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;
2727 entry = &lo->ldo_comp_entries[comp_idx];
2728 if (lod_comp_inited(entry))
2729 return entry->llc_stripe_count;
2730 if (entry->llc_stripe_count == LOV_ALL_STRIPES)
2731 return lod_get_stripe_count_plain(lod, lo,
2732 entry->llc_stripe_count,
2733 entry->llc_pattern &
2734 LOV_PATTERN_OVERSTRIPING,
2737 return lod_get_stripe_count(lod, lo, comp_idx, entry->llc_stripe_count,
2738 entry->llc_pattern & LOV_PATTERN_OVERSTRIPING,
2742 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2744 int magic, size = 0, i;
2745 struct lod_layout_component *comp_entries;
2747 bool is_composite, is_foreign = false;
2750 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2751 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2753 lo->ldo_def_striping->lds_def_striping_is_composite;
2755 comp_cnt = lo->ldo_comp_cnt;
2756 comp_entries = lo->ldo_comp_entries;
2757 is_composite = lo->ldo_is_composite;
2758 is_foreign = lo->ldo_is_foreign;
2762 return lo->ldo_foreign_lov_size;
2764 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2766 size = sizeof(struct lov_comp_md_v1) +
2767 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2768 LASSERT(size % sizeof(__u64) == 0);
2771 for (i = 0; i < comp_cnt; i++) {
2774 if (comp_entries[i].llc_magic == LOV_MAGIC_FOREIGN) {
2775 size += lov_foreign_md_size(comp_entries[i].llc_length);
2777 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 :
2779 stripe_count = lod_comp_entry_stripe_count(lo, i,
2781 if (!is_dir && is_composite)
2782 lod_comp_shrink_stripe_count(&comp_entries[i],
2785 size += lov_user_md_size(stripe_count, magic);
2787 LASSERT(size % sizeof(__u64) == 0);
2793 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2794 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2797 * \param[in] env execution environment
2798 * \param[in] dt dt_object to add components on
2799 * \param[in] buf buffer contains components to be added
2800 * \parem[in] th thandle
2802 * \retval 0 on success
2803 * \retval negative errno on failure
2805 static int lod_declare_layout_add(const struct lu_env *env,
2806 struct dt_object *dt,
2807 const struct lu_buf *buf,
2810 struct lod_thread_info *info = lod_env_info(env);
2811 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2812 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2813 struct dt_object *next = dt_object_child(dt);
2814 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
2815 struct lod_object *lo = lod_dt_obj(dt);
2816 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2818 int i, rc, array_cnt, old_array_cnt;
2821 LASSERT(lo->ldo_is_composite);
2823 if (lo->ldo_flr_state != LCM_FL_NONE)
2826 rc = lod_verify_striping(env, d, lo, buf, false);
2830 magic = comp_v1->lcm_magic;
2831 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2832 lustre_swab_lov_comp_md_v1(comp_v1);
2833 magic = comp_v1->lcm_magic;
2836 if (magic != LOV_USER_MAGIC_COMP_V1)
2839 mutex_lock(&lo->ldo_layout_mutex);
2841 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2842 OBD_ALLOC_PTR_ARRAY(comp_array, array_cnt);
2843 if (comp_array == NULL) {
2844 mutex_unlock(&lo->ldo_layout_mutex);
2849 memcpy(comp_array, lo->ldo_comp_entries,
2850 sizeof(*comp_array) * lo->ldo_comp_cnt);
2852 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2853 struct lov_user_md_v1 *v1;
2854 struct lu_extent *ext;
2856 v1 = (struct lov_user_md *)((char *)comp_v1 +
2857 comp_v1->lcm_entries[i].lcme_offset);
2858 ext = &comp_v1->lcm_entries[i].lcme_extent;
2860 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2861 lod_comp->llc_extent.e_start = ext->e_start;
2862 lod_comp->llc_extent.e_end = ext->e_end;
2863 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2864 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2866 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2867 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2868 lod_comp->llc_pattern = v1->lmm_pattern;
2870 * limit stripe count so that it's less than/equal to
2871 * extent_size / stripe_size.
2873 * Note: extension size reused llc_stripe_size field and
2874 * uninstantiated component could be defined with
2875 * extent_start == extent_end as extension component will
2878 if (!(lod_comp->llc_flags & LCME_FL_EXTENSION) &&
2879 (lod_comp_inited(lod_comp) ||
2880 lod_comp->llc_extent.e_start <
2881 lod_comp->llc_extent.e_end) &&
2882 lod_comp->llc_stripe_count != LOV_ALL_STRIPES &&
2883 ext->e_end != OBD_OBJECT_EOF &&
2884 (__u64)(lod_comp->llc_stripe_count *
2885 lod_comp->llc_stripe_size) >
2886 (ext->e_end - ext->e_start))
2887 lod_comp->llc_stripe_count =
2888 DIV_ROUND_UP(ext->e_end - ext->e_start,
2889 lod_comp->llc_stripe_size);
2890 lod_adjust_stripe_info(lod_comp, desc, 0);
2892 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2893 struct lov_user_md_v3 *v3 = (typeof(*v3) *) v1;
2895 if (v3->lmm_pool_name[0] != '\0' &&
2896 !lov_pool_is_ignored(v3->lmm_pool_name)) {
2897 rc = lod_set_pool(&lod_comp->llc_pool,
2905 old_array = lo->ldo_comp_entries;
2906 old_array_cnt = lo->ldo_comp_cnt;
2908 lo->ldo_comp_entries = comp_array;
2909 lo->ldo_comp_cnt = array_cnt;
2911 /* No need to increase layout generation here, it will be increased
2912 * later when generating component ID for the new components */
2914 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2915 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2916 XATTR_NAME_LOV, 0, th);
2918 lo->ldo_comp_entries = old_array;
2919 lo->ldo_comp_cnt = old_array_cnt;
2923 OBD_FREE_PTR_ARRAY(old_array, old_array_cnt);
2925 LASSERT(lo->ldo_mirror_count == 1);
2926 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2928 mutex_unlock(&lo->ldo_layout_mutex);
2933 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2934 lod_comp = &comp_array[i];
2935 if (lod_comp->llc_pool != NULL) {
2936 OBD_FREE(lod_comp->llc_pool,
2937 strlen(lod_comp->llc_pool) + 1);
2938 lod_comp->llc_pool = NULL;
2941 OBD_FREE_PTR_ARRAY(comp_array, array_cnt);
2942 mutex_unlock(&lo->ldo_layout_mutex);
2948 * lod_last_non_stale_mirror() - Check if a mirror is the last non-stale mirror.
2949 * @mirror_id: Mirror id to be checked.
2952 * This function checks if a mirror with specified @mirror_id is the last
2953 * non-stale mirror of a LOD object @lo.
2955 * Return: true or false.
2958 bool lod_last_non_stale_mirror(__u16 mirror_id, struct lod_object *lo)
2960 struct lod_layout_component *lod_comp;
2961 bool has_stale_flag;
2964 for (i = 0; i < lo->ldo_mirror_count; i++) {
2965 if (lo->ldo_mirrors[i].lme_id == mirror_id ||
2966 lo->ldo_mirrors[i].lme_stale)
2969 has_stale_flag = false;
2970 lod_foreach_mirror_comp(lod_comp, lo, i) {
2971 if (lod_comp->llc_flags & LCME_FL_STALE) {
2972 has_stale_flag = true;
2976 if (!has_stale_flag)
2984 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2985 * the '$field' can only be 'flags' now. The xattr value is binary
2986 * lov_comp_md_v1 which contains the component ID(s) and the value of
2987 * the field to be modified.
2988 * Please update allowed_lustre_lov macro if $field groks more values
2991 * \param[in] env execution environment
2992 * \param[in] dt dt_object to be modified
2993 * \param[in] op operation string, like "set.flags"
2994 * \param[in] buf buffer contains components to be set
2995 * \parem[in] th thandle
2997 * \retval 0 on success
2998 * \retval negative errno on failure
3000 static int lod_declare_layout_set(const struct lu_env *env,
3001 struct dt_object *dt,
3002 char *op, const struct lu_buf *buf,
3005 struct lod_layout_component *lod_comp;
3006 struct lod_thread_info *info = lod_env_info(env);
3007 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3008 struct lod_object *lo = lod_dt_obj(dt);
3009 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3012 bool changed = false;
3015 /* Please update allowed_lustre_lov macro if op
3016 * groks more values in the future
3018 if (strcmp(op, "set.flags") != 0) {
3019 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
3020 lod2obd(d)->obd_name, op);
3024 magic = comp_v1->lcm_magic;
3025 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
3026 lustre_swab_lov_comp_md_v1(comp_v1);
3027 magic = comp_v1->lcm_magic;
3030 if (magic != LOV_USER_MAGIC_COMP_V1)
3033 if (comp_v1->lcm_entry_count == 0) {
3034 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
3035 lod2obd(d)->obd_name);
3039 mutex_lock(&lo->ldo_layout_mutex);
3040 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
3041 __u32 id = comp_v1->lcm_entries[i].lcme_id;
3042 __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
3043 __u32 mirror_flag = flags & LCME_MIRROR_FLAGS;
3044 __u16 mirror_id = mirror_id_of(id);
3045 bool neg = flags & LCME_FL_NEG;
3047 if (flags & LCME_FL_INIT) {
3049 lod_striping_free_nolock(env, lo);
3050 mutex_unlock(&lo->ldo_layout_mutex);
3054 flags &= ~(LCME_MIRROR_FLAGS | LCME_FL_NEG);
3055 for (j = 0; j < lo->ldo_comp_cnt; j++) {
3056 lod_comp = &lo->ldo_comp_entries[j];
3058 /* lfs only put one flag in each entry */
3059 if ((flags && id != lod_comp->llc_id) ||
3060 (mirror_flag && mirror_id !=
3061 mirror_id_of(lod_comp->llc_id)))
3066 lod_comp->llc_flags &= ~flags;
3068 lod_comp->llc_flags &= ~mirror_flag;
3071 if ((flags & LCME_FL_STALE) &&
3072 lod_last_non_stale_mirror(mirror_id,
3075 &lo->ldo_layout_mutex);
3078 lod_comp->llc_flags |= flags;
3081 lod_comp->llc_flags |= mirror_flag;
3082 if (mirror_flag & LCME_FL_NOSYNC)
3083 lod_comp->llc_timestamp =
3084 ktime_get_real_seconds();
3090 mutex_unlock(&lo->ldo_layout_mutex);
3093 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
3094 lod2obd(d)->obd_name);
3098 lod_obj_inc_layout_gen(lo);
3100 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3101 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
3102 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3107 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
3108 * and the xattr value is a unique component ID or a special lcme_id.
3110 * \param[in] env execution environment
3111 * \param[in] dt dt_object to be operated on
3112 * \param[in] buf buffer contains component ID or lcme_id
3113 * \parem[in] th thandle
3115 * \retval 0 on success
3116 * \retval negative errno on failure
3118 static int lod_declare_layout_del(const struct lu_env *env,
3119 struct dt_object *dt,
3120 const struct lu_buf *buf,
3123 struct lod_thread_info *info = lod_env_info(env);
3124 struct dt_object *next = dt_object_child(dt);
3125 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3126 struct lod_object *lo = lod_dt_obj(dt);
3127 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3128 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3129 __u32 magic, id, flags, neg_flags = 0;
3133 LASSERT(lo->ldo_is_composite);
3135 if (lo->ldo_flr_state != LCM_FL_NONE)
3138 magic = comp_v1->lcm_magic;
3139 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
3140 lustre_swab_lov_comp_md_v1(comp_v1);
3141 magic = comp_v1->lcm_magic;
3144 if (magic != LOV_USER_MAGIC_COMP_V1)
3147 id = comp_v1->lcm_entries[0].lcme_id;
3148 flags = comp_v1->lcm_entries[0].lcme_flags;
3150 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
3151 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
3152 lod2obd(d)->obd_name, id, flags);
3156 if (id != LCME_ID_INVAL && flags != 0) {
3157 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
3158 lod2obd(d)->obd_name);
3162 if (id == LCME_ID_INVAL && !flags) {
3163 CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
3164 lod2obd(d)->obd_name);
3168 if (flags & LCME_FL_NEG) {
3169 neg_flags = flags & ~LCME_FL_NEG;
3173 mutex_lock(&lo->ldo_layout_mutex);
3175 left = lo->ldo_comp_cnt;
3177 mutex_unlock(&lo->ldo_layout_mutex);
3181 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3182 struct lod_layout_component *lod_comp;
3184 lod_comp = &lo->ldo_comp_entries[i];
3186 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
3188 else if (flags && !(flags & lod_comp->llc_flags))
3190 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
3193 if (left != (i + 1)) {
3194 CDEBUG(D_LAYOUT, "%s: this deletion will create "
3195 "a hole.\n", lod2obd(d)->obd_name);
3196 mutex_unlock(&lo->ldo_layout_mutex);
3201 /* Mark the component as deleted */
3202 lod_comp->llc_id = LCME_ID_INVAL;
3204 /* Not instantiated component */
3205 if (lod_comp->llc_stripe == NULL)
3208 LASSERT(lod_comp->llc_stripe_count > 0);
3209 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
3210 struct dt_object *obj = lod_comp->llc_stripe[j];
3214 rc = lod_sub_declare_destroy(env, obj, th);
3216 mutex_unlock(&lo->ldo_layout_mutex);
3222 LASSERTF(left >= 0, "left = %d\n", left);
3223 if (left == lo->ldo_comp_cnt) {
3224 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
3225 lod2obd(d)->obd_name, id);
3226 mutex_unlock(&lo->ldo_layout_mutex);
3230 mutex_unlock(&lo->ldo_layout_mutex);
3232 memset(attr, 0, sizeof(*attr));
3233 attr->la_valid = LA_SIZE;
3234 rc = lod_sub_declare_attr_set(env, next, attr, th);
3239 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3240 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
3241 XATTR_NAME_LOV, 0, th);
3243 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
3250 * Declare layout add/set/del operations issued by special xattr names:
3252 * XATTR_LUSTRE_LOV.add add component(s) to existing file
3253 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
3254 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
3256 * \param[in] env execution environment
3257 * \param[in] dt object
3258 * \param[in] name name of xattr
3259 * \param[in] buf lu_buf contains xattr value
3260 * \param[in] th transaction handle
3262 * \retval 0 on success
3263 * \retval negative if failed
3265 static int lod_declare_modify_layout(const struct lu_env *env,
3266 struct dt_object *dt,
3268 const struct lu_buf *buf,
3271 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3272 struct lod_object *lo = lod_dt_obj(dt);
3274 int rc, len = strlen(XATTR_LUSTRE_LOV);
3277 LASSERT(dt_object_exists(dt));
3279 if (strlen(name) <= len || name[len] != '.') {
3280 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
3281 lod2obd(d)->obd_name, name);
3286 rc = lod_striping_load(env, lo);
3290 /* the layout to be modified must be a composite layout */
3291 if (!lo->ldo_is_composite) {
3292 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
3293 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3294 GOTO(unlock, rc = -EINVAL);
3297 op = (char *)name + len;
3298 if (strcmp(op, "add") == 0) {
3299 rc = lod_declare_layout_add(env, dt, buf, th);
3300 } else if (strcmp(op, "del") == 0) {
3301 rc = lod_declare_layout_del(env, dt, buf, th);
3302 } else if (strncmp(op, "set", strlen("set")) == 0) {
3303 rc = lod_declare_layout_set(env, dt, op, buf, th);
3305 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
3306 lod2obd(d)->obd_name, name);
3307 GOTO(unlock, rc = -ENOTSUPP);
3311 lod_striping_free(env, lo);
3317 * Convert a plain file lov_mds_md to a composite layout.
3319 * \param[in,out] info the thread info::lti_ea_store buffer contains little
3320 * endian plain file layout
3322 * \retval 0 on success, <0 on failure
3324 static int lod_layout_convert(struct lod_thread_info *info)
3326 struct lov_mds_md *lmm = info->lti_ea_store;
3327 struct lov_mds_md *lmm_save;
3328 struct lov_comp_md_v1 *lcm;
3329 struct lov_comp_md_entry_v1 *lcme;
3335 /* realloc buffer to a composite layout which contains one component */
3336 blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3337 le32_to_cpu(lmm->lmm_magic));
3338 size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
3340 OBD_ALLOC_LARGE(lmm_save, blob_size);
3342 GOTO(out, rc = -ENOMEM);
3344 memcpy(lmm_save, lmm, blob_size);
3346 if (info->lti_ea_store_size < size) {
3347 rc = lod_ea_store_resize(info, size);
3352 lcm = info->lti_ea_store;
3353 memset(lcm, 0, sizeof(*lcm) + sizeof(*lcme));
3354 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
3355 lcm->lcm_size = cpu_to_le32(size);
3356 lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
3357 lmm_save->lmm_layout_gen));
3358 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
3359 lcm->lcm_entry_count = cpu_to_le16(1);
3361 lcme = &lcm->lcm_entries[0];
3362 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
3363 lcme->lcme_extent.e_start = 0;
3364 lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
3365 lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
3366 lcme->lcme_size = cpu_to_le32(blob_size);
3368 memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
3373 OBD_FREE_LARGE(lmm_save, blob_size);
3378 * Merge layouts to form a mirrored file.
3380 static int lod_declare_layout_merge(const struct lu_env *env,
3381 struct dt_object *dt,
3382 const struct lu_buf *mbuf,
3385 struct lod_thread_info *info = lod_env_info(env);
3386 struct lu_attr *layout_attr = &info->lti_layout_attr;
3387 struct lu_buf *buf = &info->lti_buf;
3388 struct lod_object *lo = lod_dt_obj(dt);
3389 struct lov_comp_md_v1 *lcm;
3390 struct lov_comp_md_v1 *cur_lcm;
3391 struct lov_comp_md_v1 *merge_lcm;
3392 struct lov_comp_md_entry_v1 *lcme;
3393 struct lov_mds_md_v1 *lmm;
3396 __u16 cur_entry_count;
3397 __u16 merge_entry_count;
3399 __u16 mirror_id = 0;
3406 merge_lcm = mbuf->lb_buf;
3407 if (mbuf->lb_len < sizeof(*merge_lcm))
3410 /* must be an existing layout from disk */
3411 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
3414 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
3416 /* do not allow to merge two mirrored files */
3417 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
3420 /* verify the target buffer */
3421 rc = lod_get_lov_ea(env, lo);
3423 RETURN(rc ? : -ENODATA);
3425 cur_lcm = info->lti_ea_store;
3426 switch (le32_to_cpu(cur_lcm->lcm_magic)) {
3429 rc = lod_layout_convert(info);
3431 case LOV_MAGIC_COMP_V1:
3441 /* info->lti_ea_store could be reallocated in lod_layout_convert() */
3442 cur_lcm = info->lti_ea_store;
3443 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
3445 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
3446 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
3447 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
3450 /* size of new layout */
3451 size = le32_to_cpu(cur_lcm->lcm_size) +
3452 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
3454 memset(buf, 0, sizeof(*buf));
3455 lu_buf_alloc(buf, size);
3456 if (buf->lb_buf == NULL)
3460 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
3462 offset = sizeof(*lcm) +
3463 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
3464 for (i = 0; i < cur_entry_count; i++) {
3465 struct lov_comp_md_entry_v1 *cur_lcme;
3467 lcme = &lcm->lcm_entries[i];
3468 cur_lcme = &cur_lcm->lcm_entries[i];
3470 lcme->lcme_offset = cpu_to_le32(offset);
3471 memcpy((char *)lcm + offset,
3472 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
3473 le32_to_cpu(lcme->lcme_size));
3475 offset += le32_to_cpu(lcme->lcme_size);
3477 if (mirror_count == 1 &&
3478 mirror_id_of(le32_to_cpu(lcme->lcme_id)) == 0) {
3479 /* Add mirror from a non-flr file, create new mirror ID.
3480 * Otherwise, keep existing mirror's component ID, used
3481 * for mirror extension.
3483 id = pflr_id(1, i + 1);
3484 lcme->lcme_id = cpu_to_le32(id);
3487 id = max(le32_to_cpu(lcme->lcme_id), id);
3490 mirror_id = mirror_id_of(id) + 1;
3492 /* check if first entry in new layout is DOM */
3493 lmm = (struct lov_mds_md_v1 *)((char *)merge_lcm +
3494 merge_lcm->lcm_entries[0].lcme_offset);
3495 merge_has_dom = lov_pattern(le32_to_cpu(lmm->lmm_pattern)) &
3498 for (i = 0; i < merge_entry_count; i++) {
3499 struct lov_comp_md_entry_v1 *merge_lcme;
3501 merge_lcme = &merge_lcm->lcm_entries[i];
3502 lcme = &lcm->lcm_entries[cur_entry_count + i];
3504 *lcme = *merge_lcme;
3505 lcme->lcme_offset = cpu_to_le32(offset);
3506 if (merge_has_dom && i == 0)
3507 lcme->lcme_flags |= cpu_to_le32(LCME_FL_STALE);
3509 id = pflr_id(mirror_id, i + 1);
3510 lcme->lcme_id = cpu_to_le32(id);
3512 memcpy((char *)lcm + offset,
3513 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
3514 le32_to_cpu(lcme->lcme_size));
3516 offset += le32_to_cpu(lcme->lcme_size);
3519 /* fixup layout information */
3520 lcm->lcm_size = cpu_to_le32(size);
3521 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
3522 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
3523 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NONE)
3524 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
3526 rc = lod_striping_reload(env, lo, buf, 0);
3530 lod_obj_inc_layout_gen(lo);
3531 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3533 /* transfer layout version to OST objects. */
3534 if (lo->ldo_mirror_count > 1) {
3535 struct lod_obj_stripe_cb_data data = { {0} };
3537 layout_attr->la_valid = LA_LAYOUT_VERSION;
3538 layout_attr->la_layout_version = 0;
3539 data.locd_attr = layout_attr;
3540 data.locd_declare = true;
3541 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
3542 rc = lod_obj_for_each_stripe(env, lo, th, &data);
3547 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
3548 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3556 * Split layouts, just set the LOVEA with the layout from mbuf.
3558 static int lod_declare_layout_split(const struct lu_env *env,
3559 struct dt_object *dt, const struct lu_buf *mbuf,
3562 struct lod_thread_info *info = lod_env_info(env);
3563 struct lu_attr *layout_attr = &info->lti_layout_attr;
3564 struct lod_object *lo = lod_dt_obj(dt);
3565 struct lov_comp_md_v1 *lcm = mbuf->lb_buf;
3569 rc = lod_striping_reload(env, lo, mbuf, LVF_ALL_STALE);
3573 lod_obj_inc_layout_gen(lo);
3574 /* fix on-disk layout gen */
3575 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3577 /* transfer layout version to OST objects. */
3578 if (lo->ldo_mirror_count > 1) {
3579 struct lod_obj_stripe_cb_data data = { {0} };
3581 layout_attr->la_valid = LA_LAYOUT_VERSION;
3582 layout_attr->la_layout_version = 0;
3583 data.locd_attr = layout_attr;
3584 data.locd_declare = true;
3585 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
3586 rc = lod_obj_for_each_stripe(env, lo, th, &data);
3591 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), mbuf,
3592 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3596 static int lod_layout_declare_or_purge_mirror(const struct lu_env *env,
3597 struct dt_object *dt, const struct lu_buf *buf,
3598 struct thandle *th, bool declare)
3600 struct lod_thread_info *info = lod_env_info(env);
3601 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3602 struct lod_object *lo = lod_dt_obj(dt);
3603 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3604 struct lov_comp_md_entry_v1 *entry;
3605 struct lov_mds_md_v1 *lmm;
3606 struct dt_object **sub_objs = NULL;
3607 int rc = 0, i, k, array_count = 0;
3612 * other ops (like lod_declare_destroy) could destroying sub objects
3615 mutex_lock(&lo->ldo_layout_mutex);
3618 /* prepare sub-objects array */
3619 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
3620 entry = &comp_v1->lcm_entries[i];
3622 if (!(entry->lcme_flags & LCME_FL_INIT))
3625 lmm = (struct lov_mds_md_v1 *)
3626 ((char *)comp_v1 + entry->lcme_offset);
3627 array_count += lmm->lmm_stripe_count;
3629 OBD_ALLOC_PTR_ARRAY(sub_objs, array_count);
3630 if (sub_objs == NULL) {
3631 mutex_unlock(&lo->ldo_layout_mutex);
3636 k = 0; /* sub_objs index */
3637 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
3638 struct lov_ost_data_v1 *objs;
3639 struct lu_object *o, *n;
3640 struct dt_object *dto;
3641 struct lu_device *nd;
3642 struct lov_mds_md_v3 *v3;
3646 entry = &comp_v1->lcm_entries[i];
3648 if (!(entry->lcme_flags & LCME_FL_INIT))
3651 lmm = (struct lov_mds_md_v1 *)
3652 ((char *)comp_v1 + entry->lcme_offset);
3653 v3 = (struct lov_mds_md_v3 *)lmm;
3654 if (lmm->lmm_magic == LOV_MAGIC_V3)
3655 objs = &v3->lmm_objects[0];
3657 objs = &lmm->lmm_objects[0];
3659 for (j = 0; j < lmm->lmm_stripe_count; j++) {
3660 idx = objs[j].l_ost_idx;
3661 rc = ostid_to_fid(&info->lti_fid, &objs[j].l_ost_oi,
3666 if (!fid_is_sane(&info->lti_fid)) {
3667 CERROR("%s: sub-object insane fid "DFID"\n",
3668 lod2obd(d)->obd_name,
3669 PFID(&info->lti_fid));
3670 GOTO(out, rc = -EINVAL);
3673 lod_getref(&d->lod_ost_descs);
3675 rc = validate_lod_and_idx(d, idx);
3677 lod_putref(d, &d->lod_ost_descs);
3681 nd = &OST_TGT(d, idx)->ltd_tgt->dd_lu_dev;
3682 lod_putref(d, &d->lod_ost_descs);
3684 o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
3686 GOTO(out, rc = PTR_ERR(o));
3688 n = lu_object_locate(o->lo_header, nd->ld_type);
3690 lu_object_put(env, n);
3691 GOTO(out, rc = -ENOENT);
3694 dto = container_of(n, struct dt_object, do_lu);
3697 rc = lod_sub_declare_destroy(env, dto, th);
3698 dt_object_put(env, dto);
3703 * collect to-be-destroyed sub objects, the
3704 * reference would be released after actual
3710 } /* for each stripe */
3711 } /* for each component in the mirror */
3716 /* destroy the sub objects */
3717 for (; i < k; i++) {
3718 rc = lod_sub_destroy(env, sub_objs[i], th);
3721 dt_object_put(env, sub_objs[i]);
3725 * if a sub object destroy failed, we'd release sub objects
3726 * reference get from above sub_objs collection.
3729 dt_object_put(env, sub_objs[i]);
3731 OBD_FREE_PTR_ARRAY(sub_objs, array_count);
3733 mutex_unlock(&lo->ldo_layout_mutex);
3739 * Purge layouts, delete sub objects in the mirror stored in the vic_buf,
3740 * and set the LOVEA with the layout from mbuf.
3742 static int lod_declare_layout_purge(const struct lu_env *env,
3743 struct dt_object *dt, const struct lu_buf *buf,
3746 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3747 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3752 if (le32_to_cpu(comp_v1->lcm_magic) != LOV_MAGIC_COMP_V1) {
3753 CERROR("%s: invalid layout magic %#x != %#x\n",
3754 lod2obd(d)->obd_name, le32_to_cpu(comp_v1->lcm_magic),
3759 if (cpu_to_le32(LOV_MAGIC_COMP_V1) != LOV_MAGIC_COMP_V1)
3760 lustre_swab_lov_comp_md_v1(comp_v1);
3762 /* from now on, @buf contains cpu endian data */
3764 if (comp_v1->lcm_mirror_count != 0) {
3765 CERROR("%s: can only purge one mirror from "DFID"\n",
3766 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3770 /* delcare sub objects deletion in the mirror stored in @buf */
3771 rc = lod_layout_declare_or_purge_mirror(env, dt, buf, th, true);
3775 /* delete sub objects from the mirror stored in @buf */
3776 static int lod_layout_purge(const struct lu_env *env, struct dt_object *dt,
3777 const struct lu_buf *buf, struct thandle *th)
3782 rc = lod_layout_declare_or_purge_mirror(env, dt, buf, th, false);
3787 * Implementation of dt_object_operations::do_declare_xattr_set.
3789 * \see dt_object_operations::do_declare_xattr_set() in the API description
3792 * the extension to the API:
3793 * - declaring LOVEA requests striping creation
3794 * - LU_XATTR_REPLACE means layout swap
3796 static int lod_declare_xattr_set(const struct lu_env *env,
3797 struct dt_object *dt,
3798 const struct lu_buf *buf,
3799 const char *name, int fl,
3802 struct lod_thread_info *info = lod_env_info(env);
3803 struct dt_object *next = dt_object_child(dt);
3804 struct lu_attr *attr = &info->lti_attr;
3805 struct lod_object *lo = lod_dt_obj(dt);
3810 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
3811 if ((S_ISREG(mode) || mode == 0) &&
3812 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE | LU_XATTR_SPLIT |
3814 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3815 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
3817 * this is a request to create object's striping.
3819 * allow to declare predefined striping on a new (!mode) object
3820 * which is supposed to be replay of regular file creation
3821 * (when LOV setting is declared)
3823 * LU_XATTR_REPLACE is set to indicate a layout swap
3825 if (dt_object_exists(dt)) {
3826 rc = dt_attr_get(env, next, attr);
3830 memset(attr, 0, sizeof(*attr));
3831 attr->la_valid = LA_TYPE | LA_MODE;
3832 attr->la_mode = S_IFREG;
3834 rc = lod_declare_striped_create(env, dt, attr, buf, th);
3835 } else if (fl & LU_XATTR_MERGE) {
3836 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3837 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3838 rc = lod_declare_layout_merge(env, dt, buf, th);
3839 } else if (fl & LU_XATTR_SPLIT) {
3840 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3841 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3842 rc = lod_declare_layout_split(env, dt, buf, th);
3843 } else if (fl & LU_XATTR_PURGE) {
3844 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3845 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3846 rc = lod_declare_layout_purge(env, dt, buf, th);
3847 } else if (S_ISREG(mode) &&
3848 strlen(name) >= sizeof(XATTR_LUSTRE_LOV) + 3 &&
3849 allowed_lustre_lov(name)) {
3851 * this is a request to modify object's striping.
3852 * add/set/del component(s).
3854 if (!dt_object_exists(dt))
3857 rc = lod_declare_modify_layout(env, dt, name, buf, th);
3858 } else if (S_ISDIR(mode)) {
3859 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
3860 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3861 rc = lod_replace_parent_fid(env, dt, buf, th, true);
3863 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
3867 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3868 strcmp(name, XATTR_LUSTRE_LOV) == 0 || allowed_lustre_lov(name)))
3869 rc = lod_save_layout_gen_intrans(info, lo);
3875 * Apply xattr changes to the object.
3877 * Applies xattr changes to the object and the stripes if the latter exist.
3879 * \param[in] env execution environment
3880 * \param[in] dt object
3881 * \param[in] buf buffer pointing to the new value of xattr
3882 * \param[in] name name of xattr
3883 * \param[in] fl flags
3884 * \param[in] th transaction handle
3886 * \retval 0 on success
3887 * \retval negative if failed
3889 static int lod_xattr_set_internal(const struct lu_env *env,
3890 struct dt_object *dt,
3891 const struct lu_buf *buf,
3892 const char *name, int fl,
3895 struct dt_object *next = dt_object_child(dt);
3896 struct lod_object *lo = lod_dt_obj(dt);
3901 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3902 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3905 /* Note: Do not set LinkEA on sub-stripes, otherwise
3906 * it will confuse the fid2path process(see mdt_path_current()).
3907 * The linkEA between master and sub-stripes is set in
3908 * lod_xattr_set_lmv(). */
3909 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
3912 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3913 if (!lo->ldo_stripe[i])
3916 if (!dt_object_exists(lo->ldo_stripe[i]))
3919 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
3929 * Delete an extended attribute.
3931 * Deletes specified xattr from the object and the stripes if the latter exist.
3933 * \param[in] env execution environment
3934 * \param[in] dt object
3935 * \param[in] name name of xattr
3936 * \param[in] th transaction handle
3938 * \retval 0 on success
3939 * \retval negative if failed
3941 static int lod_xattr_del_internal(const struct lu_env *env,
3942 struct dt_object *dt,
3943 const char *name, struct thandle *th)
3945 struct dt_object *next = dt_object_child(dt);
3946 struct lod_object *lo = lod_dt_obj(dt);
3952 rc = lod_sub_xattr_del(env, next, name, th);
3953 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3956 if (lo->ldo_dir_stripe_count == 0)
3959 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3960 if (!lo->ldo_stripe[i])
3963 if (!dt_object_exists(lo->ldo_stripe[i]))
3966 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3975 * Set default striping on a directory.
3977 * Sets specified striping on a directory object unless it matches the default
3978 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3979 * EA. This striping will be used when regular file is being created in this
3982 * \param[in] env execution environment
3983 * \param[in] dt the striped object
3984 * \param[in] buf buffer with the striping
3985 * \param[in] name name of EA
3986 * \param[in] fl xattr flag (see OSD API description)
3987 * \param[in] th transaction handle
3989 * \retval 0 on success
3990 * \retval negative if failed
3992 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
3993 struct dt_object *dt,
3994 const struct lu_buf *buf,
3995 const char *name, int fl,
3998 struct lov_user_md_v1 *lum;
3999 struct lov_user_md_v3 *v3 = NULL;
4000 const char *pool_name = NULL;
4005 LASSERT(buf != NULL && buf->lb_buf != NULL);
4008 switch (lum->lmm_magic) {
4009 case LOV_USER_MAGIC_SPECIFIC:
4010 case LOV_USER_MAGIC_V3:
4012 if (lov_pool_is_reserved(v3->lmm_pool_name))
4013 memset(v3->lmm_pool_name, 0, sizeof(v3->lmm_pool_name));
4014 else if (v3->lmm_pool_name[0] != '\0')
4015 pool_name = v3->lmm_pool_name;
4017 case LOV_USER_MAGIC_V1:
4018 /* if { size, offset, count } = { 0, -1, 0 } and no pool
4019 * (i.e. all default values specified) then delete default
4020 * striping from dir. */
4022 "set default striping: sz %u # %u offset %d %s %s\n",
4023 (unsigned)lum->lmm_stripe_size,
4024 (unsigned)lum->lmm_stripe_count,
4025 (int)lum->lmm_stripe_offset,
4026 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
4028 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
4029 lum->lmm_stripe_count,
4030 lum->lmm_stripe_offset,
4033 case LOV_USER_MAGIC_COMP_V1:
4035 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lum;
4036 struct lov_comp_md_entry_v1 *lcme;
4039 comp_cnt = le16_to_cpu(lcm->lcm_entry_count);
4040 for (i = 0; i < comp_cnt; i++) {
4041 lcme = &lcm->lcm_entries[i];
4042 if (lcme->lcme_flags & cpu_to_le32(LCME_FL_EXTENSION)) {
4043 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
4052 CERROR("Invalid magic %x\n", lum->lmm_magic);
4057 rc = lod_xattr_del_internal(env, dt, name, th);
4061 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4067 static int lod_get_default_lov_striping(const struct lu_env *env,
4068 struct lod_object *lo,
4069 struct lod_default_striping *lds,
4070 struct dt_allocation_hint *ah);
4073 * Helper function to convert compound layout to compound layout with
4076 * Copy lcm_entries array of \a src to \a tgt. Replace lov_user_md_v1
4077 * components of \a src with lov_user_md_v3 using \a pool.
4079 * \param[in] src source layout
4080 * \param[in] pool pool to use in \a tgt
4081 * \param[out] tgt target layout
4083 static void embed_pool_to_comp_v1(const struct lov_comp_md_v1 *src,
4085 struct lov_comp_md_v1 *tgt)
4088 struct lov_user_md_v1 *lum;
4089 struct lov_user_md_v3 *lum3;
4090 struct lov_comp_md_entry_v1 *entry;
4094 entry = tgt->lcm_entries;
4096 for (i = 0; i < le16_to_cpu(src->lcm_entry_count); i++, entry++) {
4097 *entry = src->lcm_entries[i];
4098 offset = le32_to_cpu(src->lcm_entries[i].lcme_offset);
4099 entry->lcme_offset = cpu_to_le32(offset + shift);
4101 lum = (struct lov_user_md_v1 *)((char *)src + offset);
4102 lum3 = (struct lov_user_md_v3 *)((char *)tgt + offset + shift);
4103 *(struct lov_user_md_v1 *)lum3 = *lum;
4104 if (lum->lmm_pattern & cpu_to_le32(LOV_PATTERN_MDT)) {
4105 lum3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
4107 lum3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4108 entry->lcme_size = cpu_to_le32(sizeof(*lum3));
4109 strscpy(lum3->lmm_pool_name, pool,
4110 sizeof(lum3->lmm_pool_name));
4111 shift += sizeof(*lum3) - sizeof(*lum);
4117 * Set default striping on a directory.
4119 * Sets specified striping on a directory object unless it matches the default
4120 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
4121 * EA. This striping will be used when regular file is being created in this
4123 * If current default striping includes a pool but specifed striping
4124 * does not - retain the pool if it exists.
4126 * \param[in] env execution environment
4127 * \param[in] dt the striped object
4128 * \param[in] buf buffer with the striping
4129 * \param[in] name name of EA
4130 * \param[in] fl xattr flag (see OSD API description)
4131 * \param[in] th transaction handle
4133 * \retval 0 on success
4134 * \retval negative if failed
4136 static int lod_xattr_set_default_lov_on_dir(const struct lu_env *env,
4137 struct dt_object *dt,
4138 const struct lu_buf *buf,
4139 const char *name, int fl,
4142 struct lod_default_striping *lds = lod_lds_buf_get(env);
4143 struct lov_user_md_v1 *v1 = buf->lb_buf;
4144 char pool[LOV_MAXPOOLNAME + 1];
4150 /* get existing striping config */
4151 rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds, NULL);
4155 memset(pool, 0, sizeof(pool));
4156 if (lds->lds_def_striping_set == 1)
4157 lod_layout_get_pool(lds->lds_def_comp_entries,
4158 lds->lds_def_comp_cnt, pool,
4161 is_del = LOVEA_DELETE_VALUES(v1->lmm_stripe_size,
4162 v1->lmm_stripe_count,
4163 v1->lmm_stripe_offset,
4166 /* Retain the pool name if it is not given */
4167 if (v1->lmm_magic == LOV_USER_MAGIC_V1 && pool[0] != '\0' &&
4169 struct lod_thread_info *info = lod_env_info(env);
4170 struct lov_user_md_v3 *v3 = info->lti_ea_store;
4172 memset(v3, 0, sizeof(*v3));
4173 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4174 v3->lmm_pattern = cpu_to_le32(v1->lmm_pattern);
4175 v3->lmm_stripe_count = cpu_to_le32(v1->lmm_stripe_count);
4176 v3->lmm_stripe_offset = cpu_to_le32(v1->lmm_stripe_offset);
4177 v3->lmm_stripe_size = cpu_to_le32(v1->lmm_stripe_size);
4179 strscpy(v3->lmm_pool_name, pool, sizeof(v3->lmm_pool_name));
4181 info->lti_buf.lb_buf = v3;
4182 info->lti_buf.lb_len = sizeof(*v3);
4183 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4185 } else if (v1->lmm_magic == LOV_USER_MAGIC_COMP_V1 &&
4186 pool[0] != '\0' && !is_del) {
4188 * try to retain the pool from default layout if the
4189 * specified component layout does not provide pool
4192 struct lod_thread_info *info = lod_env_info(env);
4193 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
4194 struct lov_comp_md_v1 *comp_v1p;
4195 struct lov_user_md_v1 *lum;
4199 struct lov_comp_md_entry_v1 *entry;
4202 entry_count = le16_to_cpu(comp_v1->lcm_entry_count);
4203 size = sizeof(*comp_v1) +
4204 entry_count * sizeof(comp_v1->lcm_entries[0]);
4205 entry = comp_v1->lcm_entries;
4206 for (i = 0; i < entry_count; i++, entry++) {
4207 offset = le32_to_cpu(entry->lcme_offset);
4208 lum = (struct lov_user_md_v1 *)((char *)comp_v1 +
4210 if (le32_to_cpu(lum->lmm_magic) != LOV_USER_MAGIC_V1)
4211 /* the i-th component includes pool info */
4213 if (lum->lmm_pattern & cpu_to_le32(LOV_PATTERN_MDT))
4214 size += sizeof(struct lov_user_md_v1);
4216 size += sizeof(struct lov_user_md_v3);
4219 if (i == entry_count) {
4221 * re-compose the layout to include the pool for
4224 if (info->lti_ea_store_size < size)
4225 rc = lod_ea_store_resize(info, size);
4228 comp_v1p = info->lti_ea_store;
4229 *comp_v1p = *comp_v1;
4230 comp_v1p->lcm_size = cpu_to_le32(size);
4231 embed_pool_to_comp_v1(comp_v1, pool, comp_v1p);
4233 info->lti_buf.lb_buf = comp_v1p;
4234 info->lti_buf.lb_len = size;
4235 rc = lod_xattr_set_lov_on_dir(env, dt,
4240 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl,
4244 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl, th);
4247 if (lds->lds_def_striping_set == 1 && lds->lds_def_comp_entries != NULL)
4248 lod_free_def_comp_entries(lds);
4254 * Set default striping on a directory object.
4256 * Sets specified striping on a directory object unless it matches the default
4257 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
4258 * EA. This striping will be used when a new directory is being created in the
4261 * \param[in] env execution environment
4262 * \param[in] dt the striped object
4263 * \param[in] buf buffer with the striping
4264 * \param[in] name name of EA
4265 * \param[in] fl xattr flag (see OSD API description)
4266 * \param[in] th transaction handle
4268 * \retval 0 on success
4269 * \retval negative if failed
4271 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
4272 struct dt_object *dt,
4273 const struct lu_buf *buf,
4274 const char *name, int fl,
4277 struct lmv_user_md_v1 *lum;
4282 LASSERT(buf != NULL && buf->lb_buf != NULL);
4286 "set default stripe_count # %u stripe_offset %d hash %u\n",
4287 le32_to_cpu(lum->lum_stripe_count),
4288 (int)le32_to_cpu(lum->lum_stripe_offset),
4289 le32_to_cpu(lum->lum_hash_type));
4291 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
4292 le32_to_cpu(lum->lum_stripe_offset)) &&
4293 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
4294 rc = lod_xattr_del_internal(env, dt, name, th);
4298 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4307 * Turn directory into a striped directory.
4309 * During replay the client sends the striping created before MDT
4310 * failure, then the layer above LOD sends this defined striping
4311 * using ->do_xattr_set(), so LOD uses this method to replay creation
4312 * of the stripes. Notice the original information for the striping
4313 * (#stripes, FIDs, etc) was transferred in declare path.
4315 * \param[in] env execution environment
4316 * \param[in] dt the striped object
4317 * \param[in] buf buf lmv_user_md for create, or lmv_mds_md for replay
4318 * \param[in] name not used currently
4319 * \param[in] fl xattr flag (see OSD API description)
4320 * \param[in] th transaction handle
4322 * \retval 0 on success
4323 * \retval negative if failed
4325 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
4326 const struct lu_buf *buf, const char *name,
4327 int fl, struct thandle *th)
4329 struct lod_object *lo = lod_dt_obj(dt);
4330 struct lod_thread_info *info = lod_env_info(env);
4331 struct lu_attr *attr = &info->lti_attr;
4332 struct dt_object_format *dof = &info->lti_format;
4333 struct lu_buf lmv_buf;
4334 struct lu_buf slave_lmv_buf;
4335 struct lmv_user_md *lum = buf->lb_buf;
4336 struct lmv_mds_md_v1 *lmm;
4337 struct lmv_mds_md_v1 *slave_lmm = NULL;
4338 struct dt_insert_rec *rec = &info->lti_dt_rec;
4343 /* lum is used to know whether it's replay */
4345 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4348 /* The stripes are supposed to be allocated in declare phase,
4349 * if there are no stripes being allocated, it will skip */
4350 if (lo->ldo_dir_stripe_count == 0) {
4351 if (lo->ldo_is_foreign) {
4352 rc = lod_sub_xattr_set(env, dt_object_child(dt), buf,
4353 XATTR_NAME_LMV, fl, th);
4360 rc = dt_attr_get(env, dt_object_child(dt), attr);
4364 attr->la_valid &= LA_ATIME | LA_MTIME | LA_CTIME | LA_FLAGS |
4365 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
4366 dof->dof_type = DFT_DIR;
4368 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
4371 lmm = lmv_buf.lb_buf;
4373 OBD_ALLOC_PTR(slave_lmm);
4374 if (slave_lmm == NULL)
4377 lod_prep_slave_lmv_md(slave_lmm, lmm);
4378 slave_lmv_buf.lb_buf = slave_lmm;
4379 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
4381 rec->rec_type = S_IFDIR;
4382 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4383 struct dt_object *dto = lo->ldo_stripe[i];
4384 char *stripe_name = info->lti_key;
4385 struct lu_name *sname;
4386 struct linkea_data ldata = { NULL };
4387 struct lu_buf linkea_buf;
4388 bool stripe_created = false;
4390 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
4394 /* fail a remote stripe creation */
4395 if (i && CFS_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_CREATE))
4398 /* if it's replay by client request, and stripe exists on remote
4399 * MDT, it means mkdir was partially executed: stripe was
4400 * created on remote MDT successfully, but target not in last
4403 if (unlikely((le32_to_cpu(lum->lum_magic) == LMV_MAGIC_V1) &&
4404 dt_object_exists(dto) && dt_object_remote(dto)))
4405 stripe_created = true;
4407 /* don't create stripe if:
4408 * 1. it's source stripe of migrating directory
4409 * 2. it's existed stripe of splitting directory
4411 if ((lod_is_migrating(lo) && i >= lo->ldo_dir_migrate_offset) ||
4412 (lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) {
4413 if (!dt_object_exists(dto))
4414 GOTO(out, rc = -EINVAL);
4415 } else if (!stripe_created) {
4416 dt_write_lock(env, dto, DT_TGT_CHILD);
4417 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
4419 dt_write_unlock(env, dto);
4423 rc = lod_sub_ref_add(env, dto, th);
4424 dt_write_unlock(env, dto);
4428 rec->rec_fid = lu_object_fid(&dto->do_lu);
4429 rc = lod_sub_insert(env, dto,
4430 (const struct dt_rec *)rec,
4431 (const struct dt_key *)dot, th);
4436 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
4437 cfs_fail_val != i) {
4438 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
4440 slave_lmm->lmv_master_mdt_index =
4443 slave_lmm->lmv_master_mdt_index =
4446 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
4447 XATTR_NAME_LMV, 0, th);
4452 /* don't insert stripe if it's existed stripe of splitting
4453 * directory (this directory is striped).
4454 * NB, plain directory will insert itself as the first
4457 if (lod_is_splitting(lo) && lo->ldo_dir_split_offset > 1 &&
4458 lo->ldo_dir_split_offset > i)
4461 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
4463 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4464 PFID(lu_object_fid(&dto->do_lu)), i + 1);
4466 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4467 PFID(lu_object_fid(&dto->do_lu)), i);
4469 if (!stripe_created) {
4470 rec->rec_fid = lu_object_fid(&dt->do_lu);
4471 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
4472 (const struct dt_key *)dotdot, th);
4476 sname = lod_name_get(env, stripe_name,
4477 strlen(stripe_name));
4478 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
4479 sname, lu_object_fid(&dt->do_lu));
4483 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
4484 linkea_buf.lb_len = ldata.ld_leh->leh_len;
4485 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
4486 XATTR_NAME_LINK, 0, th);
4491 rec->rec_fid = lu_object_fid(&dto->do_lu);
4492 rc = lod_sub_insert(env, dt_object_child(dt),
4493 (const struct dt_rec *)rec,
4494 (const struct dt_key *)stripe_name, th);
4498 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
4503 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
4504 rc = lod_sub_xattr_set(env, dt_object_child(dt),
4505 &lmv_buf, XATTR_NAME_LMV, fl, th);
4507 if (slave_lmm != NULL)
4508 OBD_FREE_PTR(slave_lmm);
4514 * Helper function to declare/execute creation of a striped directory
4516 * Called in declare/create object path, prepare striping for a directory
4517 * and prepare defaults data striping for the objects to be created in
4518 * that directory. Notice the function calls "declaration" or "execution"
4519 * methods depending on \a declare param. This is a consequence of the
4520 * current approach while we don't have natural distributed transactions:
4521 * we basically execute non-local updates in the declare phase. So, the
4522 * arguments for the both phases are the same and this is the reason for
4523 * this function to exist.
4525 * \param[in] env execution environment
4526 * \param[in] dt object
4527 * \param[in] attr attributes the stripes will be created with
4528 * \param[in] lmu lmv_user_md if MDT indices are specified
4529 * \param[in] dof format of stripes (see OSD API description)
4530 * \param[in] th transaction handle
4531 * \param[in] declare where to call "declare" or "execute" methods
4533 * \retval 0 on success
4534 * \retval negative if failed
4536 static int lod_dir_striping_create_internal(const struct lu_env *env,
4537 struct dt_object *dt,
4538 struct lu_attr *attr,
4539 const struct lu_buf *lmu,
4540 struct dt_object_format *dof,
4544 struct lod_thread_info *info = lod_env_info(env);
4545 struct lod_object *lo = lod_dt_obj(dt);
4546 const struct lod_default_striping *lds = lo->ldo_def_striping;
4550 LASSERT(ergo(lds != NULL,
4551 lds->lds_def_striping_set ||
4552 lds->lds_dir_def_striping_set));
4555 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
4556 lo->ldo_dir_stripe_offset)) {
4558 /* mkdir by default LMV */
4559 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
4560 int stripe_count = lo->ldo_dir_stripe_count;
4562 if (info->lti_ea_store_size < sizeof(*v1)) {
4563 rc = lod_ea_store_resize(info, sizeof(*v1));
4566 v1 = info->lti_ea_store;
4569 memset(v1, 0, sizeof(*v1));
4570 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
4571 v1->lum_stripe_count = cpu_to_le32(stripe_count);
4572 v1->lum_stripe_offset =
4573 cpu_to_le32(lo->ldo_dir_stripe_offset);
4575 info->lti_buf.lb_buf = v1;
4576 info->lti_buf.lb_len = sizeof(*v1);
4577 lmu = &info->lti_buf;
4581 rc = lod_declare_xattr_set_lmv(env, dt, attr, lmu, dof,
4584 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
4588 } else if (lmu->lb_buf) {
4589 /* foreign LMV EA case */
4591 struct lmv_foreign_md *lfm = lmu->lb_buf;
4593 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN)
4594 rc = lod_declare_xattr_set_lmv(env, dt, attr,
4596 } else if (lo->ldo_is_foreign) {
4597 LASSERT(lo->ldo_foreign_lmv != NULL &&
4598 lo->ldo_foreign_lmv_size > 0);
4599 info->lti_buf.lb_buf = lo->ldo_foreign_lmv;
4600 info->lti_buf.lb_len = lo->ldo_foreign_lmv_size;
4601 lmu = &info->lti_buf;
4602 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
4607 /* Transfer default LMV striping from the parent */
4608 if (lds != NULL && lds->lds_dir_def_striping_set &&
4609 lds->lds_dir_def_max_inherit != LMV_INHERIT_END &&
4610 lds->lds_dir_def_max_inherit != LMV_INHERIT_NONE &&
4611 !(LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
4612 lds->lds_dir_def_stripe_offset) &&
4613 le32_to_cpu(lds->lds_dir_def_hash_type) !=
4614 LMV_HASH_TYPE_UNKNOWN)) {
4615 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
4617 if (info->lti_ea_store_size < sizeof(*v1)) {
4618 rc = lod_ea_store_resize(info, sizeof(*v1));
4621 v1 = info->lti_ea_store;
4624 memset(v1, 0, sizeof(*v1));
4625 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
4626 v1->lum_stripe_count =
4627 cpu_to_le32(lds->lds_dir_def_stripe_count);
4628 v1->lum_stripe_offset =
4629 cpu_to_le32(lds->lds_dir_def_stripe_offset);
4631 cpu_to_le32(lds->lds_dir_def_hash_type);
4632 v1->lum_max_inherit =
4633 lmv_inherit_next(lds->lds_dir_def_max_inherit);
4634 v1->lum_max_inherit_rr =
4635 lmv_inherit_rr_next(lds->lds_dir_def_max_inherit_rr);
4637 info->lti_buf.lb_buf = v1;
4638 info->lti_buf.lb_len = sizeof(*v1);
4640 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4641 XATTR_NAME_DEFAULT_LMV,
4644 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
4646 XATTR_NAME_DEFAULT_LMV, 0,
4652 /* Transfer default LOV striping from the parent */
4653 if (lds != NULL && lds->lds_def_striping_set &&
4654 lds->lds_def_comp_cnt != 0) {
4655 struct lov_mds_md *lmm;
4656 int lmm_size = lod_comp_md_size(lo, true);
4658 if (info->lti_ea_store_size < lmm_size) {
4659 rc = lod_ea_store_resize(info, lmm_size);
4663 lmm = info->lti_ea_store;
4665 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
4669 info->lti_buf.lb_buf = lmm;
4670 info->lti_buf.lb_len = lmm_size;
4673 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4674 XATTR_NAME_LOV, 0, th);
4676 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4677 XATTR_NAME_LOV, 0, th);
4682 /* ldo_def_striping is not allocated, clear after use, in case directory
4683 * layout is changed later.
4686 lo->ldo_def_striping = NULL;
4691 static int lod_declare_dir_striping_create(const struct lu_env *env,
4692 struct dt_object *dt,
4693 struct lu_attr *attr,
4695 struct dt_object_format *dof,
4698 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
4702 static int lod_dir_striping_create(const struct lu_env *env,
4703 struct dt_object *dt,
4704 struct lu_attr *attr,
4705 const struct lu_buf *lmu,
4706 struct dt_object_format *dof,
4709 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
4714 * Make LOV EA for striped object.
4716 * Generate striping information and store it in the LOV EA of the given
4717 * object. The caller must ensure nobody else is calling the function
4718 * against the object concurrently. The transaction must be started.
4719 * FLDB service must be running as well; it's used to map FID to the target,
4720 * which is stored in LOV EA.
4722 * \param[in] env execution environment for this thread
4723 * \param[in] lo LOD object
4724 * \param[in] th transaction handle
4726 * \retval 0 if LOV EA is stored successfully
4727 * \retval negative error number on failure
4729 static int lod_generate_and_set_lovea(const struct lu_env *env,
4730 struct lod_object *lo,
4733 struct lod_thread_info *info = lod_env_info(env);
4734 struct dt_object *next = dt_object_child(&lo->ldo_obj);
4735 struct lov_mds_md_v1 *lmm;
4741 if (lo->ldo_comp_cnt == 0 && !lo->ldo_is_foreign) {
4742 lod_striping_free_nolock(env, lo);
4743 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
4747 lmm_size = lod_comp_md_size(lo, false);
4748 if (info->lti_ea_store_size < lmm_size) {
4749 rc = lod_ea_store_resize(info, lmm_size);
4753 lmm = info->lti_ea_store;
4755 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
4759 info->lti_buf.lb_buf = lmm;
4760 info->lti_buf.lb_len = lmm_size;
4761 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
4762 XATTR_NAME_LOV, 0, th);
4766 static __u32 lod_gen_component_id(struct lod_object *lo,
4767 int mirror_id, int comp_idx);
4770 * Repeat an existing component
4772 * Creates a new layout by replicating an existing component. Uses striping
4773 * policy from previous component as a template for the striping for the new
4776 * New component starts with zero length, will be extended (or removed) before
4777 * returning layout to client.
4779 * NB: Reallocates layout components array (lo->ldo_comp_entries), invalidating
4780 * any pre-existing pointers to components. Handle with care.
4782 * \param[in] env execution environment for this thread
4783 * \param[in,out] lo object to update the layout of
4784 * \param[in] index index of component to copy
4786 * \retval 0 on success
4787 * \retval negative errno on error
4789 static int lod_layout_repeat_comp(const struct lu_env *env,
4790 struct lod_object *lo, int index)
4792 struct lod_layout_component *lod_comp;
4793 struct lod_layout_component *new_comp = NULL;
4794 struct lod_layout_component *comp_array;
4795 int rc = 0, i, new_cnt = lo->ldo_comp_cnt + 1;
4800 lod_comp = &lo->ldo_comp_entries[index];
4801 LASSERT(lod_comp_inited(lod_comp) && lod_comp->llc_id != LCME_ID_INVAL);
4803 CDEBUG(D_LAYOUT, "repeating component %d\n", index);
4805 OBD_ALLOC_PTR_ARRAY(comp_array, new_cnt);
4806 if (comp_array == NULL)
4807 GOTO(out, rc = -ENOMEM);
4809 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4810 memcpy(&comp_array[i + offset], &lo->ldo_comp_entries[i],
4811 sizeof(*comp_array));
4813 /* Duplicate this component in to the next slot */
4815 new_comp = &comp_array[i + 1];
4816 memcpy(&comp_array[i + 1], &lo->ldo_comp_entries[i],
4817 sizeof(*comp_array));
4818 /* We must now skip this new component when copying */
4823 /* Set up copied component */
4824 new_comp->llc_flags &= ~LCME_FL_INIT;
4825 new_comp->llc_stripe = NULL;
4826 new_comp->llc_stripes_allocated = 0;
4827 new_comp->llc_ost_indices = NULL;
4828 new_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4829 /* for uninstantiated components, layout gen stores default stripe
4831 new_comp->llc_layout_gen = lod_comp->llc_stripe_offset;
4832 /* This makes the repeated component zero-length, placed at the end of
4833 * the preceding component */
4834 new_comp->llc_extent.e_start = new_comp->llc_extent.e_end;
4835 new_comp->llc_timestamp = lod_comp->llc_timestamp;
4836 new_comp->llc_pool = NULL;
4838 rc = lod_set_pool(&new_comp->llc_pool, lod_comp->llc_pool);
4842 if (new_comp->llc_ostlist.op_array) {
4843 __u32 *op_array = NULL;
4845 OBD_ALLOC(op_array, new_comp->llc_ostlist.op_size);
4847 GOTO(out, rc = -ENOMEM);
4848 memcpy(op_array, &new_comp->llc_ostlist.op_array,
4849 new_comp->llc_ostlist.op_size);
4850 new_comp->llc_ostlist.op_array = op_array;
4853 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
4854 lo->ldo_comp_entries = comp_array;
4855 lo->ldo_comp_cnt = new_cnt;
4857 /* Generate an id for the new component */
4858 mirror_id = mirror_id_of(new_comp->llc_id);
4859 new_comp->llc_id = LCME_ID_INVAL;
4860 new_comp->llc_id = lod_gen_component_id(lo, mirror_id, index + 1);
4861 if (new_comp->llc_id == LCME_ID_INVAL)
4862 GOTO(out, rc = -ERANGE);
4867 OBD_FREE_PTR_ARRAY(comp_array, new_cnt);
4872 static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
4876 /* clear memory region that will be used for layout change */
4877 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
4878 info->lti_count = 0;
4880 if (info->lti_comp_size >= comp_cnt)
4883 if (info->lti_comp_size > 0) {
4884 OBD_FREE_PTR_ARRAY(info->lti_comp_idx, info->lti_comp_size);
4885 info->lti_comp_size = 0;
4888 OBD_ALLOC_PTR_ARRAY(info->lti_comp_idx, comp_cnt);
4889 if (!info->lti_comp_idx)
4892 info->lti_comp_size = comp_cnt;
4897 * Prepare new layout minus deleted components
4899 * Removes components marked for deletion (LCME_ID_INVAL) by copying to a new
4900 * layout and skipping those components. Removes stripe objects if any exist.
4903 * Reallocates layout components array (lo->ldo_comp_entries), invalidating
4904 * any pre-existing pointers to components.
4906 * Caller is responsible for updating mirror end (ldo_mirror[].lme_end).
4908 * \param[in] env execution environment for this thread
4909 * \param[in,out] lo object to update the layout of
4910 * \param[in] th transaction handle for this operation
4912 * \retval # of components deleted
4913 * \retval negative errno on error
4915 static int lod_layout_del_prep_layout(const struct lu_env *env,
4916 struct lod_object *lo,
4919 struct lod_layout_component *lod_comp;
4920 struct lod_thread_info *info = lod_env_info(env);
4921 int rc = 0, i, j, deleted = 0;
4925 LASSERT(lo->ldo_is_composite);
4926 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
4928 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
4932 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4933 lod_comp = &lo->ldo_comp_entries[i];
4935 if (lod_comp->llc_id != LCME_ID_INVAL) {
4936 /* Build array of things to keep */
4937 info->lti_comp_idx[info->lti_count++] = i;
4941 if (lod_comp->llc_magic == LOV_MAGIC_FOREIGN)
4944 lod_obj_set_pool(lo, i, NULL);
4945 if (lod_comp->llc_ostlist.op_array) {
4946 OBD_FREE(lod_comp->llc_ostlist.op_array,
4947 lod_comp->llc_ostlist.op_size);
4948 lod_comp->llc_ostlist.op_array = NULL;
4949 lod_comp->llc_ostlist.op_size = 0;
4953 CDEBUG(D_LAYOUT, "deleting comp %d, left %d\n", i,
4954 lo->ldo_comp_cnt - deleted);
4956 /* No striping info for this component */
4957 if (lod_comp->llc_stripe == NULL)
4960 LASSERT(lod_comp->llc_stripe_count > 0);
4961 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4962 struct dt_object *obj = lod_comp->llc_stripe[j];
4967 /* components which are not init have no sub objects
4969 if (lod_comp_inited(lod_comp)) {
4970 rc = lod_sub_destroy(env, obj, th);
4975 lu_object_put(env, &obj->do_lu);
4976 lod_comp->llc_stripe[j] = NULL;
4978 OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe,
4979 lod_comp->llc_stripes_allocated);
4980 lod_comp->llc_stripe = NULL;
4981 OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices,
4982 lod_comp->llc_stripes_allocated);
4983 lod_comp->llc_ost_indices = NULL;
4984 lod_comp->llc_stripes_allocated = 0;
4987 /* info->lti_count has the amount of left components */
4988 LASSERTF(info->lti_count >= 0 && info->lti_count < lo->ldo_comp_cnt,
4989 "left = %d, lo->ldo_comp_cnt %d\n", (int)info->lti_count,
4990 (int)lo->ldo_comp_cnt);
4992 if (info->lti_count > 0) {
4993 struct lod_layout_component *comp_array;
4995 OBD_ALLOC_PTR_ARRAY(comp_array, info->lti_count);
4996 if (comp_array == NULL)
4997 GOTO(out, rc = -ENOMEM);
4999 for (i = 0; i < info->lti_count; i++) {
5000 memcpy(&comp_array[i],
5001 &lo->ldo_comp_entries[info->lti_comp_idx[i]],
5002 sizeof(*comp_array));
5005 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
5006 lo->ldo_comp_entries = comp_array;
5007 lo->ldo_comp_cnt = info->lti_count;
5009 lod_free_comp_entries(lo);
5014 return rc ? rc : deleted;
5018 * Delete layout component(s)
5020 * This function sets up the layout data in the env and does the setattrs
5021 * required to write out the new layout. The layout itself is modified in
5022 * lod_layout_del_prep_layout.
5024 * \param[in] env execution environment for this thread
5025 * \param[in] dt object
5026 * \param[in] th transaction handle
5028 * \retval 0 on success
5029 * \retval negative error number on failure
5031 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
5034 struct lod_object *lo = lod_dt_obj(dt);
5035 struct dt_object *next = dt_object_child(dt);
5036 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
5039 LASSERT(lo->ldo_mirror_count == 1);
5041 mutex_lock(&lo->ldo_layout_mutex);
5043 rc = lod_layout_del_prep_layout(env, lo, th);
5047 /* Only do this if we didn't delete all components */
5048 if (lo->ldo_comp_cnt > 0) {
5049 lo->ldo_mirrors[0].lme_end = lo->ldo_comp_cnt - 1;
5050 lod_obj_inc_layout_gen(lo);
5053 LASSERT(dt_object_exists(dt));
5054 rc = dt_attr_get(env, next, attr);
5058 if (attr->la_size > 0) {
5060 attr->la_valid = LA_SIZE;
5061 rc = lod_sub_attr_set(env, next, attr, th);
5066 rc = lod_generate_and_set_lovea(env, lo, th);
5070 lod_striping_free_nolock(env, lo);
5072 mutex_unlock(&lo->ldo_layout_mutex);
5079 * Implementation of dt_object_operations::do_xattr_set.
5081 * Sets specified extended attribute on the object. Three types of EAs are
5083 * LOV EA - stores striping for a regular file or default striping (when set
5085 * LMV EA - stores a marker for the striped directories
5086 * DMV EA - stores default directory striping
5088 * When striping is applied to a non-striped existing object (this is called
5089 * late striping), then LOD notices the caller wants to turn the object into a
5090 * striped one. The stripe objects are created and appropriate EA is set:
5091 * LOV EA storing all the stripes directly or LMV EA storing just a small header
5092 * with striping configuration.
5094 * \see dt_object_operations::do_xattr_set() in the API description for details.
5096 static int lod_xattr_set(const struct lu_env *env,
5097 struct dt_object *dt, const struct lu_buf *buf,
5098 const char *name, int fl, struct thandle *th)
5100 struct lod_thread_info *info = lod_env_info(env);
5101 struct dt_object *next = dt_object_child(dt);
5102 struct lu_attr *layout_attr = &info->lti_layout_attr;
5103 struct lod_object *lo = lod_dt_obj(dt);
5104 struct lod_obj_stripe_cb_data data = { {0} };
5109 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
5110 !strcmp(name, XATTR_NAME_LMV)) {
5112 case LU_XATTR_CREATE:
5113 rc = lod_dir_striping_create(env, dt, NULL, buf, NULL,
5117 case LU_XATTR_REPLACE:
5118 rc = lod_dir_layout_set(env, dt, buf, fl, th);
5125 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
5126 strcmp(name, XATTR_NAME_LOV) == 0) {
5127 rc = lod_xattr_set_default_lov_on_dir(env, dt, buf, name, fl,
5130 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
5131 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
5133 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
5136 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
5137 (strcmp(name, XATTR_NAME_LOV) == 0 ||
5138 strcmp(name, XATTR_LUSTRE_LOV) == 0 ||
5139 allowed_lustre_lov(name))) {
5140 /* layout has been changed by others in the transaction */
5141 rc = lod_check_layout_gen_intrans(info, lo);
5144 "%s: obj "DFID" gen changed from %d to %d in transaction, retry the transaction\n",
5145 dt->do_lu.lo_dev->ld_obd->obd_name,
5146 PFID(lu_object_fid(&dt->do_lu)),
5147 info->lti_gen[rc - 1], lo->ldo_layout_gen);
5151 /* in case of lov EA swap, just set it
5152 * if not, it is a replay so check striping match what we
5153 * already have during req replay, declare_xattr_set()
5154 * defines striping, then create() does the work */
5155 if (fl & LU_XATTR_REPLACE) {
5156 /* free stripes, then update disk */
5157 lod_striping_free(env, lod_dt_obj(dt));
5159 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
5160 } else if (fl & LU_XATTR_SPLIT) {
5161 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
5165 rc = lod_striping_reload(env, lo, buf, LVF_ALL_STALE);
5169 if (lo->ldo_mirror_count > 1 &&
5170 layout_attr->la_valid & LA_LAYOUT_VERSION) {
5172 layout_attr->la_layout_version =
5174 data.locd_attr = layout_attr;
5175 data.locd_declare = false;
5176 data.locd_stripe_cb =
5177 lod_obj_stripe_attr_set_cb;
5178 rc = lod_obj_for_each_stripe(env, lo, th,
5183 } else if (fl & LU_XATTR_PURGE) {
5184 rc = lod_layout_purge(env, dt, buf, th);
5185 } else if (dt_object_remote(dt)) {
5186 /* This only happens during migration, see
5187 * mdd_migrate_create(), in which Master MDT will
5188 * create a remote target object, and only set
5189 * (migrating) stripe EA on the remote object,
5190 * and does not need creating each stripes. */
5191 rc = lod_sub_xattr_set(env, next, buf, name,
5193 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
5194 /* delete component(s) */
5195 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
5196 rc = lod_layout_del(env, dt, th);
5199 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
5200 * it's going to create file with specified
5201 * component(s), the striping must have not being
5202 * cached in this case;
5204 * Otherwise, it's going to add/change component(s) to
5205 * an existing file, the striping must have been cached
5208 if (!(fl & LU_XATTR_MERGE))
5209 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
5210 !strcmp(name, XATTR_NAME_LOV),
5211 !lod_dt_obj(dt)->ldo_comp_cached));
5213 rc = lod_striped_create(env, dt, NULL, NULL, th);
5217 if (fl & LU_XATTR_MERGE && lo->ldo_mirror_count > 1 &&
5218 layout_attr->la_valid & LA_LAYOUT_VERSION) {
5219 /* mirror merge exec phase */
5220 layout_attr->la_layout_version =
5222 data.locd_attr = layout_attr;
5223 data.locd_declare = false;
5224 data.locd_stripe_cb =
5225 lod_obj_stripe_attr_set_cb;
5226 rc = lod_obj_for_each_stripe(env, lo, th,
5233 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
5234 rc = lod_replace_parent_fid(env, dt, buf, th, false);
5239 /* then all other xattr */
5240 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
5246 * Implementation of dt_object_operations::do_declare_xattr_del.
5248 * \see dt_object_operations::do_declare_xattr_del() in the API description
5251 static int lod_declare_xattr_del(const struct lu_env *env,
5252 struct dt_object *dt, const char *name,
5255 struct lod_object *lo = lod_dt_obj(dt);
5256 struct dt_object *next = dt_object_child(dt);
5261 rc = lod_sub_declare_xattr_del(env, next, name, th);
5265 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
5268 /* NB: don't delete stripe LMV, because when we do this, normally we
5269 * will remove stripes, besides, if directory LMV is corrupt, this will
5270 * prevent deleting its LMV and fixing it (via LFSCK).
5272 if (!strcmp(name, XATTR_NAME_LMV))
5275 rc = lod_striping_load(env, lo);
5279 if (lo->ldo_dir_stripe_count == 0)
5282 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5283 struct dt_object *dto = lo->ldo_stripe[i];
5288 if (!dt_object_exists(dto))
5291 rc = lod_sub_declare_xattr_del(env, dto, name, th);
5300 * Implementation of dt_object_operations::do_xattr_del.
5302 * If EA storing a regular striping is being deleted, then release
5303 * all the references to the stripe objects in core.
5305 * \see dt_object_operations::do_xattr_del() in the API description for details.
5307 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
5308 const char *name, struct thandle *th)
5314 if (!strcmp(name, XATTR_NAME_LOV) || !strcmp(name, XATTR_NAME_LMV))
5315 lod_striping_free(env, lod_dt_obj(dt));
5317 rc = lod_xattr_del_internal(env, dt, name, th);
5323 * Implementation of dt_object_operations::do_xattr_list.
5325 * \see dt_object_operations::do_xattr_list() in the API description
5328 static int lod_xattr_list(const struct lu_env *env,
5329 struct dt_object *dt, const struct lu_buf *buf)
5331 return dt_xattr_list(env, dt_object_child(dt), buf);
5334 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
5336 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
5340 * Copy OST list from layout provided by user.
5342 * \param[in] lod_comp layout_component to be filled
5343 * \param[in] v3 LOV EA V3 user data
5345 * \retval 0 on success
5346 * \retval negative if failed
5348 int lod_comp_copy_ost_lists(struct lod_layout_component *lod_comp,
5349 struct lov_user_md_v3 *v3)
5355 if (v3->lmm_stripe_offset == LOV_OFFSET_DEFAULT)
5356 v3->lmm_stripe_offset = v3->lmm_objects[0].l_ost_idx;
5358 if (lod_comp->llc_ostlist.op_array) {
5359 if (lod_comp->llc_ostlist.op_size >=
5360 v3->lmm_stripe_count * sizeof(__u32)) {
5361 lod_comp->llc_ostlist.op_count =
5362 v3->lmm_stripe_count;
5365 OBD_FREE(lod_comp->llc_ostlist.op_array,
5366 lod_comp->llc_ostlist.op_size);
5369 /* copy ost list from lmm */
5370 lod_comp->llc_ostlist.op_count = v3->lmm_stripe_count;
5371 lod_comp->llc_ostlist.op_size = v3->lmm_stripe_count * sizeof(__u32);
5372 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
5373 lod_comp->llc_ostlist.op_size);
5374 if (!lod_comp->llc_ostlist.op_array)
5377 for (j = 0; j < v3->lmm_stripe_count; j++) {
5378 lod_comp->llc_ostlist.op_array[j] =
5379 v3->lmm_objects[j].l_ost_idx;
5387 * Get default striping.
5389 * \param[in] env execution environment
5390 * \param[in] lo object
5391 * \param[out] lds default striping
5393 * \retval 0 on success
5394 * \retval negative if failed
5396 static int lod_get_default_lov_striping(const struct lu_env *env,
5397 struct lod_object *lo,
5398 struct lod_default_striping *lds,
5399 struct dt_allocation_hint *dah)
5401 struct lod_thread_info *info = lod_env_info(env);
5402 struct lov_user_md_v1 *v1 = NULL;
5403 struct lov_user_md_v3 *v3 = NULL;
5404 struct lov_comp_md_v1 *lcm = NULL;
5406 int append_stripe_count = dah != NULL ? dah->dah_append_stripe_count : 0;
5407 const char *append_pool = (dah != NULL &&
5408 dah->dah_append_pool != NULL &&
5409 dah->dah_append_pool[0] != '\0') ?
5410 dah->dah_append_pool : NULL;
5411 __u16 entry_count = 1;
5412 __u16 mirror_count = 0;
5413 bool want_composite = false;
5418 lds->lds_def_striping_set = 0;
5420 rc = lod_get_lov_ea(env, lo);
5424 if (rc < (typeof(rc))sizeof(struct lov_user_md))
5427 magic = *(__u32 *)info->lti_ea_store;
5428 if (magic == __swab32(LOV_USER_MAGIC_V1)) {
5429 lustre_swab_lov_user_md_v1(info->lti_ea_store);
5430 } else if (magic == __swab32(LOV_USER_MAGIC_V3)) {
5431 lustre_swab_lov_user_md_v3(info->lti_ea_store);
5432 } else if (magic == __swab32(LOV_USER_MAGIC_SPECIFIC)) {
5433 v3 = (struct lov_user_md_v3 *)info->lti_ea_store;
5434 lustre_swab_lov_user_md_v3(v3);
5435 lustre_swab_lov_user_md_objects(v3->lmm_objects,
5436 v3->lmm_stripe_count);
5437 } else if (magic == __swab32(LOV_USER_MAGIC_COMP_V1) ||
5438 magic == __swab32(LOV_USER_MAGIC_SEL)) {
5439 lustre_swab_lov_comp_md_v1(info->lti_ea_store);
5445 case LOV_USER_MAGIC_SPECIFIC:
5446 v1 = info->lti_ea_store;
5448 case LOV_MAGIC_COMP_V1:
5450 lcm = info->lti_ea_store;
5451 entry_count = lcm->lcm_entry_count;
5452 if (entry_count == 0)
5455 mirror_count = lcm->lcm_mirror_count + 1;
5456 want_composite = true;
5462 if (append_stripe_count != 0 || append_pool != NULL) {
5465 want_composite = false;
5468 /* realloc default comp entries if necessary */
5469 rc = lod_def_striping_comp_resize(lds, entry_count);
5473 lds->lds_def_comp_cnt = entry_count;
5474 lds->lds_def_striping_is_composite = want_composite;
5475 lds->lds_def_mirror_cnt = mirror_count;
5477 for (i = 0; i < entry_count; i++) {
5478 struct lod_layout_component *llc = &lds->lds_def_comp_entries[i];
5482 * reset llc values, llc_stripes is always NULL in the
5483 * default striping template, llc_pool will be reset
5484 * later below using lod_set_pool().
5486 * XXX At this point llc_pool may point to valid (!)
5487 * kmalloced strings from previous RPCs.
5489 memset(llc, 0, offsetof(typeof(*llc), llc_pool));
5492 v1 = (struct lov_user_md *)((char *)lcm +
5493 lcm->lcm_entries[i].lcme_offset);
5495 if (want_composite) {
5496 llc->llc_extent = lcm->lcm_entries[i].lcme_extent;
5497 /* We only inherit certain flags from the layout */
5498 llc->llc_flags = lcm->lcm_entries[i].lcme_flags &
5499 LCME_TEMPLATE_FLAGS;
5503 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",
5504 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
5507 v1->lmm_stripe_count,
5508 v1->lmm_stripe_size,
5509 v1->lmm_stripe_offset,
5511 append_stripe_count);
5513 if (!lov_pattern_supported(v1->lmm_pattern) &&
5514 !(v1->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
5515 lod_free_def_comp_entries(lds);
5519 llc->llc_stripe_count = v1->lmm_stripe_count;
5520 llc->llc_stripe_size = v1->lmm_stripe_size;
5521 llc->llc_stripe_offset = v1->lmm_stripe_offset;
5522 llc->llc_pattern = v1->lmm_pattern;
5524 if (append_stripe_count != 0 || append_pool != NULL)
5525 llc->llc_pattern = LOV_PATTERN_RAID0;
5527 if (append_stripe_count != 0)
5528 llc->llc_stripe_count = append_stripe_count;
5531 if (append_pool != NULL) {
5533 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
5534 /* XXX: sanity check here */
5535 v3 = (struct lov_user_md_v3 *)v1;
5536 if (v3->lmm_pool_name[0] != '\0')
5537 pool = v3->lmm_pool_name;
5540 lod_set_pool(&llc->llc_pool, pool);
5542 if (v1->lmm_magic == LOV_USER_MAGIC_SPECIFIC &&
5543 append_stripe_count == 0 &&
5544 append_pool == NULL) {
5545 v3 = (struct lov_user_md_v3 *)v1;
5546 rc = lod_comp_copy_ost_lists(llc, v3);
5549 } else if (llc->llc_ostlist.op_array &&
5550 llc->llc_ostlist.op_count) {
5551 for (j = 0; j < llc->llc_ostlist.op_count; j++)
5552 llc->llc_ostlist.op_array[j] = -1;
5553 llc->llc_ostlist.op_count = 0;
5557 lds->lds_def_striping_set = 1;
5561 static inline void lod_lum2lds(struct lod_default_striping *lds,
5562 const struct lmv_user_md *lum)
5564 lds->lds_dir_def_stripe_count = le32_to_cpu(lum->lum_stripe_count);
5565 lds->lds_dir_def_stripe_offset = le32_to_cpu(lum->lum_stripe_offset);
5566 lds->lds_dir_def_hash_type = le32_to_cpu(lum->lum_hash_type);
5567 lds->lds_dir_def_max_inherit = lum->lum_max_inherit;
5568 lds->lds_dir_def_max_inherit_rr = lum->lum_max_inherit_rr;
5569 lds->lds_dir_def_striping_set = 1;
5573 * Get default directory striping.
5575 * \param[in] env execution environment
5576 * \param[in] lo object
5577 * \param[out] lds default striping
5579 * \retval 0 on success
5580 * \retval negative if failed
5582 static int lod_get_default_lmv_striping(const struct lu_env *env,
5583 struct lod_object *lo,
5584 struct lod_default_striping *lds)
5586 struct lmv_user_md *lmu;
5589 lds->lds_dir_def_striping_set = 0;
5591 rc = lod_get_default_lmv_ea(env, lo);
5595 if (rc >= (int)sizeof(*lmu)) {
5596 struct lod_thread_info *info = lod_env_info(env);
5598 lmu = info->lti_ea_store;
5599 lod_lum2lds(lds, lmu);
5606 * Get default striping in the object.
5608 * Get object default striping and default directory striping.
5610 * \param[in] env execution environment
5611 * \param[in] lo object
5612 * \param[out] lds default striping
5614 * \retval 0 on success
5615 * \retval negative if failed
5617 static int lod_get_default_striping(const struct lu_env *env,
5618 struct lod_object *lo,
5619 struct dt_allocation_hint *ah,
5620 struct lod_default_striping *lds)
5624 rc = lod_get_default_lov_striping(env, lo, lds, NULL);
5625 if (lds->lds_def_striping_set) {
5626 struct lod_thread_info *info = lod_env_info(env);
5627 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5629 rc = lod_verify_striping(env, d, lo, &info->lti_buf, false);
5631 lds->lds_def_striping_set = 0;
5634 if (ah->dah_eadata_is_dmv) {
5635 lod_lum2lds(lds, ah->dah_eadata);
5636 } else if (ah->dah_dmv_imp_inherit) {
5637 lds->lds_dir_def_striping_set = 0;
5639 rc1 = lod_get_default_lmv_striping(env, lo, lds);
5640 if (rc == 0 && rc1 < 0)
5648 * Apply default striping on object.
5650 * If object striping pattern is not set, set to the one in default striping.
5651 * The default striping is from parent or fs.
5653 * \param[in] lo new object
5654 * \param[in] lds default striping
5655 * \param[in] mode new object's mode
5657 static void lod_striping_from_default(struct lod_object *lo,
5658 const struct lod_default_striping *lds,
5661 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5664 if (lds->lds_def_striping_set && S_ISREG(mode)) {
5665 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
5667 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
5668 lds->lds_def_comp_cnt);
5672 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
5673 if (lds->lds_def_mirror_cnt > 1)
5674 lo->ldo_flr_state = LCM_FL_RDONLY;
5676 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5677 struct lod_layout_component *obj_comp =
5678 &lo->ldo_comp_entries[i];
5679 struct lod_layout_component *def_comp =
5680 &lds->lds_def_comp_entries[i];
5683 "inherit "DFID" file layout from default: flags=%#x size=%u nr=%u offset=%u pattern=%#x pool=%s\n",
5684 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
5685 def_comp->llc_flags,
5686 def_comp->llc_stripe_size,
5687 def_comp->llc_stripe_count,
5688 def_comp->llc_stripe_offset,
5689 def_comp->llc_pattern,
5690 def_comp->llc_pool ?: "");
5692 *obj_comp = *def_comp;
5693 if (def_comp->llc_pool != NULL) {
5694 /* pointer was copied from def_comp */
5695 obj_comp->llc_pool = NULL;
5696 lod_obj_set_pool(lo, i, def_comp->llc_pool);
5700 if (def_comp->llc_ostlist.op_array &&
5701 def_comp->llc_ostlist.op_count) {
5702 OBD_ALLOC(obj_comp->llc_ostlist.op_array,
5703 obj_comp->llc_ostlist.op_size);
5704 if (!obj_comp->llc_ostlist.op_array)
5706 memcpy(obj_comp->llc_ostlist.op_array,
5707 def_comp->llc_ostlist.op_array,
5708 obj_comp->llc_ostlist.op_size);
5709 } else if (def_comp->llc_ostlist.op_array) {
5710 obj_comp->llc_ostlist.op_array = NULL;
5714 * Don't initialize these fields for plain layout
5715 * (v1/v3) here, they are inherited in the order of
5716 * 'parent' -> 'fs default (root)' -> 'global default
5717 * values for stripe_count & stripe_size'.
5719 * see lod_ah_init().
5721 if (!lo->ldo_is_composite)
5724 lod_adjust_stripe_info(obj_comp, desc, 0);
5726 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
5727 if (lo->ldo_dir_stripe_count == 0)
5728 lo->ldo_dir_stripe_count =
5729 lds->lds_dir_def_stripe_count;
5730 if (lo->ldo_dir_stripe_offset == -1)
5731 lo->ldo_dir_stripe_offset =
5732 lds->lds_dir_def_stripe_offset;
5733 if (lo->ldo_dir_hash_type == LMV_HASH_TYPE_UNKNOWN)
5734 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
5737 "inherit "DFID" dir layout from default: count=%hu offset=%u hash_type=%x\n",
5738 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
5739 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
5740 lo->ldo_dir_hash_type);
5744 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root,
5745 const char *append_pool)
5747 struct lod_layout_component *lod_comp;
5749 if (lo->ldo_comp_cnt == 0)
5752 if (lo->ldo_is_composite)
5755 lod_comp = &lo->ldo_comp_entries[0];
5757 if (lod_comp->llc_stripe_count <= 0 ||
5758 lod_comp->llc_stripe_size <= 0)
5761 if (from_root && (lod_comp->llc_pool == NULL ||
5762 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
5765 if (append_pool && append_pool[0])
5772 * Implementation of dt_object_operations::do_ah_init.
5774 * This method is used to make a decision on the striping configuration for the
5775 * object being created. It can be taken from the \a parent object if it exists,
5776 * or filesystem's default. The resulting configuration (number of stripes,
5777 * stripe size/offset, pool name, hash_type, etc.) is stored in the object
5778 * itself and will be used by the methods like ->doo_declare_create().
5780 * \see dt_object_operations::do_ah_init() in the API description for details.
5782 static void lod_ah_init(const struct lu_env *env,
5783 struct dt_allocation_hint *ah,
5784 struct dt_object *parent,
5785 struct dt_object *child,
5788 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
5789 struct lod_thread_info *info = lod_env_info(env);
5790 struct lod_default_striping *lds = lod_lds_buf_get(env);
5791 struct dt_object *nextp = NULL;
5792 struct dt_object *nextc;
5793 struct lod_object *lp = NULL;
5794 struct lod_object *lc;
5795 struct lov_desc *desc;
5796 struct lod_layout_component *lod_comp;
5802 if (ah->dah_append_stripe_count == -1)
5803 ah->dah_append_stripe_count =
5804 d->lod_ost_descs.ltd_lov_desc.ld_tgt_count;
5806 if (likely(parent)) {
5807 nextp = dt_object_child(parent);
5808 lp = lod_dt_obj(parent);
5811 nextc = dt_object_child(child);
5812 lc = lod_dt_obj(child);
5814 LASSERT(!lod_obj_is_striped(child));
5815 /* default layout template may have been set on the regular file
5816 * when this is called from mdd_create_data() */
5817 if (S_ISREG(child_mode))
5818 lod_free_comp_entries(lc);
5820 if (!dt_object_exists(nextc))
5821 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
5823 if (S_ISDIR(child_mode)) {
5824 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
5825 int max_stripe_count;
5827 /* other default values are 0 */
5828 lc->ldo_dir_stripe_offset = LMV_OFFSET_DEFAULT;
5830 /* no default striping configuration is needed for
5833 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5834 le32_to_cpu(lum1->lum_magic) == LMV_MAGIC_FOREIGN) {
5835 lc->ldo_is_foreign = true;
5836 /* keep stripe_count 0 and stripe_offset -1 */
5837 CDEBUG(D_INFO, "no default striping for foreign dir\n");
5841 if (likely(lp != NULL))
5842 lod_get_default_striping(env, lp, ah, lds);
5844 /* It should always honour the specified stripes */
5845 if (ah->dah_eadata && ah->dah_eadata_len &&
5846 !ah->dah_eadata_is_dmv &&
5847 (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
5848 le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC ||
5849 le32_to_cpu(lum1->lum_magic) == LMV_MAGIC_V1)) {
5850 lc->ldo_dir_stripe_count =
5851 le32_to_cpu(lum1->lum_stripe_count);
5852 lc->ldo_dir_stripe_offset =
5853 le32_to_cpu(lum1->lum_stripe_offset);
5854 lc->ldo_dir_hash_type =
5855 le32_to_cpu(lum1->lum_hash_type);
5857 "set dirstripe: count %hu, offset %d, hash %x\n",
5858 lc->ldo_dir_stripe_count,
5859 (int)lc->ldo_dir_stripe_offset,
5860 lc->ldo_dir_hash_type);
5862 if (d->lod_mdt_descs.ltd_lmv_desc.ld_active_tgt_count &&
5863 lc->ldo_dir_stripe_count < 2 &&
5864 lum1->lum_max_inherit != LMV_INHERIT_NONE) {
5865 /* when filesystem-wide default LMV is set, dirs
5866 * will be created on MDT by space usage, but if
5867 * dir is created with "lfs mkdir -c 1 ...", its
5868 * subdirs should be kept on the same MDT. To
5869 * guarantee this, set default LMV for such dir.
5871 lds->lds_dir_def_stripe_count =
5872 le32_to_cpu(lum1->lum_stripe_count);
5873 /* if "-1" stripe offset is set, save current
5874 * MDT index in default LMV.
5876 if (le32_to_cpu(lum1->lum_stripe_offset) ==
5878 lds->lds_dir_def_stripe_offset =
5879 lod2lu_dev(d)->ld_site->ld_seq_site->ss_node_id;
5881 lds->lds_dir_def_stripe_offset =
5882 le32_to_cpu(lum1->lum_stripe_offset);
5883 lds->lds_dir_def_hash_type =
5884 le32_to_cpu(lum1->lum_hash_type);
5885 lds->lds_dir_def_max_inherit =
5886 lum1->lum_max_inherit;
5887 /* it will be decreased by 1 later in setting */
5888 if (lum1->lum_max_inherit >= LMV_INHERIT_END &&
5889 lum1->lum_max_inherit < LMV_INHERIT_MAX)
5890 lds->lds_dir_def_max_inherit++;
5891 lds->lds_dir_def_max_inherit_rr =
5892 lum1->lum_max_inherit_rr;
5893 lds->lds_dir_def_striping_set = 1;
5894 /* don't inherit LOV from ROOT */
5895 if (lds->lds_def_striping_set &&
5896 fid_is_root(lod_object_fid(lp)))
5897 lds->lds_def_striping_set = 0;
5898 lc->ldo_def_striping = lds;
5899 } else if (lds->lds_def_striping_set &&
5900 !fid_is_root(lod_object_fid(lp))) {
5901 /* don't inherit default LMV for "lfs mkdir" */
5902 lds->lds_dir_def_striping_set = 0;
5903 lc->ldo_def_striping = lds;
5906 /* inherit default striping except ROOT */
5907 if ((lds->lds_def_striping_set ||
5908 lds->lds_dir_def_striping_set) &&
5909 !fid_is_root(lod_object_fid(lp)))
5910 lc->ldo_def_striping = lds;
5912 /* transfer defaults LMV to new directory */
5913 lod_striping_from_default(lc, lds, child_mode);
5915 /* set count 0 to create normal directory */
5916 if (lc->ldo_dir_stripe_count == 1)
5917 lc->ldo_dir_stripe_count = 0;
5919 /* do not save default LMV on server */
5920 if (ah->dah_dmv_imp_inherit) {
5921 lds->lds_dir_def_striping_set = 0;
5922 if (!lds->lds_def_striping_set)
5923 lc->ldo_def_striping = NULL;
5927 /* shrink the stripe count to max_mdt_stripecount if it is -1
5928 * and max_mdt_stripecount is not 0
5930 if (lc->ldo_dir_stripe_count == (__u16)(-1) &&
5931 d->lod_max_mdt_stripecount)
5932 lc->ldo_dir_stripe_count = d->lod_max_mdt_stripecount;
5934 max_stripe_count = d->lod_remote_mdt_count + 1;
5935 if (lc->ldo_dir_hash_type & LMV_HASH_FLAG_OVERSTRIPED)
5937 max_stripe_count * LMV_MAX_STRIPES_PER_MDT;
5939 /* shrink the stripe_count to max stripe count */
5940 if (lc->ldo_dir_stripe_count > max_stripe_count &&
5941 !CFS_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)) {
5942 lc->ldo_dir_stripe_count = max_stripe_count;
5943 if (lc->ldo_dir_stripe_count == 1)
5944 lc->ldo_dir_stripe_count = 0;
5947 if (!lmv_is_known_hash_type(lc->ldo_dir_hash_type))
5948 lc->ldo_dir_hash_type =
5949 (lc->ldo_dir_hash_type & LMV_HASH_FLAG_KNOWN) |
5950 d->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
5952 /* make sure all fscrypt metadata stays on same mdt */
5953 if (child->do_lu.lo_header->loh_attr & LOHA_FSCRYPT_MD) {
5954 lc->ldo_dir_stripe_count = 0;
5955 lds->lds_dir_def_stripe_offset =
5956 lod2lu_dev(d)->ld_site->ld_seq_site->ss_node_id;
5957 lds->lds_dir_def_striping_set = 1;
5958 lc->ldo_def_striping = lds;
5961 CDEBUG(D_INFO, "final dir stripe_count=%hu offset=%d hash=%x\n",
5962 lc->ldo_dir_stripe_count,
5963 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
5968 /* child object regular file*/
5970 if (!lod_object_will_be_striped(S_ISREG(child_mode),
5971 lu_object_fid(&child->do_lu)))
5974 /* If object is going to be striped over OSTs, transfer default
5975 * striping information to the child, so that we can use it
5976 * during declaration and creation.
5978 * Try from the parent first.
5980 if (likely(lp != NULL)) {
5981 rc = lod_get_default_lov_striping(env, lp, lds, ah);
5982 if (rc == 0 && lds->lds_def_striping_set) {
5983 rc = lod_verify_striping(env, d, lp, &info->lti_buf,
5986 lod_striping_from_default(lc, lds, child_mode);
5990 /* Initialize lod_device::lod_md_root object reference */
5991 if (d->lod_md_root == NULL) {
5992 struct dt_object *root;
5993 struct lod_object *lroot;
5995 lu_root_fid(&info->lti_fid);
5996 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
5997 if (!IS_ERR(root)) {
5998 lroot = lod_dt_obj(root);
6000 spin_lock(&d->lod_lock);
6001 if (d->lod_md_root != NULL)
6002 dt_object_put(env, &d->lod_md_root->ldo_obj);
6003 d->lod_md_root = lroot;
6004 spin_unlock(&d->lod_lock);
6008 /* try inherit layout from the root object (fs default) when:
6009 * - parent does not have default layout; or
6010 * - parent has plain(v1/v3) default layout, and some attributes
6011 * are not specified in the default layout;
6013 if (d->lod_md_root != NULL &&
6014 lod_need_inherit_more(lc, true, ah->dah_append_pool)) {
6015 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds,
6017 if (rc || !lds->lds_def_striping_set)
6020 rc = lod_verify_striping(env, d, d->lod_md_root, &info->lti_buf,
6025 if (lc->ldo_comp_cnt == 0) {
6026 lod_striping_from_default(lc, lds, child_mode);
6027 } else if (!lds->lds_def_striping_is_composite) {
6028 struct lod_layout_component *def_comp;
6030 LASSERT(!lc->ldo_is_composite);
6031 lod_comp = &lc->ldo_comp_entries[0];
6032 def_comp = &lds->lds_def_comp_entries[0];
6034 if (lod_comp->llc_stripe_count <= 0)
6035 lod_comp->llc_stripe_count =
6036 def_comp->llc_stripe_count;
6037 if (lod_comp->llc_stripe_size <= 0)
6038 lod_comp->llc_stripe_size =
6039 def_comp->llc_stripe_size;
6040 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT &&
6041 (!lod_comp->llc_pool || !lod_comp->llc_pool[0]))
6042 lod_comp->llc_stripe_offset =
6043 def_comp->llc_stripe_offset;
6044 if (lod_comp->llc_pool == NULL)
6045 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
6050 * fs default striping may not be explicitly set, or historically set
6051 * in config log, use them.
6053 if (lod_need_inherit_more(lc, false, ah->dah_append_pool)) {
6054 if (lc->ldo_comp_cnt == 0) {
6055 rc = lod_alloc_comp_entries(lc, 0, 1);
6057 /* fail to allocate memory, will create a
6058 * non-striped file. */
6060 lc->ldo_is_composite = 0;
6061 lod_comp = &lc->ldo_comp_entries[0];
6062 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
6064 LASSERT(!lc->ldo_is_composite);
6065 lod_comp = &lc->ldo_comp_entries[0];
6066 desc = &d->lod_ost_descs.ltd_lov_desc;
6067 lod_adjust_stripe_info(lod_comp, desc,
6068 ah->dah_append_stripe_count);
6069 if (ah->dah_append_pool && ah->dah_append_pool[0])
6070 lod_obj_set_pool(lc, 0, ah->dah_append_pool);
6077 * Size initialization on late striping.
6079 * Propagate the size of a truncated object to a deferred striping.
6080 * This function handles a special case when truncate was done on a
6081 * non-striped object and now while the striping is being created
6082 * we can't lose that size, so we have to propagate it to the stripes
6085 * \param[in] env execution environment
6086 * \param[in] dt object
6087 * \param[in] th transaction handle
6089 * \retval 0 on success
6090 * \retval negative if failed
6092 static int lod_declare_init_size(const struct lu_env *env,
6093 struct dt_object *dt, struct thandle *th)
6095 struct dt_object *next = dt_object_child(dt);
6096 struct lod_object *lo = lod_dt_obj(dt);
6097 struct dt_object **objects = NULL;
6098 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
6099 uint64_t size, offs;
6100 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
6101 struct lu_extent size_ext;
6104 if (!lod_obj_is_striped(dt))
6107 rc = dt_attr_get(env, next, attr);
6108 LASSERT(attr->la_valid & LA_SIZE);
6112 size = attr->la_size;
6116 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
6117 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6118 struct lod_layout_component *lod_comp;
6119 struct lu_extent *extent;
6121 lod_comp = &lo->ldo_comp_entries[i];
6123 if (lod_comp->llc_stripe == NULL)
6126 extent = &lod_comp->llc_extent;
6127 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
6128 if (!lo->ldo_is_composite ||
6129 lu_extent_is_overlapped(extent, &size_ext)) {
6130 objects = lod_comp->llc_stripe;
6131 stripe_count = lod_comp->llc_stripe_count;
6132 stripe_size = lod_comp->llc_stripe_size;
6135 if (stripe_count == 0)
6138 LASSERT(objects != NULL && stripe_size != 0);
6139 do_div(size, stripe_size);
6140 stripe = do_div(size, stripe_count);
6141 LASSERT(objects[stripe] != NULL);
6143 size = size * stripe_size;
6144 offs = attr->la_size;
6145 size += do_div(offs, stripe_size);
6147 attr->la_valid = LA_SIZE;
6148 attr->la_size = size;
6150 rc = lod_sub_declare_attr_set(env, objects[stripe],
6159 * Declare creation of striped object.
6161 * The function declares creation stripes for a regular object. The function
6162 * also declares whether the stripes will be created with non-zero size if
6163 * previously size was set non-zero on the master object. If object \a dt is
6164 * not local, then only fully defined striping can be applied in \a lovea.
6165 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
6168 * \param[in] env execution environment
6169 * \param[in] dt object
6170 * \param[in] attr attributes the stripes will be created with
6171 * \param[in] lovea a buffer containing striping description
6172 * \param[in] th transaction handle
6174 * \retval 0 on success
6175 * \retval negative if failed
6177 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
6178 struct lu_attr *attr,
6179 const struct lu_buf *lovea, struct thandle *th)
6181 struct lod_thread_info *info = lod_env_info(env);
6182 struct dt_object *next = dt_object_child(dt);
6183 struct lod_object *lo = lod_dt_obj(dt);
6187 if (CFS_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
6188 GOTO(out, rc = -ENOMEM);
6190 if (!dt_object_remote(next)) {
6191 /* choose OST and generate appropriate objects */
6192 rc = lod_prepare_create(env, lo, attr, lovea, th);
6197 * declare storage for striping data
6199 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6201 /* LOD can not choose OST objects for remote objects, i.e.
6202 * stripes must be ready before that. Right now, it can only
6203 * happen during migrate, i.e. migrate process needs to create
6204 * remote regular file (mdd_migrate_create), then the migrate
6205 * process will provide stripeEA. */
6206 LASSERT(lovea != NULL);
6207 info->lti_buf = *lovea;
6210 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
6211 XATTR_NAME_LOV, 0, th);
6216 * if striping is created with local object's size > 0,
6217 * we have to propagate this size to specific object
6218 * the case is possible only when local object was created previously
6220 if (dt_object_exists(next))
6221 rc = lod_declare_init_size(env, dt, th);
6224 /* failed to create striping or to set initial size, let's reset
6225 * config so that others don't get confused */
6227 lod_striping_free(env, lo);
6233 * Whether subdirectories under \a dt should be created on MDTs by space QoS
6235 * If LMV_HASH_FLAG_SPACE is set on directory default layout, its subdirectories
6236 * should be created on MDT by space QoS.
6238 * \param[in] env execution environment
6239 * \param[in] dev lu device
6240 * \param[in] dt object
6242 * \retval 1 if directory should create subdir by space usage
6244 * \retval -ev if failed
6246 static inline int dt_object_qos_mkdir(const struct lu_env *env,
6247 struct lu_device *dev,
6248 struct dt_object *dt)
6250 struct lod_thread_info *info = lod_env_info(env);
6251 struct lu_object *obj;
6252 struct lod_object *lo;
6253 struct lmv_user_md *lmu;
6256 obj = lu_object_find_slice(env, dev, lu_object_fid(&dt->do_lu), NULL);
6258 return PTR_ERR(obj);
6260 lo = lu2lod_obj(obj);
6262 rc = lod_get_default_lmv_ea(env, lo);
6263 dt_object_put(env, dt);
6267 if (rc < (int)sizeof(*lmu))
6270 lmu = info->lti_ea_store;
6271 return le32_to_cpu(lmu->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
6275 * Implementation of dt_object_operations::do_declare_create.
6277 * The method declares creation of a new object. If the object will be striped,
6278 * then helper functions are called to find FIDs for the stripes, declare
6279 * creation of the stripes and declare initialization of the striping
6280 * information to be stored in the master object.
6282 * \see dt_object_operations::do_declare_create() in the API description
6285 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
6286 struct lu_attr *attr,
6287 struct dt_allocation_hint *hint,
6288 struct dt_object_format *dof, struct thandle *th)
6290 struct dt_object *next = dt_object_child(dt);
6291 struct lod_object *lo = lod_dt_obj(dt);
6300 * first of all, we declare creation of local object
6302 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
6307 * it's lod_ah_init() that has decided the object will be striped
6309 if (dof->dof_type == DFT_REGULAR) {
6310 /* callers don't want stripes */
6311 /* XXX: all tricky interactions with ->ah_make_hint() decided
6312 * to use striping, then ->declare_create() behaving differently
6313 * should be cleaned */
6314 if (dof->u.dof_reg.striped != 0)
6315 rc = lod_declare_striped_create(env, dt, attr,
6317 } else if (dof->dof_type == DFT_DIR) {
6318 struct seq_server_site *ss;
6319 struct lu_buf buf = { NULL };
6321 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
6323 /* If the parent has default stripeEA, and client
6324 * did not find it before sending create request,
6325 * then MDT will return -EREMOTE, and client will
6326 * retrieve the default stripeEA and re-create the
6329 * Note: if dah_eadata != NULL, it means creating the
6330 * striped directory with specified stripeEA, then it
6331 * should ignore the default stripeEA */
6332 if (hint != NULL && hint->dah_eadata == NULL) {
6333 if (CFS_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
6334 GOTO(out, rc = -EREMOTE);
6336 if (lo->ldo_dir_stripe_offset != LMV_OFFSET_DEFAULT &&
6337 lo->ldo_dir_stripe_offset != ss->ss_node_id) {
6338 struct lod_device *lod;
6339 struct lu_tgt_desc *mdt = NULL;
6340 bool found_mdt = false;
6342 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6343 lod_foreach_mdt(lod, mdt) {
6344 if (mdt->ltd_index ==
6345 lo->ldo_dir_stripe_offset) {
6351 /* If the MDT indicated by stripe_offset can be
6352 * found, then tell client to resend the create
6353 * request to the correct MDT, otherwise return
6354 * error to client */
6356 GOTO(out, rc = -EREMOTE);
6358 GOTO(out, rc = -EINVAL);
6360 } else if (hint && hint->dah_eadata) {
6361 buf.lb_buf = (void *)hint->dah_eadata;
6362 buf.lb_len = hint->dah_eadata_len;
6365 rc = lod_declare_dir_striping_create(env, dt, attr, &buf, dof,
6369 /* failed to create striping or to set initial size, let's reset
6370 * config so that others don't get confused */
6372 lod_striping_free(env, lo);
6377 * Generate component ID for new created component.
6379 * \param[in] lo LOD object
6380 * \param[in] comp_idx index of ldo_comp_entries
6382 * \retval component ID on success
6383 * \retval LCME_ID_INVAL on failure
6385 static __u32 lod_gen_component_id(struct lod_object *lo,
6386 int mirror_id, int comp_idx)
6388 struct lod_layout_component *lod_comp;
6389 __u32 id, start, end;
6392 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
6394 lod_obj_inc_layout_gen(lo);
6395 id = lo->ldo_layout_gen;
6396 if (likely(id <= SEQ_ID_MAX))
6397 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
6399 /* Layout generation wraps, need to check collisions. */
6400 start = id & SEQ_ID_MASK;
6403 for (id = start; id <= end; id++) {
6404 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6405 lod_comp = &lo->ldo_comp_entries[i];
6406 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
6409 /* Found the ununsed ID */
6410 if (i == lo->ldo_comp_cnt)
6411 RETURN(pflr_id(mirror_id, id));
6414 if (end == SEQ_ID_MAX) {
6415 end = min_t(__u32, start, SEQ_ID_MAX) - 1;
6420 RETURN(LCME_ID_INVAL);
6424 * Creation of a striped regular object.
6426 * The function is called to create the stripe objects for a regular
6427 * striped file. This can happen at the initial object creation or
6428 * when the caller asks LOD to do so using ->do_xattr_set() method
6429 * (so called late striping). Notice all the information are already
6430 * prepared in the form of the list of objects (ldo_stripe field).
6431 * This is done during declare phase.
6433 * \param[in] env execution environment
6434 * \param[in] dt object
6435 * \param[in] attr attributes the stripes will be created with
6436 * \param[in] dof format of stripes (see OSD API description)
6437 * \param[in] th transaction handle
6439 * \retval 0 on success
6440 * \retval negative if failed
6442 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
6443 struct lu_attr *attr, struct dt_object_format *dof,
6446 struct lod_layout_component *lod_comp;
6447 struct lod_object *lo = lod_dt_obj(dt);
6452 mutex_lock(&lo->ldo_layout_mutex);
6454 LASSERT((lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL) ||
6455 lo->ldo_is_foreign);
6457 mirror_id = 0; /* non-flr file's mirror_id is 0 */
6458 if (lo->ldo_mirror_count > 1) {
6459 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6460 lod_comp = &lo->ldo_comp_entries[i];
6461 if (lod_comp->llc_id != LCME_ID_INVAL &&
6462 mirror_id_of(lod_comp->llc_id) > mirror_id)
6463 mirror_id = mirror_id_of(lod_comp->llc_id);
6467 /* create all underlying objects */
6468 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6469 lod_comp = &lo->ldo_comp_entries[i];
6471 if (lod_comp->llc_id == LCME_ID_INVAL) {
6472 /* only the component of FLR layout with more than 1
6473 * mirror has mirror ID in its component ID.
6475 if (lod_comp->llc_extent.e_start == 0 &&
6476 lo->ldo_mirror_count > 1)
6479 lod_comp->llc_id = lod_gen_component_id(lo,
6481 if (lod_comp->llc_id == LCME_ID_INVAL)
6482 GOTO(out, rc = -ERANGE);
6485 if (lod_comp_inited(lod_comp))
6488 if (lod_comp->llc_magic == LOV_MAGIC_FOREIGN) {
6489 lod_comp_set_init(lod_comp);
6493 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
6494 lod_comp_set_init(lod_comp);
6496 if (lov_pattern(lod_comp->llc_pattern) & LOV_PATTERN_MDT)
6497 lod_comp_set_init(lod_comp);
6499 if (lod_comp->llc_stripe == NULL)
6502 LASSERT(lod_comp->llc_stripe_count);
6503 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6504 struct dt_object *object = lod_comp->llc_stripe[j];
6505 LASSERT(object != NULL);
6506 rc = lod_sub_create(env, object, attr, NULL, dof, th);
6510 lod_comp_set_init(lod_comp);
6513 rc = lod_fill_mirrors(lo);
6517 lo->ldo_comp_cached = 1;
6519 rc = lod_generate_and_set_lovea(env, lo, th);
6523 mutex_unlock(&lo->ldo_layout_mutex);
6528 lod_striping_free_nolock(env, lo);
6529 mutex_unlock(&lo->ldo_layout_mutex);
6534 static inline bool lod_obj_is_dom(struct dt_object *dt)
6536 struct lod_object *lo = lod_dt_obj(dt);
6538 if (!dt_object_exists(dt_object_child(dt)))
6541 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
6544 if (!lo->ldo_comp_cnt)
6547 return (lov_pattern(lo->ldo_comp_entries[0].llc_pattern) &
6552 * Implementation of dt_object_operations::do_create.
6554 * If any of preceeding methods (like ->do_declare_create(),
6555 * ->do_ah_init(), etc) chose to create a striped object,
6556 * then this method will create the master and the stripes.
6558 * \see dt_object_operations::do_create() in the API description for details.
6560 static int lod_create(const struct lu_env *env, struct dt_object *dt,
6561 struct lu_attr *attr, struct dt_allocation_hint *hint,
6562 struct dt_object_format *dof, struct thandle *th)
6567 /* create local object */
6568 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
6572 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
6573 (lod_obj_is_striped(dt) || lod_obj_is_dom(dt)) &&
6574 dof->u.dof_reg.striped != 0) {
6575 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
6576 rc = lod_striped_create(env, dt, attr, dof, th);
6583 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
6584 struct dt_object *dt, struct thandle *th,
6585 int comp_idx, int stripe_idx,
6586 struct lod_obj_stripe_cb_data *data)
6588 if (data->locd_declare)
6589 return lod_sub_declare_destroy(env, dt, th);
6591 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
6592 stripe_idx == cfs_fail_val)
6593 return lod_sub_destroy(env, dt, th);
6599 * Implementation of dt_object_operations::do_declare_destroy.
6601 * If the object is a striped directory, then the function declares reference
6602 * removal from the master object (this is an index) to the stripes and declares
6603 * destroy of all the stripes. In all the cases, it declares an intention to
6604 * destroy the object itself.
6606 * \see dt_object_operations::do_declare_destroy() in the API description
6609 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
6612 struct dt_object *next = dt_object_child(dt);
6613 struct lod_object *lo = lod_dt_obj(dt);
6614 struct lod_thread_info *info = lod_env_info(env);
6615 struct dt_object *stripe;
6616 char *stripe_name = info->lti_key;
6622 * load striping information, notice we don't do this when object
6623 * is being initialized as we don't need this information till
6624 * few specific cases like destroy, chown
6626 rc = lod_striping_load(env, lo);
6630 /* declare destroy for all underlying objects */
6631 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6632 rc = next->do_ops->do_index_try(env, next,
6633 &dt_directory_features);
6637 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6638 stripe = lo->ldo_stripe[i];
6642 rc = lod_sub_declare_ref_del(env, next, th);
6646 snprintf(stripe_name, sizeof(info->lti_key),
6648 PFID(lu_object_fid(&stripe->do_lu)), i);
6649 rc = lod_sub_declare_delete(env, next,
6650 (const struct dt_key *)stripe_name, th);
6657 * we declare destroy for the local object
6659 rc = lod_sub_declare_destroy(env, next, th);
6663 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
6664 CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
6667 if (!lod_obj_is_striped(dt))
6670 /* declare destroy all striped objects */
6671 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6672 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6673 stripe = lo->ldo_stripe[i];
6677 if (!dt_object_exists(stripe))
6680 rc = lod_sub_declare_ref_del(env, stripe, th);
6684 rc = lod_sub_declare_destroy(env, stripe, th);
6689 struct lod_obj_stripe_cb_data data = { { 0 } };
6691 data.locd_declare = true;
6692 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
6693 rc = lod_obj_for_each_stripe(env, lo, th, &data);
6700 * Implementation of dt_object_operations::do_destroy.
6702 * If the object is a striped directory, then the function removes references
6703 * from the master object (this is an index) to the stripes and destroys all
6704 * the stripes. In all the cases, the function destroys the object itself.
6706 * \see dt_object_operations::do_destroy() in the API description for details.
6708 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
6711 struct dt_object *next = dt_object_child(dt);
6712 struct lod_object *lo = lod_dt_obj(dt);
6713 struct lod_thread_info *info = lod_env_info(env);
6714 char *stripe_name = info->lti_key;
6715 struct dt_object *stripe;
6721 /* destroy sub-stripe of master object */
6722 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6723 rc = next->do_ops->do_index_try(env, next,
6724 &dt_directory_features);
6728 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6729 stripe = lo->ldo_stripe[i];
6733 rc = lod_sub_ref_del(env, next, th);
6737 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
6738 PFID(lu_object_fid(&stripe->do_lu)), i);
6740 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
6741 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
6742 PFID(lu_object_fid(&stripe->do_lu)));
6744 rc = lod_sub_delete(env, next,
6745 (const struct dt_key *)stripe_name, th);
6751 rc = lod_sub_destroy(env, next, th);
6755 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
6756 CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
6759 if (!lod_obj_is_striped(dt))
6762 /* destroy all striped objects */
6763 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6764 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6765 stripe = lo->ldo_stripe[i];
6769 if (!dt_object_exists(stripe))
6772 if (!CFS_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
6773 i == cfs_fail_val) {
6774 dt_write_lock(env, stripe, DT_TGT_CHILD);
6775 rc = lod_sub_ref_del(env, stripe, th);
6776 dt_write_unlock(env, stripe);
6780 rc = lod_sub_destroy(env, stripe, th);
6786 struct lod_obj_stripe_cb_data data = { { 0 } };
6788 data.locd_declare = false;
6789 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
6790 rc = lod_obj_for_each_stripe(env, lo, th, &data);
6797 * Implementation of dt_object_operations::do_declare_ref_add.
6799 * \see dt_object_operations::do_declare_ref_add() in the API description
6802 static int lod_declare_ref_add(const struct lu_env *env,
6803 struct dt_object *dt, struct thandle *th)
6805 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
6809 * Implementation of dt_object_operations::do_ref_add.
6811 * \see dt_object_operations::do_ref_add() in the API description for details.
6813 static int lod_ref_add(const struct lu_env *env,
6814 struct dt_object *dt, struct thandle *th)
6816 return lod_sub_ref_add(env, dt_object_child(dt), th);
6820 * Implementation of dt_object_operations::do_declare_ref_del.
6822 * \see dt_object_operations::do_declare_ref_del() in the API description
6825 static int lod_declare_ref_del(const struct lu_env *env,
6826 struct dt_object *dt, struct thandle *th)
6828 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
6832 * Implementation of dt_object_operations::do_ref_del
6834 * \see dt_object_operations::do_ref_del() in the API description for details.
6836 static int lod_ref_del(const struct lu_env *env,
6837 struct dt_object *dt, struct thandle *th)
6839 return lod_sub_ref_del(env, dt_object_child(dt), th);
6843 * Implementation of dt_object_operations::do_object_sync.
6845 * \see dt_object_operations::do_object_sync() in the API description
6848 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
6849 __u64 start, __u64 end)
6851 return dt_object_sync(env, dt_object_child(dt), start, end);
6855 * Implementation of dt_object_operations::do_object_unlock.
6857 * Used to release LDLM lock(s).
6859 * \see dt_object_operations::do_object_unlock() in the API description
6862 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
6863 struct ldlm_enqueue_info *einfo,
6864 union ldlm_policy_data *policy)
6866 struct lod_object *lo = lod_dt_obj(dt);
6867 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
6868 int slave_locks_size;
6872 if (slave_locks == NULL)
6875 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
6876 /* Note: for remote lock for single stripe dir, MDT will cancel
6877 * the lock by lockh directly */
6878 LASSERT(!dt_object_remote(dt_object_child(dt)));
6880 /* locks were unlocked in MDT layer */
6881 for (i = 0; i < slave_locks->ha_count; i++)
6882 LASSERT(!lustre_handle_is_used(&slave_locks->ha_handles[i]));
6885 * NB, ha_count may not equal to ldo_dir_stripe_count, because dir
6886 * layout may change, e.g., shrink dir layout after migration.
6888 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6889 if (lo->ldo_stripe[i])
6890 dt_invalidate(env, lo->ldo_stripe[i]);
6893 slave_locks_size = offsetof(typeof(*slave_locks),
6894 ha_handles[slave_locks->ha_count]);
6895 OBD_FREE(slave_locks, slave_locks_size);
6896 einfo->ei_cbdata = NULL;
6902 * Implementation of dt_object_operations::do_object_lock.
6904 * Used to get LDLM lock on the non-striped and striped objects.
6906 * \see dt_object_operations::do_object_lock() in the API description
6909 static int lod_object_lock(const struct lu_env *env,
6910 struct dt_object *dt,
6911 struct lustre_handle *lh,
6912 struct ldlm_enqueue_info *einfo,
6913 union ldlm_policy_data *policy)
6915 struct lod_object *lo = lod_dt_obj(dt);
6916 int slave_locks_size;
6917 struct lustre_handle_array *slave_locks = NULL;
6922 /* remote object lock */
6923 if (!einfo->ei_enq_slave) {
6924 LASSERT(dt_object_remote(dt));
6925 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
6929 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
6932 rc = lod_striping_load(env, lo);
6937 if (lo->ldo_dir_stripe_count <= 1)
6940 slave_locks_size = offsetof(typeof(*slave_locks),
6941 ha_handles[lo->ldo_dir_stripe_count]);
6942 /* Freed in lod_object_unlock */
6943 OBD_ALLOC(slave_locks, slave_locks_size);
6946 slave_locks->ha_count = lo->ldo_dir_stripe_count;
6948 /* striped directory lock */
6949 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6950 struct lustre_handle lockh;
6951 struct ldlm_res_id *res_id;
6952 struct dt_object *stripe;
6954 stripe = lo->ldo_stripe[i];
6958 res_id = &lod_env_info(env)->lti_res_id;
6959 fid_build_reg_res_name(lu_object_fid(&stripe->do_lu), res_id);
6960 einfo->ei_res_id = res_id;
6962 if (dt_object_remote(stripe)) {
6963 set_bit(i, (void *)slave_locks->ha_map);
6964 rc = dt_object_lock(env, stripe, &lockh, einfo, policy);
6966 struct ldlm_namespace *ns = einfo->ei_namespace;
6967 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
6968 ldlm_completion_callback completion = einfo->ei_cb_cp;
6969 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
6971 LASSERT(ns != NULL);
6972 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS,
6973 policy, einfo->ei_mode,
6974 &dlmflags, blocking,
6976 NULL, 0, LVB_T_NONE,
6981 ldlm_lock_decref_and_cancel(
6982 &slave_locks->ha_handles[i],
6984 OBD_FREE(slave_locks, slave_locks_size);
6987 slave_locks->ha_handles[i] = lockh;
6989 einfo->ei_cbdata = slave_locks;
6995 * Implementation of dt_object_operations::do_invalidate.
6997 * \see dt_object_operations::do_invalidate() in the API description for details
6999 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
7001 return dt_invalidate(env, dt_object_child(dt));
7004 static int lod_declare_instantiate_components(const struct lu_env *env,
7005 struct lod_object *lo,
7009 struct lod_thread_info *info = lod_env_info(env);
7014 LASSERT(info->lti_count < lo->ldo_comp_cnt);
7016 for (i = 0; i < info->lti_count; i++) {
7017 rc = lod_qos_prep_create(env, lo, NULL, th,
7018 info->lti_comp_idx[i], reserve);
7024 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
7025 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
7026 &info->lti_buf, XATTR_NAME_LOV, 0, th);
7033 * Check OSTs for an existing component for further extension
7035 * Checks if OSTs are still healthy and not out of space. Gets free space
7036 * on OSTs (relative to allocation watermark rmb_low) and compares to
7037 * the proposed new_end for this component.
7039 * Decides whether or not to extend a component on its current OSTs.
7041 * \param[in] env execution environment for this thread
7042 * \param[in] lo object we're checking
7043 * \param[in] index index of this component
7044 * \param[in] extension_size extension size for this component
7045 * \param[in] extent layout extent for requested operation
7046 * \param[in] comp_extent extension component extent
7047 * \param[in] write if this is write operation
7049 * \retval true - OK to extend on current OSTs
7050 * \retval false - do not extend on current OSTs
7052 static bool lod_sel_osts_allowed(const struct lu_env *env,
7053 struct lod_object *lo,
7054 int index, __u64 reserve,
7055 struct lu_extent *extent,
7056 struct lu_extent *comp_extent, int write)
7058 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[index];
7059 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7060 struct lod_thread_info *tinfo = lod_env_info(env);
7061 struct obd_statfs *sfs = &tinfo->lti_osfs;
7062 __u64 available = 0;
7068 LASSERT(lod_comp->llc_stripe_count != 0);
7070 lod_getref(&lod->lod_ost_descs);
7071 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
7072 int index = lod_comp->llc_ost_indices[i];
7073 struct lod_tgt_desc *ost = OST_TGT(lod, index);
7074 struct obd_statfs_info info = { 0 };
7075 int j, repeated = 0;
7079 /* Get the number of times this OST repeats in this component.
7080 * Note: inter-component repeats are not counted as this is
7081 * considered as a rare case: we try to not repeat OST in other
7082 * components if possible. */
7083 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
7084 if (index != lod_comp->llc_ost_indices[j])
7087 /* already handled */
7093 if (j < lod_comp->llc_stripe_count)
7096 if (!test_bit(index, lod->lod_ost_bitmap)) {
7097 CDEBUG(D_LAYOUT, "ost %d no longer present\n", index);
7102 rc = dt_statfs_info(env, ost->ltd_tgt, sfs, &info);
7104 CDEBUG(D_LAYOUT, "statfs failed for ost %d, error %d\n",
7110 if (sfs->os_state & OS_STATFS_ENOSPC ||
7111 sfs->os_state & OS_STATFS_READONLY ||
7112 sfs->os_state & OS_STATFS_NOCREATE ||
7113 sfs->os_state & OS_STATFS_DEGRADED) {
7115 "OST%04x unusable for SEL extension, state %x\n",
7116 index, sfs->os_state);
7122 available = sfs->os_bavail * sfs->os_bsize;
7123 /* 'available' is relative to the allocation threshold */
7124 available -= (__u64) info.os_reserved_mb_low << 20;
7126 CDEBUG(D_LAYOUT, "ost %d lowwm: %d highwm: %d, "
7127 "%llu %% blocks available, %llu %% blocks free\n",
7128 index, info.os_reserved_mb_low, info.os_reserved_mb_high,
7129 (100ull * sfs->os_bavail) / sfs->os_blocks,
7130 (100ull * sfs->os_bfree) / sfs->os_blocks);
7132 if (reserve * repeated > available) {
7134 CDEBUG(D_LAYOUT, "low space on ost %d, available %llu "
7135 "< extension size %llu repeated %d\n", index,
7136 available, reserve, repeated);
7140 lod_putref(lod, &lod->lod_ost_descs);
7146 * Adjust extents after component removal
7148 * When we remove an extension component, we move the start of the next
7149 * component to match the start of the extension component, so no space is left
7152 * \param[in] env execution environment for this thread
7153 * \param[in] lo object
7154 * \param[in] max_comp layout component
7155 * \param[in] index index of this component
7157 * \retval 0 on success
7158 * \retval negative errno on error
7160 static void lod_sel_adjust_extents(const struct lu_env *env,
7161 struct lod_object *lo,
7162 int max_comp, int index)
7164 struct lod_layout_component *lod_comp = NULL;
7165 struct lod_layout_component *next = NULL;
7166 struct lod_layout_component *prev = NULL;
7167 __u64 new_start = 0;
7171 /* Extension space component */
7172 lod_comp = &lo->ldo_comp_entries[index];
7173 next = &lo->ldo_comp_entries[index + 1];
7174 prev = &lo->ldo_comp_entries[index - 1];
7176 LASSERT(lod_comp != NULL && prev != NULL && next != NULL);
7177 LASSERT(lod_comp->llc_flags & LCME_FL_EXTENSION);
7179 /* Previous is being removed */
7180 if (prev && prev->llc_id == LCME_ID_INVAL)
7181 new_start = prev->llc_extent.e_start;
7183 new_start = lod_comp->llc_extent.e_start;
7185 for (i = index + 1; i < max_comp; i++) {
7186 lod_comp = &lo->ldo_comp_entries[i];
7188 start = lod_comp->llc_extent.e_start;
7189 lod_comp->llc_extent.e_start = new_start;
7191 /* We only move zero length extendable components */
7192 if (!(start == lod_comp->llc_extent.e_end))
7195 LASSERT(!(lod_comp->llc_flags & LCME_FL_INIT));
7197 lod_comp->llc_extent.e_end = new_start;
7201 /* Calculate the proposed 'new end' for a component we're extending */
7202 static __u64 lod_extension_new_end(__u64 extension_size, __u64 extent_end,
7203 __u32 stripe_size, __u64 component_end,
7204 __u64 extension_end)
7208 LASSERT(extension_size != 0 && stripe_size != 0);
7210 /* Round up to extension size */
7211 if (extent_end == OBD_OBJECT_EOF) {
7212 new_end = OBD_OBJECT_EOF;
7214 /* Add at least extension_size to the previous component_end,
7215 * covering the req layout extent */
7216 new_end = max(extent_end - component_end, extension_size);
7217 new_end = roundup(new_end, extension_size);
7218 new_end += component_end;
7220 /* Component end must be min stripe size aligned */
7221 if (new_end % stripe_size) {
7222 CDEBUG(D_LAYOUT, "new component end is not aligned "
7223 "by the stripe size %u: [%llu, %llu) ext size "
7224 "%llu new end %llu, aligning\n",
7225 stripe_size, component_end, extent_end,
7226 extension_size, new_end);
7227 new_end = roundup(new_end, stripe_size);
7231 if (new_end < extent_end)
7232 new_end = OBD_OBJECT_EOF;
7235 /* Don't extend past the end of the extension component */
7236 if (new_end > extension_end)
7237 new_end = extension_end;
7243 * Calculate the exact reservation (per-OST extension_size) on the OSTs being
7244 * instantiated. It needs to be calculated in advance and taken into account at
7245 * the instantiation time, because otherwise lod_statfs_and_check() may consider
7246 * an OST as OK, but SEL needs its extension_size to fit the free space and the
7247 * OST may turn out to be low-on-space, thus inappropriate OST may be used and
7250 * \param[in] lod_comp lod component we are checking
7252 * \retval size to reserved on each OST of lod_comp's stripe.
7254 static __u64 lod_sel_stripe_reserved(struct lod_layout_component *lod_comp)
7256 /* extension_size is file level, so we must divide by stripe count to
7257 * compare it to available space on a single OST */
7258 return lod_comp->llc_stripe_size * SEL_UNIT_SIZE /
7259 lod_comp->llc_stripe_count;
7262 /* As lod_sel_handler() could be re-entered for the same component several
7263 * times, this is the data for the next call. Fields could be changed to
7264 * component indexes when needed, (e.g. if there is no need to instantiate
7265 * all the previous components up to the current position) to tell the caller
7266 * where to start over from. */
7273 * Process extent updates for a particular layout component
7275 * Handle layout updates for a particular extension space component touched by
7276 * a layout update operation. Core function of self-extending PFL feature.
7278 * In general, this function processes exactly *one* stage of an extension
7279 * operation, modifying the layout accordingly, then returns to the caller.
7280 * The caller is responsible for restarting processing with the new layout,
7281 * which may repeatedly return to this function until the extension updates
7284 * This function does one of a few things to the layout:
7285 * 1. Extends the component before the current extension space component to
7286 * allow it to accomodate the requested operation (if space/policy permit that
7287 * component to continue on its current OSTs)
7289 * 2. If extension of the existing component fails, we do one of two things:
7290 * a. If there is a component after the extension space, we remove the
7291 * extension space component, move the start of the next component down
7292 * accordingly, then notify the caller to restart processing w/the new
7294 * b. If there is no following component, we try repeating the current
7295 * component, creating a new component using the current one as a
7296 * template (keeping its stripe properties but not specific striping),
7297 * and try assigning striping for this component. If there is sufficient
7298 * free space on the OSTs chosen for this component, it is instantiated
7299 * and i/o continues there.
7301 * If there is not sufficient space on the new OSTs, we remove this new
7302 * component & extend the current component.
7304 * Note further that uninited components followed by extension space can be zero
7305 * length meaning that we will try to extend them before initializing them, and
7306 * if that fails, they will be removed without initialization.
7308 * 3. If we extend to/beyond the end of an extension space component, that
7309 * component is exhausted (all of its range has been given to real components),
7310 * so we remove it and restart processing.
7312 * \param[in] env execution environment for this thread
7313 * \param[in,out] lo object to update the layout of
7314 * \param[in] extent layout extent for requested operation, update
7315 * layout to fit this operation
7316 * \param[in] th transaction handle for this operation
7317 * \param[in,out] max_comp the highest comp for the portion of the layout
7318 * we are operating on (For FLR, the chosen
7319 * replica). Updated because we may remove
7321 * \param[in] index index of the extension space component we're
7323 * \param[in] write if this is write op
7324 * \param[in,out] force if the extension is to be forced; set here
7325 to force it on the 2nd call for the same
7328 * \retval 0 on success
7329 * \retval negative errno on error
7331 static int lod_sel_handler(const struct lu_env *env,
7332 struct lod_object *lo,
7333 struct lu_extent *extent,
7334 struct thandle *th, int *max_comp,
7335 int index, int write,
7336 struct sel_data *sd)
7338 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7339 struct lod_thread_info *info = lod_env_info(env);
7340 struct lod_layout_component *lod_comp;
7341 struct lod_layout_component *prev;
7342 struct lod_layout_component *next = NULL;
7343 __u64 extension_size, reserve;
7350 /* First component cannot be extension space */
7352 CERROR("%s: "DFID" first component cannot be extension space\n",
7353 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
7357 lod_comp = &lo->ldo_comp_entries[index];
7358 prev = &lo->ldo_comp_entries[index - 1];
7359 if ((index + 1) < *max_comp)
7360 next = &lo->ldo_comp_entries[index + 1];
7362 /* extension size uses the stripe size field as KiB */
7363 extension_size = lod_comp->llc_stripe_size * SEL_UNIT_SIZE;
7365 CDEBUG(D_LAYOUT, "prev start %llu, extension start %llu, extension end"
7366 " %llu, extension size %llu\n", prev->llc_extent.e_start,
7367 lod_comp->llc_extent.e_start, lod_comp->llc_extent.e_end,
7370 /* Two extension space components cannot be adjacent & extension space
7371 * components cannot be init */
7372 if ((prev->llc_flags & LCME_FL_EXTENSION) ||
7373 !(ergo(next, !(next->llc_flags & LCME_FL_EXTENSION))) ||
7374 lod_comp_inited(lod_comp)) {
7375 CERROR("%s: "DFID" invalid extension space components\n",
7376 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
7380 reserve = lod_sel_stripe_reserved(lod_comp);
7382 if (!prev->llc_stripe) {
7383 CDEBUG(D_LAYOUT, "Previous component not inited\n");
7384 info->lti_count = 1;
7385 info->lti_comp_idx[0] = index - 1;
7386 rc = lod_declare_instantiate_components(env, lo, th, reserve);
7387 /* ENOSPC tells us we can't use this component. If there is
7388 * a next or we are repeating, we either spill over (next) or
7389 * extend the original comp (repeat). Otherwise, return the
7390 * error to the user. */
7391 if (rc == -ENOSPC && (next || sd->sd_repeat))
7397 if (sd->sd_force == 0 && rc == 0)
7398 rc = !lod_sel_osts_allowed(env, lo, index - 1, reserve, extent,
7399 &lod_comp->llc_extent, write);
7401 repeated = !!(sd->sd_repeat);
7405 /* Extend previous component */
7407 new_end = lod_extension_new_end(extension_size, extent->e_end,
7408 prev->llc_stripe_size,
7409 prev->llc_extent.e_end,
7410 lod_comp->llc_extent.e_end);
7412 CDEBUG(D_LAYOUT, "new end %llu\n", new_end);
7413 lod_comp->llc_extent.e_start = new_end;
7414 prev->llc_extent.e_end = new_end;
7416 if (prev->llc_extent.e_end == lod_comp->llc_extent.e_end) {
7417 CDEBUG(D_LAYOUT, "Extension component exhausted\n");
7418 lod_comp->llc_id = LCME_ID_INVAL;
7422 /* rc == 1, failed to extend current component */
7425 /* Normal 'spillover' case - Remove the extension
7426 * space component & bring down the start of the next
7428 lod_comp->llc_id = LCME_ID_INVAL;
7430 if (!(prev->llc_flags & LCME_FL_INIT)) {
7431 prev->llc_id = LCME_ID_INVAL;
7434 lod_sel_adjust_extents(env, lo, *max_comp, index);
7435 } else if (lod_comp_inited(prev)) {
7436 /* If there is no next, and the previous component is
7437 * INIT'ed, try repeating the previous component. */
7438 LASSERT(repeated == 0);
7439 rc = lod_layout_repeat_comp(env, lo, index - 1);
7443 /* The previous component is a repeated component.
7444 * Record this so we don't keep trying to repeat it. */
7447 /* If the previous component is not INIT'ed, this may
7448 * be a component we have just instantiated but failed
7449 * to extend. Or even a repeated component we failed
7450 * to prepare a striping for. Do not repeat but instead
7451 * remove the repeated component & force the extention
7452 * of the original one */
7455 prev->llc_id = LCME_ID_INVAL;
7462 rc = lod_layout_del_prep_layout(env, lo, NULL);
7465 LASSERTF(-rc == change,
7466 "number deleted %d != requested %d\n", -rc,
7469 *max_comp = *max_comp + change;
7471 /* lod_del_prep_layout reallocates ldo_comp_entries, so we must
7472 * refresh these pointers before using them */
7473 lod_comp = &lo->ldo_comp_entries[index];
7474 prev = &lo->ldo_comp_entries[index - 1];
7475 CDEBUG(D_LAYOUT, "After extent updates: prev start %llu, current start "
7476 "%llu, current end %llu max_comp %d ldo_comp_cnt %d\n",
7477 prev->llc_extent.e_start, lod_comp->llc_extent.e_start,
7478 lod_comp->llc_extent.e_end, *max_comp, lo->ldo_comp_cnt);
7480 /* Layout changed successfully */
7485 * Declare layout extent updates
7487 * Handles extensions. Identifies extension components touched by current
7488 * operation and passes them to processing function.
7490 * Restarts with updated layouts from the processing function until the current
7491 * operation no longer touches an extension space component.
7493 * \param[in] env execution environment for this thread
7494 * \param[in,out] lo object to update the layout of
7495 * \param[in] extent layout extent for requested operation, update layout to
7496 * fit this operation
7497 * \param[in] th transaction handle for this operation
7498 * \param[in] pick identifies chosen mirror for FLR layouts
7499 * \param[in] write if this is write op
7501 * \retval 1 on layout changed, 0 on no change
7502 * \retval negative errno on error
7504 static int lod_declare_update_extents(const struct lu_env *env,
7505 struct lod_object *lo, struct lu_extent *extent,
7506 struct thandle *th, int pick, int write)
7508 struct lod_thread_info *info = lod_env_info(env);
7509 struct lod_layout_component *lod_comp;
7510 bool layout_changed = false;
7511 struct sel_data sd = { 0 };
7519 /* This makes us work on the components of the chosen mirror */
7520 if (lo->ldo_mirrors) {
7521 start_index = lo->ldo_mirrors[pick].lme_start;
7522 max_comp = lo->ldo_mirrors[pick].lme_end + 1;
7525 max_comp = lo->ldo_comp_cnt;
7527 if (lo->ldo_flr_state == LCM_FL_NONE)
7528 LASSERT(start_index == 0 && max_comp == lo->ldo_comp_cnt);
7530 CDEBUG(D_LAYOUT, "extent->e_start %llu, extent->e_end %llu\n",
7531 extent->e_start, extent->e_end);
7532 for (i = start_index; i < max_comp; i++) {
7533 lod_comp = &lo->ldo_comp_entries[i];
7535 /* We've passed all components of interest */
7536 if (lod_comp->llc_extent.e_start >= extent->e_end)
7539 if (lod_comp->llc_flags & LCME_FL_EXTENSION) {
7540 layout_changed = true;
7541 rc = lod_sel_handler(env, lo, extent, th, &max_comp,
7546 /* Nothing has changed behind the prev one */
7552 /* We may have added or removed components. If so, we must update the
7553 * start & ends of all the mirrors after the current one, and the end
7554 * of the current mirror. */
7555 if (lo->ldo_mirrors) {
7556 change = max_comp - 1 - lo->ldo_mirrors[pick].lme_end;
7558 lo->ldo_mirrors[pick].lme_end += change;
7559 for (i = pick + 1; i < lo->ldo_mirror_count; i++) {
7560 lo->ldo_mirrors[i].lme_start += change;
7561 lo->ldo_mirrors[i].lme_end += change;
7568 /* The amount of components has changed, adjust the lti_comp_idx */
7569 rc2 = lod_layout_data_init(info, lo->ldo_comp_cnt);
7571 return rc < 0 ? rc : rc2 < 0 ? rc2 : layout_changed;
7574 /* If striping is already instantiated or INIT'ed DOM? */
7575 static bool lod_is_instantiation_needed(struct lod_layout_component *comp)
7577 if (comp->llc_magic == LOV_MAGIC_FOREIGN)
7580 return !(((lov_pattern(comp->llc_pattern) & LOV_PATTERN_MDT) &&
7581 lod_comp_inited(comp)) || comp->llc_stripe);
7585 * Declare layout update for a non-FLR layout.
7587 * \param[in] env execution environment for this thread
7588 * \param[in,out] lo object to update the layout of
7589 * \param[in] layout layout intent for requested operation, "update" is
7590 * a process of reacting to this
7591 * \param[in] buf buffer containing lov ea (see comment on usage inline)
7592 * \param[in] th transaction handle for this operation
7594 * \retval 0 on success
7595 * \retval negative errno on error
7597 static int lod_declare_update_plain(const struct lu_env *env,
7598 struct lod_object *lo, struct layout_intent *layout,
7599 const struct lu_buf *buf, struct thandle *th)
7601 struct lod_thread_info *info = lod_env_info(env);
7602 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7603 struct lod_layout_component *lod_comp;
7604 struct lov_comp_md_v1 *comp_v1 = NULL;
7605 bool layout_changed = false;
7606 bool replay = false;
7610 LASSERT(lo->ldo_flr_state == LCM_FL_NONE);
7613 * In case the client is passing lovea, which only happens during
7614 * the replay of layout intent write RPC for now, we may need to
7615 * parse the lovea and apply new layout configuration.
7617 if (buf && buf->lb_len) {
7618 struct lov_user_md_v1 *v1 = buf->lb_buf;
7620 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
7621 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
7622 LOV_MAGIC_COMP_V1)) {
7623 CERROR("%s: the replay buffer of layout extend "
7624 "(magic %#x) does not contain expected "
7625 "composite layout.\n",
7626 lod2obd(d)->obd_name, v1->lmm_magic);
7627 GOTO(out, rc = -EINVAL);
7630 rc = lod_use_defined_striping(env, lo, buf);
7633 lo->ldo_comp_cached = 1;
7635 rc = lod_get_lov_ea(env, lo);
7638 /* old on-disk EA is stored in info->lti_buf */
7639 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
7641 layout_changed = true;
7643 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
7647 /* non replay path */
7648 rc = lod_striping_load(env, lo);
7653 /* Make sure defined layout covers the requested write range. */
7654 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
7655 if (lo->ldo_comp_cnt > 1 &&
7656 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
7657 lod_comp->llc_extent.e_end < layout->lai_extent.e_end) {
7658 CDEBUG_LIMIT(replay ? D_ERROR : D_LAYOUT,
7659 "%s: the defined layout [0, %#llx) does not "
7660 "covers the write range "DEXT"\n",
7661 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
7662 PEXT(&layout->lai_extent));
7663 GOTO(out, rc = -EINVAL);
7666 CDEBUG(D_LAYOUT, "%s: "DFID": update components "DEXT"\n",
7667 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
7668 PEXT(&layout->lai_extent));
7671 rc = lod_declare_update_extents(env, lo, &layout->lai_extent,
7672 th, 0, layout->lai_opc == LAYOUT_INTENT_WRITE);
7676 layout_changed = true;
7680 * Iterate ld->ldo_comp_entries, find the component whose extent under
7681 * the write range and not instantianted.
7683 for (i = 0; i < lo->ldo_comp_cnt; i++) {
7684 lod_comp = &lo->ldo_comp_entries[i];
7686 if (lod_comp->llc_extent.e_start >= layout->lai_extent.e_end)
7690 /* If striping is instantiated or INIT'ed DOM skip */
7691 if (!lod_is_instantiation_needed(lod_comp))
7695 * In replay path, lod_comp is the EA passed by
7696 * client replay buffer, comp_v1 is the pre-recovery
7697 * on-disk EA, we'd sift out those components which
7698 * were init-ed in the on-disk EA.
7700 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
7705 * this component hasn't instantiated in normal path, or during
7706 * replay it needs replay the instantiation.
7709 /* A released component is being extended */
7710 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
7711 GOTO(out, rc = -EINVAL);
7713 LASSERT(info->lti_comp_idx != NULL);
7714 info->lti_comp_idx[info->lti_count++] = i;
7715 layout_changed = true;
7718 if (!layout_changed)
7721 lod_obj_inc_layout_gen(lo);
7722 rc = lod_declare_instantiate_components(env, lo, th, 0);
7726 lod_striping_free(env, lo);
7730 static inline int lod_comp_index(struct lod_object *lo,
7731 struct lod_layout_component *lod_comp)
7733 LASSERT(lod_comp >= lo->ldo_comp_entries &&
7734 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
7736 return lod_comp - lo->ldo_comp_entries;
7740 * Stale other mirrors by writing extent.
7742 static int lod_stale_components(const struct lu_env *env, struct lod_object *lo,
7743 int primary, struct lu_extent *extent,
7746 struct lod_layout_component *pri_comp, *lod_comp;
7747 struct lu_extent pri_extent;
7752 /* The writing extent decides which components in the primary
7753 * are affected... */
7754 CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
7757 lod_foreach_mirror_comp(pri_comp, lo, primary) {
7758 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
7761 CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
7762 lod_comp_index(lo, pri_comp),
7763 PEXT(&pri_comp->llc_extent));
7765 pri_extent.e_start = pri_comp->llc_extent.e_start;
7766 pri_extent.e_end = pri_comp->llc_extent.e_end;
7768 for (i = 0; i < lo->ldo_mirror_count; i++) {
7772 rc = lod_declare_update_extents(env, lo, &pri_extent,
7774 /* if update_extents changed the layout, it may have
7775 * reallocated the component array, so start over to
7776 * avoid using stale pointers */
7782 /* ... and then stale other components that are
7783 * overlapping with primary components */
7784 lod_foreach_mirror_comp(lod_comp, lo, i) {
7785 if (!lu_extent_is_overlapped(
7787 &lod_comp->llc_extent))
7790 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
7791 i, lod_comp_index(lo, lod_comp));
7793 lod_comp->llc_flags |= LCME_FL_STALE;
7794 lo->ldo_mirrors[i].lme_stale = 1;
7795 if (lod_is_hsm(lod_comp))
7796 lod_comp->llc_foreign_flags |= HS_DIRTY;
7805 * check an OST's availability
7806 * \param[in] env execution environment
7807 * \param[in] lo lod object
7808 * \param[in] dt dt object
7809 * \param[in] index mirror index
7811 * \retval negative if failed
7812 * \retval 1 if \a dt is available
7813 * \retval 0 if \a dt is not available
7815 static inline int lod_check_ost_avail(const struct lu_env *env,
7816 struct lod_object *lo,
7817 struct dt_object *dt, int index)
7819 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7820 struct lod_tgt_desc *ost;
7822 int type = LU_SEQ_RANGE_OST;
7825 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &idx, &type);
7827 CERROR("%s: can't locate "DFID":rc = %d\n",
7828 lod2obd(lod)->obd_name, PFID(lu_object_fid(&dt->do_lu)),
7833 ost = OST_TGT(lod, idx);
7834 if (ost->ltd_active == 0) {
7835 CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail\n",
7836 PFID(lod_object_fid(lo)), index, idx);
7844 * Pick primary mirror for write
7845 * \param[in] env execution environment
7846 * \param[in] lo object
7847 * \param[in] extent write range
7849 static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo,
7850 struct lu_extent *extent)
7852 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7853 unsigned int seq = 0;
7854 struct lod_layout_component *lod_comp;
7856 int picked = -1, second_pick = -1, third_pick = -1;
7859 if (CFS_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
7860 get_random_bytes(&seq, sizeof(seq));
7861 seq %= lo->ldo_mirror_count;
7865 * Pick a mirror as the primary, and check the availability of OSTs.
7867 * This algo can be revised later after knowing the topology of
7870 lod_qos_statfs_update(env, lod, &lod->lod_ost_descs);
7872 rc = lod_fill_mirrors(lo);
7876 for (i = 0; i < lo->ldo_mirror_count; i++) {
7877 bool ost_avail = true;
7878 int index = (i + seq) % lo->ldo_mirror_count;
7880 if (lo->ldo_mirrors[index].lme_stale) {
7881 CDEBUG(D_LAYOUT, DFID": mirror %d stale\n",
7882 PFID(lod_object_fid(lo)), index);
7886 /* 2nd pick is for the primary mirror containing unavail OST */
7887 if (lo->ldo_mirrors[index].lme_prefer && second_pick < 0)
7888 second_pick = index;
7890 /* 3rd pick is for non-primary mirror containing unavail OST */
7891 if (second_pick < 0 && third_pick < 0)
7895 * we found a non-primary 1st pick, we'd like to find a
7896 * potential pirmary mirror.
7898 if (picked >= 0 && !lo->ldo_mirrors[index].lme_prefer)
7901 /* check the availability of OSTs */
7902 lod_foreach_mirror_comp(lod_comp, lo, index) {
7903 if (!lod_comp_inited(lod_comp) || !lod_comp->llc_stripe)
7906 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
7907 struct dt_object *dt = lod_comp->llc_stripe[j];
7909 rc = lod_check_ost_avail(env, lo, dt, index);
7916 } /* for all dt object in one component */
7919 } /* for all components in a mirror */
7922 * the OSTs where allocated objects locates in the components
7923 * of the mirror are available.
7928 /* this mirror has all OSTs available */
7932 * primary with all OSTs are available, this is the perfect
7935 if (lo->ldo_mirrors[index].lme_prefer)
7937 } /* for all mirrors */
7939 /* failed to pick a sound mirror, lower our expectation */
7941 picked = second_pick;
7943 picked = third_pick;
7950 static int lod_prepare_resync_mirror(const struct lu_env *env,
7951 struct lod_object *lo,
7954 struct lod_thread_info *info = lod_env_info(env);
7955 struct lod_layout_component *lod_comp;
7956 bool neg = !!(MIRROR_ID_NEG & mirror_id);
7959 mirror_id &= ~MIRROR_ID_NEG;
7961 for (i = 0; i < lo->ldo_mirror_count; i++) {
7962 if ((!neg && lo->ldo_mirrors[i].lme_id != mirror_id) ||
7963 (neg && lo->ldo_mirrors[i].lme_id == mirror_id))
7966 lod_foreach_mirror_comp(lod_comp, lo, i) {
7967 if (lod_comp_inited(lod_comp))
7970 info->lti_comp_idx[info->lti_count++] =
7971 lod_comp_index(lo, lod_comp);
7979 * figure out the components should be instantiated for resync.
7981 static int lod_prepare_resync(const struct lu_env *env, struct lod_object *lo,
7982 struct lu_extent *extent)
7984 struct lod_thread_info *info = lod_env_info(env);
7985 struct lod_layout_component *lod_comp;
7986 unsigned int need_sync = 0;
7990 DFID": instantiate all stale components in "DEXT"\n",
7991 PFID(lod_object_fid(lo)), PEXT(extent));
7994 * instantiate all components within this extent, even non-stale
7997 for (i = 0; i < lo->ldo_mirror_count; i++) {
7998 if (!lo->ldo_mirrors[i].lme_stale)
8001 lod_foreach_mirror_comp(lod_comp, lo, i) {
8002 if (!lu_extent_is_overlapped(extent,
8003 &lod_comp->llc_extent))
8008 if (lod_comp_inited(lod_comp))
8011 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
8012 i, lod_comp_index(lo, lod_comp));
8013 info->lti_comp_idx[info->lti_count++] =
8014 lod_comp_index(lo, lod_comp);
8018 return need_sync ? 0 : -EALREADY;
8021 static struct lod_layout_component *
8022 lod_locate_comp_hsm(struct lod_object *lo, int *hsm_mirror_id)
8024 struct lod_layout_component *lod_comp = NULL;
8027 if (!lo->ldo_is_composite)
8030 for (i = 0; i < lo->ldo_mirror_count; i++) {
8032 * FIXME: In the current design, there is only one HSM
8033 * mirror component in range [0, EOF] for a FLR file. This
8034 * should be fixed to support multiple HSM mirror components
8035 * with different HSM backend types and partial file ranges
8038 if (lo->ldo_mirrors[i].lme_hsm) {
8044 start_idx = lo->ldo_mirrors[i].lme_start;
8045 end_idx = lo->ldo_mirrors[i].lme_end;
8046 LASSERT(start_idx == end_idx);
8047 lod_comp = &lo->ldo_comp_entries[start_idx];
8048 LASSERT(lo->ldo_is_composite && lod_is_hsm(lod_comp) &&
8049 lod_comp->llc_extent.e_start == 0 &&
8050 lod_comp->llc_extent.e_end == LUSTRE_EOF);
8058 static int lod_declare_pccro_set(const struct lu_env *env,
8059 struct dt_object *dt, struct thandle *th)
8061 struct lod_thread_info *info = lod_env_info(env);
8062 struct lu_buf *buf = &info->lti_buf;
8063 struct lod_object *lo = lod_dt_obj(dt);
8064 struct lod_layout_component *lod_comp;
8065 struct lod_layout_component *comp_array;
8066 struct lod_mirror_entry *mirror_array;
8076 rc = lod_striping_load(env, lo);
8080 if (lo->ldo_flr_state & LCM_FL_PCC_RDONLY)
8083 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
8087 lod_comp = lod_locate_comp_hsm(lo, &hsm_mirror_id);
8089 if (lod_comp->llc_foreign_flags & HS_PCCRO) {
8090 CDEBUG(D_LAYOUT, "bad HSM flags: %#x\n",
8091 lod_comp->llc_foreign_flags);
8095 lod_obj_inc_layout_gen(lo);
8096 lod_comp->llc_foreign_flags |= HS_PCCRO;
8097 lod_comp->llc_foreign_flags &= ~HS_DIRTY;
8098 lod_comp->llc_flags &= ~LCME_FL_STALE;
8099 lo->ldo_mirrors[hsm_mirror_id].lme_stale = 0;
8100 lo->ldo_flr_state |= LCM_FL_PCC_RDONLY;
8101 buf->lb_len = lod_comp_md_size(lo, false);
8102 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
8103 buf, XATTR_NAME_LOV, 0, th);
8108 * Create an new composite layout with only one HSM component.
8109 * Field @lhm_archive_uuid is used to be the identifier within HSM
8110 * backend for the archive copy. In the PCC case with a POSIX archive,
8111 * This can just be the original inode FID. This is important because
8112 * the inode FID may change due to layout swaps or migration to a new
8113 * MDT, and we do not want that to cause problems with finding the copy
8116 mirror_cnt = lo->ldo_mirror_count + 1;
8117 if (!lo->ldo_is_composite) {
8118 LASSERT(lo->ldo_mirror_count == 0);
8122 OBD_ALLOC_PTR_ARRAY(mirror_array, mirror_cnt);
8123 if (mirror_array == NULL)
8126 new_cnt = lo->ldo_comp_cnt + 1;
8127 OBD_ALLOC_PTR_ARRAY(comp_array, new_cnt);
8128 if (comp_array == NULL) {
8129 OBD_FREE_PTR_ARRAY(mirror_array, mirror_cnt);
8134 for (i = 0; i < lo->ldo_comp_cnt; i++) {
8135 lod_comp = &lo->ldo_comp_entries[i];
8138 * Add mirror from a non-flr file, create new mirror ID.
8139 * Otherwise, keep existing mirror's component ID, used
8140 * for mirror extension.
8142 if (lo->ldo_mirror_count == 0 &&
8143 mirror_id_of(lod_comp->llc_id) == 0)
8144 lod_comp->llc_id = pflr_id(1, i + 1);
8146 if (lod_comp->llc_id != LCME_ID_INVAL &&
8147 mirror_id_of(lod_comp->llc_id) > mirror_id)
8148 mirror_id = mirror_id_of(lod_comp->llc_id);
8150 if (!lo->ldo_is_composite) {
8151 lod_comp->llc_extent.e_start = 0;
8152 lod_comp->llc_extent.e_end = LUSTRE_EOF;
8153 lod_comp_set_init(lod_comp);
8157 memcpy(comp_array, lo->ldo_comp_entries,
8158 sizeof(*comp_array) * lo->ldo_comp_cnt);
8160 lod_comp = &comp_array[new_cnt - 1];
8161 lod_comp->llc_magic = LOV_MAGIC_FOREIGN;
8162 lod_comp->llc_extent.e_start = 0;
8163 lod_comp->llc_extent.e_end = LUSTRE_EOF;
8164 lod_comp->llc_length = sizeof(struct lov_hsm_base);
8165 lod_comp->llc_type = LU_FOREIGN_TYPE_PCCRO;
8166 lod_comp->llc_foreign_flags = HS_EXISTS | HS_ARCHIVED | HS_PCCRO;
8167 memset(&lod_comp->llc_hsm, 0, sizeof(lod_comp->llc_hsm));
8169 if (lo->ldo_mirrors)
8170 OBD_FREE_PTR_ARRAY(lo->ldo_mirrors, lo->ldo_mirror_count);
8171 OBD_FREE_PTR_ARRAY(lo->ldo_comp_entries, lo->ldo_comp_cnt);
8174 * The @ldo_mirror will be refilled by lod_fill_mirrors() when
8175 * call lod_striped_create() for layout change.
8177 lo->ldo_mirrors = mirror_array;
8178 lo->ldo_mirror_count = mirror_cnt;
8179 lo->ldo_comp_entries = comp_array;
8180 lo->ldo_comp_cnt = new_cnt;
8181 lo->ldo_is_composite = 1;
8184 lod_comp->llc_id = LCME_ID_INVAL;
8185 lod_comp->llc_id = lod_gen_component_id(lo, mirror_id, new_cnt - 1);
8187 if (lo->ldo_flr_state == LCM_FL_NONE)
8188 lo->ldo_flr_state = LCM_FL_RDONLY;
8189 lo->ldo_flr_state |= LCM_FL_PCC_RDONLY;
8190 buf->lb_len = lod_comp_md_size(lo, false);
8191 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
8192 buf, XATTR_NAME_LOV, 0, th);
8194 lod_striping_free(env, lo);
8200 * TODO: When clear LCM_FL_PCC_RDONLY flag from the layouts, it means the file
8201 * is going to be modified. Currently it needs two RPCs: first one is to clear
8202 * LCM_FL_PCC_RDONLY flag; the second one is to pick primary mirror and mark
8203 * the file as LCM_FL_WRITE_PENDING.
8204 * These two RPCs can be combined in one RPC call.
8206 static int lod_declare_pccro_clear(const struct lu_env *env,
8207 struct dt_object *dt, struct thandle *th)
8209 struct lod_thread_info *info = lod_env_info(env);
8210 struct lod_object *lo = lod_dt_obj(dt);
8211 struct lod_layout_component *lod_comp;
8216 rc = lod_striping_load(env, lo);
8220 if (!(lo->ldo_flr_state & LCM_FL_PCC_RDONLY))
8223 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
8227 lod_comp = lod_locate_comp_hsm(lo, NULL);
8228 if (lod_comp == NULL) {
8229 CDEBUG(D_LAYOUT, "Not found any HSM component\n");
8230 GOTO(out, rc = -EINVAL);
8233 lod_comp->llc_foreign_flags &= ~HS_PCCRO;
8234 lo->ldo_flr_state &= ~LCM_FL_PCC_RDONLY;
8235 lod_obj_inc_layout_gen(lo);
8236 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
8237 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
8238 &info->lti_buf, XATTR_NAME_LOV, 0, th);
8241 lod_striping_free(env, lo);
8246 static int lod_declare_update_pccro(const struct lu_env *env,
8247 struct dt_object *dt,
8248 struct md_layout_change *mlc,
8251 struct layout_intent *intent = mlc->mlc_intent;
8254 switch (intent->lai_opc) {
8255 case LAYOUT_INTENT_PCCRO_SET:
8256 rc = lod_declare_pccro_set(env, dt, th);
8258 case LAYOUT_INTENT_PCCRO_CLEAR:
8259 rc = lod_declare_pccro_clear(env, dt, th);
8269 static int lod_declare_update_rdonly(const struct lu_env *env,
8270 struct lod_object *lo, struct md_layout_change *mlc,
8273 struct lod_thread_info *info = lod_env_info(env);
8274 struct lu_attr *layout_attr = &info->lti_layout_attr;
8275 struct lod_layout_component *lod_comp;
8276 struct lu_extent extent = { 0 };
8280 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
8281 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
8282 mlc->mlc_opc == MD_LAYOUT_RESYNC);
8283 LASSERT(lo->ldo_mirror_count > 0);
8285 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
8286 struct layout_intent *layout = mlc->mlc_intent;
8287 int write = layout->lai_opc == LAYOUT_INTENT_WRITE;
8290 extent = layout->lai_extent;
8291 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
8292 PFID(lod_object_fid(lo)), PEXT(&extent));
8294 picked = lod_primary_pick(env, lo, &extent);
8298 CDEBUG(D_LAYOUT, DFID": picked mirror id %u as primary\n",
8299 PFID(lod_object_fid(lo)),
8300 lo->ldo_mirrors[picked].lme_id);
8302 /* Update extents of primary before staling */
8303 rc = lod_declare_update_extents(env, lo, &extent, th, picked,
8308 if (layout->lai_opc == LAYOUT_INTENT_TRUNC) {
8310 * trunc transfers [0, size) in the intent extent, we'd
8311 * stale components overlapping [size, eof).
8313 extent.e_start = extent.e_end;
8314 extent.e_end = OBD_OBJECT_EOF;
8317 /* stale overlapping components from other mirrors */
8318 rc = lod_stale_components(env, lo, picked, &extent, th);
8322 /* restore truncate intent extent */
8323 if (layout->lai_opc == LAYOUT_INTENT_TRUNC)
8324 extent.e_end = extent.e_start;
8326 /* instantiate components for the picked mirror, start from 0 */
8329 lod_foreach_mirror_comp(lod_comp, lo, picked) {
8330 if (!lu_extent_is_overlapped(&extent,
8331 &lod_comp->llc_extent))
8334 if (!lod_is_instantiation_needed(lod_comp))
8337 info->lti_comp_idx[info->lti_count++] =
8338 lod_comp_index(lo, lod_comp);
8341 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
8342 } else { /* MD_LAYOUT_RESYNC */
8346 * could contain multiple non-stale mirrors, so we need to
8347 * prep uninited all components assuming any non-stale mirror
8348 * could be picked as the primary mirror.
8350 if (mlc->mlc_mirror_id == 0) {
8352 for (i = 0; i < lo->ldo_mirror_count; i++) {
8353 if (lo->ldo_mirrors[i].lme_stale)
8356 lod_foreach_mirror_comp(lod_comp, lo, i) {
8357 if (!lod_comp_inited(lod_comp))
8361 lod_comp->llc_extent.e_end)
8363 lod_comp->llc_extent.e_end;
8366 rc = lod_prepare_resync(env, lo, &extent);
8370 /* mirror write, try to init its all components */
8371 rc = lod_prepare_resync_mirror(env, lo,
8372 mlc->mlc_mirror_id);
8377 /* change the file state to SYNC_PENDING */
8378 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
8381 /* Reset the layout version once it's becoming too large.
8382 * This way it can make sure that the layout version is
8383 * monotonously increased in this writing era. */
8384 lod_obj_inc_layout_gen(lo);
8386 rc = lod_declare_instantiate_components(env, lo, th, 0);
8390 layout_attr->la_valid = LA_LAYOUT_VERSION;
8391 layout_attr->la_layout_version = 0;
8392 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
8393 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
8394 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
8400 lod_striping_free(env, lo);
8404 static int lod_declare_update_write_pending(const struct lu_env *env,
8405 struct lod_object *lo, struct md_layout_change *mlc,
8408 struct lod_thread_info *info = lod_env_info(env);
8409 struct lu_attr *layout_attr = &info->lti_layout_attr;
8410 struct lod_layout_component *lod_comp;
8411 struct lu_extent extent = { 0 };
8417 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
8418 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
8419 mlc->mlc_opc == MD_LAYOUT_RESYNC);
8421 /* look for the first preferred mirror */
8422 for (i = 0; i < lo->ldo_mirror_count; i++) {
8423 if (lo->ldo_mirrors[i].lme_stale)
8425 if (lo->ldo_mirrors[i].lme_prefer == 0)
8427 if (lo->ldo_mirrors[i].lme_hsm)
8434 /* no primary, use any in-sync */
8435 for (i = 0; i < lo->ldo_mirror_count; i++) {
8436 if (lo->ldo_mirrors[i].lme_stale)
8442 CERROR(DFID ": doesn't have a primary mirror\n",
8443 PFID(lod_object_fid(lo)));
8444 GOTO(out, rc = -ENODATA);
8448 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
8449 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
8451 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
8453 /* for LAYOUT_WRITE opc, it has to do the following operations:
8454 * 1. stale overlapping componets from stale mirrors;
8455 * 2. instantiate components of the primary mirror;
8456 * 3. transfter layout version to all objects of the primary;
8458 * for LAYOUT_RESYNC opc, it will do:
8459 * 1. instantiate components of all stale mirrors;
8460 * 2. transfer layout version to all objects to close write era. */
8462 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
8463 struct layout_intent *layout = mlc->mlc_intent;
8464 int write = layout->lai_opc == LAYOUT_INTENT_WRITE;
8466 LASSERT(layout != NULL);
8468 extent = layout->lai_extent;
8470 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
8471 PFID(lod_object_fid(lo)), PEXT(&extent));
8473 /* 1. Update extents of primary before staling */
8474 rc = lod_declare_update_extents(env, lo, &extent, th, primary,
8479 if (layout->lai_opc == LAYOUT_INTENT_TRUNC) {
8481 * trunc transfers [0, size) in the intent extent, we'd
8482 * stale components overlapping [size, eof).
8484 extent.e_start = extent.e_end;
8485 extent.e_end = OBD_OBJECT_EOF;
8488 /* 2. stale overlapping components */
8489 rc = lod_stale_components(env, lo, primary, &extent, th);
8493 /* 3. find the components which need instantiating.
8494 * instantiate [0, mlc->mlc_intent->e_end) */
8496 /* restore truncate intent extent */
8497 if (layout->lai_opc == LAYOUT_INTENT_TRUNC)
8498 extent.e_end = extent.e_start;
8501 lod_foreach_mirror_comp(lod_comp, lo, primary) {
8502 if (!lu_extent_is_overlapped(&extent,
8503 &lod_comp->llc_extent))
8506 if (!lod_is_instantiation_needed(lod_comp))
8509 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
8510 primary, lod_comp_index(lo, lod_comp));
8511 info->lti_comp_idx[info->lti_count++] =
8512 lod_comp_index(lo, lod_comp);
8514 } else { /* MD_LAYOUT_RESYNC */
8515 if (mlc->mlc_mirror_id == 0) {
8517 lod_foreach_mirror_comp(lod_comp, lo, primary) {
8518 if (!lod_comp_inited(lod_comp))
8521 extent.e_end = lod_comp->llc_extent.e_end;
8524 rc = lod_prepare_resync(env, lo, &extent);
8528 /* mirror write, try to init its all components */
8529 rc = lod_prepare_resync_mirror(env, lo,
8530 mlc->mlc_mirror_id);
8535 /* change the file state to SYNC_PENDING */
8536 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
8539 rc = lod_declare_instantiate_components(env, lo, th, 0);
8543 lod_obj_inc_layout_gen(lo);
8545 /* 3. transfer layout version to OST objects.
8546 * transfer new layout version to OST objects so that stale writes
8547 * can be denied. It also ends an era of writing by setting
8548 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
8549 * send write RPC; only resync RPCs could do it. */
8550 layout_attr->la_valid = LA_LAYOUT_VERSION;
8551 layout_attr->la_layout_version = 0;
8552 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
8553 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
8554 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
8559 lod_striping_free(env, lo);
8563 static int lod_declare_update_sync_pending(const struct lu_env *env,
8564 struct lod_object *lo, struct md_layout_change *mlc,
8567 struct lod_thread_info *info = lod_env_info(env);
8568 struct lu_attr *layout_attr = &info->lti_layout_attr;
8569 unsigned sync_components = 0;
8570 unsigned resync_components = 0;
8575 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
8576 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
8577 mlc->mlc_opc == MD_LAYOUT_WRITE);
8579 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
8580 PFID(lod_object_fid(lo)), mlc->mlc_opc);
8582 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
8583 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
8584 PFID(lod_object_fid(lo)));
8586 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
8587 return lod_declare_update_write_pending(env, lo, mlc, th);
8590 /* MD_LAYOUT_RESYNC_DONE */
8592 for (i = 0; i < lo->ldo_comp_cnt; i++) {
8593 struct lod_layout_component *lod_comp;
8596 lod_comp = &lo->ldo_comp_entries[i];
8598 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
8603 for (j = 0; j < mlc->mlc_resync_count; j++) {
8604 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
8607 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
8608 lod_comp->llc_flags &= ~LCME_FL_STALE;
8609 resync_components++;
8615 for (i = 0; i < mlc->mlc_resync_count; i++) {
8616 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
8619 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
8620 "or already synced\n", PFID(lod_object_fid(lo)),
8621 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
8622 GOTO(out, rc = -EINVAL);
8625 if (!sync_components || (mlc->mlc_resync_count && !resync_components)) {
8626 CDEBUG(D_LAYOUT, DFID": no mirror in sync\n",
8627 PFID(lod_object_fid(lo)));
8629 /* tend to return an error code here to prevent
8630 * the MDT from setting SoM attribute */
8631 GOTO(out, rc = -EINVAL);
8634 CDEBUG(D_LAYOUT, DFID": synced %u resynced %u/%zu components\n",
8635 PFID(lod_object_fid(lo)),
8636 sync_components, resync_components, mlc->mlc_resync_count);
8638 lo->ldo_flr_state = LCM_FL_RDONLY;
8639 lod_obj_inc_layout_gen(lo);
8641 layout_attr->la_valid = LA_LAYOUT_VERSION;
8642 layout_attr->la_layout_version = 0;
8643 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
8647 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
8648 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
8649 &info->lti_buf, XATTR_NAME_LOV, 0, th);
8654 lod_striping_free(env, lo);
8658 typedef int (*mlc_handler)(const struct lu_env *env, struct dt_object *dt,
8659 const struct md_layout_change *mlc,
8660 struct thandle *th);
8663 * Attach stripes after target's for migrating directory. NB, we
8664 * only need to declare this, the actual work is done inside
8665 * lod_xattr_set_lmv().
8667 * \param[in] env execution environment
8668 * \param[in] dt target object
8669 * \param[in] mlc layout change data
8670 * \param[in] th transaction handle
8672 * \retval 0 on success
8673 * \retval negative if failed
8675 static int lod_dir_declare_layout_attach(const struct lu_env *env,
8676 struct dt_object *dt,
8677 const struct md_layout_change *mlc,
8680 struct lod_thread_info *info = lod_env_info(env);
8681 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
8682 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
8683 struct lod_object *lo = lod_dt_obj(dt);
8684 struct dt_object *next = dt_object_child(dt);
8685 struct dt_object_format *dof = &info->lti_format;
8686 struct lmv_mds_md_v1 *lmv = mlc->mlc_buf.lb_buf;
8687 struct dt_object **stripes;
8688 __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
8689 struct lu_fid *fid = &info->lti_fid;
8690 struct lod_tgt_desc *tgt;
8691 struct dt_object *dto;
8692 struct dt_device *tgt_dt;
8693 int type = LU_SEQ_RANGE_ANY;
8694 struct dt_insert_rec *rec = &info->lti_dt_rec;
8695 char *stripe_name = info->lti_key;
8696 struct lu_name *sname;
8697 struct linkea_data ldata = { NULL };
8698 struct lu_buf linkea_buf;
8705 if (!lmv_is_sane(lmv))
8708 if (!dt_try_as_dir(env, dt, false))
8711 dof->dof_type = DFT_DIR;
8713 OBD_ALLOC_PTR_ARRAY(stripes, (lo->ldo_dir_stripe_count + stripe_count));
8717 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
8718 stripes[i] = lo->ldo_stripe[i];
8720 rec->rec_type = S_IFDIR;
8722 for (i = 0; i < stripe_count; i++) {
8724 &lmv->lmv_stripe_fids[i]);
8725 if (!fid_is_sane(fid))
8728 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
8732 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
8733 tgt_dt = lod->lod_child;
8735 tgt = LTD_TGT(ltd, idx);
8737 GOTO(out, rc = -ESTALE);
8738 tgt_dt = tgt->ltd_tgt;
8741 dto = dt_locate_at(env, tgt_dt, fid,
8742 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
8745 GOTO(out, rc = PTR_ERR(dto));
8747 stripes[i + lo->ldo_dir_stripe_count] = dto;
8749 if (!dt_try_as_dir(env, dto, true))
8750 GOTO(out, rc = -ENOTDIR);
8752 rc = lod_sub_declare_ref_add(env, dto, th);
8756 rec->rec_fid = lu_object_fid(&dto->do_lu);
8757 rc = lod_sub_declare_insert(env, dto,
8758 (const struct dt_rec *)rec,
8759 (const struct dt_key *)dot, th);
8763 rc = lod_sub_declare_insert(env, dto,
8764 (const struct dt_rec *)rec,
8765 (const struct dt_key *)dotdot, th);
8769 rc = lod_sub_declare_xattr_set(env, dto, &mlc->mlc_buf,
8770 XATTR_NAME_LMV, 0, th);
8774 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
8775 PFID(lu_object_fid(&dto->do_lu)),
8776 i + lo->ldo_dir_stripe_count);
8778 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
8779 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
8780 sname, lu_object_fid(&dt->do_lu));
8784 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
8785 linkea_buf.lb_len = ldata.ld_leh->leh_len;
8786 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
8787 XATTR_NAME_LINK, 0, th);
8791 rc = lod_sub_declare_insert(env, next,
8792 (const struct dt_rec *)rec,
8793 (const struct dt_key *)stripe_name,
8798 rc = lod_sub_declare_ref_add(env, next, th);
8804 OBD_FREE_PTR_ARRAY(lo->ldo_stripe,
8805 lo->ldo_dir_stripes_allocated);
8806 lo->ldo_stripe = stripes;
8807 lo->ldo_is_foreign = 0;
8808 lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count;
8809 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type);
8810 lo->ldo_dir_stripe_count += stripe_count;
8811 lo->ldo_dir_layout_version++;
8812 lo->ldo_dir_stripes_allocated += stripe_count;
8814 /* plain directory split creates target as a plain directory, while
8815 * after source attached as the first stripe, it becomes a striped
8816 * directory, set correct do_index_ops, otherwise it can't be unlinked.
8818 dt->do_index_ops = &lod_striped_index_ops;
8822 i = lo->ldo_dir_stripe_count;
8823 while (i < lo->ldo_dir_stripe_count + stripe_count && stripes[i])
8824 dt_object_put(env, stripes[i++]);
8826 OBD_FREE_PTR_ARRAY(stripes, stripe_count + lo->ldo_dir_stripe_count);
8830 static int lod_dir_declare_layout_detach(const struct lu_env *env,
8831 struct dt_object *dt,
8832 const struct md_layout_change *unused,
8835 struct lod_thread_info *info = lod_env_info(env);
8836 struct lod_object *lo = lod_dt_obj(dt);
8837 struct dt_object *next = dt_object_child(dt);
8838 char *stripe_name = info->lti_key;
8839 struct dt_object *dto;
8843 if (!dt_try_as_dir(env, dt, true))
8846 if (!lo->ldo_dir_stripe_count)
8847 return lod_sub_declare_delete(env, next,
8848 (const struct dt_key *)dotdot, th);
8850 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8851 dto = lo->ldo_stripe[i];
8855 if (!dt_try_as_dir(env, dto, true))
8858 rc = lod_sub_declare_delete(env, dto,
8859 (const struct dt_key *)dotdot, th);
8863 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8864 PFID(lu_object_fid(&dto->do_lu)), i);
8866 rc = lod_sub_declare_delete(env, next,
8867 (const struct dt_key *)stripe_name, th);
8871 rc = lod_sub_declare_ref_del(env, next, th);
8879 static int dt_dir_is_empty(const struct lu_env *env,
8880 struct dt_object *obj)
8883 const struct dt_it_ops *iops;
8888 if (!dt_try_as_dir(env, obj, true))
8891 iops = &obj->do_index_ops->dio_it;
8892 it = iops->init(env, obj, LUDA_64BITHASH);
8894 RETURN(PTR_ERR(it));
8896 rc = iops->get(env, it, (const struct dt_key *)"");
8900 for (rc = 0, i = 0; rc == 0 && i < 3; ++i)
8901 rc = iops->next(env, it);
8907 /* Huh? Index contains no zero key? */
8912 iops->fini(env, it);
8917 static int lod_dir_declare_layout_shrink(const struct lu_env *env,
8918 struct dt_object *dt,
8919 const struct md_layout_change *mlc,
8922 struct lod_thread_info *info = lod_env_info(env);
8923 struct lod_object *lo = lod_dt_obj(dt);
8924 struct dt_object *next = dt_object_child(dt);
8925 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
8926 char *stripe_name = info->lti_key;
8927 struct lu_buf *lmv_buf = &info->lti_buf;
8928 __u32 final_stripe_count;
8929 struct dt_object *dto;
8935 if (!dt_try_as_dir(env, dt, true))
8938 /* shouldn't be called on plain directory */
8939 LASSERT(lo->ldo_dir_stripe_count);
8941 lmv_buf->lb_buf = &info->lti_lmv.lmv_md_v1;
8942 lmv_buf->lb_len = sizeof(info->lti_lmv.lmv_md_v1);
8944 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
8945 LASSERT(final_stripe_count &&
8946 final_stripe_count < lo->ldo_dir_stripe_count);
8948 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8949 dto = lo->ldo_stripe[i];
8953 if (i < final_stripe_count) {
8954 rc = lod_sub_declare_xattr_set(env, dto, lmv_buf,
8956 LU_XATTR_REPLACE, th);
8963 rc = dt_dir_is_empty(env, dto);
8967 rc = lod_sub_declare_ref_del(env, dto, th);
8971 rc = lod_sub_declare_destroy(env, dto, th);
8975 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8976 PFID(lu_object_fid(&dto->do_lu)), i);
8978 rc = lod_sub_declare_delete(env, next,
8979 (const struct dt_key *)stripe_name, th);
8983 rc = lod_sub_declare_ref_del(env, next, th);
8988 rc = lod_sub_declare_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
8989 LU_XATTR_REPLACE, th);
8994 * Allocate stripes for split directory.
8996 * \param[in] env execution environment
8997 * \param[in] dt target object
8998 * \param[in] mlc layout change data
8999 * \param[in] th transaction handle
9001 * \retval 0 on success
9002 * \retval negative if failed
9004 static int lod_dir_declare_layout_split(const struct lu_env *env,
9005 struct dt_object *dt,
9006 const struct md_layout_change *mlc,
9009 struct lod_thread_info *info = lod_env_info(env);
9010 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
9011 struct lod_object *lo = lod_dt_obj(dt);
9012 struct dt_object_format *dof = &info->lti_format;
9013 struct lmv_user_md_v1 *lum = mlc->mlc_spec->u.sp_ea.eadata;
9014 struct dt_object **stripes;
9015 int mdt_count = lod->lod_remote_mdt_count + 1;
9023 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
9024 LASSERT(le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT);
9026 saved_count = lo->ldo_dir_stripes_allocated;
9027 stripe_count = le32_to_cpu(lum->lum_stripe_count);
9028 if (stripe_count <= saved_count)
9031 /* if the split target is overstriped, we need to put that flag in the
9032 * current layout so it can allocate the larger number of stripes
9034 * Note we need to pick up any hash *flags* which affect allocation
9035 * *before* allocation, so they're used in allocating the directory,
9036 * rather than after when we finalize directory setup (at the end of
9039 if (le32_to_cpu(lum->lum_hash_type) & LMV_HASH_FLAG_OVERSTRIPED)
9040 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_OVERSTRIPED;
9042 if (!(lo->ldo_dir_hash_type & LMV_HASH_FLAG_OVERSTRIPED) &&
9043 stripe_count > mdt_count) {
9045 } else if ((lo->ldo_dir_hash_type & LMV_HASH_FLAG_OVERSTRIPED) &&
9046 (stripe_count > mdt_count * LMV_MAX_STRIPES_PER_MDT ||
9047 /* a single MDT doesn't initialize the infrastructure for striped
9048 * directories, so we just don't support overstriping in that case
9054 dof->dof_type = DFT_DIR;
9056 OBD_ALLOC(stripes, sizeof(*stripes) * stripe_count);
9060 for (i = 0; i < lo->ldo_dir_stripes_allocated; i++)
9061 stripes[i] = lo->ldo_stripe[i];
9063 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
9065 rc = lod_mdt_alloc_qos(env, lo, stripes, saved_count, stripe_count);
9067 rc = lod_mdt_alloc_rr(env, lo, stripes, saved_count,
9070 OBD_FREE(stripes, sizeof(*stripes) * stripe_count);
9074 LASSERT(rc > saved_count);
9075 OBD_FREE(lo->ldo_stripe,
9076 sizeof(*stripes) * lo->ldo_dir_stripes_allocated);
9077 lo->ldo_stripe = stripes;
9078 lo->ldo_is_foreign = 0;
9079 lo->ldo_dir_striped = 1;
9080 lo->ldo_dir_stripe_count = rc;
9081 lo->ldo_dir_stripes_allocated = stripe_count;
9082 lo->ldo_dir_split_hash = lo->ldo_dir_hash_type;
9083 lo->ldo_dir_hash_type = le32_to_cpu(lum->lum_hash_type);
9084 if (!lmv_is_known_hash_type(lo->ldo_dir_hash_type))
9085 lo->ldo_dir_hash_type =
9086 lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
9087 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_SPLIT | LMV_HASH_FLAG_MIGRATION;
9088 lo->ldo_dir_split_offset = saved_count;
9089 lo->ldo_dir_layout_version++;
9090 lo->ldo_dir_stripe_loaded = 1;
9092 rc = lod_dir_declare_create_stripes(env, dt, mlc->mlc_attr, dof, th);
9094 lod_striping_free(env, lo);
9100 * detach all stripes from dir master object, NB, stripes are not destroyed, but
9101 * deleted from it's parent namespace, this function is called in two places:
9102 * 1. mdd_migrate_mdt() detach stripes from source, and attach them to
9104 * 2. mdd_dir_layout_update() detach stripe before turning 1-stripe directory to
9105 * a plain directory.
9107 * \param[in] env execution environment
9108 * \param[in] dt target object
9109 * \param[in] mlc layout change data
9110 * \param[in] th transaction handle
9112 * \retval 0 on success
9113 * \retval negative if failed
9115 static int lod_dir_layout_detach(const struct lu_env *env,
9116 struct dt_object *dt,
9117 const struct md_layout_change *mlc,
9120 struct lod_thread_info *info = lod_env_info(env);
9121 struct lod_object *lo = lod_dt_obj(dt);
9122 struct dt_object *next = dt_object_child(dt);
9123 char *stripe_name = info->lti_key;
9124 struct dt_object *dto;
9130 if (!lo->ldo_dir_stripe_count) {
9131 /* plain directory delete .. */
9132 rc = lod_sub_delete(env, next,
9133 (const struct dt_key *)dotdot, th);
9137 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
9138 dto = lo->ldo_stripe[i];
9142 rc = lod_sub_delete(env, dto,
9143 (const struct dt_key *)dotdot, th);
9147 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
9148 PFID(lu_object_fid(&dto->do_lu)), i);
9150 rc = lod_sub_delete(env, next,
9151 (const struct dt_key *)stripe_name, th);
9155 rc = lod_sub_ref_del(env, next, th);
9160 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
9161 dto = lo->ldo_stripe[i];
9163 dt_object_put(env, dto);
9165 OBD_FREE_PTR_ARRAY(lo->ldo_stripe, lo->ldo_dir_stripes_allocated);
9166 lo->ldo_stripe = NULL;
9167 lo->ldo_dir_stripes_allocated = 0;
9168 lo->ldo_dir_stripe_count = 0;
9169 dt->do_index_ops = &lod_index_ops;
9174 static int lod_dir_layout_shrink(const struct lu_env *env,
9175 struct dt_object *dt,
9176 const struct md_layout_change *mlc,
9179 struct lod_thread_info *info = lod_env_info(env);
9180 struct lod_object *lo = lod_dt_obj(dt);
9181 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
9182 struct dt_object *next = dt_object_child(dt);
9183 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
9184 __u32 final_stripe_count;
9185 char *stripe_name = info->lti_key;
9186 struct dt_object *dto;
9187 struct lu_buf *lmv_buf = &info->lti_buf;
9188 struct lmv_mds_md_v1 *lmv = &info->lti_lmv.lmv_md_v1;
9190 int type = LU_SEQ_RANGE_ANY;
9196 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
9198 lmv_buf->lb_buf = lmv;
9199 lmv_buf->lb_len = sizeof(*lmv);
9200 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
9201 lmv->lmv_stripe_count = cpu_to_le32(final_stripe_count);
9202 lmv->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type) &
9203 cpu_to_le32(LMV_HASH_TYPE_MASK |
9204 LMV_HASH_FLAG_FIXED);
9205 lmv->lmv_layout_version =
9206 cpu_to_le32(lo->ldo_dir_layout_version + 1);
9207 lmv->lmv_migrate_offset = 0;
9208 lmv->lmv_migrate_hash = 0;
9210 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
9211 dto = lo->ldo_stripe[i];
9215 if (i < final_stripe_count) {
9216 rc = lod_fld_lookup(env, lod,
9217 lu_object_fid(&dto->do_lu),
9222 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
9223 rc = lod_sub_xattr_set(env, dto, lmv_buf,
9225 LU_XATTR_REPLACE, th);
9232 dt_write_lock(env, dto, DT_TGT_CHILD);
9233 rc = lod_sub_ref_del(env, dto, th);
9234 dt_write_unlock(env, dto);
9238 rc = lod_sub_destroy(env, dto, th);
9242 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
9243 PFID(lu_object_fid(&dto->do_lu)), i);
9245 rc = lod_sub_delete(env, next,
9246 (const struct dt_key *)stripe_name, th);
9250 rc = lod_sub_ref_del(env, next, th);
9255 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &mdtidx,
9260 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
9261 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
9262 rc = lod_sub_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
9263 LU_XATTR_REPLACE, th);
9267 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
9268 dto = lo->ldo_stripe[i];
9270 dt_object_put(env, dto);
9272 lo->ldo_dir_stripe_count = final_stripe_count;
9277 static mlc_handler dir_mlc_declare_ops[MD_LAYOUT_MAX] = {
9278 [MD_LAYOUT_ATTACH] = lod_dir_declare_layout_attach,
9279 [MD_LAYOUT_DETACH] = lod_dir_declare_layout_detach,
9280 [MD_LAYOUT_SHRINK] = lod_dir_declare_layout_shrink,
9281 [MD_LAYOUT_SPLIT] = lod_dir_declare_layout_split,
9284 static mlc_handler dir_mlc_ops[MD_LAYOUT_MAX] = {
9285 [MD_LAYOUT_DETACH] = lod_dir_layout_detach,
9286 [MD_LAYOUT_SHRINK] = lod_dir_layout_shrink,
9289 static int lod_declare_layout_change(const struct lu_env *env,
9290 struct dt_object *dt, struct md_layout_change *mlc,
9293 struct lod_thread_info *info = lod_env_info(env);
9294 struct lod_object *lo = lod_dt_obj(dt);
9299 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
9300 LASSERT(dir_mlc_declare_ops[mlc->mlc_opc]);
9301 rc = dir_mlc_declare_ops[mlc->mlc_opc](env, dt, mlc, th);
9305 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
9306 dt_object_remote(dt_object_child(dt)))
9309 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
9310 struct layout_intent *intent = mlc->mlc_intent;
9312 if (intent->lai_opc == LAYOUT_INTENT_PCCRO_SET ||
9313 intent->lai_opc == LAYOUT_INTENT_PCCRO_CLEAR) {
9314 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
9317 rc = lod_declare_update_pccro(env, dt, mlc, th);
9322 rc = lod_striping_load(env, lo);
9326 LASSERT(lo->ldo_comp_cnt > 0);
9328 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
9332 switch (lo->ldo_flr_state) {
9334 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
9338 rc = lod_declare_update_rdonly(env, lo, mlc, th);
9340 case LCM_FL_WRITE_PENDING:
9341 rc = lod_declare_update_write_pending(env, lo, mlc, th);
9343 case LCM_FL_SYNC_PENDING:
9344 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
9351 rc = lod_save_layout_gen_intrans(info, lo);
9358 * Instantiate layout component objects which covers the intent write offset.
9360 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
9361 struct md_layout_change *mlc, struct thandle *th)
9363 struct lod_thread_info *info = lod_env_info(env);
9364 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
9365 struct lu_attr *layout_attr = &info->lti_layout_attr;
9366 struct lod_object *lo = lod_dt_obj(dt);
9371 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
9372 LASSERT(dir_mlc_ops[mlc->mlc_opc]);
9373 rc = dir_mlc_ops[mlc->mlc_opc](env, dt, mlc, th);
9377 rc = lod_check_layout_gen_intrans(info, lo);
9380 "%s: obj "DFID" gen changed from %d to %d in transaction, retry the transaction \n",
9381 dt->do_lu.lo_dev->ld_obd->obd_name,
9382 PFID(lu_object_fid(&dt->do_lu)),
9383 info->lti_gen[rc - 1], lo->ldo_layout_gen);
9387 rc = lod_striped_create(env, dt, attr, NULL, th);
9388 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
9389 layout_attr->la_layout_version |= lo->ldo_layout_gen;
9390 rc = lod_attr_set(env, dt, layout_attr, th);
9396 const struct dt_object_operations lod_obj_ops = {
9397 .do_read_lock = lod_read_lock,
9398 .do_write_lock = lod_write_lock,
9399 .do_read_unlock = lod_read_unlock,
9400 .do_write_unlock = lod_write_unlock,
9401 .do_write_locked = lod_write_locked,
9402 .do_attr_get = lod_attr_get,
9403 .do_declare_attr_set = lod_declare_attr_set,
9404 .do_attr_set = lod_attr_set,
9405 .do_xattr_get = lod_xattr_get,
9406 .do_declare_xattr_set = lod_declare_xattr_set,
9407 .do_xattr_set = lod_xattr_set,
9408 .do_declare_xattr_del = lod_declare_xattr_del,
9409 .do_xattr_del = lod_xattr_del,
9410 .do_xattr_list = lod_xattr_list,
9411 .do_ah_init = lod_ah_init,
9412 .do_declare_create = lod_declare_create,
9413 .do_create = lod_create,
9414 .do_declare_destroy = lod_declare_destroy,
9415 .do_destroy = lod_destroy,
9416 .do_index_try = lod_index_try,
9417 .do_declare_ref_add = lod_declare_ref_add,
9418 .do_ref_add = lod_ref_add,
9419 .do_declare_ref_del = lod_declare_ref_del,
9420 .do_ref_del = lod_ref_del,
9421 .do_object_sync = lod_object_sync,
9422 .do_object_lock = lod_object_lock,
9423 .do_object_unlock = lod_object_unlock,
9424 .do_invalidate = lod_invalidate,
9425 .do_declare_layout_change = lod_declare_layout_change,
9426 .do_layout_change = lod_layout_change,
9430 * Implementation of dt_body_operations::dbo_read.
9432 * \see dt_body_operations::dbo_read() in the API description for details.
9434 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
9435 struct lu_buf *buf, loff_t *pos)
9437 struct dt_object *next = dt_object_child(dt);
9439 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
9440 S_ISLNK(dt->do_lu.lo_header->loh_attr));
9441 return next->do_body_ops->dbo_read(env, next, buf, pos);
9445 * Implementation of dt_body_operations::dbo_declare_write.
9447 * \see dt_body_operations::dbo_declare_write() in the API description
9450 static ssize_t lod_declare_write(const struct lu_env *env,
9451 struct dt_object *dt,
9452 const struct lu_buf *buf, loff_t pos,
9455 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
9459 * Implementation of dt_body_operations::dbo_write.
9461 * \see dt_body_operations::dbo_write() in the API description for details.
9463 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
9464 const struct lu_buf *buf, loff_t *pos,
9467 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
9468 S_ISLNK(dt->do_lu.lo_header->loh_attr));
9469 return lod_sub_write(env, dt_object_child(dt), buf, pos, th);
9472 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
9473 __u64 start, __u64 end, struct thandle *th)
9475 if (dt_object_remote(dt))
9478 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
9481 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
9482 __u64 start, __u64 end, struct thandle *th)
9484 if (dt_object_remote(dt))
9487 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
9488 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
9492 * different type of files use the same body_ops because object may be created
9493 * in OUT, where there is no chance to set correct body_ops for each type, so
9494 * body_ops themselves will check file type inside, see lod_read/write/punch for
9497 static const struct dt_body_operations lod_body_ops = {
9498 .dbo_read = lod_read,
9499 .dbo_declare_write = lod_declare_write,
9500 .dbo_write = lod_write,
9501 .dbo_declare_punch = lod_declare_punch,
9502 .dbo_punch = lod_punch,
9506 * Implementation of lu_object_operations::loo_object_init.
9508 * The function determines the type and the index of the target device using
9509 * sequence of the object's FID. Then passes control down to the
9510 * corresponding device:
9511 * OSD for the local objects, OSP for remote
9513 * \see lu_object_operations::loo_object_init() in the API description
9516 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
9517 const struct lu_object_conf *conf)
9519 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
9520 struct lu_device *cdev = NULL;
9521 struct lu_object *cobj;
9522 struct lod_tgt_descs *ltd = NULL;
9523 struct lod_tgt_desc *tgt;
9525 int type = LU_SEQ_RANGE_ANY;
9529 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
9533 if (type == LU_SEQ_RANGE_MDT &&
9534 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
9535 cdev = &lod->lod_child->dd_lu_dev;
9536 } else if (type == LU_SEQ_RANGE_MDT) {
9537 ltd = &lod->lod_mdt_descs;
9539 } else if (type == LU_SEQ_RANGE_OST) {
9540 ltd = &lod->lod_ost_descs;
9547 if (ltd->ltd_tgts_size > idx &&
9548 test_bit(idx, ltd->ltd_tgt_bitmap)) {
9549 tgt = LTD_TGT(ltd, idx);
9551 LASSERT(tgt != NULL);
9552 LASSERT(tgt->ltd_tgt != NULL);
9554 cdev = &(tgt->ltd_tgt->dd_lu_dev);
9556 lod_putref(lod, ltd);
9559 if (unlikely(cdev == NULL))
9562 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
9563 if (unlikely(cobj == NULL))
9566 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
9568 lu_object_add(lo, cobj);
9575 * Release resources associated with striping.
9577 * If the object is striped (regular or directory), then release
9578 * the stripe objects references and free the ldo_stripe array.
9580 * \param[in] env execution environment
9581 * \param[in] lo object
9583 void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo)
9585 struct lod_layout_component *lod_comp;
9586 __u32 obj_attr = lo->ldo_obj.do_lu.lo_header->loh_attr;
9589 if (unlikely(lo->ldo_is_foreign)) {
9590 if (S_ISREG(obj_attr)) {
9591 lod_free_foreign_lov(lo);
9592 lo->ldo_comp_cached = 0;
9593 } else if (S_ISDIR(obj_attr)) {
9594 lod_free_foreign_lmv(lo);
9595 lo->ldo_dir_stripe_loaded = 0;
9597 } else if (lo->ldo_stripe != NULL) {
9598 LASSERT(lo->ldo_comp_entries == NULL);
9599 LASSERT(lo->ldo_dir_stripes_allocated > 0);
9601 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
9602 if (lo->ldo_stripe[i])
9603 dt_object_put(env, lo->ldo_stripe[i]);
9606 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
9607 OBD_FREE(lo->ldo_stripe, j);
9608 lo->ldo_stripe = NULL;
9609 lo->ldo_dir_stripes_allocated = 0;
9610 lo->ldo_dir_stripe_loaded = 0;
9611 lo->ldo_dir_stripe_count = 0;
9612 lo->ldo_obj.do_index_ops = NULL;
9613 } else if (lo->ldo_comp_entries != NULL) {
9614 for (i = 0; i < lo->ldo_comp_cnt; i++) {
9615 /* free lod_layout_component::llc_stripe array */
9616 lod_comp = &lo->ldo_comp_entries[i];
9618 /* HSM layout component */
9619 if (lod_comp->llc_magic == LOV_MAGIC_FOREIGN)
9621 if (lod_comp->llc_stripe == NULL)
9623 LASSERT(lod_comp->llc_stripes_allocated != 0);
9624 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
9625 if (lod_comp->llc_stripe[j] != NULL)
9627 &lod_comp->llc_stripe[j]->do_lu);
9629 OBD_FREE_PTR_ARRAY(lod_comp->llc_stripe,
9630 lod_comp->llc_stripes_allocated);
9631 lod_comp->llc_stripe = NULL;
9632 OBD_FREE_PTR_ARRAY(lod_comp->llc_ost_indices,
9633 lod_comp->llc_stripes_allocated);
9634 lod_comp->llc_ost_indices = NULL;
9635 lod_comp->llc_stripes_allocated = 0;
9637 lod_free_comp_entries(lo);
9638 lo->ldo_comp_cached = 0;
9642 void lod_striping_free(const struct lu_env *env, struct lod_object *lo)
9644 mutex_lock(&lo->ldo_layout_mutex);
9645 lod_striping_free_nolock(env, lo);
9646 mutex_unlock(&lo->ldo_layout_mutex);
9650 * Implementation of lu_object_operations::loo_object_free.
9652 * \see lu_object_operations::loo_object_free() in the API description
9655 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
9657 struct lod_object *lo = lu2lod_obj(o);
9659 /* release all underlying object pinned */
9660 lod_striping_free(env, lo);
9662 /* lo doesn't contain a lu_object_header, so we don't need call_rcu */
9663 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
9667 * Implementation of lu_object_operations::loo_object_release.
9669 * \see lu_object_operations::loo_object_release() in the API description
9672 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
9674 /* XXX: shouldn't we release everything here in case if object
9675 * creation failed before? */
9679 * Implementation of lu_object_operations::loo_object_print.
9681 * \see lu_object_operations::loo_object_print() in the API description
9684 static int lod_object_print(const struct lu_env *env, void *cookie,
9685 lu_printer_t p, const struct lu_object *l)
9687 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
9689 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
9692 const struct lu_object_operations lod_lu_obj_ops = {
9693 .loo_object_init = lod_object_init,
9694 .loo_object_free = lod_object_free,
9695 .loo_object_release = lod_object_release,
9696 .loo_object_print = lod_object_print,