4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * lustre/lod/lod_object.c
31 * This file contains implementations of methods for the OSD API
32 * for the Logical Object Device (LOD) layer, which provides a virtual
33 * local OSD object interface to the MDD layer, and abstracts the
34 * addressing of local (OSD) and remote (OSP) objects. The API is
35 * described in the file lustre/include/dt_object.h and in
36 * Documentation/osd-api.txt.
38 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
41 #define DEBUG_SUBSYSTEM S_MDS
43 #include <linux/random.h>
46 #include <obd_class.h>
47 #include <obd_support.h>
49 #include <lustre_fid.h>
50 #include <lustre_linkea.h>
51 #include <lustre_lmv.h>
52 #include <uapi/linux/lustre/lustre_param.h>
53 #include <lustre_swab.h>
54 #include <uapi/linux/lustre/lustre_ver.h>
55 #include <lprocfs_status.h>
56 #include <md_object.h>
58 #include "lod_internal.h"
60 static const char dot[] = ".";
61 static const char dotdot[] = "..";
64 * Implementation of dt_index_operations::dio_lookup
66 * Used with regular (non-striped) objects.
68 * \see dt_index_operations::dio_lookup() in the API description for details.
70 static int lod_lookup(const struct lu_env *env, struct dt_object *dt,
71 struct dt_rec *rec, const struct dt_key *key)
73 struct dt_object *next = dt_object_child(dt);
74 return next->do_index_ops->dio_lookup(env, next, rec, key);
78 * Implementation of dt_index_operations::dio_declare_insert.
80 * Used with regular (non-striped) objects.
82 * \see dt_index_operations::dio_declare_insert() in the API description
85 static int lod_declare_insert(const struct lu_env *env, struct dt_object *dt,
86 const struct dt_rec *rec,
87 const struct dt_key *key, struct thandle *th)
89 return lod_sub_declare_insert(env, dt_object_child(dt), rec, key, th);
93 * Implementation of dt_index_operations::dio_insert.
95 * Used with regular (non-striped) objects
97 * \see dt_index_operations::dio_insert() in the API description for details.
99 static int lod_insert(const struct lu_env *env, struct dt_object *dt,
100 const struct dt_rec *rec, const struct dt_key *key,
103 return lod_sub_insert(env, dt_object_child(dt), rec, key, th);
107 * Implementation of dt_index_operations::dio_declare_delete.
109 * Used with regular (non-striped) objects.
111 * \see dt_index_operations::dio_declare_delete() in the API description
114 static int lod_declare_delete(const struct lu_env *env, struct dt_object *dt,
115 const struct dt_key *key, struct thandle *th)
117 return lod_sub_declare_delete(env, dt_object_child(dt), key, th);
121 * Implementation of dt_index_operations::dio_delete.
123 * Used with regular (non-striped) objects.
125 * \see dt_index_operations::dio_delete() in the API description for details.
127 static int lod_delete(const struct lu_env *env, struct dt_object *dt,
128 const struct dt_key *key, struct thandle *th)
130 return lod_sub_delete(env, dt_object_child(dt), key, th);
134 * Implementation of dt_it_ops::init.
136 * Used with regular (non-striped) objects.
138 * \see dt_it_ops::init() in the API description for details.
140 static struct dt_it *lod_it_init(const struct lu_env *env,
141 struct dt_object *dt, __u32 attr)
143 struct dt_object *next = dt_object_child(dt);
144 struct lod_it *it = &lod_env_info(env)->lti_it;
145 struct dt_it *it_next;
147 it_next = next->do_index_ops->dio_it.init(env, next, attr);
151 /* currently we do not use more than one iterator per thread
152 * so we store it in thread info. if at some point we need
153 * more active iterators in a single thread, we can allocate
155 LASSERT(it->lit_obj == NULL);
157 it->lit_it = it_next;
160 return (struct dt_it *)it;
163 #define LOD_CHECK_IT(env, it) \
165 LASSERT((it)->lit_obj != NULL); \
166 LASSERT((it)->lit_it != NULL); \
170 * Implementation of dt_index_operations::dio_it.fini.
172 * Used with regular (non-striped) objects.
174 * \see dt_index_operations::dio_it.fini() in the API description for details.
176 static void lod_it_fini(const struct lu_env *env, struct dt_it *di)
178 struct lod_it *it = (struct lod_it *)di;
180 LOD_CHECK_IT(env, it);
181 it->lit_obj->do_index_ops->dio_it.fini(env, it->lit_it);
183 /* the iterator not in use any more */
189 * Implementation of dt_it_ops::get.
191 * Used with regular (non-striped) objects.
193 * \see dt_it_ops::get() in the API description for details.
195 static int lod_it_get(const struct lu_env *env, struct dt_it *di,
196 const struct dt_key *key)
198 const struct lod_it *it = (const struct lod_it *)di;
200 LOD_CHECK_IT(env, it);
201 return it->lit_obj->do_index_ops->dio_it.get(env, it->lit_it, key);
205 * Implementation of dt_it_ops::put.
207 * Used with regular (non-striped) objects.
209 * \see dt_it_ops::put() in the API description for details.
211 static void lod_it_put(const struct lu_env *env, struct dt_it *di)
213 struct lod_it *it = (struct lod_it *)di;
215 LOD_CHECK_IT(env, it);
216 return it->lit_obj->do_index_ops->dio_it.put(env, it->lit_it);
220 * Implementation of dt_it_ops::next.
222 * Used with regular (non-striped) objects
224 * \see dt_it_ops::next() in the API description for details.
226 static int lod_it_next(const struct lu_env *env, struct dt_it *di)
228 struct lod_it *it = (struct lod_it *)di;
230 LOD_CHECK_IT(env, it);
231 return it->lit_obj->do_index_ops->dio_it.next(env, it->lit_it);
235 * Implementation of dt_it_ops::key.
237 * Used with regular (non-striped) objects.
239 * \see dt_it_ops::key() in the API description for details.
241 static struct dt_key *lod_it_key(const struct lu_env *env,
242 const struct dt_it *di)
244 const struct lod_it *it = (const struct lod_it *)di;
246 LOD_CHECK_IT(env, it);
247 return it->lit_obj->do_index_ops->dio_it.key(env, it->lit_it);
251 * Implementation of dt_it_ops::key_size.
253 * Used with regular (non-striped) objects.
255 * \see dt_it_ops::key_size() in the API description for details.
257 static int lod_it_key_size(const struct lu_env *env, const struct dt_it *di)
259 struct lod_it *it = (struct lod_it *)di;
261 LOD_CHECK_IT(env, it);
262 return it->lit_obj->do_index_ops->dio_it.key_size(env, it->lit_it);
266 * Implementation of dt_it_ops::rec.
268 * Used with regular (non-striped) objects.
270 * \see dt_it_ops::rec() in the API description for details.
272 static int lod_it_rec(const struct lu_env *env, const struct dt_it *di,
273 struct dt_rec *rec, __u32 attr)
275 const struct lod_it *it = (const struct lod_it *)di;
277 LOD_CHECK_IT(env, it);
278 return it->lit_obj->do_index_ops->dio_it.rec(env, it->lit_it, rec,
283 * Implementation of dt_it_ops::rec_size.
285 * Used with regular (non-striped) objects.
287 * \see dt_it_ops::rec_size() in the API description for details.
289 static int lod_it_rec_size(const struct lu_env *env, const struct dt_it *di,
292 const struct lod_it *it = (const struct lod_it *)di;
294 LOD_CHECK_IT(env, it);
295 return it->lit_obj->do_index_ops->dio_it.rec_size(env, it->lit_it,
300 * Implementation of dt_it_ops::store.
302 * Used with regular (non-striped) objects.
304 * \see dt_it_ops::store() in the API description for details.
306 static __u64 lod_it_store(const struct lu_env *env, const struct dt_it *di)
308 const struct lod_it *it = (const struct lod_it *)di;
310 LOD_CHECK_IT(env, it);
311 return it->lit_obj->do_index_ops->dio_it.store(env, it->lit_it);
315 * Implementation of dt_it_ops::load.
317 * Used with regular (non-striped) objects.
319 * \see dt_it_ops::load() in the API description for details.
321 static int lod_it_load(const struct lu_env *env, const struct dt_it *di,
324 const struct lod_it *it = (const struct lod_it *)di;
326 LOD_CHECK_IT(env, it);
327 return it->lit_obj->do_index_ops->dio_it.load(env, it->lit_it, hash);
331 * Implementation of dt_it_ops::key_rec.
333 * Used with regular (non-striped) objects.
335 * \see dt_it_ops::rec() in the API description for details.
337 static int lod_it_key_rec(const struct lu_env *env, const struct dt_it *di,
340 const struct lod_it *it = (const struct lod_it *)di;
342 LOD_CHECK_IT(env, it);
343 return it->lit_obj->do_index_ops->dio_it.key_rec(env, it->lit_it,
347 static const struct dt_index_operations lod_index_ops = {
348 .dio_lookup = lod_lookup,
349 .dio_declare_insert = lod_declare_insert,
350 .dio_insert = lod_insert,
351 .dio_declare_delete = lod_declare_delete,
352 .dio_delete = lod_delete,
360 .key_size = lod_it_key_size,
362 .rec_size = lod_it_rec_size,
363 .store = lod_it_store,
365 .key_rec = lod_it_key_rec,
370 * Implementation of dt_index_operations::dio_lookup
372 * Used with striped directories.
374 * \see dt_index_operations::dio_lookup() in the API description for details.
376 static int lod_striped_lookup(const struct lu_env *env, struct dt_object *dt,
377 struct dt_rec *rec, const struct dt_key *key)
379 struct lod_object *lo = lod_dt_obj(dt);
380 struct dt_object *next;
381 const char *name = (const char *)key;
383 LASSERT(lo->ldo_dir_stripe_count > 0);
385 if (strcmp(name, dot) == 0) {
386 struct lu_fid *fid = (struct lu_fid *)rec;
388 *fid = *lod_object_fid(lo);
392 if (strcmp(name, dotdot) == 0) {
393 next = dt_object_child(dt);
397 index = __lmv_name_to_stripe_index(lo->ldo_dir_hash_type,
398 lo->ldo_dir_stripe_count,
399 lo->ldo_dir_migrate_hash,
400 lo->ldo_dir_migrate_offset,
401 name, strlen(name), true);
405 next = lo->ldo_stripe[index];
406 if (!next || !dt_object_exists(next))
410 return next->do_index_ops->dio_lookup(env, next, rec, key);
414 * Implementation of dt_it_ops::init.
416 * Used with striped objects. Internally just initializes the iterator
417 * on the first stripe.
419 * \see dt_it_ops::init() in the API description for details.
421 static struct dt_it *lod_striped_it_init(const struct lu_env *env,
422 struct dt_object *dt, __u32 attr)
424 struct lod_object *lo = lod_dt_obj(dt);
425 struct dt_object *next;
426 struct lod_it *it = &lod_env_info(env)->lti_it;
427 struct dt_it *it_next;
430 LASSERT(lo->ldo_dir_stripe_count > 0);
433 next = lo->ldo_stripe[index];
434 if (next && dt_object_exists(next))
436 } while (++index < lo->ldo_dir_stripe_count);
438 /* no valid stripe */
439 if (!next || !dt_object_exists(next))
440 return ERR_PTR(-ENODEV);
442 LASSERT(next->do_index_ops != NULL);
444 it_next = next->do_index_ops->dio_it.init(env, next, attr);
448 /* currently we do not use more than one iterator per thread
449 * so we store it in thread info. if at some point we need
450 * more active iterators in a single thread, we can allocate
452 LASSERT(it->lit_obj == NULL);
454 it->lit_stripe_index = index;
456 it->lit_it = it_next;
459 return (struct dt_it *)it;
462 #define LOD_CHECK_STRIPED_IT(env, it, lo) \
464 LASSERT((it)->lit_obj != NULL); \
465 LASSERT((it)->lit_it != NULL); \
466 LASSERT((lo)->ldo_dir_stripe_count > 0); \
467 LASSERT((it)->lit_stripe_index < (lo)->ldo_dir_stripe_count); \
471 * Implementation of dt_it_ops::fini.
473 * Used with striped objects.
475 * \see dt_it_ops::fini() in the API description for details.
477 static void lod_striped_it_fini(const struct lu_env *env, struct dt_it *di)
479 struct lod_it *it = (struct lod_it *)di;
480 struct lod_object *lo = lod_dt_obj(it->lit_obj);
481 struct dt_object *next;
483 /* If lit_it == NULL, then it means the sub_it has been finished,
484 * which only happens in failure cases, see lod_striped_it_next() */
485 if (it->lit_it != NULL) {
486 LOD_CHECK_STRIPED_IT(env, it, lo);
488 next = lo->ldo_stripe[it->lit_stripe_index];
490 LASSERT(next->do_index_ops != NULL);
491 next->do_index_ops->dio_it.fini(env, it->lit_it);
495 /* the iterator not in use any more */
498 it->lit_stripe_index = 0;
502 * Implementation of dt_it_ops::get.
504 * Right now it's not used widely, only to reset the iterator to the
505 * initial position. It should be possible to implement a full version
506 * which chooses a correct stripe to be able to position with any key.
508 * \see dt_it_ops::get() in the API description for details.
510 static int lod_striped_it_get(const struct lu_env *env, struct dt_it *di,
511 const struct dt_key *key)
513 const struct lod_it *it = (const struct lod_it *)di;
514 struct lod_object *lo = lod_dt_obj(it->lit_obj);
515 struct dt_object *next;
517 LOD_CHECK_STRIPED_IT(env, it, lo);
519 next = lo->ldo_stripe[it->lit_stripe_index];
520 LASSERT(next != NULL);
521 LASSERT(dt_object_exists(next));
522 LASSERT(next->do_index_ops != NULL);
524 return next->do_index_ops->dio_it.get(env, it->lit_it, key);
528 * Implementation of dt_it_ops::put.
530 * Used with striped objects.
532 * \see dt_it_ops::put() in the API description for details.
534 static void lod_striped_it_put(const struct lu_env *env, struct dt_it *di)
536 struct lod_it *it = (struct lod_it *)di;
537 struct lod_object *lo = lod_dt_obj(it->lit_obj);
538 struct dt_object *next;
541 * If lit_it == NULL, then it means the sub_it has been finished,
542 * which only happens in failure cases, see lod_striped_it_next()
547 LOD_CHECK_STRIPED_IT(env, it, lo);
549 next = lo->ldo_stripe[it->lit_stripe_index];
550 LASSERT(next != NULL);
551 LASSERT(next->do_index_ops != NULL);
553 return next->do_index_ops->dio_it.put(env, it->lit_it);
557 * Implementation of dt_it_ops::next.
559 * Used with striped objects. When the end of the current stripe is
560 * reached, the method takes the next stripe's iterator.
562 * \see dt_it_ops::next() in the API description for details.
564 static int lod_striped_it_next(const struct lu_env *env, struct dt_it *di)
566 struct lod_it *it = (struct lod_it *)di;
567 struct lod_object *lo = lod_dt_obj(it->lit_obj);
568 struct dt_object *next;
569 struct dt_it *it_next;
575 LOD_CHECK_STRIPED_IT(env, it, lo);
577 next = lo->ldo_stripe[it->lit_stripe_index];
578 LASSERT(next != NULL);
579 LASSERT(dt_object_exists(next));
580 LASSERT(next->do_index_ops != NULL);
582 rc = next->do_index_ops->dio_it.next(env, it->lit_it);
586 if (rc == 0 && it->lit_stripe_index == 0)
589 if (rc == 0 && it->lit_stripe_index > 0) {
590 struct lu_dirent *ent;
592 ent = (struct lu_dirent *)lod_env_info(env)->lti_key;
594 rc = next->do_index_ops->dio_it.rec(env, it->lit_it,
595 (struct dt_rec *)ent,
600 /* skip . and .. for slave stripe */
601 if ((strncmp(ent->lde_name, ".",
602 le16_to_cpu(ent->lde_namelen)) == 0 &&
603 le16_to_cpu(ent->lde_namelen) == 1) ||
604 (strncmp(ent->lde_name, "..",
605 le16_to_cpu(ent->lde_namelen)) == 0 &&
606 le16_to_cpu(ent->lde_namelen) == 2))
612 next->do_index_ops->dio_it.put(env, it->lit_it);
613 next->do_index_ops->dio_it.fini(env, it->lit_it);
616 /* go to next stripe */
617 index = it->lit_stripe_index;
618 while (++index < lo->ldo_dir_stripe_count) {
619 next = lo->ldo_stripe[index];
623 if (!dt_object_exists(next))
626 rc = next->do_ops->do_index_try(env, next,
627 &dt_directory_features);
631 LASSERT(next->do_index_ops != NULL);
633 it_next = next->do_index_ops->dio_it.init(env, next,
636 RETURN(PTR_ERR(it_next));
638 rc = next->do_index_ops->dio_it.get(env, it_next,
639 (const struct dt_key *)"");
641 RETURN(rc == 0 ? -EIO : rc);
643 it->lit_it = it_next;
644 it->lit_stripe_index = index;
653 * Implementation of dt_it_ops::key.
655 * Used with striped objects.
657 * \see dt_it_ops::key() in the API description for details.
659 static struct dt_key *lod_striped_it_key(const struct lu_env *env,
660 const struct dt_it *di)
662 const struct lod_it *it = (const struct lod_it *)di;
663 struct lod_object *lo = lod_dt_obj(it->lit_obj);
664 struct dt_object *next;
666 LOD_CHECK_STRIPED_IT(env, it, lo);
668 next = lo->ldo_stripe[it->lit_stripe_index];
669 LASSERT(next != NULL);
670 LASSERT(next->do_index_ops != NULL);
672 return next->do_index_ops->dio_it.key(env, it->lit_it);
676 * Implementation of dt_it_ops::key_size.
678 * Used with striped objects.
680 * \see dt_it_ops::size() in the API description for details.
682 static int lod_striped_it_key_size(const struct lu_env *env,
683 const struct dt_it *di)
685 struct lod_it *it = (struct lod_it *)di;
686 struct lod_object *lo = lod_dt_obj(it->lit_obj);
687 struct dt_object *next;
689 LOD_CHECK_STRIPED_IT(env, it, lo);
691 next = lo->ldo_stripe[it->lit_stripe_index];
692 LASSERT(next != NULL);
693 LASSERT(next->do_index_ops != NULL);
695 return next->do_index_ops->dio_it.key_size(env, it->lit_it);
699 * Implementation of dt_it_ops::rec.
701 * Used with striped objects.
703 * \see dt_it_ops::rec() in the API description for details.
705 static int lod_striped_it_rec(const struct lu_env *env, const struct dt_it *di,
706 struct dt_rec *rec, __u32 attr)
708 const struct lod_it *it = (const struct lod_it *)di;
709 struct lod_object *lo = lod_dt_obj(it->lit_obj);
710 struct dt_object *next;
712 LOD_CHECK_STRIPED_IT(env, it, lo);
714 next = lo->ldo_stripe[it->lit_stripe_index];
715 LASSERT(next != NULL);
716 LASSERT(next->do_index_ops != NULL);
718 return next->do_index_ops->dio_it.rec(env, it->lit_it, rec, attr);
722 * Implementation of dt_it_ops::rec_size.
724 * Used with striped objects.
726 * \see dt_it_ops::rec_size() in the API description for details.
728 static int lod_striped_it_rec_size(const struct lu_env *env,
729 const struct dt_it *di, __u32 attr)
731 struct lod_it *it = (struct lod_it *)di;
732 struct lod_object *lo = lod_dt_obj(it->lit_obj);
733 struct dt_object *next;
735 LOD_CHECK_STRIPED_IT(env, it, lo);
737 next = lo->ldo_stripe[it->lit_stripe_index];
738 LASSERT(next != NULL);
739 LASSERT(next->do_index_ops != NULL);
741 return next->do_index_ops->dio_it.rec_size(env, it->lit_it, attr);
745 * Implementation of dt_it_ops::store.
747 * Used with striped objects.
749 * \see dt_it_ops::store() in the API description for details.
751 static __u64 lod_striped_it_store(const struct lu_env *env,
752 const struct dt_it *di)
754 const struct lod_it *it = (const struct lod_it *)di;
755 struct lod_object *lo = lod_dt_obj(it->lit_obj);
756 struct dt_object *next;
758 LOD_CHECK_STRIPED_IT(env, it, lo);
760 next = lo->ldo_stripe[it->lit_stripe_index];
761 LASSERT(next != NULL);
762 LASSERT(next->do_index_ops != NULL);
764 return next->do_index_ops->dio_it.store(env, it->lit_it);
768 * Implementation of dt_it_ops::load.
770 * Used with striped objects.
772 * \see dt_it_ops::load() in the API description for details.
774 static int lod_striped_it_load(const struct lu_env *env,
775 const struct dt_it *di, __u64 hash)
777 const struct lod_it *it = (const struct lod_it *)di;
778 struct lod_object *lo = lod_dt_obj(it->lit_obj);
779 struct dt_object *next;
781 LOD_CHECK_STRIPED_IT(env, it, lo);
783 next = lo->ldo_stripe[it->lit_stripe_index];
784 LASSERT(next != NULL);
785 LASSERT(next->do_index_ops != NULL);
787 return next->do_index_ops->dio_it.load(env, it->lit_it, hash);
790 static const struct dt_index_operations lod_striped_index_ops = {
791 .dio_lookup = lod_striped_lookup,
792 .dio_declare_insert = lod_declare_insert,
793 .dio_insert = lod_insert,
794 .dio_declare_delete = lod_declare_delete,
795 .dio_delete = lod_delete,
797 .init = lod_striped_it_init,
798 .fini = lod_striped_it_fini,
799 .get = lod_striped_it_get,
800 .put = lod_striped_it_put,
801 .next = lod_striped_it_next,
802 .key = lod_striped_it_key,
803 .key_size = lod_striped_it_key_size,
804 .rec = lod_striped_it_rec,
805 .rec_size = lod_striped_it_rec_size,
806 .store = lod_striped_it_store,
807 .load = lod_striped_it_load,
812 * Append the FID for each shard of the striped directory after the
813 * given LMV EA header.
815 * To simplify striped directory and the consistency verification,
816 * we only store the LMV EA header on disk, for both master object
817 * and slave objects. When someone wants to know the whole LMV EA,
818 * such as client readdir(), we can build the entrie LMV EA on the
819 * MDT side (in RAM) via iterating the sub-directory entries that
820 * are contained in the master object of the stripe directory.
822 * For the master object of the striped directroy, the valid name
823 * for each shard is composed of the ${shard_FID}:${shard_idx}.
825 * There may be holes in the LMV EA if some shards' name entries
826 * are corrupted or lost.
828 * \param[in] env pointer to the thread context
829 * \param[in] lo pointer to the master object of the striped directory
830 * \param[in] buf pointer to the lu_buf which will hold the LMV EA
831 * \param[in] resize whether re-allocate the buffer if it is not big enough
833 * \retval positive size of the LMV EA
834 * \retval 0 for nothing to be loaded
835 * \retval negative error number on failure
837 int lod_load_lmv_shards(const struct lu_env *env, struct lod_object *lo,
838 struct lu_buf *buf, bool resize)
840 struct lu_dirent *ent =
841 (struct lu_dirent *)lod_env_info(env)->lti_key;
842 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
843 struct dt_object *obj = dt_object_child(&lo->ldo_obj);
844 struct lmv_mds_md_v1 *lmv1 = buf->lb_buf;
846 const struct dt_it_ops *iops;
848 __u32 magic = le32_to_cpu(lmv1->lmv_magic);
853 if (magic != LMV_MAGIC_V1)
856 stripes = le32_to_cpu(lmv1->lmv_stripe_count);
860 rc = lmv_mds_md_size(stripes, magic);
864 if (buf->lb_len < lmv1_size) {
873 lu_buf_alloc(buf, lmv1_size);
878 memcpy(buf->lb_buf, tbuf.lb_buf, tbuf.lb_len);
881 if (unlikely(!dt_try_as_dir(env, obj, true)))
884 memset(&lmv1->lmv_stripe_fids[0], 0, stripes * sizeof(struct lu_fid));
885 iops = &obj->do_index_ops->dio_it;
886 it = iops->init(env, obj, LUDA_64BITHASH);
890 rc = iops->load(env, it, 0);
892 rc = iops->next(env, it);
897 char name[FID_LEN + 2] = "";
902 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
908 fid_le_to_cpu(&fid, &ent->lde_fid);
909 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
910 if (ent->lde_name[0] == '.') {
911 if (ent->lde_namelen == 1)
914 if (ent->lde_namelen == 2 && ent->lde_name[1] == '.')
918 len = scnprintf(name, sizeof(name),
919 DFID":", PFID(&ent->lde_fid));
920 /* The ent->lde_name is composed of ${FID}:${index} */
921 if (ent->lde_namelen < len + 1 ||
922 memcmp(ent->lde_name, name, len) != 0) {
923 CDEBUG_LIMIT(lod->lod_lmv_failout ? D_ERROR : D_INFO,
924 "%s: invalid shard name %.*s with the FID "DFID" for the striped directory "DFID", %s\n",
925 lod2obd(lod)->obd_name, ent->lde_namelen,
926 ent->lde_name, PFID(&fid),
927 PFID(lu_object_fid(&obj->do_lu)),
928 lod->lod_lmv_failout ? "failout" : "skip");
930 if (lod->lod_lmv_failout)
938 if (ent->lde_name[len] < '0' ||
939 ent->lde_name[len] > '9') {
940 CDEBUG_LIMIT(lod->lod_lmv_failout ?
942 "%s: invalid shard name %.*s with the FID "DFID" for the striped directory "DFID", %s\n",
943 lod2obd(lod)->obd_name,
945 ent->lde_name, PFID(&fid),
946 PFID(lu_object_fid(&obj->do_lu)),
947 lod->lod_lmv_failout ?
950 if (lod->lod_lmv_failout)
956 index = index * 10 + ent->lde_name[len++] - '0';
957 } while (len < ent->lde_namelen);
959 if (len == ent->lde_namelen) {
960 /* Out of LMV EA range. */
961 if (index >= stripes) {
962 CERROR("%s: the shard %.*s for the striped "
963 "directory "DFID" is out of the known "
964 "LMV EA range [0 - %u], failout\n",
965 lod2obd(lod)->obd_name, ent->lde_namelen,
967 PFID(lu_object_fid(&obj->do_lu)),
973 /* The slot has been occupied. */
974 if (!fid_is_zero(&lmv1->lmv_stripe_fids[index]) &&
975 !CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME)) {
979 &lmv1->lmv_stripe_fids[index]);
980 CERROR("%s: both the shard "DFID" and "DFID
981 " for the striped directory "DFID
982 " claim the same LMV EA slot at the "
983 "index %d, failout\n",
984 lod2obd(lod)->obd_name,
985 PFID(&fid0), PFID(&fid),
986 PFID(lu_object_fid(&obj->do_lu)), index);
991 /* stored as LE mode */
992 lmv1->lmv_stripe_fids[index] = ent->lde_fid;
995 rc = iops->next(env, it);
1000 iops->fini(env, it);
1002 RETURN(rc > 0 ? lmv_mds_md_size(stripes, magic) : rc);
1006 * Implementation of dt_object_operations::do_index_try.
1008 * \see dt_object_operations::do_index_try() in the API description for details.
1010 static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
1011 const struct dt_index_features *feat)
1013 struct lod_object *lo = lod_dt_obj(dt);
1014 struct dt_object *next = dt_object_child(dt);
1018 LASSERT(next->do_ops);
1019 LASSERT(next->do_ops->do_index_try);
1021 rc = lod_striping_load(env, lo);
1025 rc = next->do_ops->do_index_try(env, next, feat);
1029 if (lo->ldo_dir_stripe_count > 0) {
1032 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1033 if (!lo->ldo_stripe[i])
1035 if (!dt_object_exists(lo->ldo_stripe[i]))
1037 rc = lo->ldo_stripe[i]->do_ops->do_index_try(env,
1038 lo->ldo_stripe[i], feat);
1042 dt->do_index_ops = &lod_striped_index_ops;
1044 dt->do_index_ops = &lod_index_ops;
1051 * Implementation of dt_object_operations::do_read_lock.
1053 * \see dt_object_operations::do_read_lock() in the API description for details.
1055 static void lod_read_lock(const struct lu_env *env, struct dt_object *dt,
1058 dt_read_lock(env, dt_object_child(dt), role);
1062 * Implementation of dt_object_operations::do_write_lock.
1064 * \see dt_object_operations::do_write_lock() in the API description for
1067 static void lod_write_lock(const struct lu_env *env, struct dt_object *dt,
1070 dt_write_lock(env, dt_object_child(dt), role);
1074 * Implementation of dt_object_operations::do_read_unlock.
1076 * \see dt_object_operations::do_read_unlock() in the API description for
1079 static void lod_read_unlock(const struct lu_env *env, struct dt_object *dt)
1081 dt_read_unlock(env, dt_object_child(dt));
1085 * Implementation of dt_object_operations::do_write_unlock.
1087 * \see dt_object_operations::do_write_unlock() in the API description for
1090 static void lod_write_unlock(const struct lu_env *env, struct dt_object *dt)
1092 dt_write_unlock(env, dt_object_child(dt));
1096 * Implementation of dt_object_operations::do_write_locked.
1098 * \see dt_object_operations::do_write_locked() in the API description for
1101 static int lod_write_locked(const struct lu_env *env, struct dt_object *dt)
1103 return dt_write_locked(env, dt_object_child(dt));
1107 * Implementation of dt_object_operations::do_attr_get.
1109 * \see dt_object_operations::do_attr_get() in the API description for details.
1111 static int lod_attr_get(const struct lu_env *env,
1112 struct dt_object *dt,
1113 struct lu_attr *attr)
1115 /* Note: for striped directory, client will merge attributes
1116 * from all of the sub-stripes see lmv_merge_attr(), and there
1117 * no MDD logic depend on directory nlink/size/time, so we can
1118 * always use master inode nlink and size for now. */
1119 return dt_attr_get(env, dt_object_child(dt), attr);
1122 void lod_adjust_stripe_size(struct lod_layout_component *comp,
1123 __u32 def_stripe_size)
1125 __u64 comp_end = comp->llc_extent.e_end;
1127 /* Choose stripe size if not set. Note that default stripe size can't
1128 * be used as is, because it must be multiplier of given component end.
1129 * - first check if default stripe size can be used
1130 * - if not than select the lowest set bit from component end and use
1131 * that value as stripe size
1133 if (!comp->llc_stripe_size) {
1134 if (comp_end == LUSTRE_EOF || !(comp_end % def_stripe_size))
1135 comp->llc_stripe_size = def_stripe_size;
1137 comp->llc_stripe_size = comp_end & ~(comp_end - 1);
1139 if (comp_end != LUSTRE_EOF &&
1140 comp_end & (LOV_MIN_STRIPE_SIZE - 1)) {
1141 CWARN("Component end %llu is not a multiple of min size %u\n",
1142 comp_end, LOV_MIN_STRIPE_SIZE);
1143 comp_end = round_up(comp_end, LOV_MIN_STRIPE_SIZE);
1145 /* check stripe size is multiplier of comp_end */
1146 if (comp_end != LUSTRE_EOF &&
1147 comp_end != comp->llc_extent.e_start &&
1148 comp_end % comp->llc_stripe_size) {
1149 /* fix that even for defined stripe size but warn
1150 * about the problem, that must not happen
1152 CWARN("Component end %llu is not aligned by the stripe size %u\n",
1153 comp_end, comp->llc_stripe_size);
1154 comp->llc_stripe_size = comp_end & ~(comp_end - 1);
1159 static inline void lod_adjust_stripe_info(struct lod_layout_component *comp,
1160 struct lov_desc *desc,
1163 if (!(comp->llc_pattern & LOV_PATTERN_MDT)) {
1164 if (append_stripes) {
1165 comp->llc_stripe_count = append_stripes;
1166 } else if (!comp->llc_stripe_count) {
1167 comp->llc_stripe_count =
1168 desc->ld_default_stripe_count;
1172 lod_adjust_stripe_size(comp, desc->ld_default_stripe_size);
1175 int lod_obj_for_each_stripe(const struct lu_env *env, struct lod_object *lo,
1177 struct lod_obj_stripe_cb_data *data)
1179 struct lod_layout_component *lod_comp;
1183 mutex_lock(&lo->ldo_layout_mutex);
1184 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1185 lod_comp = &lo->ldo_comp_entries[i];
1187 if (lod_comp->llc_magic == LOV_MAGIC_FOREIGN)
1190 if (lod_comp->llc_stripe == NULL)
1193 /* has stripe but not inited yet, this component has been
1194 * declared to be created, but hasn't created yet.
1196 if (!lod_comp_inited(lod_comp))
1199 if (data->locd_comp_skip_cb &&
1200 data->locd_comp_skip_cb(env, lo, i, data))
1203 if (data->locd_comp_cb) {
1204 rc = data->locd_comp_cb(env, lo, i, data);
1209 /* could used just to do sth about component, not each
1212 if (!data->locd_stripe_cb)
1215 LASSERT(lod_comp->llc_stripe_count > 0);
1216 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
1217 struct dt_object *dt = lod_comp->llc_stripe[j];
1221 rc = data->locd_stripe_cb(env, lo, dt, th, i, j, data);
1227 mutex_unlock(&lo->ldo_layout_mutex);
1232 lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
1233 struct dt_object *dt, struct thandle *th,
1234 int comp_idx, int stripe_idx,
1235 struct lod_obj_stripe_cb_data *data)
1237 if (data->locd_declare)
1238 return lod_sub_declare_attr_set(env, dt, data->locd_attr, th);
1240 if (data->locd_attr->la_valid & LA_LAYOUT_VERSION) {
1241 CDEBUG(D_LAYOUT, DFID": set layout version: %u, comp_idx: %d\n",
1242 PFID(lu_object_fid(&dt->do_lu)),
1243 data->locd_attr->la_layout_version, comp_idx);
1246 return lod_sub_attr_set(env, dt, data->locd_attr, th);
1250 * Implementation of dt_object_operations::do_declare_attr_set.
1252 * If the object is striped, then apply the changes to all the stripes.
1254 * \see dt_object_operations::do_declare_attr_set() in the API description
1257 static int lod_declare_attr_set(const struct lu_env *env,
1258 struct dt_object *dt,
1259 const struct lu_attr *attr,
1262 struct dt_object *next = dt_object_child(dt);
1263 struct lod_object *lo = lod_dt_obj(dt);
1268 * declare setattr on the local object
1270 rc = lod_sub_declare_attr_set(env, next, attr, th);
1274 /* osp_declare_attr_set() ignores all attributes other than
1275 * UID, GID, PROJID, and size, and osp_attr_set() ignores all
1276 * but UID, GID and PROJID. Declaration of size attr setting
1277 * happens through lod_declare_init_size(), and not through
1278 * this function. Therefore we need not load striping unless
1279 * ownership is changing. This should save memory and (we hope)
1280 * speed up rename().
1282 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1283 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1286 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1289 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID | LA_MODE |
1290 LA_ATIME | LA_MTIME | LA_CTIME |
1295 * load striping information, notice we don't do this when object
1296 * is being initialized as we don't need this information till
1297 * few specific cases like destroy, chown
1299 rc = lod_striping_load(env, lo);
1303 if (!lod_obj_is_striped(dt))
1307 * if object is striped declare changes on the stripes
1309 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1310 LASSERT(lo->ldo_stripe);
1311 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1312 if (lo->ldo_stripe[i] == NULL)
1314 if (!dt_object_exists(lo->ldo_stripe[i]))
1316 rc = lod_sub_declare_attr_set(env, lo->ldo_stripe[i],
1322 struct lod_obj_stripe_cb_data data = { { 0 } };
1324 data.locd_attr = attr;
1325 data.locd_declare = true;
1326 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1327 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1333 if (!dt_object_exists(next) || dt_object_remote(next) ||
1334 !S_ISREG(attr->la_mode))
1337 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1338 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
1342 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) ||
1343 CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1344 struct lod_thread_info *info = lod_env_info(env);
1345 struct lu_buf *buf = &info->lti_buf;
1347 buf->lb_buf = info->lti_ea_store;
1348 buf->lb_len = info->lti_ea_store_size;
1349 rc = lod_sub_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1350 LU_XATTR_REPLACE, th);
1357 * Implementation of dt_object_operations::do_attr_set.
1359 * If the object is striped, then apply the changes to all or subset of
1360 * the stripes depending on the object type and specific attributes.
1362 * \see dt_object_operations::do_attr_set() in the API description for details.
1364 static int lod_attr_set(const struct lu_env *env,
1365 struct dt_object *dt,
1366 const struct lu_attr *attr,
1369 struct dt_object *next = dt_object_child(dt);
1370 struct lod_object *lo = lod_dt_obj(dt);
1375 * apply changes to the local object
1377 rc = lod_sub_attr_set(env, next, attr, th);
1381 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1382 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1385 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1388 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE | LA_PROJID |
1389 LA_ATIME | LA_MTIME | LA_CTIME |
1394 /* FIXME: a tricky case in the code path of mdd_layout_change():
1395 * the in-memory striping information has been freed in lod_xattr_set()
1396 * due to layout change. It has to load stripe here again. It only
1397 * changes flags of layout so declare_attr_set() is still accurate */
1398 rc = lod_striping_load(env, lo);
1402 if (!lod_obj_is_striped(dt))
1406 * if object is striped, apply changes to all the stripes
1408 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1409 LASSERT(lo->ldo_stripe);
1410 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1411 if (unlikely(lo->ldo_stripe[i] == NULL))
1414 if ((dt_object_exists(lo->ldo_stripe[i]) == 0))
1417 rc = lod_sub_attr_set(env, lo->ldo_stripe[i], attr, th);
1422 struct lod_obj_stripe_cb_data data = { { 0 } };
1424 data.locd_attr = attr;
1425 data.locd_declare = false;
1426 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1427 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1433 if (!dt_object_exists(next) || dt_object_remote(next) ||
1434 !S_ISREG(attr->la_mode))
1437 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1438 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
1442 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE)) {
1443 struct lod_thread_info *info = lod_env_info(env);
1444 struct lu_buf *buf = &info->lti_buf;
1445 struct ost_id *oi = &info->lti_ostid;
1446 struct lu_fid *fid = &info->lti_fid;
1447 struct lov_mds_md_v1 *lmm;
1448 struct lov_ost_data_v1 *objs;
1451 rc = lod_get_lov_ea(env, lo);
1455 buf->lb_buf = info->lti_ea_store;
1456 buf->lb_len = info->lti_ea_store_size;
1457 lmm = info->lti_ea_store;
1458 magic = le32_to_cpu(lmm->lmm_magic);
1459 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1460 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1461 struct lov_comp_md_entry_v1 *lcme =
1462 &lcm->lcm_entries[0];
1464 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
1465 magic = le32_to_cpu(lmm->lmm_magic);
1468 if (magic == LOV_MAGIC_V1)
1469 objs = &(lmm->lmm_objects[0]);
1471 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1472 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1473 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1475 fid_to_ostid(fid, oi);
1476 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1478 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1479 LU_XATTR_REPLACE, th);
1480 } else if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1481 struct lod_thread_info *info = lod_env_info(env);
1482 struct lu_buf *buf = &info->lti_buf;
1483 struct lov_comp_md_v1 *lcm;
1484 struct lov_comp_md_entry_v1 *lcme;
1486 rc = lod_get_lov_ea(env, lo);
1490 buf->lb_buf = info->lti_ea_store;
1491 buf->lb_len = info->lti_ea_store_size;
1493 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1 &&
1494 le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_SEL)
1497 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1498 lcme = &lcm->lcm_entries[0];
1499 le64_add_cpu(&lcme->lcme_extent.e_start, 1);
1500 le64_add_cpu(&lcme->lcme_extent.e_end, -1);
1502 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1503 LU_XATTR_REPLACE, th);
1510 * Implementation of dt_object_operations::do_xattr_get.
1512 * If LOV EA is requested from the root object and it's not
1513 * found, then return default striping for the filesystem.
1515 * \see dt_object_operations::do_xattr_get() in the API description for details.
1517 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1518 struct lu_buf *buf, const char *name)
1520 struct lod_thread_info *info = lod_env_info(env);
1521 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1526 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1527 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1528 struct lmv_mds_md_v1 *lmv1;
1529 struct lmv_foreign_md *lfm;
1532 if (rc > (typeof(rc))sizeof(*lmv1))
1535 /* short (<= sizeof(struct lmv_mds_md_v1)) foreign LMV case */
1536 /* XXX empty foreign LMV is not allowed */
1537 if (rc <= offsetof(typeof(*lfm), lfm_value))
1538 RETURN(rc = rc > 0 ? -EINVAL : rc);
1540 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1541 BUILD_BUG_ON(sizeof(*lmv1) > sizeof(info->lti_key));
1543 /* lti_buf is large enough for *lmv1 or a short
1544 * (<= sizeof(struct lmv_mds_md_v1)) foreign LMV
1546 info->lti_buf.lb_buf = info->lti_key;
1547 info->lti_buf.lb_len = sizeof(*lmv1);
1548 rc = dt_xattr_get(env, dt_object_child(dt),
1549 &info->lti_buf, name);
1550 if (unlikely(rc <= offsetof(typeof(*lfm),
1552 RETURN(rc = rc > 0 ? -EINVAL : rc);
1554 lfm = info->lti_buf.lb_buf;
1555 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN)
1558 if (unlikely(rc != sizeof(*lmv1)))
1559 RETURN(rc = rc > 0 ? -EINVAL : rc);
1561 lmv1 = info->lti_buf.lb_buf;
1562 /* The on-disk LMV EA only contains header, but the
1563 * returned LMV EA size should contain the space for
1564 * the FIDs of all shards of the striped directory. */
1565 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1566 rc = lmv_mds_md_size(
1567 le32_to_cpu(lmv1->lmv_stripe_count),
1568 le32_to_cpu(lmv1->lmv_magic));
1571 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1574 if (rc != sizeof(*lmv1))
1575 RETURN(rc = rc > 0 ? -EINVAL : rc);
1577 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1581 RETURN(rc = rc1 != 0 ? rc1 : rc);
1584 if ((rc > 0) && buf->lb_buf && strcmp(name, XATTR_NAME_LOV) == 0) {
1585 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1587 if (lcm->lcm_magic == cpu_to_le32(LOV_MAGIC_SEL))
1588 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
1591 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1595 * XXX: Only used by lfsck
1597 * lod returns default striping on the real root of the device
1598 * this is like the root stores default striping for the whole
1599 * filesystem. historically we've been using a different approach
1600 * and store it in the config.
1602 dt_root_get(env, dev->lod_child, &info->lti_fid);
1603 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1605 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1606 struct lov_user_md *lum = buf->lb_buf;
1607 struct lov_desc *desc = &dev->lod_ost_descs.ltd_lov_desc;
1609 if (buf->lb_buf == NULL) {
1611 } else if (buf->lb_len >= sizeof(*lum)) {
1612 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1613 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1614 lmm_oi_set_id(&lum->lmm_oi, 0);
1615 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1616 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1617 lum->lmm_stripe_size = cpu_to_le32(
1618 desc->ld_default_stripe_size);
1619 lum->lmm_stripe_count = cpu_to_le16(
1620 desc->ld_default_stripe_count);
1621 lum->lmm_stripe_offset = cpu_to_le16(
1622 desc->ld_default_stripe_offset);
1635 * Checks that the magic of the stripe is sane.
1637 * \param[in] lod lod device
1638 * \param[in] lum a buffer storing LMV EA to verify
1640 * \retval 0 if the EA is sane
1641 * \retval negative otherwise
1643 static int lod_verify_md_striping(struct lod_device *lod,
1644 const struct lmv_user_md_v1 *lum)
1646 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1647 CERROR("%s: invalid lmv_user_md: magic = %x, "
1648 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1649 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1650 (int)le32_to_cpu(lum->lum_stripe_offset),
1651 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1659 * Initialize LMV EA for a slave.
1661 * Initialize slave's LMV EA from the master's LMV EA.
1663 * \param[in] master_lmv a buffer containing master's EA
1664 * \param[out] slave_lmv a buffer where slave's EA will be stored
1667 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1668 const struct lmv_mds_md_v1 *master_lmv)
1670 *slave_lmv = *master_lmv;
1671 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1677 * Generate LMV EA from the object passed as \a dt. The object must have
1678 * the stripes created and initialized.
1680 * \param[in] env execution environment
1681 * \param[in] dt object
1682 * \param[out] lmv_buf buffer storing generated LMV EA
1684 * \retval 0 on success
1685 * \retval negative if failed
1687 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1688 struct lu_buf *lmv_buf)
1690 struct lod_thread_info *info = lod_env_info(env);
1691 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1692 struct lod_object *lo = lod_dt_obj(dt);
1693 struct lmv_mds_md_v1 *lmm1;
1695 int type = LU_SEQ_RANGE_ANY;
1700 LASSERT(lo->ldo_dir_striped != 0);
1701 LASSERT(lo->ldo_dir_stripe_count > 0);
1702 stripe_count = lo->ldo_dir_stripe_count;
1703 /* Only store the LMV EA heahder on the disk. */
1704 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1705 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1709 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1712 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1713 memset(lmm1, 0, sizeof(*lmm1));
1714 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1715 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1716 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1717 lmm1->lmv_layout_version = cpu_to_le32(lo->ldo_dir_layout_version);
1718 if (lod_is_layout_changing(lo)) {
1719 lmm1->lmv_migrate_hash = cpu_to_le32(lo->ldo_dir_migrate_hash);
1720 lmm1->lmv_migrate_offset =
1721 cpu_to_le32(lo->ldo_dir_migrate_offset);
1723 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1728 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1729 lmv_buf->lb_buf = info->lti_ea_store;
1730 lmv_buf->lb_len = sizeof(*lmm1);
1736 * Create in-core represenation for a striped directory.
1738 * Parse the buffer containing LMV EA and instantiate LU objects
1739 * representing the stripe objects. The pointers to the objects are
1740 * stored in ldo_stripe field of \a lo. This function is used when
1741 * we need to access an already created object (i.e. load from a disk).
1743 * \param[in] env execution environment
1744 * \param[in] lo lod object
1745 * \param[in] buf buffer containing LMV EA
1747 * \retval 0 on success
1748 * \retval negative if failed
1750 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1751 const struct lu_buf *buf)
1753 struct lod_thread_info *info = lod_env_info(env);
1754 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1755 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1756 struct dt_object **stripe;
1757 union lmv_mds_md *lmm = buf->lb_buf;
1758 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1759 struct lu_fid *fid = &info->lti_fid;
1764 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1766 /* XXX may be useless as not called for foreign LMV ?? */
1767 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_FOREIGN)
1770 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1771 lo->ldo_dir_slave_stripe = 1;
1775 if (!lmv_is_sane(lmv1))
1778 LASSERT(lo->ldo_stripe == NULL);
1779 OBD_ALLOC_PTR_ARRAY(stripe, le32_to_cpu(lmv1->lmv_stripe_count));
1783 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1784 struct dt_device *tgt_dt;
1785 struct dt_object *dto;
1786 int type = LU_SEQ_RANGE_ANY;
1789 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1790 if (!fid_is_sane(fid)) {
1795 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1799 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1800 tgt_dt = lod->lod_child;
1802 struct lod_tgt_desc *tgt;
1804 tgt = LTD_TGT(ltd, idx);
1806 GOTO(out, rc = -ESTALE);
1807 tgt_dt = tgt->ltd_tgt;
1810 dto = dt_locate_at(env, tgt_dt, fid,
1811 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1814 GOTO(out, rc = PTR_ERR(dto));
1819 lo->ldo_stripe = stripe;
1820 lo->ldo_is_foreign = 0;
1821 lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1822 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1823 lo->ldo_dir_layout_version = le32_to_cpu(lmv1->lmv_layout_version);
1824 lo->ldo_dir_migrate_offset = le32_to_cpu(lmv1->lmv_migrate_offset);
1825 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv1->lmv_migrate_hash);
1826 lo->ldo_dir_hash_type = le32_to_cpu(lmv1->lmv_hash_type);
1828 lod_striping_free_nolock(env, lo);
1834 * Declare create a striped directory.
1836 * Declare creating a striped directory with a given stripe pattern on the
1837 * specified MDTs. A striped directory is represented as a regular directory
1838 * - an index listing all the stripes. The stripes point back to the master
1839 * object with ".." and LinkEA. The master object gets LMV EA which
1840 * identifies it as a striped directory. The function allocates FIDs
1843 * \param[in] env execution environment
1844 * \param[in] dt object
1845 * \param[in] attr attributes to initialize the objects with
1846 * \param[in] dof type of objects to be created
1847 * \param[in] th transaction handle
1849 * \retval 0 on success
1850 * \retval negative if failed
1852 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1853 struct dt_object *dt,
1854 struct lu_attr *attr,
1855 struct dt_object_format *dof,
1858 struct lod_thread_info *info = lod_env_info(env);
1859 struct lu_buf lmv_buf;
1860 struct lu_buf slave_lmv_buf;
1861 struct lmv_mds_md_v1 *lmm;
1862 struct lmv_mds_md_v1 *slave_lmm = NULL;
1863 struct dt_insert_rec *rec = &info->lti_dt_rec;
1864 struct lod_object *lo = lod_dt_obj(dt);
1869 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1872 lmm = lmv_buf.lb_buf;
1874 OBD_ALLOC_PTR(slave_lmm);
1875 if (slave_lmm == NULL)
1876 GOTO(out, rc = -ENOMEM);
1878 lod_prep_slave_lmv_md(slave_lmm, lmm);
1879 slave_lmv_buf.lb_buf = slave_lmm;
1880 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1882 if (!dt_try_as_dir(env, dt_object_child(dt), false))
1883 GOTO(out, rc = -EINVAL);
1885 rec->rec_type = S_IFDIR;
1886 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1887 struct dt_object *dto = lo->ldo_stripe[i];
1888 char *stripe_name = info->lti_key;
1889 struct lu_name *sname;
1890 struct linkea_data ldata = { NULL };
1891 struct lu_buf linkea_buf;
1893 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
1897 /* directory split skip create for existing stripes */
1898 if (!(lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) {
1899 rc = lod_sub_declare_create(env, dto, attr, NULL, dof,
1904 if (!dt_try_as_dir(env, dto, false))
1905 GOTO(out, rc = -EINVAL);
1907 rc = lod_sub_declare_ref_add(env, dto, th);
1911 rec->rec_fid = lu_object_fid(&dto->do_lu);
1912 rc = lod_sub_declare_insert(env, dto,
1913 (const struct dt_rec *)rec,
1914 (const struct dt_key *)dot,
1919 /* master stripe FID will be put to .. */
1920 rec->rec_fid = lu_object_fid(&dt->do_lu);
1921 rc = lod_sub_declare_insert(env, dto,
1922 (const struct dt_rec *)rec,
1923 (const struct dt_key *)dotdot,
1928 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1930 snprintf(stripe_name, sizeof(info->lti_key),
1932 PFID(lu_object_fid(&dto->do_lu)),
1935 snprintf(stripe_name, sizeof(info->lti_key),
1937 PFID(lu_object_fid(&dto->do_lu)), i);
1939 sname = lod_name_get(env, stripe_name,
1940 strlen(stripe_name));
1941 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1942 sname, lu_object_fid(&dt->do_lu));
1946 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1947 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1948 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
1949 XATTR_NAME_LINK, 0, th);
1953 rec->rec_fid = lu_object_fid(&dto->do_lu);
1954 rc = lod_sub_declare_insert(env, dt_object_child(dt),
1955 (const struct dt_rec *)rec,
1956 (const struct dt_key *)stripe_name, th);
1960 rc = lod_sub_declare_ref_add(env, dt_object_child(dt),
1966 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1967 cfs_fail_val != i) {
1968 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1970 slave_lmm->lmv_master_mdt_index =
1973 slave_lmm->lmv_master_mdt_index =
1975 rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
1976 XATTR_NAME_LMV, 0, th);
1982 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
1983 &lmv_buf, XATTR_NAME_LMV, 0, th);
1987 if (slave_lmm != NULL)
1988 OBD_FREE_PTR(slave_lmm);
1994 * Allocate a striping on a predefined set of MDTs.
1996 * Allocates new striping using the MDT index range provided by the data from
1997 * the lum_obejcts contained in the lmv_user_md passed to this method if
1998 * \a is_specific is true; or allocates new layout starting from MDT index in
1999 * lo->ldo_dir_stripe_offset. The exact order of MDTs is not important and
2000 * varies depending on MDT status. The number of stripes needed and stripe
2001 * offset are taken from the object. If that number cannot be met, then the
2002 * function returns an error and then it's the caller's responsibility to
2003 * release the stripes allocated. All the internal structures are protected,
2004 * but no concurrent allocation is allowed on the same objects.
2006 * \param[in] env execution environment for this thread
2007 * \param[in] lo LOD object
2008 * \param[out] stripes striping created
2009 * \param[out] mdt_indices MDT indices of striping created
2010 * \param[in] is_specific true if the MDTs are provided by lum; false if
2011 * only the starting MDT index is provided
2013 * \retval positive stripes allocated, including the first stripe allocated
2015 * \retval negative errno on failure
2017 static int lod_mdt_alloc_specific(const struct lu_env *env,
2018 struct lod_object *lo,
2019 struct dt_object **stripes,
2020 __u32 *mdt_indices, bool is_specific)
2022 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
2023 struct lu_tgt_descs *ltd = &lod->lod_mdt_descs;
2024 struct lu_tgt_desc *tgt = NULL;
2025 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2026 struct dt_device *tgt_dt = NULL;
2027 struct lu_fid fid = { 0 };
2028 struct dt_object *dto;
2030 u32 stripe_count = lo->ldo_dir_stripe_count;
2036 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
2037 if (!is_specific && stripe_count > 1)
2038 /* Set the start index for the 2nd stripe allocation */
2039 mdt_indices[1] = (mdt_indices[0] + 1) %
2040 (lod->lod_remote_mdt_count + 1);
2042 for (; stripe_idx < stripe_count; stripe_idx++) {
2043 /* Try to find next avaible target */
2044 idx = mdt_indices[stripe_idx];
2045 for (j = 0; j < lod->lod_remote_mdt_count;
2046 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
2047 bool already_allocated = false;
2050 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
2051 idx, lod->lod_remote_mdt_count + 1, stripe_idx);
2053 if (likely(!is_specific &&
2054 !CFS_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
2055 /* check whether the idx already exists
2056 * in current allocated array */
2057 for (k = 0; k < stripe_idx; k++) {
2058 if (mdt_indices[k] == idx) {
2059 already_allocated = true;
2064 if (already_allocated)
2068 /* Sigh, this index is not in the bitmap, let's check
2069 * next available target */
2070 if (!test_bit(idx, ltd->ltd_tgt_bitmap) &&
2071 idx != master_index)
2074 if (idx == master_index) {
2075 /* Allocate the FID locally */
2076 tgt_dt = lod->lod_child;
2077 rc = dt_fid_alloc(env, tgt_dt, &fid, NULL,
2084 /* check the status of the OSP */
2085 tgt = LTD_TGT(ltd, idx);
2089 tgt_dt = tgt->ltd_tgt;
2090 if (!tgt->ltd_active)
2091 /* this OSP doesn't feel well */
2094 if (tgt->ltd_statfs.os_state & OS_STATFS_NOCREATE)
2097 rc = dt_fid_alloc(env, tgt_dt, &fid, NULL, NULL);
2104 /* Can not allocate more stripes */
2105 if (j == lod->lod_remote_mdt_count) {
2106 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
2107 lod2obd(lod)->obd_name, stripe_count,
2112 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
2113 idx, stripe_idx, PFID(&fid));
2114 mdt_indices[stripe_idx] = idx;
2115 /* Set the start index for next stripe allocation */
2116 if (!is_specific && stripe_idx < stripe_count - 1) {
2118 * for large dir test, put all other slaves on one
2119 * remote MDT, otherwise we may save too many local
2120 * slave locks which will exceed RS_MAX_LOCKS.
2122 if (unlikely(CFS_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)))
2124 mdt_indices[stripe_idx + 1] = (idx + 1) %
2125 (lod->lod_remote_mdt_count + 1);
2127 /* tgt_dt and fid must be ready after search avaible OSP
2128 * in the above loop */
2129 LASSERT(tgt_dt != NULL);
2130 LASSERT(fid_is_sane(&fid));
2132 /* fail a remote stripe FID allocation */
2133 if (stripe_idx && CFS_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_FID))
2136 dto = dt_locate_at(env, tgt_dt, &fid,
2137 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
2144 stripes[stripe_idx] = dto;
2150 for (j = 1; j < stripe_idx; j++) {
2151 LASSERT(stripes[j] != NULL);
2152 dt_object_put(env, stripes[j]);
2158 static int lod_prep_md_striped_create(const struct lu_env *env,
2159 struct dt_object *dt,
2160 struct lu_attr *attr,
2161 const struct lmv_user_md_v1 *lum,
2162 struct dt_object_format *dof,
2165 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
2166 struct lod_object *lo = lod_dt_obj(dt);
2167 struct dt_object **stripes;
2168 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2169 struct lu_fid fid = { 0 };
2176 /* The lum has been verifed in lod_verify_md_striping */
2177 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
2178 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
2180 stripe_count = lo->ldo_dir_stripe_count;
2182 OBD_ALLOC_PTR_ARRAY(stripes, stripe_count);
2186 /* Allocate the first stripe locally */
2187 rc = dt_fid_alloc(env, lod->lod_child, &fid, NULL, NULL);
2191 stripes[0] = dt_locate_at(env, lod->lod_child, &fid,
2192 dt->do_lu.lo_dev->ld_site->ls_top_dev, &conf);
2193 if (IS_ERR(stripes[0]))
2194 GOTO(out, rc = PTR_ERR(stripes[0]));
2196 if (lo->ldo_dir_stripe_offset == LMV_OFFSET_DEFAULT) {
2197 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
2198 rc = lod_mdt_alloc_qos(env, lo, stripes, 1, stripe_count);
2200 rc = lod_mdt_alloc_rr(env, lo, stripes, 1,
2204 bool is_specific = false;
2206 OBD_ALLOC_PTR_ARRAY(idx_array, stripe_count);
2208 GOTO(out, rc = -ENOMEM);
2210 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
2212 for (i = 0; i < stripe_count; i++)
2214 le32_to_cpu(lum->lum_objects[i].lum_mds);
2217 /* stripe 0 is local */
2219 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
2220 rc = lod_mdt_alloc_specific(env, lo, stripes, idx_array,
2222 OBD_FREE_PTR_ARRAY(idx_array, stripe_count);
2230 lo->ldo_dir_striped = 1;
2231 lo->ldo_stripe = stripes;
2232 lo->ldo_dir_stripe_count = rc;
2233 lo->ldo_dir_stripes_allocated = stripe_count;
2235 lo->ldo_dir_stripe_loaded = 1;
2237 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2239 lod_striping_free(env, lo);
2245 if (!IS_ERR_OR_NULL(stripes[0]))
2246 dt_object_put(env, stripes[0]);
2247 for (i = 1; i < stripe_count; i++)
2248 LASSERT(!stripes[i]);
2249 OBD_FREE_PTR_ARRAY(stripes, stripe_count);
2256 * Alloc cached foreign LOV
2258 * \param[in] lo object
2259 * \param[in] size size of foreign LOV
2261 * \retval 0 on success
2262 * \retval negative if failed
2264 int lod_alloc_foreign_lov(struct lod_object *lo, size_t size)
2266 OBD_ALLOC_LARGE(lo->ldo_foreign_lov, size);
2267 if (lo->ldo_foreign_lov == NULL)
2269 lo->ldo_foreign_lov_size = size;
2270 lo->ldo_is_foreign = 1;
2276 * Free cached foreign LOV
2278 * \param[in] lo object
2280 void lod_free_foreign_lov(struct lod_object *lo)
2282 if (lo->ldo_foreign_lov != NULL)
2283 OBD_FREE_LARGE(lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
2284 lo->ldo_foreign_lov = NULL;
2285 lo->ldo_foreign_lov_size = 0;
2286 lo->ldo_is_foreign = 0;
2291 * Alloc cached foreign LMV
2293 * \param[in] lo object
2294 * \param[in] size size of foreign LMV
2296 * \retval 0 on success
2297 * \retval negative if failed
2299 static int lod_alloc_foreign_lmv(struct lod_object *lo, size_t size)
2301 OBD_ALLOC_LARGE(lo->ldo_foreign_lmv, size);
2302 if (lo->ldo_foreign_lmv == NULL)
2304 lo->ldo_foreign_lmv_size = size;
2305 lo->ldo_is_foreign = 1;
2310 static int lod_prep_md_replayed_create(const struct lu_env *env,
2311 struct dt_object *dt,
2312 struct lu_attr *attr,
2313 const struct lu_buf *lmv_buf,
2314 struct dt_object_format *dof,
2317 struct lod_object *lo = lod_dt_obj(dt);
2322 mutex_lock(&lo->ldo_layout_mutex);
2323 rc = lod_parse_dir_striping(env, lo, lmv_buf);
2325 lo->ldo_dir_stripe_loaded = 1;
2326 lo->ldo_dir_striped = 1;
2327 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2329 mutex_unlock(&lo->ldo_layout_mutex);
2336 * Free cached foreign LMV
2338 * \param[in] lo object
2340 static void lod_free_foreign_lmv(struct lod_object *lo)
2342 if (lo->ldo_foreign_lmv != NULL)
2343 OBD_FREE_LARGE(lo->ldo_foreign_lmv, lo->ldo_foreign_lmv_size);
2344 lo->ldo_foreign_lmv = NULL;
2345 lo->ldo_foreign_lmv_size = 0;
2346 lo->ldo_is_foreign = 0;
2350 * Declare create striped md object.
2352 * The function declares intention to create a striped directory. This is a
2353 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2354 * is to verify pattern \a lum_buf is good. Check that function for the details.
2356 * \param[in] env execution environment
2357 * \param[in] dt object
2358 * \param[in] attr attributes to initialize the objects with
2359 * \param[in] lum_buf a pattern specifying the number of stripes and
2361 * \param[in] dof type of objects to be created
2362 * \param[in] th transaction handle
2364 * \retval 0 on success
2365 * \retval negative if failed
2368 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2369 struct dt_object *dt,
2370 struct lu_attr *attr,
2371 const struct lu_buf *lum_buf,
2372 struct dt_object_format *dof,
2375 struct lod_object *lo = lod_dt_obj(dt);
2376 struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
2380 LASSERT(lum != NULL);
2383 "lum magic=%x hash=%x count=%u offset=%d inherit=%u rr=%u\n",
2384 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_hash_type),
2385 le32_to_cpu(lum->lum_stripe_count),
2386 (int)le32_to_cpu(lum->lum_stripe_offset),
2387 lum->lum_max_inherit, lum->lum_max_inherit_rr);
2389 if (lo->ldo_dir_stripe_count == 0) {
2390 if (lo->ldo_is_foreign) {
2391 rc = lod_alloc_foreign_lmv(lo, lum_buf->lb_len);
2394 memcpy(lo->ldo_foreign_lmv, lum, lum_buf->lb_len);
2395 lo->ldo_dir_stripe_loaded = 1;
2400 /* client replay striped directory creation with LMV, this happens when
2401 * all involved MDTs were rebooted, or MDT recovery was aborted.
2403 if (le32_to_cpu(lum->lum_magic) == LMV_MAGIC_V1)
2404 rc = lod_prep_md_replayed_create(env, dt, attr, lum_buf, dof,
2407 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2409 /* failed to create striping, let's reset
2410 * config so that others don't get confused */
2411 lod_striping_free(env, lo);
2417 * Set or replace striped directory layout, and LFSCK may set layout on a plain
2418 * directory, so don't check stripe count.
2420 * \param[in] env execution environment
2421 * \param[in] dt target object
2422 * \param[in] lmv_buf LMV buf which contains source stripe FIDs
2423 * \param[in] fl set or replace
2424 * \param[in] th transaction handle
2426 * \retval 0 on success
2427 * \retval negative if failed
2429 static int lod_dir_layout_set(const struct lu_env *env,
2430 struct dt_object *dt,
2431 const struct lu_buf *lmv_buf,
2435 struct dt_object *next = dt_object_child(dt);
2436 struct lod_object *lo = lod_dt_obj(dt);
2437 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2438 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
2439 struct lmv_mds_md_v1 *slave_lmv;
2440 struct lu_buf slave_buf;
2446 if (!lmv_is_sane2(lmv))
2449 /* adjust hash for dir merge, which may not be set in user command */
2450 if (lmv_is_merging(lmv) &&
2451 !(lmv->lmv_migrate_hash & LMV_HASH_TYPE_MASK))
2452 lmv->lmv_merge_hash |=
2453 lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern &
2456 LMV_DEBUG(D_INFO, lmv, "set");
2458 rc = lod_sub_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV, fl, th);
2462 /* directory restripe may update stripe LMV directly */
2463 if (!lo->ldo_dir_stripe_count)
2466 lo->ldo_dir_hash_type = le32_to_cpu(lmv->lmv_hash_type);
2467 lo->ldo_dir_migrate_offset = le32_to_cpu(lmv->lmv_migrate_offset);
2468 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_migrate_hash);
2469 lo->ldo_dir_layout_version = le32_to_cpu(lmv->lmv_layout_version);
2471 OBD_ALLOC_PTR(slave_lmv);
2475 lod_prep_slave_lmv_md(slave_lmv, lmv);
2476 slave_buf.lb_buf = slave_lmv;
2477 slave_buf.lb_len = sizeof(*slave_lmv);
2479 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2480 if (!lo->ldo_stripe[i])
2483 if (!dt_object_exists(lo->ldo_stripe[i]))
2486 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], &slave_buf,
2487 XATTR_NAME_LMV, fl, th);
2492 OBD_FREE_PTR(slave_lmv);
2498 * Implementation of dt_object_operations::do_declare_xattr_set.
2500 * Used with regular (non-striped) objects. Basically it
2501 * initializes the striping information and applies the
2502 * change to all the stripes.
2504 * \see dt_object_operations::do_declare_xattr_set() in the API description
2507 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2508 struct dt_object *dt,
2509 const struct lu_buf *buf,
2510 const char *name, int fl,
2513 struct dt_object *next = dt_object_child(dt);
2514 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2515 struct lod_object *lo = lod_dt_obj(dt);
2520 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2521 struct lmv_user_md_v1 *lum;
2523 LASSERT(buf != NULL && buf->lb_buf != NULL);
2525 rc = lod_verify_md_striping(d, lum);
2528 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2529 rc = lod_verify_striping(env, d, lo, buf, false);
2534 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2538 /* Note: Do not set LinkEA on sub-stripes, otherwise
2539 * it will confuse the fid2path process(see mdt_path_current()).
2540 * The linkEA between master and sub-stripes is set in
2541 * lod_xattr_set_lmv(). */
2542 if (strcmp(name, XATTR_NAME_LINK) == 0)
2545 /* set xattr to each stripes, if needed */
2546 rc = lod_striping_load(env, lo);
2550 if (lo->ldo_dir_stripe_count == 0)
2553 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2554 if (!lo->ldo_stripe[i])
2557 if (!dt_object_exists(lo->ldo_stripe[i]))
2560 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2570 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2571 struct lod_object *lo,
2572 struct dt_object *dt, struct thandle *th,
2573 int comp_idx, int stripe_idx,
2574 struct lod_obj_stripe_cb_data *data)
2576 struct lod_thread_info *info = lod_env_info(env);
2577 struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
2578 struct filter_fid *ff = &info->lti_ff;
2579 struct lu_buf *buf = &info->lti_buf;
2583 buf->lb_len = sizeof(*ff);
2584 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2592 * locd_buf is set if it's called by dir migration, which doesn't check
2595 if (data->locd_buf) {
2596 memset(ff, 0, sizeof(*ff));
2597 ff->ff_parent = *(struct lu_fid *)data->locd_buf->lb_buf;
2599 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
2601 if (lu_fid_eq(lod_object_fid(lo), &ff->ff_parent) &&
2602 ff->ff_layout.ol_comp_id == comp->llc_id)
2605 memset(ff, 0, sizeof(*ff));
2606 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2609 /* rewrite filter_fid */
2610 ff->ff_parent.f_ver = stripe_idx;
2611 ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
2612 ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
2613 ff->ff_layout.ol_comp_id = comp->llc_id;
2614 ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
2615 ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
2616 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
2618 if (data->locd_declare)
2619 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2620 LU_XATTR_REPLACE, th);
2622 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2623 LU_XATTR_REPLACE, th);
2629 * Reset parent FID on OST object
2631 * Replace parent FID with @dt object FID, which is only called during migration
2632 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2633 * the FID is changed.
2635 * \param[in] env execution environment
2636 * \param[in] dt dt_object whose stripes's parent FID will be reset
2637 * \parem[in] th thandle
2638 * \param[in] declare if it is declare
2640 * \retval 0 if reset succeeds
2641 * \retval negative errno if reset fails
2643 static int lod_replace_parent_fid(const struct lu_env *env,
2644 struct dt_object *dt,
2645 const struct lu_buf *buf,
2646 struct thandle *th, bool declare)
2648 struct lod_object *lo = lod_dt_obj(dt);
2649 struct lod_thread_info *info = lod_env_info(env);
2650 struct filter_fid *ff;
2651 struct lod_obj_stripe_cb_data data = { { 0 } };
2655 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2657 /* set xattr to each stripes, if needed */
2658 rc = lod_striping_load(env, lo);
2662 if (!lod_obj_is_striped(dt))
2665 if (info->lti_ea_store_size < sizeof(*ff)) {
2666 rc = lod_ea_store_resize(info, sizeof(*ff));
2671 data.locd_declare = declare;
2672 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2673 data.locd_buf = buf;
2674 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2679 __u16 lod_comp_entry_stripe_count(struct lod_object *lo, int comp_idx,
2682 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2683 struct lod_layout_component *entry;
2684 enum lod_uses_hint flags = LOD_USES_ASSIGNED_STRIPE;
2689 entry = &lo->ldo_comp_entries[comp_idx];
2690 if (lod_comp_inited(entry))
2691 return entry->llc_stripe_count;
2692 if (entry->llc_stripe_count == LOV_ALL_STRIPES)
2693 return lod_get_stripe_count_plain(lod, lo,
2694 entry->llc_stripe_count,
2695 entry->llc_pattern &
2696 LOV_PATTERN_OVERSTRIPING,
2699 return lod_get_stripe_count(lod, lo, comp_idx, entry->llc_stripe_count,
2700 entry->llc_pattern & LOV_PATTERN_OVERSTRIPING,
2704 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2706 int magic, size = 0, i;
2707 struct lod_layout_component *comp_entries;
2709 bool is_composite, is_foreign = false;
2712 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2713 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2715 lo->ldo_def_striping->lds_def_striping_is_composite;
2717 comp_cnt = lo->ldo_comp_cnt;
2718 comp_entries = lo->ldo_comp_entries;
2719 is_composite = lo->ldo_is_composite;
2720 is_foreign = lo->ldo_is_foreign;
2724 return lo->ldo_foreign_lov_size;
2726 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2728 size = sizeof(struct lov_comp_md_v1) +
2729 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2730 LASSERT(size % sizeof(__u64) == 0);
2733 for (i = 0; i < comp_cnt; i++) {
2736 if (comp_entries[i].llc_magic == LOV_MAGIC_FOREIGN) {
2737 size += lov_foreign_md_size(comp_entries[i].llc_length);
2739 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 :
2741 stripe_count = lod_comp_entry_stripe_count(lo, i,
2743 if (!is_dir && is_composite)
2744 lod_comp_shrink_stripe_count(&comp_entries[i],
2747 size += lov_user_md_size(stripe_count, magic);
2749 LASSERT(size % sizeof(__u64) == 0);
2755 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2756 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2759 * \param[in] env execution environment
2760 * \param[in] dt dt_object to add components on
2761 * \param[in] buf buffer contains components to be added
2762 * \parem[in] th thandle
2764 * \retval 0 on success
2765 * \retval negative errno on failure
2767 static int lod_declare_layout_add(const struct lu_env *env,
2768 struct dt_object *dt,
2769 const struct lu_buf *buf,
2772 struct lod_thread_info *info = lod_env_info(env);
2773 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2774 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2775 struct dt_object *next = dt_object_child(dt);
2776 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
2777 struct lod_object *lo = lod_dt_obj(dt);
2778 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2780 int i, rc, array_cnt, old_array_cnt;
2783 LASSERT(lo->ldo_is_composite);
2785 if (lo->ldo_flr_state != LCM_FL_NONE)
2788 rc = lod_verify_striping(env, d, lo, buf, false);
2792 magic = comp_v1->lcm_magic;
2793 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2794 lustre_swab_lov_comp_md_v1(comp_v1);
2795 magic = comp_v1->lcm_magic;
2798 if (magic != LOV_USER_MAGIC_COMP_V1)
2801 mutex_lock(&lo->ldo_layout_mutex);
2803 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2804 OBD_ALLOC_PTR_ARRAY(comp_array, array_cnt);
2805 if (comp_array == NULL) {
2806 mutex_unlock(&lo->ldo_layout_mutex);
2811 memcpy(comp_array, lo->ldo_comp_entries,
2812 sizeof(*comp_array) * lo->ldo_comp_cnt);
2814 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2815 struct lov_user_md_v1 *v1;
2816 struct lu_extent *ext;
2818 v1 = (struct lov_user_md *)((char *)comp_v1 +
2819 comp_v1->lcm_entries[i].lcme_offset);
2820 ext = &comp_v1->lcm_entries[i].lcme_extent;
2822 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2823 lod_comp->llc_extent.e_start = ext->e_start;
2824 lod_comp->llc_extent.e_end = ext->e_end;
2825 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2826 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2828 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2829 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2831 * limit stripe count so that it's less than/equal to
2832 * extent_size / stripe_size.
2834 * Note: extension size reused llc_stripe_size field and
2835 * uninstantiated component could be defined with
2836 * extent_start == extent_end as extension component will
2839 if (!(lod_comp->llc_flags & LCME_FL_EXTENSION) &&
2840 (lod_comp_inited(lod_comp) ||
2841 lod_comp->llc_extent.e_start <
2842 lod_comp->llc_extent.e_end) &&
2843 lod_comp->llc_stripe_count != LOV_ALL_STRIPES &&
2844 ext->e_end != OBD_OBJECT_EOF &&
2845 (__u64)(lod_comp->llc_stripe_count *
2846 lod_comp->llc_stripe_size) >
2847 (ext->e_end - ext->e_start))
2848 lod_comp->llc_stripe_count =
2849 DIV_ROUND_UP(ext->e_end - ext->e_start,
2850 lod_comp->llc_stripe_size);
2851 lod_adjust_stripe_info(lod_comp, desc, 0);
2853 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2854 struct lov_user_md_v3 *v3 = (typeof(*v3) *) v1;
2856 if (v3->lmm_pool_name[0] != '\0' &&
2857 !lov_pool_is_ignored(v3->lmm_pool_name)) {
2858 rc = lod_set_pool(&lod_comp->llc_pool,
2866 old_array = lo->ldo_comp_entries;
2867 old_array_cnt = lo->ldo_comp_cnt;
2869 lo->ldo_comp_entries = comp_array;
2870 lo->ldo_comp_cnt = array_cnt;
2872 /* No need to increase layout generation here, it will be increased
2873 * later when generating component ID for the new components */
2875 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2876 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2877 XATTR_NAME_LOV, 0, th);
2879 lo->ldo_comp_entries = old_array;
2880 lo->ldo_comp_cnt = old_array_cnt;
2884 OBD_FREE_PTR_ARRAY(old_array, old_array_cnt);
2886 LASSERT(lo->ldo_mirror_count == 1);
2887 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2889 mutex_unlock(&lo->ldo_layout_mutex);
2894 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2895 lod_comp = &comp_array[i];
2896 if (lod_comp->llc_pool != NULL) {
2897 OBD_FREE(lod_comp->llc_pool,
2898 strlen(lod_comp->llc_pool) + 1);
2899 lod_comp->llc_pool = NULL;
2902 OBD_FREE_PTR_ARRAY(comp_array, array_cnt);
2903 mutex_unlock(&lo->ldo_layout_mutex);
2909 * lod_last_non_stale_mirror() - Check if a mirror is the last non-stale mirror.
2910 * @mirror_id: Mirror id to be checked.
2913 * This function checks if a mirror with specified @mirror_id is the last
2914 * non-stale mirror of a LOD object @lo.
2916 * Return: true or false.
2919 bool lod_last_non_stale_mirror(__u16 mirror_id, struct lod_object *lo)
2921 struct lod_layout_component *lod_comp;
2922 bool has_stale_flag;
2925 for (i = 0; i < lo->ldo_mirror_count; i++) {
2926 if (lo->ldo_mirrors[i].lme_id == mirror_id ||
2927 lo->ldo_mirrors[i].lme_stale)
2930 has_stale_flag = false;
2931 lod_foreach_mirror_comp(lod_comp, lo, i) {
2932 if (lod_comp->llc_flags & LCME_FL_STALE) {
2933 has_stale_flag = true;
2937 if (!has_stale_flag)
2945 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2946 * the '$field' can only be 'flags' now. The xattr value is binary
2947 * lov_comp_md_v1 which contains the component ID(s) and the value of
2948 * the field to be modified.
2949 * Please update allowed_lustre_lov macro if $field groks more values
2952 * \param[in] env execution environment
2953 * \param[in] dt dt_object to be modified
2954 * \param[in] op operation string, like "set.flags"
2955 * \param[in] buf buffer contains components to be set
2956 * \parem[in] th thandle
2958 * \retval 0 on success
2959 * \retval negative errno on failure
2961 static int lod_declare_layout_set(const struct lu_env *env,
2962 struct dt_object *dt,
2963 char *op, const struct lu_buf *buf,
2966 struct lod_layout_component *lod_comp;
2967 struct lod_thread_info *info = lod_env_info(env);
2968 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2969 struct lod_object *lo = lod_dt_obj(dt);
2970 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2973 bool changed = false;
2976 /* Please update allowed_lustre_lov macro if op
2977 * groks more values in the future
2979 if (strcmp(op, "set.flags") != 0) {
2980 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2981 lod2obd(d)->obd_name, op);
2985 magic = comp_v1->lcm_magic;
2986 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2987 lustre_swab_lov_comp_md_v1(comp_v1);
2988 magic = comp_v1->lcm_magic;
2991 if (magic != LOV_USER_MAGIC_COMP_V1)
2994 if (comp_v1->lcm_entry_count == 0) {
2995 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2996 lod2obd(d)->obd_name);
3000 mutex_lock(&lo->ldo_layout_mutex);
3001 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
3002 __u32 id = comp_v1->lcm_entries[i].lcme_id;
3003 __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
3004 __u32 mirror_flag = flags & LCME_MIRROR_FLAGS;
3005 __u16 mirror_id = mirror_id_of(id);
3006 bool neg = flags & LCME_FL_NEG;
3008 if (flags & LCME_FL_INIT) {
3010 lod_striping_free_nolock(env, lo);
3011 mutex_unlock(&lo->ldo_layout_mutex);
3015 flags &= ~(LCME_MIRROR_FLAGS | LCME_FL_NEG);
3016 for (j = 0; j < lo->ldo_comp_cnt; j++) {
3017 lod_comp = &lo->ldo_comp_entries[j];
3019 /* lfs only put one flag in each entry */
3020 if ((flags && id != lod_comp->llc_id) ||
3021 (mirror_flag && mirror_id !=
3022 mirror_id_of(lod_comp->llc_id)))
3027 lod_comp->llc_flags &= ~flags;
3029 lod_comp->llc_flags &= ~mirror_flag;
3032 if ((flags & LCME_FL_STALE) &&
3033 lod_last_non_stale_mirror(mirror_id,
3036 &lo->ldo_layout_mutex);
3039 lod_comp->llc_flags |= flags;
3042 lod_comp->llc_flags |= mirror_flag;
3043 if (mirror_flag & LCME_FL_NOSYNC)
3044 lod_comp->llc_timestamp =
3045 ktime_get_real_seconds();
3051 mutex_unlock(&lo->ldo_layout_mutex);
3054 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
3055 lod2obd(d)->obd_name);
3059 lod_obj_inc_layout_gen(lo);
3061 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3062 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
3063 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3068 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
3069 * and the xattr value is a unique component ID or a special lcme_id.
3071 * \param[in] env execution environment
3072 * \param[in] dt dt_object to be operated on
3073 * \param[in] buf buffer contains component ID or lcme_id
3074 * \parem[in] th thandle
3076 * \retval 0 on success
3077 * \retval negative errno on failure
3079 static int lod_declare_layout_del(const struct lu_env *env,
3080 struct dt_object *dt,
3081 const struct lu_buf *buf,
3084 struct lod_thread_info *info = lod_env_info(env);
3085 struct dt_object *next = dt_object_child(dt);
3086 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3087 struct lod_object *lo = lod_dt_obj(dt);
3088 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3089 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3090 __u32 magic, id, flags, neg_flags = 0;
3094 LASSERT(lo->ldo_is_composite);
3096 if (lo->ldo_flr_state != LCM_FL_NONE)
3099 magic = comp_v1->lcm_magic;
3100 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
3101 lustre_swab_lov_comp_md_v1(comp_v1);
3102 magic = comp_v1->lcm_magic;
3105 if (magic != LOV_USER_MAGIC_COMP_V1)
3108 id = comp_v1->lcm_entries[0].lcme_id;
3109 flags = comp_v1->lcm_entries[0].lcme_flags;
3111 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
3112 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
3113 lod2obd(d)->obd_name, id, flags);
3117 if (id != LCME_ID_INVAL && flags != 0) {
3118 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
3119 lod2obd(d)->obd_name);
3123 if (id == LCME_ID_INVAL && !flags) {
3124 CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
3125 lod2obd(d)->obd_name);
3129 if (flags & LCME_FL_NEG) {
3130 neg_flags = flags & ~LCME_FL_NEG;
3134 mutex_lock(&lo->ldo_layout_mutex);
3136 left = lo->ldo_comp_cnt;
3138 mutex_unlock(&lo->ldo_layout_mutex);
3142 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3143 struct lod_layout_component *lod_comp;
3145 lod_comp = &lo->ldo_comp_entries[i];
3147 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
3149 else if (flags && !(flags & lod_comp->llc_flags))
3151 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
3154 if (left != (i + 1)) {
3155 CDEBUG(D_LAYOUT, "%s: this deletion will create "
3156 "a hole.\n", lod2obd(d)->obd_name);
3157 mutex_unlock(&lo->ldo_layout_mutex);
3162 /* Mark the component as deleted */
3163 lod_comp->llc_id = LCME_ID_INVAL;
3165 /* Not instantiated component */
3166 if (lod_comp->llc_stripe == NULL)
3169 LASSERT(lod_comp->llc_stripe_count > 0);
3170 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
3171 struct dt_object *obj = lod_comp->llc_stripe[j];
3175 rc = lod_sub_declare_destroy(env, obj, th);
3177 mutex_unlock(&lo->ldo_layout_mutex);
3183 LASSERTF(left >= 0, "left = %d\n", left);
3184 if (left == lo->ldo_comp_cnt) {
3185 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
3186 lod2obd(d)->obd_name, id);
3187 mutex_unlock(&lo->ldo_layout_mutex);
3191 mutex_unlock(&lo->ldo_layout_mutex);
3193 memset(attr, 0, sizeof(*attr));
3194 attr->la_valid = LA_SIZE;
3195 rc = lod_sub_declare_attr_set(env, next, attr, th);
3200 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3201 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
3202 XATTR_NAME_LOV, 0, th);
3204 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
3211 * Declare layout add/set/del operations issued by special xattr names:
3213 * XATTR_LUSTRE_LOV.add add component(s) to existing file
3214 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
3215 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
3217 * \param[in] env execution environment
3218 * \param[in] dt object
3219 * \param[in] name name of xattr
3220 * \param[in] buf lu_buf contains xattr value
3221 * \param[in] th transaction handle
3223 * \retval 0 on success
3224 * \retval negative if failed
3226 static int lod_declare_modify_layout(const struct lu_env *env,
3227 struct dt_object *dt,
3229 const struct lu_buf *buf,
3232 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3233 struct lod_object *lo = lod_dt_obj(dt);
3235 int rc, len = strlen(XATTR_LUSTRE_LOV);
3238 LASSERT(dt_object_exists(dt));
3240 if (strlen(name) <= len || name[len] != '.') {
3241 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
3242 lod2obd(d)->obd_name, name);
3247 rc = lod_striping_load(env, lo);
3251 /* the layout to be modified must be a composite layout */
3252 if (!lo->ldo_is_composite) {
3253 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
3254 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3255 GOTO(unlock, rc = -EINVAL);
3258 op = (char *)name + len;
3259 if (strcmp(op, "add") == 0) {
3260 rc = lod_declare_layout_add(env, dt, buf, th);
3261 } else if (strcmp(op, "del") == 0) {
3262 rc = lod_declare_layout_del(env, dt, buf, th);
3263 } else if (strncmp(op, "set", strlen("set")) == 0) {
3264 rc = lod_declare_layout_set(env, dt, op, buf, th);
3266 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
3267 lod2obd(d)->obd_name, name);
3268 GOTO(unlock, rc = -ENOTSUPP);
3272 lod_striping_free(env, lo);
3278 * Convert a plain file lov_mds_md to a composite layout.
3280 * \param[in,out] info the thread info::lti_ea_store buffer contains little
3281 * endian plain file layout
3283 * \retval 0 on success, <0 on failure
3285 static int lod_layout_convert(struct lod_thread_info *info)
3287 struct lov_mds_md *lmm = info->lti_ea_store;
3288 struct lov_mds_md *lmm_save;
3289 struct lov_comp_md_v1 *lcm;
3290 struct lov_comp_md_entry_v1 *lcme;
3296 /* realloc buffer to a composite layout which contains one component */
3297 blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3298 le32_to_cpu(lmm->lmm_magic));
3299 size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
3301 OBD_ALLOC_LARGE(lmm_save, blob_size);
3303 GOTO(out, rc = -ENOMEM);
3305 memcpy(lmm_save, lmm, blob_size);
3307 if (info->lti_ea_store_size < size) {
3308 rc = lod_ea_store_resize(info, size);
3313 lcm = info->lti_ea_store;
3314 memset(lcm, 0, sizeof(*lcm) + sizeof(*lcme));
3315 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
3316 lcm->lcm_size = cpu_to_le32(size);
3317 lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
3318 lmm_save->lmm_layout_gen));
3319 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
3320 lcm->lcm_entry_count = cpu_to_le16(1);
3322 lcme = &lcm->lcm_entries[0];
3323 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
3324 lcme->lcme_extent.e_start = 0;
3325 lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
3326 lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
3327 lcme->lcme_size = cpu_to_le32(blob_size);
3329 memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
3334 OBD_FREE_LARGE(lmm_save, blob_size);
3339 * Merge layouts to form a mirrored file.
3341 static int lod_declare_layout_merge(const struct lu_env *env,
3342 struct dt_object *dt,
3343 const struct lu_buf *mbuf,
3346 struct lod_thread_info *info = lod_env_info(env);
3347 struct lu_attr *layout_attr = &info->lti_layout_attr;
3348 struct lu_buf *buf = &info->lti_buf;
3349 struct lod_object *lo = lod_dt_obj(dt);
3350 struct lov_comp_md_v1 *lcm;
3351 struct lov_comp_md_v1 *cur_lcm;
3352 struct lov_comp_md_v1 *merge_lcm;
3353 struct lov_comp_md_entry_v1 *lcme;
3354 struct lov_mds_md_v1 *lmm;
3357 __u16 cur_entry_count;
3358 __u16 merge_entry_count;
3360 __u16 mirror_id = 0;
3367 merge_lcm = mbuf->lb_buf;
3368 if (mbuf->lb_len < sizeof(*merge_lcm))
3371 /* must be an existing layout from disk */
3372 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
3375 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
3377 /* do not allow to merge two mirrored files */
3378 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
3381 /* verify the target buffer */
3382 rc = lod_get_lov_ea(env, lo);
3384 RETURN(rc ? : -ENODATA);
3386 cur_lcm = info->lti_ea_store;
3387 switch (le32_to_cpu(cur_lcm->lcm_magic)) {
3390 rc = lod_layout_convert(info);
3392 case LOV_MAGIC_COMP_V1:
3402 /* info->lti_ea_store could be reallocated in lod_layout_convert() */
3403 cur_lcm = info->lti_ea_store;
3404 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
3406 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
3407 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
3408 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
3411 /* size of new layout */
3412 size = le32_to_cpu(cur_lcm->lcm_size) +
3413 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
3415 memset(buf, 0, sizeof(*buf));
3416 lu_buf_alloc(buf, size);
3417 if (buf->lb_buf == NULL)
3421 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
3423 offset = sizeof(*lcm) +
3424 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
3425 for (i = 0; i < cur_entry_count; i++) {
3426 struct lov_comp_md_entry_v1 *cur_lcme;
3428 lcme = &lcm->lcm_entries[i];
3429 cur_lcme = &cur_lcm->lcm_entries[i];
3431 lcme->lcme_offset = cpu_to_le32(offset);
3432 memcpy((char *)lcm + offset,
3433 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
3434 le32_to_cpu(lcme->lcme_size));
3436 offset += le32_to_cpu(lcme->lcme_size);
3438 if (mirror_count == 1 &&
3439 mirror_id_of(le32_to_cpu(lcme->lcme_id)) == 0) {
3440 /* Add mirror from a non-flr file, create new mirror ID.
3441 * Otherwise, keep existing mirror's component ID, used
3442 * for mirror extension.
3444 id = pflr_id(1, i + 1);
3445 lcme->lcme_id = cpu_to_le32(id);
3448 id = max(le32_to_cpu(lcme->lcme_id), id);
3451 mirror_id = mirror_id_of(id) + 1;
3453 /* check if first entry in new layout is DOM */
3454 lmm = (struct lov_mds_md_v1 *)((char *)merge_lcm +
3455 merge_lcm->lcm_entries[0].lcme_offset);
3456 merge_has_dom = lov_pattern(le32_to_cpu(lmm->lmm_pattern)) &
3459 for (i = 0; i < merge_entry_count; i++) {
3460 struct lov_comp_md_entry_v1 *merge_lcme;
3462 merge_lcme = &merge_lcm->lcm_entries[i];
3463 lcme = &lcm->lcm_entries[cur_entry_count + i];
3465 *lcme = *merge_lcme;
3466 lcme->lcme_offset = cpu_to_le32(offset);
3467 if (merge_has_dom && i == 0)
3468 lcme->lcme_flags |= cpu_to_le32(LCME_FL_STALE);
3470 id = pflr_id(mirror_id, i + 1);
3471 lcme->lcme_id = cpu_to_le32(id);
3473 memcpy((char *)lcm + offset,
3474 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
3475 le32_to_cpu(lcme->lcme_size));
3477 offset += le32_to_cpu(lcme->lcme_size);
3480 /* fixup layout information */
3481 lcm->lcm_size = cpu_to_le32(size);
3482 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
3483 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
3484 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NONE)
3485 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
3487 rc = lod_striping_reload(env, lo, buf, 0);
3491 lod_obj_inc_layout_gen(lo);
3492 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3494 /* transfer layout version to OST objects. */
3495 if (lo->ldo_mirror_count > 1) {
3496 struct lod_obj_stripe_cb_data data = { {0} };
3498 layout_attr->la_valid = LA_LAYOUT_VERSION;
3499 layout_attr->la_layout_version = 0;
3500 data.locd_attr = layout_attr;
3501 data.locd_declare = true;
3502 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
3503 rc = lod_obj_for_each_stripe(env, lo, th, &data);
3508 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
3509 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3517 * Split layouts, just set the LOVEA with the layout from mbuf.
3519 static int lod_declare_layout_split(const struct lu_env *env,
3520 struct dt_object *dt, const struct lu_buf *mbuf,
3523 struct lod_thread_info *info = lod_env_info(env);
3524 struct lu_attr *layout_attr = &info->lti_layout_attr;
3525 struct lod_object *lo = lod_dt_obj(dt);
3526 struct lov_comp_md_v1 *lcm = mbuf->lb_buf;
3530 rc = lod_striping_reload(env, lo, mbuf, LVF_ALL_STALE);
3534 lod_obj_inc_layout_gen(lo);
3535 /* fix on-disk layout gen */
3536 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3538 /* transfer layout version to OST objects. */
3539 if (lo->ldo_mirror_count > 1) {
3540 struct lod_obj_stripe_cb_data data = { {0} };
3542 layout_attr->la_valid = LA_LAYOUT_VERSION;
3543 layout_attr->la_layout_version = 0;
3544 data.locd_attr = layout_attr;
3545 data.locd_declare = true;
3546 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
3547 rc = lod_obj_for_each_stripe(env, lo, th, &data);
3552 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), mbuf,
3553 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3557 static int lod_layout_declare_or_purge_mirror(const struct lu_env *env,
3558 struct dt_object *dt, const struct lu_buf *buf,
3559 struct thandle *th, bool declare)
3561 struct lod_thread_info *info = lod_env_info(env);
3562 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3563 struct lod_object *lo = lod_dt_obj(dt);
3564 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3565 struct lov_comp_md_entry_v1 *entry;
3566 struct lov_mds_md_v1 *lmm;
3567 struct dt_object **sub_objs = NULL;
3568 int rc = 0, i, k, array_count = 0;
3573 * other ops (like lod_declare_destroy) could destroying sub objects
3576 mutex_lock(&lo->ldo_layout_mutex);
3579 /* prepare sub-objects array */
3580 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
3581 entry = &comp_v1->lcm_entries[i];
3583 if (!(entry->lcme_flags & LCME_FL_INIT))
3586 lmm = (struct lov_mds_md_v1 *)
3587 ((char *)comp_v1 + entry->lcme_offset);
3588 array_count += lmm->lmm_stripe_count;
3590 OBD_ALLOC_PTR_ARRAY(sub_objs, array_count);
3591 if (sub_objs == NULL) {
3592 mutex_unlock(&lo->ldo_layout_mutex);
3597 k = 0; /* sub_objs index */
3598 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
3599 struct lov_ost_data_v1 *objs;
3600 struct lu_object *o, *n;
3601 struct dt_object *dto;
3602 struct lu_device *nd;
3603 struct lov_mds_md_v3 *v3;
3607 entry = &comp_v1->lcm_entries[i];
3609 if (!(entry->lcme_flags & LCME_FL_INIT))
3612 lmm = (struct lov_mds_md_v1 *)
3613 ((char *)comp_v1 + entry->lcme_offset);
3614 v3 = (struct lov_mds_md_v3 *)lmm;
3615 if (lmm->lmm_magic == LOV_MAGIC_V3)
3616 objs = &v3->lmm_objects[0];
3618 objs = &lmm->lmm_objects[0];
3620 for (j = 0; j < lmm->lmm_stripe_count; j++) {
3621 idx = objs[j].l_ost_idx;
3622 rc = ostid_to_fid(&info->lti_fid, &objs[j].l_ost_oi,
3627 if (!fid_is_sane(&info->lti_fid)) {
3628 CERROR("%s: sub-object insane fid "DFID"\n",
3629 lod2obd(d)->obd_name,
3630 PFID(&info->lti_fid));
3631 GOTO(out, rc = -EINVAL);
3634 lod_getref(&d->lod_ost_descs);
3636 rc = validate_lod_and_idx(d, idx);
3638 lod_putref(d, &d->lod_ost_descs);
3642 nd = &OST_TGT(d, idx)->ltd_tgt->dd_lu_dev;
3643 lod_putref(d, &d->lod_ost_descs);
3645 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
3647 GOTO(out, rc = PTR_ERR(o));
3649 n = lu_object_locate(o->lo_header, nd->ld_type);
3651 lu_object_put(env, n);
3652 GOTO(out, rc = -ENOENT);
3655 dto = container_of(n, struct dt_object, do_lu);
3658 rc = lod_sub_declare_destroy(env, dto, th);
3659 dt_object_put(env, dto);
3664 * collect to-be-destroyed sub objects, the
3665 * reference would be released after actual
3671 } /* for each stripe */
3672 } /* for each component in the mirror */
3677 /* destroy the sub objects */
3678 for (; i < k; i++) {
3679 rc = lod_sub_destroy(env, sub_objs[i], th);
3682 dt_object_put(env, sub_objs[i]);
3686 * if a sub object destroy failed, we'd release sub objects
3687 * reference get from above sub_objs collection.
3690 dt_object_put(env, sub_objs[i]);
3692 OBD_FREE_PTR_ARRAY(sub_objs, array_count);
3694 mutex_unlock(&lo->ldo_layout_mutex);
3700 * Purge layouts, delete sub objects in the mirror stored in the vic_buf,
3701 * and set the LOVEA with the layout from mbuf.
3703 static int lod_declare_layout_purge(const struct lu_env *env,
3704 struct dt_object *dt, const struct lu_buf *buf,
3707 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3708 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3713 if (le32_to_cpu(comp_v1->lcm_magic) != LOV_MAGIC_COMP_V1) {
3714 CERROR("%s: invalid layout magic %#x != %#x\n",
3715 lod2obd(d)->obd_name, le32_to_cpu(comp_v1->lcm_magic),
3720 if (cpu_to_le32(LOV_MAGIC_COMP_V1) != LOV_MAGIC_COMP_V1)
3721 lustre_swab_lov_comp_md_v1(comp_v1);
3723 /* from now on, @buf contains cpu endian data */
3725 if (comp_v1->lcm_mirror_count != 0) {
3726 CERROR("%s: can only purge one mirror from "DFID"\n",
3727 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3731 /* delcare sub objects deletion in the mirror stored in @buf */
3732 rc = lod_layout_declare_or_purge_mirror(env, dt, buf, th, true);
3736 /* delete sub objects from the mirror stored in @buf */
3737 static int lod_layout_purge(const struct lu_env *env, struct dt_object *dt,
3738 const struct lu_buf *buf, struct thandle *th)
3743 rc = lod_layout_declare_or_purge_mirror(env, dt, buf, th, false);
3748 * Implementation of dt_object_operations::do_declare_xattr_set.
3750 * \see dt_object_operations::do_declare_xattr_set() in the API description
3753 * the extension to the API:
3754 * - declaring LOVEA requests striping creation
3755 * - LU_XATTR_REPLACE means layout swap
3757 static int lod_declare_xattr_set(const struct lu_env *env,
3758 struct dt_object *dt,
3759 const struct lu_buf *buf,
3760 const char *name, int fl,
3763 struct dt_object *next = dt_object_child(dt);
3764 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3769 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
3770 if ((S_ISREG(mode) || mode == 0) &&
3771 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE | LU_XATTR_SPLIT |
3773 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3774 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
3776 * this is a request to create object's striping.
3778 * allow to declare predefined striping on a new (!mode) object
3779 * which is supposed to be replay of regular file creation
3780 * (when LOV setting is declared)
3782 * LU_XATTR_REPLACE is set to indicate a layout swap
3784 if (dt_object_exists(dt)) {
3785 rc = dt_attr_get(env, next, attr);
3789 memset(attr, 0, sizeof(*attr));
3790 attr->la_valid = LA_TYPE | LA_MODE;
3791 attr->la_mode = S_IFREG;
3793 rc = lod_declare_striped_create(env, dt, attr, buf, th);
3794 } else if (fl & LU_XATTR_MERGE) {
3795 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3796 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3797 rc = lod_declare_layout_merge(env, dt, buf, th);
3798 } else if (fl & LU_XATTR_SPLIT) {
3799 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3800 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3801 rc = lod_declare_layout_split(env, dt, buf, th);
3802 } else if (fl & LU_XATTR_PURGE) {
3803 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3804 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3805 rc = lod_declare_layout_purge(env, dt, buf, th);
3806 } else if (S_ISREG(mode) &&
3807 strlen(name) >= sizeof(XATTR_LUSTRE_LOV) + 3 &&
3808 allowed_lustre_lov(name)) {
3810 * this is a request to modify object's striping.
3811 * add/set/del component(s).
3813 if (!dt_object_exists(dt))
3816 rc = lod_declare_modify_layout(env, dt, name, buf, th);
3817 } else if (S_ISDIR(mode)) {
3818 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
3819 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3820 rc = lod_replace_parent_fid(env, dt, buf, th, true);
3822 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
3829 * Apply xattr changes to the object.
3831 * Applies xattr changes to the object and the stripes if the latter exist.
3833 * \param[in] env execution environment
3834 * \param[in] dt object
3835 * \param[in] buf buffer pointing to the new value of xattr
3836 * \param[in] name name of xattr
3837 * \param[in] fl flags
3838 * \param[in] th transaction handle
3840 * \retval 0 on success
3841 * \retval negative if failed
3843 static int lod_xattr_set_internal(const struct lu_env *env,
3844 struct dt_object *dt,
3845 const struct lu_buf *buf,
3846 const char *name, int fl,
3849 struct dt_object *next = dt_object_child(dt);
3850 struct lod_object *lo = lod_dt_obj(dt);
3855 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3856 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3859 /* Note: Do not set LinkEA on sub-stripes, otherwise
3860 * it will confuse the fid2path process(see mdt_path_current()).
3861 * The linkEA between master and sub-stripes is set in
3862 * lod_xattr_set_lmv(). */
3863 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
3866 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3867 if (!lo->ldo_stripe[i])
3870 if (!dt_object_exists(lo->ldo_stripe[i]))
3873 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
3883 * Delete an extended attribute.
3885 * Deletes specified xattr from the object and the stripes if the latter exist.
3887 * \param[in] env execution environment
3888 * \param[in] dt object
3889 * \param[in] name name of xattr
3890 * \param[in] th transaction handle
3892 * \retval 0 on success
3893 * \retval negative if failed
3895 static int lod_xattr_del_internal(const struct lu_env *env,
3896 struct dt_object *dt,
3897 const char *name, struct thandle *th)
3899 struct dt_object *next = dt_object_child(dt);
3900 struct lod_object *lo = lod_dt_obj(dt);
3906 rc = lod_sub_xattr_del(env, next, name, th);
3907 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3910 if (lo->ldo_dir_stripe_count == 0)
3913 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3914 if (!lo->ldo_stripe[i])
3917 if (!dt_object_exists(lo->ldo_stripe[i]))
3920 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3929 * Set default striping on a directory.
3931 * Sets specified striping on a directory object unless it matches the default
3932 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3933 * EA. This striping will be used when regular file is being created in this
3936 * \param[in] env execution environment
3937 * \param[in] dt the striped object
3938 * \param[in] buf buffer with the striping
3939 * \param[in] name name of EA
3940 * \param[in] fl xattr flag (see OSD API description)
3941 * \param[in] th transaction handle
3943 * \retval 0 on success
3944 * \retval negative if failed
3946 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
3947 struct dt_object *dt,
3948 const struct lu_buf *buf,
3949 const char *name, int fl,
3952 struct lov_user_md_v1 *lum;
3953 struct lov_user_md_v3 *v3 = NULL;
3954 const char *pool_name = NULL;
3959 LASSERT(buf != NULL && buf->lb_buf != NULL);
3962 switch (lum->lmm_magic) {
3963 case LOV_USER_MAGIC_SPECIFIC:
3964 case LOV_USER_MAGIC_V3:
3966 if (lov_pool_is_reserved(v3->lmm_pool_name))
3967 memset(v3->lmm_pool_name, 0, sizeof(v3->lmm_pool_name));
3968 else if (v3->lmm_pool_name[0] != '\0')
3969 pool_name = v3->lmm_pool_name;
3971 case LOV_USER_MAGIC_V1:
3972 /* if { size, offset, count } = { 0, -1, 0 } and no pool
3973 * (i.e. all default values specified) then delete default
3974 * striping from dir. */
3976 "set default striping: sz %u # %u offset %d %s %s\n",
3977 (unsigned)lum->lmm_stripe_size,
3978 (unsigned)lum->lmm_stripe_count,
3979 (int)lum->lmm_stripe_offset,
3980 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
3982 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
3983 lum->lmm_stripe_count,
3984 lum->lmm_stripe_offset,
3987 case LOV_USER_MAGIC_COMP_V1:
3989 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lum;
3990 struct lov_comp_md_entry_v1 *lcme;
3993 comp_cnt = le16_to_cpu(lcm->lcm_entry_count);
3994 for (i = 0; i < comp_cnt; i++) {
3995 lcme = &lcm->lcm_entries[i];
3996 if (lcme->lcme_flags & cpu_to_le32(LCME_FL_EXTENSION)) {
3997 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
4006 CERROR("Invalid magic %x\n", lum->lmm_magic);
4011 rc = lod_xattr_del_internal(env, dt, name, th);
4015 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4021 static int lod_get_default_lov_striping(const struct lu_env *env,
4022 struct lod_object *lo,
4023 struct lod_default_striping *lds,
4024 struct dt_allocation_hint *ah);
4027 * Helper function to convert compound layout to compound layout with
4030 * Copy lcm_entries array of \a src to \a tgt. Replace lov_user_md_v1
4031 * components of \a src with lov_user_md_v3 using \a pool.
4033 * \param[in] src source layout
4034 * \param[in] pool pool to use in \a tgt
4035 * \param[out] tgt target layout
4037 static void embed_pool_to_comp_v1(const struct lov_comp_md_v1 *src,
4039 struct lov_comp_md_v1 *tgt)
4042 struct lov_user_md_v1 *lum;
4043 struct lov_user_md_v3 *lum3;
4044 struct lov_comp_md_entry_v1 *entry;
4048 entry = tgt->lcm_entries;
4050 for (i = 0; i < le16_to_cpu(src->lcm_entry_count); i++, entry++) {
4051 *entry = src->lcm_entries[i];
4052 offset = le32_to_cpu(src->lcm_entries[i].lcme_offset);
4053 entry->lcme_offset = cpu_to_le32(offset + shift);
4055 lum = (struct lov_user_md_v1 *)((char *)src + offset);
4056 lum3 = (struct lov_user_md_v3 *)((char *)tgt + offset + shift);
4057 *(struct lov_user_md_v1 *)lum3 = *lum;
4058 if (lum->lmm_pattern & cpu_to_le32(LOV_PATTERN_MDT)) {
4059 lum3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
4061 lum3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4062 entry->lcme_size = cpu_to_le32(sizeof(*lum3));
4063 strlcpy(lum3->lmm_pool_name, pool,
4064 sizeof(lum3->lmm_pool_name));
4065 shift += sizeof(*lum3) - sizeof(*lum);
4071 * Set default striping on a directory.
4073 * Sets specified striping on a directory object unless it matches the default
4074 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
4075 * EA. This striping will be used when regular file is being created in this
4077 * If current default striping includes a pool but specifed striping
4078 * does not - retain the pool if it exists.
4080 * \param[in] env execution environment
4081 * \param[in] dt the striped object
4082 * \param[in] buf buffer with the striping
4083 * \param[in] name name of EA
4084 * \param[in] fl xattr flag (see OSD API description)
4085 * \param[in] th transaction handle
4087 * \retval 0 on success
4088 * \retval negative if failed
4090 static int lod_xattr_set_default_lov_on_dir(const struct lu_env *env,
4091 struct dt_object *dt,
4092 const struct lu_buf *buf,
4093 const char *name, int fl,
4096 struct lod_default_striping *lds = lod_lds_buf_get(env);
4097 struct lov_user_md_v1 *v1 = buf->lb_buf;
4098 char pool[LOV_MAXPOOLNAME + 1];
4104 /* get existing striping config */
4105 rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds, NULL);
4109 memset(pool, 0, sizeof(pool));
4110 if (lds->lds_def_striping_set == 1)
4111 lod_layout_get_pool(lds->lds_def_comp_entries,
4112 lds->lds_def_comp_cnt, pool,
4115 is_del = LOVEA_DELETE_VALUES(v1->lmm_stripe_size,
4116 v1->lmm_stripe_count,
4117 v1->lmm_stripe_offset,
4120 /* Retain the pool name if it is not given */
4121 if (v1->lmm_magic == LOV_USER_MAGIC_V1 && pool[0] != '\0' &&
4123 struct lod_thread_info *info = lod_env_info(env);
4124 struct lov_user_md_v3 *v3 = info->lti_ea_store;
4126 memset(v3, 0, sizeof(*v3));
4127 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4128 v3->lmm_pattern = cpu_to_le32(v1->lmm_pattern);
4129 v3->lmm_stripe_count = cpu_to_le32(v1->lmm_stripe_count);
4130 v3->lmm_stripe_offset = cpu_to_le32(v1->lmm_stripe_offset);
4131 v3->lmm_stripe_size = cpu_to_le32(v1->lmm_stripe_size);
4133 strlcpy(v3->lmm_pool_name, pool, sizeof(v3->lmm_pool_name));
4135 info->lti_buf.lb_buf = v3;
4136 info->lti_buf.lb_len = sizeof(*v3);
4137 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4139 } else if (v1->lmm_magic == LOV_USER_MAGIC_COMP_V1 &&
4140 pool[0] != '\0' && !is_del) {
4142 * try to retain the pool from default layout if the
4143 * specified component layout does not provide pool
4146 struct lod_thread_info *info = lod_env_info(env);
4147 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
4148 struct lov_comp_md_v1 *comp_v1p;
4149 struct lov_user_md_v1 *lum;
4153 struct lov_comp_md_entry_v1 *entry;
4156 entry_count = le16_to_cpu(comp_v1->lcm_entry_count);
4157 size = sizeof(*comp_v1) +
4158 entry_count * sizeof(comp_v1->lcm_entries[0]);
4159 entry = comp_v1->lcm_entries;
4160 for (i = 0; i < entry_count; i++, entry++) {
4161 offset = le32_to_cpu(entry->lcme_offset);
4162 lum = (struct lov_user_md_v1 *)((char *)comp_v1 +
4164 if (le32_to_cpu(lum->lmm_magic) != LOV_USER_MAGIC_V1)
4165 /* the i-th component includes pool info */
4167 if (lum->lmm_pattern & cpu_to_le32(LOV_PATTERN_MDT))
4168 size += sizeof(struct lov_user_md_v1);
4170 size += sizeof(struct lov_user_md_v3);
4173 if (i == entry_count) {
4175 * re-compose the layout to include the pool for
4178 if (info->lti_ea_store_size < size)
4179 rc = lod_ea_store_resize(info, size);
4182 comp_v1p = info->lti_ea_store;
4183 *comp_v1p = *comp_v1;
4184 comp_v1p->lcm_size = cpu_to_le32(size);
4185 embed_pool_to_comp_v1(comp_v1, pool, comp_v1p);
4187 info->lti_buf.lb_buf = comp_v1p;
4188 info->lti_buf.lb_len = size;
4189 rc = lod_xattr_set_lov_on_dir(env, dt,
4194 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl,
4198 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl, th);
4201 if (lds->lds_def_striping_set == 1 && lds->lds_def_comp_entries != NULL)
4202 lod_free_def_comp_entries(lds);
4208 * Set default striping on a directory object.
4210 * Sets specified striping on a directory object unless it matches the default
4211 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
4212 * EA. This striping will be used when a new directory is being created in the
4215 * \param[in] env execution environment
4216 * \param[in] dt the striped object
4217 * \param[in] buf buffer with the striping
4218 * \param[in] name name of EA
4219 * \param[in] fl xattr flag (see OSD API description)
4220 * \param[in] th transaction handle
4222 * \retval 0 on success
4223 * \retval negative if failed
4225 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
4226 struct dt_object *dt,
4227 const struct lu_buf *buf,
4228 const char *name, int fl,
4231 struct lmv_user_md_v1 *lum;
4236 LASSERT(buf != NULL && buf->lb_buf != NULL);
4240 "set default stripe_count # %u stripe_offset %d hash %u\n",
4241 le32_to_cpu(lum->lum_stripe_count),
4242 (int)le32_to_cpu(lum->lum_stripe_offset),
4243 le32_to_cpu(lum->lum_hash_type));
4245 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
4246 le32_to_cpu(lum->lum_stripe_offset)) &&
4247 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
4248 rc = lod_xattr_del_internal(env, dt, name, th);
4252 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4261 * Turn directory into a striped directory.
4263 * During replay the client sends the striping created before MDT
4264 * failure, then the layer above LOD sends this defined striping
4265 * using ->do_xattr_set(), so LOD uses this method to replay creation
4266 * of the stripes. Notice the original information for the striping
4267 * (#stripes, FIDs, etc) was transferred in declare path.
4269 * \param[in] env execution environment
4270 * \param[in] dt the striped object
4271 * \param[in] buf buf lmv_user_md for create, or lmv_mds_md for replay
4272 * \param[in] name not used currently
4273 * \param[in] fl xattr flag (see OSD API description)
4274 * \param[in] th transaction handle
4276 * \retval 0 on success
4277 * \retval negative if failed
4279 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
4280 const struct lu_buf *buf, const char *name,
4281 int fl, struct thandle *th)
4283 struct lod_object *lo = lod_dt_obj(dt);
4284 struct lod_thread_info *info = lod_env_info(env);
4285 struct lu_attr *attr = &info->lti_attr;
4286 struct dt_object_format *dof = &info->lti_format;
4287 struct lu_buf lmv_buf;
4288 struct lu_buf slave_lmv_buf;
4289 struct lmv_user_md *lum = buf->lb_buf;
4290 struct lmv_mds_md_v1 *lmm;
4291 struct lmv_mds_md_v1 *slave_lmm = NULL;
4292 struct dt_insert_rec *rec = &info->lti_dt_rec;
4297 /* lum is used to know whether it's replay */
4299 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4302 /* The stripes are supposed to be allocated in declare phase,
4303 * if there are no stripes being allocated, it will skip */
4304 if (lo->ldo_dir_stripe_count == 0) {
4305 if (lo->ldo_is_foreign) {
4306 rc = lod_sub_xattr_set(env, dt_object_child(dt), buf,
4307 XATTR_NAME_LMV, fl, th);
4314 rc = dt_attr_get(env, dt_object_child(dt), attr);
4318 attr->la_valid &= LA_ATIME | LA_MTIME | LA_CTIME | LA_FLAGS |
4319 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
4320 dof->dof_type = DFT_DIR;
4322 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
4325 lmm = lmv_buf.lb_buf;
4327 OBD_ALLOC_PTR(slave_lmm);
4328 if (slave_lmm == NULL)
4331 lod_prep_slave_lmv_md(slave_lmm, lmm);
4332 slave_lmv_buf.lb_buf = slave_lmm;
4333 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
4335 rec->rec_type = S_IFDIR;
4336 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4337 struct dt_object *dto = lo->ldo_stripe[i];
4338 char *stripe_name = info->lti_key;
4339 struct lu_name *sname;
4340 struct linkea_data ldata = { NULL };
4341 struct lu_buf linkea_buf;
4342 bool stripe_created = false;
4344 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
4348 /* fail a remote stripe creation */
4349 if (i && CFS_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_CREATE))
4352 /* if it's replay by client request, and stripe exists on remote
4353 * MDT, it means mkdir was partially executed: stripe was
4354 * created on remote MDT successfully, but target not in last
4357 if (unlikely((le32_to_cpu(lum->lum_magic) == LMV_MAGIC_V1) &&
4358 dt_object_exists(dto) && dt_object_remote(dto)))
4359 stripe_created = true;
4361 /* don't create stripe if:
4362 * 1. it's source stripe of migrating directory
4363 * 2. it's existed stripe of splitting directory
4365 if ((lod_is_migrating(lo) && i >= lo->ldo_dir_migrate_offset) ||
4366 (lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) {
4367 if (!dt_object_exists(dto))
4368 GOTO(out, rc = -EINVAL);
4369 } else if (!stripe_created) {
4370 dt_write_lock(env, dto, DT_TGT_CHILD);
4371 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
4373 dt_write_unlock(env, dto);
4377 rc = lod_sub_ref_add(env, dto, th);
4378 dt_write_unlock(env, dto);
4382 rec->rec_fid = lu_object_fid(&dto->do_lu);
4383 rc = lod_sub_insert(env, dto,
4384 (const struct dt_rec *)rec,
4385 (const struct dt_key *)dot, th);
4390 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
4391 cfs_fail_val != i) {
4392 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
4394 slave_lmm->lmv_master_mdt_index =
4397 slave_lmm->lmv_master_mdt_index =
4400 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
4401 XATTR_NAME_LMV, 0, th);
4406 /* don't insert stripe if it's existed stripe of splitting
4407 * directory (this directory is striped).
4408 * NB, plain directory will insert itself as the first
4411 if (lod_is_splitting(lo) && lo->ldo_dir_split_offset > 1 &&
4412 lo->ldo_dir_split_offset > i)
4415 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
4417 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4418 PFID(lu_object_fid(&dto->do_lu)), i + 1);
4420 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4421 PFID(lu_object_fid(&dto->do_lu)), i);
4423 if (!stripe_created) {
4424 rec->rec_fid = lu_object_fid(&dt->do_lu);
4425 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
4426 (const struct dt_key *)dotdot, th);
4430 sname = lod_name_get(env, stripe_name,
4431 strlen(stripe_name));
4432 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
4433 sname, lu_object_fid(&dt->do_lu));
4437 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
4438 linkea_buf.lb_len = ldata.ld_leh->leh_len;
4439 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
4440 XATTR_NAME_LINK, 0, th);
4445 rec->rec_fid = lu_object_fid(&dto->do_lu);
4446 rc = lod_sub_insert(env, dt_object_child(dt),
4447 (const struct dt_rec *)rec,
4448 (const struct dt_key *)stripe_name, th);
4452 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
4457 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
4458 rc = lod_sub_xattr_set(env, dt_object_child(dt),
4459 &lmv_buf, XATTR_NAME_LMV, fl, th);
4461 if (slave_lmm != NULL)
4462 OBD_FREE_PTR(slave_lmm);
4468 * Helper function to declare/execute creation of a striped directory
4470 * Called in declare/create object path, prepare striping for a directory
4471 * and prepare defaults data striping for the objects to be created in
4472 * that directory. Notice the function calls "declaration" or "execution"
4473 * methods depending on \a declare param. This is a consequence of the
4474 * current approach while we don't have natural distributed transactions:
4475 * we basically execute non-local updates in the declare phase. So, the
4476 * arguments for the both phases are the same and this is the reason for
4477 * this function to exist.
4479 * \param[in] env execution environment
4480 * \param[in] dt object
4481 * \param[in] attr attributes the stripes will be created with
4482 * \param[in] lmu lmv_user_md if MDT indices are specified
4483 * \param[in] dof format of stripes (see OSD API description)
4484 * \param[in] th transaction handle
4485 * \param[in] declare where to call "declare" or "execute" methods
4487 * \retval 0 on success
4488 * \retval negative if failed
4490 static int lod_dir_striping_create_internal(const struct lu_env *env,
4491 struct dt_object *dt,
4492 struct lu_attr *attr,
4493 const struct lu_buf *lmu,
4494 struct dt_object_format *dof,
4498 struct lod_thread_info *info = lod_env_info(env);
4499 struct lod_object *lo = lod_dt_obj(dt);
4500 const struct lod_default_striping *lds = lo->ldo_def_striping;
4504 LASSERT(ergo(lds != NULL,
4505 lds->lds_def_striping_set ||
4506 lds->lds_dir_def_striping_set));
4509 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
4510 lo->ldo_dir_stripe_offset)) {
4512 /* mkdir by default LMV */
4513 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
4514 int stripe_count = lo->ldo_dir_stripe_count;
4516 if (info->lti_ea_store_size < sizeof(*v1)) {
4517 rc = lod_ea_store_resize(info, sizeof(*v1));
4520 v1 = info->lti_ea_store;
4523 memset(v1, 0, sizeof(*v1));
4524 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
4525 v1->lum_stripe_count = cpu_to_le32(stripe_count);
4526 v1->lum_stripe_offset =
4527 cpu_to_le32(lo->ldo_dir_stripe_offset);
4529 info->lti_buf.lb_buf = v1;
4530 info->lti_buf.lb_len = sizeof(*v1);
4531 lmu = &info->lti_buf;
4535 rc = lod_declare_xattr_set_lmv(env, dt, attr, lmu, dof,
4538 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
4542 } else if (lmu->lb_buf) {
4543 /* foreign LMV EA case */
4545 struct lmv_foreign_md *lfm = lmu->lb_buf;
4547 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN)
4548 rc = lod_declare_xattr_set_lmv(env, dt, attr,
4550 } else if (lo->ldo_is_foreign) {
4551 LASSERT(lo->ldo_foreign_lmv != NULL &&
4552 lo->ldo_foreign_lmv_size > 0);
4553 info->lti_buf.lb_buf = lo->ldo_foreign_lmv;
4554 info->lti_buf.lb_len = lo->ldo_foreign_lmv_size;
4555 lmu = &info->lti_buf;
4556 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
4561 /* Transfer default LMV striping from the parent */
4562 if (lds != NULL && lds->lds_dir_def_striping_set &&
4563 lds->lds_dir_def_max_inherit != LMV_INHERIT_END &&
4564 lds->lds_dir_def_max_inherit != LMV_INHERIT_NONE &&
4565 !(LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
4566 lds->lds_dir_def_stripe_offset) &&
4567 le32_to_cpu(lds->lds_dir_def_hash_type) !=
4568 LMV_HASH_TYPE_UNKNOWN)) {
4569 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
4571 if (info->lti_ea_store_size < sizeof(*v1)) {
4572 rc = lod_ea_store_resize(info, sizeof(*v1));
4575 v1 = info->lti_ea_store;
4578 memset(v1, 0, sizeof(*v1));
4579 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
4580 v1->lum_stripe_count =
4581 cpu_to_le32(lds->lds_dir_def_stripe_count);
4582 v1->lum_stripe_offset =
4583 cpu_to_le32(lds->lds_dir_def_stripe_offset);
4585 cpu_to_le32(lds->lds_dir_def_hash_type);
4586 v1->lum_max_inherit =
4587 lmv_inherit_next(lds->lds_dir_def_max_inherit);
4588 v1->lum_max_inherit_rr =
4589 lmv_inherit_rr_next(lds->lds_dir_def_max_inherit_rr);
4591 info->lti_buf.lb_buf = v1;
4592 info->lti_buf.lb_len = sizeof(*v1);
4594 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4595 XATTR_NAME_DEFAULT_LMV,
4598 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
4600 XATTR_NAME_DEFAULT_LMV, 0,
4606 /* Transfer default LOV striping from the parent */
4607 if (lds != NULL && lds->lds_def_striping_set &&
4608 lds->lds_def_comp_cnt != 0) {
4609 struct lov_mds_md *lmm;
4610 int lmm_size = lod_comp_md_size(lo, true);
4612 if (info->lti_ea_store_size < lmm_size) {
4613 rc = lod_ea_store_resize(info, lmm_size);
4617 lmm = info->lti_ea_store;
4619 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
4623 info->lti_buf.lb_buf = lmm;
4624 info->lti_buf.lb_len = lmm_size;
4627 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4628 XATTR_NAME_LOV, 0, th);
4630 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4631 XATTR_NAME_LOV, 0, th);
4636 /* ldo_def_striping is not allocated, clear after use, in case directory
4637 * layout is changed later.
4640 lo->ldo_def_striping = NULL;
4645 static int lod_declare_dir_striping_create(const struct lu_env *env,
4646 struct dt_object *dt,
4647 struct lu_attr *attr,
4649 struct dt_object_format *dof,
4652 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
4656 static int lod_dir_striping_create(const struct lu_env *env,
4657 struct dt_object *dt,
4658 struct lu_attr *attr,
4659 const struct lu_buf *lmu,
4660 struct dt_object_format *dof,
4663 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
4668 * Make LOV EA for striped object.
4670 * Generate striping information and store it in the LOV EA of the given
4671 * object. The caller must ensure nobody else is calling the function
4672 * against the object concurrently. The transaction must be started.
4673 * FLDB service must be running as well; it's used to map FID to the target,
4674 * which is stored in LOV EA.
4676 * \param[in] env execution environment for this thread
4677 * \param[in] lo LOD object
4678 * \param[in] th transaction handle
4680 * \retval 0 if LOV EA is stored successfully
4681 * \retval negative error number on failure
4683 static int lod_generate_and_set_lovea(const struct lu_env *env,
4684 struct lod_object *lo,
4687 struct lod_thread_info *info = lod_env_info(env);
4688 struct dt_object *next = dt_object_child(&lo->ldo_obj);
4689 struct lov_mds_md_v1 *lmm;
4695 if (lo->ldo_comp_cnt == 0 && !lo->ldo_is_foreign) {
4696 lod_striping_free_nolock(env, lo);
4697 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
4701 lmm_size = lod_comp_md_size(lo, false);
4702 if (info->lti_ea_store_size < lmm_size) {
4703 rc = lod_ea_store_resize(info, lmm_size);
4707 lmm = info->lti_ea_store;
4709 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
4713 info->lti_buf.lb_buf = lmm;
4714 info->lti_buf.lb_len = lmm_size;
4715 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
4716 XATTR_NAME_LOV, 0, th);
4720 static __u32 lod_gen_component_id(struct lod_object *lo,
4721 int mirror_id, int comp_idx);
4724 * Repeat an existing component
4726 * Creates a new layout by replicating an existing component. Uses striping
4727 * policy from previous component as a template for the striping for the new
4730 * New component starts with zero length, will be extended (or removed) before
4731 * returning layout to client.
4733 * NB: Reallocates layout components array (lo->ldo_comp_entries), invalidating
4734 * any pre-existing pointers to components. Handle with care.
4736 * \param[in] env execution environment for this thread
4737 * \param[in,out] lo object to update the layout of
4738 * \param[in] index index of component to copy
4740 * \retval 0 on success
4741 * \retval negative errno on error
4743 static int lod_layout_repeat_comp(const struct lu_env *env,
4744 struct lod_object *lo, int index)
4746 struct lod_layout_component *lod_comp;
4747 struct lod_layout_component *new_comp = NULL;
4748 struct lod_layout_component *comp_array;
4749 int rc = 0, i, new_cnt = lo->ldo_comp_cnt + 1;
4754 lod_comp = &lo->ldo_comp_entries[index];
4755 LASSERT(lod_comp_inited(lod_comp) && lod_comp->llc_id != LCME_ID_INVAL);
4757 CDEBUG(D_LAYOUT, "repeating component %d\n", index);
4759 OBD_ALLOC_PTR_ARRAY(comp_array, new_cnt);
4760 if (comp_array == NULL)
4761 GOTO(out, rc = -ENOMEM);
4763 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4764 memcpy(&comp_array[i + offset], &lo->ldo_comp_entries[i],
4765 sizeof(*comp_array));
4767 /* Duplicate this component in to the next slot */
4769 new_comp = &comp_array[i + 1];
4770 memcpy(&comp_array[i + 1], &lo->ldo_comp_entries[i],
4771 sizeof(*comp_array));
4772 /* We must now skip this new component when copying */
4777 /* Set up copied component */
4778 new_comp->llc_flags &= ~LCME_FL_INIT;
4779 new_comp->llc_stripe = NULL;
4780 new_comp->llc_stripes_allocated = 0;
4781 new_comp->llc_ost_indices = NULL;
4782 new_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4783 /* for uninstantiated components, layout gen stores default stripe
4785 new_comp->llc_layout_gen = lod_comp->llc_stripe_offset;
4786 /* This makes the repeated component zero-length, placed at the end of
4787 * the preceding component */
4788 new_comp->llc_extent.e_start = new_comp->llc_extent.e_end;
4789 new_comp->llc_timestamp = lod_comp->llc_timestamp;
4790 new_comp->llc_pool = NULL;
4792 rc = lod_set_pool(&new_comp->llc_pool, lod_comp->llc_pool);
4796 if (new_comp->llc_ostlist.op_array) {
4797 __u32 *op_array = NULL;
4799 OBD_ALLOC(op_array, new_comp->llc_ostlist.op_size);
4801 GOTO(out, rc = -ENOMEM);
4802 memcpy(op_array, &new_comp->llc_ostlist.op_array,
4803 new_comp->llc_ostlist.op_size);
4804 new_comp->llc_ostlist.op_array = op_array;
4807 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
4808 lo->ldo_comp_entries = comp_array;
4809 lo->ldo_comp_cnt = new_cnt;
4811 /* Generate an id for the new component */
4812 mirror_id = mirror_id_of(new_comp->llc_id);
4813 new_comp->llc_id = LCME_ID_INVAL;
4814 new_comp->llc_id = lod_gen_component_id(lo, mirror_id, index + 1);
4815 if (new_comp->llc_id == LCME_ID_INVAL)
4816 GOTO(out, rc = -ERANGE);
4821 OBD_FREE_PTR_ARRAY(comp_array, new_cnt);
4826 static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
4830 /* clear memory region that will be used for layout change */
4831 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
4832 info->lti_count = 0;
4834 if (info->lti_comp_size >= comp_cnt)
4837 if (info->lti_comp_size > 0) {
4838 OBD_FREE_PTR_ARRAY(info->lti_comp_idx, info->lti_comp_size);
4839 info->lti_comp_size = 0;
4842 OBD_ALLOC_PTR_ARRAY(info->lti_comp_idx, comp_cnt);
4843 if (!info->lti_comp_idx)
4846 info->lti_comp_size = comp_cnt;
4851 * Prepare new layout minus deleted components
4853 * Removes components marked for deletion (LCME_ID_INVAL) by copying to a new
4854 * layout and skipping those components. Removes stripe objects if any exist.
4857 * Reallocates layout components array (lo->ldo_comp_entries), invalidating
4858 * any pre-existing pointers to components.
4860 * Caller is responsible for updating mirror end (ldo_mirror[].lme_end).
4862 * \param[in] env execution environment for this thread
4863 * \param[in,out] lo object to update the layout of
4864 * \param[in] th transaction handle for this operation
4866 * \retval # of components deleted
4867 * \retval negative errno on error
4869 static int lod_layout_del_prep_layout(const struct lu_env *env,
4870 struct lod_object *lo,
4873 struct lod_layout_component *lod_comp;
4874 struct lod_thread_info *info = lod_env_info(env);
4875 int rc = 0, i, j, deleted = 0;
4879 LASSERT(lo->ldo_is_composite);
4880 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
4882 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
4886 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4887 lod_comp = &lo->ldo_comp_entries[i];
4889 if (lod_comp->llc_id != LCME_ID_INVAL) {
4890 /* Build array of things to keep */
4891 info->lti_comp_idx[info->lti_count++] = i;
4895 if (lod_comp->llc_magic == LOV_MAGIC_FOREIGN)
4898 lod_obj_set_pool(lo, i, NULL);
4899 if (lod_comp->llc_ostlist.op_array) {
4900 OBD_FREE(lod_comp->llc_ostlist.op_array,
4901 lod_comp->llc_ostlist.op_size);
4902 lod_comp->llc_ostlist.op_array = NULL;
4903 lod_comp->llc_ostlist.op_size = 0;
4907 CDEBUG(D_LAYOUT, "deleting comp %d, left %d\n", i,
4908 lo->ldo_comp_cnt - deleted);
4910 /* No striping info for this component */
4911 if (lod_comp->llc_stripe == NULL)
4914 LASSERT(lod_comp->llc_stripe_count > 0);
4915 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4916 struct dt_object *obj = lod_comp->llc_stripe[j];
4921 /* components which are not init have no sub objects
4923 if (lod_comp_inited(lod_comp)) {
4924 rc = lod_sub_destroy(env, obj, th);
4929 lu_object_put(env, &obj->do_lu);
4930 lod_comp->llc_stripe[j] = NULL;
4932 OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe,
4933 lod_comp->llc_stripes_allocated);
4934 lod_comp->llc_stripe = NULL;
4935 OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices,
4936 lod_comp->llc_stripes_allocated);
4937 lod_comp->llc_ost_indices = NULL;
4938 lod_comp->llc_stripes_allocated = 0;
4941 /* info->lti_count has the amount of left components */
4942 LASSERTF(info->lti_count >= 0 && info->lti_count < lo->ldo_comp_cnt,
4943 "left = %d, lo->ldo_comp_cnt %d\n", (int)info->lti_count,
4944 (int)lo->ldo_comp_cnt);
4946 if (info->lti_count > 0) {
4947 struct lod_layout_component *comp_array;
4949 OBD_ALLOC_PTR_ARRAY(comp_array, info->lti_count);
4950 if (comp_array == NULL)
4951 GOTO(out, rc = -ENOMEM);
4953 for (i = 0; i < info->lti_count; i++) {
4954 memcpy(&comp_array[i],
4955 &lo->ldo_comp_entries[info->lti_comp_idx[i]],
4956 sizeof(*comp_array));
4959 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
4960 lo->ldo_comp_entries = comp_array;
4961 lo->ldo_comp_cnt = info->lti_count;
4963 lod_free_comp_entries(lo);
4968 return rc ? rc : deleted;
4972 * Delete layout component(s)
4974 * This function sets up the layout data in the env and does the setattrs
4975 * required to write out the new layout. The layout itself is modified in
4976 * lod_layout_del_prep_layout.
4978 * \param[in] env execution environment for this thread
4979 * \param[in] dt object
4980 * \param[in] th transaction handle
4982 * \retval 0 on success
4983 * \retval negative error number on failure
4985 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
4988 struct lod_object *lo = lod_dt_obj(dt);
4989 struct dt_object *next = dt_object_child(dt);
4990 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4993 LASSERT(lo->ldo_mirror_count == 1);
4995 mutex_lock(&lo->ldo_layout_mutex);
4997 rc = lod_layout_del_prep_layout(env, lo, th);
5001 /* Only do this if we didn't delete all components */
5002 if (lo->ldo_comp_cnt > 0) {
5003 lo->ldo_mirrors[0].lme_end = lo->ldo_comp_cnt - 1;
5004 lod_obj_inc_layout_gen(lo);
5007 LASSERT(dt_object_exists(dt));
5008 rc = dt_attr_get(env, next, attr);
5012 if (attr->la_size > 0) {
5014 attr->la_valid = LA_SIZE;
5015 rc = lod_sub_attr_set(env, next, attr, th);
5020 rc = lod_generate_and_set_lovea(env, lo, th);
5024 lod_striping_free_nolock(env, lo);
5026 mutex_unlock(&lo->ldo_layout_mutex);
5033 * Implementation of dt_object_operations::do_xattr_set.
5035 * Sets specified extended attribute on the object. Three types of EAs are
5037 * LOV EA - stores striping for a regular file or default striping (when set
5039 * LMV EA - stores a marker for the striped directories
5040 * DMV EA - stores default directory striping
5042 * When striping is applied to a non-striped existing object (this is called
5043 * late striping), then LOD notices the caller wants to turn the object into a
5044 * striped one. The stripe objects are created and appropriate EA is set:
5045 * LOV EA storing all the stripes directly or LMV EA storing just a small header
5046 * with striping configuration.
5048 * \see dt_object_operations::do_xattr_set() in the API description for details.
5050 static int lod_xattr_set(const struct lu_env *env,
5051 struct dt_object *dt, const struct lu_buf *buf,
5052 const char *name, int fl, struct thandle *th)
5054 struct dt_object *next = dt_object_child(dt);
5055 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
5056 struct lod_object *lo = lod_dt_obj(dt);
5057 struct lod_obj_stripe_cb_data data = { {0} };
5062 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
5063 !strcmp(name, XATTR_NAME_LMV)) {
5065 case LU_XATTR_CREATE:
5066 rc = lod_dir_striping_create(env, dt, NULL, buf, NULL,
5070 case LU_XATTR_REPLACE:
5071 rc = lod_dir_layout_set(env, dt, buf, fl, th);
5078 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
5079 strcmp(name, XATTR_NAME_LOV) == 0) {
5080 rc = lod_xattr_set_default_lov_on_dir(env, dt, buf, name, fl,
5083 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
5084 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
5086 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
5089 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
5090 (strcmp(name, XATTR_NAME_LOV) == 0 ||
5091 strcmp(name, XATTR_LUSTRE_LOV) == 0 ||
5092 allowed_lustre_lov(name))) {
5093 /* in case of lov EA swap, just set it
5094 * if not, it is a replay so check striping match what we
5095 * already have during req replay, declare_xattr_set()
5096 * defines striping, then create() does the work */
5097 if (fl & LU_XATTR_REPLACE) {
5098 /* free stripes, then update disk */
5099 lod_striping_free(env, lod_dt_obj(dt));
5101 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
5102 } else if (fl & LU_XATTR_SPLIT) {
5103 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
5107 rc = lod_striping_reload(env, lo, buf, LVF_ALL_STALE);
5111 if (lo->ldo_mirror_count > 1 &&
5112 layout_attr->la_valid & LA_LAYOUT_VERSION) {
5114 layout_attr->la_layout_version =
5116 data.locd_attr = layout_attr;
5117 data.locd_declare = false;
5118 data.locd_stripe_cb =
5119 lod_obj_stripe_attr_set_cb;
5120 rc = lod_obj_for_each_stripe(env, lo, th,
5125 } else if (fl & LU_XATTR_PURGE) {
5126 rc = lod_layout_purge(env, dt, buf, th);
5127 } else if (dt_object_remote(dt)) {
5128 /* This only happens during migration, see
5129 * mdd_migrate_create(), in which Master MDT will
5130 * create a remote target object, and only set
5131 * (migrating) stripe EA on the remote object,
5132 * and does not need creating each stripes. */
5133 rc = lod_sub_xattr_set(env, next, buf, name,
5135 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
5136 /* delete component(s) */
5137 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
5138 rc = lod_layout_del(env, dt, th);
5141 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
5142 * it's going to create create file with specified
5143 * component(s), the striping must have not being
5144 * cached in this case;
5146 * Otherwise, it's going to add/change component(s) to
5147 * an existing file, the striping must have been cached
5150 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
5151 !strcmp(name, XATTR_NAME_LOV),
5152 !lod_dt_obj(dt)->ldo_comp_cached));
5154 rc = lod_striped_create(env, dt, NULL, NULL, th);
5158 if (fl & LU_XATTR_MERGE && lo->ldo_mirror_count > 1 &&
5159 layout_attr->la_valid & LA_LAYOUT_VERSION) {
5160 /* mirror merge exec phase */
5161 layout_attr->la_layout_version =
5163 data.locd_attr = layout_attr;
5164 data.locd_declare = false;
5165 data.locd_stripe_cb =
5166 lod_obj_stripe_attr_set_cb;
5167 rc = lod_obj_for_each_stripe(env, lo, th,
5174 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
5175 rc = lod_replace_parent_fid(env, dt, buf, th, false);
5180 /* then all other xattr */
5181 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
5187 * Implementation of dt_object_operations::do_declare_xattr_del.
5189 * \see dt_object_operations::do_declare_xattr_del() in the API description
5192 static int lod_declare_xattr_del(const struct lu_env *env,
5193 struct dt_object *dt, const char *name,
5196 struct lod_object *lo = lod_dt_obj(dt);
5197 struct dt_object *next = dt_object_child(dt);
5202 rc = lod_sub_declare_xattr_del(env, next, name, th);
5206 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
5209 /* NB: don't delete stripe LMV, because when we do this, normally we
5210 * will remove stripes, besides, if directory LMV is corrupt, this will
5211 * prevent deleting its LMV and fixing it (via LFSCK).
5213 if (!strcmp(name, XATTR_NAME_LMV))
5216 rc = lod_striping_load(env, lo);
5220 if (lo->ldo_dir_stripe_count == 0)
5223 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5224 struct dt_object *dto = lo->ldo_stripe[i];
5229 if (!dt_object_exists(dto))
5232 rc = lod_sub_declare_xattr_del(env, dto, name, th);
5241 * Implementation of dt_object_operations::do_xattr_del.
5243 * If EA storing a regular striping is being deleted, then release
5244 * all the references to the stripe objects in core.
5246 * \see dt_object_operations::do_xattr_del() in the API description for details.
5248 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
5249 const char *name, struct thandle *th)
5255 if (!strcmp(name, XATTR_NAME_LOV) || !strcmp(name, XATTR_NAME_LMV))
5256 lod_striping_free(env, lod_dt_obj(dt));
5258 rc = lod_xattr_del_internal(env, dt, name, th);
5264 * Implementation of dt_object_operations::do_xattr_list.
5266 * \see dt_object_operations::do_xattr_list() in the API description
5269 static int lod_xattr_list(const struct lu_env *env,
5270 struct dt_object *dt, const struct lu_buf *buf)
5272 return dt_xattr_list(env, dt_object_child(dt), buf);
5275 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
5277 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
5281 * Copy OST list from layout provided by user.
5283 * \param[in] lod_comp layout_component to be filled
5284 * \param[in] v3 LOV EA V3 user data
5286 * \retval 0 on success
5287 * \retval negative if failed
5289 int lod_comp_copy_ost_lists(struct lod_layout_component *lod_comp,
5290 struct lov_user_md_v3 *v3)
5296 if (v3->lmm_stripe_offset == LOV_OFFSET_DEFAULT)
5297 v3->lmm_stripe_offset = v3->lmm_objects[0].l_ost_idx;
5299 if (lod_comp->llc_ostlist.op_array) {
5300 if (lod_comp->llc_ostlist.op_size >=
5301 v3->lmm_stripe_count * sizeof(__u32)) {
5302 lod_comp->llc_ostlist.op_count =
5303 v3->lmm_stripe_count;
5306 OBD_FREE(lod_comp->llc_ostlist.op_array,
5307 lod_comp->llc_ostlist.op_size);
5310 /* copy ost list from lmm */
5311 lod_comp->llc_ostlist.op_count = v3->lmm_stripe_count;
5312 lod_comp->llc_ostlist.op_size = v3->lmm_stripe_count * sizeof(__u32);
5313 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
5314 lod_comp->llc_ostlist.op_size);
5315 if (!lod_comp->llc_ostlist.op_array)
5318 for (j = 0; j < v3->lmm_stripe_count; j++) {
5319 lod_comp->llc_ostlist.op_array[j] =
5320 v3->lmm_objects[j].l_ost_idx;
5328 * Get default striping.
5330 * \param[in] env execution environment
5331 * \param[in] lo object
5332 * \param[out] lds default striping
5334 * \retval 0 on success
5335 * \retval negative if failed
5337 static int lod_get_default_lov_striping(const struct lu_env *env,
5338 struct lod_object *lo,
5339 struct lod_default_striping *lds,
5340 struct dt_allocation_hint *dah)
5342 struct lod_thread_info *info = lod_env_info(env);
5343 struct lov_user_md_v1 *v1 = NULL;
5344 struct lov_user_md_v3 *v3 = NULL;
5345 struct lov_comp_md_v1 *lcm = NULL;
5347 int append_stripe_count = dah != NULL ? dah->dah_append_stripe_count : 0;
5348 const char *append_pool = (dah != NULL &&
5349 dah->dah_append_pool != NULL &&
5350 dah->dah_append_pool[0] != '\0') ?
5351 dah->dah_append_pool : NULL;
5352 __u16 entry_count = 1;
5353 __u16 mirror_count = 0;
5354 bool want_composite = false;
5359 lds->lds_def_striping_set = 0;
5361 rc = lod_get_lov_ea(env, lo);
5365 if (rc < (typeof(rc))sizeof(struct lov_user_md))
5368 magic = *(__u32 *)info->lti_ea_store;
5369 if (magic == __swab32(LOV_USER_MAGIC_V1)) {
5370 lustre_swab_lov_user_md_v1(info->lti_ea_store);
5371 } else if (magic == __swab32(LOV_USER_MAGIC_V3)) {
5372 lustre_swab_lov_user_md_v3(info->lti_ea_store);
5373 } else if (magic == __swab32(LOV_USER_MAGIC_SPECIFIC)) {
5374 v3 = (struct lov_user_md_v3 *)info->lti_ea_store;
5375 lustre_swab_lov_user_md_v3(v3);
5376 lustre_swab_lov_user_md_objects(v3->lmm_objects,
5377 v3->lmm_stripe_count);
5378 } else if (magic == __swab32(LOV_USER_MAGIC_COMP_V1) ||
5379 magic == __swab32(LOV_USER_MAGIC_SEL)) {
5380 lustre_swab_lov_comp_md_v1(info->lti_ea_store);
5386 case LOV_USER_MAGIC_SPECIFIC:
5387 v1 = info->lti_ea_store;
5389 case LOV_MAGIC_COMP_V1:
5391 lcm = info->lti_ea_store;
5392 entry_count = lcm->lcm_entry_count;
5393 if (entry_count == 0)
5396 mirror_count = lcm->lcm_mirror_count + 1;
5397 want_composite = true;
5403 if (append_stripe_count != 0 || append_pool != NULL) {
5406 want_composite = false;
5409 /* realloc default comp entries if necessary */
5410 rc = lod_def_striping_comp_resize(lds, entry_count);
5414 lds->lds_def_comp_cnt = entry_count;
5415 lds->lds_def_striping_is_composite = want_composite;
5416 lds->lds_def_mirror_cnt = mirror_count;
5418 for (i = 0; i < entry_count; i++) {
5419 struct lod_layout_component *llc = &lds->lds_def_comp_entries[i];
5423 * reset llc values, llc_stripes is always NULL in the
5424 * default striping template, llc_pool will be reset
5425 * later below using lod_set_pool().
5427 * XXX At this point llc_pool may point to valid (!)
5428 * kmalloced strings from previous RPCs.
5430 memset(llc, 0, offsetof(typeof(*llc), llc_pool));
5433 v1 = (struct lov_user_md *)((char *)lcm +
5434 lcm->lcm_entries[i].lcme_offset);
5436 if (want_composite) {
5437 llc->llc_extent = lcm->lcm_entries[i].lcme_extent;
5438 /* We only inherit certain flags from the layout */
5439 llc->llc_flags = lcm->lcm_entries[i].lcme_flags &
5440 LCME_TEMPLATE_FLAGS;
5444 CDEBUG(D_LAYOUT, DFID" magic = %#08x, pattern = %#x, stripe_count = %hu, stripe_size = %u, stripe_offset = %hu, append_pool = '%s', append_stripe_count = %d\n",
5445 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
5448 v1->lmm_stripe_count,
5449 v1->lmm_stripe_size,
5450 v1->lmm_stripe_offset,
5452 append_stripe_count);
5454 if (!lov_pattern_supported(v1->lmm_pattern) &&
5455 !(v1->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
5456 lod_free_def_comp_entries(lds);
5460 llc->llc_stripe_count = v1->lmm_stripe_count;
5461 llc->llc_stripe_size = v1->lmm_stripe_size;
5462 llc->llc_stripe_offset = v1->lmm_stripe_offset;
5463 llc->llc_pattern = v1->lmm_pattern;
5465 if (append_stripe_count != 0 || append_pool != NULL)
5466 llc->llc_pattern = LOV_PATTERN_RAID0;
5468 if (append_stripe_count != 0)
5469 llc->llc_stripe_count = append_stripe_count;
5472 if (append_pool != NULL) {
5474 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
5475 /* XXX: sanity check here */
5476 v3 = (struct lov_user_md_v3 *)v1;
5477 if (v3->lmm_pool_name[0] != '\0')
5478 pool = v3->lmm_pool_name;
5481 lod_set_pool(&llc->llc_pool, pool);
5483 if (v1->lmm_magic == LOV_USER_MAGIC_SPECIFIC &&
5484 append_stripe_count == 0 &&
5485 append_pool == NULL) {
5486 v3 = (struct lov_user_md_v3 *)v1;
5487 rc = lod_comp_copy_ost_lists(llc, v3);
5490 } else if (llc->llc_ostlist.op_array &&
5491 llc->llc_ostlist.op_count) {
5492 for (j = 0; j < llc->llc_ostlist.op_count; j++)
5493 llc->llc_ostlist.op_array[j] = -1;
5494 llc->llc_ostlist.op_count = 0;
5498 lds->lds_def_striping_set = 1;
5502 static inline void lod_lum2lds(struct lod_default_striping *lds,
5503 const struct lmv_user_md *lum)
5505 lds->lds_dir_def_stripe_count = le32_to_cpu(lum->lum_stripe_count);
5506 lds->lds_dir_def_stripe_offset = le32_to_cpu(lum->lum_stripe_offset);
5507 lds->lds_dir_def_hash_type = le32_to_cpu(lum->lum_hash_type);
5508 lds->lds_dir_def_max_inherit = lum->lum_max_inherit;
5509 lds->lds_dir_def_max_inherit_rr = lum->lum_max_inherit_rr;
5510 lds->lds_dir_def_striping_set = 1;
5514 * Get default directory striping.
5516 * \param[in] env execution environment
5517 * \param[in] lo object
5518 * \param[out] lds default striping
5520 * \retval 0 on success
5521 * \retval negative if failed
5523 static int lod_get_default_lmv_striping(const struct lu_env *env,
5524 struct lod_object *lo,
5525 struct lod_default_striping *lds)
5527 struct lmv_user_md *lmu;
5530 lds->lds_dir_def_striping_set = 0;
5532 rc = lod_get_default_lmv_ea(env, lo);
5536 if (rc >= (int)sizeof(*lmu)) {
5537 struct lod_thread_info *info = lod_env_info(env);
5539 lmu = info->lti_ea_store;
5540 lod_lum2lds(lds, lmu);
5547 * Get default striping in the object.
5549 * Get object default striping and default directory striping.
5551 * \param[in] env execution environment
5552 * \param[in] lo object
5553 * \param[out] lds default striping
5555 * \retval 0 on success
5556 * \retval negative if failed
5558 static int lod_get_default_striping(const struct lu_env *env,
5559 struct lod_object *lo,
5560 struct dt_allocation_hint *ah,
5561 struct lod_default_striping *lds)
5565 rc = lod_get_default_lov_striping(env, lo, lds, NULL);
5566 if (lds->lds_def_striping_set) {
5567 struct lod_thread_info *info = lod_env_info(env);
5568 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5570 rc = lod_verify_striping(env, d, lo, &info->lti_buf, false);
5572 lds->lds_def_striping_set = 0;
5575 if (ah->dah_eadata_is_dmv) {
5576 lod_lum2lds(lds, ah->dah_eadata);
5577 } else if (ah->dah_dmv_imp_inherit) {
5578 lds->lds_dir_def_striping_set = 0;
5580 rc1 = lod_get_default_lmv_striping(env, lo, lds);
5581 if (rc == 0 && rc1 < 0)
5589 * Apply default striping on object.
5591 * If object striping pattern is not set, set to the one in default striping.
5592 * The default striping is from parent or fs.
5594 * \param[in] lo new object
5595 * \param[in] lds default striping
5596 * \param[in] mode new object's mode
5598 static void lod_striping_from_default(struct lod_object *lo,
5599 const struct lod_default_striping *lds,
5602 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5605 if (lds->lds_def_striping_set && S_ISREG(mode)) {
5606 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
5608 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
5609 lds->lds_def_comp_cnt);
5613 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
5614 if (lds->lds_def_mirror_cnt > 1)
5615 lo->ldo_flr_state = LCM_FL_RDONLY;
5617 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5618 struct lod_layout_component *obj_comp =
5619 &lo->ldo_comp_entries[i];
5620 struct lod_layout_component *def_comp =
5621 &lds->lds_def_comp_entries[i];
5624 "inherit "DFID" file layout from default: flags=%#x size=%u nr=%u offset=%u pattern=%#x pool=%s\n",
5625 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
5626 def_comp->llc_flags,
5627 def_comp->llc_stripe_size,
5628 def_comp->llc_stripe_count,
5629 def_comp->llc_stripe_offset,
5630 def_comp->llc_pattern,
5631 def_comp->llc_pool ?: "");
5633 *obj_comp = *def_comp;
5634 if (def_comp->llc_pool != NULL) {
5635 /* pointer was copied from def_comp */
5636 obj_comp->llc_pool = NULL;
5637 lod_obj_set_pool(lo, i, def_comp->llc_pool);
5641 if (def_comp->llc_ostlist.op_array &&
5642 def_comp->llc_ostlist.op_count) {
5643 OBD_ALLOC(obj_comp->llc_ostlist.op_array,
5644 obj_comp->llc_ostlist.op_size);
5645 if (!obj_comp->llc_ostlist.op_array)
5647 memcpy(obj_comp->llc_ostlist.op_array,
5648 def_comp->llc_ostlist.op_array,
5649 obj_comp->llc_ostlist.op_size);
5650 } else if (def_comp->llc_ostlist.op_array) {
5651 obj_comp->llc_ostlist.op_array = NULL;
5655 * Don't initialize these fields for plain layout
5656 * (v1/v3) here, they are inherited in the order of
5657 * 'parent' -> 'fs default (root)' -> 'global default
5658 * values for stripe_count & stripe_size'.
5660 * see lod_ah_init().
5662 if (!lo->ldo_is_composite)
5665 lod_adjust_stripe_info(obj_comp, desc, 0);
5667 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
5668 if (lo->ldo_dir_stripe_count == 0)
5669 lo->ldo_dir_stripe_count =
5670 lds->lds_dir_def_stripe_count;
5671 if (lo->ldo_dir_stripe_offset == -1)
5672 lo->ldo_dir_stripe_offset =
5673 lds->lds_dir_def_stripe_offset;
5674 if (lo->ldo_dir_hash_type == LMV_HASH_TYPE_UNKNOWN)
5675 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
5678 "inherit "DFID" dir layout from default: count=%hu offset=%u hash_type=%x\n",
5679 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
5680 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
5681 lo->ldo_dir_hash_type);
5685 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root,
5686 const char *append_pool)
5688 struct lod_layout_component *lod_comp;
5690 if (lo->ldo_comp_cnt == 0)
5693 if (lo->ldo_is_composite)
5696 lod_comp = &lo->ldo_comp_entries[0];
5698 if (lod_comp->llc_stripe_count <= 0 ||
5699 lod_comp->llc_stripe_size <= 0)
5702 if (from_root && (lod_comp->llc_pool == NULL ||
5703 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
5706 if (append_pool && append_pool[0])
5713 * Implementation of dt_object_operations::do_ah_init.
5715 * This method is used to make a decision on the striping configuration for the
5716 * object being created. It can be taken from the \a parent object if it exists,
5717 * or filesystem's default. The resulting configuration (number of stripes,
5718 * stripe size/offset, pool name, hash_type, etc.) is stored in the object
5719 * itself and will be used by the methods like ->doo_declare_create().
5721 * \see dt_object_operations::do_ah_init() in the API description for details.
5723 static void lod_ah_init(const struct lu_env *env,
5724 struct dt_allocation_hint *ah,
5725 struct dt_object *parent,
5726 struct dt_object *child,
5729 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
5730 struct lod_thread_info *info = lod_env_info(env);
5731 struct lod_default_striping *lds = lod_lds_buf_get(env);
5732 struct dt_object *nextp = NULL;
5733 struct dt_object *nextc;
5734 struct lod_object *lp = NULL;
5735 struct lod_object *lc;
5736 struct lov_desc *desc;
5737 struct lod_layout_component *lod_comp;
5743 if (ah->dah_append_stripe_count == -1)
5744 ah->dah_append_stripe_count =
5745 d->lod_ost_descs.ltd_lov_desc.ld_tgt_count;
5747 if (likely(parent)) {
5748 nextp = dt_object_child(parent);
5749 lp = lod_dt_obj(parent);
5752 nextc = dt_object_child(child);
5753 lc = lod_dt_obj(child);
5755 LASSERT(!lod_obj_is_striped(child));
5756 /* default layout template may have been set on the regular file
5757 * when this is called from mdd_create_data() */
5758 if (S_ISREG(child_mode))
5759 lod_free_comp_entries(lc);
5761 if (!dt_object_exists(nextc))
5762 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
5764 if (S_ISDIR(child_mode)) {
5765 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
5767 /* other default values are 0 */
5768 lc->ldo_dir_stripe_offset = LMV_OFFSET_DEFAULT;
5770 /* no default striping configuration is needed for
5773 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5774 le32_to_cpu(lum1->lum_magic) == LMV_MAGIC_FOREIGN) {
5775 lc->ldo_is_foreign = true;
5776 /* keep stripe_count 0 and stripe_offset -1 */
5777 CDEBUG(D_INFO, "no default striping for foreign dir\n");
5781 if (likely(lp != NULL))
5782 lod_get_default_striping(env, lp, ah, lds);
5784 /* It should always honour the specified stripes */
5785 if (ah->dah_eadata && ah->dah_eadata_len &&
5786 !ah->dah_eadata_is_dmv &&
5787 (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
5788 le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC ||
5789 le32_to_cpu(lum1->lum_magic) == LMV_MAGIC_V1)) {
5790 lc->ldo_dir_stripe_count =
5791 le32_to_cpu(lum1->lum_stripe_count);
5792 lc->ldo_dir_stripe_offset =
5793 le32_to_cpu(lum1->lum_stripe_offset);
5794 lc->ldo_dir_hash_type =
5795 le32_to_cpu(lum1->lum_hash_type);
5797 "set dirstripe: count %hu, offset %d, hash %x\n",
5798 lc->ldo_dir_stripe_count,
5799 (int)lc->ldo_dir_stripe_offset,
5800 lc->ldo_dir_hash_type);
5802 if (d->lod_mdt_descs.ltd_lmv_desc.ld_active_tgt_count &&
5803 lc->ldo_dir_stripe_count < 2 &&
5804 lum1->lum_max_inherit != LMV_INHERIT_NONE) {
5805 /* when filesystem-wide default LMV is set, dirs
5806 * will be created on MDT by space usage, but if
5807 * dir is created with "lfs mkdir -c 1 ...", its
5808 * subdirs should be kept on the same MDT. To
5809 * guarantee this, set default LMV for such dir.
5811 lds->lds_dir_def_stripe_count =
5812 le32_to_cpu(lum1->lum_stripe_count);
5813 /* if "-1" stripe offset is set, save current
5814 * MDT index in default LMV.
5816 if (le32_to_cpu(lum1->lum_stripe_offset) ==
5818 lds->lds_dir_def_stripe_offset =
5819 lod2lu_dev(d)->ld_site->ld_seq_site->ss_node_id;
5821 lds->lds_dir_def_stripe_offset =
5822 le32_to_cpu(lum1->lum_stripe_offset);
5823 lds->lds_dir_def_hash_type =
5824 le32_to_cpu(lum1->lum_hash_type);
5825 lds->lds_dir_def_max_inherit =
5826 lum1->lum_max_inherit;
5827 /* it will be decreased by 1 later in setting */
5828 if (lum1->lum_max_inherit >= LMV_INHERIT_END &&
5829 lum1->lum_max_inherit < LMV_INHERIT_MAX)
5830 lds->lds_dir_def_max_inherit++;
5831 lds->lds_dir_def_max_inherit_rr =
5832 lum1->lum_max_inherit_rr;
5833 lds->lds_dir_def_striping_set = 1;
5834 /* don't inherit LOV from ROOT */
5835 if (lds->lds_def_striping_set &&
5836 fid_is_root(lod_object_fid(lp)))
5837 lds->lds_def_striping_set = 0;
5838 lc->ldo_def_striping = lds;
5839 } else if (lds->lds_def_striping_set &&
5840 !fid_is_root(lod_object_fid(lp))) {
5841 /* don't inherit default LMV for "lfs mkdir" */
5842 lds->lds_dir_def_striping_set = 0;
5843 lc->ldo_def_striping = lds;
5846 /* inherit default striping except ROOT */
5847 if ((lds->lds_def_striping_set ||
5848 lds->lds_dir_def_striping_set) &&
5849 !fid_is_root(lod_object_fid(lp)))
5850 lc->ldo_def_striping = lds;
5852 /* transfer defaults LMV to new directory */
5853 lod_striping_from_default(lc, lds, child_mode);
5855 /* set count 0 to create normal directory */
5856 if (lc->ldo_dir_stripe_count == 1)
5857 lc->ldo_dir_stripe_count = 0;
5859 /* do not save default LMV on server */
5860 if (ah->dah_dmv_imp_inherit) {
5861 lds->lds_dir_def_striping_set = 0;
5862 if (!lds->lds_def_striping_set)
5863 lc->ldo_def_striping = NULL;
5867 /* shrink the stripe count to max_mdt_stripecount if it is -1
5868 * and max_mdt_stripecount is not 0
5870 if (lc->ldo_dir_stripe_count == (__u16)(-1) &&
5871 d->lod_max_mdt_stripecount)
5872 lc->ldo_dir_stripe_count = d->lod_max_mdt_stripecount;
5874 /* shrink the stripe_count to the avaible MDT count */
5875 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
5876 !CFS_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)) {
5877 lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
5878 if (lc->ldo_dir_stripe_count == 1)
5879 lc->ldo_dir_stripe_count = 0;
5882 if (!lmv_is_known_hash_type(lc->ldo_dir_hash_type))
5883 lc->ldo_dir_hash_type =
5884 (lc->ldo_dir_hash_type & LMV_HASH_FLAG_KNOWN) |
5885 d->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
5887 /* make sure all fscrypt metadata stays on same mdt */
5888 if (child->do_lu.lo_header->loh_attr & LOHA_FSCRYPT_MD) {
5889 lc->ldo_dir_stripe_count = 0;
5890 lds->lds_dir_def_stripe_offset =
5891 lod2lu_dev(d)->ld_site->ld_seq_site->ss_node_id;
5892 lds->lds_dir_def_striping_set = 1;
5893 lc->ldo_def_striping = lds;
5896 CDEBUG(D_INFO, "final dir stripe_count=%hu offset=%d hash=%u\n",
5897 lc->ldo_dir_stripe_count,
5898 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
5903 /* child object regular file*/
5905 if (!lod_object_will_be_striped(S_ISREG(child_mode),
5906 lu_object_fid(&child->do_lu)))
5909 /* If object is going to be striped over OSTs, transfer default
5910 * striping information to the child, so that we can use it
5911 * during declaration and creation.
5913 * Try from the parent first.
5915 if (likely(lp != NULL)) {
5916 rc = lod_get_default_lov_striping(env, lp, lds, ah);
5917 if (rc == 0 && lds->lds_def_striping_set) {
5918 rc = lod_verify_striping(env, d, lp, &info->lti_buf,
5921 lod_striping_from_default(lc, lds, child_mode);
5925 /* Initialize lod_device::lod_md_root object reference */
5926 if (d->lod_md_root == NULL) {
5927 struct dt_object *root;
5928 struct lod_object *lroot;
5930 lu_root_fid(&info->lti_fid);
5931 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
5932 if (!IS_ERR(root)) {
5933 lroot = lod_dt_obj(root);
5935 spin_lock(&d->lod_lock);
5936 if (d->lod_md_root != NULL)
5937 dt_object_put(env, &d->lod_md_root->ldo_obj);
5938 d->lod_md_root = lroot;
5939 spin_unlock(&d->lod_lock);
5943 /* try inherit layout from the root object (fs default) when:
5944 * - parent does not have default layout; or
5945 * - parent has plain(v1/v3) default layout, and some attributes
5946 * are not specified in the default layout;
5948 if (d->lod_md_root != NULL &&
5949 lod_need_inherit_more(lc, true, ah->dah_append_pool)) {
5950 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds,
5952 if (rc || !lds->lds_def_striping_set)
5955 rc = lod_verify_striping(env, d, d->lod_md_root, &info->lti_buf,
5960 if (lc->ldo_comp_cnt == 0) {
5961 lod_striping_from_default(lc, lds, child_mode);
5962 } else if (!lds->lds_def_striping_is_composite) {
5963 struct lod_layout_component *def_comp;
5965 LASSERT(!lc->ldo_is_composite);
5966 lod_comp = &lc->ldo_comp_entries[0];
5967 def_comp = &lds->lds_def_comp_entries[0];
5969 if (lod_comp->llc_stripe_count <= 0)
5970 lod_comp->llc_stripe_count =
5971 def_comp->llc_stripe_count;
5972 if (lod_comp->llc_stripe_size <= 0)
5973 lod_comp->llc_stripe_size =
5974 def_comp->llc_stripe_size;
5975 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT &&
5976 (!lod_comp->llc_pool || !lod_comp->llc_pool[0]))
5977 lod_comp->llc_stripe_offset =
5978 def_comp->llc_stripe_offset;
5979 if (lod_comp->llc_pool == NULL)
5980 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
5985 * fs default striping may not be explicitly set, or historically set
5986 * in config log, use them.
5988 if (lod_need_inherit_more(lc, false, ah->dah_append_pool)) {
5989 if (lc->ldo_comp_cnt == 0) {
5990 rc = lod_alloc_comp_entries(lc, 0, 1);
5992 /* fail to allocate memory, will create a
5993 * non-striped file. */
5995 lc->ldo_is_composite = 0;
5996 lod_comp = &lc->ldo_comp_entries[0];
5997 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
5999 LASSERT(!lc->ldo_is_composite);
6000 lod_comp = &lc->ldo_comp_entries[0];
6001 desc = &d->lod_ost_descs.ltd_lov_desc;
6002 lod_adjust_stripe_info(lod_comp, desc,
6003 ah->dah_append_stripe_count);
6004 if (ah->dah_append_pool && ah->dah_append_pool[0])
6005 lod_obj_set_pool(lc, 0, ah->dah_append_pool);
6012 * Size initialization on late striping.
6014 * Propagate the size of a truncated object to a deferred striping.
6015 * This function handles a special case when truncate was done on a
6016 * non-striped object and now while the striping is being created
6017 * we can't lose that size, so we have to propagate it to the stripes
6020 * \param[in] env execution environment
6021 * \param[in] dt object
6022 * \param[in] th transaction handle
6024 * \retval 0 on success
6025 * \retval negative if failed
6027 static int lod_declare_init_size(const struct lu_env *env,
6028 struct dt_object *dt, struct thandle *th)
6030 struct dt_object *next = dt_object_child(dt);
6031 struct lod_object *lo = lod_dt_obj(dt);
6032 struct dt_object **objects = NULL;
6033 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
6034 uint64_t size, offs;
6035 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
6036 struct lu_extent size_ext;
6039 if (!lod_obj_is_striped(dt))
6042 rc = dt_attr_get(env, next, attr);
6043 LASSERT(attr->la_valid & LA_SIZE);
6047 size = attr->la_size;
6051 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
6052 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6053 struct lod_layout_component *lod_comp;
6054 struct lu_extent *extent;
6056 lod_comp = &lo->ldo_comp_entries[i];
6058 if (lod_comp->llc_stripe == NULL)
6061 extent = &lod_comp->llc_extent;
6062 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
6063 if (!lo->ldo_is_composite ||
6064 lu_extent_is_overlapped(extent, &size_ext)) {
6065 objects = lod_comp->llc_stripe;
6066 stripe_count = lod_comp->llc_stripe_count;
6067 stripe_size = lod_comp->llc_stripe_size;
6070 if (stripe_count == 0)
6073 LASSERT(objects != NULL && stripe_size != 0);
6074 do_div(size, stripe_size);
6075 stripe = do_div(size, stripe_count);
6076 LASSERT(objects[stripe] != NULL);
6078 size = size * stripe_size;
6079 offs = attr->la_size;
6080 size += do_div(offs, stripe_size);
6082 attr->la_valid = LA_SIZE;
6083 attr->la_size = size;
6085 rc = lod_sub_declare_attr_set(env, objects[stripe],
6094 * Declare creation of striped object.
6096 * The function declares creation stripes for a regular object. The function
6097 * also declares whether the stripes will be created with non-zero size if
6098 * previously size was set non-zero on the master object. If object \a dt is
6099 * not local, then only fully defined striping can be applied in \a lovea.
6100 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
6103 * \param[in] env execution environment
6104 * \param[in] dt object
6105 * \param[in] attr attributes the stripes will be created with
6106 * \param[in] lovea a buffer containing striping description
6107 * \param[in] th transaction handle
6109 * \retval 0 on success
6110 * \retval negative if failed
6112 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
6113 struct lu_attr *attr,
6114 const struct lu_buf *lovea, struct thandle *th)
6116 struct lod_thread_info *info = lod_env_info(env);
6117 struct dt_object *next = dt_object_child(dt);
6118 struct lod_object *lo = lod_dt_obj(dt);
6122 if (CFS_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
6123 GOTO(out, rc = -ENOMEM);
6125 if (!dt_object_remote(next)) {
6126 /* choose OST and generate appropriate objects */
6127 rc = lod_prepare_create(env, lo, attr, lovea, th);
6132 * declare storage for striping data
6134 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6136 /* LOD can not choose OST objects for remote objects, i.e.
6137 * stripes must be ready before that. Right now, it can only
6138 * happen during migrate, i.e. migrate process needs to create
6139 * remote regular file (mdd_migrate_create), then the migrate
6140 * process will provide stripeEA. */
6141 LASSERT(lovea != NULL);
6142 info->lti_buf = *lovea;
6145 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
6146 XATTR_NAME_LOV, 0, th);
6151 * if striping is created with local object's size > 0,
6152 * we have to propagate this size to specific object
6153 * the case is possible only when local object was created previously
6155 if (dt_object_exists(next))
6156 rc = lod_declare_init_size(env, dt, th);
6159 /* failed to create striping or to set initial size, let's reset
6160 * config so that others don't get confused */
6162 lod_striping_free(env, lo);
6168 * Whether subdirectories under \a dt should be created on MDTs by space QoS
6170 * If LMV_HASH_FLAG_SPACE is set on directory default layout, its subdirectories
6171 * should be created on MDT by space QoS.
6173 * \param[in] env execution environment
6174 * \param[in] dev lu device
6175 * \param[in] dt object
6177 * \retval 1 if directory should create subdir by space usage
6179 * \retval -ev if failed
6181 static inline int dt_object_qos_mkdir(const struct lu_env *env,
6182 struct lu_device *dev,
6183 struct dt_object *dt)
6185 struct lod_thread_info *info = lod_env_info(env);
6186 struct lu_object *obj;
6187 struct lod_object *lo;
6188 struct lmv_user_md *lmu;
6191 obj = lu_object_find_slice(env, dev, lu_object_fid(&dt->do_lu), NULL);
6193 return PTR_ERR(obj);
6195 lo = lu2lod_obj(obj);
6197 rc = lod_get_default_lmv_ea(env, lo);
6198 dt_object_put(env, dt);
6202 if (rc < (int)sizeof(*lmu))
6205 lmu = info->lti_ea_store;
6206 return le32_to_cpu(lmu->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
6210 * Implementation of dt_object_operations::do_declare_create.
6212 * The method declares creation of a new object. If the object will be striped,
6213 * then helper functions are called to find FIDs for the stripes, declare
6214 * creation of the stripes and declare initialization of the striping
6215 * information to be stored in the master object.
6217 * \see dt_object_operations::do_declare_create() in the API description
6220 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
6221 struct lu_attr *attr,
6222 struct dt_allocation_hint *hint,
6223 struct dt_object_format *dof, struct thandle *th)
6225 struct dt_object *next = dt_object_child(dt);
6226 struct lod_object *lo = lod_dt_obj(dt);
6235 * first of all, we declare creation of local object
6237 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
6242 * it's lod_ah_init() that has decided the object will be striped
6244 if (dof->dof_type == DFT_REGULAR) {
6245 /* callers don't want stripes */
6246 /* XXX: all tricky interactions with ->ah_make_hint() decided
6247 * to use striping, then ->declare_create() behaving differently
6248 * should be cleaned */
6249 if (dof->u.dof_reg.striped != 0)
6250 rc = lod_declare_striped_create(env, dt, attr,
6252 } else if (dof->dof_type == DFT_DIR) {
6253 struct seq_server_site *ss;
6254 struct lu_buf buf = { NULL };
6256 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
6258 /* If the parent has default stripeEA, and client
6259 * did not find it before sending create request,
6260 * then MDT will return -EREMOTE, and client will
6261 * retrieve the default stripeEA and re-create the
6264 * Note: if dah_eadata != NULL, it means creating the
6265 * striped directory with specified stripeEA, then it
6266 * should ignore the default stripeEA */
6267 if (hint != NULL && hint->dah_eadata == NULL) {
6268 if (CFS_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
6269 GOTO(out, rc = -EREMOTE);
6271 if (lo->ldo_dir_stripe_offset != LMV_OFFSET_DEFAULT &&
6272 lo->ldo_dir_stripe_offset != ss->ss_node_id) {
6273 struct lod_device *lod;
6274 struct lu_tgt_desc *mdt = NULL;
6275 bool found_mdt = false;
6277 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6278 lod_foreach_mdt(lod, mdt) {
6279 if (mdt->ltd_index ==
6280 lo->ldo_dir_stripe_offset) {
6286 /* If the MDT indicated by stripe_offset can be
6287 * found, then tell client to resend the create
6288 * request to the correct MDT, otherwise return
6289 * error to client */
6291 GOTO(out, rc = -EREMOTE);
6293 GOTO(out, rc = -EINVAL);
6295 } else if (hint && hint->dah_eadata) {
6296 buf.lb_buf = (void *)hint->dah_eadata;
6297 buf.lb_len = hint->dah_eadata_len;
6300 rc = lod_declare_dir_striping_create(env, dt, attr, &buf, dof,
6304 /* failed to create striping or to set initial size, let's reset
6305 * config so that others don't get confused */
6307 lod_striping_free(env, lo);
6312 * Generate component ID for new created component.
6314 * \param[in] lo LOD object
6315 * \param[in] comp_idx index of ldo_comp_entries
6317 * \retval component ID on success
6318 * \retval LCME_ID_INVAL on failure
6320 static __u32 lod_gen_component_id(struct lod_object *lo,
6321 int mirror_id, int comp_idx)
6323 struct lod_layout_component *lod_comp;
6324 __u32 id, start, end;
6327 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
6329 lod_obj_inc_layout_gen(lo);
6330 id = lo->ldo_layout_gen;
6331 if (likely(id <= SEQ_ID_MAX))
6332 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
6334 /* Layout generation wraps, need to check collisions. */
6335 start = id & SEQ_ID_MASK;
6338 for (id = start; id <= end; id++) {
6339 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6340 lod_comp = &lo->ldo_comp_entries[i];
6341 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
6344 /* Found the ununsed ID */
6345 if (i == lo->ldo_comp_cnt)
6346 RETURN(pflr_id(mirror_id, id));
6349 if (end == SEQ_ID_MAX) {
6350 end = min_t(__u32, start, SEQ_ID_MAX) - 1;
6355 RETURN(LCME_ID_INVAL);
6359 * Creation of a striped regular object.
6361 * The function is called to create the stripe objects for a regular
6362 * striped file. This can happen at the initial object creation or
6363 * when the caller asks LOD to do so using ->do_xattr_set() method
6364 * (so called late striping). Notice all the information are already
6365 * prepared in the form of the list of objects (ldo_stripe field).
6366 * This is done during declare phase.
6368 * \param[in] env execution environment
6369 * \param[in] dt object
6370 * \param[in] attr attributes the stripes will be created with
6371 * \param[in] dof format of stripes (see OSD API description)
6372 * \param[in] th transaction handle
6374 * \retval 0 on success
6375 * \retval negative if failed
6377 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
6378 struct lu_attr *attr, struct dt_object_format *dof,
6381 struct lod_layout_component *lod_comp;
6382 struct lod_object *lo = lod_dt_obj(dt);
6387 mutex_lock(&lo->ldo_layout_mutex);
6389 LASSERT((lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL) ||
6390 lo->ldo_is_foreign);
6392 mirror_id = 0; /* non-flr file's mirror_id is 0 */
6393 if (lo->ldo_mirror_count > 1) {
6394 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6395 lod_comp = &lo->ldo_comp_entries[i];
6396 if (lod_comp->llc_id != LCME_ID_INVAL &&
6397 mirror_id_of(lod_comp->llc_id) > mirror_id)
6398 mirror_id = mirror_id_of(lod_comp->llc_id);
6402 /* create all underlying objects */
6403 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6404 lod_comp = &lo->ldo_comp_entries[i];
6406 if (lod_comp->llc_id == LCME_ID_INVAL) {
6407 /* only the component of FLR layout with more than 1
6408 * mirror has mirror ID in its component ID.
6410 if (lod_comp->llc_extent.e_start == 0 &&
6411 lo->ldo_mirror_count > 1)
6414 lod_comp->llc_id = lod_gen_component_id(lo,
6416 if (lod_comp->llc_id == LCME_ID_INVAL)
6417 GOTO(out, rc = -ERANGE);
6420 if (lod_comp_inited(lod_comp))
6423 if (lod_comp->llc_magic == LOV_MAGIC_FOREIGN) {
6424 lod_comp_set_init(lod_comp);
6428 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
6429 lod_comp_set_init(lod_comp);
6431 if (lov_pattern(lod_comp->llc_pattern) & LOV_PATTERN_MDT)
6432 lod_comp_set_init(lod_comp);
6434 if (lod_comp->llc_stripe == NULL)
6437 LASSERT(lod_comp->llc_stripe_count);
6438 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6439 struct dt_object *object = lod_comp->llc_stripe[j];
6440 LASSERT(object != NULL);
6441 rc = lod_sub_create(env, object, attr, NULL, dof, th);
6445 lod_comp_set_init(lod_comp);
6448 rc = lod_fill_mirrors(lo);
6452 lo->ldo_comp_cached = 1;
6454 rc = lod_generate_and_set_lovea(env, lo, th);
6458 mutex_unlock(&lo->ldo_layout_mutex);
6463 lod_striping_free_nolock(env, lo);
6464 mutex_unlock(&lo->ldo_layout_mutex);
6469 static inline bool lod_obj_is_dom(struct dt_object *dt)
6471 struct lod_object *lo = lod_dt_obj(dt);
6473 if (!dt_object_exists(dt_object_child(dt)))
6476 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
6479 if (!lo->ldo_comp_cnt)
6482 return (lov_pattern(lo->ldo_comp_entries[0].llc_pattern) &
6487 * Implementation of dt_object_operations::do_create.
6489 * If any of preceeding methods (like ->do_declare_create(),
6490 * ->do_ah_init(), etc) chose to create a striped object,
6491 * then this method will create the master and the stripes.
6493 * \see dt_object_operations::do_create() in the API description for details.
6495 static int lod_create(const struct lu_env *env, struct dt_object *dt,
6496 struct lu_attr *attr, struct dt_allocation_hint *hint,
6497 struct dt_object_format *dof, struct thandle *th)
6502 /* create local object */
6503 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
6507 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
6508 (lod_obj_is_striped(dt) || lod_obj_is_dom(dt)) &&
6509 dof->u.dof_reg.striped != 0) {
6510 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
6511 rc = lod_striped_create(env, dt, attr, dof, th);
6518 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
6519 struct dt_object *dt, struct thandle *th,
6520 int comp_idx, int stripe_idx,
6521 struct lod_obj_stripe_cb_data *data)
6523 if (data->locd_declare)
6524 return lod_sub_declare_destroy(env, dt, th);
6526 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
6527 stripe_idx == cfs_fail_val)
6528 return lod_sub_destroy(env, dt, th);
6534 * Implementation of dt_object_operations::do_declare_destroy.
6536 * If the object is a striped directory, then the function declares reference
6537 * removal from the master object (this is an index) to the stripes and declares
6538 * destroy of all the stripes. In all the cases, it declares an intention to
6539 * destroy the object itself.
6541 * \see dt_object_operations::do_declare_destroy() in the API description
6544 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
6547 struct dt_object *next = dt_object_child(dt);
6548 struct lod_object *lo = lod_dt_obj(dt);
6549 struct lod_thread_info *info = lod_env_info(env);
6550 struct dt_object *stripe;
6551 char *stripe_name = info->lti_key;
6557 * load striping information, notice we don't do this when object
6558 * is being initialized as we don't need this information till
6559 * few specific cases like destroy, chown
6561 rc = lod_striping_load(env, lo);
6565 /* declare destroy for all underlying objects */
6566 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6567 rc = next->do_ops->do_index_try(env, next,
6568 &dt_directory_features);
6572 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6573 stripe = lo->ldo_stripe[i];
6577 rc = lod_sub_declare_ref_del(env, next, th);
6581 snprintf(stripe_name, sizeof(info->lti_key),
6583 PFID(lu_object_fid(&stripe->do_lu)), i);
6584 rc = lod_sub_declare_delete(env, next,
6585 (const struct dt_key *)stripe_name, th);
6592 * we declare destroy for the local object
6594 rc = lod_sub_declare_destroy(env, next, th);
6598 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
6599 CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
6602 if (!lod_obj_is_striped(dt))
6605 /* declare destroy all striped objects */
6606 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6607 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6608 stripe = lo->ldo_stripe[i];
6612 if (!dt_object_exists(stripe))
6615 rc = lod_sub_declare_ref_del(env, stripe, th);
6619 rc = lod_sub_declare_destroy(env, stripe, th);
6624 struct lod_obj_stripe_cb_data data = { { 0 } };
6626 data.locd_declare = true;
6627 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
6628 rc = lod_obj_for_each_stripe(env, lo, th, &data);
6635 * Implementation of dt_object_operations::do_destroy.
6637 * If the object is a striped directory, then the function removes references
6638 * from the master object (this is an index) to the stripes and destroys all
6639 * the stripes. In all the cases, the function destroys the object itself.
6641 * \see dt_object_operations::do_destroy() in the API description for details.
6643 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
6646 struct dt_object *next = dt_object_child(dt);
6647 struct lod_object *lo = lod_dt_obj(dt);
6648 struct lod_thread_info *info = lod_env_info(env);
6649 char *stripe_name = info->lti_key;
6650 struct dt_object *stripe;
6656 /* destroy sub-stripe of master object */
6657 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6658 rc = next->do_ops->do_index_try(env, next,
6659 &dt_directory_features);
6663 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6664 stripe = lo->ldo_stripe[i];
6668 rc = lod_sub_ref_del(env, next, th);
6672 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
6673 PFID(lu_object_fid(&stripe->do_lu)), i);
6675 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
6676 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
6677 PFID(lu_object_fid(&stripe->do_lu)));
6679 rc = lod_sub_delete(env, next,
6680 (const struct dt_key *)stripe_name, th);
6686 rc = lod_sub_destroy(env, next, th);
6690 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
6691 CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
6694 if (!lod_obj_is_striped(dt))
6697 /* destroy all striped objects */
6698 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6699 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6700 stripe = lo->ldo_stripe[i];
6704 if (!dt_object_exists(stripe))
6707 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
6708 i == cfs_fail_val) {
6709 dt_write_lock(env, stripe, DT_TGT_CHILD);
6710 rc = lod_sub_ref_del(env, stripe, th);
6711 dt_write_unlock(env, stripe);
6715 rc = lod_sub_destroy(env, stripe, th);
6721 struct lod_obj_stripe_cb_data data = { { 0 } };
6723 data.locd_declare = false;
6724 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
6725 rc = lod_obj_for_each_stripe(env, lo, th, &data);
6732 * Implementation of dt_object_operations::do_declare_ref_add.
6734 * \see dt_object_operations::do_declare_ref_add() in the API description
6737 static int lod_declare_ref_add(const struct lu_env *env,
6738 struct dt_object *dt, struct thandle *th)
6740 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
6744 * Implementation of dt_object_operations::do_ref_add.
6746 * \see dt_object_operations::do_ref_add() in the API description for details.
6748 static int lod_ref_add(const struct lu_env *env,
6749 struct dt_object *dt, struct thandle *th)
6751 return lod_sub_ref_add(env, dt_object_child(dt), th);
6755 * Implementation of dt_object_operations::do_declare_ref_del.
6757 * \see dt_object_operations::do_declare_ref_del() in the API description
6760 static int lod_declare_ref_del(const struct lu_env *env,
6761 struct dt_object *dt, struct thandle *th)
6763 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
6767 * Implementation of dt_object_operations::do_ref_del
6769 * \see dt_object_operations::do_ref_del() in the API description for details.
6771 static int lod_ref_del(const struct lu_env *env,
6772 struct dt_object *dt, struct thandle *th)
6774 return lod_sub_ref_del(env, dt_object_child(dt), th);
6778 * Implementation of dt_object_operations::do_object_sync.
6780 * \see dt_object_operations::do_object_sync() in the API description
6783 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
6784 __u64 start, __u64 end)
6786 return dt_object_sync(env, dt_object_child(dt), start, end);
6790 * Implementation of dt_object_operations::do_object_unlock.
6792 * Used to release LDLM lock(s).
6794 * \see dt_object_operations::do_object_unlock() in the API description
6797 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
6798 struct ldlm_enqueue_info *einfo,
6799 union ldlm_policy_data *policy)
6801 struct lod_object *lo = lod_dt_obj(dt);
6802 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
6803 int slave_locks_size;
6807 if (slave_locks == NULL)
6810 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
6811 /* Note: for remote lock for single stripe dir, MDT will cancel
6812 * the lock by lockh directly */
6813 LASSERT(!dt_object_remote(dt_object_child(dt)));
6815 /* locks were unlocked in MDT layer */
6816 for (i = 0; i < slave_locks->ha_count; i++)
6817 LASSERT(!lustre_handle_is_used(&slave_locks->ha_handles[i]));
6820 * NB, ha_count may not equal to ldo_dir_stripe_count, because dir
6821 * layout may change, e.g., shrink dir layout after migration.
6823 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6824 if (lo->ldo_stripe[i])
6825 dt_invalidate(env, lo->ldo_stripe[i]);
6828 slave_locks_size = offsetof(typeof(*slave_locks),
6829 ha_handles[slave_locks->ha_count]);
6830 OBD_FREE(slave_locks, slave_locks_size);
6831 einfo->ei_cbdata = NULL;
6837 * Implementation of dt_object_operations::do_object_lock.
6839 * Used to get LDLM lock on the non-striped and striped objects.
6841 * \see dt_object_operations::do_object_lock() in the API description
6844 static int lod_object_lock(const struct lu_env *env,
6845 struct dt_object *dt,
6846 struct lustre_handle *lh,
6847 struct ldlm_enqueue_info *einfo,
6848 union ldlm_policy_data *policy)
6850 struct lod_object *lo = lod_dt_obj(dt);
6851 int slave_locks_size;
6852 struct lustre_handle_array *slave_locks = NULL;
6857 /* remote object lock */
6858 if (!einfo->ei_enq_slave) {
6859 LASSERT(dt_object_remote(dt));
6860 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
6864 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
6867 rc = lod_striping_load(env, lo);
6872 if (lo->ldo_dir_stripe_count <= 1)
6875 slave_locks_size = offsetof(typeof(*slave_locks),
6876 ha_handles[lo->ldo_dir_stripe_count]);
6877 /* Freed in lod_object_unlock */
6878 OBD_ALLOC(slave_locks, slave_locks_size);
6881 slave_locks->ha_count = lo->ldo_dir_stripe_count;
6883 /* striped directory lock */
6884 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6885 struct lustre_handle lockh;
6886 struct ldlm_res_id *res_id;
6887 struct dt_object *stripe;
6889 stripe = lo->ldo_stripe[i];
6893 res_id = &lod_env_info(env)->lti_res_id;
6894 fid_build_reg_res_name(lu_object_fid(&stripe->do_lu), res_id);
6895 einfo->ei_res_id = res_id;
6897 if (dt_object_remote(stripe)) {
6898 set_bit(i, (void *)slave_locks->ha_map);
6899 rc = dt_object_lock(env, stripe, &lockh, einfo, policy);
6901 struct ldlm_namespace *ns = einfo->ei_namespace;
6902 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
6903 ldlm_completion_callback completion = einfo->ei_cb_cp;
6904 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
6906 LASSERT(ns != NULL);
6907 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS,
6908 policy, einfo->ei_mode,
6909 &dlmflags, blocking,
6911 NULL, 0, LVB_T_NONE,
6916 ldlm_lock_decref_and_cancel(
6917 &slave_locks->ha_handles[i],
6919 OBD_FREE(slave_locks, slave_locks_size);
6922 slave_locks->ha_handles[i] = lockh;
6924 einfo->ei_cbdata = slave_locks;
6930 * Implementation of dt_object_operations::do_invalidate.
6932 * \see dt_object_operations::do_invalidate() in the API description for details
6934 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
6936 return dt_invalidate(env, dt_object_child(dt));
6939 static int lod_declare_instantiate_components(const struct lu_env *env,
6940 struct lod_object *lo,
6944 struct lod_thread_info *info = lod_env_info(env);
6949 LASSERT(info->lti_count < lo->ldo_comp_cnt);
6951 for (i = 0; i < info->lti_count; i++) {
6952 rc = lod_qos_prep_create(env, lo, NULL, th,
6953 info->lti_comp_idx[i], reserve);
6959 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6960 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
6961 &info->lti_buf, XATTR_NAME_LOV, 0, th);
6968 * Check OSTs for an existing component for further extension
6970 * Checks if OSTs are still healthy and not out of space. Gets free space
6971 * on OSTs (relative to allocation watermark rmb_low) and compares to
6972 * the proposed new_end for this component.
6974 * Decides whether or not to extend a component on its current OSTs.
6976 * \param[in] env execution environment for this thread
6977 * \param[in] lo object we're checking
6978 * \param[in] index index of this component
6979 * \param[in] extension_size extension size for this component
6980 * \param[in] extent layout extent for requested operation
6981 * \param[in] comp_extent extension component extent
6982 * \param[in] write if this is write operation
6984 * \retval true - OK to extend on current OSTs
6985 * \retval false - do not extend on current OSTs
6987 static bool lod_sel_osts_allowed(const struct lu_env *env,
6988 struct lod_object *lo,
6989 int index, __u64 reserve,
6990 struct lu_extent *extent,
6991 struct lu_extent *comp_extent, int write)
6993 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[index];
6994 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6995 struct lod_thread_info *tinfo = lod_env_info(env);
6996 struct obd_statfs *sfs = &tinfo->lti_osfs;
6997 __u64 available = 0;
7003 LASSERT(lod_comp->llc_stripe_count != 0);
7005 lod_getref(&lod->lod_ost_descs);
7006 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
7007 int index = lod_comp->llc_ost_indices[i];
7008 struct lod_tgt_desc *ost = OST_TGT(lod, index);
7009 struct obd_statfs_info info = { 0 };
7010 int j, repeated = 0;
7014 /* Get the number of times this OST repeats in this component.
7015 * Note: inter-component repeats are not counted as this is
7016 * considered as a rare case: we try to not repeat OST in other
7017 * components if possible. */
7018 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
7019 if (index != lod_comp->llc_ost_indices[j])
7022 /* already handled */
7028 if (j < lod_comp->llc_stripe_count)
7031 if (!test_bit(index, lod->lod_ost_bitmap)) {
7032 CDEBUG(D_LAYOUT, "ost %d no longer present\n", index);
7037 rc = dt_statfs_info(env, ost->ltd_tgt, sfs, &info);
7039 CDEBUG(D_LAYOUT, "statfs failed for ost %d, error %d\n",
7045 if (sfs->os_state & OS_STATFS_ENOSPC ||
7046 sfs->os_state & OS_STATFS_READONLY ||
7047 sfs->os_state & OS_STATFS_NOCREATE ||
7048 sfs->os_state & OS_STATFS_DEGRADED) {
7050 "OST%04x unusable for SEL extension, state %x\n",
7051 index, sfs->os_state);
7057 available = sfs->os_bavail * sfs->os_bsize;
7058 /* 'available' is relative to the allocation threshold */
7059 available -= (__u64) info.os_reserved_mb_low << 20;
7061 CDEBUG(D_LAYOUT, "ost %d lowwm: %d highwm: %d, "
7062 "%llu %% blocks available, %llu %% blocks free\n",
7063 index, info.os_reserved_mb_low, info.os_reserved_mb_high,
7064 (100ull * sfs->os_bavail) / sfs->os_blocks,
7065 (100ull * sfs->os_bfree) / sfs->os_blocks);
7067 if (reserve * repeated > available) {
7069 CDEBUG(D_LAYOUT, "low space on ost %d, available %llu "
7070 "< extension size %llu repeated %d\n", index,
7071 available, reserve, repeated);
7075 lod_putref(lod, &lod->lod_ost_descs);
7081 * Adjust extents after component removal
7083 * When we remove an extension component, we move the start of the next
7084 * component to match the start of the extension component, so no space is left
7087 * \param[in] env execution environment for this thread
7088 * \param[in] lo object
7089 * \param[in] max_comp layout component
7090 * \param[in] index index of this component
7092 * \retval 0 on success
7093 * \retval negative errno on error
7095 static void lod_sel_adjust_extents(const struct lu_env *env,
7096 struct lod_object *lo,
7097 int max_comp, int index)
7099 struct lod_layout_component *lod_comp = NULL;
7100 struct lod_layout_component *next = NULL;
7101 struct lod_layout_component *prev = NULL;
7102 __u64 new_start = 0;
7106 /* Extension space component */
7107 lod_comp = &lo->ldo_comp_entries[index];
7108 next = &lo->ldo_comp_entries[index + 1];
7109 prev = &lo->ldo_comp_entries[index - 1];
7111 LASSERT(lod_comp != NULL && prev != NULL && next != NULL);
7112 LASSERT(lod_comp->llc_flags & LCME_FL_EXTENSION);
7114 /* Previous is being removed */
7115 if (prev && prev->llc_id == LCME_ID_INVAL)
7116 new_start = prev->llc_extent.e_start;
7118 new_start = lod_comp->llc_extent.e_start;
7120 for (i = index + 1; i < max_comp; i++) {
7121 lod_comp = &lo->ldo_comp_entries[i];
7123 start = lod_comp->llc_extent.e_start;
7124 lod_comp->llc_extent.e_start = new_start;
7126 /* We only move zero length extendable components */
7127 if (!(start == lod_comp->llc_extent.e_end))
7130 LASSERT(!(lod_comp->llc_flags & LCME_FL_INIT));
7132 lod_comp->llc_extent.e_end = new_start;
7136 /* Calculate the proposed 'new end' for a component we're extending */
7137 static __u64 lod_extension_new_end(__u64 extension_size, __u64 extent_end,
7138 __u32 stripe_size, __u64 component_end,
7139 __u64 extension_end)
7143 LASSERT(extension_size != 0 && stripe_size != 0);
7145 /* Round up to extension size */
7146 if (extent_end == OBD_OBJECT_EOF) {
7147 new_end = OBD_OBJECT_EOF;
7149 /* Add at least extension_size to the previous component_end,
7150 * covering the req layout extent */
7151 new_end = max(extent_end - component_end, extension_size);
7152 new_end = roundup(new_end, extension_size);
7153 new_end += component_end;
7155 /* Component end must be min stripe size aligned */
7156 if (new_end % stripe_size) {
7157 CDEBUG(D_LAYOUT, "new component end is not aligned "
7158 "by the stripe size %u: [%llu, %llu) ext size "
7159 "%llu new end %llu, aligning\n",
7160 stripe_size, component_end, extent_end,
7161 extension_size, new_end);
7162 new_end = roundup(new_end, stripe_size);
7166 if (new_end < extent_end)
7167 new_end = OBD_OBJECT_EOF;
7170 /* Don't extend past the end of the extension component */
7171 if (new_end > extension_end)
7172 new_end = extension_end;
7178 * Calculate the exact reservation (per-OST extension_size) on the OSTs being
7179 * instantiated. It needs to be calculated in advance and taken into account at
7180 * the instantiation time, because otherwise lod_statfs_and_check() may consider
7181 * an OST as OK, but SEL needs its extension_size to fit the free space and the
7182 * OST may turn out to be low-on-space, thus inappropriate OST may be used and
7185 * \param[in] lod_comp lod component we are checking
7187 * \retval size to reserved on each OST of lod_comp's stripe.
7189 static __u64 lod_sel_stripe_reserved(struct lod_layout_component *lod_comp)
7191 /* extension_size is file level, so we must divide by stripe count to
7192 * compare it to available space on a single OST */
7193 return lod_comp->llc_stripe_size * SEL_UNIT_SIZE /
7194 lod_comp->llc_stripe_count;
7197 /* As lod_sel_handler() could be re-entered for the same component several
7198 * times, this is the data for the next call. Fields could be changed to
7199 * component indexes when needed, (e.g. if there is no need to instantiate
7200 * all the previous components up to the current position) to tell the caller
7201 * where to start over from. */
7208 * Process extent updates for a particular layout component
7210 * Handle layout updates for a particular extension space component touched by
7211 * a layout update operation. Core function of self-extending PFL feature.
7213 * In general, this function processes exactly *one* stage of an extension
7214 * operation, modifying the layout accordingly, then returns to the caller.
7215 * The caller is responsible for restarting processing with the new layout,
7216 * which may repeatedly return to this function until the extension updates
7219 * This function does one of a few things to the layout:
7220 * 1. Extends the component before the current extension space component to
7221 * allow it to accomodate the requested operation (if space/policy permit that
7222 * component to continue on its current OSTs)
7224 * 2. If extension of the existing component fails, we do one of two things:
7225 * a. If there is a component after the extension space, we remove the
7226 * extension space component, move the start of the next component down
7227 * accordingly, then notify the caller to restart processing w/the new
7229 * b. If there is no following component, we try repeating the current
7230 * component, creating a new component using the current one as a
7231 * template (keeping its stripe properties but not specific striping),
7232 * and try assigning striping for this component. If there is sufficient
7233 * free space on the OSTs chosen for this component, it is instantiated
7234 * and i/o continues there.
7236 * If there is not sufficient space on the new OSTs, we remove this new
7237 * component & extend the current component.
7239 * Note further that uninited components followed by extension space can be zero
7240 * length meaning that we will try to extend them before initializing them, and
7241 * if that fails, they will be removed without initialization.
7243 * 3. If we extend to/beyond the end of an extension space component, that
7244 * component is exhausted (all of its range has been given to real components),
7245 * so we remove it and restart processing.
7247 * \param[in] env execution environment for this thread
7248 * \param[in,out] lo object to update the layout of
7249 * \param[in] extent layout extent for requested operation, update
7250 * layout to fit this operation
7251 * \param[in] th transaction handle for this operation
7252 * \param[in,out] max_comp the highest comp for the portion of the layout
7253 * we are operating on (For FLR, the chosen
7254 * replica). Updated because we may remove
7256 * \param[in] index index of the extension space component we're
7258 * \param[in] write if this is write op
7259 * \param[in,out] force if the extension is to be forced; set here
7260 to force it on the 2nd call for the same
7263 * \retval 0 on success
7264 * \retval negative errno on error
7266 static int lod_sel_handler(const struct lu_env *env,
7267 struct lod_object *lo,
7268 struct lu_extent *extent,
7269 struct thandle *th, int *max_comp,
7270 int index, int write,
7271 struct sel_data *sd)
7273 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7274 struct lod_thread_info *info = lod_env_info(env);
7275 struct lod_layout_component *lod_comp;
7276 struct lod_layout_component *prev;
7277 struct lod_layout_component *next = NULL;
7278 __u64 extension_size, reserve;
7285 /* First component cannot be extension space */
7287 CERROR("%s: "DFID" first component cannot be extension space\n",
7288 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
7292 lod_comp = &lo->ldo_comp_entries[index];
7293 prev = &lo->ldo_comp_entries[index - 1];
7294 if ((index + 1) < *max_comp)
7295 next = &lo->ldo_comp_entries[index + 1];
7297 /* extension size uses the stripe size field as KiB */
7298 extension_size = lod_comp->llc_stripe_size * SEL_UNIT_SIZE;
7300 CDEBUG(D_LAYOUT, "prev start %llu, extension start %llu, extension end"
7301 " %llu, extension size %llu\n", prev->llc_extent.e_start,
7302 lod_comp->llc_extent.e_start, lod_comp->llc_extent.e_end,
7305 /* Two extension space components cannot be adjacent & extension space
7306 * components cannot be init */
7307 if ((prev->llc_flags & LCME_FL_EXTENSION) ||
7308 !(ergo(next, !(next->llc_flags & LCME_FL_EXTENSION))) ||
7309 lod_comp_inited(lod_comp)) {
7310 CERROR("%s: "DFID" invalid extension space components\n",
7311 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
7315 reserve = lod_sel_stripe_reserved(lod_comp);
7317 if (!prev->llc_stripe) {
7318 CDEBUG(D_LAYOUT, "Previous component not inited\n");
7319 info->lti_count = 1;
7320 info->lti_comp_idx[0] = index - 1;
7321 rc = lod_declare_instantiate_components(env, lo, th, reserve);
7322 /* ENOSPC tells us we can't use this component. If there is
7323 * a next or we are repeating, we either spill over (next) or
7324 * extend the original comp (repeat). Otherwise, return the
7325 * error to the user. */
7326 if (rc == -ENOSPC && (next || sd->sd_repeat))
7332 if (sd->sd_force == 0 && rc == 0)
7333 rc = !lod_sel_osts_allowed(env, lo, index - 1, reserve, extent,
7334 &lod_comp->llc_extent, write);
7336 repeated = !!(sd->sd_repeat);
7340 /* Extend previous component */
7342 new_end = lod_extension_new_end(extension_size, extent->e_end,
7343 prev->llc_stripe_size,
7344 prev->llc_extent.e_end,
7345 lod_comp->llc_extent.e_end);
7347 CDEBUG(D_LAYOUT, "new end %llu\n", new_end);
7348 lod_comp->llc_extent.e_start = new_end;
7349 prev->llc_extent.e_end = new_end;
7351 if (prev->llc_extent.e_end == lod_comp->llc_extent.e_end) {
7352 CDEBUG(D_LAYOUT, "Extension component exhausted\n");
7353 lod_comp->llc_id = LCME_ID_INVAL;
7357 /* rc == 1, failed to extend current component */
7360 /* Normal 'spillover' case - Remove the extension
7361 * space component & bring down the start of the next
7363 lod_comp->llc_id = LCME_ID_INVAL;
7365 if (!(prev->llc_flags & LCME_FL_INIT)) {
7366 prev->llc_id = LCME_ID_INVAL;
7369 lod_sel_adjust_extents(env, lo, *max_comp, index);
7370 } else if (lod_comp_inited(prev)) {
7371 /* If there is no next, and the previous component is
7372 * INIT'ed, try repeating the previous component. */
7373 LASSERT(repeated == 0);
7374 rc = lod_layout_repeat_comp(env, lo, index - 1);
7378 /* The previous component is a repeated component.
7379 * Record this so we don't keep trying to repeat it. */
7382 /* If the previous component is not INIT'ed, this may
7383 * be a component we have just instantiated but failed
7384 * to extend. Or even a repeated component we failed
7385 * to prepare a striping for. Do not repeat but instead
7386 * remove the repeated component & force the extention
7387 * of the original one */
7390 prev->llc_id = LCME_ID_INVAL;
7397 rc = lod_layout_del_prep_layout(env, lo, NULL);
7400 LASSERTF(-rc == change,
7401 "number deleted %d != requested %d\n", -rc,
7404 *max_comp = *max_comp + change;
7406 /* lod_del_prep_layout reallocates ldo_comp_entries, so we must
7407 * refresh these pointers before using them */
7408 lod_comp = &lo->ldo_comp_entries[index];
7409 prev = &lo->ldo_comp_entries[index - 1];
7410 CDEBUG(D_LAYOUT, "After extent updates: prev start %llu, current start "
7411 "%llu, current end %llu max_comp %d ldo_comp_cnt %d\n",
7412 prev->llc_extent.e_start, lod_comp->llc_extent.e_start,
7413 lod_comp->llc_extent.e_end, *max_comp, lo->ldo_comp_cnt);
7415 /* Layout changed successfully */
7420 * Declare layout extent updates
7422 * Handles extensions. Identifies extension components touched by current
7423 * operation and passes them to processing function.
7425 * Restarts with updated layouts from the processing function until the current
7426 * operation no longer touches an extension space component.
7428 * \param[in] env execution environment for this thread
7429 * \param[in,out] lo object to update the layout of
7430 * \param[in] extent layout extent for requested operation, update layout to
7431 * fit this operation
7432 * \param[in] th transaction handle for this operation
7433 * \param[in] pick identifies chosen mirror for FLR layouts
7434 * \param[in] write if this is write op
7436 * \retval 1 on layout changed, 0 on no change
7437 * \retval negative errno on error
7439 static int lod_declare_update_extents(const struct lu_env *env,
7440 struct lod_object *lo, struct lu_extent *extent,
7441 struct thandle *th, int pick, int write)
7443 struct lod_thread_info *info = lod_env_info(env);
7444 struct lod_layout_component *lod_comp;
7445 bool layout_changed = false;
7446 struct sel_data sd = { 0 };
7454 /* This makes us work on the components of the chosen mirror */
7455 start_index = lo->ldo_mirrors[pick].lme_start;
7456 max_comp = lo->ldo_mirrors[pick].lme_end + 1;
7457 if (lo->ldo_flr_state == LCM_FL_NONE)
7458 LASSERT(start_index == 0 && max_comp == lo->ldo_comp_cnt);
7460 CDEBUG(D_LAYOUT, "extent->e_start %llu, extent->e_end %llu\n",
7461 extent->e_start, extent->e_end);
7462 for (i = start_index; i < max_comp; i++) {
7463 lod_comp = &lo->ldo_comp_entries[i];
7465 /* We've passed all components of interest */
7466 if (lod_comp->llc_extent.e_start >= extent->e_end)
7469 if (lod_comp->llc_flags & LCME_FL_EXTENSION) {
7470 layout_changed = true;
7471 rc = lod_sel_handler(env, lo, extent, th, &max_comp,
7476 /* Nothing has changed behind the prev one */
7482 /* We may have added or removed components. If so, we must update the
7483 * start & ends of all the mirrors after the current one, and the end
7484 * of the current mirror. */
7485 change = max_comp - 1 - lo->ldo_mirrors[pick].lme_end;
7487 lo->ldo_mirrors[pick].lme_end += change;
7488 for (i = pick + 1; i < lo->ldo_mirror_count; i++) {
7489 lo->ldo_mirrors[i].lme_start += change;
7490 lo->ldo_mirrors[i].lme_end += change;
7496 /* The amount of components has changed, adjust the lti_comp_idx */
7497 rc2 = lod_layout_data_init(info, lo->ldo_comp_cnt);
7499 return rc < 0 ? rc : rc2 < 0 ? rc2 : layout_changed;
7502 /* If striping is already instantiated or INIT'ed DOM? */
7503 static bool lod_is_instantiation_needed(struct lod_layout_component *comp)
7505 if (comp->llc_magic == LOV_MAGIC_FOREIGN)
7508 return !(((lov_pattern(comp->llc_pattern) & LOV_PATTERN_MDT) &&
7509 lod_comp_inited(comp)) || comp->llc_stripe);
7513 * Declare layout update for a non-FLR layout.
7515 * \param[in] env execution environment for this thread
7516 * \param[in,out] lo object to update the layout of
7517 * \param[in] layout layout intent for requested operation, "update" is
7518 * a process of reacting to this
7519 * \param[in] buf buffer containing lov ea (see comment on usage inline)
7520 * \param[in] th transaction handle for this operation
7522 * \retval 0 on success
7523 * \retval negative errno on error
7525 static int lod_declare_update_plain(const struct lu_env *env,
7526 struct lod_object *lo, struct layout_intent *layout,
7527 const struct lu_buf *buf, struct thandle *th)
7529 struct lod_thread_info *info = lod_env_info(env);
7530 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7531 struct lod_layout_component *lod_comp;
7532 struct lov_comp_md_v1 *comp_v1 = NULL;
7533 bool layout_changed = false;
7534 bool replay = false;
7538 LASSERT(lo->ldo_flr_state == LCM_FL_NONE);
7541 * In case the client is passing lovea, which only happens during
7542 * the replay of layout intent write RPC for now, we may need to
7543 * parse the lovea and apply new layout configuration.
7545 if (buf && buf->lb_len) {
7546 struct lov_user_md_v1 *v1 = buf->lb_buf;
7548 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
7549 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
7550 LOV_MAGIC_COMP_V1)) {
7551 CERROR("%s: the replay buffer of layout extend "
7552 "(magic %#x) does not contain expected "
7553 "composite layout.\n",
7554 lod2obd(d)->obd_name, v1->lmm_magic);
7555 GOTO(out, rc = -EINVAL);
7558 rc = lod_use_defined_striping(env, lo, buf);
7561 lo->ldo_comp_cached = 1;
7563 rc = lod_get_lov_ea(env, lo);
7566 /* old on-disk EA is stored in info->lti_buf */
7567 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
7569 layout_changed = true;
7571 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
7575 /* non replay path */
7576 rc = lod_striping_load(env, lo);
7581 /* Make sure defined layout covers the requested write range. */
7582 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
7583 if (lo->ldo_comp_cnt > 1 &&
7584 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
7585 lod_comp->llc_extent.e_end < layout->lai_extent.e_end) {
7586 CDEBUG_LIMIT(replay ? D_ERROR : D_LAYOUT,
7587 "%s: the defined layout [0, %#llx) does not "
7588 "covers the write range "DEXT"\n",
7589 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
7590 PEXT(&layout->lai_extent));
7591 GOTO(out, rc = -EINVAL);
7594 CDEBUG(D_LAYOUT, "%s: "DFID": update components "DEXT"\n",
7595 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
7596 PEXT(&layout->lai_extent));
7599 rc = lod_declare_update_extents(env, lo, &layout->lai_extent,
7600 th, 0, layout->lai_opc == LAYOUT_INTENT_WRITE);
7604 layout_changed = true;
7608 * Iterate ld->ldo_comp_entries, find the component whose extent under
7609 * the write range and not instantianted.
7611 for (i = 0; i < lo->ldo_comp_cnt; i++) {
7612 lod_comp = &lo->ldo_comp_entries[i];
7614 if (lod_comp->llc_extent.e_start >= layout->lai_extent.e_end)
7618 /* If striping is instantiated or INIT'ed DOM skip */
7619 if (!lod_is_instantiation_needed(lod_comp))
7623 * In replay path, lod_comp is the EA passed by
7624 * client replay buffer, comp_v1 is the pre-recovery
7625 * on-disk EA, we'd sift out those components which
7626 * were init-ed in the on-disk EA.
7628 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
7633 * this component hasn't instantiated in normal path, or during
7634 * replay it needs replay the instantiation.
7637 /* A released component is being extended */
7638 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
7639 GOTO(out, rc = -EINVAL);
7641 LASSERT(info->lti_comp_idx != NULL);
7642 info->lti_comp_idx[info->lti_count++] = i;
7643 layout_changed = true;
7646 if (!layout_changed)
7649 lod_obj_inc_layout_gen(lo);
7650 rc = lod_declare_instantiate_components(env, lo, th, 0);
7654 lod_striping_free(env, lo);
7658 static inline int lod_comp_index(struct lod_object *lo,
7659 struct lod_layout_component *lod_comp)
7661 LASSERT(lod_comp >= lo->ldo_comp_entries &&
7662 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
7664 return lod_comp - lo->ldo_comp_entries;
7668 * Stale other mirrors by writing extent.
7670 static int lod_stale_components(const struct lu_env *env, struct lod_object *lo,
7671 int primary, struct lu_extent *extent,
7674 struct lod_layout_component *pri_comp, *lod_comp;
7675 struct lu_extent pri_extent;
7680 /* The writing extent decides which components in the primary
7681 * are affected... */
7682 CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
7685 lod_foreach_mirror_comp(pri_comp, lo, primary) {
7686 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
7689 CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
7690 lod_comp_index(lo, pri_comp),
7691 PEXT(&pri_comp->llc_extent));
7693 pri_extent.e_start = pri_comp->llc_extent.e_start;
7694 pri_extent.e_end = pri_comp->llc_extent.e_end;
7696 for (i = 0; i < lo->ldo_mirror_count; i++) {
7699 rc = lod_declare_update_extents(env, lo, &pri_extent,
7701 /* if update_extents changed the layout, it may have
7702 * reallocated the component array, so start over to
7703 * avoid using stale pointers */
7709 /* ... and then stale other components that are
7710 * overlapping with primary components */
7711 lod_foreach_mirror_comp(lod_comp, lo, i) {
7712 if (!lu_extent_is_overlapped(
7714 &lod_comp->llc_extent))
7717 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
7718 i, lod_comp_index(lo, lod_comp));
7720 lod_comp->llc_flags |= LCME_FL_STALE;
7721 lo->ldo_mirrors[i].lme_stale = 1;
7722 if (lod_is_hsm(lod_comp))
7723 lod_comp->llc_foreign_flags |= HS_DIRTY;
7732 * check an OST's availability
7733 * \param[in] env execution environment
7734 * \param[in] lo lod object
7735 * \param[in] dt dt object
7736 * \param[in] index mirror index
7738 * \retval negative if failed
7739 * \retval 1 if \a dt is available
7740 * \retval 0 if \a dt is not available
7742 static inline int lod_check_ost_avail(const struct lu_env *env,
7743 struct lod_object *lo,
7744 struct dt_object *dt, int index)
7746 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7747 struct lod_tgt_desc *ost;
7749 int type = LU_SEQ_RANGE_OST;
7752 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &idx, &type);
7754 CERROR("%s: can't locate "DFID":rc = %d\n",
7755 lod2obd(lod)->obd_name, PFID(lu_object_fid(&dt->do_lu)),
7760 ost = OST_TGT(lod, idx);
7761 if (ost->ltd_active == 0) {
7762 CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail\n",
7763 PFID(lod_object_fid(lo)), index, idx);
7771 * Pick primary mirror for write
7772 * \param[in] env execution environment
7773 * \param[in] lo object
7774 * \param[in] extent write range
7776 static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo,
7777 struct lu_extent *extent)
7779 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7780 unsigned int seq = 0;
7781 struct lod_layout_component *lod_comp;
7783 int picked = -1, second_pick = -1, third_pick = -1;
7786 if (CFS_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
7787 get_random_bytes(&seq, sizeof(seq));
7788 seq %= lo->ldo_mirror_count;
7792 * Pick a mirror as the primary, and check the availability of OSTs.
7794 * This algo can be revised later after knowing the topology of
7797 lod_qos_statfs_update(env, lod, &lod->lod_ost_descs);
7799 rc = lod_fill_mirrors(lo);
7803 for (i = 0; i < lo->ldo_mirror_count; i++) {
7804 bool ost_avail = true;
7805 int index = (i + seq) % lo->ldo_mirror_count;
7807 if (lo->ldo_mirrors[index].lme_stale) {
7808 CDEBUG(D_LAYOUT, DFID": mirror %d stale\n",
7809 PFID(lod_object_fid(lo)), index);
7813 /* 2nd pick is for the primary mirror containing unavail OST */
7814 if (lo->ldo_mirrors[index].lme_prefer && second_pick < 0)
7815 second_pick = index;
7817 /* 3rd pick is for non-primary mirror containing unavail OST */
7818 if (second_pick < 0 && third_pick < 0)
7822 * we found a non-primary 1st pick, we'd like to find a
7823 * potential pirmary mirror.
7825 if (picked >= 0 && !lo->ldo_mirrors[index].lme_prefer)
7828 /* check the availability of OSTs */
7829 lod_foreach_mirror_comp(lod_comp, lo, index) {
7830 if (!lod_comp_inited(lod_comp) || !lod_comp->llc_stripe)
7833 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
7834 struct dt_object *dt = lod_comp->llc_stripe[j];
7836 rc = lod_check_ost_avail(env, lo, dt, index);
7843 } /* for all dt object in one component */
7846 } /* for all components in a mirror */
7849 * the OSTs where allocated objects locates in the components
7850 * of the mirror are available.
7855 /* this mirror has all OSTs available */
7859 * primary with all OSTs are available, this is the perfect
7862 if (lo->ldo_mirrors[index].lme_prefer)
7864 } /* for all mirrors */
7866 /* failed to pick a sound mirror, lower our expectation */
7868 picked = second_pick;
7870 picked = third_pick;
7877 static int lod_prepare_resync_mirror(const struct lu_env *env,
7878 struct lod_object *lo,
7881 struct lod_thread_info *info = lod_env_info(env);
7882 struct lod_layout_component *lod_comp;
7883 bool neg = !!(MIRROR_ID_NEG & mirror_id);
7886 mirror_id &= ~MIRROR_ID_NEG;
7888 for (i = 0; i < lo->ldo_mirror_count; i++) {
7889 if ((!neg && lo->ldo_mirrors[i].lme_id != mirror_id) ||
7890 (neg && lo->ldo_mirrors[i].lme_id == mirror_id))
7893 lod_foreach_mirror_comp(lod_comp, lo, i) {
7894 if (lod_comp_inited(lod_comp))
7897 info->lti_comp_idx[info->lti_count++] =
7898 lod_comp_index(lo, lod_comp);
7906 * figure out the components should be instantiated for resync.
7908 static int lod_prepare_resync(const struct lu_env *env, struct lod_object *lo,
7909 struct lu_extent *extent)
7911 struct lod_thread_info *info = lod_env_info(env);
7912 struct lod_layout_component *lod_comp;
7913 unsigned int need_sync = 0;
7917 DFID": instantiate all stale components in "DEXT"\n",
7918 PFID(lod_object_fid(lo)), PEXT(extent));
7921 * instantiate all components within this extent, even non-stale
7924 for (i = 0; i < lo->ldo_mirror_count; i++) {
7925 if (!lo->ldo_mirrors[i].lme_stale)
7928 lod_foreach_mirror_comp(lod_comp, lo, i) {
7929 if (!lu_extent_is_overlapped(extent,
7930 &lod_comp->llc_extent))
7935 if (lod_comp_inited(lod_comp))
7938 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
7939 i, lod_comp_index(lo, lod_comp));
7940 info->lti_comp_idx[info->lti_count++] =
7941 lod_comp_index(lo, lod_comp);
7945 return need_sync ? 0 : -EALREADY;
7948 static struct lod_layout_component *
7949 lod_locate_comp_hsm(struct lod_object *lo, int *hsm_mirror_id)
7951 struct lod_layout_component *lod_comp = NULL;
7954 if (!lo->ldo_is_composite)
7957 for (i = 0; i < lo->ldo_mirror_count; i++) {
7959 * FIXME: In the current design, there is only one HSM
7960 * mirror component in range [0, EOF] for a FLR file. This
7961 * should be fixed to support multiple HSM mirror components
7962 * with different HSM backend types and partial file ranges
7965 if (lo->ldo_mirrors[i].lme_hsm) {
7971 start_idx = lo->ldo_mirrors[i].lme_start;
7972 end_idx = lo->ldo_mirrors[i].lme_end;
7973 LASSERT(start_idx == end_idx);
7974 lod_comp = &lo->ldo_comp_entries[start_idx];
7975 LASSERT(lo->ldo_is_composite && lod_is_hsm(lod_comp) &&
7976 lod_comp->llc_extent.e_start == 0 &&
7977 lod_comp->llc_extent.e_end == LUSTRE_EOF);
7985 static int lod_declare_pccro_set(const struct lu_env *env,
7986 struct dt_object *dt, struct thandle *th)
7988 struct lod_thread_info *info = lod_env_info(env);
7989 struct lu_buf *buf = &info->lti_buf;
7990 struct lod_object *lo = lod_dt_obj(dt);
7991 struct lod_layout_component *lod_comp;
7992 struct lod_layout_component *comp_array;
7993 struct lod_mirror_entry *mirror_array;
8003 rc = lod_striping_load(env, lo);
8007 if (lo->ldo_flr_state & LCM_FL_PCC_RDONLY)
8010 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
8014 lod_comp = lod_locate_comp_hsm(lo, &hsm_mirror_id);
8016 if (lod_comp->llc_foreign_flags & HS_PCCRO) {
8017 CDEBUG(D_LAYOUT, "bad HSM flags: %#x\n",
8018 lod_comp->llc_foreign_flags);
8022 lod_obj_inc_layout_gen(lo);
8023 lod_comp->llc_foreign_flags |= HS_PCCRO;
8024 lod_comp->llc_foreign_flags &= ~HS_DIRTY;
8025 lod_comp->llc_flags &= ~LCME_FL_STALE;
8026 lo->ldo_mirrors[hsm_mirror_id].lme_stale = 0;
8027 lo->ldo_flr_state |= LCM_FL_PCC_RDONLY;
8028 buf->lb_len = lod_comp_md_size(lo, false);
8029 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
8030 buf, XATTR_NAME_LOV, 0, th);
8035 * Create an new composite layout with only one HSM component.
8036 * Field @lhm_archive_uuid is used to be the identifier within HSM
8037 * backend for the archive copy. In the PCC case with a POSIX archive,
8038 * This can just be the original inode FID. This is important because
8039 * the inode FID may change due to layout swaps or migration to a new
8040 * MDT, and we do not want that to cause problems with finding the copy
8043 mirror_cnt = lo->ldo_mirror_count + 1;
8044 if (!lo->ldo_is_composite) {
8045 LASSERT(lo->ldo_mirror_count == 0);
8049 OBD_ALLOC_PTR_ARRAY(mirror_array, mirror_cnt);
8050 if (mirror_array == NULL)
8053 new_cnt = lo->ldo_comp_cnt + 1;
8054 OBD_ALLOC_PTR_ARRAY(comp_array, new_cnt);
8055 if (comp_array == NULL) {
8056 OBD_FREE_PTR_ARRAY(mirror_array, mirror_cnt);
8061 for (i = 0; i < lo->ldo_comp_cnt; i++) {
8062 lod_comp = &lo->ldo_comp_entries[i];
8065 * Add mirror from a non-flr file, create new mirror ID.
8066 * Otherwise, keep existing mirror's component ID, used
8067 * for mirror extension.
8069 if (lo->ldo_mirror_count == 0 &&
8070 mirror_id_of(lod_comp->llc_id) == 0)
8071 lod_comp->llc_id = pflr_id(1, i + 1);
8073 if (lod_comp->llc_id != LCME_ID_INVAL &&
8074 mirror_id_of(lod_comp->llc_id) > mirror_id)
8075 mirror_id = mirror_id_of(lod_comp->llc_id);
8077 if (!lo->ldo_is_composite) {
8078 lod_comp->llc_extent.e_start = 0;
8079 lod_comp->llc_extent.e_end = LUSTRE_EOF;
8080 lod_comp_set_init(lod_comp);
8084 memcpy(comp_array, lo->ldo_comp_entries,
8085 sizeof(*comp_array) * lo->ldo_comp_cnt);
8087 lod_comp = &comp_array[new_cnt - 1];
8088 lod_comp->llc_magic = LOV_MAGIC_FOREIGN;
8089 lod_comp->llc_extent.e_start = 0;
8090 lod_comp->llc_extent.e_end = LUSTRE_EOF;
8091 lod_comp->llc_length = sizeof(struct lov_hsm_base);
8092 lod_comp->llc_type = LU_FOREIGN_TYPE_PCCRO;
8093 lod_comp->llc_foreign_flags = HS_EXISTS | HS_ARCHIVED | HS_PCCRO;
8094 memset(&lod_comp->llc_hsm, 0, sizeof(lod_comp->llc_hsm));
8096 if (lo->ldo_mirrors)
8097 OBD_FREE_PTR_ARRAY(lo->ldo_mirrors, lo->ldo_mirror_count);
8098 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
8101 * The @ldo_mirror will be refilled by lod_fill_mirrors() when
8102 * call lod_striped_create() for layout change.
8104 lo->ldo_mirrors = mirror_array;
8105 lo->ldo_mirror_count = mirror_cnt;
8106 lo->ldo_comp_entries = comp_array;
8107 lo->ldo_comp_cnt = new_cnt;
8108 lo->ldo_is_composite = 1;
8111 lod_comp->llc_id = LCME_ID_INVAL;
8112 lod_comp->llc_id = lod_gen_component_id(lo, mirror_id, new_cnt - 1);
8114 if (lo->ldo_flr_state == LCM_FL_NONE)
8115 lo->ldo_flr_state = LCM_FL_RDONLY;
8116 lo->ldo_flr_state |= LCM_FL_PCC_RDONLY;
8117 buf->lb_len = lod_comp_md_size(lo, false);
8118 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
8119 buf, XATTR_NAME_LOV, 0, th);
8121 lod_striping_free(env, lo);
8127 * TODO: When clear LCM_FL_PCC_RDONLY flag from the layouts, it means the file
8128 * is going to be modified. Currently it needs two RPCs: first one is to clear
8129 * LCM_FL_PCC_RDONLY flag; the second one is to pick primary mirror and mark
8130 * the file as LCM_FL_WRITE_PENDING.
8131 * These two RPCs can be combined in one RPC call.
8133 static int lod_declare_pccro_clear(const struct lu_env *env,
8134 struct dt_object *dt, struct thandle *th)
8136 struct lod_thread_info *info = lod_env_info(env);
8137 struct lod_object *lo = lod_dt_obj(dt);
8138 struct lod_layout_component *lod_comp;
8143 rc = lod_striping_load(env, lo);
8147 if (!(lo->ldo_flr_state & LCM_FL_PCC_RDONLY))
8150 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
8154 lod_comp = lod_locate_comp_hsm(lo, NULL);
8155 if (lod_comp == NULL) {
8156 CDEBUG(D_LAYOUT, "Not found any HSM component\n");
8157 GOTO(out, rc = -EINVAL);
8160 lod_comp->llc_foreign_flags &= ~HS_PCCRO;
8161 lo->ldo_flr_state &= ~LCM_FL_PCC_RDONLY;
8162 lod_obj_inc_layout_gen(lo);
8163 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
8164 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
8165 &info->lti_buf, XATTR_NAME_LOV, 0, th);
8168 lod_striping_free(env, lo);
8173 static int lod_declare_update_pccro(const struct lu_env *env,
8174 struct dt_object *dt,
8175 struct md_layout_change *mlc,
8178 struct layout_intent *intent = mlc->mlc_intent;
8181 switch (intent->lai_opc) {
8182 case LAYOUT_INTENT_PCCRO_SET:
8183 rc = lod_declare_pccro_set(env, dt, th);
8185 case LAYOUT_INTENT_PCCRO_CLEAR:
8186 rc = lod_declare_pccro_clear(env, dt, th);
8196 static int lod_declare_update_rdonly(const struct lu_env *env,
8197 struct lod_object *lo, struct md_layout_change *mlc,
8200 struct lod_thread_info *info = lod_env_info(env);
8201 struct lu_attr *layout_attr = &info->lti_layout_attr;
8202 struct lod_layout_component *lod_comp;
8203 struct lu_extent extent = { 0 };
8207 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
8208 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
8209 mlc->mlc_opc == MD_LAYOUT_RESYNC);
8210 LASSERT(lo->ldo_mirror_count > 0);
8212 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
8213 struct layout_intent *layout = mlc->mlc_intent;
8214 int write = layout->lai_opc == LAYOUT_INTENT_WRITE;
8217 extent = layout->lai_extent;
8218 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
8219 PFID(lod_object_fid(lo)), PEXT(&extent));
8221 picked = lod_primary_pick(env, lo, &extent);
8225 CDEBUG(D_LAYOUT, DFID": picked mirror id %u as primary\n",
8226 PFID(lod_object_fid(lo)),
8227 lo->ldo_mirrors[picked].lme_id);
8229 /* Update extents of primary before staling */
8230 rc = lod_declare_update_extents(env, lo, &extent, th, picked,
8235 if (layout->lai_opc == LAYOUT_INTENT_TRUNC) {
8237 * trunc transfers [0, size) in the intent extent, we'd
8238 * stale components overlapping [size, eof).
8240 extent.e_start = extent.e_end;
8241 extent.e_end = OBD_OBJECT_EOF;
8244 /* stale overlapping components from other mirrors */
8245 rc = lod_stale_components(env, lo, picked, &extent, th);
8249 /* restore truncate intent extent */
8250 if (layout->lai_opc == LAYOUT_INTENT_TRUNC)
8251 extent.e_end = extent.e_start;
8253 /* instantiate components for the picked mirror, start from 0 */
8256 lod_foreach_mirror_comp(lod_comp, lo, picked) {
8257 if (!lu_extent_is_overlapped(&extent,
8258 &lod_comp->llc_extent))
8261 if (!lod_is_instantiation_needed(lod_comp))
8264 info->lti_comp_idx[info->lti_count++] =
8265 lod_comp_index(lo, lod_comp);
8268 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
8269 } else { /* MD_LAYOUT_RESYNC */
8273 * could contain multiple non-stale mirrors, so we need to
8274 * prep uninited all components assuming any non-stale mirror
8275 * could be picked as the primary mirror.
8277 if (mlc->mlc_mirror_id == 0) {
8279 for (i = 0; i < lo->ldo_mirror_count; i++) {
8280 if (lo->ldo_mirrors[i].lme_stale)
8283 lod_foreach_mirror_comp(lod_comp, lo, i) {
8284 if (!lod_comp_inited(lod_comp))
8288 lod_comp->llc_extent.e_end)
8290 lod_comp->llc_extent.e_end;
8293 rc = lod_prepare_resync(env, lo, &extent);
8297 /* mirror write, try to init its all components */
8298 rc = lod_prepare_resync_mirror(env, lo,
8299 mlc->mlc_mirror_id);
8304 /* change the file state to SYNC_PENDING */
8305 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
8308 /* Reset the layout version once it's becoming too large.
8309 * This way it can make sure that the layout version is
8310 * monotonously increased in this writing era. */
8311 lod_obj_inc_layout_gen(lo);
8313 rc = lod_declare_instantiate_components(env, lo, th, 0);
8317 layout_attr->la_valid = LA_LAYOUT_VERSION;
8318 layout_attr->la_layout_version = 0;
8319 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
8320 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
8321 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
8327 lod_striping_free(env, lo);
8331 static int lod_declare_update_write_pending(const struct lu_env *env,
8332 struct lod_object *lo, struct md_layout_change *mlc,
8335 struct lod_thread_info *info = lod_env_info(env);
8336 struct lu_attr *layout_attr = &info->lti_layout_attr;
8337 struct lod_layout_component *lod_comp;
8338 struct lu_extent extent = { 0 };
8344 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
8345 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
8346 mlc->mlc_opc == MD_LAYOUT_RESYNC);
8348 /* look for the first preferred mirror */
8349 for (i = 0; i < lo->ldo_mirror_count; i++) {
8350 if (lo->ldo_mirrors[i].lme_stale)
8352 if (lo->ldo_mirrors[i].lme_prefer == 0)
8354 if (lo->ldo_mirrors[i].lme_hsm)
8361 /* no primary, use any in-sync */
8362 for (i = 0; i < lo->ldo_mirror_count; i++) {
8363 if (lo->ldo_mirrors[i].lme_stale)
8369 CERROR(DFID ": doesn't have a primary mirror\n",
8370 PFID(lod_object_fid(lo)));
8371 GOTO(out, rc = -ENODATA);
8375 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
8376 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
8378 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
8380 /* for LAYOUT_WRITE opc, it has to do the following operations:
8381 * 1. stale overlapping componets from stale mirrors;
8382 * 2. instantiate components of the primary mirror;
8383 * 3. transfter layout version to all objects of the primary;
8385 * for LAYOUT_RESYNC opc, it will do:
8386 * 1. instantiate components of all stale mirrors;
8387 * 2. transfer layout version to all objects to close write era. */
8389 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
8390 struct layout_intent *layout = mlc->mlc_intent;
8391 int write = layout->lai_opc == LAYOUT_INTENT_WRITE;
8393 LASSERT(layout != NULL);
8395 extent = layout->lai_extent;
8397 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
8398 PFID(lod_object_fid(lo)), PEXT(&extent));
8400 /* 1. Update extents of primary before staling */
8401 rc = lod_declare_update_extents(env, lo, &extent, th, primary,
8406 if (layout->lai_opc == LAYOUT_INTENT_TRUNC) {
8408 * trunc transfers [0, size) in the intent extent, we'd
8409 * stale components overlapping [size, eof).
8411 extent.e_start = extent.e_end;
8412 extent.e_end = OBD_OBJECT_EOF;
8415 /* 2. stale overlapping components */
8416 rc = lod_stale_components(env, lo, primary, &extent, th);
8420 /* 3. find the components which need instantiating.
8421 * instantiate [0, mlc->mlc_intent->e_end) */
8423 /* restore truncate intent extent */
8424 if (layout->lai_opc == LAYOUT_INTENT_TRUNC)
8425 extent.e_end = extent.e_start;
8428 lod_foreach_mirror_comp(lod_comp, lo, primary) {
8429 if (!lu_extent_is_overlapped(&extent,
8430 &lod_comp->llc_extent))
8433 if (!lod_is_instantiation_needed(lod_comp))
8436 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
8437 primary, lod_comp_index(lo, lod_comp));
8438 info->lti_comp_idx[info->lti_count++] =
8439 lod_comp_index(lo, lod_comp);
8441 } else { /* MD_LAYOUT_RESYNC */
8442 if (mlc->mlc_mirror_id == 0) {
8444 lod_foreach_mirror_comp(lod_comp, lo, primary) {
8445 if (!lod_comp_inited(lod_comp))
8448 extent.e_end = lod_comp->llc_extent.e_end;
8451 rc = lod_prepare_resync(env, lo, &extent);
8455 /* mirror write, try to init its all components */
8456 rc = lod_prepare_resync_mirror(env, lo,
8457 mlc->mlc_mirror_id);
8462 /* change the file state to SYNC_PENDING */
8463 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
8466 rc = lod_declare_instantiate_components(env, lo, th, 0);
8470 lod_obj_inc_layout_gen(lo);
8472 /* 3. transfer layout version to OST objects.
8473 * transfer new layout version to OST objects so that stale writes
8474 * can be denied. It also ends an era of writing by setting
8475 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
8476 * send write RPC; only resync RPCs could do it. */
8477 layout_attr->la_valid = LA_LAYOUT_VERSION;
8478 layout_attr->la_layout_version = 0;
8479 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
8480 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
8481 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
8486 lod_striping_free(env, lo);
8490 static int lod_declare_update_sync_pending(const struct lu_env *env,
8491 struct lod_object *lo, struct md_layout_change *mlc,
8494 struct lod_thread_info *info = lod_env_info(env);
8495 struct lu_attr *layout_attr = &info->lti_layout_attr;
8496 unsigned sync_components = 0;
8497 unsigned resync_components = 0;
8502 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
8503 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
8504 mlc->mlc_opc == MD_LAYOUT_WRITE);
8506 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
8507 PFID(lod_object_fid(lo)), mlc->mlc_opc);
8509 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
8510 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
8511 PFID(lod_object_fid(lo)));
8513 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
8514 return lod_declare_update_write_pending(env, lo, mlc, th);
8517 /* MD_LAYOUT_RESYNC_DONE */
8519 for (i = 0; i < lo->ldo_comp_cnt; i++) {
8520 struct lod_layout_component *lod_comp;
8523 lod_comp = &lo->ldo_comp_entries[i];
8525 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
8530 for (j = 0; j < mlc->mlc_resync_count; j++) {
8531 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
8534 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
8535 lod_comp->llc_flags &= ~LCME_FL_STALE;
8536 resync_components++;
8542 for (i = 0; i < mlc->mlc_resync_count; i++) {
8543 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
8546 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
8547 "or already synced\n", PFID(lod_object_fid(lo)),
8548 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
8549 GOTO(out, rc = -EINVAL);
8552 if (!sync_components || (mlc->mlc_resync_count && !resync_components)) {
8553 CDEBUG(D_LAYOUT, DFID": no mirror in sync\n",
8554 PFID(lod_object_fid(lo)));
8556 /* tend to return an error code here to prevent
8557 * the MDT from setting SoM attribute */
8558 GOTO(out, rc = -EINVAL);
8561 CDEBUG(D_LAYOUT, DFID": synced %u resynced %u/%zu components\n",
8562 PFID(lod_object_fid(lo)),
8563 sync_components, resync_components, mlc->mlc_resync_count);
8565 lo->ldo_flr_state = LCM_FL_RDONLY;
8566 lod_obj_inc_layout_gen(lo);
8568 layout_attr->la_valid = LA_LAYOUT_VERSION;
8569 layout_attr->la_layout_version = 0;
8570 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
8574 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
8575 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
8576 &info->lti_buf, XATTR_NAME_LOV, 0, th);
8581 lod_striping_free(env, lo);
8585 typedef int (*mlc_handler)(const struct lu_env *env, struct dt_object *dt,
8586 const struct md_layout_change *mlc,
8587 struct thandle *th);
8590 * Attach stripes after target's for migrating directory. NB, we
8591 * only need to declare this, the actual work is done inside
8592 * lod_xattr_set_lmv().
8594 * \param[in] env execution environment
8595 * \param[in] dt target object
8596 * \param[in] mlc layout change data
8597 * \param[in] th transaction handle
8599 * \retval 0 on success
8600 * \retval negative if failed
8602 static int lod_dir_declare_layout_attach(const struct lu_env *env,
8603 struct dt_object *dt,
8604 const struct md_layout_change *mlc,
8607 struct lod_thread_info *info = lod_env_info(env);
8608 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
8609 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
8610 struct lod_object *lo = lod_dt_obj(dt);
8611 struct dt_object *next = dt_object_child(dt);
8612 struct dt_object_format *dof = &info->lti_format;
8613 struct lmv_mds_md_v1 *lmv = mlc->mlc_buf.lb_buf;
8614 struct dt_object **stripes;
8615 __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
8616 struct lu_fid *fid = &info->lti_fid;
8617 struct lod_tgt_desc *tgt;
8618 struct dt_object *dto;
8619 struct dt_device *tgt_dt;
8620 int type = LU_SEQ_RANGE_ANY;
8621 struct dt_insert_rec *rec = &info->lti_dt_rec;
8622 char *stripe_name = info->lti_key;
8623 struct lu_name *sname;
8624 struct linkea_data ldata = { NULL };
8625 struct lu_buf linkea_buf;
8632 if (!lmv_is_sane(lmv))
8635 if (!dt_try_as_dir(env, dt, false))
8638 dof->dof_type = DFT_DIR;
8640 OBD_ALLOC_PTR_ARRAY(stripes, (lo->ldo_dir_stripe_count + stripe_count));
8644 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
8645 stripes[i] = lo->ldo_stripe[i];
8647 rec->rec_type = S_IFDIR;
8649 for (i = 0; i < stripe_count; i++) {
8651 &lmv->lmv_stripe_fids[i]);
8652 if (!fid_is_sane(fid))
8655 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
8659 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
8660 tgt_dt = lod->lod_child;
8662 tgt = LTD_TGT(ltd, idx);
8664 GOTO(out, rc = -ESTALE);
8665 tgt_dt = tgt->ltd_tgt;
8668 dto = dt_locate_at(env, tgt_dt, fid,
8669 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
8672 GOTO(out, rc = PTR_ERR(dto));
8674 stripes[i + lo->ldo_dir_stripe_count] = dto;
8676 if (!dt_try_as_dir(env, dto, true))
8677 GOTO(out, rc = -ENOTDIR);
8679 rc = lod_sub_declare_ref_add(env, dto, th);
8683 rec->rec_fid = lu_object_fid(&dto->do_lu);
8684 rc = lod_sub_declare_insert(env, dto,
8685 (const struct dt_rec *)rec,
8686 (const struct dt_key *)dot, th);
8690 rc = lod_sub_declare_insert(env, dto,
8691 (const struct dt_rec *)rec,
8692 (const struct dt_key *)dotdot, th);
8696 rc = lod_sub_declare_xattr_set(env, dto, &mlc->mlc_buf,
8697 XATTR_NAME_LMV, 0, th);
8701 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
8702 PFID(lu_object_fid(&dto->do_lu)),
8703 i + lo->ldo_dir_stripe_count);
8705 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
8706 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
8707 sname, lu_object_fid(&dt->do_lu));
8711 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
8712 linkea_buf.lb_len = ldata.ld_leh->leh_len;
8713 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
8714 XATTR_NAME_LINK, 0, th);
8718 rc = lod_sub_declare_insert(env, next,
8719 (const struct dt_rec *)rec,
8720 (const struct dt_key *)stripe_name,
8725 rc = lod_sub_declare_ref_add(env, next, th);
8731 OBD_FREE_PTR_ARRAY(lo->ldo_stripe,
8732 lo->ldo_dir_stripes_allocated);
8733 lo->ldo_stripe = stripes;
8734 lo->ldo_is_foreign = 0;
8735 lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count;
8736 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type);
8737 lo->ldo_dir_stripe_count += stripe_count;
8738 lo->ldo_dir_layout_version++;
8739 lo->ldo_dir_stripes_allocated += stripe_count;
8741 /* plain directory split creates target as a plain directory, while
8742 * after source attached as the first stripe, it becomes a striped
8743 * directory, set correct do_index_ops, otherwise it can't be unlinked.
8745 dt->do_index_ops = &lod_striped_index_ops;
8749 i = lo->ldo_dir_stripe_count;
8750 while (i < lo->ldo_dir_stripe_count + stripe_count && stripes[i])
8751 dt_object_put(env, stripes[i++]);
8753 OBD_FREE_PTR_ARRAY(stripes, stripe_count + lo->ldo_dir_stripe_count);
8757 static int lod_dir_declare_layout_detach(const struct lu_env *env,
8758 struct dt_object *dt,
8759 const struct md_layout_change *unused,
8762 struct lod_thread_info *info = lod_env_info(env);
8763 struct lod_object *lo = lod_dt_obj(dt);
8764 struct dt_object *next = dt_object_child(dt);
8765 char *stripe_name = info->lti_key;
8766 struct dt_object *dto;
8770 if (!dt_try_as_dir(env, dt, true))
8773 if (!lo->ldo_dir_stripe_count)
8774 return lod_sub_declare_delete(env, next,
8775 (const struct dt_key *)dotdot, th);
8777 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8778 dto = lo->ldo_stripe[i];
8782 if (!dt_try_as_dir(env, dto, true))
8785 rc = lod_sub_declare_delete(env, dto,
8786 (const struct dt_key *)dotdot, th);
8790 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8791 PFID(lu_object_fid(&dto->do_lu)), i);
8793 rc = lod_sub_declare_delete(env, next,
8794 (const struct dt_key *)stripe_name, th);
8798 rc = lod_sub_declare_ref_del(env, next, th);
8806 static int dt_dir_is_empty(const struct lu_env *env,
8807 struct dt_object *obj)
8810 const struct dt_it_ops *iops;
8815 if (!dt_try_as_dir(env, obj, true))
8818 iops = &obj->do_index_ops->dio_it;
8819 it = iops->init(env, obj, LUDA_64BITHASH);
8821 RETURN(PTR_ERR(it));
8823 rc = iops->get(env, it, (const struct dt_key *)"");
8827 for (rc = 0, i = 0; rc == 0 && i < 3; ++i)
8828 rc = iops->next(env, it);
8834 /* Huh? Index contains no zero key? */
8839 iops->fini(env, it);
8844 static int lod_dir_declare_layout_shrink(const struct lu_env *env,
8845 struct dt_object *dt,
8846 const struct md_layout_change *mlc,
8849 struct lod_thread_info *info = lod_env_info(env);
8850 struct lod_object *lo = lod_dt_obj(dt);
8851 struct dt_object *next = dt_object_child(dt);
8852 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
8853 char *stripe_name = info->lti_key;
8854 struct lu_buf *lmv_buf = &info->lti_buf;
8855 __u32 final_stripe_count;
8856 struct dt_object *dto;
8862 if (!dt_try_as_dir(env, dt, true))
8865 /* shouldn't be called on plain directory */
8866 LASSERT(lo->ldo_dir_stripe_count);
8868 lmv_buf->lb_buf = &info->lti_lmv.lmv_md_v1;
8869 lmv_buf->lb_len = sizeof(info->lti_lmv.lmv_md_v1);
8871 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
8872 LASSERT(final_stripe_count &&
8873 final_stripe_count < lo->ldo_dir_stripe_count);
8875 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8876 dto = lo->ldo_stripe[i];
8880 if (i < final_stripe_count) {
8881 rc = lod_sub_declare_xattr_set(env, dto, lmv_buf,
8883 LU_XATTR_REPLACE, th);
8890 rc = dt_dir_is_empty(env, dto);
8894 rc = lod_sub_declare_ref_del(env, dto, th);
8898 rc = lod_sub_declare_destroy(env, dto, th);
8902 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8903 PFID(lu_object_fid(&dto->do_lu)), i);
8905 rc = lod_sub_declare_delete(env, next,
8906 (const struct dt_key *)stripe_name, th);
8910 rc = lod_sub_declare_ref_del(env, next, th);
8915 rc = lod_sub_declare_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
8916 LU_XATTR_REPLACE, th);
8921 * Allocate stripes for split directory.
8923 * \param[in] env execution environment
8924 * \param[in] dt target object
8925 * \param[in] mlc layout change data
8926 * \param[in] th transaction handle
8928 * \retval 0 on success
8929 * \retval negative if failed
8931 static int lod_dir_declare_layout_split(const struct lu_env *env,
8932 struct dt_object *dt,
8933 const struct md_layout_change *mlc,
8936 struct lod_thread_info *info = lod_env_info(env);
8937 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
8938 struct lod_object *lo = lod_dt_obj(dt);
8939 struct dt_object_format *dof = &info->lti_format;
8940 struct lmv_user_md_v1 *lum = mlc->mlc_spec->u.sp_ea.eadata;
8941 struct dt_object **stripes;
8949 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
8950 LASSERT(le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT);
8952 saved_count = lo->ldo_dir_stripes_allocated;
8953 stripe_count = le32_to_cpu(lum->lum_stripe_count);
8954 if (stripe_count <= saved_count)
8957 dof->dof_type = DFT_DIR;
8959 OBD_ALLOC(stripes, sizeof(*stripes) * stripe_count);
8963 for (i = 0; i < lo->ldo_dir_stripes_allocated; i++)
8964 stripes[i] = lo->ldo_stripe[i];
8966 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
8967 rc = lod_mdt_alloc_qos(env, lo, stripes, saved_count, stripe_count);
8969 rc = lod_mdt_alloc_rr(env, lo, stripes, saved_count,
8972 OBD_FREE(stripes, sizeof(*stripes) * stripe_count);
8976 LASSERT(rc > saved_count);
8977 OBD_FREE(lo->ldo_stripe,
8978 sizeof(*stripes) * lo->ldo_dir_stripes_allocated);
8979 lo->ldo_stripe = stripes;
8980 lo->ldo_is_foreign = 0;
8981 lo->ldo_dir_striped = 1;
8982 lo->ldo_dir_stripe_count = rc;
8983 lo->ldo_dir_stripes_allocated = stripe_count;
8984 lo->ldo_dir_split_hash = lo->ldo_dir_hash_type;
8985 lo->ldo_dir_hash_type = le32_to_cpu(lum->lum_hash_type);
8986 if (!lmv_is_known_hash_type(lo->ldo_dir_hash_type))
8987 lo->ldo_dir_hash_type =
8988 lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
8989 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_SPLIT | LMV_HASH_FLAG_MIGRATION;
8990 lo->ldo_dir_split_offset = saved_count;
8991 lo->ldo_dir_layout_version++;
8992 lo->ldo_dir_stripe_loaded = 1;
8994 rc = lod_dir_declare_create_stripes(env, dt, mlc->mlc_attr, dof, th);
8996 lod_striping_free(env, lo);
9002 * detach all stripes from dir master object, NB, stripes are not destroyed, but
9003 * deleted from it's parent namespace, this function is called in two places:
9004 * 1. mdd_migrate_mdt() detach stripes from source, and attach them to
9006 * 2. mdd_dir_layout_update() detach stripe before turning 1-stripe directory to
9007 * a plain directory.
9009 * \param[in] env execution environment
9010 * \param[in] dt target object
9011 * \param[in] mlc layout change data
9012 * \param[in] th transaction handle
9014 * \retval 0 on success
9015 * \retval negative if failed
9017 static int lod_dir_layout_detach(const struct lu_env *env,
9018 struct dt_object *dt,
9019 const struct md_layout_change *mlc,
9022 struct lod_thread_info *info = lod_env_info(env);
9023 struct lod_object *lo = lod_dt_obj(dt);
9024 struct dt_object *next = dt_object_child(dt);
9025 char *stripe_name = info->lti_key;
9026 struct dt_object *dto;
9032 if (!lo->ldo_dir_stripe_count) {
9033 /* plain directory delete .. */
9034 rc = lod_sub_delete(env, next,
9035 (const struct dt_key *)dotdot, th);
9039 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
9040 dto = lo->ldo_stripe[i];
9044 rc = lod_sub_delete(env, dto,
9045 (const struct dt_key *)dotdot, th);
9049 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
9050 PFID(lu_object_fid(&dto->do_lu)), i);
9052 rc = lod_sub_delete(env, next,
9053 (const struct dt_key *)stripe_name, th);
9057 rc = lod_sub_ref_del(env, next, th);
9062 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
9063 dto = lo->ldo_stripe[i];
9065 dt_object_put(env, dto);
9067 OBD_FREE_PTR_ARRAY(lo->ldo_stripe, lo->ldo_dir_stripes_allocated);
9068 lo->ldo_stripe = NULL;
9069 lo->ldo_dir_stripes_allocated = 0;
9070 lo->ldo_dir_stripe_count = 0;
9071 dt->do_index_ops = &lod_index_ops;
9076 static int lod_dir_layout_shrink(const struct lu_env *env,
9077 struct dt_object *dt,
9078 const struct md_layout_change *mlc,
9081 struct lod_thread_info *info = lod_env_info(env);
9082 struct lod_object *lo = lod_dt_obj(dt);
9083 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
9084 struct dt_object *next = dt_object_child(dt);
9085 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
9086 __u32 final_stripe_count;
9087 char *stripe_name = info->lti_key;
9088 struct dt_object *dto;
9089 struct lu_buf *lmv_buf = &info->lti_buf;
9090 struct lmv_mds_md_v1 *lmv = &info->lti_lmv.lmv_md_v1;
9092 int type = LU_SEQ_RANGE_ANY;
9098 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
9100 lmv_buf->lb_buf = lmv;
9101 lmv_buf->lb_len = sizeof(*lmv);
9102 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
9103 lmv->lmv_stripe_count = cpu_to_le32(final_stripe_count);
9104 lmv->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type) &
9105 cpu_to_le32(LMV_HASH_TYPE_MASK |
9106 LMV_HASH_FLAG_FIXED);
9107 lmv->lmv_layout_version =
9108 cpu_to_le32(lo->ldo_dir_layout_version + 1);
9109 lmv->lmv_migrate_offset = 0;
9110 lmv->lmv_migrate_hash = 0;
9112 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
9113 dto = lo->ldo_stripe[i];
9117 if (i < final_stripe_count) {
9118 rc = lod_fld_lookup(env, lod,
9119 lu_object_fid(&dto->do_lu),
9124 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
9125 rc = lod_sub_xattr_set(env, dto, lmv_buf,
9127 LU_XATTR_REPLACE, th);
9134 dt_write_lock(env, dto, DT_TGT_CHILD);
9135 rc = lod_sub_ref_del(env, dto, th);
9136 dt_write_unlock(env, dto);
9140 rc = lod_sub_destroy(env, dto, th);
9144 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
9145 PFID(lu_object_fid(&dto->do_lu)), i);
9147 rc = lod_sub_delete(env, next,
9148 (const struct dt_key *)stripe_name, th);
9152 rc = lod_sub_ref_del(env, next, th);
9157 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &mdtidx,
9162 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
9163 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
9164 rc = lod_sub_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
9165 LU_XATTR_REPLACE, th);
9169 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
9170 dto = lo->ldo_stripe[i];
9172 dt_object_put(env, dto);
9174 lo->ldo_dir_stripe_count = final_stripe_count;
9179 static mlc_handler dir_mlc_declare_ops[MD_LAYOUT_MAX] = {
9180 [MD_LAYOUT_ATTACH] = lod_dir_declare_layout_attach,
9181 [MD_LAYOUT_DETACH] = lod_dir_declare_layout_detach,
9182 [MD_LAYOUT_SHRINK] = lod_dir_declare_layout_shrink,
9183 [MD_LAYOUT_SPLIT] = lod_dir_declare_layout_split,
9186 static mlc_handler dir_mlc_ops[MD_LAYOUT_MAX] = {
9187 [MD_LAYOUT_DETACH] = lod_dir_layout_detach,
9188 [MD_LAYOUT_SHRINK] = lod_dir_layout_shrink,
9191 static int lod_declare_layout_change(const struct lu_env *env,
9192 struct dt_object *dt, struct md_layout_change *mlc,
9195 struct lod_thread_info *info = lod_env_info(env);
9196 struct lod_object *lo = lod_dt_obj(dt);
9201 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
9202 LASSERT(dir_mlc_declare_ops[mlc->mlc_opc]);
9203 rc = dir_mlc_declare_ops[mlc->mlc_opc](env, dt, mlc, th);
9207 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
9208 dt_object_remote(dt_object_child(dt)))
9211 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
9212 struct layout_intent *intent = mlc->mlc_intent;
9214 if (intent->lai_opc == LAYOUT_INTENT_PCCRO_SET ||
9215 intent->lai_opc == LAYOUT_INTENT_PCCRO_CLEAR) {
9216 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
9219 rc = lod_declare_update_pccro(env, dt, mlc, th);
9224 rc = lod_striping_load(env, lo);
9228 LASSERT(lo->ldo_comp_cnt > 0);
9230 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
9234 switch (lo->ldo_flr_state) {
9236 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
9240 rc = lod_declare_update_rdonly(env, lo, mlc, th);
9242 case LCM_FL_WRITE_PENDING:
9243 rc = lod_declare_update_write_pending(env, lo, mlc, th);
9245 case LCM_FL_SYNC_PENDING:
9246 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
9257 * Instantiate layout component objects which covers the intent write offset.
9259 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
9260 struct md_layout_change *mlc, struct thandle *th)
9262 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
9263 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
9264 struct lod_object *lo = lod_dt_obj(dt);
9269 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
9270 LASSERT(dir_mlc_ops[mlc->mlc_opc]);
9271 rc = dir_mlc_ops[mlc->mlc_opc](env, dt, mlc, th);
9275 rc = lod_striped_create(env, dt, attr, NULL, th);
9276 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
9277 layout_attr->la_layout_version |= lo->ldo_layout_gen;
9278 rc = lod_attr_set(env, dt, layout_attr, th);
9284 const struct dt_object_operations lod_obj_ops = {
9285 .do_read_lock = lod_read_lock,
9286 .do_write_lock = lod_write_lock,
9287 .do_read_unlock = lod_read_unlock,
9288 .do_write_unlock = lod_write_unlock,
9289 .do_write_locked = lod_write_locked,
9290 .do_attr_get = lod_attr_get,
9291 .do_declare_attr_set = lod_declare_attr_set,
9292 .do_attr_set = lod_attr_set,
9293 .do_xattr_get = lod_xattr_get,
9294 .do_declare_xattr_set = lod_declare_xattr_set,
9295 .do_xattr_set = lod_xattr_set,
9296 .do_declare_xattr_del = lod_declare_xattr_del,
9297 .do_xattr_del = lod_xattr_del,
9298 .do_xattr_list = lod_xattr_list,
9299 .do_ah_init = lod_ah_init,
9300 .do_declare_create = lod_declare_create,
9301 .do_create = lod_create,
9302 .do_declare_destroy = lod_declare_destroy,
9303 .do_destroy = lod_destroy,
9304 .do_index_try = lod_index_try,
9305 .do_declare_ref_add = lod_declare_ref_add,
9306 .do_ref_add = lod_ref_add,
9307 .do_declare_ref_del = lod_declare_ref_del,
9308 .do_ref_del = lod_ref_del,
9309 .do_object_sync = lod_object_sync,
9310 .do_object_lock = lod_object_lock,
9311 .do_object_unlock = lod_object_unlock,
9312 .do_invalidate = lod_invalidate,
9313 .do_declare_layout_change = lod_declare_layout_change,
9314 .do_layout_change = lod_layout_change,
9318 * Implementation of dt_body_operations::dbo_read.
9320 * \see dt_body_operations::dbo_read() in the API description for details.
9322 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
9323 struct lu_buf *buf, loff_t *pos)
9325 struct dt_object *next = dt_object_child(dt);
9327 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
9328 S_ISLNK(dt->do_lu.lo_header->loh_attr));
9329 return next->do_body_ops->dbo_read(env, next, buf, pos);
9333 * Implementation of dt_body_operations::dbo_declare_write.
9335 * \see dt_body_operations::dbo_declare_write() in the API description
9338 static ssize_t lod_declare_write(const struct lu_env *env,
9339 struct dt_object *dt,
9340 const struct lu_buf *buf, loff_t pos,
9343 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
9347 * Implementation of dt_body_operations::dbo_write.
9349 * \see dt_body_operations::dbo_write() in the API description for details.
9351 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
9352 const struct lu_buf *buf, loff_t *pos,
9355 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
9356 S_ISLNK(dt->do_lu.lo_header->loh_attr));
9357 return lod_sub_write(env, dt_object_child(dt), buf, pos, th);
9360 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
9361 __u64 start, __u64 end, struct thandle *th)
9363 if (dt_object_remote(dt))
9366 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
9369 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
9370 __u64 start, __u64 end, struct thandle *th)
9372 if (dt_object_remote(dt))
9375 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
9376 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
9380 * different type of files use the same body_ops because object may be created
9381 * in OUT, where there is no chance to set correct body_ops for each type, so
9382 * body_ops themselves will check file type inside, see lod_read/write/punch for
9385 static const struct dt_body_operations lod_body_ops = {
9386 .dbo_read = lod_read,
9387 .dbo_declare_write = lod_declare_write,
9388 .dbo_write = lod_write,
9389 .dbo_declare_punch = lod_declare_punch,
9390 .dbo_punch = lod_punch,
9394 * Implementation of lu_object_operations::loo_object_init.
9396 * The function determines the type and the index of the target device using
9397 * sequence of the object's FID. Then passes control down to the
9398 * corresponding device:
9399 * OSD for the local objects, OSP for remote
9401 * \see lu_object_operations::loo_object_init() in the API description
9404 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
9405 const struct lu_object_conf *conf)
9407 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
9408 struct lu_device *cdev = NULL;
9409 struct lu_object *cobj;
9410 struct lod_tgt_descs *ltd = NULL;
9411 struct lod_tgt_desc *tgt;
9413 int type = LU_SEQ_RANGE_ANY;
9417 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
9421 if (type == LU_SEQ_RANGE_MDT &&
9422 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
9423 cdev = &lod->lod_child->dd_lu_dev;
9424 } else if (type == LU_SEQ_RANGE_MDT) {
9425 ltd = &lod->lod_mdt_descs;
9427 } else if (type == LU_SEQ_RANGE_OST) {
9428 ltd = &lod->lod_ost_descs;
9435 if (ltd->ltd_tgts_size > idx &&
9436 test_bit(idx, ltd->ltd_tgt_bitmap)) {
9437 tgt = LTD_TGT(ltd, idx);
9439 LASSERT(tgt != NULL);
9440 LASSERT(tgt->ltd_tgt != NULL);
9442 cdev = &(tgt->ltd_tgt->dd_lu_dev);
9444 lod_putref(lod, ltd);
9447 if (unlikely(cdev == NULL))
9450 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
9451 if (unlikely(cobj == NULL))
9454 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
9456 lu_object_add(lo, cobj);
9463 * Release resources associated with striping.
9465 * If the object is striped (regular or directory), then release
9466 * the stripe objects references and free the ldo_stripe array.
9468 * \param[in] env execution environment
9469 * \param[in] lo object
9471 void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo)
9473 struct lod_layout_component *lod_comp;
9474 __u32 obj_attr = lo->ldo_obj.do_lu.lo_header->loh_attr;
9477 if (unlikely(lo->ldo_is_foreign)) {
9478 if (S_ISREG(obj_attr)) {
9479 lod_free_foreign_lov(lo);
9480 lo->ldo_comp_cached = 0;
9481 } else if (S_ISDIR(obj_attr)) {
9482 lod_free_foreign_lmv(lo);
9483 lo->ldo_dir_stripe_loaded = 0;
9485 } else if (lo->ldo_stripe != NULL) {
9486 LASSERT(lo->ldo_comp_entries == NULL);
9487 LASSERT(lo->ldo_dir_stripes_allocated > 0);
9489 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
9490 if (lo->ldo_stripe[i])
9491 dt_object_put(env, lo->ldo_stripe[i]);
9494 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
9495 OBD_FREE(lo->ldo_stripe, j);
9496 lo->ldo_stripe = NULL;
9497 lo->ldo_dir_stripes_allocated = 0;
9498 lo->ldo_dir_stripe_loaded = 0;
9499 lo->ldo_dir_stripe_count = 0;
9500 lo->ldo_obj.do_index_ops = NULL;
9501 } else if (lo->ldo_comp_entries != NULL) {
9502 for (i = 0; i < lo->ldo_comp_cnt; i++) {
9503 /* free lod_layout_component::llc_stripe array */
9504 lod_comp = &lo->ldo_comp_entries[i];
9506 /* HSM layout component */
9507 if (lod_comp->llc_magic == LOV_MAGIC_FOREIGN)
9509 if (lod_comp->llc_stripe == NULL)
9511 LASSERT(lod_comp->llc_stripes_allocated != 0);
9512 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
9513 if (lod_comp->llc_stripe[j] != NULL)
9515 &lod_comp->llc_stripe[j]->do_lu);
9517 OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe,
9518 lod_comp->llc_stripes_allocated);
9519 lod_comp->llc_stripe = NULL;
9520 OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices,
9521 lod_comp->llc_stripes_allocated);
9522 lod_comp->llc_ost_indices = NULL;
9523 lod_comp->llc_stripes_allocated = 0;
9525 lod_free_comp_entries(lo);
9526 lo->ldo_comp_cached = 0;
9530 void lod_striping_free(const struct lu_env *env, struct lod_object *lo)
9532 mutex_lock(&lo->ldo_layout_mutex);
9533 lod_striping_free_nolock(env, lo);
9534 mutex_unlock(&lo->ldo_layout_mutex);
9538 * Implementation of lu_object_operations::loo_object_free.
9540 * \see lu_object_operations::loo_object_free() in the API description
9543 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
9545 struct lod_object *lo = lu2lod_obj(o);
9547 /* release all underlying object pinned */
9548 lod_striping_free(env, lo);
9550 /* lo doesn't contain a lu_object_header, so we don't need call_rcu */
9551 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
9555 * Implementation of lu_object_operations::loo_object_release.
9557 * \see lu_object_operations::loo_object_release() in the API description
9560 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
9562 /* XXX: shouldn't we release everything here in case if object
9563 * creation failed before? */
9567 * Implementation of lu_object_operations::loo_object_print.
9569 * \see lu_object_operations::loo_object_print() in the API description
9572 static int lod_object_print(const struct lu_env *env, void *cookie,
9573 lu_printer_t p, const struct lu_object *l)
9575 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
9577 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
9580 const struct lu_object_operations lod_lu_obj_ops = {
9581 .loo_object_init = lod_object_init,
9582 .loo_object_free = lod_object_free,
9583 .loo_object_release = lod_object_release,
9584 .loo_object_print = lod_object_print,