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) && !data->locd_declare)
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;
2051 "try idx %d, mdt cnt %u, allocated %u, specific %d count %hu offset %d hash %#X\n",
2052 idx, lod->lod_remote_mdt_count + 1, stripe_idx,
2053 is_specific, lo->ldo_dir_stripe_count,
2054 (int)lo->ldo_dir_stripe_offset,
2055 lo->ldo_dir_hash_type);
2057 if (likely(!is_specific &&
2058 !CFS_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE) &&
2059 !(lo->ldo_dir_hash_type &
2060 LMV_HASH_FLAG_OVERSTRIPED))) {
2061 /* check whether the idx already exists
2062 * in current allocated array */
2063 for (k = 0; k < stripe_idx; k++) {
2064 if (mdt_indices[k] == idx) {
2065 already_allocated = true;
2070 if (already_allocated)
2074 /* Sigh, this index is not in the bitmap, let's check
2075 * next available target */
2076 if (!test_bit(idx, ltd->ltd_tgt_bitmap) &&
2077 idx != master_index)
2080 if (idx == master_index) {
2081 /* Allocate the FID locally */
2082 tgt_dt = lod->lod_child;
2083 rc = dt_fid_alloc(env, tgt_dt, &fid, NULL,
2090 /* check the status of the OSP */
2091 tgt = LTD_TGT(ltd, idx);
2095 tgt_dt = tgt->ltd_tgt;
2096 if (!tgt->ltd_active)
2097 /* this OSP doesn't feel well */
2100 if (tgt->ltd_statfs.os_state & OS_STATFS_NOCREATE)
2103 rc = dt_fid_alloc(env, tgt_dt, &fid, NULL, NULL);
2110 /* Can not allocate more stripes */
2111 if (j == lod->lod_remote_mdt_count) {
2112 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
2113 lod2obd(lod)->obd_name, stripe_count,
2118 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
2119 idx, stripe_idx, PFID(&fid));
2120 mdt_indices[stripe_idx] = idx;
2121 /* Set the start index for next stripe allocation */
2122 if (!is_specific && stripe_idx < stripe_count - 1) {
2124 * for large dir test, put all other slaves on one
2125 * remote MDT, otherwise we may save too many local
2126 * slave locks which will exceed RS_MAX_LOCKS.
2128 if (unlikely(CFS_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)))
2130 mdt_indices[stripe_idx + 1] = (idx + 1) %
2131 (lod->lod_remote_mdt_count + 1);
2133 /* tgt_dt and fid must be ready after search avaible OSP
2134 * in the above loop */
2135 LASSERT(tgt_dt != NULL);
2136 LASSERT(fid_is_sane(&fid));
2138 /* fail a remote stripe FID allocation */
2139 if (stripe_idx && CFS_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_FID))
2142 dto = dt_locate_at(env, tgt_dt, &fid,
2143 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
2150 stripes[stripe_idx] = dto;
2156 for (j = 1; j < stripe_idx; j++) {
2157 LASSERT(stripes[j] != NULL);
2158 dt_object_put(env, stripes[j]);
2164 static int lod_prep_md_striped_create(const struct lu_env *env,
2165 struct dt_object *dt,
2166 struct lu_attr *attr,
2167 const struct lmv_user_md_v1 *lum,
2168 struct dt_object_format *dof,
2171 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
2172 struct lod_object *lo = lod_dt_obj(dt);
2173 struct dt_object **stripes;
2174 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2175 struct lu_fid fid = { 0 };
2176 int mdt_count = lod->lod_remote_mdt_count + 1;
2183 /* The lum has been verifed in lod_verify_md_striping */
2184 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
2185 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
2187 stripe_count = lo->ldo_dir_stripe_count;
2188 if (!(lo->ldo_dir_hash_type & LMV_HASH_FLAG_OVERSTRIPED) &&
2189 stripe_count > mdt_count)
2192 if ((lo->ldo_dir_hash_type & LMV_HASH_FLAG_OVERSTRIPED) &&
2193 (stripe_count > mdt_count * LMV_MAX_STRIPES_PER_MDT ||
2194 /* a single MDT doesn't initialize the infrastructure for striped
2195 * directories, so we just don't support overstriping in that case
2200 OBD_ALLOC_PTR_ARRAY(stripes, stripe_count);
2204 /* Allocate the first stripe locally */
2205 rc = dt_fid_alloc(env, lod->lod_child, &fid, NULL, NULL);
2209 stripes[0] = dt_locate_at(env, lod->lod_child, &fid,
2210 dt->do_lu.lo_dev->ld_site->ls_top_dev, &conf);
2211 if (IS_ERR(stripes[0]))
2212 GOTO(out, rc = PTR_ERR(stripes[0]));
2214 if (lo->ldo_dir_stripe_offset == LMV_OFFSET_DEFAULT) {
2215 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
2216 rc = lod_mdt_alloc_qos(env, lo, stripes, 1, stripe_count);
2218 rc = lod_mdt_alloc_rr(env, lo, stripes, 1,
2222 bool is_specific = false;
2224 OBD_ALLOC_PTR_ARRAY(idx_array, stripe_count);
2226 GOTO(out, rc = -ENOMEM);
2228 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
2229 int stripes_per_mdt;
2234 /* Verify we do not exceed the stripes per MDT limit */
2235 for (mdt = 0; mdt < mdt_count + 1; mdt++) {
2236 stripes_per_mdt = 0;
2237 for (i = 0; i < stripe_count; i++) {
2238 if (mdt == le32_to_cpu(
2239 lum->lum_objects[i].lum_mds))
2242 if (stripes_per_mdt > LMV_MAX_STRIPES_PER_MDT)
2243 GOTO(out_free, rc = -EINVAL);
2246 for (i = 0; i < stripe_count; i++)
2248 le32_to_cpu(lum->lum_objects[i].lum_mds);
2251 /* stripe 0 is local */
2253 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
2254 rc = lod_mdt_alloc_specific(env, lo, stripes, idx_array,
2257 OBD_FREE_PTR_ARRAY(idx_array, stripe_count);
2265 lo->ldo_dir_striped = 1;
2266 lo->ldo_stripe = stripes;
2267 lo->ldo_dir_stripe_count = rc;
2268 lo->ldo_dir_stripes_allocated = stripe_count;
2270 lo->ldo_dir_stripe_loaded = 1;
2272 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2274 lod_striping_free(env, lo);
2280 if (!IS_ERR_OR_NULL(stripes[0]))
2281 dt_object_put(env, stripes[0]);
2282 for (i = 1; i < stripe_count; i++)
2283 LASSERT(!stripes[i]);
2284 OBD_FREE_PTR_ARRAY(stripes, stripe_count);
2291 * Alloc cached foreign LOV
2293 * \param[in] lo object
2294 * \param[in] size size of foreign LOV
2296 * \retval 0 on success
2297 * \retval negative if failed
2299 int lod_alloc_foreign_lov(struct lod_object *lo, size_t size)
2301 OBD_ALLOC_LARGE(lo->ldo_foreign_lov, size);
2302 if (lo->ldo_foreign_lov == NULL)
2304 lo->ldo_foreign_lov_size = size;
2305 lo->ldo_is_foreign = 1;
2311 * Free cached foreign LOV
2313 * \param[in] lo object
2315 void lod_free_foreign_lov(struct lod_object *lo)
2317 if (lo->ldo_foreign_lov != NULL)
2318 OBD_FREE_LARGE(lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
2319 lo->ldo_foreign_lov = NULL;
2320 lo->ldo_foreign_lov_size = 0;
2321 lo->ldo_is_foreign = 0;
2326 * Alloc cached foreign LMV
2328 * \param[in] lo object
2329 * \param[in] size size of foreign LMV
2331 * \retval 0 on success
2332 * \retval negative if failed
2334 static int lod_alloc_foreign_lmv(struct lod_object *lo, size_t size)
2336 OBD_ALLOC_LARGE(lo->ldo_foreign_lmv, size);
2337 if (lo->ldo_foreign_lmv == NULL)
2339 lo->ldo_foreign_lmv_size = size;
2340 lo->ldo_is_foreign = 1;
2345 static int lod_prep_md_replayed_create(const struct lu_env *env,
2346 struct dt_object *dt,
2347 struct lu_attr *attr,
2348 const struct lu_buf *lmv_buf,
2349 struct dt_object_format *dof,
2352 struct lod_object *lo = lod_dt_obj(dt);
2357 mutex_lock(&lo->ldo_layout_mutex);
2358 rc = lod_parse_dir_striping(env, lo, lmv_buf);
2360 lo->ldo_dir_stripe_loaded = 1;
2361 lo->ldo_dir_striped = 1;
2362 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2364 mutex_unlock(&lo->ldo_layout_mutex);
2371 * Free cached foreign LMV
2373 * \param[in] lo object
2375 static void lod_free_foreign_lmv(struct lod_object *lo)
2377 if (lo->ldo_foreign_lmv != NULL)
2378 OBD_FREE_LARGE(lo->ldo_foreign_lmv, lo->ldo_foreign_lmv_size);
2379 lo->ldo_foreign_lmv = NULL;
2380 lo->ldo_foreign_lmv_size = 0;
2381 lo->ldo_is_foreign = 0;
2385 * Declare create striped md object.
2387 * The function declares intention to create a striped directory. This is a
2388 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2389 * is to verify pattern \a lum_buf is good. Check that function for the details.
2391 * \param[in] env execution environment
2392 * \param[in] dt object
2393 * \param[in] attr attributes to initialize the objects with
2394 * \param[in] lum_buf a pattern specifying the number of stripes and
2396 * \param[in] dof type of objects to be created
2397 * \param[in] th transaction handle
2399 * \retval 0 on success
2400 * \retval negative if failed
2403 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2404 struct dt_object *dt,
2405 struct lu_attr *attr,
2406 const struct lu_buf *lum_buf,
2407 struct dt_object_format *dof,
2410 struct lod_object *lo = lod_dt_obj(dt);
2411 struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
2415 LASSERT(lum != NULL);
2418 "lum magic=%x hash=%x count=%u offset=%d inherit=%u rr=%u\n",
2419 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_hash_type),
2420 le32_to_cpu(lum->lum_stripe_count),
2421 (int)le32_to_cpu(lum->lum_stripe_offset),
2422 lum->lum_max_inherit, lum->lum_max_inherit_rr);
2424 if (lo->ldo_dir_stripe_count == 0) {
2425 if (lo->ldo_is_foreign) {
2426 rc = lod_alloc_foreign_lmv(lo, lum_buf->lb_len);
2429 memcpy(lo->ldo_foreign_lmv, lum, lum_buf->lb_len);
2430 lo->ldo_dir_stripe_loaded = 1;
2435 /* client replay striped directory creation with LMV, this happens when
2436 * all involved MDTs were rebooted, or MDT recovery was aborted.
2438 if (le32_to_cpu(lum->lum_magic) == LMV_MAGIC_V1)
2439 rc = lod_prep_md_replayed_create(env, dt, attr, lum_buf, dof,
2442 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2444 /* failed to create striping, let's reset
2445 * config so that others don't get confused */
2446 lod_striping_free(env, lo);
2452 * Set or replace striped directory layout, and LFSCK may set layout on a plain
2453 * directory, so don't check stripe count.
2455 * \param[in] env execution environment
2456 * \param[in] dt target object
2457 * \param[in] lmv_buf LMV buf which contains source stripe FIDs
2458 * \param[in] fl set or replace
2459 * \param[in] th transaction handle
2461 * \retval 0 on success
2462 * \retval negative if failed
2464 static int lod_dir_layout_set(const struct lu_env *env,
2465 struct dt_object *dt,
2466 const struct lu_buf *lmv_buf,
2470 struct dt_object *next = dt_object_child(dt);
2471 struct lod_object *lo = lod_dt_obj(dt);
2472 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2473 struct lmv_mds_md_v1 *lmv = lmv_buf->lb_buf;
2474 struct lmv_mds_md_v1 *slave_lmv;
2475 struct lu_buf slave_buf;
2481 if (!lmv_is_sane2(lmv))
2484 /* adjust hash for dir merge, which may not be set in user command */
2485 if (lmv_is_merging(lmv) &&
2486 !(lmv->lmv_migrate_hash & LMV_HASH_TYPE_MASK))
2487 lmv->lmv_merge_hash |=
2488 lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern &
2491 LMV_DEBUG(D_INFO, lmv, "set");
2493 rc = lod_sub_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV, fl, th);
2497 /* directory restripe may update stripe LMV directly */
2498 if (!lo->ldo_dir_stripe_count)
2501 lo->ldo_dir_hash_type = le32_to_cpu(lmv->lmv_hash_type);
2502 lo->ldo_dir_migrate_offset = le32_to_cpu(lmv->lmv_migrate_offset);
2503 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_migrate_hash);
2504 lo->ldo_dir_layout_version = le32_to_cpu(lmv->lmv_layout_version);
2506 OBD_ALLOC_PTR(slave_lmv);
2510 lod_prep_slave_lmv_md(slave_lmv, lmv);
2511 slave_buf.lb_buf = slave_lmv;
2512 slave_buf.lb_len = sizeof(*slave_lmv);
2514 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2515 if (!lo->ldo_stripe[i])
2518 if (!dt_object_exists(lo->ldo_stripe[i]))
2521 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], &slave_buf,
2522 XATTR_NAME_LMV, fl, th);
2527 OBD_FREE_PTR(slave_lmv);
2533 * Implementation of dt_object_operations::do_declare_xattr_set.
2535 * Used with regular (non-striped) objects. Basically it
2536 * initializes the striping information and applies the
2537 * change to all the stripes.
2539 * \see dt_object_operations::do_declare_xattr_set() in the API description
2542 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2543 struct dt_object *dt,
2544 const struct lu_buf *buf,
2545 const char *name, int fl,
2548 struct dt_object *next = dt_object_child(dt);
2549 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2550 struct lod_object *lo = lod_dt_obj(dt);
2555 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2556 struct lmv_user_md_v1 *lum;
2558 LASSERT(buf != NULL);
2559 if (!buf->lb_buf || buf->lb_len < sizeof(*lum))
2563 rc = lod_verify_md_striping(d, lum);
2566 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2567 rc = lod_verify_striping(env, d, lo, buf, false);
2572 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2576 /* Note: Do not set LinkEA on sub-stripes, otherwise
2577 * it will confuse the fid2path process(see mdt_path_current()).
2578 * The linkEA between master and sub-stripes is set in
2579 * lod_xattr_set_lmv(). */
2580 if (strcmp(name, XATTR_NAME_LINK) == 0)
2583 /* set xattr to each stripes, if needed */
2584 rc = lod_striping_load(env, lo);
2588 if (lo->ldo_dir_stripe_count == 0)
2591 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2592 if (!lo->ldo_stripe[i])
2595 if (!dt_object_exists(lo->ldo_stripe[i]))
2598 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2608 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2609 struct lod_object *lo,
2610 struct dt_object *dt, struct thandle *th,
2611 int comp_idx, int stripe_idx,
2612 struct lod_obj_stripe_cb_data *data)
2614 struct lod_thread_info *info = lod_env_info(env);
2615 struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
2616 struct filter_fid *ff = &info->lti_ff;
2617 struct lu_buf *buf = &info->lti_buf;
2621 buf->lb_len = sizeof(*ff);
2622 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2630 * locd_buf is set if it's called by dir migration, which doesn't check
2633 if (data->locd_buf) {
2634 memset(ff, 0, sizeof(*ff));
2635 ff->ff_parent = *(struct lu_fid *)data->locd_buf->lb_buf;
2637 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
2639 if (lu_fid_eq(lod_object_fid(lo), &ff->ff_parent) &&
2640 ff->ff_layout.ol_comp_id == comp->llc_id)
2643 memset(ff, 0, sizeof(*ff));
2644 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2647 /* rewrite filter_fid */
2648 ff->ff_parent.f_ver = stripe_idx;
2649 ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
2650 ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
2651 ff->ff_layout.ol_comp_id = comp->llc_id;
2652 ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
2653 ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
2654 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
2656 if (data->locd_declare)
2657 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2658 LU_XATTR_REPLACE, th);
2660 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2661 LU_XATTR_REPLACE, th);
2667 * Reset parent FID on OST object
2669 * Replace parent FID with @dt object FID, which is only called during migration
2670 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2671 * the FID is changed.
2673 * \param[in] env execution environment
2674 * \param[in] dt dt_object whose stripes's parent FID will be reset
2675 * \parem[in] th thandle
2676 * \param[in] declare if it is declare
2678 * \retval 0 if reset succeeds
2679 * \retval negative errno if reset fails
2681 static int lod_replace_parent_fid(const struct lu_env *env,
2682 struct dt_object *dt,
2683 const struct lu_buf *buf,
2684 struct thandle *th, bool declare)
2686 struct lod_object *lo = lod_dt_obj(dt);
2687 struct lod_thread_info *info = lod_env_info(env);
2688 struct filter_fid *ff;
2689 struct lod_obj_stripe_cb_data data = { { 0 } };
2693 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2695 /* set xattr to each stripes, if needed */
2696 rc = lod_striping_load(env, lo);
2700 if (!lod_obj_is_striped(dt))
2703 if (info->lti_ea_store_size < sizeof(*ff)) {
2704 rc = lod_ea_store_resize(info, sizeof(*ff));
2709 data.locd_declare = declare;
2710 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2711 data.locd_buf = buf;
2712 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2717 __u16 lod_comp_entry_stripe_count(struct lod_object *lo, int comp_idx,
2720 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2721 struct lod_layout_component *entry;
2722 enum lod_uses_hint flags = LOD_USES_ASSIGNED_STRIPE;
2725 entry = &lo->ldo_def_striping->lds_def_comp_entries[comp_idx];
2726 return entry->llc_ostlist.op_count;
2729 entry = &lo->ldo_comp_entries[comp_idx];
2730 if (lod_comp_inited(entry))
2731 return entry->llc_stripe_count;
2732 if (entry->llc_stripe_count == LOV_ALL_STRIPES)
2733 return lod_get_stripe_count_plain(lod, lo,
2734 entry->llc_stripe_count,
2735 entry->llc_pattern &
2736 LOV_PATTERN_OVERSTRIPING,
2739 return lod_get_stripe_count(lod, lo, comp_idx, entry->llc_stripe_count,
2740 entry->llc_pattern & LOV_PATTERN_OVERSTRIPING,
2744 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2746 int magic, size = 0, i;
2747 struct lod_layout_component *comp_entries;
2749 bool is_composite, is_foreign = false;
2752 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2753 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2755 lo->ldo_def_striping->lds_def_striping_is_composite;
2757 comp_cnt = lo->ldo_comp_cnt;
2758 comp_entries = lo->ldo_comp_entries;
2759 is_composite = lo->ldo_is_composite;
2760 is_foreign = lo->ldo_is_foreign;
2764 return lo->ldo_foreign_lov_size;
2766 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2768 size = sizeof(struct lov_comp_md_v1) +
2769 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2770 LASSERT(size % sizeof(__u64) == 0);
2773 for (i = 0; i < comp_cnt; i++) {
2776 if (comp_entries[i].llc_magic == LOV_MAGIC_FOREIGN) {
2777 size += lov_foreign_md_size(comp_entries[i].llc_length);
2779 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 :
2781 stripe_count = lod_comp_entry_stripe_count(lo, i,
2783 if (!is_dir && is_composite)
2784 lod_comp_shrink_stripe_count(&comp_entries[i],
2786 if (is_dir && comp_entries[i].llc_ostlist.op_count)
2787 magic = LOV_MAGIC_SPECIFIC;
2789 size += lov_user_md_size(stripe_count, magic);
2791 LASSERT(size % sizeof(__u64) == 0);
2797 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2798 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2801 * \param[in] env execution environment
2802 * \param[in] dt dt_object to add components on
2803 * \param[in] buf buffer contains components to be added
2804 * \parem[in] th thandle
2806 * \retval 0 on success
2807 * \retval negative errno on failure
2809 static int lod_declare_layout_add(const struct lu_env *env,
2810 struct dt_object *dt,
2811 const struct lu_buf *buf,
2814 struct lod_thread_info *info = lod_env_info(env);
2815 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2816 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2817 struct dt_object *next = dt_object_child(dt);
2818 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
2819 struct lod_object *lo = lod_dt_obj(dt);
2820 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2822 int i, rc, array_cnt, old_array_cnt;
2825 LASSERT(lo->ldo_is_composite);
2827 if (lo->ldo_flr_state != LCM_FL_NONE)
2830 rc = lod_verify_striping(env, d, lo, buf, false);
2834 magic = comp_v1->lcm_magic;
2835 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2836 lustre_swab_lov_comp_md_v1(comp_v1);
2837 magic = comp_v1->lcm_magic;
2840 if (magic != LOV_USER_MAGIC_COMP_V1)
2843 mutex_lock(&lo->ldo_layout_mutex);
2845 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2846 OBD_ALLOC_PTR_ARRAY(comp_array, array_cnt);
2847 if (comp_array == NULL) {
2848 mutex_unlock(&lo->ldo_layout_mutex);
2853 memcpy(comp_array, lo->ldo_comp_entries,
2854 sizeof(*comp_array) * lo->ldo_comp_cnt);
2856 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2857 struct lov_user_md_v1 *v1;
2858 struct lu_extent *ext;
2860 v1 = (struct lov_user_md *)((char *)comp_v1 +
2861 comp_v1->lcm_entries[i].lcme_offset);
2862 ext = &comp_v1->lcm_entries[i].lcme_extent;
2864 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2865 lod_comp->llc_extent.e_start = ext->e_start;
2866 lod_comp->llc_extent.e_end = ext->e_end;
2867 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2868 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2870 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2871 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2872 lod_comp->llc_pattern = v1->lmm_pattern;
2874 * limit stripe count so that it's less than/equal to
2875 * extent_size / stripe_size.
2877 * Note: extension size reused llc_stripe_size field and
2878 * uninstantiated component could be defined with
2879 * extent_start == extent_end as extension component will
2882 if (!(lod_comp->llc_flags & LCME_FL_EXTENSION) &&
2883 (lod_comp_inited(lod_comp) ||
2884 lod_comp->llc_extent.e_start <
2885 lod_comp->llc_extent.e_end) &&
2886 lod_comp->llc_stripe_count != LOV_ALL_STRIPES &&
2887 ext->e_end != OBD_OBJECT_EOF &&
2888 (__u64)(lod_comp->llc_stripe_count *
2889 lod_comp->llc_stripe_size) >
2890 (ext->e_end - ext->e_start))
2891 lod_comp->llc_stripe_count =
2892 DIV_ROUND_UP(ext->e_end - ext->e_start,
2893 lod_comp->llc_stripe_size);
2894 lod_adjust_stripe_info(lod_comp, desc, 0);
2896 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2897 struct lov_user_md_v3 *v3 = (typeof(*v3) *) v1;
2899 if (v3->lmm_pool_name[0] != '\0' &&
2900 !lov_pool_is_ignored(v3->lmm_pool_name)) {
2901 rc = lod_set_pool(&lod_comp->llc_pool,
2909 old_array = lo->ldo_comp_entries;
2910 old_array_cnt = lo->ldo_comp_cnt;
2912 lo->ldo_comp_entries = comp_array;
2913 lo->ldo_comp_cnt = array_cnt;
2915 /* No need to increase layout generation here, it will be increased
2916 * later when generating component ID for the new components */
2918 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2919 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2920 XATTR_NAME_LOV, 0, th);
2922 lo->ldo_comp_entries = old_array;
2923 lo->ldo_comp_cnt = old_array_cnt;
2927 OBD_FREE_PTR_ARRAY(old_array, old_array_cnt);
2929 LASSERT(lo->ldo_mirror_count == 1);
2930 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2932 mutex_unlock(&lo->ldo_layout_mutex);
2937 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2938 lod_comp = &comp_array[i];
2939 if (lod_comp->llc_pool != NULL) {
2940 OBD_FREE(lod_comp->llc_pool,
2941 strlen(lod_comp->llc_pool) + 1);
2942 lod_comp->llc_pool = NULL;
2945 OBD_FREE_PTR_ARRAY(comp_array, array_cnt);
2946 mutex_unlock(&lo->ldo_layout_mutex);
2952 * lod_last_non_stale_mirror() - Check if a mirror is the last non-stale mirror.
2953 * @mirror_id: Mirror id to be checked.
2956 * This function checks if a mirror with specified @mirror_id is the last
2957 * non-stale mirror of a LOD object @lo.
2959 * Return: true or false.
2962 bool lod_last_non_stale_mirror(__u16 mirror_id, struct lod_object *lo)
2964 struct lod_layout_component *lod_comp;
2965 bool has_stale_flag;
2968 for (i = 0; i < lo->ldo_mirror_count; i++) {
2969 if (lo->ldo_mirrors[i].lme_id == mirror_id ||
2970 lo->ldo_mirrors[i].lme_stale)
2973 has_stale_flag = false;
2974 lod_foreach_mirror_comp(lod_comp, lo, i) {
2975 if (lod_comp->llc_flags & LCME_FL_STALE) {
2976 has_stale_flag = true;
2980 if (!has_stale_flag)
2988 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2989 * the '$field' can only be 'flags' now. The xattr value is binary
2990 * lov_comp_md_v1 which contains the component ID(s) and the value of
2991 * the field to be modified.
2992 * Please update allowed_lustre_lov macro if $field groks more values
2995 * \param[in] env execution environment
2996 * \param[in] dt dt_object to be modified
2997 * \param[in] op operation string, like "set.flags"
2998 * \param[in] buf buffer contains components to be set
2999 * \parem[in] th thandle
3001 * \retval 0 on success
3002 * \retval negative errno on failure
3004 static int lod_declare_layout_set(const struct lu_env *env,
3005 struct dt_object *dt,
3006 char *op, const struct lu_buf *buf,
3009 struct lod_layout_component *lod_comp;
3010 struct lod_thread_info *info = lod_env_info(env);
3011 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3012 struct lod_object *lo = lod_dt_obj(dt);
3013 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3016 bool changed = false;
3019 /* Please update allowed_lustre_lov macro if op
3020 * groks more values in the future
3022 if (strcmp(op, "set.flags") != 0) {
3023 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
3024 lod2obd(d)->obd_name, op);
3028 magic = comp_v1->lcm_magic;
3029 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
3030 lustre_swab_lov_comp_md_v1(comp_v1);
3031 magic = comp_v1->lcm_magic;
3034 if (magic != LOV_USER_MAGIC_COMP_V1)
3037 if (comp_v1->lcm_entry_count == 0) {
3038 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
3039 lod2obd(d)->obd_name);
3043 mutex_lock(&lo->ldo_layout_mutex);
3044 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
3045 __u32 id = comp_v1->lcm_entries[i].lcme_id;
3046 __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
3047 __u32 mirror_flag = flags & LCME_MIRROR_FLAGS;
3048 __u16 mirror_id = mirror_id_of(id);
3049 bool neg = flags & LCME_FL_NEG;
3051 if (flags & LCME_FL_INIT) {
3053 lod_striping_free_nolock(env, lo);
3054 mutex_unlock(&lo->ldo_layout_mutex);
3058 flags &= ~(LCME_MIRROR_FLAGS | LCME_FL_NEG);
3059 for (j = 0; j < lo->ldo_comp_cnt; j++) {
3060 lod_comp = &lo->ldo_comp_entries[j];
3062 /* lfs only put one flag in each entry */
3063 if ((flags && id != lod_comp->llc_id) ||
3064 (mirror_flag && mirror_id !=
3065 mirror_id_of(lod_comp->llc_id)))
3070 lod_comp->llc_flags &= ~flags;
3072 lod_comp->llc_flags &= ~mirror_flag;
3075 if ((flags & LCME_FL_STALE) &&
3076 lod_last_non_stale_mirror(mirror_id,
3079 &lo->ldo_layout_mutex);
3082 lod_comp->llc_flags |= flags;
3085 lod_comp->llc_flags |= mirror_flag;
3086 if (mirror_flag & LCME_FL_NOSYNC)
3087 lod_comp->llc_timestamp =
3088 ktime_get_real_seconds();
3094 mutex_unlock(&lo->ldo_layout_mutex);
3097 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
3098 lod2obd(d)->obd_name);
3102 lod_obj_inc_layout_gen(lo);
3104 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3105 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
3106 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3111 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
3112 * and the xattr value is a unique component ID or a special lcme_id.
3114 * \param[in] env execution environment
3115 * \param[in] dt dt_object to be operated on
3116 * \param[in] buf buffer contains component ID or lcme_id
3117 * \parem[in] th thandle
3119 * \retval 0 on success
3120 * \retval negative errno on failure
3122 static int lod_declare_layout_del(const struct lu_env *env,
3123 struct dt_object *dt,
3124 const struct lu_buf *buf,
3127 struct lod_thread_info *info = lod_env_info(env);
3128 struct dt_object *next = dt_object_child(dt);
3129 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3130 struct lod_object *lo = lod_dt_obj(dt);
3131 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3132 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3133 __u32 magic, id, flags, neg_flags = 0;
3137 LASSERT(lo->ldo_is_composite);
3139 if (lo->ldo_flr_state != LCM_FL_NONE)
3142 magic = comp_v1->lcm_magic;
3143 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
3144 lustre_swab_lov_comp_md_v1(comp_v1);
3145 magic = comp_v1->lcm_magic;
3148 if (magic != LOV_USER_MAGIC_COMP_V1)
3151 id = comp_v1->lcm_entries[0].lcme_id;
3152 flags = comp_v1->lcm_entries[0].lcme_flags;
3154 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
3155 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
3156 lod2obd(d)->obd_name, id, flags);
3160 if (id != LCME_ID_INVAL && flags != 0) {
3161 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
3162 lod2obd(d)->obd_name);
3166 if (id == LCME_ID_INVAL && !flags) {
3167 CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
3168 lod2obd(d)->obd_name);
3172 if (flags & LCME_FL_NEG) {
3173 neg_flags = flags & ~LCME_FL_NEG;
3177 mutex_lock(&lo->ldo_layout_mutex);
3179 left = lo->ldo_comp_cnt;
3181 mutex_unlock(&lo->ldo_layout_mutex);
3185 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3186 struct lod_layout_component *lod_comp;
3188 lod_comp = &lo->ldo_comp_entries[i];
3190 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
3192 else if (flags && !(flags & lod_comp->llc_flags))
3194 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
3197 if (left != (i + 1)) {
3198 CDEBUG(D_LAYOUT, "%s: this deletion will create "
3199 "a hole.\n", lod2obd(d)->obd_name);
3200 mutex_unlock(&lo->ldo_layout_mutex);
3205 /* Mark the component as deleted */
3206 lod_comp->llc_id = LCME_ID_INVAL;
3208 /* Not instantiated component */
3209 if (lod_comp->llc_stripe == NULL)
3212 LASSERT(lod_comp->llc_stripe_count > 0);
3213 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
3214 struct dt_object *obj = lod_comp->llc_stripe[j];
3218 rc = lod_sub_declare_destroy(env, obj, th);
3220 mutex_unlock(&lo->ldo_layout_mutex);
3226 LASSERTF(left >= 0, "left = %d\n", left);
3227 if (left == lo->ldo_comp_cnt) {
3228 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
3229 lod2obd(d)->obd_name, id);
3230 mutex_unlock(&lo->ldo_layout_mutex);
3234 mutex_unlock(&lo->ldo_layout_mutex);
3236 memset(attr, 0, sizeof(*attr));
3237 attr->la_valid = LA_SIZE;
3238 rc = lod_sub_declare_attr_set(env, next, attr, th);
3243 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3244 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
3245 XATTR_NAME_LOV, 0, th);
3247 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
3254 * Declare layout add/set/del operations issued by special xattr names:
3256 * XATTR_LUSTRE_LOV.add add component(s) to existing file
3257 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
3258 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
3260 * \param[in] env execution environment
3261 * \param[in] dt object
3262 * \param[in] name name of xattr
3263 * \param[in] buf lu_buf contains xattr value
3264 * \param[in] th transaction handle
3266 * \retval 0 on success
3267 * \retval negative if failed
3269 static int lod_declare_modify_layout(const struct lu_env *env,
3270 struct dt_object *dt,
3272 const struct lu_buf *buf,
3275 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3276 struct lod_object *lo = lod_dt_obj(dt);
3278 int rc, len = strlen(XATTR_LUSTRE_LOV);
3281 LASSERT(dt_object_exists(dt));
3283 if (strlen(name) <= len || name[len] != '.') {
3284 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
3285 lod2obd(d)->obd_name, name);
3290 rc = lod_striping_load(env, lo);
3294 /* the layout to be modified must be a composite layout */
3295 if (!lo->ldo_is_composite) {
3296 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
3297 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3298 GOTO(unlock, rc = -EINVAL);
3301 op = (char *)name + len;
3302 if (strcmp(op, "add") == 0) {
3303 rc = lod_declare_layout_add(env, dt, buf, th);
3304 } else if (strcmp(op, "del") == 0) {
3305 rc = lod_declare_layout_del(env, dt, buf, th);
3306 } else if (strncmp(op, "set", strlen("set")) == 0) {
3307 rc = lod_declare_layout_set(env, dt, op, buf, th);
3309 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
3310 lod2obd(d)->obd_name, name);
3311 GOTO(unlock, rc = -ENOTSUPP);
3315 lod_striping_free(env, lo);
3321 * Convert a plain file lov_mds_md to a composite layout.
3323 * \param[in,out] info the thread info::lti_ea_store buffer contains little
3324 * endian plain file layout
3326 * \retval 0 on success, <0 on failure
3328 static int lod_layout_convert(struct lod_thread_info *info)
3330 struct lov_mds_md *lmm = info->lti_ea_store;
3331 struct lov_mds_md *lmm_save;
3332 struct lov_comp_md_v1 *lcm;
3333 struct lov_comp_md_entry_v1 *lcme;
3339 /* realloc buffer to a composite layout which contains one component */
3340 blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3341 le32_to_cpu(lmm->lmm_magic));
3342 size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
3344 OBD_ALLOC_LARGE(lmm_save, blob_size);
3346 GOTO(out, rc = -ENOMEM);
3348 memcpy(lmm_save, lmm, blob_size);
3350 if (info->lti_ea_store_size < size) {
3351 rc = lod_ea_store_resize(info, size);
3356 lcm = info->lti_ea_store;
3357 memset(lcm, 0, sizeof(*lcm) + sizeof(*lcme));
3358 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
3359 lcm->lcm_size = cpu_to_le32(size);
3360 lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
3361 lmm_save->lmm_layout_gen));
3362 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
3363 lcm->lcm_entry_count = cpu_to_le16(1);
3365 lcme = &lcm->lcm_entries[0];
3366 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
3367 lcme->lcme_extent.e_start = 0;
3368 lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
3369 lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
3370 lcme->lcme_size = cpu_to_le32(blob_size);
3372 memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
3377 OBD_FREE_LARGE(lmm_save, blob_size);
3382 * Merge layouts to form a mirrored file.
3384 static int lod_declare_layout_merge(const struct lu_env *env,
3385 struct dt_object *dt,
3386 const struct lu_buf *mbuf,
3389 struct lod_thread_info *info = lod_env_info(env);
3390 struct lu_attr *layout_attr = &info->lti_layout_attr;
3391 struct lu_buf *buf = &info->lti_buf;
3392 struct lod_object *lo = lod_dt_obj(dt);
3393 struct lov_comp_md_v1 *lcm;
3394 struct lov_comp_md_v1 *cur_lcm;
3395 struct lov_comp_md_v1 *merge_lcm;
3396 struct lov_comp_md_entry_v1 *lcme;
3397 struct lov_mds_md_v1 *lmm;
3400 __u16 cur_entry_count;
3401 __u16 merge_entry_count;
3403 __u16 mirror_id = 0;
3410 merge_lcm = mbuf->lb_buf;
3411 if (mbuf->lb_len < sizeof(*merge_lcm))
3414 /* must be an existing layout from disk */
3415 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
3418 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
3420 /* do not allow to merge two mirrored files */
3421 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
3424 /* verify the target buffer */
3425 rc = lod_get_lov_ea(env, lo);
3427 RETURN(rc ? : -ENODATA);
3429 cur_lcm = info->lti_ea_store;
3430 switch (le32_to_cpu(cur_lcm->lcm_magic)) {
3433 rc = lod_layout_convert(info);
3435 case LOV_MAGIC_COMP_V1:
3445 /* info->lti_ea_store could be reallocated in lod_layout_convert() */
3446 cur_lcm = info->lti_ea_store;
3447 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
3449 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
3450 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
3451 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
3454 /* size of new layout */
3455 size = le32_to_cpu(cur_lcm->lcm_size) +
3456 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
3458 memset(buf, 0, sizeof(*buf));
3459 lu_buf_alloc(buf, size);
3460 if (buf->lb_buf == NULL)
3464 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
3466 offset = sizeof(*lcm) +
3467 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
3468 for (i = 0; i < cur_entry_count; i++) {
3469 struct lov_comp_md_entry_v1 *cur_lcme;
3471 lcme = &lcm->lcm_entries[i];
3472 cur_lcme = &cur_lcm->lcm_entries[i];
3474 lcme->lcme_offset = cpu_to_le32(offset);
3475 memcpy((char *)lcm + offset,
3476 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
3477 le32_to_cpu(lcme->lcme_size));
3479 offset += le32_to_cpu(lcme->lcme_size);
3481 if (mirror_count == 1 &&
3482 mirror_id_of(le32_to_cpu(lcme->lcme_id)) == 0) {
3483 /* Add mirror from a non-flr file, create new mirror ID.
3484 * Otherwise, keep existing mirror's component ID, used
3485 * for mirror extension.
3487 id = pflr_id(1, i + 1);
3488 lcme->lcme_id = cpu_to_le32(id);
3491 id = max(le32_to_cpu(lcme->lcme_id), id);
3494 mirror_id = mirror_id_of(id) + 1;
3496 /* check if first entry in new layout is DOM */
3497 lmm = (struct lov_mds_md_v1 *)((char *)merge_lcm +
3498 merge_lcm->lcm_entries[0].lcme_offset);
3499 merge_has_dom = lov_pattern(le32_to_cpu(lmm->lmm_pattern)) &
3502 for (i = 0; i < merge_entry_count; i++) {
3503 struct lov_comp_md_entry_v1 *merge_lcme;
3505 merge_lcme = &merge_lcm->lcm_entries[i];
3506 lcme = &lcm->lcm_entries[cur_entry_count + i];
3508 *lcme = *merge_lcme;
3509 lcme->lcme_offset = cpu_to_le32(offset);
3510 if (merge_has_dom && i == 0)
3511 lcme->lcme_flags |= cpu_to_le32(LCME_FL_STALE);
3513 id = pflr_id(mirror_id, i + 1);
3514 lcme->lcme_id = cpu_to_le32(id);
3516 memcpy((char *)lcm + offset,
3517 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
3518 le32_to_cpu(lcme->lcme_size));
3520 offset += le32_to_cpu(lcme->lcme_size);
3523 /* fixup layout information */
3524 lcm->lcm_size = cpu_to_le32(size);
3525 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
3526 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
3527 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NONE)
3528 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
3530 rc = lod_striping_reload(env, lo, buf, 0);
3534 lod_obj_inc_layout_gen(lo);
3535 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3537 /* transfer layout version to OST objects. */
3538 if (lo->ldo_mirror_count > 1) {
3539 struct lod_obj_stripe_cb_data data = { {0} };
3541 layout_attr->la_valid = LA_LAYOUT_VERSION;
3542 layout_attr->la_layout_version = 0;
3543 data.locd_attr = layout_attr;
3544 data.locd_declare = true;
3545 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
3546 rc = lod_obj_for_each_stripe(env, lo, th, &data);
3551 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
3552 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3560 * Split layouts, just set the LOVEA with the layout from mbuf.
3562 static int lod_declare_layout_split(const struct lu_env *env,
3563 struct dt_object *dt, const struct lu_buf *mbuf,
3566 struct lod_thread_info *info = lod_env_info(env);
3567 struct lu_attr *layout_attr = &info->lti_layout_attr;
3568 struct lod_object *lo = lod_dt_obj(dt);
3569 struct lov_comp_md_v1 *lcm = mbuf->lb_buf;
3573 rc = lod_striping_reload(env, lo, mbuf, LVF_ALL_STALE);
3577 lod_obj_inc_layout_gen(lo);
3578 /* fix on-disk layout gen */
3579 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3581 /* transfer layout version to OST objects. */
3582 if (lo->ldo_mirror_count > 1) {
3583 struct lod_obj_stripe_cb_data data = { {0} };
3585 layout_attr->la_valid = LA_LAYOUT_VERSION;
3586 layout_attr->la_layout_version = 0;
3587 data.locd_attr = layout_attr;
3588 data.locd_declare = true;
3589 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
3590 rc = lod_obj_for_each_stripe(env, lo, th, &data);
3595 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), mbuf,
3596 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3600 static int lod_layout_declare_or_purge_mirror(const struct lu_env *env,
3601 struct dt_object *dt, const struct lu_buf *buf,
3602 struct thandle *th, bool declare)
3604 struct lod_thread_info *info = lod_env_info(env);
3605 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3606 struct lod_object *lo = lod_dt_obj(dt);
3607 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3608 struct lov_comp_md_entry_v1 *entry;
3609 struct lov_mds_md_v1 *lmm;
3610 struct dt_object **sub_objs = NULL;
3611 int rc = 0, i, k, array_count = 0;
3616 * other ops (like lod_declare_destroy) could destroying sub objects
3619 mutex_lock(&lo->ldo_layout_mutex);
3622 /* prepare sub-objects array */
3623 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
3624 entry = &comp_v1->lcm_entries[i];
3626 if (!(entry->lcme_flags & LCME_FL_INIT))
3629 lmm = (struct lov_mds_md_v1 *)
3630 ((char *)comp_v1 + entry->lcme_offset);
3631 array_count += lmm->lmm_stripe_count;
3633 OBD_ALLOC_PTR_ARRAY(sub_objs, array_count);
3634 if (sub_objs == NULL) {
3635 mutex_unlock(&lo->ldo_layout_mutex);
3640 k = 0; /* sub_objs index */
3641 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
3642 struct lov_ost_data_v1 *objs;
3643 struct lu_object *o, *n;
3644 struct dt_object *dto;
3645 struct lu_device *nd;
3646 struct lov_mds_md_v3 *v3;
3650 entry = &comp_v1->lcm_entries[i];
3652 if (!(entry->lcme_flags & LCME_FL_INIT))
3655 lmm = (struct lov_mds_md_v1 *)
3656 ((char *)comp_v1 + entry->lcme_offset);
3657 v3 = (struct lov_mds_md_v3 *)lmm;
3658 if (lmm->lmm_magic == LOV_MAGIC_V3)
3659 objs = &v3->lmm_objects[0];
3661 objs = &lmm->lmm_objects[0];
3663 for (j = 0; j < lmm->lmm_stripe_count; j++) {
3664 idx = objs[j].l_ost_idx;
3665 rc = ostid_to_fid(&info->lti_fid, &objs[j].l_ost_oi,
3670 if (!fid_is_sane(&info->lti_fid)) {
3671 CERROR("%s: sub-object insane fid "DFID"\n",
3672 lod2obd(d)->obd_name,
3673 PFID(&info->lti_fid));
3674 GOTO(out, rc = -EINVAL);
3677 lod_getref(&d->lod_ost_descs);
3679 rc = validate_lod_and_idx(d, idx);
3681 lod_putref(d, &d->lod_ost_descs);
3685 nd = &OST_TGT(d, idx)->ltd_tgt->dd_lu_dev;
3686 lod_putref(d, &d->lod_ost_descs);
3688 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
3690 GOTO(out, rc = PTR_ERR(o));
3692 n = lu_object_locate(o->lo_header, nd->ld_type);
3694 lu_object_put(env, n);
3695 GOTO(out, rc = -ENOENT);
3698 dto = container_of(n, struct dt_object, do_lu);
3701 rc = lod_sub_declare_destroy(env, dto, th);
3702 dt_object_put(env, dto);
3707 * collect to-be-destroyed sub objects, the
3708 * reference would be released after actual
3714 } /* for each stripe */
3715 } /* for each component in the mirror */
3720 /* destroy the sub objects */
3721 for (; i < k; i++) {
3722 rc = lod_sub_destroy(env, sub_objs[i], th);
3725 dt_object_put(env, sub_objs[i]);
3729 * if a sub object destroy failed, we'd release sub objects
3730 * reference get from above sub_objs collection.
3733 dt_object_put(env, sub_objs[i]);
3735 OBD_FREE_PTR_ARRAY(sub_objs, array_count);
3737 mutex_unlock(&lo->ldo_layout_mutex);
3743 * Purge layouts, delete sub objects in the mirror stored in the vic_buf,
3744 * and set the LOVEA with the layout from mbuf.
3746 static int lod_declare_layout_purge(const struct lu_env *env,
3747 struct dt_object *dt, const struct lu_buf *buf,
3750 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3751 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3756 if (le32_to_cpu(comp_v1->lcm_magic) != LOV_MAGIC_COMP_V1) {
3757 CERROR("%s: invalid layout magic %#x != %#x\n",
3758 lod2obd(d)->obd_name, le32_to_cpu(comp_v1->lcm_magic),
3763 if (cpu_to_le32(LOV_MAGIC_COMP_V1) != LOV_MAGIC_COMP_V1)
3764 lustre_swab_lov_comp_md_v1(comp_v1);
3766 /* from now on, @buf contains cpu endian data */
3768 if (comp_v1->lcm_mirror_count != 0) {
3769 CERROR("%s: can only purge one mirror from "DFID"\n",
3770 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3774 /* delcare sub objects deletion in the mirror stored in @buf */
3775 rc = lod_layout_declare_or_purge_mirror(env, dt, buf, th, true);
3779 /* delete sub objects from the mirror stored in @buf */
3780 static int lod_layout_purge(const struct lu_env *env, struct dt_object *dt,
3781 const struct lu_buf *buf, struct thandle *th)
3786 rc = lod_layout_declare_or_purge_mirror(env, dt, buf, th, false);
3791 * Implementation of dt_object_operations::do_declare_xattr_set.
3793 * \see dt_object_operations::do_declare_xattr_set() in the API description
3796 * the extension to the API:
3797 * - declaring LOVEA requests striping creation
3798 * - LU_XATTR_REPLACE means layout swap
3800 static int lod_declare_xattr_set(const struct lu_env *env,
3801 struct dt_object *dt,
3802 const struct lu_buf *buf,
3803 const char *name, int fl,
3806 struct lod_thread_info *info = lod_env_info(env);
3807 struct dt_object *next = dt_object_child(dt);
3808 struct lu_attr *attr = &info->lti_attr;
3809 struct lod_object *lo = lod_dt_obj(dt);
3814 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
3815 if ((S_ISREG(mode) || mode == 0) &&
3816 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE | LU_XATTR_SPLIT |
3818 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3819 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
3821 * this is a request to create object's striping.
3823 * allow to declare predefined striping on a new (!mode) object
3824 * which is supposed to be replay of regular file creation
3825 * (when LOV setting is declared)
3827 * LU_XATTR_REPLACE is set to indicate a layout swap
3829 if (dt_object_exists(dt)) {
3830 rc = dt_attr_get(env, next, attr);
3834 memset(attr, 0, sizeof(*attr));
3835 attr->la_valid = LA_TYPE | LA_MODE;
3836 attr->la_mode = S_IFREG;
3838 rc = lod_declare_striped_create(env, dt, attr, buf, th);
3839 } else if (fl & LU_XATTR_MERGE) {
3840 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3841 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3842 rc = lod_declare_layout_merge(env, dt, buf, th);
3843 } else if (fl & LU_XATTR_SPLIT) {
3844 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3845 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3846 rc = lod_declare_layout_split(env, dt, buf, th);
3847 } else if (fl & LU_XATTR_PURGE) {
3848 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3849 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3850 rc = lod_declare_layout_purge(env, dt, buf, th);
3851 } else if (S_ISREG(mode) &&
3852 strlen(name) >= sizeof(XATTR_LUSTRE_LOV) + 3 &&
3853 allowed_lustre_lov(name)) {
3855 * this is a request to modify object's striping.
3856 * add/set/del component(s).
3858 if (!dt_object_exists(dt))
3861 rc = lod_declare_modify_layout(env, dt, name, buf, th);
3862 } else if (S_ISDIR(mode)) {
3863 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
3864 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3865 rc = lod_replace_parent_fid(env, dt, buf, th, true);
3867 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
3871 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3872 strcmp(name, XATTR_LUSTRE_LOV) == 0 || allowed_lustre_lov(name)))
3873 rc = lod_save_layout_gen_intrans(info, lo);
3879 * Apply xattr changes to the object.
3881 * Applies xattr changes to the object and the stripes if the latter exist.
3883 * \param[in] env execution environment
3884 * \param[in] dt object
3885 * \param[in] buf buffer pointing to the new value of xattr
3886 * \param[in] name name of xattr
3887 * \param[in] fl flags
3888 * \param[in] th transaction handle
3890 * \retval 0 on success
3891 * \retval negative if failed
3893 static int lod_xattr_set_internal(const struct lu_env *env,
3894 struct dt_object *dt,
3895 const struct lu_buf *buf,
3896 const char *name, int fl,
3899 struct dt_object *next = dt_object_child(dt);
3900 struct lod_object *lo = lod_dt_obj(dt);
3905 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3906 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3909 /* Note: Do not set LinkEA on sub-stripes, otherwise
3910 * it will confuse the fid2path process(see mdt_path_current()).
3911 * The linkEA between master and sub-stripes is set in
3912 * lod_xattr_set_lmv(). */
3913 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
3916 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3917 if (!lo->ldo_stripe[i])
3920 if (!dt_object_exists(lo->ldo_stripe[i]))
3923 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
3933 * Delete an extended attribute.
3935 * Deletes specified xattr from the object and the stripes if the latter exist.
3937 * \param[in] env execution environment
3938 * \param[in] dt object
3939 * \param[in] name name of xattr
3940 * \param[in] th transaction handle
3942 * \retval 0 on success
3943 * \retval negative if failed
3945 static int lod_xattr_del_internal(const struct lu_env *env,
3946 struct dt_object *dt,
3947 const char *name, struct thandle *th)
3949 struct dt_object *next = dt_object_child(dt);
3950 struct lod_object *lo = lod_dt_obj(dt);
3956 rc = lod_sub_xattr_del(env, next, name, th);
3957 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3960 if (lo->ldo_dir_stripe_count == 0)
3963 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3964 if (!lo->ldo_stripe[i])
3967 if (!dt_object_exists(lo->ldo_stripe[i]))
3970 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3979 * Set default striping on a directory.
3981 * Sets specified striping on a directory object unless it matches the default
3982 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3983 * EA. This striping will be used when regular file is being created in this
3986 * \param[in] env execution environment
3987 * \param[in] dt the striped object
3988 * \param[in] buf buffer with the striping
3989 * \param[in] name name of EA
3990 * \param[in] fl xattr flag (see OSD API description)
3991 * \param[in] th transaction handle
3993 * \retval 0 on success
3994 * \retval negative if failed
3996 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
3997 struct dt_object *dt,
3998 const struct lu_buf *buf,
3999 const char *name, int fl,
4002 struct lov_user_md_v1 *lum;
4003 struct lov_user_md_v3 *v3 = NULL;
4004 const char *pool_name = NULL;
4009 LASSERT(buf != NULL && buf->lb_buf != NULL);
4012 switch (lum->lmm_magic) {
4013 case LOV_USER_MAGIC_SPECIFIC:
4014 case LOV_USER_MAGIC_V3:
4016 if (lov_pool_is_reserved(v3->lmm_pool_name))
4017 memset(v3->lmm_pool_name, 0, sizeof(v3->lmm_pool_name));
4018 else if (v3->lmm_pool_name[0] != '\0')
4019 pool_name = v3->lmm_pool_name;
4021 case LOV_USER_MAGIC_V1:
4022 /* if { size, offset, count } = { 0, -1, 0 } and no pool
4023 * (i.e. all default values specified) then delete default
4024 * striping from dir. */
4026 "set default striping: sz %u # %u offset %d %s %s\n",
4027 (unsigned)lum->lmm_stripe_size,
4028 (unsigned)lum->lmm_stripe_count,
4029 (int)lum->lmm_stripe_offset,
4030 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
4032 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
4033 lum->lmm_stripe_count,
4034 lum->lmm_stripe_offset,
4037 case LOV_USER_MAGIC_COMP_V1:
4039 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lum;
4040 struct lov_comp_md_entry_v1 *lcme;
4043 comp_cnt = le16_to_cpu(lcm->lcm_entry_count);
4044 for (i = 0; i < comp_cnt; i++) {
4045 lcme = &lcm->lcm_entries[i];
4046 if (lcme->lcme_flags & cpu_to_le32(LCME_FL_EXTENSION)) {
4047 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
4056 CERROR("Invalid magic %x\n", lum->lmm_magic);
4061 rc = lod_xattr_del_internal(env, dt, name, th);
4065 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4071 static int lod_get_default_lov_striping(const struct lu_env *env,
4072 struct lod_object *lo,
4073 struct lod_default_striping *lds,
4074 struct dt_allocation_hint *ah);
4077 * Helper function to convert compound layout to compound layout with
4080 * Copy lcm_entries array of \a src to \a tgt. Replace lov_user_md_v1
4081 * components of \a src with lov_user_md_v3 using \a pool.
4083 * \param[in] src source layout
4084 * \param[in] pool pool to use in \a tgt
4085 * \param[out] tgt target layout
4087 static void embed_pool_to_comp_v1(const struct lov_comp_md_v1 *src,
4089 struct lov_comp_md_v1 *tgt)
4092 struct lov_user_md_v1 *lum;
4093 struct lov_user_md_v3 *lum3;
4094 struct lov_comp_md_entry_v1 *entry;
4098 entry = tgt->lcm_entries;
4100 for (i = 0; i < le16_to_cpu(src->lcm_entry_count); i++, entry++) {
4101 *entry = src->lcm_entries[i];
4102 offset = le32_to_cpu(src->lcm_entries[i].lcme_offset);
4103 entry->lcme_offset = cpu_to_le32(offset + shift);
4105 lum = (struct lov_user_md_v1 *)((char *)src + offset);
4106 lum3 = (struct lov_user_md_v3 *)((char *)tgt + offset + shift);
4107 *(struct lov_user_md_v1 *)lum3 = *lum;
4108 if (lum->lmm_pattern & cpu_to_le32(LOV_PATTERN_MDT)) {
4109 lum3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
4111 lum3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4112 entry->lcme_size = cpu_to_le32(sizeof(*lum3));
4113 strscpy(lum3->lmm_pool_name, pool,
4114 sizeof(lum3->lmm_pool_name));
4115 shift += sizeof(*lum3) - sizeof(*lum);
4121 * Set default striping on a directory.
4123 * Sets specified striping on a directory object unless it matches the default
4124 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
4125 * EA. This striping will be used when regular file is being created in this
4127 * If current default striping includes a pool but specifed striping
4128 * does not - retain the pool if it exists.
4130 * \param[in] env execution environment
4131 * \param[in] dt the striped object
4132 * \param[in] buf buffer with the striping
4133 * \param[in] name name of EA
4134 * \param[in] fl xattr flag (see OSD API description)
4135 * \param[in] th transaction handle
4137 * \retval 0 on success
4138 * \retval negative if failed
4140 static int lod_xattr_set_default_lov_on_dir(const struct lu_env *env,
4141 struct dt_object *dt,
4142 const struct lu_buf *buf,
4143 const char *name, int fl,
4146 struct lod_default_striping *lds = lod_lds_buf_get(env);
4147 struct lov_user_md_v1 *v1 = buf->lb_buf;
4148 char pool[LOV_MAXPOOLNAME + 1];
4154 /* get existing striping config */
4155 rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds, NULL);
4159 memset(pool, 0, sizeof(pool));
4160 if (lds->lds_def_striping_set == 1)
4161 lod_layout_get_pool(lds->lds_def_comp_entries,
4162 lds->lds_def_comp_cnt, pool,
4165 is_del = LOVEA_DELETE_VALUES(v1->lmm_stripe_size,
4166 v1->lmm_stripe_count,
4167 v1->lmm_stripe_offset,
4170 /* Retain the pool name if it is not given */
4171 if (v1->lmm_magic == LOV_USER_MAGIC_V1 && pool[0] != '\0' &&
4173 struct lod_thread_info *info = lod_env_info(env);
4174 struct lov_user_md_v3 *v3 = info->lti_ea_store;
4176 memset(v3, 0, sizeof(*v3));
4177 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4178 v3->lmm_pattern = cpu_to_le32(v1->lmm_pattern);
4179 v3->lmm_stripe_count = cpu_to_le32(v1->lmm_stripe_count);
4180 v3->lmm_stripe_offset = cpu_to_le32(v1->lmm_stripe_offset);
4181 v3->lmm_stripe_size = cpu_to_le32(v1->lmm_stripe_size);
4183 strscpy(v3->lmm_pool_name, pool, sizeof(v3->lmm_pool_name));
4185 info->lti_buf.lb_buf = v3;
4186 info->lti_buf.lb_len = sizeof(*v3);
4187 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4189 } else if (v1->lmm_magic == LOV_USER_MAGIC_COMP_V1 &&
4190 pool[0] != '\0' && !is_del) {
4192 * try to retain the pool from default layout if the
4193 * specified component layout does not provide pool
4196 struct lod_thread_info *info = lod_env_info(env);
4197 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
4198 struct lov_comp_md_v1 *comp_v1p;
4199 struct lov_user_md_v1 *lum;
4203 struct lov_comp_md_entry_v1 *entry;
4206 entry_count = le16_to_cpu(comp_v1->lcm_entry_count);
4207 size = sizeof(*comp_v1) +
4208 entry_count * sizeof(comp_v1->lcm_entries[0]);
4209 entry = comp_v1->lcm_entries;
4210 for (i = 0; i < entry_count; i++, entry++) {
4211 offset = le32_to_cpu(entry->lcme_offset);
4212 lum = (struct lov_user_md_v1 *)((char *)comp_v1 +
4214 if (le32_to_cpu(lum->lmm_magic) != LOV_USER_MAGIC_V1)
4215 /* the i-th component includes pool info */
4217 if (lum->lmm_pattern & cpu_to_le32(LOV_PATTERN_MDT))
4218 size += sizeof(struct lov_user_md_v1);
4220 size += sizeof(struct lov_user_md_v3);
4223 if (i == entry_count) {
4225 * re-compose the layout to include the pool for
4228 if (info->lti_ea_store_size < size)
4229 rc = lod_ea_store_resize(info, size);
4232 comp_v1p = info->lti_ea_store;
4233 *comp_v1p = *comp_v1;
4234 comp_v1p->lcm_size = cpu_to_le32(size);
4235 embed_pool_to_comp_v1(comp_v1, pool, comp_v1p);
4237 info->lti_buf.lb_buf = comp_v1p;
4238 info->lti_buf.lb_len = size;
4239 rc = lod_xattr_set_lov_on_dir(env, dt,
4244 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl,
4248 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl, th);
4251 if (lds->lds_def_striping_set == 1 && lds->lds_def_comp_entries != NULL)
4252 lod_free_def_comp_entries(lds);
4258 * Set default striping on a directory object.
4260 * Sets specified striping on a directory object unless it matches the default
4261 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
4262 * EA. This striping will be used when a new directory is being created in the
4265 * \param[in] env execution environment
4266 * \param[in] dt the striped object
4267 * \param[in] buf buffer with the striping
4268 * \param[in] name name of EA
4269 * \param[in] fl xattr flag (see OSD API description)
4270 * \param[in] th transaction handle
4272 * \retval 0 on success
4273 * \retval negative if failed
4275 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
4276 struct dt_object *dt,
4277 const struct lu_buf *buf,
4278 const char *name, int fl,
4281 struct lmv_user_md_v1 *lum;
4286 LASSERT(buf != NULL && buf->lb_buf != NULL);
4290 "set default stripe_count # %u stripe_offset %d hash %u\n",
4291 le32_to_cpu(lum->lum_stripe_count),
4292 (int)le32_to_cpu(lum->lum_stripe_offset),
4293 le32_to_cpu(lum->lum_hash_type));
4295 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
4296 le32_to_cpu(lum->lum_stripe_offset)) &&
4297 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
4298 rc = lod_xattr_del_internal(env, dt, name, th);
4302 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4311 * Turn directory into a striped directory.
4313 * During replay the client sends the striping created before MDT
4314 * failure, then the layer above LOD sends this defined striping
4315 * using ->do_xattr_set(), so LOD uses this method to replay creation
4316 * of the stripes. Notice the original information for the striping
4317 * (#stripes, FIDs, etc) was transferred in declare path.
4319 * \param[in] env execution environment
4320 * \param[in] dt the striped object
4321 * \param[in] buf buf lmv_user_md for create, or lmv_mds_md for replay
4322 * \param[in] name not used currently
4323 * \param[in] fl xattr flag (see OSD API description)
4324 * \param[in] th transaction handle
4326 * \retval 0 on success
4327 * \retval negative if failed
4329 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
4330 const struct lu_buf *buf, const char *name,
4331 int fl, struct thandle *th)
4333 struct lod_object *lo = lod_dt_obj(dt);
4334 struct lod_thread_info *info = lod_env_info(env);
4335 struct lu_attr *attr = &info->lti_attr;
4336 struct dt_object_format *dof = &info->lti_format;
4337 struct lu_buf lmv_buf;
4338 struct lu_buf slave_lmv_buf;
4339 struct lmv_user_md *lum = buf->lb_buf;
4340 struct lmv_mds_md_v1 *lmm;
4341 struct lmv_mds_md_v1 *slave_lmm = NULL;
4342 struct dt_insert_rec *rec = &info->lti_dt_rec;
4347 /* lum is used to know whether it's replay */
4349 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4352 /* The stripes are supposed to be allocated in declare phase,
4353 * if there are no stripes being allocated, it will skip */
4354 if (lo->ldo_dir_stripe_count == 0) {
4355 if (lo->ldo_is_foreign) {
4356 rc = lod_sub_xattr_set(env, dt_object_child(dt), buf,
4357 XATTR_NAME_LMV, fl, th);
4364 rc = dt_attr_get(env, dt_object_child(dt), attr);
4368 attr->la_valid &= LA_ATIME | LA_MTIME | LA_CTIME | LA_FLAGS |
4369 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
4370 dof->dof_type = DFT_DIR;
4372 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
4375 lmm = lmv_buf.lb_buf;
4377 OBD_ALLOC_PTR(slave_lmm);
4378 if (slave_lmm == NULL)
4381 lod_prep_slave_lmv_md(slave_lmm, lmm);
4382 slave_lmv_buf.lb_buf = slave_lmm;
4383 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
4385 rec->rec_type = S_IFDIR;
4386 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4387 struct dt_object *dto = lo->ldo_stripe[i];
4388 char *stripe_name = info->lti_key;
4389 struct lu_name *sname;
4390 struct linkea_data ldata = { NULL };
4391 struct lu_buf linkea_buf;
4392 bool stripe_created = false;
4394 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
4398 /* fail a remote stripe creation */
4399 if (i && CFS_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_CREATE))
4402 /* if it's replay by client request, and stripe exists on remote
4403 * MDT, it means mkdir was partially executed: stripe was
4404 * created on remote MDT successfully, but target not in last
4407 if (unlikely((le32_to_cpu(lum->lum_magic) == LMV_MAGIC_V1) &&
4408 dt_object_exists(dto) && dt_object_remote(dto)))
4409 stripe_created = true;
4411 /* don't create stripe if:
4412 * 1. it's source stripe of migrating directory
4413 * 2. it's existed stripe of splitting directory
4415 if ((lod_is_migrating(lo) && i >= lo->ldo_dir_migrate_offset) ||
4416 (lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) {
4417 if (!dt_object_exists(dto))
4418 GOTO(out, rc = -EINVAL);
4419 } else if (!stripe_created) {
4420 dt_write_lock(env, dto, DT_TGT_CHILD);
4421 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
4423 dt_write_unlock(env, dto);
4427 rc = lod_sub_ref_add(env, dto, th);
4428 dt_write_unlock(env, dto);
4432 rec->rec_fid = lu_object_fid(&dto->do_lu);
4433 rc = lod_sub_insert(env, dto,
4434 (const struct dt_rec *)rec,
4435 (const struct dt_key *)dot, th);
4440 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
4441 cfs_fail_val != i) {
4442 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
4444 slave_lmm->lmv_master_mdt_index =
4447 slave_lmm->lmv_master_mdt_index =
4450 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
4451 XATTR_NAME_LMV, 0, th);
4456 /* don't insert stripe if it's existed stripe of splitting
4457 * directory (this directory is striped).
4458 * NB, plain directory will insert itself as the first
4461 if (lod_is_splitting(lo) && lo->ldo_dir_split_offset > 1 &&
4462 lo->ldo_dir_split_offset > i)
4465 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
4467 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4468 PFID(lu_object_fid(&dto->do_lu)), i + 1);
4470 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4471 PFID(lu_object_fid(&dto->do_lu)), i);
4473 if (!stripe_created) {
4474 rec->rec_fid = lu_object_fid(&dt->do_lu);
4475 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
4476 (const struct dt_key *)dotdot, th);
4480 sname = lod_name_get(env, stripe_name,
4481 strlen(stripe_name));
4482 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
4483 sname, lu_object_fid(&dt->do_lu));
4487 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
4488 linkea_buf.lb_len = ldata.ld_leh->leh_len;
4489 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
4490 XATTR_NAME_LINK, 0, th);
4495 rec->rec_fid = lu_object_fid(&dto->do_lu);
4496 rc = lod_sub_insert(env, dt_object_child(dt),
4497 (const struct dt_rec *)rec,
4498 (const struct dt_key *)stripe_name, th);
4502 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
4507 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
4508 rc = lod_sub_xattr_set(env, dt_object_child(dt),
4509 &lmv_buf, XATTR_NAME_LMV, fl, th);
4511 if (slave_lmm != NULL)
4512 OBD_FREE_PTR(slave_lmm);
4518 * Helper function to declare/execute creation of a striped directory
4520 * Called in declare/create object path, prepare striping for a directory
4521 * and prepare defaults data striping for the objects to be created in
4522 * that directory. Notice the function calls "declaration" or "execution"
4523 * methods depending on \a declare param. This is a consequence of the
4524 * current approach while we don't have natural distributed transactions:
4525 * we basically execute non-local updates in the declare phase. So, the
4526 * arguments for the both phases are the same and this is the reason for
4527 * this function to exist.
4529 * \param[in] env execution environment
4530 * \param[in] dt object
4531 * \param[in] attr attributes the stripes will be created with
4532 * \param[in] lmu lmv_user_md if MDT indices are specified
4533 * \param[in] dof format of stripes (see OSD API description)
4534 * \param[in] th transaction handle
4535 * \param[in] declare where to call "declare" or "execute" methods
4537 * \retval 0 on success
4538 * \retval negative if failed
4540 static int lod_dir_striping_create_internal(const struct lu_env *env,
4541 struct dt_object *dt,
4542 struct lu_attr *attr,
4543 const struct lu_buf *lmu,
4544 struct dt_object_format *dof,
4548 struct lod_thread_info *info = lod_env_info(env);
4549 struct lod_object *lo = lod_dt_obj(dt);
4550 const struct lod_default_striping *lds = lo->ldo_def_striping;
4554 LASSERT(ergo(lds != NULL,
4555 lds->lds_def_striping_set ||
4556 lds->lds_dir_def_striping_set));
4559 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
4560 lo->ldo_dir_stripe_offset)) {
4562 /* mkdir by default LMV */
4563 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
4564 int stripe_count = lo->ldo_dir_stripe_count;
4566 if (info->lti_ea_store_size < sizeof(*v1)) {
4567 rc = lod_ea_store_resize(info, sizeof(*v1));
4570 v1 = info->lti_ea_store;
4573 memset(v1, 0, sizeof(*v1));
4574 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
4575 v1->lum_stripe_count = cpu_to_le32(stripe_count);
4576 v1->lum_stripe_offset =
4577 cpu_to_le32(lo->ldo_dir_stripe_offset);
4579 info->lti_buf.lb_buf = v1;
4580 info->lti_buf.lb_len = sizeof(*v1);
4581 lmu = &info->lti_buf;
4585 rc = lod_declare_xattr_set_lmv(env, dt, attr, lmu, dof,
4588 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
4592 } else if (lmu->lb_buf) {
4593 /* foreign LMV EA case */
4595 struct lmv_foreign_md *lfm = lmu->lb_buf;
4597 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN)
4598 rc = lod_declare_xattr_set_lmv(env, dt, attr,
4600 } else if (lo->ldo_is_foreign) {
4601 LASSERT(lo->ldo_foreign_lmv != NULL &&
4602 lo->ldo_foreign_lmv_size > 0);
4603 info->lti_buf.lb_buf = lo->ldo_foreign_lmv;
4604 info->lti_buf.lb_len = lo->ldo_foreign_lmv_size;
4605 lmu = &info->lti_buf;
4606 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
4611 /* Transfer default LMV striping from the parent */
4612 if (lds != NULL && lds->lds_dir_def_striping_set &&
4613 lds->lds_dir_def_max_inherit != LMV_INHERIT_END &&
4614 lds->lds_dir_def_max_inherit != LMV_INHERIT_NONE &&
4615 !(LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
4616 lds->lds_dir_def_stripe_offset) &&
4617 le32_to_cpu(lds->lds_dir_def_hash_type) !=
4618 LMV_HASH_TYPE_UNKNOWN)) {
4619 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
4621 if (info->lti_ea_store_size < sizeof(*v1)) {
4622 rc = lod_ea_store_resize(info, sizeof(*v1));
4625 v1 = info->lti_ea_store;
4628 memset(v1, 0, sizeof(*v1));
4629 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
4630 v1->lum_stripe_count =
4631 cpu_to_le32(lds->lds_dir_def_stripe_count);
4632 v1->lum_stripe_offset =
4633 cpu_to_le32(lds->lds_dir_def_stripe_offset);
4635 cpu_to_le32(lds->lds_dir_def_hash_type);
4636 v1->lum_max_inherit =
4637 lmv_inherit_next(lds->lds_dir_def_max_inherit);
4638 v1->lum_max_inherit_rr =
4639 lmv_inherit_rr_next(lds->lds_dir_def_max_inherit_rr);
4641 info->lti_buf.lb_buf = v1;
4642 info->lti_buf.lb_len = sizeof(*v1);
4644 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4645 XATTR_NAME_DEFAULT_LMV,
4648 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
4650 XATTR_NAME_DEFAULT_LMV, 0,
4656 /* Transfer default LOV striping from the parent */
4657 if (lds != NULL && lds->lds_def_striping_set &&
4658 lds->lds_def_comp_cnt != 0) {
4659 struct lov_mds_md *lmm;
4660 int lmm_size = lod_comp_md_size(lo, true);
4662 if (info->lti_ea_store_size < lmm_size) {
4663 rc = lod_ea_store_resize(info, lmm_size);
4667 lmm = info->lti_ea_store;
4669 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
4673 info->lti_buf.lb_buf = lmm;
4674 info->lti_buf.lb_len = lmm_size;
4677 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4678 XATTR_NAME_LOV, 0, th);
4680 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4681 XATTR_NAME_LOV, 0, th);
4686 /* ldo_def_striping is not allocated, clear after use, in case directory
4687 * layout is changed later.
4690 lo->ldo_def_striping = NULL;
4695 static int lod_declare_dir_striping_create(const struct lu_env *env,
4696 struct dt_object *dt,
4697 struct lu_attr *attr,
4699 struct dt_object_format *dof,
4702 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
4706 static int lod_dir_striping_create(const struct lu_env *env,
4707 struct dt_object *dt,
4708 struct lu_attr *attr,
4709 const struct lu_buf *lmu,
4710 struct dt_object_format *dof,
4713 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
4718 * Make LOV EA for striped object.
4720 * Generate striping information and store it in the LOV EA of the given
4721 * object. The caller must ensure nobody else is calling the function
4722 * against the object concurrently. The transaction must be started.
4723 * FLDB service must be running as well; it's used to map FID to the target,
4724 * which is stored in LOV EA.
4726 * \param[in] env execution environment for this thread
4727 * \param[in] lo LOD object
4728 * \param[in] th transaction handle
4730 * \retval 0 if LOV EA is stored successfully
4731 * \retval negative error number on failure
4733 static int lod_generate_and_set_lovea(const struct lu_env *env,
4734 struct lod_object *lo,
4737 struct lod_thread_info *info = lod_env_info(env);
4738 struct dt_object *next = dt_object_child(&lo->ldo_obj);
4739 struct lov_mds_md_v1 *lmm;
4745 if (lo->ldo_comp_cnt == 0 && !lo->ldo_is_foreign) {
4746 lod_striping_free_nolock(env, lo);
4747 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
4751 lmm_size = lod_comp_md_size(lo, false);
4752 if (info->lti_ea_store_size < lmm_size) {
4753 rc = lod_ea_store_resize(info, lmm_size);
4757 lmm = info->lti_ea_store;
4759 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
4763 info->lti_buf.lb_buf = lmm;
4764 info->lti_buf.lb_len = lmm_size;
4765 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
4766 XATTR_NAME_LOV, 0, th);
4770 static __u32 lod_gen_component_id(struct lod_object *lo,
4771 int mirror_id, int comp_idx);
4774 * Repeat an existing component
4776 * Creates a new layout by replicating an existing component. Uses striping
4777 * policy from previous component as a template for the striping for the new
4780 * New component starts with zero length, will be extended (or removed) before
4781 * returning layout to client.
4783 * NB: Reallocates layout components array (lo->ldo_comp_entries), invalidating
4784 * any pre-existing pointers to components. Handle with care.
4786 * \param[in] env execution environment for this thread
4787 * \param[in,out] lo object to update the layout of
4788 * \param[in] index index of component to copy
4790 * \retval 0 on success
4791 * \retval negative errno on error
4793 static int lod_layout_repeat_comp(const struct lu_env *env,
4794 struct lod_object *lo, int index)
4796 struct lod_layout_component *lod_comp;
4797 struct lod_layout_component *new_comp = NULL;
4798 struct lod_layout_component *comp_array;
4799 int rc = 0, i, new_cnt = lo->ldo_comp_cnt + 1;
4804 lod_comp = &lo->ldo_comp_entries[index];
4805 LASSERT(lod_comp_inited(lod_comp) && lod_comp->llc_id != LCME_ID_INVAL);
4807 CDEBUG(D_LAYOUT, "repeating component %d\n", index);
4809 OBD_ALLOC_PTR_ARRAY(comp_array, new_cnt);
4810 if (comp_array == NULL)
4811 GOTO(out, rc = -ENOMEM);
4813 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4814 memcpy(&comp_array[i + offset], &lo->ldo_comp_entries[i],
4815 sizeof(*comp_array));
4817 /* Duplicate this component in to the next slot */
4819 new_comp = &comp_array[i + 1];
4820 memcpy(&comp_array[i + 1], &lo->ldo_comp_entries[i],
4821 sizeof(*comp_array));
4822 /* We must now skip this new component when copying */
4827 /* Set up copied component */
4828 new_comp->llc_flags &= ~LCME_FL_INIT;
4829 new_comp->llc_stripe = NULL;
4830 new_comp->llc_stripes_allocated = 0;
4831 new_comp->llc_ost_indices = NULL;
4832 new_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4833 /* for uninstantiated components, layout gen stores default stripe
4835 new_comp->llc_layout_gen = lod_comp->llc_stripe_offset;
4836 /* This makes the repeated component zero-length, placed at the end of
4837 * the preceding component */
4838 new_comp->llc_extent.e_start = new_comp->llc_extent.e_end;
4839 new_comp->llc_timestamp = lod_comp->llc_timestamp;
4840 new_comp->llc_pool = NULL;
4842 rc = lod_set_pool(&new_comp->llc_pool, lod_comp->llc_pool);
4846 if (new_comp->llc_ostlist.op_array) {
4847 __u32 *op_array = NULL;
4849 OBD_ALLOC(op_array, new_comp->llc_ostlist.op_size);
4851 GOTO(out, rc = -ENOMEM);
4852 memcpy(op_array, &new_comp->llc_ostlist.op_array,
4853 new_comp->llc_ostlist.op_size);
4854 new_comp->llc_ostlist.op_array = op_array;
4857 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
4858 lo->ldo_comp_entries = comp_array;
4859 lo->ldo_comp_cnt = new_cnt;
4861 /* Generate an id for the new component */
4862 mirror_id = mirror_id_of(new_comp->llc_id);
4863 new_comp->llc_id = LCME_ID_INVAL;
4864 new_comp->llc_id = lod_gen_component_id(lo, mirror_id, index + 1);
4865 if (new_comp->llc_id == LCME_ID_INVAL)
4866 GOTO(out, rc = -ERANGE);
4871 OBD_FREE_PTR_ARRAY(comp_array, new_cnt);
4876 static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
4880 /* clear memory region that will be used for layout change */
4881 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
4882 info->lti_count = 0;
4884 if (info->lti_comp_size >= comp_cnt)
4887 if (info->lti_comp_size > 0) {
4888 OBD_FREE_PTR_ARRAY(info->lti_comp_idx, info->lti_comp_size);
4889 info->lti_comp_size = 0;
4892 OBD_ALLOC_PTR_ARRAY(info->lti_comp_idx, comp_cnt);
4893 if (!info->lti_comp_idx)
4896 info->lti_comp_size = comp_cnt;
4901 * Prepare new layout minus deleted components
4903 * Removes components marked for deletion (LCME_ID_INVAL) by copying to a new
4904 * layout and skipping those components. Removes stripe objects if any exist.
4907 * Reallocates layout components array (lo->ldo_comp_entries), invalidating
4908 * any pre-existing pointers to components.
4910 * Caller is responsible for updating mirror end (ldo_mirror[].lme_end).
4912 * \param[in] env execution environment for this thread
4913 * \param[in,out] lo object to update the layout of
4914 * \param[in] th transaction handle for this operation
4916 * \retval # of components deleted
4917 * \retval negative errno on error
4919 static int lod_layout_del_prep_layout(const struct lu_env *env,
4920 struct lod_object *lo,
4923 struct lod_layout_component *lod_comp;
4924 struct lod_thread_info *info = lod_env_info(env);
4925 int rc = 0, i, j, deleted = 0;
4929 LASSERT(lo->ldo_is_composite);
4930 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
4932 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
4936 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4937 lod_comp = &lo->ldo_comp_entries[i];
4939 if (lod_comp->llc_id != LCME_ID_INVAL) {
4940 /* Build array of things to keep */
4941 info->lti_comp_idx[info->lti_count++] = i;
4945 if (lod_comp->llc_magic == LOV_MAGIC_FOREIGN)
4948 lod_obj_set_pool(lo, i, NULL);
4949 if (lod_comp->llc_ostlist.op_array) {
4950 OBD_FREE(lod_comp->llc_ostlist.op_array,
4951 lod_comp->llc_ostlist.op_size);
4952 lod_comp->llc_ostlist.op_array = NULL;
4953 lod_comp->llc_ostlist.op_size = 0;
4957 CDEBUG(D_LAYOUT, "deleting comp %d, left %d\n", i,
4958 lo->ldo_comp_cnt - deleted);
4960 /* No striping info for this component */
4961 if (lod_comp->llc_stripe == NULL)
4964 LASSERT(lod_comp->llc_stripe_count > 0);
4965 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4966 struct dt_object *obj = lod_comp->llc_stripe[j];
4971 /* components which are not init have no sub objects
4973 if (lod_comp_inited(lod_comp)) {
4974 rc = lod_sub_destroy(env, obj, th);
4979 lu_object_put(env, &obj->do_lu);
4980 lod_comp->llc_stripe[j] = NULL;
4982 OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe,
4983 lod_comp->llc_stripes_allocated);
4984 lod_comp->llc_stripe = NULL;
4985 OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices,
4986 lod_comp->llc_stripes_allocated);
4987 lod_comp->llc_ost_indices = NULL;
4988 lod_comp->llc_stripes_allocated = 0;
4991 /* info->lti_count has the amount of left components */
4992 LASSERTF(info->lti_count >= 0 && info->lti_count < lo->ldo_comp_cnt,
4993 "left = %d, lo->ldo_comp_cnt %d\n", (int)info->lti_count,
4994 (int)lo->ldo_comp_cnt);
4996 if (info->lti_count > 0) {
4997 struct lod_layout_component *comp_array;
4999 OBD_ALLOC_PTR_ARRAY(comp_array, info->lti_count);
5000 if (comp_array == NULL)
5001 GOTO(out, rc = -ENOMEM);
5003 for (i = 0; i < info->lti_count; i++) {
5004 memcpy(&comp_array[i],
5005 &lo->ldo_comp_entries[info->lti_comp_idx[i]],
5006 sizeof(*comp_array));
5009 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
5010 lo->ldo_comp_entries = comp_array;
5011 lo->ldo_comp_cnt = info->lti_count;
5013 lod_free_comp_entries(lo);
5018 return rc ? rc : deleted;
5022 * Delete layout component(s)
5024 * This function sets up the layout data in the env and does the setattrs
5025 * required to write out the new layout. The layout itself is modified in
5026 * lod_layout_del_prep_layout.
5028 * \param[in] env execution environment for this thread
5029 * \param[in] dt object
5030 * \param[in] th transaction handle
5032 * \retval 0 on success
5033 * \retval negative error number on failure
5035 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
5038 struct lod_object *lo = lod_dt_obj(dt);
5039 struct dt_object *next = dt_object_child(dt);
5040 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
5043 LASSERT(lo->ldo_mirror_count == 1);
5045 mutex_lock(&lo->ldo_layout_mutex);
5047 rc = lod_layout_del_prep_layout(env, lo, th);
5051 /* Only do this if we didn't delete all components */
5052 if (lo->ldo_comp_cnt > 0) {
5053 lo->ldo_mirrors[0].lme_end = lo->ldo_comp_cnt - 1;
5054 lod_obj_inc_layout_gen(lo);
5057 LASSERT(dt_object_exists(dt));
5058 rc = dt_attr_get(env, next, attr);
5062 if (attr->la_size > 0) {
5064 attr->la_valid = LA_SIZE;
5065 rc = lod_sub_attr_set(env, next, attr, th);
5070 rc = lod_generate_and_set_lovea(env, lo, th);
5074 lod_striping_free_nolock(env, lo);
5076 mutex_unlock(&lo->ldo_layout_mutex);
5083 * Implementation of dt_object_operations::do_xattr_set.
5085 * Sets specified extended attribute on the object. Three types of EAs are
5087 * LOV EA - stores striping for a regular file or default striping (when set
5089 * LMV EA - stores a marker for the striped directories
5090 * DMV EA - stores default directory striping
5092 * When striping is applied to a non-striped existing object (this is called
5093 * late striping), then LOD notices the caller wants to turn the object into a
5094 * striped one. The stripe objects are created and appropriate EA is set:
5095 * LOV EA storing all the stripes directly or LMV EA storing just a small header
5096 * with striping configuration.
5098 * \see dt_object_operations::do_xattr_set() in the API description for details.
5100 static int lod_xattr_set(const struct lu_env *env,
5101 struct dt_object *dt, const struct lu_buf *buf,
5102 const char *name, int fl, struct thandle *th)
5104 struct lod_thread_info *info = lod_env_info(env);
5105 struct dt_object *next = dt_object_child(dt);
5106 struct lu_attr *layout_attr = &info->lti_layout_attr;
5107 struct lod_object *lo = lod_dt_obj(dt);
5108 struct lod_obj_stripe_cb_data data = { {0} };
5113 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
5114 !strcmp(name, XATTR_NAME_LMV)) {
5116 case LU_XATTR_CREATE:
5117 rc = lod_dir_striping_create(env, dt, NULL, buf, NULL,
5121 case LU_XATTR_REPLACE:
5122 rc = lod_dir_layout_set(env, dt, buf, fl, th);
5129 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
5130 strcmp(name, XATTR_NAME_LOV) == 0) {
5131 rc = lod_xattr_set_default_lov_on_dir(env, dt, buf, name, fl,
5134 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
5135 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
5137 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
5140 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
5141 (strcmp(name, XATTR_NAME_LOV) == 0 ||
5142 strcmp(name, XATTR_LUSTRE_LOV) == 0 ||
5143 allowed_lustre_lov(name))) {
5144 /* layout has been changed by others in the transaction */
5145 rc = lod_check_layout_gen_intrans(info, lo);
5148 "%s: obj "DFID" gen changed from %d to %d in transaction, retry the transaction\n",
5149 dt->do_lu.lo_dev->ld_obd->obd_name,
5150 PFID(lu_object_fid(&dt->do_lu)),
5151 info->lti_gen[rc - 1], lo->ldo_layout_gen);
5155 /* in case of lov EA swap, just set it
5156 * if not, it is a replay so check striping match what we
5157 * already have during req replay, declare_xattr_set()
5158 * defines striping, then create() does the work */
5159 if (fl & LU_XATTR_REPLACE) {
5160 /* free stripes, then update disk */
5161 lod_striping_free(env, lod_dt_obj(dt));
5163 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
5164 } else if (fl & LU_XATTR_SPLIT) {
5165 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
5169 rc = lod_striping_reload(env, lo, buf, LVF_ALL_STALE);
5173 if (lo->ldo_mirror_count > 1 &&
5174 layout_attr->la_valid & LA_LAYOUT_VERSION) {
5176 layout_attr->la_layout_version =
5178 data.locd_attr = layout_attr;
5179 data.locd_declare = false;
5180 data.locd_stripe_cb =
5181 lod_obj_stripe_attr_set_cb;
5182 rc = lod_obj_for_each_stripe(env, lo, th,
5187 } else if (fl & LU_XATTR_PURGE) {
5188 rc = lod_layout_purge(env, dt, buf, th);
5189 } else if (dt_object_remote(dt)) {
5190 /* This only happens during migration, see
5191 * mdd_migrate_create(), in which Master MDT will
5192 * create a remote target object, and only set
5193 * (migrating) stripe EA on the remote object,
5194 * and does not need creating each stripes. */
5195 rc = lod_sub_xattr_set(env, next, buf, name,
5197 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
5198 /* delete component(s) */
5199 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
5200 rc = lod_layout_del(env, dt, th);
5203 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
5204 * it's going to create file with specified
5205 * component(s), the striping must have not being
5206 * cached in this case;
5208 * Otherwise, it's going to add/change component(s) to
5209 * an existing file, the striping must have been cached
5212 if (!(fl & LU_XATTR_MERGE))
5213 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
5214 !strcmp(name, XATTR_NAME_LOV),
5215 !lod_dt_obj(dt)->ldo_comp_cached));
5217 rc = lod_striped_create(env, dt, NULL, NULL, th);
5221 if (fl & LU_XATTR_MERGE && lo->ldo_mirror_count > 1 &&
5222 layout_attr->la_valid & LA_LAYOUT_VERSION) {
5223 /* mirror merge exec phase */
5224 layout_attr->la_layout_version =
5226 data.locd_attr = layout_attr;
5227 data.locd_declare = false;
5228 data.locd_stripe_cb =
5229 lod_obj_stripe_attr_set_cb;
5230 rc = lod_obj_for_each_stripe(env, lo, th,
5237 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
5238 rc = lod_replace_parent_fid(env, dt, buf, th, false);
5243 /* then all other xattr */
5244 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
5250 * Implementation of dt_object_operations::do_declare_xattr_del.
5252 * \see dt_object_operations::do_declare_xattr_del() in the API description
5255 static int lod_declare_xattr_del(const struct lu_env *env,
5256 struct dt_object *dt, const char *name,
5259 struct lod_object *lo = lod_dt_obj(dt);
5260 struct dt_object *next = dt_object_child(dt);
5265 rc = lod_sub_declare_xattr_del(env, next, name, th);
5269 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
5272 /* NB: don't delete stripe LMV, because when we do this, normally we
5273 * will remove stripes, besides, if directory LMV is corrupt, this will
5274 * prevent deleting its LMV and fixing it (via LFSCK).
5276 if (!strcmp(name, XATTR_NAME_LMV))
5279 rc = lod_striping_load(env, lo);
5283 if (lo->ldo_dir_stripe_count == 0)
5286 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5287 struct dt_object *dto = lo->ldo_stripe[i];
5292 if (!dt_object_exists(dto))
5295 rc = lod_sub_declare_xattr_del(env, dto, name, th);
5304 * Implementation of dt_object_operations::do_xattr_del.
5306 * If EA storing a regular striping is being deleted, then release
5307 * all the references to the stripe objects in core.
5309 * \see dt_object_operations::do_xattr_del() in the API description for details.
5311 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
5312 const char *name, struct thandle *th)
5318 if (!strcmp(name, XATTR_NAME_LOV) || !strcmp(name, XATTR_NAME_LMV))
5319 lod_striping_free(env, lod_dt_obj(dt));
5321 rc = lod_xattr_del_internal(env, dt, name, th);
5327 * Implementation of dt_object_operations::do_xattr_list.
5329 * \see dt_object_operations::do_xattr_list() in the API description
5332 static int lod_xattr_list(const struct lu_env *env,
5333 struct dt_object *dt, const struct lu_buf *buf)
5335 return dt_xattr_list(env, dt_object_child(dt), buf);
5338 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
5340 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
5344 * Copy OST list from layout provided by user.
5346 * \param[in] lod_comp layout_component to be filled
5347 * \param[in] v3 LOV EA V3 user data
5349 * \retval 0 on success
5350 * \retval negative if failed
5352 int lod_comp_copy_ost_lists(struct lod_layout_component *lod_comp,
5353 struct lov_user_md_v3 *v3)
5359 if (v3->lmm_stripe_offset == LOV_OFFSET_DEFAULT)
5360 v3->lmm_stripe_offset = v3->lmm_objects[0].l_ost_idx;
5362 if (lod_comp->llc_ostlist.op_array) {
5363 if (lod_comp->llc_ostlist.op_size >=
5364 v3->lmm_stripe_count * sizeof(__u32)) {
5365 lod_comp->llc_ostlist.op_count =
5366 v3->lmm_stripe_count;
5369 OBD_FREE(lod_comp->llc_ostlist.op_array,
5370 lod_comp->llc_ostlist.op_size);
5373 /* copy ost list from lmm */
5374 lod_comp->llc_ostlist.op_count = v3->lmm_stripe_count;
5375 lod_comp->llc_ostlist.op_size = v3->lmm_stripe_count * sizeof(__u32);
5376 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
5377 lod_comp->llc_ostlist.op_size);
5378 if (!lod_comp->llc_ostlist.op_array)
5381 for (j = 0; j < v3->lmm_stripe_count; j++) {
5382 lod_comp->llc_ostlist.op_array[j] =
5383 v3->lmm_objects[j].l_ost_idx;
5391 * Get default striping.
5393 * \param[in] env execution environment
5394 * \param[in] lo object
5395 * \param[out] lds default striping
5397 * \retval 0 on success
5398 * \retval negative if failed
5400 static int lod_get_default_lov_striping(const struct lu_env *env,
5401 struct lod_object *lo,
5402 struct lod_default_striping *lds,
5403 struct dt_allocation_hint *dah)
5405 struct lod_thread_info *info = lod_env_info(env);
5406 struct lov_user_md_v1 *v1 = NULL;
5407 struct lov_user_md_v3 *v3 = NULL;
5408 struct lov_comp_md_v1 *lcm = NULL;
5410 int append_stripe_count = dah != NULL ? dah->dah_append_stripe_count : 0;
5411 const char *append_pool = (dah != NULL &&
5412 dah->dah_append_pool != NULL &&
5413 dah->dah_append_pool[0] != '\0') ?
5414 dah->dah_append_pool : NULL;
5415 __u16 entry_count = 1;
5416 __u16 mirror_count = 0;
5417 bool want_composite = false;
5422 lds->lds_def_striping_set = 0;
5424 rc = lod_get_lov_ea(env, lo);
5428 if (rc < (typeof(rc))sizeof(struct lov_user_md))
5431 magic = *(__u32 *)info->lti_ea_store;
5432 if (magic == __swab32(LOV_USER_MAGIC_V1)) {
5433 lustre_swab_lov_user_md_v1(info->lti_ea_store);
5434 } else if (magic == __swab32(LOV_USER_MAGIC_V3)) {
5435 lustre_swab_lov_user_md_v3(info->lti_ea_store);
5436 } else if (magic == __swab32(LOV_USER_MAGIC_SPECIFIC)) {
5437 v3 = (struct lov_user_md_v3 *)info->lti_ea_store;
5438 lustre_swab_lov_user_md_v3(v3);
5439 lustre_swab_lov_user_md_objects(v3->lmm_objects,
5440 v3->lmm_stripe_count);
5441 } else if (magic == __swab32(LOV_USER_MAGIC_COMP_V1) ||
5442 magic == __swab32(LOV_USER_MAGIC_SEL)) {
5443 lustre_swab_lov_comp_md_v1(info->lti_ea_store);
5449 case LOV_USER_MAGIC_SPECIFIC:
5450 v1 = info->lti_ea_store;
5452 case LOV_MAGIC_COMP_V1:
5454 lcm = info->lti_ea_store;
5455 entry_count = lcm->lcm_entry_count;
5456 if (entry_count == 0)
5459 mirror_count = lcm->lcm_mirror_count + 1;
5460 want_composite = true;
5466 if (append_stripe_count != 0 || append_pool != NULL) {
5469 want_composite = false;
5472 /* realloc default comp entries if necessary */
5473 rc = lod_def_striping_comp_resize(lds, entry_count);
5477 lds->lds_def_comp_cnt = entry_count;
5478 lds->lds_def_striping_is_composite = want_composite;
5479 lds->lds_def_mirror_cnt = mirror_count;
5481 for (i = 0; i < entry_count; i++) {
5482 struct lod_layout_component *llc = &lds->lds_def_comp_entries[i];
5486 * reset llc values, llc_stripes is always NULL in the
5487 * default striping template, llc_pool will be reset
5488 * later below using lod_set_pool().
5490 * XXX At this point llc_pool may point to valid (!)
5491 * kmalloced strings from previous RPCs.
5493 memset(llc, 0, offsetof(typeof(*llc), llc_pool));
5496 v1 = (struct lov_user_md *)((char *)lcm +
5497 lcm->lcm_entries[i].lcme_offset);
5499 if (want_composite) {
5500 llc->llc_extent = lcm->lcm_entries[i].lcme_extent;
5501 /* We only inherit certain flags from the layout */
5502 llc->llc_flags = lcm->lcm_entries[i].lcme_flags &
5503 LCME_TEMPLATE_FLAGS;
5507 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",
5508 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
5511 v1->lmm_stripe_count,
5512 v1->lmm_stripe_size,
5513 v1->lmm_stripe_offset,
5515 append_stripe_count);
5517 if (!lov_pattern_supported(v1->lmm_pattern) &&
5518 !(v1->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
5519 lod_free_def_comp_entries(lds);
5523 llc->llc_stripe_count = v1->lmm_stripe_count;
5524 llc->llc_stripe_size = v1->lmm_stripe_size;
5525 llc->llc_stripe_offset = v1->lmm_stripe_offset;
5526 llc->llc_pattern = v1->lmm_pattern;
5528 if (append_stripe_count != 0 || append_pool != NULL)
5529 llc->llc_pattern = LOV_PATTERN_RAID0;
5531 if (append_stripe_count != 0)
5532 llc->llc_stripe_count = append_stripe_count;
5535 if (append_pool != NULL) {
5537 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
5538 /* XXX: sanity check here */
5539 v3 = (struct lov_user_md_v3 *)v1;
5540 if (v3->lmm_pool_name[0] != '\0')
5541 pool = v3->lmm_pool_name;
5544 lod_set_pool(&llc->llc_pool, pool);
5546 if (v1->lmm_magic == LOV_USER_MAGIC_SPECIFIC &&
5547 append_stripe_count == 0 &&
5548 append_pool == NULL) {
5549 v3 = (struct lov_user_md_v3 *)v1;
5550 rc = lod_comp_copy_ost_lists(llc, v3);
5553 } else if (llc->llc_ostlist.op_array &&
5554 llc->llc_ostlist.op_count) {
5555 for (j = 0; j < llc->llc_ostlist.op_count; j++)
5556 llc->llc_ostlist.op_array[j] = -1;
5557 llc->llc_ostlist.op_count = 0;
5561 lds->lds_def_striping_set = 1;
5565 static inline void lod_lum2lds(struct lod_default_striping *lds,
5566 const struct lmv_user_md *lum)
5568 lds->lds_dir_def_stripe_count = le32_to_cpu(lum->lum_stripe_count);
5569 lds->lds_dir_def_stripe_offset = le32_to_cpu(lum->lum_stripe_offset);
5570 lds->lds_dir_def_hash_type = le32_to_cpu(lum->lum_hash_type);
5571 lds->lds_dir_def_max_inherit = lum->lum_max_inherit;
5572 lds->lds_dir_def_max_inherit_rr = lum->lum_max_inherit_rr;
5573 lds->lds_dir_def_striping_set = 1;
5577 * Get default directory striping.
5579 * \param[in] env execution environment
5580 * \param[in] lo object
5581 * \param[out] lds default striping
5583 * \retval 0 on success
5584 * \retval negative if failed
5586 static int lod_get_default_lmv_striping(const struct lu_env *env,
5587 struct lod_object *lo,
5588 struct lod_default_striping *lds)
5590 struct lmv_user_md *lmu;
5593 lds->lds_dir_def_striping_set = 0;
5595 rc = lod_get_default_lmv_ea(env, lo);
5599 if (rc >= (int)sizeof(*lmu)) {
5600 struct lod_thread_info *info = lod_env_info(env);
5602 lmu = info->lti_ea_store;
5603 lod_lum2lds(lds, lmu);
5610 * Get default striping in the object.
5612 * Get object default striping and default directory striping.
5614 * \param[in] env execution environment
5615 * \param[in] lo object
5616 * \param[out] lds default striping
5618 * \retval 0 on success
5619 * \retval negative if failed
5621 static int lod_get_default_striping(const struct lu_env *env,
5622 struct lod_object *lo,
5623 struct dt_allocation_hint *ah,
5624 struct lod_default_striping *lds)
5628 rc = lod_get_default_lov_striping(env, lo, lds, NULL);
5629 if (lds->lds_def_striping_set) {
5630 struct lod_thread_info *info = lod_env_info(env);
5631 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5633 rc = lod_verify_striping(env, d, lo, &info->lti_buf, false);
5635 lds->lds_def_striping_set = 0;
5638 if (ah->dah_eadata_is_dmv) {
5639 lod_lum2lds(lds, ah->dah_eadata);
5640 } else if (ah->dah_dmv_imp_inherit) {
5641 lds->lds_dir_def_striping_set = 0;
5643 rc1 = lod_get_default_lmv_striping(env, lo, lds);
5644 if (rc == 0 && rc1 < 0)
5652 * Apply default striping on object.
5654 * If object striping pattern is not set, set to the one in default striping.
5655 * The default striping is from parent or fs.
5657 * \param[in] lo new object
5658 * \param[in] lds default striping
5659 * \param[in] mode new object's mode
5661 static void lod_striping_from_default(struct lod_object *lo,
5662 const struct lod_default_striping *lds,
5665 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5668 if (lds->lds_def_striping_set && S_ISREG(mode)) {
5669 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
5671 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
5672 lds->lds_def_comp_cnt);
5676 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
5677 if (lds->lds_def_mirror_cnt > 1)
5678 lo->ldo_flr_state = LCM_FL_RDONLY;
5680 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5681 struct lod_layout_component *obj_comp =
5682 &lo->ldo_comp_entries[i];
5683 struct lod_layout_component *def_comp =
5684 &lds->lds_def_comp_entries[i];
5687 "inherit "DFID" file layout from default: flags=%#x size=%u nr=%u offset=%u pattern=%#x pool=%s\n",
5688 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
5689 def_comp->llc_flags,
5690 def_comp->llc_stripe_size,
5691 def_comp->llc_stripe_count,
5692 def_comp->llc_stripe_offset,
5693 def_comp->llc_pattern,
5694 def_comp->llc_pool ?: "");
5696 *obj_comp = *def_comp;
5697 if (def_comp->llc_pool != NULL) {
5698 /* pointer was copied from def_comp */
5699 obj_comp->llc_pool = NULL;
5700 lod_obj_set_pool(lo, i, def_comp->llc_pool);
5704 if (def_comp->llc_ostlist.op_array &&
5705 def_comp->llc_ostlist.op_count) {
5706 OBD_ALLOC(obj_comp->llc_ostlist.op_array,
5707 obj_comp->llc_ostlist.op_size);
5708 if (!obj_comp->llc_ostlist.op_array)
5710 memcpy(obj_comp->llc_ostlist.op_array,
5711 def_comp->llc_ostlist.op_array,
5712 obj_comp->llc_ostlist.op_size);
5713 } else if (def_comp->llc_ostlist.op_array) {
5714 obj_comp->llc_ostlist.op_array = NULL;
5718 * Don't initialize these fields for plain layout
5719 * (v1/v3) here, they are inherited in the order of
5720 * 'parent' -> 'fs default (root)' -> 'global default
5721 * values for stripe_count & stripe_size'.
5723 * see lod_ah_init().
5725 if (!lo->ldo_is_composite)
5728 lod_adjust_stripe_info(obj_comp, desc, 0);
5730 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
5731 if (lo->ldo_dir_stripe_count == 0)
5732 lo->ldo_dir_stripe_count =
5733 lds->lds_dir_def_stripe_count;
5734 if (lo->ldo_dir_stripe_offset == -1)
5735 lo->ldo_dir_stripe_offset =
5736 lds->lds_dir_def_stripe_offset;
5737 if (lo->ldo_dir_hash_type == LMV_HASH_TYPE_UNKNOWN)
5738 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
5741 "inherit "DFID" dir layout from default: count=%hu offset=%u hash_type=%x\n",
5742 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
5743 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
5744 lo->ldo_dir_hash_type);
5748 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root,
5749 const char *append_pool)
5751 struct lod_layout_component *lod_comp;
5753 if (lo->ldo_comp_cnt == 0)
5756 if (lo->ldo_is_composite)
5759 lod_comp = &lo->ldo_comp_entries[0];
5761 if (lod_comp->llc_stripe_count <= 0 ||
5762 lod_comp->llc_stripe_size <= 0)
5765 if (from_root && (lod_comp->llc_pool == NULL ||
5766 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
5769 if (append_pool && append_pool[0])
5776 * Implementation of dt_object_operations::do_ah_init.
5778 * This method is used to make a decision on the striping configuration for the
5779 * object being created. It can be taken from the \a parent object if it exists,
5780 * or filesystem's default. The resulting configuration (number of stripes,
5781 * stripe size/offset, pool name, hash_type, etc.) is stored in the object
5782 * itself and will be used by the methods like ->doo_declare_create().
5784 * \see dt_object_operations::do_ah_init() in the API description for details.
5786 static void lod_ah_init(const struct lu_env *env,
5787 struct dt_allocation_hint *ah,
5788 struct dt_object *parent,
5789 struct dt_object *child,
5792 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
5793 struct lod_thread_info *info = lod_env_info(env);
5794 struct lod_default_striping *lds = lod_lds_buf_get(env);
5795 struct dt_object *nextp = NULL;
5796 struct dt_object *nextc;
5797 struct lod_object *lp = NULL;
5798 struct lod_object *lc;
5799 struct lov_desc *desc;
5800 struct lod_layout_component *lod_comp;
5806 if (ah->dah_append_stripe_count == -1)
5807 ah->dah_append_stripe_count =
5808 d->lod_ost_descs.ltd_lov_desc.ld_tgt_count;
5810 if (likely(parent)) {
5811 nextp = dt_object_child(parent);
5812 lp = lod_dt_obj(parent);
5815 nextc = dt_object_child(child);
5816 lc = lod_dt_obj(child);
5818 LASSERT(!lod_obj_is_striped(child));
5819 /* default layout template may have been set on the regular file
5820 * when this is called from mdd_create_data() */
5821 if (S_ISREG(child_mode))
5822 lod_free_comp_entries(lc);
5824 if (!dt_object_exists(nextc))
5825 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
5827 if (S_ISDIR(child_mode)) {
5828 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
5829 int max_stripe_count;
5831 /* other default values are 0 */
5832 lc->ldo_dir_stripe_offset = LMV_OFFSET_DEFAULT;
5834 /* no default striping configuration is needed for
5837 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5838 le32_to_cpu(lum1->lum_magic) == LMV_MAGIC_FOREIGN) {
5839 lc->ldo_is_foreign = true;
5840 /* keep stripe_count 0 and stripe_offset -1 */
5841 CDEBUG(D_INFO, "no default striping for foreign dir\n");
5845 if (likely(lp != NULL))
5846 lod_get_default_striping(env, lp, ah, lds);
5848 /* It should always honour the specified stripes */
5849 if (ah->dah_eadata && ah->dah_eadata_len &&
5850 !ah->dah_eadata_is_dmv &&
5851 (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
5852 le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC ||
5853 le32_to_cpu(lum1->lum_magic) == LMV_MAGIC_V1)) {
5854 lc->ldo_dir_stripe_count =
5855 le32_to_cpu(lum1->lum_stripe_count);
5856 lc->ldo_dir_stripe_offset =
5857 le32_to_cpu(lum1->lum_stripe_offset);
5858 lc->ldo_dir_hash_type =
5859 le32_to_cpu(lum1->lum_hash_type);
5861 "set dirstripe: count %hu, offset %d, hash %x\n",
5862 lc->ldo_dir_stripe_count,
5863 (int)lc->ldo_dir_stripe_offset,
5864 lc->ldo_dir_hash_type);
5866 if (d->lod_mdt_descs.ltd_lmv_desc.ld_active_tgt_count &&
5867 lc->ldo_dir_stripe_count < 2 &&
5868 lum1->lum_max_inherit != LMV_INHERIT_NONE) {
5869 /* when filesystem-wide default LMV is set, dirs
5870 * will be created on MDT by space usage, but if
5871 * dir is created with "lfs mkdir -c 1 ...", its
5872 * subdirs should be kept on the same MDT. To
5873 * guarantee this, set default LMV for such dir.
5875 lds->lds_dir_def_stripe_count =
5876 le32_to_cpu(lum1->lum_stripe_count);
5877 /* if "-1" stripe offset is set, save current
5878 * MDT index in default LMV.
5880 if (le32_to_cpu(lum1->lum_stripe_offset) ==
5882 lds->lds_dir_def_stripe_offset =
5883 lod2lu_dev(d)->ld_site->ld_seq_site->ss_node_id;
5885 lds->lds_dir_def_stripe_offset =
5886 le32_to_cpu(lum1->lum_stripe_offset);
5887 lds->lds_dir_def_hash_type =
5888 le32_to_cpu(lum1->lum_hash_type);
5889 lds->lds_dir_def_max_inherit =
5890 lum1->lum_max_inherit;
5891 /* it will be decreased by 1 later in setting */
5892 if (lum1->lum_max_inherit >= LMV_INHERIT_END &&
5893 lum1->lum_max_inherit < LMV_INHERIT_MAX)
5894 lds->lds_dir_def_max_inherit++;
5895 lds->lds_dir_def_max_inherit_rr =
5896 lum1->lum_max_inherit_rr;
5897 lds->lds_dir_def_striping_set = 1;
5898 /* don't inherit LOV from ROOT */
5899 if (lds->lds_def_striping_set &&
5900 fid_is_root(lod_object_fid(lp)))
5901 lds->lds_def_striping_set = 0;
5902 lc->ldo_def_striping = lds;
5903 } else if (lds->lds_def_striping_set &&
5904 !fid_is_root(lod_object_fid(lp))) {
5905 /* don't inherit default LMV for "lfs mkdir" */
5906 lds->lds_dir_def_striping_set = 0;
5907 lc->ldo_def_striping = lds;
5910 /* inherit default striping except ROOT */
5911 if ((lds->lds_def_striping_set ||
5912 lds->lds_dir_def_striping_set) &&
5913 !fid_is_root(lod_object_fid(lp)))
5914 lc->ldo_def_striping = lds;
5916 /* transfer defaults LMV to new directory */
5917 lod_striping_from_default(lc, lds, child_mode);
5919 /* set count 0 to create normal directory */
5920 if (lc->ldo_dir_stripe_count == 1)
5921 lc->ldo_dir_stripe_count = 0;
5923 /* do not save default LMV on server */
5924 if (ah->dah_dmv_imp_inherit) {
5925 lds->lds_dir_def_striping_set = 0;
5926 if (!lds->lds_def_striping_set)
5927 lc->ldo_def_striping = NULL;
5931 /* shrink the stripe count to max_mdt_stripecount if it is -1
5932 * and max_mdt_stripecount is not 0
5934 if (lc->ldo_dir_stripe_count == (__u16)(-1) &&
5935 d->lod_max_mdt_stripecount)
5936 lc->ldo_dir_stripe_count = d->lod_max_mdt_stripecount;
5938 max_stripe_count = d->lod_remote_mdt_count + 1;
5939 if (lc->ldo_dir_hash_type & LMV_HASH_FLAG_OVERSTRIPED)
5941 max_stripe_count * LMV_MAX_STRIPES_PER_MDT;
5943 /* shrink the stripe_count to max stripe count */
5944 if (lc->ldo_dir_stripe_count > max_stripe_count &&
5945 !CFS_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)) {
5946 lc->ldo_dir_stripe_count = max_stripe_count;
5947 if (lc->ldo_dir_stripe_count == 1)
5948 lc->ldo_dir_stripe_count = 0;
5951 if (!lmv_is_known_hash_type(lc->ldo_dir_hash_type))
5952 lc->ldo_dir_hash_type =
5953 (lc->ldo_dir_hash_type & LMV_HASH_FLAG_KNOWN) |
5954 d->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
5956 /* make sure all fscrypt metadata stays on same mdt */
5957 if (child->do_lu.lo_header->loh_attr & LOHA_FSCRYPT_MD) {
5958 lc->ldo_dir_stripe_count = 0;
5959 lds->lds_dir_def_stripe_offset =
5960 lod2lu_dev(d)->ld_site->ld_seq_site->ss_node_id;
5961 lds->lds_dir_def_striping_set = 1;
5962 lc->ldo_def_striping = lds;
5965 CDEBUG(D_INFO, "final dir stripe_count=%hu offset=%d hash=%x\n",
5966 lc->ldo_dir_stripe_count,
5967 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
5972 /* child object regular file*/
5974 if (!lod_object_will_be_striped(S_ISREG(child_mode),
5975 lu_object_fid(&child->do_lu)))
5978 /* If object is going to be striped over OSTs, transfer default
5979 * striping information to the child, so that we can use it
5980 * during declaration and creation.
5982 * Try from the parent first.
5984 if (likely(lp != NULL)) {
5985 rc = lod_get_default_lov_striping(env, lp, lds, ah);
5986 if (rc == 0 && lds->lds_def_striping_set) {
5987 rc = lod_verify_striping(env, d, lp, &info->lti_buf,
5990 lod_striping_from_default(lc, lds, child_mode);
5994 /* Initialize lod_device::lod_md_root object reference */
5995 if (d->lod_md_root == NULL) {
5996 struct dt_object *root;
5997 struct lod_object *lroot;
5999 lu_root_fid(&info->lti_fid);
6000 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
6001 if (!IS_ERR(root)) {
6002 lroot = lod_dt_obj(root);
6004 spin_lock(&d->lod_lock);
6005 if (d->lod_md_root != NULL)
6006 dt_object_put(env, &d->lod_md_root->ldo_obj);
6007 d->lod_md_root = lroot;
6008 spin_unlock(&d->lod_lock);
6012 /* try inherit layout from the root object (fs default) when:
6013 * - parent does not have default layout; or
6014 * - parent has plain(v1/v3) default layout, and some attributes
6015 * are not specified in the default layout;
6017 if (d->lod_md_root != NULL &&
6018 lod_need_inherit_more(lc, true, ah->dah_append_pool)) {
6019 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds,
6021 if (rc || !lds->lds_def_striping_set)
6024 rc = lod_verify_striping(env, d, d->lod_md_root, &info->lti_buf,
6029 if (lc->ldo_comp_cnt == 0) {
6030 lod_striping_from_default(lc, lds, child_mode);
6031 } else if (!lds->lds_def_striping_is_composite) {
6032 struct lod_layout_component *def_comp;
6034 LASSERT(!lc->ldo_is_composite);
6035 lod_comp = &lc->ldo_comp_entries[0];
6036 def_comp = &lds->lds_def_comp_entries[0];
6038 if (lod_comp->llc_stripe_count <= 0)
6039 lod_comp->llc_stripe_count =
6040 def_comp->llc_stripe_count;
6041 if (lod_comp->llc_stripe_size <= 0)
6042 lod_comp->llc_stripe_size =
6043 def_comp->llc_stripe_size;
6044 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT &&
6045 (!lod_comp->llc_pool || !lod_comp->llc_pool[0]))
6046 lod_comp->llc_stripe_offset =
6047 def_comp->llc_stripe_offset;
6048 if (lod_comp->llc_pool == NULL)
6049 lod_qos_set_pool(lc, 0, def_comp->llc_pool);
6054 * fs default striping may not be explicitly set, or historically set
6055 * in config log, use them.
6057 if (lod_need_inherit_more(lc, false, ah->dah_append_pool)) {
6058 if (lc->ldo_comp_cnt == 0) {
6059 rc = lod_alloc_comp_entries(lc, 0, 1);
6061 /* fail to allocate memory, will create a
6062 * non-striped file. */
6064 lc->ldo_is_composite = 0;
6065 lod_comp = &lc->ldo_comp_entries[0];
6066 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
6068 LASSERT(!lc->ldo_is_composite);
6069 lod_comp = &lc->ldo_comp_entries[0];
6070 desc = &d->lod_ost_descs.ltd_lov_desc;
6071 lod_adjust_stripe_info(lod_comp, desc,
6072 ah->dah_append_stripe_count);
6073 if (ah->dah_append_pool && ah->dah_append_pool[0])
6074 lod_qos_set_pool(lc, 0, ah->dah_append_pool);
6081 * Size initialization on late striping.
6083 * Propagate the size of a truncated object to a deferred striping.
6084 * This function handles a special case when truncate was done on a
6085 * non-striped object and now while the striping is being created
6086 * we can't lose that size, so we have to propagate it to the stripes
6089 * \param[in] env execution environment
6090 * \param[in] dt object
6091 * \param[in] th transaction handle
6093 * \retval 0 on success
6094 * \retval negative if failed
6096 static int lod_declare_init_size(const struct lu_env *env,
6097 struct dt_object *dt, struct thandle *th)
6099 struct dt_object *next = dt_object_child(dt);
6100 struct lod_object *lo = lod_dt_obj(dt);
6101 struct dt_object **objects = NULL;
6102 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
6103 uint64_t size, offs;
6104 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
6105 struct lu_extent size_ext;
6108 if (!lod_obj_is_striped(dt))
6111 rc = dt_attr_get(env, next, attr);
6112 LASSERT(attr->la_valid & LA_SIZE);
6116 size = attr->la_size;
6120 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
6121 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6122 struct lod_layout_component *lod_comp;
6123 struct lu_extent *extent;
6125 lod_comp = &lo->ldo_comp_entries[i];
6127 if (lod_comp->llc_stripe == NULL)
6130 extent = &lod_comp->llc_extent;
6131 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
6132 if (!lo->ldo_is_composite ||
6133 lu_extent_is_overlapped(extent, &size_ext)) {
6134 objects = lod_comp->llc_stripe;
6135 stripe_count = lod_comp->llc_stripe_count;
6136 stripe_size = lod_comp->llc_stripe_size;
6139 if (stripe_count == 0)
6142 LASSERT(objects != NULL && stripe_size != 0);
6143 do_div(size, stripe_size);
6144 stripe = do_div(size, stripe_count);
6145 LASSERT(objects[stripe] != NULL);
6147 size = size * stripe_size;
6148 offs = attr->la_size;
6149 size += do_div(offs, stripe_size);
6151 attr->la_valid = LA_SIZE;
6152 attr->la_size = size;
6154 rc = lod_sub_declare_attr_set(env, objects[stripe],
6163 * Declare creation of striped object.
6165 * The function declares creation stripes for a regular object. The function
6166 * also declares whether the stripes will be created with non-zero size if
6167 * previously size was set non-zero on the master object. If object \a dt is
6168 * not local, then only fully defined striping can be applied in \a lovea.
6169 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
6172 * \param[in] env execution environment
6173 * \param[in] dt object
6174 * \param[in] attr attributes the stripes will be created with
6175 * \param[in] lovea a buffer containing striping description
6176 * \param[in] th transaction handle
6178 * \retval 0 on success
6179 * \retval negative if failed
6181 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
6182 struct lu_attr *attr,
6183 const struct lu_buf *lovea, struct thandle *th)
6185 struct lod_thread_info *info = lod_env_info(env);
6186 struct dt_object *next = dt_object_child(dt);
6187 struct lod_object *lo = lod_dt_obj(dt);
6191 if (CFS_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
6192 GOTO(out, rc = -ENOMEM);
6194 if (!dt_object_remote(next)) {
6195 /* choose OST and generate appropriate objects */
6196 rc = lod_prepare_create(env, lo, attr, lovea, th);
6201 * declare storage for striping data
6203 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6205 /* LOD can not choose OST objects for remote objects, i.e.
6206 * stripes must be ready before that. Right now, it can only
6207 * happen during migrate, i.e. migrate process needs to create
6208 * remote regular file (mdd_migrate_create), then the migrate
6209 * process will provide stripeEA. */
6210 LASSERT(lovea != NULL);
6211 info->lti_buf = *lovea;
6214 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
6215 XATTR_NAME_LOV, 0, th);
6220 * if striping is created with local object's size > 0,
6221 * we have to propagate this size to specific object
6222 * the case is possible only when local object was created previously
6224 if (dt_object_exists(next))
6225 rc = lod_declare_init_size(env, dt, th);
6228 /* failed to create striping or to set initial size, let's reset
6229 * config so that others don't get confused */
6231 lod_striping_free(env, lo);
6237 * Whether subdirectories under \a dt should be created on MDTs by space QoS
6239 * If LMV_HASH_FLAG_SPACE is set on directory default layout, its subdirectories
6240 * should be created on MDT by space QoS.
6242 * \param[in] env execution environment
6243 * \param[in] dev lu device
6244 * \param[in] dt object
6246 * \retval 1 if directory should create subdir by space usage
6248 * \retval -ev if failed
6250 static inline int dt_object_qos_mkdir(const struct lu_env *env,
6251 struct lu_device *dev,
6252 struct dt_object *dt)
6254 struct lod_thread_info *info = lod_env_info(env);
6255 struct lu_object *obj;
6256 struct lod_object *lo;
6257 struct lmv_user_md *lmu;
6260 obj = lu_object_find_slice(env, dev, lu_object_fid(&dt->do_lu), NULL);
6262 return PTR_ERR(obj);
6264 lo = lu2lod_obj(obj);
6266 rc = lod_get_default_lmv_ea(env, lo);
6267 dt_object_put(env, dt);
6271 if (rc < (int)sizeof(*lmu))
6274 lmu = info->lti_ea_store;
6275 return le32_to_cpu(lmu->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
6279 * Implementation of dt_object_operations::do_declare_create.
6281 * The method declares creation of a new object. If the object will be striped,
6282 * then helper functions are called to find FIDs for the stripes, declare
6283 * creation of the stripes and declare initialization of the striping
6284 * information to be stored in the master object.
6286 * \see dt_object_operations::do_declare_create() in the API description
6289 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
6290 struct lu_attr *attr,
6291 struct dt_allocation_hint *hint,
6292 struct dt_object_format *dof, struct thandle *th)
6294 struct dt_object *next = dt_object_child(dt);
6295 struct lod_object *lo = lod_dt_obj(dt);
6304 * first of all, we declare creation of local object
6306 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
6311 * it's lod_ah_init() that has decided the object will be striped
6313 if (dof->dof_type == DFT_REGULAR) {
6314 /* callers don't want stripes */
6315 /* XXX: all tricky interactions with ->ah_make_hint() decided
6316 * to use striping, then ->declare_create() behaving differently
6317 * should be cleaned */
6318 if (dof->u.dof_reg.striped != 0)
6319 rc = lod_declare_striped_create(env, dt, attr,
6321 } else if (dof->dof_type == DFT_DIR) {
6322 struct seq_server_site *ss;
6323 struct lu_buf buf = { NULL };
6325 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
6327 /* If the parent has default stripeEA, and client
6328 * did not find it before sending create request,
6329 * then MDT will return -EREMOTE, and client will
6330 * retrieve the default stripeEA and re-create the
6333 * Note: if dah_eadata != NULL, it means creating the
6334 * striped directory with specified stripeEA, then it
6335 * should ignore the default stripeEA */
6336 if (hint != NULL && hint->dah_eadata == NULL) {
6337 if (CFS_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
6338 GOTO(out, rc = -EREMOTE);
6340 if (lo->ldo_dir_stripe_offset != LMV_OFFSET_DEFAULT &&
6341 lo->ldo_dir_stripe_offset != ss->ss_node_id) {
6342 struct lod_device *lod;
6343 struct lu_tgt_desc *mdt = NULL;
6344 bool found_mdt = false;
6346 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6347 lod_foreach_mdt(lod, mdt) {
6348 if (mdt->ltd_index ==
6349 lo->ldo_dir_stripe_offset) {
6355 /* If the MDT indicated by stripe_offset can be
6356 * found, then tell client to resend the create
6357 * request to the correct MDT, otherwise return
6358 * error to client */
6360 GOTO(out, rc = -EREMOTE);
6362 GOTO(out, rc = -EINVAL);
6364 } else if (hint && hint->dah_eadata) {
6365 buf.lb_buf = (void *)hint->dah_eadata;
6366 buf.lb_len = hint->dah_eadata_len;
6369 rc = lod_declare_dir_striping_create(env, dt, attr, &buf, dof,
6373 /* failed to create striping or to set initial size, let's reset
6374 * config so that others don't get confused */
6376 lod_striping_free(env, lo);
6381 * Generate component ID for new created component.
6383 * \param[in] lo LOD object
6384 * \param[in] comp_idx index of ldo_comp_entries
6386 * \retval component ID on success
6387 * \retval LCME_ID_INVAL on failure
6389 static __u32 lod_gen_component_id(struct lod_object *lo,
6390 int mirror_id, int comp_idx)
6392 struct lod_layout_component *lod_comp;
6393 __u32 id, start, end;
6396 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
6398 lod_obj_inc_layout_gen(lo);
6399 id = lo->ldo_layout_gen;
6400 if (likely(id <= SEQ_ID_MAX))
6401 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
6403 /* Layout generation wraps, need to check collisions. */
6404 start = id & SEQ_ID_MASK;
6407 for (id = start; id <= end; id++) {
6408 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6409 lod_comp = &lo->ldo_comp_entries[i];
6410 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
6413 /* Found the ununsed ID */
6414 if (i == lo->ldo_comp_cnt)
6415 RETURN(pflr_id(mirror_id, id));
6418 if (end == SEQ_ID_MAX) {
6419 end = min_t(__u32, start, SEQ_ID_MAX) - 1;
6424 RETURN(LCME_ID_INVAL);
6428 * Creation of a striped regular object.
6430 * The function is called to create the stripe objects for a regular
6431 * striped file. This can happen at the initial object creation or
6432 * when the caller asks LOD to do so using ->do_xattr_set() method
6433 * (so called late striping). Notice all the information are already
6434 * prepared in the form of the list of objects (ldo_stripe field).
6435 * This is done during declare phase.
6437 * \param[in] env execution environment
6438 * \param[in] dt object
6439 * \param[in] attr attributes the stripes will be created with
6440 * \param[in] dof format of stripes (see OSD API description)
6441 * \param[in] th transaction handle
6443 * \retval 0 on success
6444 * \retval negative if failed
6446 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
6447 struct lu_attr *attr, struct dt_object_format *dof,
6450 struct lod_layout_component *lod_comp;
6451 struct lod_object *lo = lod_dt_obj(dt);
6456 mutex_lock(&lo->ldo_layout_mutex);
6458 LASSERT((lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL) ||
6459 lo->ldo_is_foreign);
6461 mirror_id = 0; /* non-flr file's mirror_id is 0 */
6462 if (lo->ldo_mirror_count > 1) {
6463 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6464 lod_comp = &lo->ldo_comp_entries[i];
6465 if (lod_comp->llc_id != LCME_ID_INVAL &&
6466 mirror_id_of(lod_comp->llc_id) > mirror_id)
6467 mirror_id = mirror_id_of(lod_comp->llc_id);
6471 /* create all underlying objects */
6472 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6473 lod_comp = &lo->ldo_comp_entries[i];
6475 if (lod_comp->llc_id == LCME_ID_INVAL) {
6476 /* only the component of FLR layout with more than 1
6477 * mirror has mirror ID in its component ID.
6479 if (lod_comp->llc_extent.e_start == 0 &&
6480 lo->ldo_mirror_count > 1)
6483 lod_comp->llc_id = lod_gen_component_id(lo,
6485 if (lod_comp->llc_id == LCME_ID_INVAL)
6486 GOTO(out, rc = -ERANGE);
6489 if (lod_comp_inited(lod_comp))
6492 if (lod_comp->llc_magic == LOV_MAGIC_FOREIGN) {
6493 lod_comp_set_init(lod_comp);
6497 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
6498 lod_comp_set_init(lod_comp);
6500 if (lov_pattern(lod_comp->llc_pattern) & LOV_PATTERN_MDT)
6501 lod_comp_set_init(lod_comp);
6503 if (lod_comp->llc_stripe == NULL)
6506 LASSERT(lod_comp->llc_stripe_count);
6507 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6508 struct dt_object *object = lod_comp->llc_stripe[j];
6509 LASSERT(object != NULL);
6510 rc = lod_sub_create(env, object, attr, NULL, dof, th);
6514 lod_comp_set_init(lod_comp);
6517 rc = lod_fill_mirrors(lo);
6521 lo->ldo_comp_cached = 1;
6523 rc = lod_generate_and_set_lovea(env, lo, th);
6527 mutex_unlock(&lo->ldo_layout_mutex);
6532 lod_striping_free_nolock(env, lo);
6533 mutex_unlock(&lo->ldo_layout_mutex);
6538 static inline bool lod_obj_is_dom(struct dt_object *dt)
6540 struct lod_object *lo = lod_dt_obj(dt);
6542 if (!dt_object_exists(dt_object_child(dt)))
6545 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
6548 if (!lo->ldo_comp_cnt)
6551 return (lov_pattern(lo->ldo_comp_entries[0].llc_pattern) &
6556 * Implementation of dt_object_operations::do_create.
6558 * If any of preceeding methods (like ->do_declare_create(),
6559 * ->do_ah_init(), etc) chose to create a striped object,
6560 * then this method will create the master and the stripes.
6562 * \see dt_object_operations::do_create() in the API description for details.
6564 static int lod_create(const struct lu_env *env, struct dt_object *dt,
6565 struct lu_attr *attr, struct dt_allocation_hint *hint,
6566 struct dt_object_format *dof, struct thandle *th)
6571 /* create local object */
6572 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
6576 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
6577 (lod_obj_is_striped(dt) || lod_obj_is_dom(dt)) &&
6578 dof->u.dof_reg.striped != 0) {
6579 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
6580 rc = lod_striped_create(env, dt, attr, dof, th);
6587 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
6588 struct dt_object *dt, struct thandle *th,
6589 int comp_idx, int stripe_idx,
6590 struct lod_obj_stripe_cb_data *data)
6592 if (data->locd_declare)
6593 return lod_sub_declare_destroy(env, dt, th);
6595 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
6596 stripe_idx == cfs_fail_val)
6597 return lod_sub_destroy(env, dt, th);
6603 * Implementation of dt_object_operations::do_declare_destroy.
6605 * If the object is a striped directory, then the function declares reference
6606 * removal from the master object (this is an index) to the stripes and declares
6607 * destroy of all the stripes. In all the cases, it declares an intention to
6608 * destroy the object itself.
6610 * \see dt_object_operations::do_declare_destroy() in the API description
6613 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
6616 struct dt_object *next = dt_object_child(dt);
6617 struct lod_object *lo = lod_dt_obj(dt);
6618 struct lod_thread_info *info = lod_env_info(env);
6619 struct dt_object *stripe;
6620 char *stripe_name = info->lti_key;
6626 * load striping information, notice we don't do this when object
6627 * is being initialized as we don't need this information till
6628 * few specific cases like destroy, chown
6630 rc = lod_striping_load(env, lo);
6634 /* declare destroy for all underlying objects */
6635 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6636 rc = next->do_ops->do_index_try(env, next,
6637 &dt_directory_features);
6641 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6642 stripe = lo->ldo_stripe[i];
6646 rc = lod_sub_declare_ref_del(env, next, th);
6650 snprintf(stripe_name, sizeof(info->lti_key),
6652 PFID(lu_object_fid(&stripe->do_lu)), i);
6653 rc = lod_sub_declare_delete(env, next,
6654 (const struct dt_key *)stripe_name, th);
6661 * we declare destroy for the local object
6663 rc = lod_sub_declare_destroy(env, next, th);
6667 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
6668 CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
6671 if (!lod_obj_is_striped(dt))
6674 /* declare destroy all striped objects */
6675 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6676 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6677 stripe = lo->ldo_stripe[i];
6681 if (!dt_object_exists(stripe))
6684 rc = lod_sub_declare_ref_del(env, stripe, th);
6688 rc = lod_sub_declare_destroy(env, stripe, th);
6693 struct lod_obj_stripe_cb_data data = { { 0 } };
6695 data.locd_declare = true;
6696 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
6697 rc = lod_obj_for_each_stripe(env, lo, th, &data);
6704 * Implementation of dt_object_operations::do_destroy.
6706 * If the object is a striped directory, then the function removes references
6707 * from the master object (this is an index) to the stripes and destroys all
6708 * the stripes. In all the cases, the function destroys the object itself.
6710 * \see dt_object_operations::do_destroy() in the API description for details.
6712 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
6715 struct dt_object *next = dt_object_child(dt);
6716 struct lod_object *lo = lod_dt_obj(dt);
6717 struct lod_thread_info *info = lod_env_info(env);
6718 char *stripe_name = info->lti_key;
6719 struct dt_object *stripe;
6725 /* destroy sub-stripe of master object */
6726 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6727 rc = next->do_ops->do_index_try(env, next,
6728 &dt_directory_features);
6732 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6733 stripe = lo->ldo_stripe[i];
6737 rc = lod_sub_ref_del(env, next, th);
6741 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
6742 PFID(lu_object_fid(&stripe->do_lu)), i);
6744 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
6745 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
6746 PFID(lu_object_fid(&stripe->do_lu)));
6748 rc = lod_sub_delete(env, next,
6749 (const struct dt_key *)stripe_name, th);
6755 rc = lod_sub_destroy(env, next, th);
6759 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
6760 CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
6763 if (!lod_obj_is_striped(dt))
6766 /* destroy all striped objects */
6767 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6768 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6769 stripe = lo->ldo_stripe[i];
6773 if (!dt_object_exists(stripe))
6776 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
6777 i == cfs_fail_val) {
6778 dt_write_lock(env, stripe, DT_TGT_CHILD);
6779 rc = lod_sub_ref_del(env, stripe, th);
6780 dt_write_unlock(env, stripe);
6784 rc = lod_sub_destroy(env, stripe, th);
6790 struct lod_obj_stripe_cb_data data = { { 0 } };
6792 data.locd_declare = false;
6793 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
6794 rc = lod_obj_for_each_stripe(env, lo, th, &data);
6801 * Implementation of dt_object_operations::do_declare_ref_add.
6803 * \see dt_object_operations::do_declare_ref_add() in the API description
6806 static int lod_declare_ref_add(const struct lu_env *env,
6807 struct dt_object *dt, struct thandle *th)
6809 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
6813 * Implementation of dt_object_operations::do_ref_add.
6815 * \see dt_object_operations::do_ref_add() in the API description for details.
6817 static int lod_ref_add(const struct lu_env *env,
6818 struct dt_object *dt, struct thandle *th)
6820 return lod_sub_ref_add(env, dt_object_child(dt), th);
6824 * Implementation of dt_object_operations::do_declare_ref_del.
6826 * \see dt_object_operations::do_declare_ref_del() in the API description
6829 static int lod_declare_ref_del(const struct lu_env *env,
6830 struct dt_object *dt, struct thandle *th)
6832 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
6836 * Implementation of dt_object_operations::do_ref_del
6838 * \see dt_object_operations::do_ref_del() in the API description for details.
6840 static int lod_ref_del(const struct lu_env *env,
6841 struct dt_object *dt, struct thandle *th)
6843 return lod_sub_ref_del(env, dt_object_child(dt), th);
6847 * Implementation of dt_object_operations::do_object_sync.
6849 * \see dt_object_operations::do_object_sync() in the API description
6852 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
6853 __u64 start, __u64 end)
6855 return dt_object_sync(env, dt_object_child(dt), start, end);
6859 * Implementation of dt_object_operations::do_object_unlock.
6861 * Used to release LDLM lock(s).
6863 * \see dt_object_operations::do_object_unlock() in the API description
6866 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
6867 struct ldlm_enqueue_info *einfo,
6868 union ldlm_policy_data *policy)
6870 struct lod_object *lo = lod_dt_obj(dt);
6871 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
6872 int slave_locks_size;
6876 if (slave_locks == NULL)
6879 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
6880 /* Note: for remote lock for single stripe dir, MDT will cancel
6881 * the lock by lockh directly */
6882 LASSERT(!dt_object_remote(dt_object_child(dt)));
6884 /* locks were unlocked in MDT layer */
6885 for (i = 0; i < slave_locks->ha_count; i++)
6886 LASSERT(!lustre_handle_is_used(&slave_locks->ha_handles[i]));
6889 * NB, ha_count may not equal to ldo_dir_stripe_count, because dir
6890 * layout may change, e.g., shrink dir layout after migration.
6892 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6893 if (lo->ldo_stripe[i])
6894 dt_invalidate(env, lo->ldo_stripe[i]);
6897 slave_locks_size = offsetof(typeof(*slave_locks),
6898 ha_handles[slave_locks->ha_count]);
6899 OBD_FREE(slave_locks, slave_locks_size);
6900 einfo->ei_cbdata = NULL;
6906 * Implementation of dt_object_operations::do_object_lock.
6908 * Used to get LDLM lock on the non-striped and striped objects.
6910 * \see dt_object_operations::do_object_lock() in the API description
6913 static int lod_object_lock(const struct lu_env *env,
6914 struct dt_object *dt,
6915 struct lustre_handle *lh,
6916 struct ldlm_enqueue_info *einfo,
6917 union ldlm_policy_data *policy)
6919 struct lod_object *lo = lod_dt_obj(dt);
6920 int slave_locks_size;
6921 struct lustre_handle_array *slave_locks = NULL;
6926 /* remote object lock */
6927 if (!einfo->ei_enq_slave) {
6928 LASSERT(dt_object_remote(dt));
6929 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
6933 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
6936 rc = lod_striping_load(env, lo);
6941 if (lo->ldo_dir_stripe_count <= 1)
6944 slave_locks_size = offsetof(typeof(*slave_locks),
6945 ha_handles[lo->ldo_dir_stripe_count]);
6946 /* Freed in lod_object_unlock */
6947 OBD_ALLOC(slave_locks, slave_locks_size);
6950 slave_locks->ha_count = lo->ldo_dir_stripe_count;
6952 /* striped directory lock */
6953 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6954 struct lustre_handle lockh;
6955 struct ldlm_res_id *res_id;
6956 struct dt_object *stripe;
6958 stripe = lo->ldo_stripe[i];
6962 res_id = &lod_env_info(env)->lti_res_id;
6963 fid_build_reg_res_name(lu_object_fid(&stripe->do_lu), res_id);
6964 einfo->ei_res_id = res_id;
6966 if (dt_object_remote(stripe)) {
6967 set_bit(i, (void *)slave_locks->ha_map);
6968 rc = dt_object_lock(env, stripe, &lockh, einfo, policy);
6970 struct ldlm_namespace *ns = einfo->ei_namespace;
6971 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
6972 ldlm_completion_callback completion = einfo->ei_cb_cp;
6973 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
6975 LASSERT(ns != NULL);
6976 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS,
6977 policy, einfo->ei_mode,
6978 &dlmflags, blocking,
6980 NULL, 0, LVB_T_NONE,
6985 ldlm_lock_decref_and_cancel(
6986 &slave_locks->ha_handles[i],
6988 OBD_FREE(slave_locks, slave_locks_size);
6991 slave_locks->ha_handles[i] = lockh;
6993 einfo->ei_cbdata = slave_locks;
6999 * Implementation of dt_object_operations::do_invalidate.
7001 * \see dt_object_operations::do_invalidate() in the API description for details
7003 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
7005 return dt_invalidate(env, dt_object_child(dt));
7008 static int lod_declare_instantiate_components(const struct lu_env *env,
7009 struct lod_object *lo,
7013 struct lod_thread_info *info = lod_env_info(env);
7018 LASSERT(info->lti_count < lo->ldo_comp_cnt);
7020 for (i = 0; i < info->lti_count; i++) {
7021 rc = lod_qos_prep_create(env, lo, NULL, th,
7022 info->lti_comp_idx[i], reserve);
7028 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
7029 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
7030 &info->lti_buf, XATTR_NAME_LOV, 0, th);
7037 * Check OSTs for an existing component for further extension
7039 * Checks if OSTs are still healthy and not out of space. Gets free space
7040 * on OSTs (relative to allocation watermark rmb_low) and compares to
7041 * the proposed new_end for this component.
7043 * Decides whether or not to extend a component on its current OSTs.
7045 * \param[in] env execution environment for this thread
7046 * \param[in] lo object we're checking
7047 * \param[in] index index of this component
7048 * \param[in] extension_size extension size for this component
7049 * \param[in] extent layout extent for requested operation
7050 * \param[in] comp_extent extension component extent
7051 * \param[in] write if this is write operation
7053 * \retval true - OK to extend on current OSTs
7054 * \retval false - do not extend on current OSTs
7056 static bool lod_sel_osts_allowed(const struct lu_env *env,
7057 struct lod_object *lo,
7058 int index, __u64 reserve,
7059 struct lu_extent *extent,
7060 struct lu_extent *comp_extent, int write)
7062 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[index];
7063 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7064 struct lod_thread_info *tinfo = lod_env_info(env);
7065 struct obd_statfs *sfs = &tinfo->lti_osfs;
7066 __u64 available = 0;
7072 LASSERT(lod_comp->llc_stripe_count != 0);
7074 lod_getref(&lod->lod_ost_descs);
7075 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
7076 int index = lod_comp->llc_ost_indices[i];
7077 struct lod_tgt_desc *ost = OST_TGT(lod, index);
7078 struct obd_statfs_info info = { 0 };
7079 int j, repeated = 0;
7083 /* Get the number of times this OST repeats in this component.
7084 * Note: inter-component repeats are not counted as this is
7085 * considered as a rare case: we try to not repeat OST in other
7086 * components if possible. */
7087 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
7088 if (index != lod_comp->llc_ost_indices[j])
7091 /* already handled */
7097 if (j < lod_comp->llc_stripe_count)
7100 if (!test_bit(index, lod->lod_ost_bitmap)) {
7101 CDEBUG(D_LAYOUT, "ost %d no longer present\n", index);
7106 rc = dt_statfs_info(env, ost->ltd_tgt, sfs, &info);
7108 CDEBUG(D_LAYOUT, "statfs failed for ost %d, error %d\n",
7114 if (sfs->os_state & OS_STATFS_ENOSPC ||
7115 sfs->os_state & OS_STATFS_READONLY ||
7116 sfs->os_state & OS_STATFS_NOCREATE ||
7117 sfs->os_state & OS_STATFS_DEGRADED) {
7119 "OST%04x unusable for SEL extension, state %x\n",
7120 index, sfs->os_state);
7126 available = sfs->os_bavail * sfs->os_bsize;
7127 /* 'available' is relative to the allocation threshold */
7128 available -= (__u64) info.os_reserved_mb_low << 20;
7130 CDEBUG(D_LAYOUT, "ost %d lowwm: %d highwm: %d, "
7131 "%llu %% blocks available, %llu %% blocks free\n",
7132 index, info.os_reserved_mb_low, info.os_reserved_mb_high,
7133 (100ull * sfs->os_bavail) / sfs->os_blocks,
7134 (100ull * sfs->os_bfree) / sfs->os_blocks);
7136 if (reserve * repeated > available) {
7138 CDEBUG(D_LAYOUT, "low space on ost %d, available %llu "
7139 "< extension size %llu repeated %d\n", index,
7140 available, reserve, repeated);
7144 lod_putref(lod, &lod->lod_ost_descs);
7150 * Adjust extents after component removal
7152 * When we remove an extension component, we move the start of the next
7153 * component to match the start of the extension component, so no space is left
7156 * \param[in] env execution environment for this thread
7157 * \param[in] lo object
7158 * \param[in] max_comp layout component
7159 * \param[in] index index of this component
7161 * \retval 0 on success
7162 * \retval negative errno on error
7164 static void lod_sel_adjust_extents(const struct lu_env *env,
7165 struct lod_object *lo,
7166 int max_comp, int index)
7168 struct lod_layout_component *lod_comp = NULL;
7169 struct lod_layout_component *next = NULL;
7170 struct lod_layout_component *prev = NULL;
7171 __u64 new_start = 0;
7175 /* Extension space component */
7176 lod_comp = &lo->ldo_comp_entries[index];
7177 next = &lo->ldo_comp_entries[index + 1];
7178 prev = &lo->ldo_comp_entries[index - 1];
7180 LASSERT(lod_comp != NULL && prev != NULL && next != NULL);
7181 LASSERT(lod_comp->llc_flags & LCME_FL_EXTENSION);
7183 /* Previous is being removed */
7184 if (prev && prev->llc_id == LCME_ID_INVAL)
7185 new_start = prev->llc_extent.e_start;
7187 new_start = lod_comp->llc_extent.e_start;
7189 for (i = index + 1; i < max_comp; i++) {
7190 lod_comp = &lo->ldo_comp_entries[i];
7192 start = lod_comp->llc_extent.e_start;
7193 lod_comp->llc_extent.e_start = new_start;
7195 /* We only move zero length extendable components */
7196 if (!(start == lod_comp->llc_extent.e_end))
7199 LASSERT(!(lod_comp->llc_flags & LCME_FL_INIT));
7201 lod_comp->llc_extent.e_end = new_start;
7205 /* Calculate the proposed 'new end' for a component we're extending */
7206 static __u64 lod_extension_new_end(__u64 extension_size, __u64 extent_end,
7207 __u32 stripe_size, __u64 component_end,
7208 __u64 extension_end)
7212 LASSERT(extension_size != 0 && stripe_size != 0);
7214 /* Round up to extension size */
7215 if (extent_end == OBD_OBJECT_EOF) {
7216 new_end = OBD_OBJECT_EOF;
7218 /* Add at least extension_size to the previous component_end,
7219 * covering the req layout extent */
7220 new_end = max(extent_end - component_end, extension_size);
7221 new_end = roundup(new_end, extension_size);
7222 new_end += component_end;
7224 /* Component end must be min stripe size aligned */
7225 if (new_end % stripe_size) {
7226 CDEBUG(D_LAYOUT, "new component end is not aligned "
7227 "by the stripe size %u: [%llu, %llu) ext size "
7228 "%llu new end %llu, aligning\n",
7229 stripe_size, component_end, extent_end,
7230 extension_size, new_end);
7231 new_end = roundup(new_end, stripe_size);
7235 if (new_end < extent_end)
7236 new_end = OBD_OBJECT_EOF;
7239 /* Don't extend past the end of the extension component */
7240 if (new_end > extension_end)
7241 new_end = extension_end;
7247 * Calculate the exact reservation (per-OST extension_size) on the OSTs being
7248 * instantiated. It needs to be calculated in advance and taken into account at
7249 * the instantiation time, because otherwise lod_statfs_and_check() may consider
7250 * an OST as OK, but SEL needs its extension_size to fit the free space and the
7251 * OST may turn out to be low-on-space, thus inappropriate OST may be used and
7254 * \param[in] lod_comp lod component we are checking
7256 * \retval size to reserved on each OST of lod_comp's stripe.
7258 static __u64 lod_sel_stripe_reserved(struct lod_layout_component *lod_comp)
7260 /* extension_size is file level, so we must divide by stripe count to
7261 * compare it to available space on a single OST */
7262 return lod_comp->llc_stripe_size * SEL_UNIT_SIZE /
7263 lod_comp->llc_stripe_count;
7266 /* As lod_sel_handler() could be re-entered for the same component several
7267 * times, this is the data for the next call. Fields could be changed to
7268 * component indexes when needed, (e.g. if there is no need to instantiate
7269 * all the previous components up to the current position) to tell the caller
7270 * where to start over from. */
7277 * Process extent updates for a particular layout component
7279 * Handle layout updates for a particular extension space component touched by
7280 * a layout update operation. Core function of self-extending PFL feature.
7282 * In general, this function processes exactly *one* stage of an extension
7283 * operation, modifying the layout accordingly, then returns to the caller.
7284 * The caller is responsible for restarting processing with the new layout,
7285 * which may repeatedly return to this function until the extension updates
7288 * This function does one of a few things to the layout:
7289 * 1. Extends the component before the current extension space component to
7290 * allow it to accomodate the requested operation (if space/policy permit that
7291 * component to continue on its current OSTs)
7293 * 2. If extension of the existing component fails, we do one of two things:
7294 * a. If there is a component after the extension space, we remove the
7295 * extension space component, move the start of the next component down
7296 * accordingly, then notify the caller to restart processing w/the new
7298 * b. If there is no following component, we try repeating the current
7299 * component, creating a new component using the current one as a
7300 * template (keeping its stripe properties but not specific striping),
7301 * and try assigning striping for this component. If there is sufficient
7302 * free space on the OSTs chosen for this component, it is instantiated
7303 * and i/o continues there.
7305 * If there is not sufficient space on the new OSTs, we remove this new
7306 * component & extend the current component.
7308 * Note further that uninited components followed by extension space can be zero
7309 * length meaning that we will try to extend them before initializing them, and
7310 * if that fails, they will be removed without initialization.
7312 * 3. If we extend to/beyond the end of an extension space component, that
7313 * component is exhausted (all of its range has been given to real components),
7314 * so we remove it and restart processing.
7316 * \param[in] env execution environment for this thread
7317 * \param[in,out] lo object to update the layout of
7318 * \param[in] extent layout extent for requested operation, update
7319 * layout to fit this operation
7320 * \param[in] th transaction handle for this operation
7321 * \param[in,out] max_comp the highest comp for the portion of the layout
7322 * we are operating on (For FLR, the chosen
7323 * replica). Updated because we may remove
7325 * \param[in] index index of the extension space component we're
7327 * \param[in] write if this is write op
7328 * \param[in,out] force if the extension is to be forced; set here
7329 to force it on the 2nd call for the same
7332 * \retval 0 on success
7333 * \retval negative errno on error
7335 static int lod_sel_handler(const struct lu_env *env,
7336 struct lod_object *lo,
7337 struct lu_extent *extent,
7338 struct thandle *th, int *max_comp,
7339 int index, int write,
7340 struct sel_data *sd)
7342 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7343 struct lod_thread_info *info = lod_env_info(env);
7344 struct lod_layout_component *lod_comp;
7345 struct lod_layout_component *prev;
7346 struct lod_layout_component *next = NULL;
7347 __u64 extension_size, reserve;
7354 /* First component cannot be extension space */
7356 CERROR("%s: "DFID" first component cannot be extension space\n",
7357 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
7361 lod_comp = &lo->ldo_comp_entries[index];
7362 prev = &lo->ldo_comp_entries[index - 1];
7363 if ((index + 1) < *max_comp)
7364 next = &lo->ldo_comp_entries[index + 1];
7366 /* extension size uses the stripe size field as KiB */
7367 extension_size = lod_comp->llc_stripe_size * SEL_UNIT_SIZE;
7369 CDEBUG(D_LAYOUT, "prev start %llu, extension start %llu, extension end"
7370 " %llu, extension size %llu\n", prev->llc_extent.e_start,
7371 lod_comp->llc_extent.e_start, lod_comp->llc_extent.e_end,
7374 /* Two extension space components cannot be adjacent & extension space
7375 * components cannot be init */
7376 if ((prev->llc_flags & LCME_FL_EXTENSION) ||
7377 !(ergo(next, !(next->llc_flags & LCME_FL_EXTENSION))) ||
7378 lod_comp_inited(lod_comp)) {
7379 CERROR("%s: "DFID" invalid extension space components\n",
7380 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
7384 reserve = lod_sel_stripe_reserved(lod_comp);
7386 if (!prev->llc_stripe) {
7387 CDEBUG(D_LAYOUT, "Previous component not inited\n");
7388 info->lti_count = 1;
7389 info->lti_comp_idx[0] = index - 1;
7390 rc = lod_declare_instantiate_components(env, lo, th, reserve);
7391 /* ENOSPC tells us we can't use this component. If there is
7392 * a next or we are repeating, we either spill over (next) or
7393 * extend the original comp (repeat). Otherwise, return the
7394 * error to the user. */
7395 if (rc == -ENOSPC && (next || sd->sd_repeat))
7401 if (sd->sd_force == 0 && rc == 0)
7402 rc = !lod_sel_osts_allowed(env, lo, index - 1, reserve, extent,
7403 &lod_comp->llc_extent, write);
7405 repeated = !!(sd->sd_repeat);
7409 /* Extend previous component */
7411 new_end = lod_extension_new_end(extension_size, extent->e_end,
7412 prev->llc_stripe_size,
7413 prev->llc_extent.e_end,
7414 lod_comp->llc_extent.e_end);
7416 CDEBUG(D_LAYOUT, "new end %llu\n", new_end);
7417 lod_comp->llc_extent.e_start = new_end;
7418 prev->llc_extent.e_end = new_end;
7420 if (prev->llc_extent.e_end == lod_comp->llc_extent.e_end) {
7421 CDEBUG(D_LAYOUT, "Extension component exhausted\n");
7422 lod_comp->llc_id = LCME_ID_INVAL;
7426 /* rc == 1, failed to extend current component */
7429 /* Normal 'spillover' case - Remove the extension
7430 * space component & bring down the start of the next
7432 lod_comp->llc_id = LCME_ID_INVAL;
7434 if (!(prev->llc_flags & LCME_FL_INIT)) {
7435 prev->llc_id = LCME_ID_INVAL;
7438 lod_sel_adjust_extents(env, lo, *max_comp, index);
7439 } else if (lod_comp_inited(prev)) {
7440 /* If there is no next, and the previous component is
7441 * INIT'ed, try repeating the previous component. */
7442 LASSERT(repeated == 0);
7443 rc = lod_layout_repeat_comp(env, lo, index - 1);
7447 /* The previous component is a repeated component.
7448 * Record this so we don't keep trying to repeat it. */
7451 /* If the previous component is not INIT'ed, this may
7452 * be a component we have just instantiated but failed
7453 * to extend. Or even a repeated component we failed
7454 * to prepare a striping for. Do not repeat but instead
7455 * remove the repeated component & force the extention
7456 * of the original one */
7459 prev->llc_id = LCME_ID_INVAL;
7466 rc = lod_layout_del_prep_layout(env, lo, NULL);
7469 LASSERTF(-rc == change,
7470 "number deleted %d != requested %d\n", -rc,
7473 *max_comp = *max_comp + change;
7475 /* lod_del_prep_layout reallocates ldo_comp_entries, so we must
7476 * refresh these pointers before using them */
7477 lod_comp = &lo->ldo_comp_entries[index];
7478 prev = &lo->ldo_comp_entries[index - 1];
7479 CDEBUG(D_LAYOUT, "After extent updates: prev start %llu, current start "
7480 "%llu, current end %llu max_comp %d ldo_comp_cnt %d\n",
7481 prev->llc_extent.e_start, lod_comp->llc_extent.e_start,
7482 lod_comp->llc_extent.e_end, *max_comp, lo->ldo_comp_cnt);
7484 /* Layout changed successfully */
7489 * Declare layout extent updates
7491 * Handles extensions. Identifies extension components touched by current
7492 * operation and passes them to processing function.
7494 * Restarts with updated layouts from the processing function until the current
7495 * operation no longer touches an extension space component.
7497 * \param[in] env execution environment for this thread
7498 * \param[in,out] lo object to update the layout of
7499 * \param[in] extent layout extent for requested operation, update layout to
7500 * fit this operation
7501 * \param[in] th transaction handle for this operation
7502 * \param[in] pick identifies chosen mirror for FLR layouts
7503 * \param[in] write if this is write op
7505 * \retval 1 on layout changed, 0 on no change
7506 * \retval negative errno on error
7508 static int lod_declare_update_extents(const struct lu_env *env,
7509 struct lod_object *lo, struct lu_extent *extent,
7510 struct thandle *th, int pick, int write)
7512 struct lod_thread_info *info = lod_env_info(env);
7513 struct lod_layout_component *lod_comp;
7514 bool layout_changed = false;
7515 struct sel_data sd = { 0 };
7523 /* This makes us work on the components of the chosen mirror */
7524 if (lo->ldo_mirrors) {
7525 start_index = lo->ldo_mirrors[pick].lme_start;
7526 max_comp = lo->ldo_mirrors[pick].lme_end + 1;
7529 max_comp = lo->ldo_comp_cnt;
7531 if (lo->ldo_flr_state == LCM_FL_NONE)
7532 LASSERT(start_index == 0 && max_comp == lo->ldo_comp_cnt);
7534 CDEBUG(D_LAYOUT, "extent->e_start %llu, extent->e_end %llu\n",
7535 extent->e_start, extent->e_end);
7536 for (i = start_index; i < max_comp; i++) {
7537 lod_comp = &lo->ldo_comp_entries[i];
7539 /* We've passed all components of interest */
7540 if (lod_comp->llc_extent.e_start >= extent->e_end)
7543 if (lod_comp->llc_flags & LCME_FL_EXTENSION) {
7544 layout_changed = true;
7545 rc = lod_sel_handler(env, lo, extent, th, &max_comp,
7550 /* Nothing has changed behind the prev one */
7556 /* We may have added or removed components. If so, we must update the
7557 * start & ends of all the mirrors after the current one, and the end
7558 * of the current mirror. */
7559 if (lo->ldo_mirrors) {
7560 change = max_comp - 1 - lo->ldo_mirrors[pick].lme_end;
7562 lo->ldo_mirrors[pick].lme_end += change;
7563 for (i = pick + 1; i < lo->ldo_mirror_count; i++) {
7564 lo->ldo_mirrors[i].lme_start += change;
7565 lo->ldo_mirrors[i].lme_end += change;
7572 /* The amount of components has changed, adjust the lti_comp_idx */
7573 rc2 = lod_layout_data_init(info, lo->ldo_comp_cnt);
7575 return rc < 0 ? rc : rc2 < 0 ? rc2 : layout_changed;
7578 /* If striping is already instantiated or INIT'ed DOM? */
7579 static bool lod_is_instantiation_needed(struct lod_layout_component *comp)
7581 if (comp->llc_magic == LOV_MAGIC_FOREIGN)
7584 return !(((lov_pattern(comp->llc_pattern) & LOV_PATTERN_MDT) &&
7585 lod_comp_inited(comp)) || comp->llc_stripe);
7589 * Declare layout update for a non-FLR layout.
7591 * \param[in] env execution environment for this thread
7592 * \param[in,out] lo object to update the layout of
7593 * \param[in] layout layout intent for requested operation, "update" is
7594 * a process of reacting to this
7595 * \param[in] buf buffer containing lov ea (see comment on usage inline)
7596 * \param[in] th transaction handle for this operation
7598 * \retval 0 on success
7599 * \retval negative errno on error
7601 static int lod_declare_update_plain(const struct lu_env *env,
7602 struct lod_object *lo, struct layout_intent *layout,
7603 const struct lu_buf *buf, struct thandle *th)
7605 struct lod_thread_info *info = lod_env_info(env);
7606 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7607 struct lod_layout_component *lod_comp;
7608 struct lov_comp_md_v1 *comp_v1 = NULL;
7609 bool layout_changed = false;
7610 bool replay = false;
7614 LASSERT(lo->ldo_flr_state == LCM_FL_NONE);
7617 * In case the client is passing lovea, which only happens during
7618 * the replay of layout intent write RPC for now, we may need to
7619 * parse the lovea and apply new layout configuration.
7621 if (buf && buf->lb_len) {
7622 struct lov_user_md_v1 *v1 = buf->lb_buf;
7624 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
7625 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
7626 LOV_MAGIC_COMP_V1)) {
7627 CERROR("%s: the replay buffer of layout extend "
7628 "(magic %#x) does not contain expected "
7629 "composite layout.\n",
7630 lod2obd(d)->obd_name, v1->lmm_magic);
7631 GOTO(out, rc = -EINVAL);
7634 rc = lod_use_defined_striping(env, lo, buf);
7637 lo->ldo_comp_cached = 1;
7639 rc = lod_get_lov_ea(env, lo);
7642 /* old on-disk EA is stored in info->lti_buf */
7643 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
7645 layout_changed = true;
7647 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
7651 /* non replay path */
7652 rc = lod_striping_load(env, lo);
7657 /* Make sure defined layout covers the requested write range. */
7658 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
7659 if (lo->ldo_comp_cnt > 1 &&
7660 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
7661 lod_comp->llc_extent.e_end < layout->lai_extent.e_end) {
7662 CDEBUG_LIMIT(replay ? D_ERROR : D_LAYOUT,
7663 "%s: the defined layout [0, %#llx) does not "
7664 "covers the write range "DEXT"\n",
7665 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
7666 PEXT(&layout->lai_extent));
7667 GOTO(out, rc = -EINVAL);
7670 CDEBUG(D_LAYOUT, "%s: "DFID": update components "DEXT"\n",
7671 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
7672 PEXT(&layout->lai_extent));
7675 rc = lod_declare_update_extents(env, lo, &layout->lai_extent,
7676 th, 0, layout->lai_opc == LAYOUT_INTENT_WRITE);
7680 layout_changed = true;
7684 * Iterate ld->ldo_comp_entries, find the component whose extent under
7685 * the write range and not instantianted.
7687 for (i = 0; i < lo->ldo_comp_cnt; i++) {
7688 lod_comp = &lo->ldo_comp_entries[i];
7690 if (lod_comp->llc_extent.e_start >= layout->lai_extent.e_end)
7694 /* If striping is instantiated or INIT'ed DOM skip */
7695 if (!lod_is_instantiation_needed(lod_comp))
7699 * In replay path, lod_comp is the EA passed by
7700 * client replay buffer, comp_v1 is the pre-recovery
7701 * on-disk EA, we'd sift out those components which
7702 * were init-ed in the on-disk EA.
7704 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
7709 * this component hasn't instantiated in normal path, or during
7710 * replay it needs replay the instantiation.
7713 /* A released component is being extended */
7714 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
7715 GOTO(out, rc = -EINVAL);
7717 LASSERT(info->lti_comp_idx != NULL);
7718 info->lti_comp_idx[info->lti_count++] = i;
7719 layout_changed = true;
7722 if (!layout_changed)
7725 lod_obj_inc_layout_gen(lo);
7726 rc = lod_declare_instantiate_components(env, lo, th, 0);
7730 lod_striping_free(env, lo);
7734 static inline int lod_comp_index(struct lod_object *lo,
7735 struct lod_layout_component *lod_comp)
7737 LASSERT(lod_comp >= lo->ldo_comp_entries &&
7738 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
7740 return lod_comp - lo->ldo_comp_entries;
7744 * Stale other mirrors by writing extent.
7746 static int lod_stale_components(const struct lu_env *env, struct lod_object *lo,
7747 int primary, struct lu_extent *extent,
7750 struct lod_layout_component *pri_comp, *lod_comp;
7751 struct lu_extent pri_extent;
7756 /* The writing extent decides which components in the primary
7757 * are affected... */
7758 CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
7761 lod_foreach_mirror_comp(pri_comp, lo, primary) {
7762 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
7765 CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
7766 lod_comp_index(lo, pri_comp),
7767 PEXT(&pri_comp->llc_extent));
7769 pri_extent.e_start = pri_comp->llc_extent.e_start;
7770 pri_extent.e_end = pri_comp->llc_extent.e_end;
7772 for (i = 0; i < lo->ldo_mirror_count; i++) {
7776 rc = lod_declare_update_extents(env, lo, &pri_extent,
7778 /* if update_extents changed the layout, it may have
7779 * reallocated the component array, so start over to
7780 * avoid using stale pointers */
7786 /* ... and then stale other components that are
7787 * overlapping with primary components */
7788 lod_foreach_mirror_comp(lod_comp, lo, i) {
7789 if (!lu_extent_is_overlapped(
7791 &lod_comp->llc_extent))
7794 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
7795 i, lod_comp_index(lo, lod_comp));
7797 lod_comp->llc_flags |= LCME_FL_STALE;
7798 lo->ldo_mirrors[i].lme_stale = 1;
7799 if (lod_is_hsm(lod_comp))
7800 lod_comp->llc_foreign_flags |= HS_DIRTY;
7809 * check an OST's availability
7810 * \param[in] env execution environment
7811 * \param[in] lo lod object
7812 * \param[in] dt dt object
7813 * \param[in] index mirror index
7815 * \retval negative if failed
7816 * \retval 1 if \a dt is available
7817 * \retval 0 if \a dt is not available
7819 static inline int lod_check_ost_avail(const struct lu_env *env,
7820 struct lod_object *lo,
7821 struct dt_object *dt, int index)
7823 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7824 struct lod_tgt_desc *ost;
7826 int type = LU_SEQ_RANGE_OST;
7829 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &idx, &type);
7831 CERROR("%s: can't locate "DFID":rc = %d\n",
7832 lod2obd(lod)->obd_name, PFID(lu_object_fid(&dt->do_lu)),
7837 ost = OST_TGT(lod, idx);
7838 if (ost->ltd_active == 0) {
7839 CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail\n",
7840 PFID(lod_object_fid(lo)), index, idx);
7848 * Pick primary mirror for write
7849 * \param[in] env execution environment
7850 * \param[in] lo object
7851 * \param[in] extent write range
7853 static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo,
7854 struct lu_extent *extent)
7856 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7857 unsigned int seq = 0;
7858 struct lod_layout_component *lod_comp;
7860 int picked = -1, second_pick = -1, third_pick = -1;
7863 if (CFS_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
7864 get_random_bytes(&seq, sizeof(seq));
7865 seq %= lo->ldo_mirror_count;
7869 * Pick a mirror as the primary, and check the availability of OSTs.
7871 * This algo can be revised later after knowing the topology of
7874 lod_qos_statfs_update(env, lod, &lod->lod_ost_descs);
7876 rc = lod_fill_mirrors(lo);
7880 for (i = 0; i < lo->ldo_mirror_count; i++) {
7881 bool ost_avail = true;
7882 int index = (i + seq) % lo->ldo_mirror_count;
7884 if (lo->ldo_mirrors[index].lme_stale) {
7885 CDEBUG(D_LAYOUT, DFID": mirror %d stale\n",
7886 PFID(lod_object_fid(lo)), index);
7890 /* 2nd pick is for the primary mirror containing unavail OST */
7891 if (lo->ldo_mirrors[index].lme_prefer && second_pick < 0)
7892 second_pick = index;
7894 /* 3rd pick is for non-primary mirror containing unavail OST */
7895 if (second_pick < 0 && third_pick < 0)
7899 * we found a non-primary 1st pick, we'd like to find a
7900 * potential pirmary mirror.
7902 if (picked >= 0 && !lo->ldo_mirrors[index].lme_prefer)
7905 /* check the availability of OSTs */
7906 lod_foreach_mirror_comp(lod_comp, lo, index) {
7907 if (!lod_comp_inited(lod_comp) || !lod_comp->llc_stripe)
7910 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
7911 struct dt_object *dt = lod_comp->llc_stripe[j];
7913 rc = lod_check_ost_avail(env, lo, dt, index);
7920 } /* for all dt object in one component */
7923 } /* for all components in a mirror */
7926 * the OSTs where allocated objects locates in the components
7927 * of the mirror are available.
7932 /* this mirror has all OSTs available */
7936 * primary with all OSTs are available, this is the perfect
7939 if (lo->ldo_mirrors[index].lme_prefer)
7941 } /* for all mirrors */
7943 /* failed to pick a sound mirror, lower our expectation */
7945 picked = second_pick;
7947 picked = third_pick;
7954 static int lod_prepare_resync_mirror(const struct lu_env *env,
7955 struct lod_object *lo,
7958 struct lod_thread_info *info = lod_env_info(env);
7959 struct lod_layout_component *lod_comp;
7960 bool neg = !!(MIRROR_ID_NEG & mirror_id);
7963 mirror_id &= ~MIRROR_ID_NEG;
7965 for (i = 0; i < lo->ldo_mirror_count; i++) {
7966 if ((!neg && lo->ldo_mirrors[i].lme_id != mirror_id) ||
7967 (neg && lo->ldo_mirrors[i].lme_id == mirror_id))
7970 lod_foreach_mirror_comp(lod_comp, lo, i) {
7971 if (lod_comp_inited(lod_comp))
7974 info->lti_comp_idx[info->lti_count++] =
7975 lod_comp_index(lo, lod_comp);
7983 * figure out the components should be instantiated for resync.
7985 static int lod_prepare_resync(const struct lu_env *env, struct lod_object *lo,
7986 struct lu_extent *extent)
7988 struct lod_thread_info *info = lod_env_info(env);
7989 struct lod_layout_component *lod_comp;
7990 unsigned int need_sync = 0;
7994 DFID": instantiate all stale components in "DEXT"\n",
7995 PFID(lod_object_fid(lo)), PEXT(extent));
7998 * instantiate all components within this extent, even non-stale
8001 for (i = 0; i < lo->ldo_mirror_count; i++) {
8002 if (!lo->ldo_mirrors[i].lme_stale)
8005 lod_foreach_mirror_comp(lod_comp, lo, i) {
8006 if (!lu_extent_is_overlapped(extent,
8007 &lod_comp->llc_extent))
8012 if (lod_comp_inited(lod_comp))
8015 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
8016 i, lod_comp_index(lo, lod_comp));
8017 info->lti_comp_idx[info->lti_count++] =
8018 lod_comp_index(lo, lod_comp);
8022 return need_sync ? 0 : -EALREADY;
8025 static struct lod_layout_component *
8026 lod_locate_comp_hsm(struct lod_object *lo, int *hsm_mirror_id)
8028 struct lod_layout_component *lod_comp = NULL;
8031 if (!lo->ldo_is_composite)
8034 for (i = 0; i < lo->ldo_mirror_count; i++) {
8036 * FIXME: In the current design, there is only one HSM
8037 * mirror component in range [0, EOF] for a FLR file. This
8038 * should be fixed to support multiple HSM mirror components
8039 * with different HSM backend types and partial file ranges
8042 if (lo->ldo_mirrors[i].lme_hsm) {
8048 start_idx = lo->ldo_mirrors[i].lme_start;
8049 end_idx = lo->ldo_mirrors[i].lme_end;
8050 LASSERT(start_idx == end_idx);
8051 lod_comp = &lo->ldo_comp_entries[start_idx];
8052 LASSERT(lo->ldo_is_composite && lod_is_hsm(lod_comp) &&
8053 lod_comp->llc_extent.e_start == 0 &&
8054 lod_comp->llc_extent.e_end == LUSTRE_EOF);
8062 static int lod_declare_pccro_set(const struct lu_env *env,
8063 struct dt_object *dt, struct thandle *th)
8065 struct lod_thread_info *info = lod_env_info(env);
8066 struct lu_buf *buf = &info->lti_buf;
8067 struct lod_object *lo = lod_dt_obj(dt);
8068 struct lod_layout_component *lod_comp;
8069 struct lod_layout_component *comp_array;
8070 struct lod_mirror_entry *mirror_array;
8080 rc = lod_striping_load(env, lo);
8084 if (lo->ldo_flr_state & LCM_FL_PCC_RDONLY)
8087 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
8091 lod_comp = lod_locate_comp_hsm(lo, &hsm_mirror_id);
8093 if (lod_comp->llc_foreign_flags & HS_PCCRO) {
8094 CDEBUG(D_LAYOUT, "bad HSM flags: %#x\n",
8095 lod_comp->llc_foreign_flags);
8099 lod_obj_inc_layout_gen(lo);
8100 lod_comp->llc_foreign_flags |= HS_PCCRO;
8101 lod_comp->llc_foreign_flags &= ~HS_DIRTY;
8102 lod_comp->llc_flags &= ~LCME_FL_STALE;
8103 lo->ldo_mirrors[hsm_mirror_id].lme_stale = 0;
8104 lo->ldo_flr_state |= LCM_FL_PCC_RDONLY;
8105 buf->lb_len = lod_comp_md_size(lo, false);
8106 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
8107 buf, XATTR_NAME_LOV, 0, th);
8112 * Create an new composite layout with only one HSM component.
8113 * Field @lhm_archive_uuid is used to be the identifier within HSM
8114 * backend for the archive copy. In the PCC case with a POSIX archive,
8115 * This can just be the original inode FID. This is important because
8116 * the inode FID may change due to layout swaps or migration to a new
8117 * MDT, and we do not want that to cause problems with finding the copy
8120 mirror_cnt = lo->ldo_mirror_count + 1;
8121 if (!lo->ldo_is_composite) {
8122 LASSERT(lo->ldo_mirror_count == 0);
8126 OBD_ALLOC_PTR_ARRAY(mirror_array, mirror_cnt);
8127 if (mirror_array == NULL)
8130 new_cnt = lo->ldo_comp_cnt + 1;
8131 OBD_ALLOC_PTR_ARRAY(comp_array, new_cnt);
8132 if (comp_array == NULL) {
8133 OBD_FREE_PTR_ARRAY(mirror_array, mirror_cnt);
8138 for (i = 0; i < lo->ldo_comp_cnt; i++) {
8139 lod_comp = &lo->ldo_comp_entries[i];
8142 * Add mirror from a non-flr file, create new mirror ID.
8143 * Otherwise, keep existing mirror's component ID, used
8144 * for mirror extension.
8146 if (lo->ldo_mirror_count == 0 &&
8147 mirror_id_of(lod_comp->llc_id) == 0)
8148 lod_comp->llc_id = pflr_id(1, i + 1);
8150 if (lod_comp->llc_id != LCME_ID_INVAL &&
8151 mirror_id_of(lod_comp->llc_id) > mirror_id)
8152 mirror_id = mirror_id_of(lod_comp->llc_id);
8154 if (!lo->ldo_is_composite) {
8155 lod_comp->llc_extent.e_start = 0;
8156 lod_comp->llc_extent.e_end = LUSTRE_EOF;
8157 lod_comp_set_init(lod_comp);
8161 memcpy(comp_array, lo->ldo_comp_entries,
8162 sizeof(*comp_array) * lo->ldo_comp_cnt);
8164 lod_comp = &comp_array[new_cnt - 1];
8165 lod_comp->llc_magic = LOV_MAGIC_FOREIGN;
8166 lod_comp->llc_extent.e_start = 0;
8167 lod_comp->llc_extent.e_end = LUSTRE_EOF;
8168 lod_comp->llc_length = sizeof(struct lov_hsm_base);
8169 lod_comp->llc_type = LU_FOREIGN_TYPE_PCCRO;
8170 lod_comp->llc_foreign_flags = HS_EXISTS | HS_ARCHIVED | HS_PCCRO;
8171 memset(&lod_comp->llc_hsm, 0, sizeof(lod_comp->llc_hsm));
8173 if (lo->ldo_mirrors)
8174 OBD_FREE_PTR_ARRAY(lo->ldo_mirrors, lo->ldo_mirror_count);
8175 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
8178 * The @ldo_mirror will be refilled by lod_fill_mirrors() when
8179 * call lod_striped_create() for layout change.
8181 lo->ldo_mirrors = mirror_array;
8182 lo->ldo_mirror_count = mirror_cnt;
8183 lo->ldo_comp_entries = comp_array;
8184 lo->ldo_comp_cnt = new_cnt;
8185 lo->ldo_is_composite = 1;
8188 lod_comp->llc_id = LCME_ID_INVAL;
8189 lod_comp->llc_id = lod_gen_component_id(lo, mirror_id, new_cnt - 1);
8191 if (lo->ldo_flr_state == LCM_FL_NONE)
8192 lo->ldo_flr_state = LCM_FL_RDONLY;
8193 lo->ldo_flr_state |= LCM_FL_PCC_RDONLY;
8194 buf->lb_len = lod_comp_md_size(lo, false);
8195 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
8196 buf, XATTR_NAME_LOV, 0, th);
8198 lod_striping_free(env, lo);
8204 * TODO: When clear LCM_FL_PCC_RDONLY flag from the layouts, it means the file
8205 * is going to be modified. Currently it needs two RPCs: first one is to clear
8206 * LCM_FL_PCC_RDONLY flag; the second one is to pick primary mirror and mark
8207 * the file as LCM_FL_WRITE_PENDING.
8208 * These two RPCs can be combined in one RPC call.
8210 static int lod_declare_pccro_clear(const struct lu_env *env,
8211 struct dt_object *dt, struct thandle *th)
8213 struct lod_thread_info *info = lod_env_info(env);
8214 struct lod_object *lo = lod_dt_obj(dt);
8215 struct lod_layout_component *lod_comp;
8220 rc = lod_striping_load(env, lo);
8224 if (!(lo->ldo_flr_state & LCM_FL_PCC_RDONLY))
8227 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
8231 lod_comp = lod_locate_comp_hsm(lo, NULL);
8232 if (lod_comp == NULL) {
8233 CDEBUG(D_LAYOUT, "Not found any HSM component\n");
8234 GOTO(out, rc = -EINVAL);
8237 lod_comp->llc_foreign_flags &= ~HS_PCCRO;
8238 lo->ldo_flr_state &= ~LCM_FL_PCC_RDONLY;
8239 lod_obj_inc_layout_gen(lo);
8240 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
8241 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
8242 &info->lti_buf, XATTR_NAME_LOV, 0, th);
8245 lod_striping_free(env, lo);
8250 static int lod_declare_update_pccro(const struct lu_env *env,
8251 struct dt_object *dt,
8252 struct md_layout_change *mlc,
8255 struct layout_intent *intent = mlc->mlc_intent;
8258 switch (intent->lai_opc) {
8259 case LAYOUT_INTENT_PCCRO_SET:
8260 rc = lod_declare_pccro_set(env, dt, th);
8262 case LAYOUT_INTENT_PCCRO_CLEAR:
8263 rc = lod_declare_pccro_clear(env, dt, th);
8273 static int lod_declare_update_rdonly(const struct lu_env *env,
8274 struct lod_object *lo, struct md_layout_change *mlc,
8277 struct lod_thread_info *info = lod_env_info(env);
8278 struct lu_attr *layout_attr = &info->lti_layout_attr;
8279 struct lod_layout_component *lod_comp;
8280 struct lu_extent extent = { 0 };
8284 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
8285 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
8286 mlc->mlc_opc == MD_LAYOUT_RESYNC);
8287 LASSERT(lo->ldo_mirror_count > 0);
8289 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
8290 struct layout_intent *layout = mlc->mlc_intent;
8291 int write = layout->lai_opc == LAYOUT_INTENT_WRITE;
8294 extent = layout->lai_extent;
8295 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
8296 PFID(lod_object_fid(lo)), PEXT(&extent));
8298 picked = lod_primary_pick(env, lo, &extent);
8302 CDEBUG(D_LAYOUT, DFID": picked mirror id %u as primary\n",
8303 PFID(lod_object_fid(lo)),
8304 lo->ldo_mirrors[picked].lme_id);
8306 /* Update extents of primary before staling */
8307 rc = lod_declare_update_extents(env, lo, &extent, th, picked,
8312 if (layout->lai_opc == LAYOUT_INTENT_TRUNC) {
8314 * trunc transfers [0, size) in the intent extent, we'd
8315 * stale components overlapping [size, eof).
8317 extent.e_start = extent.e_end;
8318 extent.e_end = OBD_OBJECT_EOF;
8321 /* stale overlapping components from other mirrors */
8322 rc = lod_stale_components(env, lo, picked, &extent, th);
8326 /* restore truncate intent extent */
8327 if (layout->lai_opc == LAYOUT_INTENT_TRUNC)
8328 extent.e_end = extent.e_start;
8330 /* instantiate components for the picked mirror, start from 0 */
8333 lod_foreach_mirror_comp(lod_comp, lo, picked) {
8334 if (!lu_extent_is_overlapped(&extent,
8335 &lod_comp->llc_extent))
8338 if (!lod_is_instantiation_needed(lod_comp))
8341 info->lti_comp_idx[info->lti_count++] =
8342 lod_comp_index(lo, lod_comp);
8345 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
8346 } else { /* MD_LAYOUT_RESYNC */
8350 * could contain multiple non-stale mirrors, so we need to
8351 * prep uninited all components assuming any non-stale mirror
8352 * could be picked as the primary mirror.
8354 if (mlc->mlc_mirror_id == 0) {
8356 for (i = 0; i < lo->ldo_mirror_count; i++) {
8357 if (lo->ldo_mirrors[i].lme_stale)
8360 lod_foreach_mirror_comp(lod_comp, lo, i) {
8361 if (!lod_comp_inited(lod_comp))
8365 lod_comp->llc_extent.e_end)
8367 lod_comp->llc_extent.e_end;
8370 rc = lod_prepare_resync(env, lo, &extent);
8374 /* mirror write, try to init its all components */
8375 rc = lod_prepare_resync_mirror(env, lo,
8376 mlc->mlc_mirror_id);
8381 /* change the file state to SYNC_PENDING */
8382 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
8385 /* Reset the layout version once it's becoming too large.
8386 * This way it can make sure that the layout version is
8387 * monotonously increased in this writing era. */
8388 lod_obj_inc_layout_gen(lo);
8390 rc = lod_declare_instantiate_components(env, lo, th, 0);
8394 layout_attr->la_valid = LA_LAYOUT_VERSION;
8395 layout_attr->la_layout_version = 0;
8396 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
8397 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
8398 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
8404 lod_striping_free(env, lo);
8408 static int lod_declare_update_write_pending(const struct lu_env *env,
8409 struct lod_object *lo, struct md_layout_change *mlc,
8412 struct lod_thread_info *info = lod_env_info(env);
8413 struct lu_attr *layout_attr = &info->lti_layout_attr;
8414 struct lod_layout_component *lod_comp;
8415 struct lu_extent extent = { 0 };
8421 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
8422 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
8423 mlc->mlc_opc == MD_LAYOUT_RESYNC);
8425 /* look for the first preferred mirror */
8426 for (i = 0; i < lo->ldo_mirror_count; i++) {
8427 if (lo->ldo_mirrors[i].lme_stale)
8429 if (lo->ldo_mirrors[i].lme_prefer == 0)
8431 if (lo->ldo_mirrors[i].lme_hsm)
8438 /* no primary, use any in-sync */
8439 for (i = 0; i < lo->ldo_mirror_count; i++) {
8440 if (lo->ldo_mirrors[i].lme_stale)
8446 CERROR(DFID ": doesn't have a primary mirror\n",
8447 PFID(lod_object_fid(lo)));
8448 GOTO(out, rc = -ENODATA);
8452 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
8453 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
8455 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
8457 /* for LAYOUT_WRITE opc, it has to do the following operations:
8458 * 1. stale overlapping componets from stale mirrors;
8459 * 2. instantiate components of the primary mirror;
8460 * 3. transfter layout version to all objects of the primary;
8462 * for LAYOUT_RESYNC opc, it will do:
8463 * 1. instantiate components of all stale mirrors;
8464 * 2. transfer layout version to all objects to close write era. */
8466 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
8467 struct layout_intent *layout = mlc->mlc_intent;
8468 int write = layout->lai_opc == LAYOUT_INTENT_WRITE;
8470 LASSERT(layout != NULL);
8472 extent = layout->lai_extent;
8474 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
8475 PFID(lod_object_fid(lo)), PEXT(&extent));
8477 /* 1. Update extents of primary before staling */
8478 rc = lod_declare_update_extents(env, lo, &extent, th, primary,
8483 if (layout->lai_opc == LAYOUT_INTENT_TRUNC) {
8485 * trunc transfers [0, size) in the intent extent, we'd
8486 * stale components overlapping [size, eof).
8488 extent.e_start = extent.e_end;
8489 extent.e_end = OBD_OBJECT_EOF;
8492 /* 2. stale overlapping components */
8493 rc = lod_stale_components(env, lo, primary, &extent, th);
8497 /* 3. find the components which need instantiating.
8498 * instantiate [0, mlc->mlc_intent->e_end) */
8500 /* restore truncate intent extent */
8501 if (layout->lai_opc == LAYOUT_INTENT_TRUNC)
8502 extent.e_end = extent.e_start;
8505 lod_foreach_mirror_comp(lod_comp, lo, primary) {
8506 if (!lu_extent_is_overlapped(&extent,
8507 &lod_comp->llc_extent))
8510 if (!lod_is_instantiation_needed(lod_comp))
8513 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
8514 primary, lod_comp_index(lo, lod_comp));
8515 info->lti_comp_idx[info->lti_count++] =
8516 lod_comp_index(lo, lod_comp);
8518 } else { /* MD_LAYOUT_RESYNC */
8519 if (mlc->mlc_mirror_id == 0) {
8521 lod_foreach_mirror_comp(lod_comp, lo, primary) {
8522 if (!lod_comp_inited(lod_comp))
8525 extent.e_end = lod_comp->llc_extent.e_end;
8528 rc = lod_prepare_resync(env, lo, &extent);
8532 /* mirror write, try to init its all components */
8533 rc = lod_prepare_resync_mirror(env, lo,
8534 mlc->mlc_mirror_id);
8539 /* change the file state to SYNC_PENDING */
8540 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
8543 rc = lod_declare_instantiate_components(env, lo, th, 0);
8547 lod_obj_inc_layout_gen(lo);
8549 /* 3. transfer layout version to OST objects.
8550 * transfer new layout version to OST objects so that stale writes
8551 * can be denied. It also ends an era of writing by setting
8552 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
8553 * send write RPC; only resync RPCs could do it. */
8554 layout_attr->la_valid = LA_LAYOUT_VERSION;
8555 layout_attr->la_layout_version = 0;
8556 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
8557 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
8558 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
8563 lod_striping_free(env, lo);
8567 static int lod_declare_update_sync_pending(const struct lu_env *env,
8568 struct lod_object *lo, struct md_layout_change *mlc,
8571 struct lod_thread_info *info = lod_env_info(env);
8572 struct lu_attr *layout_attr = &info->lti_layout_attr;
8573 unsigned sync_components = 0;
8574 unsigned resync_components = 0;
8579 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
8580 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
8581 mlc->mlc_opc == MD_LAYOUT_WRITE);
8583 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
8584 PFID(lod_object_fid(lo)), mlc->mlc_opc);
8586 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
8587 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
8588 PFID(lod_object_fid(lo)));
8590 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
8591 return lod_declare_update_write_pending(env, lo, mlc, th);
8594 /* MD_LAYOUT_RESYNC_DONE */
8596 for (i = 0; i < lo->ldo_comp_cnt; i++) {
8597 struct lod_layout_component *lod_comp;
8600 lod_comp = &lo->ldo_comp_entries[i];
8602 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
8607 for (j = 0; j < mlc->mlc_resync_count; j++) {
8608 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
8611 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
8612 lod_comp->llc_flags &= ~LCME_FL_STALE;
8613 resync_components++;
8619 for (i = 0; i < mlc->mlc_resync_count; i++) {
8620 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
8623 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
8624 "or already synced\n", PFID(lod_object_fid(lo)),
8625 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
8626 GOTO(out, rc = -EINVAL);
8629 if (!sync_components || (mlc->mlc_resync_count && !resync_components)) {
8630 CDEBUG(D_LAYOUT, DFID": no mirror in sync\n",
8631 PFID(lod_object_fid(lo)));
8633 /* tend to return an error code here to prevent
8634 * the MDT from setting SoM attribute */
8635 GOTO(out, rc = -EINVAL);
8638 CDEBUG(D_LAYOUT, DFID": synced %u resynced %u/%zu components\n",
8639 PFID(lod_object_fid(lo)),
8640 sync_components, resync_components, mlc->mlc_resync_count);
8642 lo->ldo_flr_state = LCM_FL_RDONLY;
8643 lod_obj_inc_layout_gen(lo);
8645 layout_attr->la_valid = LA_LAYOUT_VERSION;
8646 layout_attr->la_layout_version = 0;
8647 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
8651 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
8652 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
8653 &info->lti_buf, XATTR_NAME_LOV, 0, th);
8658 lod_striping_free(env, lo);
8662 typedef int (*mlc_handler)(const struct lu_env *env, struct dt_object *dt,
8663 const struct md_layout_change *mlc,
8664 struct thandle *th);
8667 * Attach stripes after target's for migrating directory. NB, we
8668 * only need to declare this, the actual work is done inside
8669 * lod_xattr_set_lmv().
8671 * \param[in] env execution environment
8672 * \param[in] dt target object
8673 * \param[in] mlc layout change data
8674 * \param[in] th transaction handle
8676 * \retval 0 on success
8677 * \retval negative if failed
8679 static int lod_dir_declare_layout_attach(const struct lu_env *env,
8680 struct dt_object *dt,
8681 const struct md_layout_change *mlc,
8684 struct lod_thread_info *info = lod_env_info(env);
8685 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
8686 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
8687 struct lod_object *lo = lod_dt_obj(dt);
8688 struct dt_object *next = dt_object_child(dt);
8689 struct dt_object_format *dof = &info->lti_format;
8690 struct lmv_mds_md_v1 *lmv = mlc->mlc_buf.lb_buf;
8691 struct dt_object **stripes;
8692 __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
8693 struct lu_fid *fid = &info->lti_fid;
8694 struct lod_tgt_desc *tgt;
8695 struct dt_object *dto;
8696 struct dt_device *tgt_dt;
8697 int type = LU_SEQ_RANGE_ANY;
8698 struct dt_insert_rec *rec = &info->lti_dt_rec;
8699 char *stripe_name = info->lti_key;
8700 struct lu_name *sname;
8701 struct linkea_data ldata = { NULL };
8702 struct lu_buf linkea_buf;
8709 if (!lmv_is_sane(lmv))
8712 if (!dt_try_as_dir(env, dt, false))
8715 dof->dof_type = DFT_DIR;
8717 OBD_ALLOC_PTR_ARRAY(stripes, (lo->ldo_dir_stripe_count + stripe_count));
8721 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
8722 stripes[i] = lo->ldo_stripe[i];
8724 rec->rec_type = S_IFDIR;
8726 for (i = 0; i < stripe_count; i++) {
8728 &lmv->lmv_stripe_fids[i]);
8729 if (!fid_is_sane(fid))
8732 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
8736 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
8737 tgt_dt = lod->lod_child;
8739 tgt = LTD_TGT(ltd, idx);
8741 GOTO(out, rc = -ESTALE);
8742 tgt_dt = tgt->ltd_tgt;
8745 dto = dt_locate_at(env, tgt_dt, fid,
8746 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
8749 GOTO(out, rc = PTR_ERR(dto));
8751 stripes[i + lo->ldo_dir_stripe_count] = dto;
8753 if (!dt_try_as_dir(env, dto, true))
8754 GOTO(out, rc = -ENOTDIR);
8756 rc = lod_sub_declare_ref_add(env, dto, th);
8760 rec->rec_fid = lu_object_fid(&dto->do_lu);
8761 rc = lod_sub_declare_insert(env, dto,
8762 (const struct dt_rec *)rec,
8763 (const struct dt_key *)dot, th);
8767 rc = lod_sub_declare_insert(env, dto,
8768 (const struct dt_rec *)rec,
8769 (const struct dt_key *)dotdot, th);
8773 rc = lod_sub_declare_xattr_set(env, dto, &mlc->mlc_buf,
8774 XATTR_NAME_LMV, 0, th);
8778 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
8779 PFID(lu_object_fid(&dto->do_lu)),
8780 i + lo->ldo_dir_stripe_count);
8782 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
8783 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
8784 sname, lu_object_fid(&dt->do_lu));
8788 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
8789 linkea_buf.lb_len = ldata.ld_leh->leh_len;
8790 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
8791 XATTR_NAME_LINK, 0, th);
8795 rc = lod_sub_declare_insert(env, next,
8796 (const struct dt_rec *)rec,
8797 (const struct dt_key *)stripe_name,
8802 rc = lod_sub_declare_ref_add(env, next, th);
8808 OBD_FREE_PTR_ARRAY(lo->ldo_stripe,
8809 lo->ldo_dir_stripes_allocated);
8810 lo->ldo_stripe = stripes;
8811 lo->ldo_is_foreign = 0;
8812 lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count;
8813 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type);
8814 lo->ldo_dir_stripe_count += stripe_count;
8815 lo->ldo_dir_layout_version++;
8816 lo->ldo_dir_stripes_allocated += stripe_count;
8818 /* plain directory split creates target as a plain directory, while
8819 * after source attached as the first stripe, it becomes a striped
8820 * directory, set correct do_index_ops, otherwise it can't be unlinked.
8822 dt->do_index_ops = &lod_striped_index_ops;
8826 i = lo->ldo_dir_stripe_count;
8827 while (i < lo->ldo_dir_stripe_count + stripe_count && stripes[i])
8828 dt_object_put(env, stripes[i++]);
8830 OBD_FREE_PTR_ARRAY(stripes, stripe_count + lo->ldo_dir_stripe_count);
8834 static int lod_dir_declare_layout_detach(const struct lu_env *env,
8835 struct dt_object *dt,
8836 const struct md_layout_change *unused,
8839 struct lod_thread_info *info = lod_env_info(env);
8840 struct lod_object *lo = lod_dt_obj(dt);
8841 struct dt_object *next = dt_object_child(dt);
8842 char *stripe_name = info->lti_key;
8843 struct dt_object *dto;
8847 if (!dt_try_as_dir(env, dt, true))
8850 if (!lo->ldo_dir_stripe_count)
8851 return lod_sub_declare_delete(env, next,
8852 (const struct dt_key *)dotdot, th);
8854 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8855 dto = lo->ldo_stripe[i];
8859 if (!dt_try_as_dir(env, dto, true))
8862 rc = lod_sub_declare_delete(env, dto,
8863 (const struct dt_key *)dotdot, th);
8867 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8868 PFID(lu_object_fid(&dto->do_lu)), i);
8870 rc = lod_sub_declare_delete(env, next,
8871 (const struct dt_key *)stripe_name, th);
8875 rc = lod_sub_declare_ref_del(env, next, th);
8883 static int dt_dir_is_empty(const struct lu_env *env,
8884 struct dt_object *obj)
8887 const struct dt_it_ops *iops;
8892 if (!dt_try_as_dir(env, obj, true))
8895 iops = &obj->do_index_ops->dio_it;
8896 it = iops->init(env, obj, LUDA_64BITHASH);
8898 RETURN(PTR_ERR(it));
8900 rc = iops->get(env, it, (const struct dt_key *)"");
8904 for (rc = 0, i = 0; rc == 0 && i < 3; ++i)
8905 rc = iops->next(env, it);
8911 /* Huh? Index contains no zero key? */
8916 iops->fini(env, it);
8921 static int lod_dir_declare_layout_shrink(const struct lu_env *env,
8922 struct dt_object *dt,
8923 const struct md_layout_change *mlc,
8926 struct lod_thread_info *info = lod_env_info(env);
8927 struct lod_object *lo = lod_dt_obj(dt);
8928 struct dt_object *next = dt_object_child(dt);
8929 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
8930 char *stripe_name = info->lti_key;
8931 struct lu_buf *lmv_buf = &info->lti_buf;
8932 __u32 final_stripe_count;
8933 struct dt_object *dto;
8939 if (!dt_try_as_dir(env, dt, true))
8942 /* shouldn't be called on plain directory */
8943 LASSERT(lo->ldo_dir_stripe_count);
8945 lmv_buf->lb_buf = &info->lti_lmv.lmv_md_v1;
8946 lmv_buf->lb_len = sizeof(info->lti_lmv.lmv_md_v1);
8948 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
8949 LASSERT(final_stripe_count &&
8950 final_stripe_count < lo->ldo_dir_stripe_count);
8952 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8953 dto = lo->ldo_stripe[i];
8957 if (i < final_stripe_count) {
8958 rc = lod_sub_declare_xattr_set(env, dto, lmv_buf,
8960 LU_XATTR_REPLACE, th);
8967 rc = dt_dir_is_empty(env, dto);
8971 rc = lod_sub_declare_ref_del(env, dto, th);
8975 rc = lod_sub_declare_destroy(env, dto, th);
8979 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8980 PFID(lu_object_fid(&dto->do_lu)), i);
8982 rc = lod_sub_declare_delete(env, next,
8983 (const struct dt_key *)stripe_name, th);
8987 rc = lod_sub_declare_ref_del(env, next, th);
8992 rc = lod_sub_declare_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
8993 LU_XATTR_REPLACE, th);
8998 * Allocate stripes for split directory.
9000 * \param[in] env execution environment
9001 * \param[in] dt target object
9002 * \param[in] mlc layout change data
9003 * \param[in] th transaction handle
9005 * \retval 0 on success
9006 * \retval negative if failed
9008 static int lod_dir_declare_layout_split(const struct lu_env *env,
9009 struct dt_object *dt,
9010 const struct md_layout_change *mlc,
9013 struct lod_thread_info *info = lod_env_info(env);
9014 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
9015 struct lod_object *lo = lod_dt_obj(dt);
9016 struct dt_object_format *dof = &info->lti_format;
9017 struct lmv_user_md_v1 *lum = mlc->mlc_spec->u.sp_ea.eadata;
9018 struct dt_object **stripes;
9019 int mdt_count = lod->lod_remote_mdt_count + 1;
9027 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
9028 LASSERT(le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT);
9030 saved_count = lo->ldo_dir_stripes_allocated;
9031 stripe_count = le32_to_cpu(lum->lum_stripe_count);
9032 if (stripe_count <= saved_count)
9035 /* if the split target is overstriped, we need to put that flag in the
9036 * current layout so it can allocate the larger number of stripes
9038 * Note we need to pick up any hash *flags* which affect allocation
9039 * *before* allocation, so they're used in allocating the directory,
9040 * rather than after when we finalize directory setup (at the end of
9043 if (le32_to_cpu(lum->lum_hash_type) & LMV_HASH_FLAG_OVERSTRIPED)
9044 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_OVERSTRIPED;
9046 if (!(lo->ldo_dir_hash_type & LMV_HASH_FLAG_OVERSTRIPED) &&
9047 stripe_count > mdt_count) {
9049 } else if ((lo->ldo_dir_hash_type & LMV_HASH_FLAG_OVERSTRIPED) &&
9050 (stripe_count > mdt_count * LMV_MAX_STRIPES_PER_MDT ||
9051 /* a single MDT doesn't initialize the infrastructure for striped
9052 * directories, so we just don't support overstriping in that case
9058 dof->dof_type = DFT_DIR;
9060 OBD_ALLOC(stripes, sizeof(*stripes) * stripe_count);
9064 for (i = 0; i < lo->ldo_dir_stripes_allocated; i++)
9065 stripes[i] = lo->ldo_stripe[i];
9067 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
9069 rc = lod_mdt_alloc_qos(env, lo, stripes, saved_count, stripe_count);
9071 rc = lod_mdt_alloc_rr(env, lo, stripes, saved_count,
9074 OBD_FREE(stripes, sizeof(*stripes) * stripe_count);
9078 LASSERT(rc > saved_count);
9079 OBD_FREE(lo->ldo_stripe,
9080 sizeof(*stripes) * lo->ldo_dir_stripes_allocated);
9081 lo->ldo_stripe = stripes;
9082 lo->ldo_is_foreign = 0;
9083 lo->ldo_dir_striped = 1;
9084 lo->ldo_dir_stripe_count = rc;
9085 lo->ldo_dir_stripes_allocated = stripe_count;
9086 lo->ldo_dir_split_hash = lo->ldo_dir_hash_type;
9087 lo->ldo_dir_hash_type = le32_to_cpu(lum->lum_hash_type);
9088 if (!lmv_is_known_hash_type(lo->ldo_dir_hash_type))
9089 lo->ldo_dir_hash_type =
9090 lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
9091 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_SPLIT | LMV_HASH_FLAG_MIGRATION;
9092 lo->ldo_dir_split_offset = saved_count;
9093 lo->ldo_dir_layout_version++;
9094 lo->ldo_dir_stripe_loaded = 1;
9096 rc = lod_dir_declare_create_stripes(env, dt, mlc->mlc_attr, dof, th);
9098 lod_striping_free(env, lo);
9104 * detach all stripes from dir master object, NB, stripes are not destroyed, but
9105 * deleted from it's parent namespace, this function is called in two places:
9106 * 1. mdd_migrate_mdt() detach stripes from source, and attach them to
9108 * 2. mdd_dir_layout_update() detach stripe before turning 1-stripe directory to
9109 * a plain directory.
9111 * \param[in] env execution environment
9112 * \param[in] dt target object
9113 * \param[in] mlc layout change data
9114 * \param[in] th transaction handle
9116 * \retval 0 on success
9117 * \retval negative if failed
9119 static int lod_dir_layout_detach(const struct lu_env *env,
9120 struct dt_object *dt,
9121 const struct md_layout_change *mlc,
9124 struct lod_thread_info *info = lod_env_info(env);
9125 struct lod_object *lo = lod_dt_obj(dt);
9126 struct dt_object *next = dt_object_child(dt);
9127 char *stripe_name = info->lti_key;
9128 struct dt_object *dto;
9134 if (!lo->ldo_dir_stripe_count) {
9135 /* plain directory delete .. */
9136 rc = lod_sub_delete(env, next,
9137 (const struct dt_key *)dotdot, th);
9141 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
9142 dto = lo->ldo_stripe[i];
9146 rc = lod_sub_delete(env, dto,
9147 (const struct dt_key *)dotdot, th);
9151 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
9152 PFID(lu_object_fid(&dto->do_lu)), i);
9154 rc = lod_sub_delete(env, next,
9155 (const struct dt_key *)stripe_name, th);
9159 rc = lod_sub_ref_del(env, next, th);
9164 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
9165 dto = lo->ldo_stripe[i];
9167 dt_object_put(env, dto);
9169 OBD_FREE_PTR_ARRAY(lo->ldo_stripe, lo->ldo_dir_stripes_allocated);
9170 lo->ldo_stripe = NULL;
9171 lo->ldo_dir_stripes_allocated = 0;
9172 lo->ldo_dir_stripe_count = 0;
9173 dt->do_index_ops = &lod_index_ops;
9178 static int lod_dir_layout_shrink(const struct lu_env *env,
9179 struct dt_object *dt,
9180 const struct md_layout_change *mlc,
9183 struct lod_thread_info *info = lod_env_info(env);
9184 struct lod_object *lo = lod_dt_obj(dt);
9185 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
9186 struct dt_object *next = dt_object_child(dt);
9187 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
9188 __u32 final_stripe_count;
9189 char *stripe_name = info->lti_key;
9190 struct dt_object *dto;
9191 struct lu_buf *lmv_buf = &info->lti_buf;
9192 struct lmv_mds_md_v1 *lmv = &info->lti_lmv.lmv_md_v1;
9194 int type = LU_SEQ_RANGE_ANY;
9200 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
9202 lmv_buf->lb_buf = lmv;
9203 lmv_buf->lb_len = sizeof(*lmv);
9204 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
9205 lmv->lmv_stripe_count = cpu_to_le32(final_stripe_count);
9206 lmv->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type) &
9207 cpu_to_le32(LMV_HASH_TYPE_MASK |
9208 LMV_HASH_FLAG_FIXED);
9209 lmv->lmv_layout_version =
9210 cpu_to_le32(lo->ldo_dir_layout_version + 1);
9211 lmv->lmv_migrate_offset = 0;
9212 lmv->lmv_migrate_hash = 0;
9214 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
9215 dto = lo->ldo_stripe[i];
9219 if (i < final_stripe_count) {
9220 rc = lod_fld_lookup(env, lod,
9221 lu_object_fid(&dto->do_lu),
9226 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
9227 rc = lod_sub_xattr_set(env, dto, lmv_buf,
9229 LU_XATTR_REPLACE, th);
9236 dt_write_lock(env, dto, DT_TGT_CHILD);
9237 rc = lod_sub_ref_del(env, dto, th);
9238 dt_write_unlock(env, dto);
9242 rc = lod_sub_destroy(env, dto, th);
9246 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
9247 PFID(lu_object_fid(&dto->do_lu)), i);
9249 rc = lod_sub_delete(env, next,
9250 (const struct dt_key *)stripe_name, th);
9254 rc = lod_sub_ref_del(env, next, th);
9259 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &mdtidx,
9264 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
9265 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
9266 rc = lod_sub_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
9267 LU_XATTR_REPLACE, th);
9271 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
9272 dto = lo->ldo_stripe[i];
9274 dt_object_put(env, dto);
9276 lo->ldo_dir_stripe_count = final_stripe_count;
9281 static mlc_handler dir_mlc_declare_ops[MD_LAYOUT_MAX] = {
9282 [MD_LAYOUT_ATTACH] = lod_dir_declare_layout_attach,
9283 [MD_LAYOUT_DETACH] = lod_dir_declare_layout_detach,
9284 [MD_LAYOUT_SHRINK] = lod_dir_declare_layout_shrink,
9285 [MD_LAYOUT_SPLIT] = lod_dir_declare_layout_split,
9288 static mlc_handler dir_mlc_ops[MD_LAYOUT_MAX] = {
9289 [MD_LAYOUT_DETACH] = lod_dir_layout_detach,
9290 [MD_LAYOUT_SHRINK] = lod_dir_layout_shrink,
9293 static int lod_declare_layout_change(const struct lu_env *env,
9294 struct dt_object *dt, struct md_layout_change *mlc,
9297 struct lod_thread_info *info = lod_env_info(env);
9298 struct lod_object *lo = lod_dt_obj(dt);
9303 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
9304 LASSERT(dir_mlc_declare_ops[mlc->mlc_opc]);
9305 rc = dir_mlc_declare_ops[mlc->mlc_opc](env, dt, mlc, th);
9309 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
9310 dt_object_remote(dt_object_child(dt)))
9313 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
9314 struct layout_intent *intent = mlc->mlc_intent;
9316 if (intent->lai_opc == LAYOUT_INTENT_PCCRO_SET ||
9317 intent->lai_opc == LAYOUT_INTENT_PCCRO_CLEAR) {
9318 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
9321 rc = lod_declare_update_pccro(env, dt, mlc, th);
9326 rc = lod_striping_load(env, lo);
9330 LASSERT(lo->ldo_comp_cnt > 0);
9332 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
9336 switch (lo->ldo_flr_state) {
9338 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
9342 rc = lod_declare_update_rdonly(env, lo, mlc, th);
9344 case LCM_FL_WRITE_PENDING:
9345 rc = lod_declare_update_write_pending(env, lo, mlc, th);
9347 case LCM_FL_SYNC_PENDING:
9348 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
9355 rc = lod_save_layout_gen_intrans(info, lo);
9362 * Instantiate layout component objects which covers the intent write offset.
9364 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
9365 struct md_layout_change *mlc, struct thandle *th)
9367 struct lod_thread_info *info = lod_env_info(env);
9368 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
9369 struct lu_attr *layout_attr = &info->lti_layout_attr;
9370 struct lod_object *lo = lod_dt_obj(dt);
9375 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
9376 LASSERT(dir_mlc_ops[mlc->mlc_opc]);
9377 rc = dir_mlc_ops[mlc->mlc_opc](env, dt, mlc, th);
9381 rc = lod_check_layout_gen_intrans(info, lo);
9384 "%s: obj "DFID" gen changed from %d to %d in transaction, retry the transaction \n",
9385 dt->do_lu.lo_dev->ld_obd->obd_name,
9386 PFID(lu_object_fid(&dt->do_lu)),
9387 info->lti_gen[rc - 1], lo->ldo_layout_gen);
9391 rc = lod_striped_create(env, dt, attr, NULL, th);
9392 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
9393 layout_attr->la_layout_version |= lo->ldo_layout_gen;
9394 rc = lod_attr_set(env, dt, layout_attr, th);
9400 const struct dt_object_operations lod_obj_ops = {
9401 .do_read_lock = lod_read_lock,
9402 .do_write_lock = lod_write_lock,
9403 .do_read_unlock = lod_read_unlock,
9404 .do_write_unlock = lod_write_unlock,
9405 .do_write_locked = lod_write_locked,
9406 .do_attr_get = lod_attr_get,
9407 .do_declare_attr_set = lod_declare_attr_set,
9408 .do_attr_set = lod_attr_set,
9409 .do_xattr_get = lod_xattr_get,
9410 .do_declare_xattr_set = lod_declare_xattr_set,
9411 .do_xattr_set = lod_xattr_set,
9412 .do_declare_xattr_del = lod_declare_xattr_del,
9413 .do_xattr_del = lod_xattr_del,
9414 .do_xattr_list = lod_xattr_list,
9415 .do_ah_init = lod_ah_init,
9416 .do_declare_create = lod_declare_create,
9417 .do_create = lod_create,
9418 .do_declare_destroy = lod_declare_destroy,
9419 .do_destroy = lod_destroy,
9420 .do_index_try = lod_index_try,
9421 .do_declare_ref_add = lod_declare_ref_add,
9422 .do_ref_add = lod_ref_add,
9423 .do_declare_ref_del = lod_declare_ref_del,
9424 .do_ref_del = lod_ref_del,
9425 .do_object_sync = lod_object_sync,
9426 .do_object_lock = lod_object_lock,
9427 .do_object_unlock = lod_object_unlock,
9428 .do_invalidate = lod_invalidate,
9429 .do_declare_layout_change = lod_declare_layout_change,
9430 .do_layout_change = lod_layout_change,
9434 * Implementation of dt_body_operations::dbo_read.
9436 * \see dt_body_operations::dbo_read() in the API description for details.
9438 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
9439 struct lu_buf *buf, loff_t *pos)
9441 struct dt_object *next = dt_object_child(dt);
9443 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
9444 S_ISLNK(dt->do_lu.lo_header->loh_attr));
9445 return next->do_body_ops->dbo_read(env, next, buf, pos);
9449 * Implementation of dt_body_operations::dbo_declare_write.
9451 * \see dt_body_operations::dbo_declare_write() in the API description
9454 static ssize_t lod_declare_write(const struct lu_env *env,
9455 struct dt_object *dt,
9456 const struct lu_buf *buf, loff_t pos,
9459 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
9463 * Implementation of dt_body_operations::dbo_write.
9465 * \see dt_body_operations::dbo_write() in the API description for details.
9467 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
9468 const struct lu_buf *buf, loff_t *pos,
9471 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
9472 S_ISLNK(dt->do_lu.lo_header->loh_attr));
9473 return lod_sub_write(env, dt_object_child(dt), buf, pos, th);
9476 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
9477 __u64 start, __u64 end, struct thandle *th)
9479 if (dt_object_remote(dt))
9482 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
9485 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
9486 __u64 start, __u64 end, struct thandle *th)
9488 if (dt_object_remote(dt))
9491 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
9492 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
9496 * different type of files use the same body_ops because object may be created
9497 * in OUT, where there is no chance to set correct body_ops for each type, so
9498 * body_ops themselves will check file type inside, see lod_read/write/punch for
9501 static const struct dt_body_operations lod_body_ops = {
9502 .dbo_read = lod_read,
9503 .dbo_declare_write = lod_declare_write,
9504 .dbo_write = lod_write,
9505 .dbo_declare_punch = lod_declare_punch,
9506 .dbo_punch = lod_punch,
9510 * Implementation of lu_object_operations::loo_object_init.
9512 * The function determines the type and the index of the target device using
9513 * sequence of the object's FID. Then passes control down to the
9514 * corresponding device:
9515 * OSD for the local objects, OSP for remote
9517 * \see lu_object_operations::loo_object_init() in the API description
9520 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
9521 const struct lu_object_conf *conf)
9523 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
9524 struct lu_device *cdev = NULL;
9525 struct lu_object *cobj;
9526 struct lod_tgt_descs *ltd = NULL;
9527 struct lod_tgt_desc *tgt;
9529 int type = LU_SEQ_RANGE_ANY;
9533 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
9537 if (type == LU_SEQ_RANGE_MDT &&
9538 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
9539 cdev = &lod->lod_child->dd_lu_dev;
9540 } else if (type == LU_SEQ_RANGE_MDT) {
9541 ltd = &lod->lod_mdt_descs;
9543 } else if (type == LU_SEQ_RANGE_OST) {
9544 ltd = &lod->lod_ost_descs;
9551 if (ltd->ltd_tgts_size > idx &&
9552 test_bit(idx, ltd->ltd_tgt_bitmap)) {
9553 tgt = LTD_TGT(ltd, idx);
9555 LASSERT(tgt != NULL);
9556 LASSERT(tgt->ltd_tgt != NULL);
9558 cdev = &(tgt->ltd_tgt->dd_lu_dev);
9560 lod_putref(lod, ltd);
9563 if (unlikely(cdev == NULL))
9566 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
9567 if (unlikely(cobj == NULL))
9570 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
9572 lu_object_add(lo, cobj);
9579 * Release resources associated with striping.
9581 * If the object is striped (regular or directory), then release
9582 * the stripe objects references and free the ldo_stripe array.
9584 * \param[in] env execution environment
9585 * \param[in] lo object
9587 void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo)
9589 struct lod_layout_component *lod_comp;
9590 __u32 obj_attr = lo->ldo_obj.do_lu.lo_header->loh_attr;
9593 if (unlikely(lo->ldo_is_foreign)) {
9594 if (S_ISREG(obj_attr)) {
9595 lod_free_foreign_lov(lo);
9596 lo->ldo_comp_cached = 0;
9597 } else if (S_ISDIR(obj_attr)) {
9598 lod_free_foreign_lmv(lo);
9599 lo->ldo_dir_stripe_loaded = 0;
9601 } else if (lo->ldo_stripe != NULL) {
9602 LASSERT(lo->ldo_comp_entries == NULL);
9603 LASSERT(lo->ldo_dir_stripes_allocated > 0);
9605 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
9606 if (lo->ldo_stripe[i])
9607 dt_object_put(env, lo->ldo_stripe[i]);
9610 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
9611 OBD_FREE(lo->ldo_stripe, j);
9612 lo->ldo_stripe = NULL;
9613 lo->ldo_dir_stripes_allocated = 0;
9614 lo->ldo_dir_stripe_loaded = 0;
9615 lo->ldo_dir_stripe_count = 0;
9616 lo->ldo_obj.do_index_ops = NULL;
9617 } else if (lo->ldo_comp_entries != NULL) {
9618 for (i = 0; i < lo->ldo_comp_cnt; i++) {
9619 /* free lod_layout_component::llc_stripe array */
9620 lod_comp = &lo->ldo_comp_entries[i];
9622 /* HSM layout component */
9623 if (lod_comp->llc_magic == LOV_MAGIC_FOREIGN)
9625 if (lod_comp->llc_stripe == NULL)
9627 LASSERT(lod_comp->llc_stripes_allocated != 0);
9628 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
9629 if (lod_comp->llc_stripe[j] != NULL)
9631 &lod_comp->llc_stripe[j]->do_lu);
9633 OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe,
9634 lod_comp->llc_stripes_allocated);
9635 lod_comp->llc_stripe = NULL;
9636 OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices,
9637 lod_comp->llc_stripes_allocated);
9638 lod_comp->llc_ost_indices = NULL;
9639 lod_comp->llc_stripes_allocated = 0;
9641 lod_free_comp_entries(lo);
9642 lo->ldo_comp_cached = 0;
9646 void lod_striping_free(const struct lu_env *env, struct lod_object *lo)
9648 mutex_lock(&lo->ldo_layout_mutex);
9649 lod_striping_free_nolock(env, lo);
9650 mutex_unlock(&lo->ldo_layout_mutex);
9654 * Implementation of lu_object_operations::loo_object_free.
9656 * \see lu_object_operations::loo_object_free() in the API description
9659 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
9661 struct lod_object *lo = lu2lod_obj(o);
9663 /* release all underlying object pinned */
9664 lod_striping_free(env, lo);
9666 /* lo doesn't contain a lu_object_header, so we don't need call_rcu */
9667 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
9671 * Implementation of lu_object_operations::loo_object_release.
9673 * \see lu_object_operations::loo_object_release() in the API description
9676 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
9678 /* XXX: shouldn't we release everything here in case if object
9679 * creation failed before? */
9683 * Implementation of lu_object_operations::loo_object_print.
9685 * \see lu_object_operations::loo_object_print() in the API description
9688 static int lod_object_print(const struct lu_env *env, void *cookie,
9689 lu_printer_t p, const struct lu_object *l)
9691 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
9693 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
9696 const struct lu_object_operations lod_lu_obj_ops = {
9697 .loo_object_init = lod_object_init,
9698 .loo_object_free = lod_object_free,
9699 .loo_object_release = lod_object_release,
9700 .loo_object_print = lod_object_print,