layout_component *lod_comp;
3052 lod_comp = &lo->ldo_comp_entries[i];
3054 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
3056 else if (flags && !(flags & lod_comp->llc_flags))
3058 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
3061 if (left != (i + 1)) {
3062 CDEBUG(D_LAYOUT, "%s: this deletion will create "
3063 "a hole.\n", lod2obd(d)->obd_name);
3068 /* Mark the component as deleted */
3069 lod_comp->llc_id = LCME_ID_INVAL;
3071 /* Not instantiated component */
3072 if (lod_comp->llc_stripe == NULL)
3075 LASSERT(lod_comp->llc_stripe_count > 0);
3076 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
3077 struct dt_object *obj = lod_comp->llc_stripe[j];
3081 rc = lod_sub_declare_destroy(env, obj, th);
3087 LASSERTF(left >= 0, "left = %d\n", left);
3088 if (left == lo->ldo_comp_cnt) {
3089 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
3090 lod2obd(d)->obd_name, id);
3094 memset(attr, 0, sizeof(*attr));
3095 attr->la_valid = LA_SIZE;
3096 rc = lod_sub_declare_attr_set(env, next, attr, th);
3101 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3102 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
3103 XATTR_NAME_LOV, 0, th);
3105 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
3112 * Declare layout add/set/del operations issued by special xattr names:
3114 * XATTR_LUSTRE_LOV.add add component(s) to existing file
3115 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
3116 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
3118 * \param[in] env execution environment
3119 * \param[in] dt object
3120 * \param[in] name name of xattr
3121 * \param[in] buf lu_buf contains xattr value
3122 * \param[in] th transaction handle
3124 * \retval 0 on success
3125 * \retval negative if failed
3127 static int lod_declare_modify_layout(const struct lu_env *env,
3128 struct dt_object *dt,
3130 const struct lu_buf *buf,
3133 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3134 struct lod_object *lo = lod_dt_obj(dt);
3136 int rc, len = strlen(XATTR_LUSTRE_LOV);
3139 LASSERT(dt_object_exists(dt));
3141 if (strlen(name) <= len || name[len] != '.') {
3142 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
3143 lod2obd(d)->obd_name, name);
3148 rc = lod_striping_load(env, lo);
3152 /* the layout to be modified must be a composite layout */
3153 if (!lo->ldo_is_composite) {
3154 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
3155 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3156 GOTO(unlock, rc = -EINVAL);
3159 op = (char *)name + len;
3160 if (strcmp(op, "add") == 0) {
3161 rc = lod_declare_layout_add(env, dt, buf, th);
3162 } else if (strcmp(op, "del") == 0) {
3163 rc = lod_declare_layout_del(env, dt, buf, th);
3164 } else if (strncmp(op, "set", strlen("set")) == 0) {
3165 rc = lod_declare_layout_set(env, dt, op, buf, th);
3167 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
3168 lod2obd(d)->obd_name, name);
3169 GOTO(unlock, rc = -ENOTSUPP);
3173 lod_striping_free(env, lo);
3179 * Convert a plain file lov_mds_md to a composite layout.
3181 * \param[in,out] info the thread info::lti_ea_store buffer contains little
3182 * endian plain file layout
3184 * \retval 0 on success, <0 on failure
3186 static int lod_layout_convert(struct lod_thread_info *info)
3188 struct lov_mds_md *lmm = info->lti_ea_store;
3189 struct lov_mds_md *lmm_save;
3190 struct lov_comp_md_v1 *lcm;
3191 struct lov_comp_md_entry_v1 *lcme;
3197 /* realloc buffer to a composite layout which contains one component */
3198 blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3199 le32_to_cpu(lmm->lmm_magic));
3200 size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
3202 OBD_ALLOC_LARGE(lmm_save, blob_size);
3204 GOTO(out, rc = -ENOMEM);
3206 memcpy(lmm_save, lmm, blob_size);
3208 if (info->lti_ea_store_size < size) {
3209 rc = lod_ea_store_resize(info, size);
3214 lcm = info->lti_ea_store;
3215 memset(lcm, 0, sizeof(*lcm) + sizeof(*lcme));
3216 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
3217 lcm->lcm_size = cpu_to_le32(size);
3218 lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
3219 lmm_save->lmm_layout_gen));
3220 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
3221 lcm->lcm_entry_count = cpu_to_le16(1);
3223 lcme = &lcm->lcm_entries[0];
3224 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
3225 lcme->lcme_extent.e_start = 0;
3226 lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
3227 lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
3228 lcme->lcme_size = cpu_to_le32(blob_size);
3230 memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
3235 OBD_FREE_LARGE(lmm_save, blob_size);
3240 * Merge layouts to form a mirrored file.
3242 static int lod_declare_layout_merge(const struct lu_env *env,
3243 struct dt_object *dt, const struct lu_buf *mbuf,
3246 struct lod_thread_info *info = lod_env_info(env);
3247 struct lu_buf *buf = &info->lti_buf;
3248 struct lod_object *lo = lod_dt_obj(dt);
3249 struct lov_comp_md_v1 *lcm;
3250 struct lov_comp_md_v1 *cur_lcm;
3251 struct lov_comp_md_v1 *merge_lcm;
3252 struct lov_comp_md_entry_v1 *lcme;
3253 struct lov_mds_md_v1 *lmm;
3256 __u16 cur_entry_count;
3257 __u16 merge_entry_count;
3259 __u16 mirror_id = 0;
3266 merge_lcm = mbuf->lb_buf;
3267 if (mbuf->lb_len < sizeof(*merge_lcm))
3270 /* must be an existing layout from disk */
3271 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
3274 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
3276 /* do not allow to merge two mirrored files */
3277 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
3280 /* verify the target buffer */
3281 rc = lod_get_lov_ea(env, lo);
3283 RETURN(rc ? : -ENODATA);
3285 cur_lcm = info->lti_ea_store;
3286 switch (le32_to_cpu(cur_lcm->lcm_magic)) {
3289 rc = lod_layout_convert(info);
3291 case LOV_MAGIC_COMP_V1:
3301 /* info->lti_ea_store could be reallocated in lod_layout_convert() */
3302 cur_lcm = info->lti_ea_store;
3303 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
3305 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
3306 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
3307 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
3310 /* size of new layout */
3311 size = le32_to_cpu(cur_lcm->lcm_size) +
3312 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
3314 memset(buf, 0, sizeof(*buf));
3315 lu_buf_alloc(buf, size);
3316 if (buf->lb_buf == NULL)
3320 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
3322 offset = sizeof(*lcm) +
3323 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
3324 for (i = 0; i < cur_entry_count; i++) {
3325 struct lov_comp_md_entry_v1 *cur_lcme;
3327 lcme = &lcm->lcm_entries[i];
3328 cur_lcme = &cur_lcm->lcm_entries[i];
3330 lcme->lcme_offset = cpu_to_le32(offset);
3331 memcpy((char *)lcm + offset,
3332 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
3333 le32_to_cpu(lcme->lcme_size));
3335 offset += le32_to_cpu(lcme->lcme_size);
3337 if (mirror_count == 1 &&
3338 mirror_id_of(le32_to_cpu(lcme->lcme_id)) == 0) {
3339 /* Add mirror from a non-flr file, create new mirror ID.
3340 * Otherwise, keep existing mirror's component ID, used
3341 * for mirror extension.
3343 id = pflr_id(1, i + 1);
3344 lcme->lcme_id = cpu_to_le32(id);
3347 id = max(le32_to_cpu(lcme->lcme_id), id);
3350 mirror_id = mirror_id_of(id) + 1;
3352 /* check if first entry in new layout is DOM */
3353 lmm = (struct lov_mds_md_v1 *)((char *)merge_lcm +
3354 merge_lcm->lcm_entries[0].lcme_offset);
3355 merge_has_dom = lov_pattern(le32_to_cpu(lmm->lmm_pattern)) ==
3358 for (i = 0; i < merge_entry_count; i++) {
3359 struct lov_comp_md_entry_v1 *merge_lcme;
3361 merge_lcme = &merge_lcm->lcm_entries[i];
3362 lcme = &lcm->lcm_entries[cur_entry_count + i];
3364 *lcme = *merge_lcme;
3365 lcme->lcme_offset = cpu_to_le32(offset);
3366 if (merge_has_dom && i == 0)
3367 lcme->lcme_flags |= cpu_to_le32(LCME_FL_STALE);
3369 id = pflr_id(mirror_id, i + 1);
3370 lcme->lcme_id = cpu_to_le32(id);
3372 memcpy((char *)lcm + offset,
3373 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
3374 le32_to_cpu(lcme->lcme_size));
3376 offset += le32_to_cpu(lcme->lcme_size);
3379 /* fixup layout information */
3380 lod_obj_inc_layout_gen(lo);
3381 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3382 lcm->lcm_size = cpu_to_le32(size);
3383 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
3384 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
3385 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NONE)
3386 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
3388 rc = lod_striping_reload(env, lo, buf);
3392 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
3393 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3401 * Split layouts, just set the LOVEA with the layout from mbuf.
3403 static int lod_declare_layout_split(const struct lu_env *env,
3404 struct dt_object *dt, const struct lu_buf *mbuf,
3407 struct lod_object *lo = lod_dt_obj(dt);
3408 struct lov_comp_md_v1 *lcm = mbuf->lb_buf;
3412 rc = lod_striping_reload(env, lo, mbuf);
3416 lod_obj_inc_layout_gen(lo);
3417 /* fix on-disk layout gen */
3418 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3420 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), mbuf,
3421 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3426 * Implementation of dt_object_operations::do_declare_xattr_set.
3428 * \see dt_object_operations::do_declare_xattr_set() in the API description
3431 * the extension to the API:
3432 * - declaring LOVEA requests striping creation
3433 * - LU_XATTR_REPLACE means layout swap
3435 static int lod_declare_xattr_set(const struct lu_env *env,
3436 struct dt_object *dt,
3437 const struct lu_buf *buf,
3438 const char *name, int fl,
3441 struct dt_object *next = dt_object_child(dt);
3442 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3447 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
3448 if ((S_ISREG(mode) || mode == 0) &&
3449 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE | LU_XATTR_SPLIT)) &&
3450 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3451 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
3453 * this is a request to create object's striping.
3455 * allow to declare predefined striping on a new (!mode) object
3456 * which is supposed to be replay of regular file creation
3457 * (when LOV setting is declared)
3459 * LU_XATTR_REPLACE is set to indicate a layout swap
3461 if (dt_object_exists(dt)) {
3462 rc = dt_attr_get(env, next, attr);
3466 memset(attr, 0, sizeof(*attr));
3467 attr->la_valid = LA_TYPE | LA_MODE;
3468 attr->la_mode = S_IFREG;
3470 rc = lod_declare_striped_create(env, dt, attr, buf, th);
3471 } else if (fl & LU_XATTR_MERGE) {
3472 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3473 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3474 rc = lod_declare_layout_merge(env, dt, buf, th);
3475 } else if (fl & LU_XATTR_SPLIT) {
3476 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3477 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3478 rc = lod_declare_layout_split(env, dt, buf, th);
3479 } else if (S_ISREG(mode) &&
3480 strlen(name) >= sizeof(XATTR_LUSTRE_LOV) + 3 &&
3481 allowed_lustre_lov(name)) {
3483 * this is a request to modify object's striping.
3484 * add/set/del component(s).
3486 if (!dt_object_exists(dt))
3489 rc = lod_declare_modify_layout(env, dt, name, buf, th);
3490 } else if (S_ISDIR(mode)) {
3491 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
3492 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3493 rc = lod_replace_parent_fid(env, dt, buf, th, true);
3495 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
3502 * Apply xattr changes to the object.
3504 * Applies xattr changes to the object and the stripes if the latter exist.
3506 * \param[in] env execution environment
3507 * \param[in] dt object
3508 * \param[in] buf buffer pointing to the new value of xattr
3509 * \param[in] name name of xattr
3510 * \param[in] fl flags
3511 * \param[in] th transaction handle
3513 * \retval 0 on success
3514 * \retval negative if failed
3516 static int lod_xattr_set_internal(const struct lu_env *env,
3517 struct dt_object *dt,
3518 const struct lu_buf *buf,
3519 const char *name, int fl,
3522 struct dt_object *next = dt_object_child(dt);
3523 struct lod_object *lo = lod_dt_obj(dt);
3528 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3529 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3532 /* Note: Do not set LinkEA on sub-stripes, otherwise
3533 * it will confuse the fid2path process(see mdt_path_current()).
3534 * The linkEA between master and sub-stripes is set in
3535 * lod_xattr_set_lmv(). */
3536 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
3539 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3540 if (!lo->ldo_stripe[i])
3543 if (!dt_object_exists(lo->ldo_stripe[i]))
3546 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
3556 * Delete an extended attribute.
3558 * Deletes specified xattr from the object and the stripes if the latter exist.
3560 * \param[in] env execution environment
3561 * \param[in] dt object
3562 * \param[in] name name of xattr
3563 * \param[in] th transaction handle
3565 * \retval 0 on success
3566 * \retval negative if failed
3568 static int lod_xattr_del_internal(const struct lu_env *env,
3569 struct dt_object *dt,
3570 const char *name, struct thandle *th)
3572 struct dt_object *next = dt_object_child(dt);
3573 struct lod_object *lo = lod_dt_obj(dt);
3579 rc = lod_sub_xattr_del(env, next, name, th);
3580 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3583 if (lo->ldo_dir_stripe_count == 0)
3586 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3587 if (!lo->ldo_stripe[i])
3590 if (!dt_object_exists(lo->ldo_stripe[i]))
3593 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3602 * Set default striping on a directory.
3604 * Sets specified striping on a directory object unless it matches the default
3605 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3606 * EA. This striping will be used when regular file is being created in this
3609 * \param[in] env execution environment
3610 * \param[in] dt the striped object
3611 * \param[in] buf buffer with the striping
3612 * \param[in] name name of EA
3613 * \param[in] fl xattr flag (see OSD API description)
3614 * \param[in] th transaction handle
3616 * \retval 0 on success
3617 * \retval negative if failed
3619 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
3620 struct dt_object *dt,
3621 const struct lu_buf *buf,
3622 const char *name, int fl,
3625 struct lov_user_md_v1 *lum;
3626 struct lov_user_md_v3 *v3 = NULL;
3627 const char *pool_name = NULL;
3632 LASSERT(buf != NULL && buf->lb_buf != NULL);
3635 switch (lum->lmm_magic) {
3636 case LOV_USER_MAGIC_SPECIFIC:
3637 case LOV_USER_MAGIC_V3:
3639 if (v3->lmm_pool_name[0] != '\0')
3640 pool_name = v3->lmm_pool_name;
3642 case LOV_USER_MAGIC_V1:
3643 /* if { size, offset, count } = { 0, -1, 0 } and no pool
3644 * (i.e. all default values specified) then delete default
3645 * striping from dir. */
3647 "set default striping: sz %u # %u offset %d %s %s\n",
3648 (unsigned)lum->lmm_stripe_size,
3649 (unsigned)lum->lmm_stripe_count,
3650 (int)lum->lmm_stripe_offset,
3651 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
3653 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
3654 lum->lmm_stripe_count,
3655 lum->lmm_stripe_offset,
3658 case LOV_USER_MAGIC_COMP_V1:
3660 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lum;
3661 struct lov_comp_md_entry_v1 *lcme;
3664 comp_cnt = le16_to_cpu(lcm->lcm_entry_count);
3665 for (i = 0; i < comp_cnt; i++) {
3666 lcme = &lcm->lcm_entries[i];
3667 if (lcme->lcme_flags & cpu_to_le32(LCME_FL_EXTENSION)) {
3668 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
3677 CERROR("Invalid magic %x\n", lum->lmm_magic);
3682 rc = lod_xattr_del_internal(env, dt, name, th);
3686 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3693 * Set default striping on a directory object.
3695 * Sets specified striping on a directory object unless it matches the default
3696 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3697 * EA. This striping will be used when a new directory is being created in the
3700 * \param[in] env execution environment
3701 * \param[in] dt the striped object
3702 * \param[in] buf buffer with the striping
3703 * \param[in] name name of EA
3704 * \param[in] fl xattr flag (see OSD API description)
3705 * \param[in] th transaction handle
3707 * \retval 0 on success
3708 * \retval negative if failed
3710 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
3711 struct dt_object *dt,
3712 const struct lu_buf *buf,
3713 const char *name, int fl,
3716 struct lmv_user_md_v1 *lum;
3721 LASSERT(buf != NULL && buf->lb_buf != NULL);
3725 "set default stripe_count # %u stripe_offset %d hash %u\n",
3726 le32_to_cpu(lum->lum_stripe_count),
3727 (int)le32_to_cpu(lum->lum_stripe_offset),
3728 le32_to_cpu(lum->lum_hash_type));
3730 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
3731 le32_to_cpu(lum->lum_stripe_offset)) &&
3732 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
3733 rc = lod_xattr_del_internal(env, dt, name, th);
3737 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3746 * Turn directory into a striped directory.
3748 * During replay the client sends the striping created before MDT
3749 * failure, then the layer above LOD sends this defined striping
3750 * using ->do_xattr_set(), so LOD uses this method to replay creation
3751 * of the stripes. Notice the original information for the striping
3752 * (#stripes, FIDs, etc) was transferred in declare path.
3754 * \param[in] env execution environment
3755 * \param[in] dt the striped object
3756 * \param[in] buf not used currently
3757 * \param[in] name not used currently
3758 * \param[in] fl xattr flag (see OSD API description)
3759 * \param[in] th transaction handle
3761 * \retval 0 on success
3762 * \retval negative if failed
3764 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
3765 const struct lu_buf *buf, const char *name,
3766 int fl, struct thandle *th)
3768 struct lod_object *lo = lod_dt_obj(dt);
3769 struct lod_thread_info *info = lod_env_info(env);
3770 struct lu_attr *attr = &info->lti_attr;
3771 struct dt_object_format *dof = &info->lti_format;
3772 struct lu_buf lmv_buf;
3773 struct lu_buf slave_lmv_buf;
3774 struct lmv_mds_md_v1 *lmm;
3775 struct lmv_mds_md_v1 *slave_lmm = NULL;
3776 struct dt_insert_rec *rec = &info->lti_dt_rec;
3781 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3784 /* The stripes are supposed to be allocated in declare phase,
3785 * if there are no stripes being allocated, it will skip */
3786 if (lo->ldo_dir_stripe_count == 0) {
3787 if (lo->ldo_dir_is_foreign) {
3788 rc = lod_sub_xattr_set(env, dt_object_child(dt), buf,
3789 XATTR_NAME_LMV, fl, th);
3796 rc = dt_attr_get(env, dt_object_child(dt), attr);
3800 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME | LA_FLAGS |
3801 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
3802 dof->dof_type = DFT_DIR;
3804 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
3807 lmm = lmv_buf.lb_buf;
3809 OBD_ALLOC_PTR(slave_lmm);
3810 if (slave_lmm == NULL)
3813 lod_prep_slave_lmv_md(slave_lmm, lmm);
3814 slave_lmv_buf.lb_buf = slave_lmm;
3815 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
3817 rec->rec_type = S_IFDIR;
3818 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3819 struct dt_object *dto = lo->ldo_stripe[i];
3820 char *stripe_name = info->lti_key;
3821 struct lu_name *sname;
3822 struct linkea_data ldata = { NULL };
3823 struct lu_buf linkea_buf;
3825 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
3829 /* fail a remote stripe creation */
3830 if (i && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_CREATE))
3833 /* don't create stripe if:
3834 * 1. it's source stripe of migrating directory
3835 * 2. it's existed stripe of splitting directory
3837 if ((lod_is_migrating(lo) && i >= lo->ldo_dir_migrate_offset) ||
3838 (lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) {
3839 if (!dt_object_exists(dto))
3840 GOTO(out, rc = -EINVAL);
3842 dt_write_lock(env, dto, DT_TGT_CHILD);
3843 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
3845 dt_write_unlock(env, dto);
3849 rc = lod_sub_ref_add(env, dto, th);
3850 dt_write_unlock(env, dto);
3854 rec->rec_fid = lu_object_fid(&dto->do_lu);
3855 rc = lod_sub_insert(env, dto,
3856 (const struct dt_rec *)rec,
3857 (const struct dt_key *)dot, th);
3862 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
3863 cfs_fail_val != i) {
3864 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
3866 slave_lmm->lmv_master_mdt_index =
3869 slave_lmm->lmv_master_mdt_index =
3872 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
3873 XATTR_NAME_LMV, 0, th);
3878 /* don't insert stripe if it's existed stripe of splitting
3879 * directory (this directory is striped).
3880 * NB, plain directory will insert itself as the first
3883 if (lod_is_splitting(lo) && lo->ldo_dir_split_offset > 1 &&
3884 lo->ldo_dir_split_offset > i)
3887 rec->rec_fid = lu_object_fid(&dt->do_lu);
3888 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
3889 (const struct dt_key *)dotdot, th);
3893 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
3895 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3896 PFID(lu_object_fid(&dto->do_lu)), i + 1);
3898 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3899 PFID(lu_object_fid(&dto->do_lu)), i);
3901 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
3902 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
3903 sname, lu_object_fid(&dt->do_lu));
3907 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
3908 linkea_buf.lb_len = ldata.ld_leh->leh_len;
3909 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
3910 XATTR_NAME_LINK, 0, th);
3914 rec->rec_fid = lu_object_fid(&dto->do_lu);
3915 rc = lod_sub_insert(env, dt_object_child(dt),
3916 (const struct dt_rec *)rec,
3917 (const struct dt_key *)stripe_name, th);
3921 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
3926 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
3927 rc = lod_sub_xattr_set(env, dt_object_child(dt),
3928 &lmv_buf, XATTR_NAME_LMV, fl, th);
3930 if (slave_lmm != NULL)
3931 OBD_FREE_PTR(slave_lmm);
3937 * Helper function to declare/execute creation of a striped directory
3939 * Called in declare/create object path, prepare striping for a directory
3940 * and prepare defaults data striping for the objects to be created in
3941 * that directory. Notice the function calls "declaration" or "execution"
3942 * methods depending on \a declare param. This is a consequence of the
3943 * current approach while we don't have natural distributed transactions:
3944 * we basically execute non-local updates in the declare phase. So, the
3945 * arguments for the both phases are the same and this is the reason for
3946 * this function to exist.
3948 * \param[in] env execution environment
3949 * \param[in] dt object
3950 * \param[in] attr attributes the stripes will be created with
3951 * \param[in] lmu lmv_user_md if MDT indices are specified
3952 * \param[in] dof format of stripes (see OSD API description)
3953 * \param[in] th transaction handle
3954 * \param[in] declare where to call "declare" or "execute" methods
3956 * \retval 0 on success
3957 * \retval negative if failed
3959 static int lod_dir_striping_create_internal(const struct lu_env *env,
3960 struct dt_object *dt,
3961 struct lu_attr *attr,
3962 const struct lu_buf *lmu,
3963 struct dt_object_format *dof,
3967 struct lod_thread_info *info = lod_env_info(env);
3968 struct lod_object *lo = lod_dt_obj(dt);
3969 const struct lod_default_striping *lds = lo->ldo_def_striping;
3973 LASSERT(ergo(lds != NULL,
3974 lds->lds_def_striping_set ||
3975 lds->lds_dir_def_striping_set));
3977 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
3978 lo->ldo_dir_stripe_offset)) {
3980 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3981 int stripe_count = lo->ldo_dir_stripe_count;
3983 if (info->lti_ea_store_size < sizeof(*v1)) {
3984 rc = lod_ea_store_resize(info, sizeof(*v1));
3987 v1 = info->lti_ea_store;
3990 memset(v1, 0, sizeof(*v1));
3991 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3992 v1->lum_stripe_count = cpu_to_le32(stripe_count);
3993 v1->lum_stripe_offset =
3994 cpu_to_le32(lo->ldo_dir_stripe_offset);
3996 info->lti_buf.lb_buf = v1;
3997 info->lti_buf.lb_len = sizeof(*v1);
3998 lmu = &info->lti_buf;
4002 rc = lod_declare_xattr_set_lmv(env, dt, attr, lmu, dof,
4005 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
4010 /* foreign LMV EA case */
4012 struct lmv_foreign_md *lfm = lmu->lb_buf;
4014 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
4015 rc = lod_declare_xattr_set_lmv(env, dt, attr,
4019 if (lo->ldo_dir_is_foreign) {
4020 LASSERT(lo->ldo_foreign_lmv != NULL &&
4021 lo->ldo_foreign_lmv_size > 0);
4022 info->lti_buf.lb_buf = lo->ldo_foreign_lmv;
4023 info->lti_buf.lb_len = lo->ldo_foreign_lmv_size;
4024 lmu = &info->lti_buf;
4025 rc = lod_xattr_set_lmv(env, dt, lmu,
4026 XATTR_NAME_LMV, 0, th);
4031 /* Transfer default LMV striping from the parent */
4032 if (lds != NULL && lds->lds_dir_def_striping_set &&
4033 !(LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
4034 lds->lds_dir_def_stripe_offset) &&
4035 le32_to_cpu(lds->lds_dir_def_hash_type) !=
4036 LMV_HASH_TYPE_UNKNOWN)) {
4037 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
4039 if (info->lti_ea_store_size < sizeof(*v1)) {
4040 rc = lod_ea_store_resize(info, sizeof(*v1));
4043 v1 = info->lti_ea_store;
4046 memset(v1, 0, sizeof(*v1));
4047 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
4048 v1->lum_stripe_count =
4049 cpu_to_le32(lds->lds_dir_def_stripe_count);
4050 v1->lum_stripe_offset =
4051 cpu_to_le32(lds->lds_dir_def_stripe_offset);
4053 cpu_to_le32(lds->lds_dir_def_hash_type);
4055 info->lti_buf.lb_buf = v1;
4056 info->lti_buf.lb_len = sizeof(*v1);
4058 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4059 XATTR_NAME_DEFAULT_LMV,
4062 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
4064 XATTR_NAME_DEFAULT_LMV, 0,
4070 /* Transfer default LOV striping from the parent */
4071 if (lds != NULL && lds->lds_def_striping_set &&
4072 lds->lds_def_comp_cnt != 0) {
4073 struct lov_mds_md *lmm;
4074 int lmm_size = lod_comp_md_size(lo, true);
4076 if (info->lti_ea_store_size < lmm_size) {
4077 rc = lod_ea_store_resize(info, lmm_size);
4081 lmm = info->lti_ea_store;
4083 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
4087 info->lti_buf.lb_buf = lmm;
4088 info->lti_buf.lb_len = lmm_size;
4091 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4092 XATTR_NAME_LOV, 0, th);
4094 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4095 XATTR_NAME_LOV, 0, th);
4103 static int lod_declare_dir_striping_create(const struct lu_env *env,
4104 struct dt_object *dt,
4105 struct lu_attr *attr,
4107 struct dt_object_format *dof,
4110 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
4114 static int lod_dir_striping_create(const struct lu_env *env,
4115 struct dt_object *dt,
4116 struct lu_attr *attr,
4117 struct dt_object_format *dof,
4120 return lod_dir_striping_create_internal(env, dt, attr, NULL, dof, th,
4125 * Make LOV EA for striped object.
4127 * Generate striping information and store it in the LOV EA of the given
4128 * object. The caller must ensure nobody else is calling the function
4129 * against the object concurrently. The transaction must be started.
4130 * FLDB service must be running as well; it's used to map FID to the target,
4131 * which is stored in LOV EA.
4133 * \param[in] env execution environment for this thread
4134 * \param[in] lo LOD object
4135 * \param[in] th transaction handle
4137 * \retval 0 if LOV EA is stored successfully
4138 * \retval negative error number on failure
4140 static int lod_generate_and_set_lovea(const struct lu_env *env,
4141 struct lod_object *lo,
4144 struct lod_thread_info *info = lod_env_info(env);
4145 struct dt_object *next = dt_object_child(&lo->ldo_obj);
4146 struct lov_mds_md_v1 *lmm;
4152 if (lo->ldo_comp_cnt == 0 && !lo->ldo_is_foreign) {
4153 lod_striping_free(env, lo);
4154 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
4158 lmm_size = lod_comp_md_size(lo, false);
4159 if (info->lti_ea_store_size < lmm_size) {
4160 rc = lod_ea_store_resize(info, lmm_size);
4164 lmm = info->lti_ea_store;
4166 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
4170 info->lti_buf.lb_buf = lmm;
4171 info->lti_buf.lb_len = lmm_size;
4172 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
4173 XATTR_NAME_LOV, 0, th);
4177 static __u32 lod_gen_component_id(struct lod_object *lo,
4178 int mirror_id, int comp_idx);
4181 * Repeat an existing component
4183 * Creates a new layout by replicating an existing component. Uses striping
4184 * policy from previous component as a template for the striping for the new
4187 * New component starts with zero length, will be extended (or removed) before
4188 * returning layout to client.
4190 * NB: Reallocates layout components array (lo->ldo_comp_entries), invalidating
4191 * any pre-existing pointers to components. Handle with care.
4193 * \param[in] env execution environment for this thread
4194 * \param[in,out] lo object to update the layout of
4195 * \param[in] index index of component to copy
4197 * \retval 0 on success
4198 * \retval negative errno on error
4200 static int lod_layout_repeat_comp(const struct lu_env *env,
4201 struct lod_object *lo, int index)
4203 struct lod_layout_component *lod_comp;
4204 struct lod_layout_component *new_comp = NULL;
4205 struct lod_layout_component *comp_array;
4206 int rc = 0, i, new_cnt = lo->ldo_comp_cnt + 1;
4211 lod_comp = &lo->ldo_comp_entries[index];
4212 LASSERT(lod_comp_inited(lod_comp) && lod_comp->llc_id != LCME_ID_INVAL);
4214 CDEBUG(D_LAYOUT, "repeating component %d\n", index);
4216 OBD_ALLOC_PTR_ARRAY(comp_array, new_cnt);
4217 if (comp_array == NULL)
4218 GOTO(out, rc = -ENOMEM);
4220 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4221 memcpy(&comp_array[i + offset], &lo->ldo_comp_entries[i],
4222 sizeof(*comp_array));
4224 /* Duplicate this component in to the next slot */
4226 new_comp = &comp_array[i + 1];
4227 memcpy(&comp_array[i + 1], &lo->ldo_comp_entries[i],
4228 sizeof(*comp_array));
4229 /* We must now skip this new component when copying */
4234 /* Set up copied component */
4235 new_comp->llc_flags &= ~LCME_FL_INIT;
4236 new_comp->llc_stripe = NULL;
4237 new_comp->llc_stripes_allocated = 0;
4238 new_comp->llc_ost_indices = NULL;
4239 new_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4240 /* for uninstantiated components, layout gen stores default stripe
4242 new_comp->llc_layout_gen = lod_comp->llc_stripe_offset;
4243 /* This makes the repeated component zero-length, placed at the end of
4244 * the preceding component */
4245 new_comp->llc_extent.e_start = new_comp->llc_extent.e_end;
4246 new_comp->llc_timestamp = lod_comp->llc_timestamp;
4247 new_comp->llc_pool = NULL;
4249 rc = lod_set_pool(&new_comp->llc_pool, lod_comp->llc_pool);
4253 if (new_comp->llc_ostlist.op_array) {
4254 __u32 *op_array = NULL;
4256 OBD_ALLOC(op_array, new_comp->llc_ostlist.op_size);
4258 GOTO(out, rc = -ENOMEM);
4259 memcpy(op_array, &new_comp->llc_ostlist.op_array,
4260 new_comp->llc_ostlist.op_size);
4261 new_comp->llc_ostlist.op_array = op_array;
4264 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
4265 lo->ldo_comp_entries = comp_array;
4266 lo->ldo_comp_cnt = new_cnt;
4268 /* Generate an id for the new component */
4269 mirror_id = mirror_id_of(new_comp->llc_id);
4270 new_comp->llc_id = LCME_ID_INVAL;
4271 new_comp->llc_id = lod_gen_component_id(lo, mirror_id, index + 1);
4272 if (new_comp->llc_id == LCME_ID_INVAL)
4273 GOTO(out, rc = -ERANGE);
4278 OBD_FREE_PTR_ARRAY(comp_array, new_cnt);
4283 static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
4287 /* clear memory region that will be used for layout change */
4288 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
4289 info->lti_count = 0;
4291 if (info->lti_comp_size >= comp_cnt)
4294 if (info->lti_comp_size > 0) {
4295 OBD_FREE_PTR_ARRAY(info->lti_comp_idx, info->lti_comp_size);
4296 info->lti_comp_size = 0;
4299 OBD_ALLOC_PTR_ARRAY(info->lti_comp_idx, comp_cnt);
4300 if (!info->lti_comp_idx)
4303 info->lti_comp_size = comp_cnt;
4308 * Prepare new layout minus deleted components
4310 * Removes components marked for deletion (LCME_ID_INVAL) by copying to a new
4311 * layout and skipping those components. Removes stripe objects if any exist.
4314 * Reallocates layout components array (lo->ldo_comp_entries), invalidating
4315 * any pre-existing pointers to components.
4317 * Caller is responsible for updating mirror end (ldo_mirror[].lme_end).
4319 * \param[in] env execution environment for this thread
4320 * \param[in,out] lo object to update the layout of
4321 * \param[in] th transaction handle for this operation
4323 * \retval # of components deleted
4324 * \retval negative errno on error
4326 static int lod_layout_del_prep_layout(const struct lu_env *env,
4327 struct lod_object *lo,
4330 struct lod_layout_component *lod_comp;
4331 struct lod_thread_info *info = lod_env_info(env);
4332 int rc = 0, i, j, deleted = 0;
4336 LASSERT(lo->ldo_is_composite);
4337 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
4339 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
4343 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4344 lod_comp = &lo->ldo_comp_entries[i];
4346 if (lod_comp->llc_id != LCME_ID_INVAL) {
4347 /* Build array of things to keep */
4348 info->lti_comp_idx[info->lti_count++] = i;
4352 lod_obj_set_pool(lo, i, NULL);
4353 if (lod_comp->llc_ostlist.op_array) {
4354 OBD_FREE(lod_comp->llc_ostlist.op_array,
4355 lod_comp->llc_ostlist.op_size);
4356 lod_comp->llc_ostlist.op_array = NULL;
4357 lod_comp->llc_ostlist.op_size = 0;
4361 CDEBUG(D_LAYOUT, "deleting comp %d, left %d\n", i,
4362 lo->ldo_comp_cnt - deleted);
4364 /* No striping info for this component */
4365 if (lod_comp->llc_stripe == NULL)
4368 LASSERT(lod_comp->llc_stripe_count > 0);
4369 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4370 struct dt_object *obj = lod_comp->llc_stripe[j];
4375 /* components which are not init have no sub objects
4377 if (lod_comp_inited(lod_comp)) {
4378 rc = lod_sub_destroy(env, obj, th);
4383 lu_object_put(env, &obj->do_lu);
4384 lod_comp->llc_stripe[j] = NULL;
4386 OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe,
4387 lod_comp->llc_stripes_allocated);
4388 lod_comp->llc_stripe = NULL;
4389 OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices,
4390 lod_comp->llc_stripes_allocated);
4391 lod_comp->llc_ost_indices = NULL;
4392 lod_comp->llc_stripes_allocated = 0;
4395 /* info->lti_count has the amount of left components */
4396 LASSERTF(info->lti_count >= 0 && info->lti_count < lo->ldo_comp_cnt,
4397 "left = %d, lo->ldo_comp_cnt %d\n", (int)info->lti_count,
4398 (int)lo->ldo_comp_cnt);
4400 if (info->lti_count > 0) {
4401 struct lod_layout_component *comp_array;
4403 OBD_ALLOC_PTR_ARRAY(comp_array, info->lti_count);
4404 if (comp_array == NULL)
4405 GOTO(out, rc = -ENOMEM);
4407 for (i = 0; i < info->lti_count; i++) {
4408 memcpy(&comp_array[i],
4409 &lo->ldo_comp_entries[info->lti_comp_idx[i]],
4410 sizeof(*comp_array));
4413 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
4414 lo->ldo_comp_entries = comp_array;
4415 lo->ldo_comp_cnt = info->lti_count;
4417 lod_free_comp_entries(lo);
4422 return rc ? rc : deleted;
4426 * Delete layout component(s)
4428 * This function sets up the layout data in the env and does the setattrs
4429 * required to write out the new layout. The layout itself is modified in
4430 * lod_layout_del_prep_layout.
4432 * \param[in] env execution environment for this thread
4433 * \param[in] dt object
4434 * \param[in] th transaction handle
4436 * \retval 0 on success
4437 * \retval negative error number on failure
4439 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
4442 struct lod_object *lo = lod_dt_obj(dt);
4443 struct dt_object *next = dt_object_child(dt);
4444 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4447 LASSERT(lo->ldo_mirror_count == 1);
4449 rc = lod_layout_del_prep_layout(env, lo, th);
4453 /* Only do this if we didn't delete all components */
4454 if (lo->ldo_comp_cnt > 0) {
4455 lo->ldo_mirrors[0].lme_end = lo->ldo_comp_cnt - 1;
4456 lod_obj_inc_layout_gen(lo);
4459 LASSERT(dt_object_exists(dt));
4460 rc = dt_attr_get(env, next, attr);
4464 if (attr->la_size > 0) {
4466 attr->la_valid = LA_SIZE;
4467 rc = lod_sub_attr_set(env, next, attr, th);
4472 rc = lod_generate_and_set_lovea(env, lo, th);
4476 lod_striping_free(env, lo);
4481 static int lod_get_default_lov_striping(const struct lu_env *env,
4482 struct lod_object *lo,
4483 struct lod_default_striping *lds,
4484 struct dt_allocation_hint *ah);
4486 * Implementation of dt_object_operations::do_xattr_set.
4488 * Sets specified extended attribute on the object. Three types of EAs are
4490 * LOV EA - stores striping for a regular file or default striping (when set
4492 * LMV EA - stores a marker for the striped directories
4493 * DMV EA - stores default directory striping
4495 * When striping is applied to a non-striped existing object (this is called
4496 * late striping), then LOD notices the caller wants to turn the object into a
4497 * striped one. The stripe objects are created and appropriate EA is set:
4498 * LOV EA storing all the stripes directly or LMV EA storing just a small header
4499 * with striping configuration.
4501 * \see dt_object_operations::do_xattr_set() in the API description for details.
4503 static int lod_xattr_set(const struct lu_env *env,
4504 struct dt_object *dt, const struct lu_buf *buf,
4505 const char *name, int fl, struct thandle *th)
4507 struct dt_object *next = dt_object_child(dt);
4512 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4513 !strcmp(name, XATTR_NAME_LMV)) {
4515 case LU_XATTR_CREATE:
4516 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
4519 case LU_XATTR_REPLACE:
4520 rc = lod_dir_layout_set(env, dt, buf, fl, th);
4527 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4528 strcmp(name, XATTR_NAME_LOV) == 0) {
4529 struct lod_default_striping *lds = lod_lds_buf_get(env);
4530 struct lov_user_md_v1 *v1 = buf->lb_buf;
4531 char pool[LOV_MAXPOOLNAME + 1];
4534 /* get existing striping config */
4535 rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds,
4540 memset(pool, 0, sizeof(pool));
4541 if (lds->lds_def_striping_set == 1)
4542 lod_layout_get_pool(lds->lds_def_comp_entries,
4543 lds->lds_def_comp_cnt, pool,
4546 is_del = LOVEA_DELETE_VALUES(v1->lmm_stripe_size,
4547 v1->lmm_stripe_count,
4548 v1->lmm_stripe_offset,
4551 /* Retain the pool name if it is not given */
4552 if (v1->lmm_magic == LOV_USER_MAGIC_V1 && pool[0] != '\0' &&
4554 struct lod_thread_info *info = lod_env_info(env);
4555 struct lov_user_md_v3 *v3 = info->lti_ea_store;
4557 memset(v3, 0, sizeof(*v3));
4558 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4559 v3->lmm_pattern = cpu_to_le32(v1->lmm_pattern);
4560 v3->lmm_stripe_count =
4561 cpu_to_le32(v1->lmm_stripe_count);
4562 v3->lmm_stripe_offset =
4563 cpu_to_le32(v1->lmm_stripe_offset);
4564 v3->lmm_stripe_size = cpu_to_le32(v1->lmm_stripe_size);
4566 strlcpy(v3->lmm_pool_name, pool,
4567 sizeof(v3->lmm_pool_name));
4569 info->lti_buf.lb_buf = v3;
4570 info->lti_buf.lb_len = sizeof(*v3);
4571 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4574 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name,
4578 if (lds->lds_def_striping_set == 1 &&
4579 lds->lds_def_comp_entries != NULL)
4580 lod_free_def_comp_entries(lds);
4583 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4584 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
4586 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
4589 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
4590 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4591 strcmp(name, XATTR_LUSTRE_LOV) == 0 ||
4592 allowed_lustre_lov(name))) {
4593 /* in case of lov EA swap, just set it
4594 * if not, it is a replay so check striping match what we
4595 * already have during req replay, declare_xattr_set()
4596 * defines striping, then create() does the work */
4597 if (fl & LU_XATTR_REPLACE) {
4598 /* free stripes, then update disk */
4599 lod_striping_free(env, lod_dt_obj(dt));
4601 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
4602 } else if (dt_object_remote(dt)) {
4603 /* This only happens during migration, see
4604 * mdd_migrate_create(), in which Master MDT will
4605 * create a remote target object, and only set
4606 * (migrating) stripe EA on the remote object,
4607 * and does not need creating each stripes. */
4608 rc = lod_sub_xattr_set(env, next, buf, name,
4610 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
4611 /* delete component(s) */
4612 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
4613 rc = lod_layout_del(env, dt, th);
4616 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
4617 * it's going to create create file with specified
4618 * component(s), the striping must have not being
4619 * cached in this case;
4621 * Otherwise, it's going to add/change component(s) to
4622 * an existing file, the striping must have been cached
4625 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
4626 !strcmp(name, XATTR_NAME_LOV),
4627 !lod_dt_obj(dt)->ldo_comp_cached));
4629 rc = lod_striped_create(env, dt, NULL, NULL, th);
4632 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4633 rc = lod_replace_parent_fid(env, dt, buf, th, false);
4638 /* then all other xattr */
4639 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4645 * Implementation of dt_object_operations::do_declare_xattr_del.
4647 * \see dt_object_operations::do_declare_xattr_del() in the API description
4650 static int lod_declare_xattr_del(const struct lu_env *env,
4651 struct dt_object *dt, const char *name,
4654 struct lod_object *lo = lod_dt_obj(dt);
4655 struct dt_object *next = dt_object_child(dt);
4660 rc = lod_sub_declare_xattr_del(env, next, name, th);
4664 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4667 /* NB: don't delete stripe LMV, because when we do this, normally we
4668 * will remove stripes, besides, if directory LMV is corrupt, this will
4669 * prevent deleting its LMV and fixing it (via LFSCK).
4671 if (!strcmp(name, XATTR_NAME_LMV))
4674 rc = lod_striping_load(env, lo);
4678 if (lo->ldo_dir_stripe_count == 0)
4681 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4682 struct dt_object *dto = lo->ldo_stripe[i];
4687 if (!dt_object_exists(dto))
4690 rc = lod_sub_declare_xattr_del(env, dto, name, th);
4699 * Implementation of dt_object_operations::do_xattr_del.
4701 * If EA storing a regular striping is being deleted, then release
4702 * all the references to the stripe objects in core.
4704 * \see dt_object_operations::do_xattr_del() in the API description for details.
4706 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
4707 const char *name, struct thandle *th)
4713 if (!strcmp(name, XATTR_NAME_LOV) || !strcmp(name, XATTR_NAME_LMV))
4714 lod_striping_free(env, lod_dt_obj(dt));
4716 rc = lod_xattr_del_internal(env, dt, name, th);
4722 * Implementation of dt_object_operations::do_xattr_list.
4724 * \see dt_object_operations::do_xattr_list() in the API description
4727 static int lod_xattr_list(const struct lu_env *env,
4728 struct dt_object *dt, const struct lu_buf *buf)
4730 return dt_xattr_list(env, dt_object_child(dt), buf);
4733 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
4735 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
4739 * Copy OST list from layout provided by user.
4741 * \param[in] lod_comp layout_component to be filled
4742 * \param[in] v3 LOV EA V3 user data
4744 * \retval 0 on success
4745 * \retval negative if failed
4747 int lod_comp_copy_ost_lists(struct lod_layout_component *lod_comp,
4748 struct lov_user_md_v3 *v3)
4754 if (v3->lmm_stripe_offset == LOV_OFFSET_DEFAULT)
4755 v3->lmm_stripe_offset = v3->lmm_objects[0].l_ost_idx;
4757 if (lod_comp->llc_ostlist.op_array) {
4758 if (lod_comp->llc_ostlist.op_size >=
4759 v3->lmm_stripe_count * sizeof(__u32)) {
4760 lod_comp->llc_ostlist.op_count =
4761 v3->lmm_stripe_count;
4764 OBD_FREE(lod_comp->llc_ostlist.op_array,
4765 lod_comp->llc_ostlist.op_size);
4768 /* copy ost list from lmm */
4769 lod_comp->llc_ostlist.op_count = v3->lmm_stripe_count;
4770 lod_comp->llc_ostlist.op_size = v3->lmm_stripe_count * sizeof(__u32);
4771 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
4772 lod_comp->llc_ostlist.op_size);
4773 if (!lod_comp->llc_ostlist.op_array)
4776 for (j = 0; j < v3->lmm_stripe_count; j++) {
4777 lod_comp->llc_ostlist.op_array[j] =
4778 v3->lmm_objects[j].l_ost_idx;
4786 * Get default striping.
4788 * \param[in] env execution environment
4789 * \param[in] lo object
4790 * \param[out] lds default striping
4792 * \retval 0 on success
4793 * \retval negative if failed
4795 static int lod_get_default_lov_striping(const struct lu_env *env,
4796 struct lod_object *lo,
4797 struct lod_default_striping *lds,
4798 struct dt_allocation_hint *ah)
4800 struct lod_thread_info *info = lod_env_info(env);
4801 struct lov_user_md_v1 *v1 = NULL;
4802 struct lov_user_md_v3 *v3 = NULL;
4803 struct lov_comp_md_v1 *comp_v1 = NULL;
4811 rc = lod_get_lov_ea(env, lo);
4815 if (rc < (typeof(rc))sizeof(struct lov_user_md))
4818 v1 = info->lti_ea_store;
4819 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
4820 lustre_swab_lov_user_md_v1(v1);
4821 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
4822 v3 = (struct lov_user_md_v3 *)v1;
4823 lustre_swab_lov_user_md_v3(v3);
4824 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_SPECIFIC)) {
4825 v3 = (struct lov_user_md_v3 *)v1;
4826 lustre_swab_lov_user_md_v3(v3);
4827 lustre_swab_lov_user_md_objects(v3->lmm_objects,
4828 v3->lmm_stripe_count);
4829 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_COMP_V1) ||
4830 v1->lmm_magic == __swab32(LOV_USER_MAGIC_SEL)) {
4831 comp_v1 = (struct lov_comp_md_v1 *)v1;
4832 lustre_swab_lov_comp_md_v1(comp_v1);
4835 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1 &&
4836 v1->lmm_magic != LOV_MAGIC_COMP_V1 &&
4837 v1->lmm_magic != LOV_MAGIC_SEL &&
4838 v1->lmm_magic != LOV_USER_MAGIC_SPECIFIC)
4841 if ((v1->lmm_magic == LOV_MAGIC_COMP_V1 ||
4842 v1->lmm_magic == LOV_MAGIC_SEL) &&
4843 !(ah && ah->dah_append_stripes)) {
4844 comp_v1 = (struct lov_comp_md_v1 *)v1;
4845 comp_cnt = comp_v1->lcm_entry_count;
4848 mirror_cnt = comp_v1->lcm_mirror_count + 1;
4856 /* realloc default comp entries if necessary */
4857 rc = lod_def_striping_comp_resize(lds, comp_cnt);
4861 lds->lds_def_comp_cnt = comp_cnt;
4862 lds->lds_def_striping_is_composite = composite;
4863 lds->lds_def_mirror_cnt = mirror_cnt;
4865 for (i = 0; i < comp_cnt; i++) {
4866 struct lod_layout_component *lod_comp;
4869 lod_comp = &lds->lds_def_comp_entries[i];
4871 * reset lod_comp values, llc_stripes is always NULL in
4872 * the default striping template, llc_pool will be reset
4875 memset(lod_comp, 0, offsetof(typeof(*lod_comp), llc_pool));
4878 v1 = (struct lov_user_md *)((char *)comp_v1 +
4879 comp_v1->lcm_entries[i].lcme_offset);
4880 lod_comp->llc_extent =
4881 comp_v1->lcm_entries[i].lcme_extent;
4882 /* We only inherit certain flags from the layout */
4883 lod_comp->llc_flags =
4884 comp_v1->lcm_entries[i].lcme_flags &
4885 LCME_TEMPLATE_FLAGS;
4888 if (!lov_pattern_supported(v1->lmm_pattern) &&
4889 !(v1->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
4890 lod_free_def_comp_entries(lds);
4894 CDEBUG(D_LAYOUT, DFID" stripe_count=%d stripe_size=%d stripe_offset=%d append_stripes=%d\n",
4895 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
4896 (int)v1->lmm_stripe_count, (int)v1->lmm_stripe_size,
4897 (int)v1->lmm_stripe_offset,
4898 ah ? ah->dah_append_stripes : 0);
4900 if (ah && ah->dah_append_stripes)
4901 lod_comp->llc_stripe_count = ah->dah_append_stripes;
4903 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
4904 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
4905 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
4906 lod_comp->llc_pattern = v1->lmm_pattern;
4909 if (ah && ah->dah_append_pool && ah->dah_append_pool[0]) {
4910 pool = ah->dah_append_pool;
4911 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
4912 /* XXX: sanity check here */
4913 v3 = (struct lov_user_md_v3 *) v1;
4914 if (v3->lmm_pool_name[0] != '\0')
4915 pool = v3->lmm_pool_name;
4917 lod_set_def_pool(lds, i, pool);
4918 if (v1->lmm_magic == LOV_USER_MAGIC_SPECIFIC) {
4919 v3 = (struct lov_user_md_v3 *)v1;
4920 rc = lod_comp_copy_ost_lists(lod_comp, v3);
4923 } else if (lod_comp->llc_ostlist.op_array &&
4924 lod_comp->llc_ostlist.op_count) {
4925 for (j = 0; j < lod_comp->llc_ostlist.op_count; j++)
4926 lod_comp->llc_ostlist.op_array[j] = -1;
4927 lod_comp->llc_ostlist.op_count = 0;
4931 lds->lds_def_striping_set = 1;
4936 * Get default directory striping.
4938 * \param[in] env execution environment
4939 * \param[in] lo object
4940 * \param[out] lds default striping
4942 * \retval 0 on success
4943 * \retval negative if failed
4945 static int lod_get_default_lmv_striping(const struct lu_env *env,
4946 struct lod_object *lo,
4947 struct lod_default_striping *lds)
4949 struct lmv_user_md *lmu;
4952 lds->lds_dir_def_striping_set = 0;
4954 rc = lod_get_default_lmv_ea(env, lo);
4958 if (rc >= (int)sizeof(*lmu)) {
4959 struct lod_thread_info *info = lod_env_info(env);
4961 lmu = info->lti_ea_store;
4963 lds->lds_dir_def_stripe_count =
4964 le32_to_cpu(lmu->lum_stripe_count);
4965 lds->lds_dir_def_stripe_offset =
4966 le32_to_cpu(lmu->lum_stripe_offset);
4967 lds->lds_dir_def_hash_type =
4968 le32_to_cpu(lmu->lum_hash_type);
4969 lds->lds_dir_def_striping_set = 1;
4976 * Get default striping in the object.
4978 * Get object default striping and default directory striping.
4980 * \param[in] env execution environment
4981 * \param[in] lo object
4982 * \param[out] lds default striping
4984 * \retval 0 on success
4985 * \retval negative if failed
4987 static int lod_get_default_striping(const struct lu_env *env,
4988 struct lod_object *lo,
4989 struct lod_default_striping *lds)
4993 rc = lod_get_default_lov_striping(env, lo, lds, NULL);
4994 rc1 = lod_get_default_lmv_striping(env, lo, lds);
4995 if (rc == 0 && rc1 < 0)
5002 * Apply default striping on object.
5004 * If object striping pattern is not set, set to the one in default striping.
5005 * The default striping is from parent or fs.
5007 * \param[in] lo new object
5008 * \param[in] lds default striping
5009 * \param[in] mode new object's mode
5011 static void lod_striping_from_default(struct lod_object *lo,
5012 const struct lod_default_striping *lds,
5015 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5018 if (lds->lds_def_striping_set && S_ISREG(mode)) {
5019 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
5021 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
5022 lds->lds_def_comp_cnt);
5026 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
5027 if (lds->lds_def_mirror_cnt > 1)
5028 lo->ldo_flr_state = LCM_FL_RDONLY;
5030 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5031 struct lod_layout_component *obj_comp =
5032 &lo->ldo_comp_entries[i];
5033 struct lod_layout_component *def_comp =
5034 &lds->lds_def_comp_entries[i];
5036 CDEBUG(D_LAYOUT, "Inherit from default: flags=%#x "
5037 "size=%hu nr=%u offset=%u pattern=%#x pool=%s\n",
5038 def_comp->llc_flags,
5039 def_comp->llc_stripe_size,
5040 def_comp->llc_stripe_count,
5041 def_comp->llc_stripe_offset,
5042 def_comp->llc_pattern,
5043 def_comp->llc_pool ?: "");
5045 *obj_comp = *def_comp;
5046 if (def_comp->llc_pool != NULL) {
5047 /* pointer was copied from def_comp */
5048 obj_comp->llc_pool = NULL;
5049 lod_obj_set_pool(lo, i, def_comp->llc_pool);
5053 if (def_comp->llc_ostlist.op_array &&
5054 def_comp->llc_ostlist.op_count) {
5055 OBD_ALLOC(obj_comp->llc_ostlist.op_array,
5056 obj_comp->llc_ostlist.op_size);
5057 if (!obj_comp->llc_ostlist.op_array)
5059 memcpy(obj_comp->llc_ostlist.op_array,
5060 def_comp->llc_ostlist.op_array,
5061 obj_comp->llc_ostlist.op_size);
5062 } else if (def_comp->llc_ostlist.op_array) {
5063 obj_comp->llc_ostlist.op_array = NULL;
5067 * Don't initialize these fields for plain layout
5068 * (v1/v3) here, they are inherited in the order of
5069 * 'parent' -> 'fs default (root)' -> 'global default
5070 * values for stripe_count & stripe_size'.
5072 * see lod_ah_init().
5074 if (!lo->ldo_is_composite)
5077 lod_adjust_stripe_info(obj_comp, desc, 0);
5079 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
5080 if (lo->ldo_dir_stripe_count == 0)
5081 lo->ldo_dir_stripe_count =
5082 lds->lds_dir_def_stripe_count;
5083 if (lo->ldo_dir_stripe_offset == -1)
5084 lo->ldo_dir_stripe_offset =
5085 lds->lds_dir_def_stripe_offset;
5086 if (lo->ldo_dir_hash_type == 0)
5087 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
5089 CDEBUG(D_LAYOUT, "striping from default dir: count:%hu, "
5090 "offset:%u, hash_type:%u\n",
5091 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
5092 lo->ldo_dir_hash_type);
5096 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root,
5099 struct lod_layout_component *lod_comp;
5101 if (lo->ldo_comp_cnt == 0)
5104 if (lo->ldo_is_composite)
5107 lod_comp = &lo->ldo_comp_entries[0];
5109 if (lod_comp->llc_stripe_count <= 0 ||
5110 lod_comp->llc_stripe_size <= 0)
5113 if (from_root && (lod_comp->llc_pool == NULL ||
5114 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
5117 if (append_pool && append_pool[0])
5124 * Implementation of dt_object_operations::do_ah_init.
5126 * This method is used to make a decision on the striping configuration for the
5127 * object being created. It can be taken from the \a parent object if it exists,
5128 * or filesystem's default. The resulting configuration (number of stripes,
5129 * stripe size/offset, pool name, etc) is stored in the object itself and will
5130 * be used by the methods like ->doo_declare_create().
5132 * \see dt_object_operations::do_ah_init() in the API description for details.
5134 static void lod_ah_init(const struct lu_env *env,
5135 struct dt_allocation_hint *ah,
5136 struct dt_object *parent,
5137 struct dt_object *child,
5140 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
5141 struct lod_thread_info *info = lod_env_info(env);
5142 struct lod_default_striping *lds = lod_lds_buf_get(env);
5143 struct dt_object *nextp = NULL;
5144 struct dt_object *nextc;
5145 struct lod_object *lp = NULL;
5146 struct lod_object *lc;
5147 struct lov_desc *desc;
5148 struct lod_layout_component *lod_comp;
5154 if (ah->dah_append_stripes == -1)
5155 ah->dah_append_stripes =
5156 d->lod_ost_descs.ltd_lov_desc.ld_tgt_count;
5158 if (likely(parent)) {
5159 nextp = dt_object_child(parent);
5160 lp = lod_dt_obj(parent);
5163 nextc = dt_object_child(child);
5164 lc = lod_dt_obj(child);
5166 LASSERT(!lod_obj_is_striped(child));
5167 /* default layout template may have been set on the regular file
5168 * when this is called from mdd_create_data() */
5169 if (S_ISREG(child_mode))
5170 lod_free_comp_entries(lc);
5172 if (!dt_object_exists(nextc))
5173 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
5175 if (S_ISDIR(child_mode)) {
5176 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
5178 /* other default values are 0 */
5179 lc->ldo_dir_stripe_offset = -1;
5181 /* no default striping configuration is needed for
5184 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5185 le32_to_cpu(lum1->lum_magic) == LMV_MAGIC_FOREIGN) {
5186 lc->ldo_dir_is_foreign = true;
5187 /* keep stripe_count 0 and stripe_offset -1 */
5188 CDEBUG(D_INFO, "no default striping for foreign dir\n");
5193 * If parent object is not root directory,
5194 * then get default striping from parent object.
5196 if (likely(lp != NULL)) {
5197 lod_get_default_striping(env, lp, lds);
5199 /* inherit default striping except ROOT */
5200 if ((lds->lds_def_striping_set ||
5201 lds->lds_dir_def_striping_set) &&
5202 !fid_is_root(lod_object_fid(lp)))
5203 lc->ldo_def_striping = lds;
5206 /* It should always honour the specified stripes */
5207 /* Note: old client (< 2.7)might also do lfs mkdir, whose EA
5208 * will have old magic. In this case, we should ignore the
5209 * stripe count and try to create dir by default stripe.
5211 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5212 (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
5213 le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC)) {
5214 lc->ldo_dir_stripe_count =
5215 le32_to_cpu(lum1->lum_stripe_count);
5216 lc->ldo_dir_stripe_offset =
5217 le32_to_cpu(lum1->lum_stripe_offset);
5218 lc->ldo_dir_hash_type =
5219 le32_to_cpu(lum1->lum_hash_type);
5221 "set dirstripe: count %hu, offset %d, hash %u\n",
5222 lc->ldo_dir_stripe_count,
5223 (int)lc->ldo_dir_stripe_offset,
5224 lc->ldo_dir_hash_type);
5226 /* transfer defaults LMV to new directory */
5227 lod_striping_from_default(lc, lds, child_mode);
5229 /* set count 0 to create normal directory */
5230 if (lc->ldo_dir_stripe_count == 1)
5231 lc->ldo_dir_stripe_count = 0;
5234 /* shrink the stripe_count to the avaible MDT count */
5235 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
5236 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)) {
5237 lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
5238 if (lc->ldo_dir_stripe_count == 1)
5239 lc->ldo_dir_stripe_count = 0;
5242 if (!(lc->ldo_dir_hash_type & LMV_HASH_TYPE_MASK))
5243 lc->ldo_dir_hash_type |=
5244 d->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
5246 CDEBUG(D_INFO, "final dir stripe [%hu %d %u]\n",
5247 lc->ldo_dir_stripe_count,
5248 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
5253 /* child object regular file*/
5255 if (!lod_object_will_be_striped(S_ISREG(child_mode),
5256 lu_object_fid(&child->do_lu)))
5259 /* If object is going to be striped over OSTs, transfer default
5260 * striping information to the child, so that we can use it
5261 * during declaration and creation.
5263 * Try from the parent first.
5265 if (likely(lp != NULL)) {
5266 rc = lod_get_default_lov_striping(env, lp, lds, ah);
5268 lod_striping_from_default(lc, lds, child_mode);
5271 /* Initialize lod_device::lod_md_root object reference */
5272 if (d->lod_md_root == NULL) {
5273 struct dt_object *root;
5274 struct lod_object *lroot;
5276 lu_root_fid(&info->lti_fid);
5277 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
5278 if (!IS_ERR(root)) {
5279 lroot = lod_dt_obj(root);
5281 spin_lock(&d->lod_lock);
5282 if (d->lod_md_root != NULL)
5283 dt_object_put(env, &d->lod_md_root->ldo_obj);
5284 d->lod_md_root = lroot;
5285 spin_unlock(&d->lod_lock);
5289 /* try inherit layout from the root object (fs default) when:
5290 * - parent does not have default layout; or
5291 * - parent has plain(v1/v3) default layout, and some attributes
5292 * are not specified in the default layout;
5294 if (d->lod_md_root != NULL &&
5295 lod_need_inherit_more(lc, true, ah->dah_append_pool)) {
5296 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds,
5300 if (lc->ldo_comp_cnt == 0) {
5301 lod_striping_from_default(lc, lds, child_mode);
5302 } else if (!lds->lds_def_striping_is_composite) {
5303 struct lod_layout_component *def_comp;
5305 LASSERT(!lc->ldo_is_composite);
5306 lod_comp = &lc->ldo_comp_entries[0];
5307 def_comp = &lds->lds_def_comp_entries[0];
5309 if (lod_comp->llc_stripe_count <= 0)
5310 lod_comp->llc_stripe_count =
5311 def_comp->llc_stripe_count;
5312 if (lod_comp->llc_stripe_size <= 0)
5313 lod_comp->llc_stripe_size =
5314 def_comp->llc_stripe_size;
5315 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT &&
5316 (!lod_comp->llc_pool || !lod_comp->llc_pool[0]))
5317 lod_comp->llc_stripe_offset =
5318 def_comp->llc_stripe_offset;
5319 if (lod_comp->llc_pool == NULL)
5320 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
5325 * fs default striping may not be explicitly set, or historically set
5326 * in config log, use them.
5328 if (lod_need_inherit_more(lc, false, ah->dah_append_pool)) {
5329 if (lc->ldo_comp_cnt == 0) {
5330 rc = lod_alloc_comp_entries(lc, 0, 1);
5332 /* fail to allocate memory, will create a
5333 * non-striped file. */
5335 lc->ldo_is_composite = 0;
5336 lod_comp = &lc->ldo_comp_entries[0];
5337 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
5339 LASSERT(!lc->ldo_is_composite);
5340 lod_comp = &lc->ldo_comp_entries[0];
5341 desc = &d->lod_ost_descs.ltd_lov_desc;
5342 lod_adjust_stripe_info(lod_comp, desc, ah->dah_append_stripes);
5343 if (ah->dah_append_pool && ah->dah_append_pool[0])
5344 lod_obj_set_pool(lc, 0, ah->dah_append_pool);
5351 * Size initialization on late striping.
5353 * Propagate the size of a truncated object to a deferred striping.
5354 * This function handles a special case when truncate was done on a
5355 * non-striped object and now while the striping is being created
5356 * we can't lose that size, so we have to propagate it to the stripes
5359 * \param[in] env execution environment
5360 * \param[in] dt object
5361 * \param[in] th transaction handle
5363 * \retval 0 on success
5364 * \retval negative if failed
5366 static int lod_declare_init_size(const struct lu_env *env,
5367 struct dt_object *dt, struct thandle *th)
5369 struct dt_object *next = dt_object_child(dt);
5370 struct lod_object *lo = lod_dt_obj(dt);
5371 struct dt_object **objects = NULL;
5372 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
5373 uint64_t size, offs;
5374 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
5375 struct lu_extent size_ext;
5378 if (!lod_obj_is_striped(dt))
5381 rc = dt_attr_get(env, next, attr);
5382 LASSERT(attr->la_valid & LA_SIZE);
5386 size = attr->la_size;
5390 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
5391 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5392 struct lod_layout_component *lod_comp;
5393 struct lu_extent *extent;
5395 lod_comp = &lo->ldo_comp_entries[i];
5397 if (lod_comp->llc_stripe == NULL)
5400 extent = &lod_comp->llc_extent;
5401 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
5402 if (!lo->ldo_is_composite ||
5403 lu_extent_is_overlapped(extent, &size_ext)) {
5404 objects = lod_comp->llc_stripe;
5405 stripe_count = lod_comp->llc_stripe_count;
5406 stripe_size = lod_comp->llc_stripe_size;
5409 if (stripe_count == 0)
5412 LASSERT(objects != NULL && stripe_size != 0);
5413 do_div(size, stripe_size);
5414 stripe = do_div(size, stripe_count);
5415 LASSERT(objects[stripe] != NULL);
5417 size = size * stripe_size;
5418 offs = attr->la_size;
5419 size += do_div(offs, stripe_size);
5421 attr->la_valid = LA_SIZE;
5422 attr->la_size = size;
5424 rc = lod_sub_declare_attr_set(env, objects[stripe],
5433 * Declare creation of striped object.
5435 * The function declares creation stripes for a regular object. The function
5436 * also declares whether the stripes will be created with non-zero size if
5437 * previously size was set non-zero on the master object. If object \a dt is
5438 * not local, then only fully defined striping can be applied in \a lovea.
5439 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
5442 * \param[in] env execution environment
5443 * \param[in] dt object
5444 * \param[in] attr attributes the stripes will be created with
5445 * \param[in] lovea a buffer containing striping description
5446 * \param[in] th transaction handle
5448 * \retval 0 on success
5449 * \retval negative if failed
5451 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
5452 struct lu_attr *attr,
5453 const struct lu_buf *lovea, struct thandle *th)
5455 struct lod_thread_info *info = lod_env_info(env);
5456 struct dt_object *next = dt_object_child(dt);
5457 struct lod_object *lo = lod_dt_obj(dt);
5461 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
5462 GOTO(out, rc = -ENOMEM);
5464 if (!dt_object_remote(next)) {
5465 /* choose OST and generate appropriate objects */
5466 rc = lod_prepare_create(env, lo, attr, lovea, th);
5471 * declare storage for striping data
5473 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
5475 /* LOD can not choose OST objects for remote objects, i.e.
5476 * stripes must be ready before that. Right now, it can only
5477 * happen during migrate, i.e. migrate process needs to create
5478 * remote regular file (mdd_migrate_create), then the migrate
5479 * process will provide stripeEA. */
5480 LASSERT(lovea != NULL);
5481 info->lti_buf = *lovea;
5484 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
5485 XATTR_NAME_LOV, 0, th);
5490 * if striping is created with local object's size > 0,
5491 * we have to propagate this size to specific object
5492 * the case is possible only when local object was created previously
5494 if (dt_object_exists(next))
5495 rc = lod_declare_init_size(env, dt, th);
5498 /* failed to create striping or to set initial size, let's reset
5499 * config so that others don't get confused */
5501 lod_striping_free(env, lo);
5507 * Whether subdirectories under \a dt should be created on MDTs by space QoS
5509 * If LMV_HASH_FLAG_SPACE is set on directory default layout, its subdirectories
5510 * should be created on MDT by space QoS.
5512 * \param[in] env execution environment
5513 * \param[in] dev lu device
5514 * \param[in] dt object
5516 * \retval 1 if directory should create subdir by space usage
5518 * \retval -ev if failed
5520 static inline int dt_object_qos_mkdir(const struct lu_env *env,
5521 struct lu_device *dev,
5522 struct dt_object *dt)
5524 struct lod_thread_info *info = lod_env_info(env);
5525 struct lu_object *obj;
5526 struct lod_object *lo;
5527 struct lmv_user_md *lmu;
5530 obj = lu_object_find_slice(env, dev, lu_object_fid(&dt->do_lu), NULL);
5532 return PTR_ERR(obj);
5534 lo = lu2lod_obj(obj);
5536 rc = lod_get_default_lmv_ea(env, lo);
5537 dt_object_put(env, dt);
5541 if (rc < (int)sizeof(*lmu))
5544 lmu = info->lti_ea_store;
5545 return le32_to_cpu(lmu->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
5549 * Implementation of dt_object_operations::do_declare_create.
5551 * The method declares creation of a new object. If the object will be striped,
5552 * then helper functions are called to find FIDs for the stripes, declare
5553 * creation of the stripes and declare initialization of the striping
5554 * information to be stored in the master object.
5556 * \see dt_object_operations::do_declare_create() in the API description
5559 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
5560 struct lu_attr *attr,
5561 struct dt_allocation_hint *hint,
5562 struct dt_object_format *dof, struct thandle *th)
5564 struct dt_object *next = dt_object_child(dt);
5565 struct lod_object *lo = lod_dt_obj(dt);
5574 * first of all, we declare creation of local object
5576 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
5581 * it's lod_ah_init() that has decided the object will be striped
5583 if (dof->dof_type == DFT_REGULAR) {
5584 /* callers don't want stripes */
5585 /* XXX: all tricky interactions with ->ah_make_hint() decided
5586 * to use striping, then ->declare_create() behaving differently
5587 * should be cleaned */
5588 if (dof->u.dof_reg.striped != 0)
5589 rc = lod_declare_striped_create(env, dt, attr,
5591 } else if (dof->dof_type == DFT_DIR) {
5592 struct seq_server_site *ss;
5593 struct lu_buf buf = { NULL };
5594 struct lu_buf *lmu = NULL;
5596 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
5598 /* If the parent has default stripeEA, and client
5599 * did not find it before sending create request,
5600 * then MDT will return -EREMOTE, and client will
5601 * retrieve the default stripeEA and re-create the
5604 * Note: if dah_eadata != NULL, it means creating the
5605 * striped directory with specified stripeEA, then it
5606 * should ignore the default stripeEA */
5607 if (hint != NULL && hint->dah_eadata == NULL) {
5608 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
5609 GOTO(out, rc = -EREMOTE);
5611 if (lo->ldo_dir_stripe_offset == LMV_OFFSET_DEFAULT) {
5612 struct lod_default_striping *lds;
5614 lds = lo->ldo_def_striping;
5616 * child and parent should be on the same MDT,
5617 * but if parent has default LMV, and the start
5618 * MDT offset is -1, it's allowed. This check
5619 * is not necessary after 2.12.22 because client
5620 * follows this already, but old client may not.
5622 if (hint->dah_parent &&
5623 dt_object_remote(hint->dah_parent) && lds &&
5624 lds->lds_dir_def_stripe_offset !=
5626 GOTO(out, rc = -EREMOTE);
5627 } else if (lo->ldo_dir_stripe_offset !=
5629 struct lod_device *lod;
5630 struct lu_tgt_desc *mdt = NULL;
5631 bool found_mdt = false;
5633 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5634 lod_foreach_mdt(lod, mdt) {
5635 if (mdt->ltd_index ==
5636 lo->ldo_dir_stripe_offset) {
5642 /* If the MDT indicated by stripe_offset can be
5643 * found, then tell client to resend the create
5644 * request to the correct MDT, otherwise return
5645 * error to client */
5647 GOTO(out, rc = -EREMOTE);
5649 GOTO(out, rc = -EINVAL);
5651 } else if (hint && hint->dah_eadata) {
5653 lmu->lb_buf = (void *)hint->dah_eadata;
5654 lmu->lb_len = hint->dah_eadata_len;
5657 rc = lod_declare_dir_striping_create(env, dt, attr, lmu, dof,
5661 /* failed to create striping or to set initial size, let's reset
5662 * config so that others don't get confused */
5664 lod_striping_free(env, lo);
5669 * Generate component ID for new created component.
5671 * \param[in] lo LOD object
5672 * \param[in] comp_idx index of ldo_comp_entries
5674 * \retval component ID on success
5675 * \retval LCME_ID_INVAL on failure
5677 static __u32 lod_gen_component_id(struct lod_object *lo,
5678 int mirror_id, int comp_idx)
5680 struct lod_layout_component *lod_comp;
5681 __u32 id, start, end;
5684 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
5686 lod_obj_inc_layout_gen(lo);
5687 id = lo->ldo_layout_gen;
5688 if (likely(id <= SEQ_ID_MAX))
5689 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
5691 /* Layout generation wraps, need to check collisions. */
5692 start = id & SEQ_ID_MASK;
5695 for (id = start; id <= end; id++) {
5696 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5697 lod_comp = &lo->ldo_comp_entries[i];
5698 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
5701 /* Found the ununsed ID */
5702 if (i == lo->ldo_comp_cnt)
5703 RETURN(pflr_id(mirror_id, id));
5705 if (end == LCME_ID_MAX) {
5707 end = min(lo->ldo_layout_gen & LCME_ID_MASK,
5708 (__u32)(LCME_ID_MAX - 1));
5712 RETURN(LCME_ID_INVAL);
5716 * Creation of a striped regular object.
5718 * The function is called to create the stripe objects for a regular
5719 * striped file. This can happen at the initial object creation or
5720 * when the caller asks LOD to do so using ->do_xattr_set() method
5721 * (so called late striping). Notice all the information are already
5722 * prepared in the form of the list of objects (ldo_stripe field).
5723 * This is done during declare phase.
5725 * \param[in] env execution environment
5726 * \param[in] dt object
5727 * \param[in] attr attributes the stripes will be created with
5728 * \param[in] dof format of stripes (see OSD API description)
5729 * \param[in] th transaction handle
5731 * \retval 0 on success
5732 * \retval negative if failed
5734 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
5735 struct lu_attr *attr, struct dt_object_format *dof,
5738 struct lod_layout_component *lod_comp;
5739 struct lod_object *lo = lod_dt_obj(dt);
5744 LASSERT((lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL) ||
5745 lo->ldo_is_foreign);
5747 mirror_id = 0; /* non-flr file's mirror_id is 0 */
5748 if (lo->ldo_mirror_count > 1) {
5749 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5750 lod_comp = &lo->ldo_comp_entries[i];
5751 if (lod_comp->llc_id != LCME_ID_INVAL &&
5752 mirror_id_of(lod_comp->llc_id) > mirror_id)
5753 mirror_id = mirror_id_of(lod_comp->llc_id);
5757 /* create all underlying objects */
5758 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5759 lod_comp = &lo->ldo_comp_entries[i];
5761 if (lod_comp->llc_id == LCME_ID_INVAL) {
5762 /* only the component of FLR layout with more than 1
5763 * mirror has mirror ID in its component ID.
5765 if (lod_comp->llc_extent.e_start == 0 &&
5766 lo->ldo_mirror_count > 1)
5769 lod_comp->llc_id = lod_gen_component_id(lo,
5771 if (lod_comp->llc_id == LCME_ID_INVAL)
5772 GOTO(out, rc = -ERANGE);
5775 if (lod_comp_inited(lod_comp))
5778 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
5779 lod_comp_set_init(lod_comp);
5781 if (lov_pattern(lod_comp->llc_pattern) == LOV_PATTERN_MDT)
5782 lod_comp_set_init(lod_comp);
5784 if (lod_comp->llc_stripe == NULL)
5787 LASSERT(lod_comp->llc_stripe_count);
5788 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
5789 struct dt_object *object = lod_comp->llc_stripe[j];
5790 LASSERT(object != NULL);
5791 rc = lod_sub_create(env, object, attr, NULL, dof, th);
5795 lod_comp_set_init(lod_comp);
5798 rc = lod_fill_mirrors(lo);
5802 rc = lod_generate_and_set_lovea(env, lo, th);
5806 lo->ldo_comp_cached = 1;
5810 lod_striping_free(env, lo);
5814 static inline bool lod_obj_is_dom(struct dt_object *dt)
5816 struct lod_object *lo = lod_dt_obj(dt);
5818 if (!dt_object_exists(dt_object_child(dt)))
5821 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
5824 if (!lo->ldo_comp_cnt)
5827 return (lov_pattern(lo->ldo_comp_entries[0].llc_pattern) ==
5832 * Implementation of dt_object_operations::do_create.
5834 * If any of preceeding methods (like ->do_declare_create(),
5835 * ->do_ah_init(), etc) chose to create a striped object,
5836 * then this method will create the master and the stripes.
5838 * \see dt_object_operations::do_create() in the API description for details.
5840 static int lod_create(const struct lu_env *env, struct dt_object *dt,
5841 struct lu_attr *attr, struct dt_allocation_hint *hint,
5842 struct dt_object_format *dof, struct thandle *th)
5847 /* create local object */
5848 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
5852 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
5853 (lod_obj_is_striped(dt) || lod_obj_is_dom(dt)) &&
5854 dof->u.dof_reg.striped != 0) {
5855 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
5856 rc = lod_striped_create(env, dt, attr, dof, th);
5863 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
5864 struct dt_object *dt, struct thandle *th,
5865 int comp_idx, int stripe_idx,
5866 struct lod_obj_stripe_cb_data *data)
5868 if (data->locd_declare)
5869 return lod_sub_declare_destroy(env, dt, th);
5870 else if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
5871 stripe_idx == cfs_fail_val)
5872 return lod_sub_destroy(env, dt, th);
5878 * Implementation of dt_object_operations::do_declare_destroy.
5880 * If the object is a striped directory, then the function declares reference
5881 * removal from the master object (this is an index) to the stripes and declares
5882 * destroy of all the stripes. In all the cases, it declares an intention to
5883 * destroy the object itself.
5885 * \see dt_object_operations::do_declare_destroy() in the API description
5888 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
5891 struct dt_object *next = dt_object_child(dt);
5892 struct lod_object *lo = lod_dt_obj(dt);
5893 struct lod_thread_info *info = lod_env_info(env);
5894 struct dt_object *stripe;
5895 char *stripe_name = info->lti_key;
5901 * load striping information, notice we don't do this when object
5902 * is being initialized as we don't need this information till
5903 * few specific cases like destroy, chown
5905 rc = lod_striping_load(env, lo);
5909 /* declare destroy for all underlying objects */
5910 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5911 rc = next->do_ops->do_index_try(env, next,
5912 &dt_directory_features);
5916 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5917 stripe = lo->ldo_stripe[i];
5921 rc = lod_sub_declare_ref_del(env, next, th);
5925 snprintf(stripe_name, sizeof(info->lti_key),
5927 PFID(lu_object_fid(&stripe->do_lu)), i);
5928 rc = lod_sub_declare_delete(env, next,
5929 (const struct dt_key *)stripe_name, th);
5936 * we declare destroy for the local object
5938 rc = lod_sub_declare_destroy(env, next, th);
5942 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
5943 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
5946 if (!lod_obj_is_striped(dt))
5949 /* declare destroy all striped objects */
5950 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5951 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5952 stripe = lo->ldo_stripe[i];
5956 if (!dt_object_exists(stripe))
5959 rc = lod_sub_declare_ref_del(env, stripe, th);
5963 rc = lod_sub_declare_destroy(env, stripe, th);
5968 struct lod_obj_stripe_cb_data data = { { 0 } };
5970 data.locd_declare = true;
5971 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
5972 rc = lod_obj_for_each_stripe(env, lo, th, &data);
5979 * Implementation of dt_object_operations::do_destroy.
5981 * If the object is a striped directory, then the function removes references
5982 * from the master object (this is an index) to the stripes and destroys all
5983 * the stripes. In all the cases, the function destroys the object itself.
5985 * \see dt_object_operations::do_destroy() in the API description for details.
5987 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
5990 struct dt_object *next = dt_object_child(dt);
5991 struct lod_object *lo = lod_dt_obj(dt);
5992 struct lod_thread_info *info = lod_env_info(env);
5993 char *stripe_name = info->lti_key;
5994 struct dt_object *stripe;
6000 /* destroy sub-stripe of master object */
6001 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6002 rc = next->do_ops->do_index_try(env, next,
6003 &dt_directory_features);
6007 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6008 stripe = lo->ldo_stripe[i];
6012 rc = lod_sub_ref_del(env, next, th);
6016 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
6017 PFID(lu_object_fid(&stripe->do_lu)), i);
6019 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
6020 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
6021 PFID(lu_object_fid(&stripe->do_lu)));
6023 rc = lod_sub_delete(env, next,
6024 (const struct dt_key *)stripe_name, th);
6030 rc = lod_sub_destroy(env, next, th);
6034 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
6035 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
6038 if (!lod_obj_is_striped(dt))
6041 /* destroy all striped objects */
6042 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6043 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6044 stripe = lo->ldo_stripe[i];
6048 if (!dt_object_exists(stripe))
6051 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
6052 i == cfs_fail_val) {
6053 dt_write_lock(env, stripe, DT_TGT_CHILD);
6054 rc = lod_sub_ref_del(env, stripe, th);
6055 dt_write_unlock(env, stripe);
6059 rc = lod_sub_destroy(env, stripe, th);
6065 struct lod_obj_stripe_cb_data data = { { 0 } };
6067 data.locd_declare = false;
6068 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
6069 rc = lod_obj_for_each_stripe(env, lo, th, &data);
6076 * Implementation of dt_object_operations::do_declare_ref_add.
6078 * \see dt_object_operations::do_declare_ref_add() in the API description
6081 static int lod_declare_ref_add(const struct lu_env *env,
6082 struct dt_object *dt, struct thandle *th)
6084 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
6088 * Implementation of dt_object_operations::do_ref_add.
6090 * \see dt_object_operations::do_ref_add() in the API description for details.
6092 static int lod_ref_add(const struct lu_env *env,
6093 struct dt_object *dt, struct thandle *th)
6095 return lod_sub_ref_add(env, dt_object_child(dt), th);
6099 * Implementation of dt_object_operations::do_declare_ref_del.
6101 * \see dt_object_operations::do_declare_ref_del() in the API description
6104 static int lod_declare_ref_del(const struct lu_env *env,
6105 struct dt_object *dt, struct thandle *th)
6107 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
6111 * Implementation of dt_object_operations::do_ref_del
6113 * \see dt_object_operations::do_ref_del() in the API description for details.
6115 static int lod_ref_del(const struct lu_env *env,
6116 struct dt_object *dt, struct thandle *th)
6118 return lod_sub_ref_del(env, dt_object_child(dt), th);
6122 * Implementation of dt_object_operations::do_object_sync.
6124 * \see dt_object_operations::do_object_sync() in the API description
6127 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
6128 __u64 start, __u64 end)
6130 return dt_object_sync(env, dt_object_child(dt), start, end);
6134 * Implementation of dt_object_operations::do_object_unlock.
6136 * Used to release LDLM lock(s).
6138 * \see dt_object_operations::do_object_unlock() in the API description
6141 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
6142 struct ldlm_enqueue_info *einfo,
6143 union ldlm_policy_data *policy)
6145 struct lod_object *lo = lod_dt_obj(dt);
6146 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
6147 int slave_locks_size;
6151 if (slave_locks == NULL)
6154 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
6155 /* Note: for remote lock for single stripe dir, MDT will cancel
6156 * the lock by lockh directly */
6157 LASSERT(!dt_object_remote(dt_object_child(dt)));
6159 /* locks were unlocked in MDT layer */
6160 for (i = 0; i < slave_locks->ha_count; i++)
6161 LASSERT(!lustre_handle_is_used(&slave_locks->ha_handles[i]));
6164 * NB, ha_count may not equal to ldo_dir_stripe_count, because dir
6165 * layout may change, e.g., shrink dir layout after migration.
6167 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6168 if (lo->ldo_stripe[i])
6169 dt_invalidate(env, lo->ldo_stripe[i]);
6172 slave_locks_size = offsetof(typeof(*slave_locks),
6173 ha_handles[slave_locks->ha_count]);
6174 OBD_FREE(slave_locks, slave_locks_size);
6175 einfo->ei_cbdata = NULL;
6181 * Implementation of dt_object_operations::do_object_lock.
6183 * Used to get LDLM lock on the non-striped and striped objects.
6185 * \see dt_object_operations::do_object_lock() in the API description
6188 static int lod_object_lock(const struct lu_env *env,
6189 struct dt_object *dt,
6190 struct lustre_handle *lh,
6191 struct ldlm_enqueue_info *einfo,
6192 union ldlm_policy_data *policy)
6194 struct lod_object *lo = lod_dt_obj(dt);
6195 int slave_locks_size;
6196 struct lustre_handle_array *slave_locks = NULL;
6201 /* remote object lock */
6202 if (!einfo->ei_enq_slave) {
6203 LASSERT(dt_object_remote(dt));
6204 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
6208 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
6211 rc = lod_striping_load(env, lo);
6216 if (lo->ldo_dir_stripe_count <= 1)
6219 slave_locks_size = offsetof(typeof(*slave_locks),
6220 ha_handles[lo->ldo_dir_stripe_count]);
6221 /* Freed in lod_object_unlock */
6222 OBD_ALLOC(slave_locks, slave_locks_size);
6225 slave_locks->ha_count = lo->ldo_dir_stripe_count;
6227 /* striped directory lock */
6228 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6229 struct lustre_handle lockh;
6230 struct ldlm_res_id *res_id;
6231 struct dt_object *stripe;
6233 stripe = lo->ldo_stripe[i];
6237 res_id = &lod_env_info(env)->lti_res_id;
6238 fid_build_reg_res_name(lu_object_fid(&stripe->do_lu), res_id);
6239 einfo->ei_res_id = res_id;
6241 if (dt_object_remote(stripe)) {
6242 set_bit(i, (void *)slave_locks->ha_map);
6243 rc = dt_object_lock(env, stripe, &lockh, einfo, policy);
6245 struct ldlm_namespace *ns = einfo->ei_namespace;
6246 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
6247 ldlm_completion_callback completion = einfo->ei_cb_cp;
6248 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
6250 if (einfo->ei_mode == LCK_PW ||
6251 einfo->ei_mode == LCK_EX)
6252 dlmflags |= LDLM_FL_COS_INCOMPAT;
6254 LASSERT(ns != NULL);
6255 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS,
6256 policy, einfo->ei_mode,
6257 &dlmflags, blocking,
6259 NULL, 0, LVB_T_NONE,
6264 ldlm_lock_decref_and_cancel(
6265 &slave_locks->ha_handles[i],
6267 OBD_FREE(slave_locks, slave_locks_size);
6270 slave_locks->ha_handles[i] = lockh;
6272 einfo->ei_cbdata = slave_locks;
6278 * Implementation of dt_object_operations::do_invalidate.
6280 * \see dt_object_operations::do_invalidate() in the API description for details
6282 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
6284 return dt_invalidate(env, dt_object_child(dt));
6287 static int lod_declare_instantiate_components(const struct lu_env *env,
6288 struct lod_object *lo,
6292 struct lod_thread_info *info = lod_env_info(env);
6297 LASSERT(info->lti_count < lo->ldo_comp_cnt);
6299 for (i = 0; i < info->lti_count; i++) {
6300 rc = lod_qos_prep_create(env, lo, NULL, th,
6301 info->lti_comp_idx[i], reserve);
6307 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6308 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
6309 &info->lti_buf, XATTR_NAME_LOV, 0, th);
6316 * Check OSTs for an existing component for further extension
6318 * Checks if OSTs are still healthy and not out of space. Gets free space
6319 * on OSTs (relative to allocation watermark rmb_low) and compares to
6320 * the proposed new_end for this component.
6322 * Decides whether or not to extend a component on its current OSTs.
6324 * \param[in] env execution environment for this thread
6325 * \param[in] lo object we're checking
6326 * \param[in] index index of this component
6327 * \param[in] extension_size extension size for this component
6328 * \param[in] extent layout extent for requested operation
6329 * \param[in] comp_extent extension component extent
6330 * \param[in] write if this is write operation
6332 * \retval true - OK to extend on current OSTs
6333 * \retval false - do not extend on current OSTs
6335 static bool lod_sel_osts_allowed(const struct lu_env *env,
6336 struct lod_object *lo,
6337 int index, __u64 reserve,
6338 struct lu_extent *extent,
6339 struct lu_extent *comp_extent, int write)
6341 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[index];
6342 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6343 struct lod_thread_info *tinfo = lod_env_info(env);
6344 struct obd_statfs *sfs = &tinfo->lti_osfs;
6345 __u64 available = 0;
6351 LASSERT(lod_comp->llc_stripe_count != 0);
6353 lod_getref(&lod->lod_ost_descs);
6354 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
6355 int index = lod_comp->llc_ost_indices[i];
6356 struct lod_tgt_desc *ost = OST_TGT(lod, index);
6357 struct obd_statfs_info info = { 0 };
6358 int j, repeated = 0;
6362 /* Get the number of times this OST repeats in this component.
6363 * Note: inter-component repeats are not counted as this is
6364 * considered as a rare case: we try to not repeat OST in other
6365 * components if possible. */
6366 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6367 if (index != lod_comp->llc_ost_indices[j])
6370 /* already handled */
6376 if (j < lod_comp->llc_stripe_count)
6379 if (!test_bit(index, lod->lod_ost_bitmap)) {
6380 CDEBUG(D_LAYOUT, "ost %d no longer present\n", index);
6385 rc = dt_statfs_info(env, ost->ltd_tgt, sfs, &info);
6387 CDEBUG(D_LAYOUT, "statfs failed for ost %d, error %d\n",
6393 if (sfs->os_state & OS_STATFS_ENOSPC ||
6394 sfs->os_state & OS_STATFS_READONLY ||
6395 sfs->os_state & OS_STATFS_DEGRADED) {
6396 CDEBUG(D_LAYOUT, "ost %d is not availble for SEL "
6397 "extension, state %u\n", index, sfs->os_state);
6403 available = sfs->os_bavail * sfs->os_bsize;
6404 /* 'available' is relative to the allocation threshold */
6405 available -= (__u64) info.os_reserved_mb_low << 20;
6407 CDEBUG(D_LAYOUT, "ost %d lowwm: %d highwm: %d, "
6408 "%llu %% blocks available, %llu %% blocks free\n",
6409 index, info.os_reserved_mb_low, info.os_reserved_mb_high,
6410 (100ull * sfs->os_bavail) / sfs->os_blocks,
6411 (100ull * sfs->os_bfree) / sfs->os_blocks);
6413 if (reserve * repeated > available) {
6415 CDEBUG(D_LAYOUT, "low space on ost %d, available %llu "
6416 "< extension size %llu repeated %d\n", index,
6417 available, reserve, repeated);
6421 lod_putref(lod, &lod->lod_ost_descs);
6427 * Adjust extents after component removal
6429 * When we remove an extension component, we move the start of the next
6430 * component to match the start of the extension component, so no space is left
6433 * \param[in] env execution environment for this thread
6434 * \param[in] lo object
6435 * \param[in] max_comp layout component
6436 * \param[in] index index of this component
6438 * \retval 0 on success
6439 * \retval negative errno on error
6441 static void lod_sel_adjust_extents(const struct lu_env *env,
6442 struct lod_object *lo,
6443 int max_comp, int index)
6445 struct lod_layout_component *lod_comp = NULL;
6446 struct lod_layout_component *next = NULL;
6447 struct lod_layout_component *prev = NULL;
6448 __u64 new_start = 0;
6452 /* Extension space component */
6453 lod_comp = &lo->ldo_comp_entries[index];
6454 next = &lo->ldo_comp_entries[index + 1];
6455 prev = &lo->ldo_comp_entries[index - 1];
6457 LASSERT(lod_comp != NULL && prev != NULL && next != NULL);
6458 LASSERT(lod_comp->llc_flags & LCME_FL_EXTENSION);
6460 /* Previous is being removed */
6461 if (prev && prev->llc_id == LCME_ID_INVAL)
6462 new_start = prev->llc_extent.e_start;
6464 new_start = lod_comp->llc_extent.e_start;
6466 for (i = index + 1; i < max_comp; i++) {
6467 lod_comp = &lo->ldo_comp_entries[i];
6469 start = lod_comp->llc_extent.e_start;
6470 lod_comp->llc_extent.e_start = new_start;
6472 /* We only move zero length extendable components */
6473 if (!(start == lod_comp->llc_extent.e_end))
6476 LASSERT(!(lod_comp->llc_flags & LCME_FL_INIT));
6478 lod_comp->llc_extent.e_end = new_start;
6482 /* Calculate the proposed 'new end' for a component we're extending */
6483 static __u64 lod_extension_new_end(__u64 extension_size, __u64 extent_end,
6484 __u32 stripe_size, __u64 component_end,
6485 __u64 extension_end)
6489 LASSERT(extension_size != 0 && stripe_size != 0);
6491 /* Round up to extension size */
6492 if (extent_end == OBD_OBJECT_EOF) {
6493 new_end = OBD_OBJECT_EOF;
6495 /* Add at least extension_size to the previous component_end,
6496 * covering the req layout extent */
6497 new_end = max(extent_end - component_end, extension_size);
6498 new_end = roundup(new_end, extension_size);
6499 new_end += component_end;
6501 /* Component end must be min stripe size aligned */
6502 if (new_end % stripe_size) {
6503 CDEBUG(D_LAYOUT, "new component end is not aligned "
6504 "by the stripe size %u: [%llu, %llu) ext size "
6505 "%llu new end %llu, aligning\n",
6506 stripe_size, component_end, extent_end,
6507 extension_size, new_end);
6508 new_end = roundup(new_end, stripe_size);
6512 if (new_end < extent_end)
6513 new_end = OBD_OBJECT_EOF;
6516 /* Don't extend past the end of the extension component */
6517 if (new_end > extension_end)
6518 new_end = extension_end;
6524 * Calculate the exact reservation (per-OST extension_size) on the OSTs being
6525 * instantiated. It needs to be calculated in advance and taken into account at
6526 * the instantiation time, because otherwise lod_statfs_and_check() may consider
6527 * an OST as OK, but SEL needs its extension_size to fit the free space and the
6528 * OST may turn out to be low-on-space, thus inappropriate OST may be used and
6531 * \param[in] lod_comp lod component we are checking
6533 * \retval size to reserved on each OST of lod_comp's stripe.
6535 static __u64 lod_sel_stripe_reserved(struct lod_layout_component *lod_comp)
6537 /* extension_size is file level, so we must divide by stripe count to
6538 * compare it to available space on a single OST */
6539 return lod_comp->llc_stripe_size * SEL_UNIT_SIZE /
6540 lod_comp->llc_stripe_count;
6543 /* As lod_sel_handler() could be re-entered for the same component several
6544 * times, this is the data for the next call. Fields could be changed to
6545 * component indexes when needed, (e.g. if there is no need to instantiate
6546 * all the previous components up to the current position) to tell the caller
6547 * where to start over from. */
6554 * Process extent updates for a particular layout component
6556 * Handle layout updates for a particular extension space component touched by
6557 * a layout update operation. Core function of self-extending PFL feature.
6559 * In general, this function processes exactly *one* stage of an extension
6560 * operation, modifying the layout accordingly, then returns to the caller.
6561 * The caller is responsible for restarting processing with the new layout,
6562 * which may repeatedly return to this function until the extension updates
6565 * This function does one of a few things to the layout:
6566 * 1. Extends the component before the current extension space component to
6567 * allow it to accomodate the requested operation (if space/policy permit that
6568 * component to continue on its current OSTs)
6570 * 2. If extension of the existing component fails, we do one of two things:
6571 * a. If there is a component after the extension space, we remove the
6572 * extension space component, move the start of the next component down
6573 * accordingly, then notify the caller to restart processing w/the new
6575 * b. If there is no following component, we try repeating the current
6576 * component, creating a new component using the current one as a
6577 * template (keeping its stripe properties but not specific striping),
6578 * and try assigning striping for this component. If there is sufficient
6579 * free space on the OSTs chosen for this component, it is instantiated
6580 * and i/o continues there.
6582 * If there is not sufficient space on the new OSTs, we remove this new
6583 * component & extend the current component.
6585 * Note further that uninited components followed by extension space can be zero
6586 * length meaning that we will try to extend them before initializing them, and
6587 * if that fails, they will be removed without initialization.
6589 * 3. If we extend to/beyond the end of an extension space component, that
6590 * component is exhausted (all of its range has been given to real components),
6591 * so we remove it and restart processing.
6593 * \param[in] env execution environment for this thread
6594 * \param[in,out] lo object to update the layout of
6595 * \param[in] extent layout extent for requested operation, update
6596 * layout to fit this operation
6597 * \param[in] th transaction handle for this operation
6598 * \param[in,out] max_comp the highest comp for the portion of the layout
6599 * we are operating on (For FLR, the chosen
6600 * replica). Updated because we may remove
6602 * \param[in] index index of the extension space component we're
6604 * \param[in] write if this is write op
6605 * \param[in,out] force if the extension is to be forced; set here
6606 to force it on the 2nd call for the same
6609 * \retval 0 on success
6610 * \retval negative errno on error
6612 static int lod_sel_handler(const struct lu_env *env,
6613 struct lod_object *lo,
6614 struct lu_extent *extent,
6615 struct thandle *th, int *max_comp,
6616 int index, int write,
6617 struct sel_data *sd)
6619 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6620 struct lod_thread_info *info = lod_env_info(env);
6621 struct lod_layout_component *lod_comp;
6622 struct lod_layout_component *prev;
6623 struct lod_layout_component *next = NULL;
6624 __u64 extension_size, reserve;
6631 /* First component cannot be extension space */
6633 CERROR("%s: "DFID" first component cannot be extension space\n",
6634 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
6638 lod_comp = &lo->ldo_comp_entries[index];
6639 prev = &lo->ldo_comp_entries[index - 1];
6640 if ((index + 1) < *max_comp)
6641 next = &lo->ldo_comp_entries[index + 1];
6643 /* extension size uses the stripe size field as KiB */
6644 extension_size = lod_comp->llc_stripe_size * SEL_UNIT_SIZE;
6646 CDEBUG(D_LAYOUT, "prev start %llu, extension start %llu, extension end"
6647 " %llu, extension size %llu\n", prev->llc_extent.e_start,
6648 lod_comp->llc_extent.e_start, lod_comp->llc_extent.e_end,
6651 /* Two extension space components cannot be adjacent & extension space
6652 * components cannot be init */
6653 if ((prev->llc_flags & LCME_FL_EXTENSION) ||
6654 !(ergo(next, !(next->llc_flags & LCME_FL_EXTENSION))) ||
6655 lod_comp_inited(lod_comp)) {
6656 CERROR("%s: "DFID" invalid extension space components\n",
6657 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
6661 reserve = lod_sel_stripe_reserved(lod_comp);
6663 if (!prev->llc_stripe) {
6664 CDEBUG(D_LAYOUT, "Previous component not inited\n");
6665 info->lti_count = 1;
6666 info->lti_comp_idx[0] = index - 1;
6667 rc = lod_declare_instantiate_components(env, lo, th, reserve);
6668 /* ENOSPC tells us we can't use this component. If there is
6669 * a next or we are repeating, we either spill over (next) or
6670 * extend the original comp (repeat). Otherwise, return the
6671 * error to the user. */
6672 if (rc == -ENOSPC && (next || sd->sd_repeat))
6678 if (sd->sd_force == 0 && rc == 0)
6679 rc = !lod_sel_osts_allowed(env, lo, index - 1, reserve, extent,
6680 &lod_comp->llc_extent, write);
6682 repeated = !!(sd->sd_repeat);
6686 /* Extend previous component */
6688 new_end = lod_extension_new_end(extension_size, extent->e_end,
6689 prev->llc_stripe_size,
6690 prev->llc_extent.e_end,
6691 lod_comp->llc_extent.e_end);
6693 CDEBUG(D_LAYOUT, "new end %llu\n", new_end);
6694 lod_comp->llc_extent.e_start = new_end;
6695 prev->llc_extent.e_end = new_end;
6697 if (prev->llc_extent.e_end == lod_comp->llc_extent.e_end) {
6698 CDEBUG(D_LAYOUT, "Extension component exhausted\n");
6699 lod_comp->llc_id = LCME_ID_INVAL;
6703 /* rc == 1, failed to extend current component */
6706 /* Normal 'spillover' case - Remove the extension
6707 * space component & bring down the start of the next
6709 lod_comp->llc_id = LCME_ID_INVAL;
6711 if (!(prev->llc_flags & LCME_FL_INIT)) {
6712 prev->llc_id = LCME_ID_INVAL;
6715 lod_sel_adjust_extents(env, lo, *max_comp, index);
6716 } else if (lod_comp_inited(prev)) {
6717 /* If there is no next, and the previous component is
6718 * INIT'ed, try repeating the previous component. */
6719 LASSERT(repeated == 0);
6720 rc = lod_layout_repeat_comp(env, lo, index - 1);
6724 /* The previous component is a repeated component.
6725 * Record this so we don't keep trying to repeat it. */
6728 /* If the previous component is not INIT'ed, this may
6729 * be a component we have just instantiated but failed
6730 * to extend. Or even a repeated component we failed
6731 * to prepare a striping for. Do not repeat but instead
6732 * remove the repeated component & force the extention
6733 * of the original one */
6736 prev->llc_id = LCME_ID_INVAL;
6743 rc = lod_layout_del_prep_layout(env, lo, NULL);
6746 LASSERTF(-rc == change,
6747 "number deleted %d != requested %d\n", -rc,
6750 *max_comp = *max_comp + change;
6752 /* lod_del_prep_layout reallocates ldo_comp_entries, so we must
6753 * refresh these pointers before using them */
6754 lod_comp = &lo->ldo_comp_entries[index];
6755 prev = &lo->ldo_comp_entries[index - 1];
6756 CDEBUG(D_LAYOUT, "After extent updates: prev start %llu, current start "
6757 "%llu, current end %llu max_comp %d ldo_comp_cnt %d\n",
6758 prev->llc_extent.e_start, lod_comp->llc_extent.e_start,
6759 lod_comp->llc_extent.e_end, *max_comp, lo->ldo_comp_cnt);
6761 /* Layout changed successfully */
6766 * Declare layout extent updates
6768 * Handles extensions. Identifies extension components touched by current
6769 * operation and passes them to processing function.
6771 * Restarts with updated layouts from the processing function until the current
6772 * operation no longer touches an extension space component.
6774 * \param[in] env execution environment for this thread
6775 * \param[in,out] lo object to update the layout of
6776 * \param[in] extent layout extent for requested operation, update layout to
6777 * fit this operation
6778 * \param[in] th transaction handle for this operation
6779 * \param[in] pick identifies chosen mirror for FLR layouts
6780 * \param[in] write if this is write op
6782 * \retval 1 on layout changed, 0 on no change
6783 * \retval negative errno on error
6785 static int lod_declare_update_extents(const struct lu_env *env,
6786 struct lod_object *lo, struct lu_extent *extent,
6787 struct thandle *th, int pick, int write)
6789 struct lod_thread_info *info = lod_env_info(env);
6790 struct lod_layout_component *lod_comp;
6791 bool layout_changed = false;
6792 struct sel_data sd = { 0 };
6800 /* This makes us work on the components of the chosen mirror */
6801 start_index = lo->ldo_mirrors[pick].lme_start;
6802 max_comp = lo->ldo_mirrors[pick].lme_end + 1;
6803 if (lo->ldo_flr_state == LCM_FL_NONE)
6804 LASSERT(start_index == 0 && max_comp == lo->ldo_comp_cnt);
6806 CDEBUG(D_LAYOUT, "extent->e_start %llu, extent->e_end %llu\n",
6807 extent->e_start, extent->e_end);
6808 for (i = start_index; i < max_comp; i++) {
6809 lod_comp = &lo->ldo_comp_entries[i];
6811 /* We've passed all components of interest */
6812 if (lod_comp->llc_extent.e_start >= extent->e_end)
6815 if (lod_comp->llc_flags & LCME_FL_EXTENSION) {
6816 layout_changed = true;
6817 rc = lod_sel_handler(env, lo, extent, th, &max_comp,
6822 /* Nothing has changed behind the prev one */
6828 /* We may have added or removed components. If so, we must update the
6829 * start & ends of all the mirrors after the current one, and the end
6830 * of the current mirror. */
6831 change = max_comp - 1 - lo->ldo_mirrors[pick].lme_end;
6833 lo->ldo_mirrors[pick].lme_end += change;
6834 for (i = pick + 1; i < lo->ldo_mirror_count; i++) {
6835 lo->ldo_mirrors[i].lme_start += change;
6836 lo->ldo_mirrors[i].lme_end += change;
6842 /* The amount of components has changed, adjust the lti_comp_idx */
6843 rc2 = lod_layout_data_init(info, lo->ldo_comp_cnt);
6845 return rc < 0 ? rc : rc2 < 0 ? rc2 : layout_changed;
6848 /* If striping is already instantiated or INIT'ed DOM? */
6849 static bool lod_is_instantiation_needed(struct lod_layout_component *comp)
6851 return !(((lov_pattern(comp->llc_pattern) == LOV_PATTERN_MDT) &&
6852 lod_comp_inited(comp)) || comp->llc_stripe);
6856 * Declare layout update for a non-FLR layout.
6858 * \param[in] env execution environment for this thread
6859 * \param[in,out] lo object to update the layout of
6860 * \param[in] layout layout intent for requested operation, "update" is
6861 * a process of reacting to this
6862 * \param[in] buf buffer containing lov ea (see comment on usage inline)
6863 * \param[in] th transaction handle for this operation
6865 * \retval 0 on success
6866 * \retval negative errno on error
6868 static int lod_declare_update_plain(const struct lu_env *env,
6869 struct lod_object *lo, struct layout_intent *layout,
6870 const struct lu_buf *buf, struct thandle *th)
6872 struct lod_thread_info *info = lod_env_info(env);
6873 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6874 struct lod_layout_component *lod_comp;
6875 struct lov_comp_md_v1 *comp_v1 = NULL;
6876 bool layout_changed = false;
6877 bool replay = false;
6881 LASSERT(lo->ldo_flr_state == LCM_FL_NONE);
6884 * In case the client is passing lovea, which only happens during
6885 * the replay of layout intent write RPC for now, we may need to
6886 * parse the lovea and apply new layout configuration.
6888 if (buf && buf->lb_len) {
6889 struct lov_user_md_v1 *v1 = buf->lb_buf;
6891 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
6892 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
6893 LOV_MAGIC_COMP_V1)) {
6894 CERROR("%s: the replay buffer of layout extend "
6895 "(magic %#x) does not contain expected "
6896 "composite layout.\n",
6897 lod2obd(d)->obd_name, v1->lmm_magic);
6898 GOTO(out, rc = -EINVAL);
6901 rc = lod_use_defined_striping(env, lo, buf);
6904 lo->ldo_comp_cached = 1;
6906 rc = lod_get_lov_ea(env, lo);
6909 /* old on-disk EA is stored in info->lti_buf */
6910 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
6912 layout_changed = true;
6914 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
6918 /* non replay path */
6919 rc = lod_striping_load(env, lo);
6924 /* Make sure defined layout covers the requested write range. */
6925 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
6926 if (lo->ldo_comp_cnt > 1 &&
6927 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
6928 lod_comp->llc_extent.e_end < layout->li_extent.e_end) {
6929 CDEBUG_LIMIT(replay ? D_ERROR : D_LAYOUT,
6930 "%s: the defined layout [0, %#llx) does not "
6931 "covers the write range "DEXT"\n",
6932 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
6933 PEXT(&layout->li_extent));
6934 GOTO(out, rc = -EINVAL);
6937 CDEBUG(D_LAYOUT, "%s: "DFID": update components "DEXT"\n",
6938 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
6939 PEXT(&layout->li_extent));
6942 rc = lod_declare_update_extents(env, lo, &layout->li_extent,
6943 th, 0, layout->li_opc == LAYOUT_INTENT_WRITE);
6947 layout_changed = true;
6951 * Iterate ld->ldo_comp_entries, find the component whose extent under
6952 * the write range and not instantianted.
6954 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6955 lod_comp = &lo->ldo_comp_entries[i];
6957 if (lod_comp->llc_extent.e_start >= layout->li_extent.e_end)
6961 /* If striping is instantiated or INIT'ed DOM skip */
6962 if (!lod_is_instantiation_needed(lod_comp))
6966 * In replay path, lod_comp is the EA passed by
6967 * client replay buffer, comp_v1 is the pre-recovery
6968 * on-disk EA, we'd sift out those components which
6969 * were init-ed in the on-disk EA.
6971 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
6976 * this component hasn't instantiated in normal path, or during
6977 * replay it needs replay the instantiation.
6980 /* A released component is being extended */
6981 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
6982 GOTO(out, rc = -EINVAL);
6984 LASSERT(info->lti_comp_idx != NULL);
6985 info->lti_comp_idx[info->lti_count++] = i;
6986 layout_changed = true;
6989 if (!layout_changed)
6992 lod_obj_inc_layout_gen(lo);
6993 rc = lod_declare_instantiate_components(env, lo, th, 0);
6997 lod_striping_free(env, lo);
7001 static inline int lod_comp_index(struct lod_object *lo,
7002 struct lod_layout_component *lod_comp)
7004 LASSERT(lod_comp >= lo->ldo_comp_entries &&
7005 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
7007 return lod_comp - lo->ldo_comp_entries;
7011 * Stale other mirrors by writing extent.
7013 static int lod_stale_components(const struct lu_env *env, struct lod_object *lo,
7014 int primary, struct lu_extent *extent,
7017 struct lod_layout_component *pri_comp, *lod_comp;
7018 struct lu_extent pri_extent;
7023 /* The writing extent decides which components in the primary
7024 * are affected... */
7025 CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
7028 lod_foreach_mirror_comp(pri_comp, lo, primary) {
7029 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
7032 CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
7033 lod_comp_index(lo, pri_comp),
7034 PEXT(&pri_comp->llc_extent));
7036 pri_extent.e_start = pri_comp->llc_extent.e_start;
7037 pri_extent.e_end = pri_comp->llc_extent.e_end;
7039 for (i = 0; i < lo->ldo_mirror_count; i++) {
7042 rc = lod_declare_update_extents(env, lo, &pri_extent,
7044 /* if update_extents changed the layout, it may have
7045 * reallocated the component array, so start over to
7046 * avoid using stale pointers */
7052 /* ... and then stale other components that are
7053 * overlapping with primary components */
7054 lod_foreach_mirror_comp(lod_comp, lo, i) {
7055 if (!lu_extent_is_overlapped(
7057 &lod_comp->llc_extent))
7060 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
7061 i, lod_comp_index(lo, lod_comp));
7063 lod_comp->llc_flags |= LCME_FL_STALE;
7064 lo->ldo_mirrors[i].lme_stale = 1;
7073 * check an OST's availability
7074 * \param[in] env execution environment
7075 * \param[in] lo lod object
7076 * \param[in] dt dt object
7077 * \param[in] index mirror index
7079 * \retval negative if failed
7080 * \retval 1 if \a dt is available
7081 * \retval 0 if \a dt is not available
7083 static inline int lod_check_ost_avail(const struct lu_env *env,
7084 struct lod_object *lo,
7085 struct dt_object *dt, int index)
7087 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7088 struct lod_tgt_desc *ost;
7090 int type = LU_SEQ_RANGE_OST;
7093 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &idx, &type);
7095 CERROR("%s: can't locate "DFID":rc = %d\n",
7096 lod2obd(lod)->obd_name, PFID(lu_object_fid(&dt->do_lu)),
7101 ost = OST_TGT(lod, idx);
7102 if (ost->ltd_statfs.os_state &
7103 (OS_STATFS_READONLY | OS_STATFS_ENOSPC | OS_STATFS_ENOINO |
7104 OS_STATFS_NOPRECREATE) ||
7105 ost->ltd_active == 0) {
7106 CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail, rc = %d\n",
7107 PFID(lod_object_fid(lo)), index, idx, rc);
7115 * Pick primary mirror for write
7116 * \param[in] env execution environment
7117 * \param[in] lo object
7118 * \param[in] extent write range
7120 static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo,
7121 struct lu_extent *extent)
7123 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7124 unsigned int seq = 0;
7125 struct lod_layout_component *lod_comp;
7127 int picked = -1, second_pick = -1, third_pick = -1;
7130 if (OBD_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
7131 get_random_bytes(&seq, sizeof(seq));
7132 seq %= lo->ldo_mirror_count;
7136 * Pick a mirror as the primary, and check the availability of OSTs.
7138 * This algo can be revised later after knowing the topology of
7141 lod_qos_statfs_update(env, lod, &lod->lod_ost_descs);
7142 for (i = 0; i < lo->ldo_mirror_count; i++) {
7143 bool ost_avail = true;
7144 int index = (i + seq) % lo->ldo_mirror_count;
7146 if (lo->ldo_mirrors[index].lme_stale) {
7147 CDEBUG(D_LAYOUT, DFID": mirror %d stale\n",
7148 PFID(lod_object_fid(lo)), index);
7152 /* 2nd pick is for the primary mirror containing unavail OST */
7153 if (lo->ldo_mirrors[index].lme_primary && second_pick < 0)
7154 second_pick = index;
7156 /* 3rd pick is for non-primary mirror containing unavail OST */
7157 if (second_pick < 0 && third_pick < 0)
7161 * we found a non-primary 1st pick, we'd like to find a
7162 * potential pirmary mirror.
7164 if (picked >= 0 && !lo->ldo_mirrors[index].lme_primary)
7167 /* check the availability of OSTs */
7168 lod_foreach_mirror_comp(lod_comp, lo, index) {
7169 if (!lod_comp_inited(lod_comp) || !lod_comp->llc_stripe)
7172 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
7173 struct dt_object *dt = lod_comp->llc_stripe[j];
7175 rc = lod_check_ost_avail(env, lo, dt, index);
7182 } /* for all dt object in one component */
7185 } /* for all components in a mirror */
7188 * the OSTs where allocated objects locates in the components
7189 * of the mirror are available.
7194 /* this mirror has all OSTs available */
7198 * primary with all OSTs are available, this is the perfect
7201 if (lo->ldo_mirrors[index].lme_primary)
7203 } /* for all mirrors */
7205 /* failed to pick a sound mirror, lower our expectation */
7207 picked = second_pick;
7209 picked = third_pick;
7216 static int lod_prepare_resync_mirror(const struct lu_env *env,
7217 struct lod_object *lo,
7220 struct lod_thread_info *info = lod_env_info(env);
7221 struct lod_layout_component *lod_comp;
7222 bool neg = !!(MIRROR_ID_NEG & mirror_id);
7225 mirror_id &= ~MIRROR_ID_NEG;
7227 for (i = 0; i < lo->ldo_mirror_count; i++) {
7228 if ((!neg && lo->ldo_mirrors[i].lme_id != mirror_id) ||
7229 (neg && lo->ldo_mirrors[i].lme_id == mirror_id))
7232 lod_foreach_mirror_comp(lod_comp, lo, i) {
7233 if (lod_comp_inited(lod_comp))
7236 info->lti_comp_idx[info->lti_count++] =
7237 lod_comp_index(lo, lod_comp);
7245 * figure out the components should be instantiated for resync.
7247 static int lod_prepare_resync(const struct lu_env *env, struct lod_object *lo,
7248 struct lu_extent *extent)
7250 struct lod_thread_info *info = lod_env_info(env);
7251 struct lod_layout_component *lod_comp;
7252 unsigned int need_sync = 0;
7256 DFID": instantiate all stale components in "DEXT"\n",
7257 PFID(lod_object_fid(lo)), PEXT(extent));
7260 * instantiate all components within this extent, even non-stale
7263 for (i = 0; i < lo->ldo_mirror_count; i++) {
7264 if (!lo->ldo_mirrors[i].lme_stale)
7267 lod_foreach_mirror_comp(lod_comp, lo, i) {
7268 if (!lu_extent_is_overlapped(extent,
7269 &lod_comp->llc_extent))
7274 if (lod_comp_inited(lod_comp))
7277 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
7278 i, lod_comp_index(lo, lod_comp));
7279 info->lti_comp_idx[info->lti_count++] =
7280 lod_comp_index(lo, lod_comp);
7284 return need_sync ? 0 : -EALREADY;
7287 static int lod_declare_update_rdonly(const struct lu_env *env,
7288 struct lod_object *lo, struct md_layout_change *mlc,
7291 struct lod_thread_info *info = lod_env_info(env);
7292 struct lu_attr *layout_attr = &info->lti_layout_attr;
7293 struct lod_layout_component *lod_comp;
7294 struct lu_extent extent = { 0 };
7298 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
7299 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
7300 mlc->mlc_opc == MD_LAYOUT_RESYNC);
7301 LASSERT(lo->ldo_mirror_count > 0);
7303 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7304 struct layout_intent *layout = mlc->mlc_intent;
7305 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
7308 extent = layout->li_extent;
7309 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
7310 PFID(lod_object_fid(lo)), PEXT(&extent));
7312 picked = lod_primary_pick(env, lo, &extent);
7316 CDEBUG(D_LAYOUT, DFID": picked mirror id %u as primary\n",
7317 PFID(lod_object_fid(lo)),
7318 lo->ldo_mirrors[picked].lme_id);
7320 /* Update extents of primary before staling */
7321 rc = lod_declare_update_extents(env, lo, &extent, th, picked,
7326 if (layout->li_opc == LAYOUT_INTENT_TRUNC) {
7328 * trunc transfers [0, size) in the intent extent, we'd
7329 * stale components overlapping [size, eof).
7331 extent.e_start = extent.e_end;
7332 extent.e_end = OBD_OBJECT_EOF;
7335 /* stale overlapping components from other mirrors */
7336 rc = lod_stale_components(env, lo, picked, &extent, th);
7340 /* restore truncate intent extent */
7341 if (layout->li_opc == LAYOUT_INTENT_TRUNC)
7342 extent.e_end = extent.e_start;
7344 /* instantiate components for the picked mirror, start from 0 */
7347 lod_foreach_mirror_comp(lod_comp, lo, picked) {
7348 if (!lu_extent_is_overlapped(&extent,
7349 &lod_comp->llc_extent))
7352 if (!lod_is_instantiation_needed(lod_comp))
7355 info->lti_comp_idx[info->lti_count++] =
7356 lod_comp_index(lo, lod_comp);
7359 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
7360 } else { /* MD_LAYOUT_RESYNC */
7364 * could contain multiple non-stale mirrors, so we need to
7365 * prep uninited all components assuming any non-stale mirror
7366 * could be picked as the primary mirror.
7368 if (mlc->mlc_mirror_id == 0) {
7370 for (i = 0; i < lo->ldo_mirror_count; i++) {
7371 if (lo->ldo_mirrors[i].lme_stale)
7374 lod_foreach_mirror_comp(lod_comp, lo, i) {
7375 if (!lod_comp_inited(lod_comp))
7379 lod_comp->llc_extent.e_end)
7381 lod_comp->llc_extent.e_end;
7384 rc = lod_prepare_resync(env, lo, &extent);
7388 /* mirror write, try to init its all components */
7389 rc = lod_prepare_resync_mirror(env, lo,
7390 mlc->mlc_mirror_id);
7395 /* change the file state to SYNC_PENDING */
7396 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
7399 /* Reset the layout version once it's becoming too large.
7400 * This way it can make sure that the layout version is
7401 * monotonously increased in this writing era. */
7402 lod_obj_inc_layout_gen(lo);
7403 if (lo->ldo_layout_gen > (LCME_ID_MAX >> 1)) {
7404 __u32 layout_version;
7406 get_random_bytes(&layout_version, sizeof(layout_version));
7407 lo->ldo_layout_gen = layout_version & 0xffff;
7410 rc = lod_declare_instantiate_components(env, lo, th, 0);
7414 layout_attr->la_valid = LA_LAYOUT_VERSION;
7415 layout_attr->la_layout_version = 0; /* set current version */
7416 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
7417 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
7418 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
7424 lod_striping_free(env, lo);
7428 static int lod_declare_update_write_pending(const struct lu_env *env,
7429 struct lod_object *lo, struct md_layout_change *mlc,
7432 struct lod_thread_info *info = lod_env_info(env);
7433 struct lu_attr *layout_attr = &info->lti_layout_attr;
7434 struct lod_layout_component *lod_comp;
7435 struct lu_extent extent = { 0 };
7441 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
7442 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
7443 mlc->mlc_opc == MD_LAYOUT_RESYNC);
7445 /* look for the primary mirror */
7446 for (i = 0; i < lo->ldo_mirror_count; i++) {
7447 if (lo->ldo_mirrors[i].lme_stale)
7449 if (lo->ldo_mirrors[i].lme_primary == 0)
7452 if (unlikely(primary >= 0)) {
7453 CERROR(DFID " has multiple primary: %u / %u\n",
7454 PFID(lod_object_fid(lo)),
7455 lo->ldo_mirrors[i].lme_id,
7456 lo->ldo_mirrors[primary].lme_id);
7463 /* no primary, use any in-sync */
7464 for (i = 0; i < lo->ldo_mirror_count; i++) {
7465 if (lo->ldo_mirrors[i].lme_stale)
7471 CERROR(DFID ": doesn't have a primary mirror\n",
7472 PFID(lod_object_fid(lo)));
7473 GOTO(out, rc = -ENODATA);
7477 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
7478 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
7480 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
7482 /* for LAYOUT_WRITE opc, it has to do the following operations:
7483 * 1. stale overlapping componets from stale mirrors;
7484 * 2. instantiate components of the primary mirror;
7485 * 3. transfter layout version to all objects of the primary;
7487 * for LAYOUT_RESYNC opc, it will do:
7488 * 1. instantiate components of all stale mirrors;
7489 * 2. transfer layout version to all objects to close write era. */
7491 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7492 struct layout_intent *layout = mlc->mlc_intent;
7493 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
7495 LASSERT(mlc->mlc_intent != NULL);
7497 extent = mlc->mlc_intent->li_extent;
7499 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
7500 PFID(lod_object_fid(lo)), PEXT(&extent));
7502 /* 1. Update extents of primary before staling */
7503 rc = lod_declare_update_extents(env, lo, &extent, th, primary,
7508 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC) {
7510 * trunc transfers [0, size) in the intent extent, we'd
7511 * stale components overlapping [size, eof).
7513 extent.e_start = extent.e_end;
7514 extent.e_end = OBD_OBJECT_EOF;
7517 /* 2. stale overlapping components */
7518 rc = lod_stale_components(env, lo, primary, &extent, th);
7522 /* 3. find the components which need instantiating.
7523 * instantiate [0, mlc->mlc_intent->e_end) */
7525 /* restore truncate intent extent */
7526 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC)
7527 extent.e_end = extent.e_start;
7530 lod_foreach_mirror_comp(lod_comp, lo, primary) {
7531 if (!lu_extent_is_overlapped(&extent,
7532 &lod_comp->llc_extent))
7535 if (!lod_is_instantiation_needed(lod_comp))
7538 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
7539 primary, lod_comp_index(lo, lod_comp));
7540 info->lti_comp_idx[info->lti_count++] =
7541 lod_comp_index(lo, lod_comp);
7543 } else { /* MD_LAYOUT_RESYNC */
7544 if (mlc->mlc_mirror_id == 0) {
7546 lod_foreach_mirror_comp(lod_comp, lo, primary) {
7547 if (!lod_comp_inited(lod_comp))
7550 extent.e_end = lod_comp->llc_extent.e_end;
7553 rc = lod_prepare_resync(env, lo, &extent);
7557 /* mirror write, try to init its all components */
7558 rc = lod_prepare_resync_mirror(env, lo,
7559 mlc->mlc_mirror_id);
7564 /* change the file state to SYNC_PENDING */
7565 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
7568 rc = lod_declare_instantiate_components(env, lo, th, 0);
7572 /* 3. transfer layout version to OST objects.
7573 * transfer new layout version to OST objects so that stale writes
7574 * can be denied. It also ends an era of writing by setting
7575 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
7576 * send write RPC; only resync RPCs could do it. */
7577 layout_attr->la_valid = LA_LAYOUT_VERSION;
7578 layout_attr->la_layout_version = 0; /* set current version */
7579 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
7580 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
7581 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
7585 lod_obj_inc_layout_gen(lo);
7588 lod_striping_free(env, lo);
7592 static int lod_declare_update_sync_pending(const struct lu_env *env,
7593 struct lod_object *lo, struct md_layout_change *mlc,
7596 struct lod_thread_info *info = lod_env_info(env);
7597 unsigned sync_components = 0;
7598 unsigned resync_components = 0;
7603 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
7604 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
7605 mlc->mlc_opc == MD_LAYOUT_WRITE);
7607 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
7608 PFID(lod_object_fid(lo)), mlc->mlc_opc);
7610 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7611 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
7612 PFID(lod_object_fid(lo)));
7614 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
7615 return lod_declare_update_write_pending(env, lo, mlc, th);
7618 /* MD_LAYOUT_RESYNC_DONE */
7620 for (i = 0; i < lo->ldo_comp_cnt; i++) {
7621 struct lod_layout_component *lod_comp;
7624 lod_comp = &lo->ldo_comp_entries[i];
7626 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
7631 for (j = 0; j < mlc->mlc_resync_count; j++) {
7632 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
7635 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
7636 lod_comp->llc_flags &= ~LCME_FL_STALE;
7637 resync_components++;
7643 for (i = 0; i < mlc->mlc_resync_count; i++) {
7644 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
7647 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
7648 "or already synced\n", PFID(lod_object_fid(lo)),
7649 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
7650 GOTO(out, rc = -EINVAL);
7653 if (!sync_components || (mlc->mlc_resync_count && !resync_components)) {
7654 CDEBUG(D_LAYOUT, DFID": no mirror in sync\n",
7655 PFID(lod_object_fid(lo)));
7657 /* tend to return an error code here to prevent
7658 * the MDT from setting SoM attribute */
7659 GOTO(out, rc = -EINVAL);
7662 CDEBUG(D_LAYOUT, DFID": synced %u resynced %u/%zu components\n",
7663 PFID(lod_object_fid(lo)),
7664 sync_components, resync_components, mlc->mlc_resync_count);
7666 lo->ldo_flr_state = LCM_FL_RDONLY;
7667 lod_obj_inc_layout_gen(lo);
7669 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
7670 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
7671 &info->lti_buf, XATTR_NAME_LOV, 0, th);
7676 lod_striping_free(env, lo);
7680 typedef int (*mlc_handler)(const struct lu_env *env, struct dt_object *dt,
7681 const struct md_layout_change *mlc,
7682 struct thandle *th);
7685 * Attach stripes after target's for migrating directory. NB, we
7686 * only need to declare this, the actual work is done inside
7687 * lod_xattr_set_lmv().
7689 * \param[in] env execution environment
7690 * \param[in] dt target object
7691 * \param[in] mlc layout change data
7692 * \param[in] th transaction handle
7694 * \retval 0 on success
7695 * \retval negative if failed
7697 static int lod_dir_declare_layout_attach(const struct lu_env *env,
7698 struct dt_object *dt,
7699 const struct md_layout_change *mlc,
7702 struct lod_thread_info *info = lod_env_info(env);
7703 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
7704 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
7705 struct lod_object *lo = lod_dt_obj(dt);
7706 struct dt_object *next = dt_object_child(dt);
7707 struct dt_object_format *dof = &info->lti_format;
7708 struct lmv_mds_md_v1 *lmv = mlc->mlc_buf.lb_buf;
7709 struct dt_object **stripes;
7710 __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
7711 struct lu_fid *fid = &info->lti_fid;
7712 struct lod_tgt_desc *tgt;
7713 struct dt_object *dto;
7714 struct dt_device *tgt_dt;
7715 int type = LU_SEQ_RANGE_ANY;
7716 struct dt_insert_rec *rec = &info->lti_dt_rec;
7717 char *stripe_name = info->lti_key;
7718 struct lu_name *sname;
7719 struct linkea_data ldata = { NULL };
7720 struct lu_buf linkea_buf;
7727 if (!lmv_is_sane(lmv))
7730 if (!dt_try_as_dir(env, dt))
7733 dof->dof_type = DFT_DIR;
7735 OBD_ALLOC_PTR_ARRAY(stripes, (lo->ldo_dir_stripe_count + stripe_count));
7739 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
7740 stripes[i] = lo->ldo_stripe[i];
7742 rec->rec_type = S_IFDIR;
7744 for (i = 0; i < stripe_count; i++) {
7746 &lmv->lmv_stripe_fids[i]);
7747 if (!fid_is_sane(fid))
7750 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
7754 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
7755 tgt_dt = lod->lod_child;
7757 tgt = LTD_TGT(ltd, idx);
7759 GOTO(out, rc = -ESTALE);
7760 tgt_dt = tgt->ltd_tgt;
7763 dto = dt_locate_at(env, tgt_dt, fid,
7764 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
7767 GOTO(out, rc = PTR_ERR(dto));
7769 stripes[i + lo->ldo_dir_stripe_count] = dto;
7771 if (!dt_try_as_dir(env, dto))
7772 GOTO(out, rc = -ENOTDIR);
7774 rc = lod_sub_declare_ref_add(env, dto, th);
7778 rec->rec_fid = lu_object_fid(&dto->do_lu);
7779 rc = lod_sub_declare_insert(env, dto,
7780 (const struct dt_rec *)rec,
7781 (const struct dt_key *)dot, th);
7785 rc = lod_sub_declare_insert(env, dto,
7786 (const struct dt_rec *)rec,
7787 (const struct dt_key *)dotdot, th);
7791 rc = lod_sub_declare_xattr_set(env, dto, &mlc->mlc_buf,
7792 XATTR_NAME_LMV, 0, th);
7796 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
7797 PFID(lu_object_fid(&dto->do_lu)),
7798 i + lo->ldo_dir_stripe_count);
7800 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
7801 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
7802 sname, lu_object_fid(&dt->do_lu));
7806 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
7807 linkea_buf.lb_len = ldata.ld_leh->leh_len;
7808 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
7809 XATTR_NAME_LINK, 0, th);
7813 rc = lod_sub_declare_insert(env, next,
7814 (const struct dt_rec *)rec,
7815 (const struct dt_key *)stripe_name,
7820 rc = lod_sub_declare_ref_add(env, next, th);
7826 OBD_FREE_PTR_ARRAY(lo->ldo_stripe,
7827 lo->ldo_dir_stripes_allocated);
7828 lo->ldo_stripe = stripes;
7829 lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count;
7830 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type);
7831 lo->ldo_dir_stripe_count += stripe_count;
7832 lo->ldo_dir_stripes_allocated += stripe_count;
7834 /* plain directory split creates target as a plain directory, while
7835 * after source attached as the first stripe, it becomes a striped
7836 * directory, set correct do_index_ops, otherwise it can't be unlinked.
7838 dt->do_index_ops = &lod_striped_index_ops;
7842 i = lo->ldo_dir_stripe_count;
7843 while (i < lo->ldo_dir_stripe_count + stripe_count && stripes[i])
7844 dt_object_put(env, stripes[i++]);
7846 OBD_FREE_PTR_ARRAY(stripes, stripe_count + lo->ldo_dir_stripe_count);
7850 static int lod_dir_declare_layout_detach(const struct lu_env *env,
7851 struct dt_object *dt,
7852 const struct md_layout_change *unused,
7855 struct lod_thread_info *info = lod_env_info(env);
7856 struct lod_object *lo = lod_dt_obj(dt);
7857 struct dt_object *next = dt_object_child(dt);
7858 char *stripe_name = info->lti_key;
7859 struct dt_object *dto;
7863 if (!dt_try_as_dir(env, dt))
7866 if (!lo->ldo_dir_stripe_count)
7867 return lod_sub_declare_delete(env, next,
7868 (const struct dt_key *)dotdot, th);
7870 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
7871 dto = lo->ldo_stripe[i];
7875 if (!dt_try_as_dir(env, dto))
7878 rc = lod_sub_declare_delete(env, dto,
7879 (const struct dt_key *)dotdot, th);
7883 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
7884 PFID(lu_object_fid(&dto->do_lu)), i);
7886 rc = lod_sub_declare_delete(env, next,
7887 (const struct dt_key *)stripe_name, th);
7891 rc = lod_sub_declare_ref_del(env, next, th);
7899 static int dt_dir_is_empty(const struct lu_env *env,
7900 struct dt_object *obj)
7903 const struct dt_it_ops *iops;
7908 if (!dt_try_as_dir(env, obj))
7911 iops = &obj->do_index_ops->dio_it;
7912 it = iops->init(env, obj, LUDA_64BITHASH);
7914 RETURN(PTR_ERR(it));
7916 rc = iops->get(env, it, (const struct dt_key *)"");
7920 for (rc = 0, i = 0; rc == 0 && i < 3; ++i)
7921 rc = iops->next(env, it);
7927 /* Huh? Index contains no zero key? */
7932 iops->fini(env, it);
7937 static int lod_dir_declare_layout_shrink(const struct lu_env *env,
7938 struct dt_object *dt,
7939 const struct md_layout_change *mlc,
7942 struct lod_thread_info *info = lod_env_info(env);
7943 struct lod_object *lo = lod_dt_obj(dt);
7944 struct dt_object *next = dt_object_child(dt);
7945 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
7946 __u32 final_stripe_count;
7947 char *stripe_name = info->lti_key;
7948 struct lu_buf *lmv_buf = &info->lti_buf;
7949 struct dt_object *dto;
7955 if (!dt_try_as_dir(env, dt))
7958 /* shouldn't be called on plain directory */
7959 LASSERT(lo->ldo_dir_stripe_count);
7961 lmv_buf->lb_buf = &info->lti_lmv.lmv_md_v1;
7962 lmv_buf->lb_len = sizeof(info->lti_lmv.lmv_md_v1);
7964 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
7965 LASSERT(final_stripe_count &&
7966 final_stripe_count < lo->ldo_dir_stripe_count);
7968 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
7969 dto = lo->ldo_stripe[i];
7973 if (i < final_stripe_count) {
7974 if (final_stripe_count == 1)
7977 rc = lod_sub_declare_xattr_set(env, dto, lmv_buf,
7979 LU_XATTR_REPLACE, th);
7986 rc = dt_dir_is_empty(env, dto);
7990 rc = lod_sub_declare_ref_del(env, dto, th);
7994 rc = lod_sub_declare_destroy(env, dto, th);
7998 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
7999 PFID(lu_object_fid(&dto->do_lu)), i);
8001 rc = lod_sub_declare_delete(env, next,
8002 (const struct dt_key *)stripe_name, th);
8006 rc = lod_sub_declare_ref_del(env, next, th);
8011 rc = lod_sub_declare_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
8012 LU_XATTR_REPLACE, th);
8017 * Allocate stripes for split directory.
8019 * \param[in] env execution environment
8020 * \param[in] dt target object
8021 * \param[in] mlc layout change data
8022 * \param[in] th transaction handle
8024 * \retval 0 on success
8025 * \retval negative if failed
8027 static int lod_dir_declare_layout_split(const struct lu_env *env,
8028 struct dt_object *dt,
8029 const struct md_layout_change *mlc,
8032 struct lod_thread_info *info = lod_env_info(env);
8033 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
8034 struct lod_object *lo = lod_dt_obj(dt);
8035 struct dt_object_format *dof = &info->lti_format;
8036 struct lmv_user_md_v1 *lum = mlc->mlc_spec->u.sp_ea.eadata;
8037 struct dt_object **stripes;
8045 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
8046 LASSERT(le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT);
8048 saved_count = lo->ldo_dir_stripes_allocated;
8049 stripe_count = le32_to_cpu(lum->lum_stripe_count);
8050 if (stripe_count <= saved_count)
8053 dof->dof_type = DFT_DIR;
8055 OBD_ALLOC(stripes, sizeof(*stripes) * stripe_count);
8059 for (i = 0; i < lo->ldo_dir_stripes_allocated; i++)
8060 stripes[i] = lo->ldo_stripe[i];
8062 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
8063 rc = lod_mdt_alloc_qos(env, lo, stripes, saved_count, stripe_count);
8065 rc = lod_mdt_alloc_rr(env, lo, stripes, saved_count,
8068 OBD_FREE(stripes, sizeof(*stripes) * stripe_count);
8072 LASSERT(rc > saved_count);
8073 OBD_FREE(lo->ldo_stripe,
8074 sizeof(*stripes) * lo->ldo_dir_stripes_allocated);
8075 lo->ldo_stripe = stripes;
8076 lo->ldo_dir_striped = 1;
8077 lo->ldo_dir_stripe_count = rc;
8078 lo->ldo_dir_stripes_allocated = stripe_count;
8079 lo->ldo_dir_split_hash = lo->ldo_dir_hash_type;
8080 lo->ldo_dir_hash_type = le32_to_cpu(lum->lum_hash_type);
8081 if (!lmv_is_known_hash_type(lo->ldo_dir_hash_type))
8082 lo->ldo_dir_hash_type =
8083 lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
8084 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_SPLIT | LMV_HASH_FLAG_MIGRATION;
8085 lo->ldo_dir_split_offset = saved_count;
8086 lo->ldo_dir_layout_version++;
8087 lo->ldo_dir_stripe_loaded = 1;
8089 rc = lod_dir_declare_create_stripes(env, dt, mlc->mlc_attr, dof, th);
8091 lod_striping_free(env, lo);
8097 * detach all stripes from dir master object, NB, stripes are not destroyed, but
8098 * deleted from it's parent namespace, this function is called in two places:
8099 * 1. mdd_migrate_mdt() detach stripes from source, and attach them to
8101 * 2. mdd_dir_layout_update() detach stripe before turning 1-stripe directory to
8102 * a plain directory.
8104 * \param[in] env execution environment
8105 * \param[in] dt target object
8106 * \param[in] mlc layout change data
8107 * \param[in] th transaction handle
8109 * \retval 0 on success
8110 * \retval negative if failed
8112 static int lod_dir_layout_detach(const struct lu_env *env,
8113 struct dt_object *dt,
8114 const struct md_layout_change *mlc,
8117 struct lod_thread_info *info = lod_env_info(env);
8118 struct lod_object *lo = lod_dt_obj(dt);
8119 struct dt_object *next = dt_object_child(dt);
8120 char *stripe_name = info->lti_key;
8121 struct dt_object *dto;
8127 if (!lo->ldo_dir_stripe_count) {
8128 /* plain directory delete .. */
8129 rc = lod_sub_delete(env, next,
8130 (const struct dt_key *)dotdot, th);
8134 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8135 dto = lo->ldo_stripe[i];
8139 rc = lod_sub_delete(env, dto,
8140 (const struct dt_key *)dotdot, th);
8144 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8145 PFID(lu_object_fid(&dto->do_lu)), i);
8147 rc = lod_sub_delete(env, next,
8148 (const struct dt_key *)stripe_name, th);
8152 rc = lod_sub_ref_del(env, next, th);
8157 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8158 dto = lo->ldo_stripe[i];
8160 dt_object_put(env, dto);
8162 OBD_FREE_PTR_ARRAY(lo->ldo_stripe, lo->ldo_dir_stripes_allocated);
8163 lo->ldo_stripe = NULL;
8164 lo->ldo_dir_stripes_allocated = 0;
8165 lo->ldo_dir_stripe_count = 0;
8166 dt->do_index_ops = &lod_index_ops;
8171 static int lod_dir_layout_shrink(const struct lu_env *env,
8172 struct dt_object *dt,
8173 const struct md_layout_change *mlc,
8176 struct lod_thread_info *info = lod_env_info(env);
8177 struct lod_object *lo = lod_dt_obj(dt);
8178 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
8179 struct dt_object *next = dt_object_child(dt);
8180 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
8181 __u32 final_stripe_count;
8182 char *stripe_name = info->lti_key;
8183 struct dt_object *dto;
8184 struct lu_buf *lmv_buf = &info->lti_buf;
8185 struct lmv_mds_md_v1 *lmv = &info->lti_lmv.lmv_md_v1;
8187 int type = LU_SEQ_RANGE_ANY;
8193 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
8195 lmv_buf->lb_buf = lmv;
8196 lmv_buf->lb_len = sizeof(*lmv);
8197 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
8198 lmv->lmv_stripe_count = cpu_to_le32(final_stripe_count);
8199 lmv->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type) &
8200 cpu_to_le32(LMV_HASH_TYPE_MASK);
8201 lmv->lmv_layout_version =
8202 cpu_to_le32(lo->ldo_dir_layout_version + 1);
8204 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8205 dto = lo->ldo_stripe[i];
8209 if (i < final_stripe_count) {
8210 /* if only one stripe left, no need to update
8211 * LMV because this stripe will replace master
8212 * object and act as a plain directory.
8214 if (final_stripe_count == 1)
8218 rc = lod_fld_lookup(env, lod,
8219 lu_object_fid(&dto->do_lu),
8224 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
8225 rc = lod_sub_xattr_set(env, dto, lmv_buf,
8227 LU_XATTR_REPLACE, th);
8234 dt_write_lock(env, dto, DT_TGT_CHILD);
8235 rc = lod_sub_ref_del(env, dto, th);
8236 dt_write_unlock(env, dto);
8240 rc = lod_sub_destroy(env, dto, th);
8244 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8245 PFID(lu_object_fid(&dto->do_lu)), i);
8247 rc = lod_sub_delete(env, next,
8248 (const struct dt_key *)stripe_name, th);
8252 rc = lod_sub_ref_del(env, next, th);
8257 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &mdtidx,
8262 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
8263 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
8264 rc = lod_sub_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
8265 LU_XATTR_REPLACE, th);
8269 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
8270 dto = lo->ldo_stripe[i];
8272 dt_object_put(env, dto);
8274 lo->ldo_dir_stripe_count = final_stripe_count;
8279 static mlc_handler dir_mlc_declare_ops[MD_LAYOUT_MAX] = {
8280 [MD_LAYOUT_ATTACH] = lod_dir_declare_layout_attach,
8281 [MD_LAYOUT_DETACH] = lod_dir_declare_layout_detach,
8282 [MD_LAYOUT_SHRINK] = lod_dir_declare_layout_shrink,
8283 [MD_LAYOUT_SPLIT] = lod_dir_declare_layout_split,
8286 static mlc_handler dir_mlc_ops[MD_LAYOUT_MAX] = {
8287 [MD_LAYOUT_DETACH] = lod_dir_layout_detach,
8288 [MD_LAYOUT_SHRINK] = lod_dir_layout_shrink,
8291 static int lod_declare_layout_change(const struct lu_env *env,
8292 struct dt_object *dt, struct md_layout_change *mlc,
8295 struct lod_thread_info *info = lod_env_info(env);
8296 struct lod_object *lo = lod_dt_obj(dt);
8301 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
8302 LASSERT(dir_mlc_declare_ops[mlc->mlc_opc]);
8303 rc = dir_mlc_declare_ops[mlc->mlc_opc](env, dt, mlc, th);
8307 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
8308 dt_object_remote(dt_object_child(dt)))
8311 rc = lod_striping_load(env, lo);
8315 LASSERT(lo->ldo_comp_cnt > 0);
8317 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
8321 switch (lo->ldo_flr_state) {
8323 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
8327 rc = lod_declare_update_rdonly(env, lo, mlc, th);
8329 case LCM_FL_WRITE_PENDING:
8330 rc = lod_declare_update_write_pending(env, lo, mlc, th);
8332 case LCM_FL_SYNC_PENDING:
8333 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
8344 * Instantiate layout component objects which covers the intent write offset.
8346 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
8347 struct md_layout_change *mlc, struct thandle *th)
8349 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
8350 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
8351 struct lod_object *lo = lod_dt_obj(dt);
8356 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
8357 LASSERT(dir_mlc_ops[mlc->mlc_opc]);
8358 rc = dir_mlc_ops[mlc->mlc_opc](env, dt, mlc, th);
8362 rc = lod_striped_create(env, dt, attr, NULL, th);
8363 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
8364 layout_attr->la_layout_version |= lo->ldo_layout_gen;
8365 rc = lod_attr_set(env, dt, layout_attr, th);
8371 struct dt_object_operations lod_obj_ops = {
8372 .do_read_lock = lod_read_lock,
8373 .do_write_lock = lod_write_lock,
8374 .do_read_unlock = lod_read_unlock,
8375 .do_write_unlock = lod_write_unlock,
8376 .do_write_locked = lod_write_locked,
8377 .do_attr_get = lod_attr_get,
8378 .do_declare_attr_set = lod_declare_attr_set,
8379 .do_attr_set = lod_attr_set,
8380 .do_xattr_get = lod_xattr_get,
8381 .do_declare_xattr_set = lod_declare_xattr_set,
8382 .do_xattr_set = lod_xattr_set,
8383 .do_declare_xattr_del = lod_declare_xattr_del,
8384 .do_xattr_del = lod_xattr_del,
8385 .do_xattr_list = lod_xattr_list,
8386 .do_ah_init = lod_ah_init,
8387 .do_declare_create = lod_declare_create,
8388 .do_create = lod_create,
8389 .do_declare_destroy = lod_declare_destroy,
8390 .do_destroy = lod_destroy,
8391 .do_index_try = lod_index_try,
8392 .do_declare_ref_add = lod_declare_ref_add,
8393 .do_ref_add = lod_ref_add,
8394 .do_declare_ref_del = lod_declare_ref_del,
8395 .do_ref_del = lod_ref_del,
8396 .do_object_sync = lod_object_sync,
8397 .do_object_lock = lod_object_lock,
8398 .do_object_unlock = lod_object_unlock,
8399 .do_invalidate = lod_invalidate,
8400 .do_declare_layout_change = lod_declare_layout_change,
8401 .do_layout_change = lod_layout_change,
8405 * Implementation of dt_body_operations::dbo_read.
8407 * \see dt_body_operations::dbo_read() in the API description for details.
8409 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
8410 struct lu_buf *buf, loff_t *pos)
8412 struct dt_object *next = dt_object_child(dt);
8414 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
8415 S_ISLNK(dt->do_lu.lo_header->loh_attr));
8416 return next->do_body_ops->dbo_read(env, next, buf, pos);
8420 * Implementation of dt_body_operations::dbo_declare_write.
8422 * \see dt_body_operations::dbo_declare_write() in the API description
8425 static ssize_t lod_declare_write(const struct lu_env *env,
8426 struct dt_object *dt,
8427 const struct lu_buf *buf, loff_t pos,
8430 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
8434 * Implementation of dt_body_operations::dbo_write.
8436 * \see dt_body_operations::dbo_write() in the API description for details.
8438 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
8439 const struct lu_buf *buf, loff_t *pos,
8442 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
8443 S_ISLNK(dt->do_lu.lo_header->loh_attr));
8444 return lod_sub_write(env, dt_object_child(dt), buf, pos, th);
8447 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
8448 __u64 start, __u64 end, struct thandle *th)
8450 if (dt_object_remote(dt))
8453 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
8456 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
8457 __u64 start, __u64 end, struct thandle *th)
8459 if (dt_object_remote(dt))
8462 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
8463 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
8467 * different type of files use the same body_ops because object may be created
8468 * in OUT, where there is no chance to set correct body_ops for each type, so
8469 * body_ops themselves will check file type inside, see lod_read/write/punch for
8472 const struct dt_body_operations lod_body_ops = {
8473 .dbo_read = lod_read,
8474 .dbo_declare_write = lod_declare_write,
8475 .dbo_write = lod_write,
8476 .dbo_declare_punch = lod_declare_punch,
8477 .dbo_punch = lod_punch,
8481 * Implementation of lu_object_operations::loo_object_init.
8483 * The function determines the type and the index of the target device using
8484 * sequence of the object's FID. Then passes control down to the
8485 * corresponding device:
8486 * OSD for the local objects, OSP for remote
8488 * \see lu_object_operations::loo_object_init() in the API description
8491 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
8492 const struct lu_object_conf *conf)
8494 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
8495 struct lu_device *cdev = NULL;
8496 struct lu_object *cobj;
8497 struct lod_tgt_descs *ltd = NULL;
8498 struct lod_tgt_desc *tgt;
8500 int type = LU_SEQ_RANGE_ANY;
8504 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
8508 if (type == LU_SEQ_RANGE_MDT &&
8509 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
8510 cdev = &lod->lod_child->dd_lu_dev;
8511 } else if (type == LU_SEQ_RANGE_MDT) {
8512 ltd = &lod->lod_mdt_descs;
8514 } else if (type == LU_SEQ_RANGE_OST) {
8515 ltd = &lod->lod_ost_descs;
8522 if (ltd->ltd_tgts_size > idx &&
8523 test_bit(idx, ltd->ltd_tgt_bitmap)) {
8524 tgt = LTD_TGT(ltd, idx);
8526 LASSERT(tgt != NULL);
8527 LASSERT(tgt->ltd_tgt != NULL);
8529 cdev = &(tgt->ltd_tgt->dd_lu_dev);
8531 lod_putref(lod, ltd);
8534 if (unlikely(cdev == NULL))
8537 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
8538 if (unlikely(cobj == NULL))
8541 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
8543 lu_object_add(lo, cobj);
8550 * Alloc cached foreign LOV
8552 * \param[in] lo object
8553 * \param[in] size size of foreign LOV
8555 * \retval 0 on success
8556 * \retval negative if failed
8558 int lod_alloc_foreign_lov(struct lod_object *lo, size_t size)
8560 OBD_ALLOC_LARGE(lo->ldo_foreign_lov, size);
8561 if (lo->ldo_foreign_lov == NULL)
8563 lo->ldo_foreign_lov_size = size;
8564 lo->ldo_is_foreign = 1;
8570 * Free cached foreign LOV
8572 * \param[in] lo object
8574 void lod_free_foreign_lov(struct lod_object *lo)
8576 if (lo->ldo_foreign_lov != NULL)
8577 OBD_FREE_LARGE(lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
8578 lo->ldo_foreign_lov = NULL;
8579 lo->ldo_foreign_lov_size = 0;
8580 lo->ldo_is_foreign = 0;
8585 * Free cached foreign LMV
8587 * \param[in] lo object
8589 void lod_free_foreign_lmv(struct lod_object *lo)
8591 if (lo->ldo_foreign_lmv != NULL)
8592 OBD_FREE_LARGE(lo->ldo_foreign_lmv, lo->ldo_foreign_lmv_size);
8593 lo->ldo_foreign_lmv = NULL;
8594 lo->ldo_foreign_lmv_size = 0;
8595 lo->ldo_dir_is_foreign = 0;
8600 * Release resources associated with striping.
8602 * If the object is striped (regular or directory), then release
8603 * the stripe objects references and free the ldo_stripe array.
8605 * \param[in] env execution environment
8606 * \param[in] lo object
8608 void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo)
8610 struct lod_layout_component *lod_comp;
8613 if (unlikely(lo->ldo_is_foreign)) {
8614 lod_free_foreign_lov(lo);
8615 lo->ldo_comp_cached = 0;
8616 } else if (unlikely(lo->ldo_dir_is_foreign)) {
8617 lod_free_foreign_lmv(lo);
8618 lo->ldo_dir_stripe_loaded = 0;
8619 } else if (lo->ldo_stripe != NULL) {
8620 LASSERT(lo->ldo_comp_entries == NULL);
8621 LASSERT(lo->ldo_dir_stripes_allocated > 0);
8623 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8624 if (lo->ldo_stripe[i])
8625 dt_object_put(env, lo->ldo_stripe[i]);
8628 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
8629 OBD_FREE(lo->ldo_stripe, j);
8630 lo->ldo_stripe = NULL;
8631 lo->ldo_dir_stripes_allocated = 0;
8632 lo->ldo_dir_stripe_loaded = 0;
8633 lo->ldo_dir_stripe_count = 0;
8634 } else if (lo->ldo_comp_entries != NULL) {
8635 for (i = 0; i < lo->ldo_comp_cnt; i++) {
8636 /* free lod_layout_component::llc_stripe array */
8637 lod_comp = &lo->ldo_comp_entries[i];
8639 if (lod_comp->llc_stripe == NULL)
8641 LASSERT(lod_comp->llc_stripes_allocated != 0);
8642 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
8643 if (lod_comp->llc_stripe[j] != NULL)
8645 &lod_comp->llc_stripe[j]->do_lu);
8647 OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe,
8648 lod_comp->llc_stripes_allocated);
8649 lod_comp->llc_stripe = NULL;
8650 OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices,
8651 lod_comp->llc_stripes_allocated);
8652 lod_comp->llc_ost_indices = NULL;
8653 lod_comp->llc_stripes_allocated = 0;
8655 lod_free_comp_entries(lo);
8656 lo->ldo_comp_cached = 0;
8660 void lod_striping_free(const struct lu_env *env, struct lod_object *lo)
8662 mutex_lock(&lo->ldo_layout_mutex);
8663 lod_striping_free_nolock(env, lo);
8664 mutex_unlock(&lo->ldo_layout_mutex);
8668 * Implementation of lu_object_operations::loo_object_free.
8670 * \see lu_object_operations::loo_object_free() in the API description
8673 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
8675 struct lod_object *lo = lu2lod_obj(o);
8677 /* release all underlying object pinned */
8678 lod_striping_free(env, lo);
8680 /* lo doesn't contain a lu_object_header, so we don't need call_rcu */
8681 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
8685 * Implementation of lu_object_operations::loo_object_release.
8687 * \see lu_object_operations::loo_object_release() in the API description
8690 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
8692 /* XXX: shouldn't we release everything here in case if object
8693 * creation failed before? */
8697 * Implementation of lu_object_operations::loo_object_print.
8699 * \see lu_object_operations::loo_object_print() in the API description
8702 static int lod_object_print(const struct lu_env *env, void *cookie,
8703 lu_printer_t p, const struct lu_object *l)
8705 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
8707 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
8710 struct lu_object_operations lod_lu_obj_ops = {
8711 .loo_object_init = lod_object_init,
8712 .loo_object_free = lod_object_free,
8713 .loo_object_release = lod_object_release,
8714 .loo_object_print = lod_object_print,