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
44 #include <obd_class.h>
45 #include <obd_support.h>
47 #include <lustre_fid.h>
48 #include <lustre_linkea.h>
49 #include <lustre_lmv.h>
50 #include <uapi/linux/lustre/lustre_param.h>
51 #include <lustre_swab.h>
52 #include <uapi/linux/lustre/lustre_ver.h>
53 #include <lprocfs_status.h>
54 #include <md_object.h>
56 #include "lod_internal.h"
58 static const char dot[] = ".";
59 static const char dotdot[] = "..";
62 * Implementation of dt_index_operations::dio_lookup
64 * Used with regular (non-striped) objects.
66 * \see dt_index_operations::dio_lookup() in the API description for details.
68 static int lod_lookup(const struct lu_env *env, struct dt_object *dt,
69 struct dt_rec *rec, const struct dt_key *key)
71 struct dt_object *next = dt_object_child(dt);
72 return next->do_index_ops->dio_lookup(env, next, rec, key);
76 * Implementation of dt_index_operations::dio_declare_insert.
78 * Used with regular (non-striped) objects.
80 * \see dt_index_operations::dio_declare_insert() in the API description
83 static int lod_declare_insert(const struct lu_env *env, struct dt_object *dt,
84 const struct dt_rec *rec,
85 const struct dt_key *key, struct thandle *th)
87 return lod_sub_declare_insert(env, dt_object_child(dt), rec, key, th);
91 * Implementation of dt_index_operations::dio_insert.
93 * Used with regular (non-striped) objects
95 * \see dt_index_operations::dio_insert() in the API description for details.
97 static int lod_insert(const struct lu_env *env, struct dt_object *dt,
98 const struct dt_rec *rec, const struct dt_key *key,
99 struct thandle *th, int ign)
101 return lod_sub_insert(env, dt_object_child(dt), rec, key, th, ign);
105 * Implementation of dt_index_operations::dio_declare_delete.
107 * Used with regular (non-striped) objects.
109 * \see dt_index_operations::dio_declare_delete() in the API description
112 static int lod_declare_delete(const struct lu_env *env, struct dt_object *dt,
113 const struct dt_key *key, struct thandle *th)
115 return lod_sub_declare_delete(env, dt_object_child(dt), key, th);
119 * Implementation of dt_index_operations::dio_delete.
121 * Used with regular (non-striped) objects.
123 * \see dt_index_operations::dio_delete() in the API description for details.
125 static int lod_delete(const struct lu_env *env, struct dt_object *dt,
126 const struct dt_key *key, struct thandle *th)
128 return lod_sub_delete(env, dt_object_child(dt), key, th);
132 * Implementation of dt_it_ops::init.
134 * Used with regular (non-striped) objects.
136 * \see dt_it_ops::init() in the API description for details.
138 static struct dt_it *lod_it_init(const struct lu_env *env,
139 struct dt_object *dt, __u32 attr)
141 struct dt_object *next = dt_object_child(dt);
142 struct lod_it *it = &lod_env_info(env)->lti_it;
143 struct dt_it *it_next;
145 it_next = next->do_index_ops->dio_it.init(env, next, attr);
149 /* currently we do not use more than one iterator per thread
150 * so we store it in thread info. if at some point we need
151 * more active iterators in a single thread, we can allocate
153 LASSERT(it->lit_obj == NULL);
155 it->lit_it = it_next;
158 return (struct dt_it *)it;
161 #define LOD_CHECK_IT(env, it) \
163 LASSERT((it)->lit_obj != NULL); \
164 LASSERT((it)->lit_it != NULL); \
168 * Implementation of dt_index_operations::dio_it.fini.
170 * Used with regular (non-striped) objects.
172 * \see dt_index_operations::dio_it.fini() in the API description for details.
174 static void lod_it_fini(const struct lu_env *env, struct dt_it *di)
176 struct lod_it *it = (struct lod_it *)di;
178 LOD_CHECK_IT(env, it);
179 it->lit_obj->do_index_ops->dio_it.fini(env, it->lit_it);
181 /* the iterator not in use any more */
187 * Implementation of dt_it_ops::get.
189 * Used with regular (non-striped) objects.
191 * \see dt_it_ops::get() in the API description for details.
193 static int lod_it_get(const struct lu_env *env, struct dt_it *di,
194 const struct dt_key *key)
196 const struct lod_it *it = (const struct lod_it *)di;
198 LOD_CHECK_IT(env, it);
199 return it->lit_obj->do_index_ops->dio_it.get(env, it->lit_it, key);
203 * Implementation of dt_it_ops::put.
205 * Used with regular (non-striped) objects.
207 * \see dt_it_ops::put() in the API description for details.
209 static void lod_it_put(const struct lu_env *env, struct dt_it *di)
211 struct lod_it *it = (struct lod_it *)di;
213 LOD_CHECK_IT(env, it);
214 return it->lit_obj->do_index_ops->dio_it.put(env, it->lit_it);
218 * Implementation of dt_it_ops::next.
220 * Used with regular (non-striped) objects
222 * \see dt_it_ops::next() in the API description for details.
224 static int lod_it_next(const struct lu_env *env, struct dt_it *di)
226 struct lod_it *it = (struct lod_it *)di;
228 LOD_CHECK_IT(env, it);
229 return it->lit_obj->do_index_ops->dio_it.next(env, it->lit_it);
233 * Implementation of dt_it_ops::key.
235 * Used with regular (non-striped) objects.
237 * \see dt_it_ops::key() in the API description for details.
239 static struct dt_key *lod_it_key(const struct lu_env *env,
240 const struct dt_it *di)
242 const struct lod_it *it = (const struct lod_it *)di;
244 LOD_CHECK_IT(env, it);
245 return it->lit_obj->do_index_ops->dio_it.key(env, it->lit_it);
249 * Implementation of dt_it_ops::key_size.
251 * Used with regular (non-striped) objects.
253 * \see dt_it_ops::key_size() in the API description for details.
255 static int lod_it_key_size(const struct lu_env *env, const struct dt_it *di)
257 struct lod_it *it = (struct lod_it *)di;
259 LOD_CHECK_IT(env, it);
260 return it->lit_obj->do_index_ops->dio_it.key_size(env, it->lit_it);
264 * Implementation of dt_it_ops::rec.
266 * Used with regular (non-striped) objects.
268 * \see dt_it_ops::rec() in the API description for details.
270 static int lod_it_rec(const struct lu_env *env, const struct dt_it *di,
271 struct dt_rec *rec, __u32 attr)
273 const struct lod_it *it = (const struct lod_it *)di;
275 LOD_CHECK_IT(env, it);
276 return it->lit_obj->do_index_ops->dio_it.rec(env, it->lit_it, rec,
281 * Implementation of dt_it_ops::rec_size.
283 * Used with regular (non-striped) objects.
285 * \see dt_it_ops::rec_size() in the API description for details.
287 static int lod_it_rec_size(const struct lu_env *env, const struct dt_it *di,
290 const struct lod_it *it = (const struct lod_it *)di;
292 LOD_CHECK_IT(env, it);
293 return it->lit_obj->do_index_ops->dio_it.rec_size(env, it->lit_it,
298 * Implementation of dt_it_ops::store.
300 * Used with regular (non-striped) objects.
302 * \see dt_it_ops::store() in the API description for details.
304 static __u64 lod_it_store(const struct lu_env *env, const struct dt_it *di)
306 const struct lod_it *it = (const struct lod_it *)di;
308 LOD_CHECK_IT(env, it);
309 return it->lit_obj->do_index_ops->dio_it.store(env, it->lit_it);
313 * Implementation of dt_it_ops::load.
315 * Used with regular (non-striped) objects.
317 * \see dt_it_ops::load() in the API description for details.
319 static int lod_it_load(const struct lu_env *env, const struct dt_it *di,
322 const struct lod_it *it = (const struct lod_it *)di;
324 LOD_CHECK_IT(env, it);
325 return it->lit_obj->do_index_ops->dio_it.load(env, it->lit_it, hash);
329 * Implementation of dt_it_ops::key_rec.
331 * Used with regular (non-striped) objects.
333 * \see dt_it_ops::rec() in the API description for details.
335 static int lod_it_key_rec(const struct lu_env *env, const struct dt_it *di,
338 const struct lod_it *it = (const struct lod_it *)di;
340 LOD_CHECK_IT(env, it);
341 return it->lit_obj->do_index_ops->dio_it.key_rec(env, it->lit_it,
345 static struct dt_index_operations lod_index_ops = {
346 .dio_lookup = lod_lookup,
347 .dio_declare_insert = lod_declare_insert,
348 .dio_insert = lod_insert,
349 .dio_declare_delete = lod_declare_delete,
350 .dio_delete = lod_delete,
358 .key_size = lod_it_key_size,
360 .rec_size = lod_it_rec_size,
361 .store = lod_it_store,
363 .key_rec = lod_it_key_rec,
368 * Implementation of dt_it_ops::init.
370 * Used with striped objects. Internally just initializes the iterator
371 * on the first stripe.
373 * \see dt_it_ops::init() in the API description for details.
375 static struct dt_it *lod_striped_it_init(const struct lu_env *env,
376 struct dt_object *dt, __u32 attr)
378 struct lod_object *lo = lod_dt_obj(dt);
379 struct dt_object *next;
380 struct lod_it *it = &lod_env_info(env)->lti_it;
381 struct dt_it *it_next;
384 LASSERT(lo->ldo_dir_stripe_count > 0);
385 next = lo->ldo_stripe[0];
386 LASSERT(next != NULL);
387 LASSERT(next->do_index_ops != NULL);
389 it_next = next->do_index_ops->dio_it.init(env, next, attr);
393 /* currently we do not use more than one iterator per thread
394 * so we store it in thread info. if at some point we need
395 * more active iterators in a single thread, we can allocate
397 LASSERT(it->lit_obj == NULL);
399 it->lit_stripe_index = 0;
401 it->lit_it = it_next;
404 return (struct dt_it *)it;
407 #define LOD_CHECK_STRIPED_IT(env, it, lo) \
409 LASSERT((it)->lit_obj != NULL); \
410 LASSERT((it)->lit_it != NULL); \
411 LASSERT((lo)->ldo_dir_stripe_count > 0); \
412 LASSERT((it)->lit_stripe_index < (lo)->ldo_dir_stripe_count); \
416 * Implementation of dt_it_ops::fini.
418 * Used with striped objects.
420 * \see dt_it_ops::fini() in the API description for details.
422 static void lod_striped_it_fini(const struct lu_env *env, struct dt_it *di)
424 struct lod_it *it = (struct lod_it *)di;
425 struct lod_object *lo = lod_dt_obj(it->lit_obj);
426 struct dt_object *next;
428 /* If lit_it == NULL, then it means the sub_it has been finished,
429 * which only happens in failure cases, see lod_striped_it_next() */
430 if (it->lit_it != NULL) {
431 LOD_CHECK_STRIPED_IT(env, it, lo);
433 next = lo->ldo_stripe[it->lit_stripe_index];
434 LASSERT(next != NULL);
435 LASSERT(next->do_index_ops != NULL);
437 next->do_index_ops->dio_it.fini(env, it->lit_it);
440 /* the iterator not in use any more */
443 it->lit_stripe_index = 0;
447 * Implementation of dt_it_ops::get.
449 * Right now it's not used widely, only to reset the iterator to the
450 * initial position. It should be possible to implement a full version
451 * which chooses a correct stripe to be able to position with any key.
453 * \see dt_it_ops::get() in the API description for details.
455 static int lod_striped_it_get(const struct lu_env *env, struct dt_it *di,
456 const struct dt_key *key)
458 const struct lod_it *it = (const struct lod_it *)di;
459 struct lod_object *lo = lod_dt_obj(it->lit_obj);
460 struct dt_object *next;
463 LOD_CHECK_STRIPED_IT(env, it, lo);
465 next = lo->ldo_stripe[it->lit_stripe_index];
466 LASSERT(next != NULL);
467 LASSERT(next->do_index_ops != NULL);
469 return next->do_index_ops->dio_it.get(env, it->lit_it, key);
473 * Implementation of dt_it_ops::put.
475 * Used with striped objects.
477 * \see dt_it_ops::put() in the API description for details.
479 static void lod_striped_it_put(const struct lu_env *env, struct dt_it *di)
481 struct lod_it *it = (struct lod_it *)di;
482 struct lod_object *lo = lod_dt_obj(it->lit_obj);
483 struct dt_object *next;
485 LOD_CHECK_STRIPED_IT(env, it, lo);
487 next = lo->ldo_stripe[it->lit_stripe_index];
488 LASSERT(next != NULL);
489 LASSERT(next->do_index_ops != NULL);
491 return next->do_index_ops->dio_it.put(env, it->lit_it);
495 * Implementation of dt_it_ops::next.
497 * Used with striped objects. When the end of the current stripe is
498 * reached, the method takes the next stripe's iterator.
500 * \see dt_it_ops::next() in the API description for details.
502 static int lod_striped_it_next(const struct lu_env *env, struct dt_it *di)
504 struct lod_it *it = (struct lod_it *)di;
505 struct lod_object *lo = lod_dt_obj(it->lit_obj);
506 struct dt_object *next;
507 struct dt_it *it_next;
511 LOD_CHECK_STRIPED_IT(env, it, lo);
513 next = lo->ldo_stripe[it->lit_stripe_index];
514 LASSERT(next != NULL);
515 LASSERT(next->do_index_ops != NULL);
517 rc = next->do_index_ops->dio_it.next(env, it->lit_it);
521 if (rc == 0 && it->lit_stripe_index == 0)
524 if (rc == 0 && it->lit_stripe_index > 0) {
525 struct lu_dirent *ent;
527 ent = (struct lu_dirent *)lod_env_info(env)->lti_key;
529 rc = next->do_index_ops->dio_it.rec(env, it->lit_it,
530 (struct dt_rec *)ent,
535 /* skip . and .. for slave stripe */
536 if ((strncmp(ent->lde_name, ".",
537 le16_to_cpu(ent->lde_namelen)) == 0 &&
538 le16_to_cpu(ent->lde_namelen) == 1) ||
539 (strncmp(ent->lde_name, "..",
540 le16_to_cpu(ent->lde_namelen)) == 0 &&
541 le16_to_cpu(ent->lde_namelen) == 2))
547 /* go to next stripe */
548 if (it->lit_stripe_index + 1 >= lo->ldo_dir_stripe_count)
551 it->lit_stripe_index++;
553 next->do_index_ops->dio_it.put(env, it->lit_it);
554 next->do_index_ops->dio_it.fini(env, it->lit_it);
557 next = lo->ldo_stripe[it->lit_stripe_index];
558 LASSERT(next != NULL);
559 rc = next->do_ops->do_index_try(env, next, &dt_directory_features);
563 LASSERT(next->do_index_ops != NULL);
565 it_next = next->do_index_ops->dio_it.init(env, next, it->lit_attr);
566 if (!IS_ERR(it_next)) {
567 it->lit_it = it_next;
570 rc = PTR_ERR(it_next);
577 * Implementation of dt_it_ops::key.
579 * Used with striped objects.
581 * \see dt_it_ops::key() in the API description for details.
583 static struct dt_key *lod_striped_it_key(const struct lu_env *env,
584 const struct dt_it *di)
586 const struct lod_it *it = (const struct lod_it *)di;
587 struct lod_object *lo = lod_dt_obj(it->lit_obj);
588 struct dt_object *next;
590 LOD_CHECK_STRIPED_IT(env, it, lo);
592 next = lo->ldo_stripe[it->lit_stripe_index];
593 LASSERT(next != NULL);
594 LASSERT(next->do_index_ops != NULL);
596 return next->do_index_ops->dio_it.key(env, it->lit_it);
600 * Implementation of dt_it_ops::key_size.
602 * Used with striped objects.
604 * \see dt_it_ops::size() in the API description for details.
606 static int lod_striped_it_key_size(const struct lu_env *env,
607 const struct dt_it *di)
609 struct lod_it *it = (struct lod_it *)di;
610 struct lod_object *lo = lod_dt_obj(it->lit_obj);
611 struct dt_object *next;
613 LOD_CHECK_STRIPED_IT(env, it, lo);
615 next = lo->ldo_stripe[it->lit_stripe_index];
616 LASSERT(next != NULL);
617 LASSERT(next->do_index_ops != NULL);
619 return next->do_index_ops->dio_it.key_size(env, it->lit_it);
623 * Implementation of dt_it_ops::rec.
625 * Used with striped objects.
627 * \see dt_it_ops::rec() in the API description for details.
629 static int lod_striped_it_rec(const struct lu_env *env, const struct dt_it *di,
630 struct dt_rec *rec, __u32 attr)
632 const struct lod_it *it = (const struct lod_it *)di;
633 struct lod_object *lo = lod_dt_obj(it->lit_obj);
634 struct dt_object *next;
636 LOD_CHECK_STRIPED_IT(env, it, lo);
638 next = lo->ldo_stripe[it->lit_stripe_index];
639 LASSERT(next != NULL);
640 LASSERT(next->do_index_ops != NULL);
642 return next->do_index_ops->dio_it.rec(env, it->lit_it, rec, attr);
646 * Implementation of dt_it_ops::rec_size.
648 * Used with striped objects.
650 * \see dt_it_ops::rec_size() in the API description for details.
652 static int lod_striped_it_rec_size(const struct lu_env *env,
653 const struct dt_it *di, __u32 attr)
655 struct lod_it *it = (struct lod_it *)di;
656 struct lod_object *lo = lod_dt_obj(it->lit_obj);
657 struct dt_object *next;
659 LOD_CHECK_STRIPED_IT(env, it, lo);
661 next = lo->ldo_stripe[it->lit_stripe_index];
662 LASSERT(next != NULL);
663 LASSERT(next->do_index_ops != NULL);
665 return next->do_index_ops->dio_it.rec_size(env, it->lit_it, attr);
669 * Implementation of dt_it_ops::store.
671 * Used with striped objects.
673 * \see dt_it_ops::store() in the API description for details.
675 static __u64 lod_striped_it_store(const struct lu_env *env,
676 const struct dt_it *di)
678 const struct lod_it *it = (const struct lod_it *)di;
679 struct lod_object *lo = lod_dt_obj(it->lit_obj);
680 struct dt_object *next;
682 LOD_CHECK_STRIPED_IT(env, it, lo);
684 next = lo->ldo_stripe[it->lit_stripe_index];
685 LASSERT(next != NULL);
686 LASSERT(next->do_index_ops != NULL);
688 return next->do_index_ops->dio_it.store(env, it->lit_it);
692 * Implementation of dt_it_ops::load.
694 * Used with striped objects.
696 * \see dt_it_ops::load() in the API description for details.
698 static int lod_striped_it_load(const struct lu_env *env,
699 const struct dt_it *di, __u64 hash)
701 const struct lod_it *it = (const struct lod_it *)di;
702 struct lod_object *lo = lod_dt_obj(it->lit_obj);
703 struct dt_object *next;
705 LOD_CHECK_STRIPED_IT(env, it, lo);
707 next = lo->ldo_stripe[it->lit_stripe_index];
708 LASSERT(next != NULL);
709 LASSERT(next->do_index_ops != NULL);
711 return next->do_index_ops->dio_it.load(env, it->lit_it, hash);
714 static struct dt_index_operations lod_striped_index_ops = {
715 .dio_lookup = lod_lookup,
716 .dio_declare_insert = lod_declare_insert,
717 .dio_insert = lod_insert,
718 .dio_declare_delete = lod_declare_delete,
719 .dio_delete = lod_delete,
721 .init = lod_striped_it_init,
722 .fini = lod_striped_it_fini,
723 .get = lod_striped_it_get,
724 .put = lod_striped_it_put,
725 .next = lod_striped_it_next,
726 .key = lod_striped_it_key,
727 .key_size = lod_striped_it_key_size,
728 .rec = lod_striped_it_rec,
729 .rec_size = lod_striped_it_rec_size,
730 .store = lod_striped_it_store,
731 .load = lod_striped_it_load,
736 * Append the FID for each shard of the striped directory after the
737 * given LMV EA header.
739 * To simplify striped directory and the consistency verification,
740 * we only store the LMV EA header on disk, for both master object
741 * and slave objects. When someone wants to know the whole LMV EA,
742 * such as client readdir(), we can build the entrie LMV EA on the
743 * MDT side (in RAM) via iterating the sub-directory entries that
744 * are contained in the master object of the stripe directory.
746 * For the master object of the striped directroy, the valid name
747 * for each shard is composed of the ${shard_FID}:${shard_idx}.
749 * There may be holes in the LMV EA if some shards' name entries
750 * are corrupted or lost.
752 * \param[in] env pointer to the thread context
753 * \param[in] lo pointer to the master object of the striped directory
754 * \param[in] buf pointer to the lu_buf which will hold the LMV EA
755 * \param[in] resize whether re-allocate the buffer if it is not big enough
757 * \retval positive size of the LMV EA
758 * \retval 0 for nothing to be loaded
759 * \retval negative error number on failure
761 int lod_load_lmv_shards(const struct lu_env *env, struct lod_object *lo,
762 struct lu_buf *buf, bool resize)
764 struct lu_dirent *ent =
765 (struct lu_dirent *)lod_env_info(env)->lti_key;
766 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
767 struct dt_object *obj = dt_object_child(&lo->ldo_obj);
768 struct lmv_mds_md_v1 *lmv1 = buf->lb_buf;
770 const struct dt_it_ops *iops;
772 __u32 magic = le32_to_cpu(lmv1->lmv_magic);
777 /* If it is not a striped directory, then load nothing. */
778 if (magic != LMV_MAGIC_V1)
781 /* If it is in migration (or failure), then load nothing. */
782 if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
785 stripes = le32_to_cpu(lmv1->lmv_stripe_count);
789 rc = lmv_mds_md_size(stripes, magic);
793 if (buf->lb_len < lmv1_size) {
802 lu_buf_alloc(buf, lmv1_size);
807 memcpy(buf->lb_buf, tbuf.lb_buf, tbuf.lb_len);
810 if (unlikely(!dt_try_as_dir(env, obj)))
813 memset(&lmv1->lmv_stripe_fids[0], 0, stripes * sizeof(struct lu_fid));
814 iops = &obj->do_index_ops->dio_it;
815 it = iops->init(env, obj, LUDA_64BITHASH);
819 rc = iops->load(env, it, 0);
821 rc = iops->next(env, it);
826 char name[FID_LEN + 2] = "";
831 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
837 fid_le_to_cpu(&fid, &ent->lde_fid);
838 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
839 if (ent->lde_name[0] == '.') {
840 if (ent->lde_namelen == 1)
843 if (ent->lde_namelen == 2 && ent->lde_name[1] == '.')
847 len = snprintf(name, sizeof(name),
848 DFID":", PFID(&ent->lde_fid));
849 /* The ent->lde_name is composed of ${FID}:${index} */
850 if (ent->lde_namelen < len + 1 ||
851 memcmp(ent->lde_name, name, len) != 0) {
852 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
853 "%s: invalid shard name %.*s with the FID "DFID
854 " for the striped directory "DFID", %s\n",
855 lod2obd(lod)->obd_name, ent->lde_namelen,
856 ent->lde_name, PFID(&fid),
857 PFID(lu_object_fid(&obj->do_lu)),
858 lod->lod_lmv_failout ? "failout" : "skip");
860 if (lod->lod_lmv_failout)
868 if (ent->lde_name[len] < '0' ||
869 ent->lde_name[len] > '9') {
870 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
871 "%s: invalid shard name %.*s with the "
872 "FID "DFID" for the striped directory "
874 lod2obd(lod)->obd_name, ent->lde_namelen,
875 ent->lde_name, PFID(&fid),
876 PFID(lu_object_fid(&obj->do_lu)),
877 lod->lod_lmv_failout ?
880 if (lod->lod_lmv_failout)
886 index = index * 10 + ent->lde_name[len++] - '0';
887 } while (len < ent->lde_namelen);
889 if (len == ent->lde_namelen) {
890 /* Out of LMV EA range. */
891 if (index >= stripes) {
892 CERROR("%s: the shard %.*s for the striped "
893 "directory "DFID" is out of the known "
894 "LMV EA range [0 - %u], failout\n",
895 lod2obd(lod)->obd_name, ent->lde_namelen,
897 PFID(lu_object_fid(&obj->do_lu)),
903 /* The slot has been occupied. */
904 if (!fid_is_zero(&lmv1->lmv_stripe_fids[index])) {
908 &lmv1->lmv_stripe_fids[index]);
909 CERROR("%s: both the shard "DFID" and "DFID
910 " for the striped directory "DFID
911 " claim the same LMV EA slot at the "
912 "index %d, failout\n",
913 lod2obd(lod)->obd_name,
914 PFID(&fid0), PFID(&fid),
915 PFID(lu_object_fid(&obj->do_lu)), index);
920 /* stored as LE mode */
921 lmv1->lmv_stripe_fids[index] = ent->lde_fid;
924 rc = iops->next(env, it);
931 RETURN(rc > 0 ? lmv_mds_md_size(stripes, magic) : rc);
935 * Implementation of dt_object_operations::do_index_try.
937 * \see dt_object_operations::do_index_try() in the API description for details.
939 static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
940 const struct dt_index_features *feat)
942 struct lod_object *lo = lod_dt_obj(dt);
943 struct dt_object *next = dt_object_child(dt);
947 LASSERT(next->do_ops);
948 LASSERT(next->do_ops->do_index_try);
950 rc = lod_load_striping_locked(env, lo);
954 rc = next->do_ops->do_index_try(env, next, feat);
958 if (lo->ldo_dir_stripe_count > 0) {
961 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
962 if (dt_object_exists(lo->ldo_stripe[i]) == 0)
964 rc = lo->ldo_stripe[i]->do_ops->do_index_try(env,
965 lo->ldo_stripe[i], feat);
969 dt->do_index_ops = &lod_striped_index_ops;
971 dt->do_index_ops = &lod_index_ops;
978 * Implementation of dt_object_operations::do_read_lock.
980 * \see dt_object_operations::do_read_lock() in the API description for details.
982 static void lod_read_lock(const struct lu_env *env, struct dt_object *dt,
985 dt_read_lock(env, dt_object_child(dt), role);
989 * Implementation of dt_object_operations::do_write_lock.
991 * \see dt_object_operations::do_write_lock() in the API description for
994 static void lod_write_lock(const struct lu_env *env, struct dt_object *dt,
997 dt_write_lock(env, dt_object_child(dt), role);
1001 * Implementation of dt_object_operations::do_read_unlock.
1003 * \see dt_object_operations::do_read_unlock() in the API description for
1006 static void lod_read_unlock(const struct lu_env *env, struct dt_object *dt)
1008 dt_read_unlock(env, dt_object_child(dt));
1012 * Implementation of dt_object_operations::do_write_unlock.
1014 * \see dt_object_operations::do_write_unlock() in the API description for
1017 static void lod_write_unlock(const struct lu_env *env, struct dt_object *dt)
1019 dt_write_unlock(env, dt_object_child(dt));
1023 * Implementation of dt_object_operations::do_write_locked.
1025 * \see dt_object_operations::do_write_locked() in the API description for
1028 static int lod_write_locked(const struct lu_env *env, struct dt_object *dt)
1030 return dt_write_locked(env, dt_object_child(dt));
1034 * Implementation of dt_object_operations::do_attr_get.
1036 * \see dt_object_operations::do_attr_get() in the API description for details.
1038 static int lod_attr_get(const struct lu_env *env,
1039 struct dt_object *dt,
1040 struct lu_attr *attr)
1042 /* Note: for striped directory, client will merge attributes
1043 * from all of the sub-stripes see lmv_merge_attr(), and there
1044 * no MDD logic depend on directory nlink/size/time, so we can
1045 * always use master inode nlink and size for now. */
1046 return dt_attr_get(env, dt_object_child(dt), attr);
1049 int lod_obj_for_each_stripe(const struct lu_env *env, struct lod_object *lo,
1051 struct lod_obj_stripe_cb_data *data)
1053 struct lod_layout_component *lod_comp;
1057 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1058 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1059 lod_comp = &lo->ldo_comp_entries[i];
1061 if (lod_comp->llc_stripe == NULL)
1064 /* has stripe but not inited yet, this component has been
1065 * declared to be created, but hasn't created yet.
1067 if (!lod_comp_inited(lod_comp))
1070 if (data->locd_comp_skip_cb &&
1071 data->locd_comp_skip_cb(env, lo, i, data))
1074 LASSERT(lod_comp->llc_stripe_count > 0);
1075 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
1076 struct dt_object *dt = lod_comp->llc_stripe[j];
1080 rc = data->locd_stripe_cb(env, lo, dt, th, i, j, data);
1088 static bool lod_obj_attr_set_comp_skip_cb(const struct lu_env *env,
1089 struct lod_object *lo, int comp_idx,
1090 struct lod_obj_stripe_cb_data *data)
1092 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[comp_idx];
1093 bool skipped = false;
1095 if (!(data->locd_attr->la_valid & LA_LAYOUT_VERSION))
1098 switch (lo->ldo_flr_state) {
1099 case LCM_FL_WRITE_PENDING: {
1102 /* skip stale components */
1103 if (lod_comp->llc_flags & LCME_FL_STALE) {
1108 /* skip valid and overlapping components, therefore any
1109 * attempts to write overlapped components will never succeed
1110 * because client will get EINPROGRESS. */
1111 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1115 if (lo->ldo_comp_entries[i].llc_flags & LCME_FL_STALE)
1118 if (lu_extent_is_overlapped(&lod_comp->llc_extent,
1119 &lo->ldo_comp_entries[i].llc_extent)) {
1127 LASSERTF(0, "impossible: %d\n", lo->ldo_flr_state);
1128 case LCM_FL_SYNC_PENDING:
1132 CDEBUG(D_LAYOUT, DFID": %s to set component %x to version: %u\n",
1133 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
1134 skipped ? "skipped" : "chose", lod_comp->llc_id,
1135 data->locd_attr->la_layout_version);
1141 lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
1142 struct dt_object *dt, struct thandle *th,
1143 int comp_idx, int stripe_idx,
1144 struct lod_obj_stripe_cb_data *data)
1146 if (data->locd_declare)
1147 return lod_sub_declare_attr_set(env, dt, data->locd_attr, th);
1149 if (data->locd_attr->la_valid & LA_LAYOUT_VERSION) {
1150 CDEBUG(D_LAYOUT, DFID": set layout version: %u, comp_idx: %d\n",
1151 PFID(lu_object_fid(&dt->do_lu)),
1152 data->locd_attr->la_layout_version, comp_idx);
1155 return lod_sub_attr_set(env, dt, data->locd_attr, th);
1159 * Implementation of dt_object_operations::do_declare_attr_set.
1161 * If the object is striped, then apply the changes to all the stripes.
1163 * \see dt_object_operations::do_declare_attr_set() in the API description
1166 static int lod_declare_attr_set(const struct lu_env *env,
1167 struct dt_object *dt,
1168 const struct lu_attr *attr,
1171 struct dt_object *next = dt_object_child(dt);
1172 struct lod_object *lo = lod_dt_obj(dt);
1177 * declare setattr on the local object
1179 rc = lod_sub_declare_attr_set(env, next, attr, th);
1183 /* osp_declare_attr_set() ignores all attributes other than
1184 * UID, GID, PROJID, and size, and osp_attr_set() ignores all
1185 * but UID, GID and PROJID. Declaration of size attr setting
1186 * happens through lod_declare_init_size(), and not through
1187 * this function. Therefore we need not load striping unless
1188 * ownership is changing. This should save memory and (we hope)
1189 * speed up rename().
1191 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1192 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1195 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1198 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID | LA_MODE |
1199 LA_ATIME | LA_MTIME | LA_CTIME |
1204 * load striping information, notice we don't do this when object
1205 * is being initialized as we don't need this information till
1206 * few specific cases like destroy, chown
1208 rc = lod_load_striping(env, lo);
1212 if (!lod_obj_is_striped(dt))
1216 * if object is striped declare changes on the stripes
1218 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1219 LASSERT(lo->ldo_stripe);
1220 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1221 if (lo->ldo_stripe[i] == NULL)
1223 rc = lod_sub_declare_attr_set(env, lo->ldo_stripe[i],
1229 struct lod_obj_stripe_cb_data data = { { 0 } };
1231 data.locd_attr = attr;
1232 data.locd_declare = true;
1233 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1234 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1240 if (!dt_object_exists(next) || dt_object_remote(next) ||
1241 !S_ISREG(attr->la_mode))
1244 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1245 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
1249 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) ||
1250 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1251 struct lod_thread_info *info = lod_env_info(env);
1252 struct lu_buf *buf = &info->lti_buf;
1254 buf->lb_buf = info->lti_ea_store;
1255 buf->lb_len = info->lti_ea_store_size;
1256 rc = lod_sub_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1257 LU_XATTR_REPLACE, th);
1264 * Implementation of dt_object_operations::do_attr_set.
1266 * If the object is striped, then apply the changes to all or subset of
1267 * the stripes depending on the object type and specific attributes.
1269 * \see dt_object_operations::do_attr_set() in the API description for details.
1271 static int lod_attr_set(const struct lu_env *env,
1272 struct dt_object *dt,
1273 const struct lu_attr *attr,
1276 struct dt_object *next = dt_object_child(dt);
1277 struct lod_object *lo = lod_dt_obj(dt);
1282 * apply changes to the local object
1284 rc = lod_sub_attr_set(env, next, attr, th);
1288 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1289 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1292 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1295 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE | LA_PROJID |
1296 LA_ATIME | LA_MTIME | LA_CTIME |
1301 /* FIXME: a tricky case in the code path of mdd_layout_change():
1302 * the in-memory striping information has been freed in lod_xattr_set()
1303 * due to layout change. It has to load stripe here again. It only
1304 * changes flags of layout so declare_attr_set() is still accurate */
1305 rc = lod_load_striping_locked(env, lo);
1309 if (!lod_obj_is_striped(dt))
1313 * if object is striped, apply changes to all the stripes
1315 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1316 LASSERT(lo->ldo_stripe);
1317 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1318 if (unlikely(lo->ldo_stripe[i] == NULL))
1321 if ((dt_object_exists(lo->ldo_stripe[i]) == 0))
1324 rc = lod_sub_attr_set(env, lo->ldo_stripe[i], attr, th);
1329 struct lod_obj_stripe_cb_data data = { { 0 } };
1331 data.locd_attr = attr;
1332 data.locd_declare = false;
1333 data.locd_comp_skip_cb = lod_obj_attr_set_comp_skip_cb;
1334 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1335 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1341 if (!dt_object_exists(next) || dt_object_remote(next) ||
1342 !S_ISREG(attr->la_mode))
1345 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1346 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
1350 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE)) {
1351 struct lod_thread_info *info = lod_env_info(env);
1352 struct lu_buf *buf = &info->lti_buf;
1353 struct ost_id *oi = &info->lti_ostid;
1354 struct lu_fid *fid = &info->lti_fid;
1355 struct lov_mds_md_v1 *lmm;
1356 struct lov_ost_data_v1 *objs;
1359 rc = lod_get_lov_ea(env, lo);
1363 buf->lb_buf = info->lti_ea_store;
1364 buf->lb_len = info->lti_ea_store_size;
1365 lmm = info->lti_ea_store;
1366 magic = le32_to_cpu(lmm->lmm_magic);
1367 if (magic == LOV_MAGIC_COMP_V1) {
1368 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1369 struct lov_comp_md_entry_v1 *lcme =
1370 &lcm->lcm_entries[0];
1372 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
1373 magic = le32_to_cpu(lmm->lmm_magic);
1376 if (magic == LOV_MAGIC_V1)
1377 objs = &(lmm->lmm_objects[0]);
1379 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1380 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1381 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1383 fid_to_ostid(fid, oi);
1384 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1386 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1387 LU_XATTR_REPLACE, th);
1388 } else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1389 struct lod_thread_info *info = lod_env_info(env);
1390 struct lu_buf *buf = &info->lti_buf;
1391 struct lov_comp_md_v1 *lcm;
1392 struct lov_comp_md_entry_v1 *lcme;
1394 rc = lod_get_lov_ea(env, lo);
1398 buf->lb_buf = info->lti_ea_store;
1399 buf->lb_len = info->lti_ea_store_size;
1401 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
1404 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1405 lcme = &lcm->lcm_entries[0];
1406 le64_add_cpu(&lcme->lcme_extent.e_start, 1);
1407 le64_add_cpu(&lcme->lcme_extent.e_end, -1);
1409 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1410 LU_XATTR_REPLACE, th);
1417 * Implementation of dt_object_operations::do_xattr_get.
1419 * If LOV EA is requested from the root object and it's not
1420 * found, then return default striping for the filesystem.
1422 * \see dt_object_operations::do_xattr_get() in the API description for details.
1424 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1425 struct lu_buf *buf, const char *name)
1427 struct lod_thread_info *info = lod_env_info(env);
1428 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1433 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1434 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1435 struct lmv_mds_md_v1 *lmv1;
1438 if (rc > (typeof(rc))sizeof(*lmv1))
1441 if (rc < (typeof(rc))sizeof(*lmv1))
1442 RETURN(rc = rc > 0 ? -EINVAL : rc);
1444 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1445 CLASSERT(sizeof(*lmv1) <= sizeof(info->lti_key));
1447 info->lti_buf.lb_buf = info->lti_key;
1448 info->lti_buf.lb_len = sizeof(*lmv1);
1449 rc = dt_xattr_get(env, dt_object_child(dt),
1450 &info->lti_buf, name);
1451 if (unlikely(rc != sizeof(*lmv1)))
1452 RETURN(rc = rc > 0 ? -EINVAL : rc);
1454 lmv1 = info->lti_buf.lb_buf;
1455 /* The on-disk LMV EA only contains header, but the
1456 * returned LMV EA size should contain the space for
1457 * the FIDs of all shards of the striped directory. */
1458 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1459 rc = lmv_mds_md_size(
1460 le32_to_cpu(lmv1->lmv_stripe_count),
1463 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1467 RETURN(rc = rc1 != 0 ? rc1 : rc);
1470 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1474 * XXX: Only used by lfsck
1476 * lod returns default striping on the real root of the device
1477 * this is like the root stores default striping for the whole
1478 * filesystem. historically we've been using a different approach
1479 * and store it in the config.
1481 dt_root_get(env, dev->lod_child, &info->lti_fid);
1482 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1484 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1485 struct lov_user_md *lum = buf->lb_buf;
1486 struct lov_desc *desc = &dev->lod_desc;
1488 if (buf->lb_buf == NULL) {
1490 } else if (buf->lb_len >= sizeof(*lum)) {
1491 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1492 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1493 lmm_oi_set_id(&lum->lmm_oi, 0);
1494 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1495 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1496 lum->lmm_stripe_size = cpu_to_le32(
1497 desc->ld_default_stripe_size);
1498 lum->lmm_stripe_count = cpu_to_le16(
1499 desc->ld_default_stripe_count);
1500 lum->lmm_stripe_offset = cpu_to_le16(
1501 desc->ld_default_stripe_offset);
1514 * Checks that the magic of the stripe is sane.
1516 * \param[in] lod lod device
1517 * \param[in] lum a buffer storing LMV EA to verify
1519 * \retval 0 if the EA is sane
1520 * \retval negative otherwise
1522 static int lod_verify_md_striping(struct lod_device *lod,
1523 const struct lmv_user_md_v1 *lum)
1525 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1526 CERROR("%s: invalid lmv_user_md: magic = %x, "
1527 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1528 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1529 (int)le32_to_cpu(lum->lum_stripe_offset),
1530 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1538 * Initialize LMV EA for a slave.
1540 * Initialize slave's LMV EA from the master's LMV EA.
1542 * \param[in] master_lmv a buffer containing master's EA
1543 * \param[out] slave_lmv a buffer where slave's EA will be stored
1546 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1547 const struct lmv_mds_md_v1 *master_lmv)
1549 *slave_lmv = *master_lmv;
1550 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1556 * Generate LMV EA from the object passed as \a dt. The object must have
1557 * the stripes created and initialized.
1559 * \param[in] env execution environment
1560 * \param[in] dt object
1561 * \param[out] lmv_buf buffer storing generated LMV EA
1563 * \retval 0 on success
1564 * \retval negative if failed
1566 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1567 struct lu_buf *lmv_buf)
1569 struct lod_thread_info *info = lod_env_info(env);
1570 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1571 struct lod_object *lo = lod_dt_obj(dt);
1572 struct lmv_mds_md_v1 *lmm1;
1574 int type = LU_SEQ_RANGE_ANY;
1579 LASSERT(lo->ldo_dir_striped != 0);
1580 LASSERT(lo->ldo_dir_stripe_count > 0);
1581 stripe_count = lo->ldo_dir_stripe_count;
1582 /* Only store the LMV EA heahder on the disk. */
1583 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1584 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1588 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1591 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1592 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1593 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1594 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1595 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1600 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1601 lmv_buf->lb_buf = info->lti_ea_store;
1602 lmv_buf->lb_len = sizeof(*lmm1);
1608 * Create in-core represenation for a striped directory.
1610 * Parse the buffer containing LMV EA and instantiate LU objects
1611 * representing the stripe objects. The pointers to the objects are
1612 * stored in ldo_stripe field of \a lo. This function is used when
1613 * we need to access an already created object (i.e. load from a disk).
1615 * \param[in] env execution environment
1616 * \param[in] lo lod object
1617 * \param[in] buf buffer containing LMV EA
1619 * \retval 0 on success
1620 * \retval negative if failed
1622 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1623 const struct lu_buf *buf)
1625 struct lod_thread_info *info = lod_env_info(env);
1626 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1627 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1628 struct dt_object **stripe;
1629 union lmv_mds_md *lmm = buf->lb_buf;
1630 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1631 struct lu_fid *fid = &info->lti_fid;
1636 if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
1639 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1640 lo->ldo_dir_slave_stripe = 1;
1644 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1647 if (le32_to_cpu(lmv1->lmv_stripe_count) < 1)
1650 LASSERT(lo->ldo_stripe == NULL);
1651 OBD_ALLOC(stripe, sizeof(stripe[0]) *
1652 (le32_to_cpu(lmv1->lmv_stripe_count)));
1656 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1657 struct dt_device *tgt_dt;
1658 struct dt_object *dto;
1659 int type = LU_SEQ_RANGE_ANY;
1662 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1663 if (!fid_is_sane(fid))
1664 GOTO(out, rc = -ESTALE);
1666 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1670 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1671 tgt_dt = lod->lod_child;
1673 struct lod_tgt_desc *tgt;
1675 tgt = LTD_TGT(ltd, idx);
1677 GOTO(out, rc = -ESTALE);
1678 tgt_dt = tgt->ltd_tgt;
1681 dto = dt_locate_at(env, tgt_dt, fid,
1682 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1685 GOTO(out, rc = PTR_ERR(dto));
1690 lo->ldo_stripe = stripe;
1691 lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1692 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1694 lod_object_free_striping(env, lo);
1700 * Declare create a striped directory.
1702 * Declare creating a striped directory with a given stripe pattern on the
1703 * specified MDTs. A striped directory is represented as a regular directory
1704 * - an index listing all the stripes. The stripes point back to the master
1705 * object with ".." and LinkEA. The master object gets LMV EA which
1706 * identifies it as a striped directory. The function allocates FIDs
1709 * \param[in] env execution environment
1710 * \param[in] dt object
1711 * \param[in] attr attributes to initialize the objects with
1712 * \param[in] dof type of objects to be created
1713 * \param[in] th transaction handle
1715 * \retval 0 on success
1716 * \retval negative if failed
1718 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1719 struct dt_object *dt,
1720 struct lu_attr *attr,
1721 struct dt_object_format *dof,
1724 struct lod_thread_info *info = lod_env_info(env);
1725 struct lu_buf lmv_buf;
1726 struct lu_buf slave_lmv_buf;
1727 struct lmv_mds_md_v1 *lmm;
1728 struct lmv_mds_md_v1 *slave_lmm = NULL;
1729 struct dt_insert_rec *rec = &info->lti_dt_rec;
1730 struct lod_object *lo = lod_dt_obj(dt);
1735 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1738 lmm = lmv_buf.lb_buf;
1740 OBD_ALLOC_PTR(slave_lmm);
1741 if (slave_lmm == NULL)
1742 GOTO(out, rc = -ENOMEM);
1744 lod_prep_slave_lmv_md(slave_lmm, lmm);
1745 slave_lmv_buf.lb_buf = slave_lmm;
1746 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1748 if (!dt_try_as_dir(env, dt_object_child(dt)))
1749 GOTO(out, rc = -EINVAL);
1751 rec->rec_type = S_IFDIR;
1752 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1753 struct dt_object *dto = lo->ldo_stripe[i];
1754 char *stripe_name = info->lti_key;
1755 struct lu_name *sname;
1756 struct linkea_data ldata = { NULL };
1757 struct lu_buf linkea_buf;
1759 rc = lod_sub_declare_create(env, dto, attr, NULL, dof, th);
1763 if (!dt_try_as_dir(env, dto))
1764 GOTO(out, rc = -EINVAL);
1766 rc = lod_sub_declare_ref_add(env, dto, th);
1770 rec->rec_fid = lu_object_fid(&dto->do_lu);
1771 rc = lod_sub_declare_insert(env, dto,
1772 (const struct dt_rec *)rec,
1773 (const struct dt_key *)dot, th);
1777 /* master stripe FID will be put to .. */
1778 rec->rec_fid = lu_object_fid(&dt->do_lu);
1779 rc = lod_sub_declare_insert(env, dto,
1780 (const struct dt_rec *)rec,
1781 (const struct dt_key *)dotdot, th);
1785 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1786 cfs_fail_val != i) {
1787 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1789 slave_lmm->lmv_master_mdt_index =
1792 slave_lmm->lmv_master_mdt_index =
1794 rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
1795 XATTR_NAME_LMV, 0, th);
1800 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1802 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1803 PFID(lu_object_fid(&dto->do_lu)), i + 1);
1805 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1806 PFID(lu_object_fid(&dto->do_lu)), i);
1808 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
1809 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1810 sname, lu_object_fid(&dt->do_lu));
1814 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1815 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1816 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
1817 XATTR_NAME_LINK, 0, th);
1821 rec->rec_fid = lu_object_fid(&dto->do_lu);
1822 rc = lod_sub_declare_insert(env, dt_object_child(dt),
1823 (const struct dt_rec *)rec,
1824 (const struct dt_key *)stripe_name,
1829 rc = lod_sub_declare_ref_add(env, dt_object_child(dt), th);
1834 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
1835 &lmv_buf, XATTR_NAME_LMV, 0, th);
1839 if (slave_lmm != NULL)
1840 OBD_FREE_PTR(slave_lmm);
1845 static int lod_prep_md_striped_create(const struct lu_env *env,
1846 struct dt_object *dt,
1847 struct lu_attr *attr,
1848 const struct lmv_user_md_v1 *lum,
1849 struct dt_object_format *dof,
1852 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1853 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1854 struct lod_object *lo = lod_dt_obj(dt);
1855 struct dt_object **stripe;
1862 bool is_specific = false;
1865 /* The lum has been verifed in lod_verify_md_striping */
1866 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1867 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
1868 LASSERT(le32_to_cpu(lum->lum_stripe_count) > 0);
1870 stripe_count = le32_to_cpu(lum->lum_stripe_count);
1872 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
1873 if (idx_array == NULL)
1876 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_count);
1878 GOTO(out_free, rc = -ENOMEM);
1880 /* Start index must be the master MDT */
1881 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1882 idx_array[0] = master_index;
1883 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
1885 for (i = 1; i < stripe_count; i++)
1886 idx_array[i] = le32_to_cpu(lum->lum_objects[i].lum_mds);
1889 for (i = 0; i < stripe_count; i++) {
1890 struct lod_tgt_desc *tgt = NULL;
1891 struct dt_object *dto;
1892 struct lu_fid fid = { 0 };
1894 struct lu_object_conf conf = { 0 };
1895 struct dt_device *tgt_dt = NULL;
1897 /* Try to find next avaible target */
1899 for (j = 0; j < lod->lod_remote_mdt_count;
1900 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
1901 bool already_allocated = false;
1904 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
1905 idx, lod->lod_remote_mdt_count + 1, i);
1907 if (likely(!is_specific &&
1908 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
1909 /* check whether the idx already exists
1910 * in current allocated array */
1911 for (k = 0; k < i; k++) {
1912 if (idx_array[k] == idx) {
1913 already_allocated = true;
1918 if (already_allocated)
1922 /* Sigh, this index is not in the bitmap, let's check
1923 * next available target */
1924 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) &&
1925 idx != master_index)
1928 if (idx == master_index) {
1929 /* Allocate the FID locally */
1930 rc = obd_fid_alloc(env, lod->lod_child_exp,
1934 tgt_dt = lod->lod_child;
1938 /* check the status of the OSP */
1939 tgt = LTD_TGT(ltd, idx);
1943 tgt_dt = tgt->ltd_tgt;
1944 rc = dt_statfs(env, tgt_dt, NULL);
1946 /* this OSP doesn't feel well */
1951 rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
1960 /* Can not allocate more stripes */
1961 if (j == lod->lod_remote_mdt_count) {
1962 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
1963 lod2obd(lod)->obd_name, stripe_count, i);
1967 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
1968 idx, i, PFID(&fid));
1970 /* Set the start index for next stripe allocation */
1971 if (!is_specific && i < stripe_count - 1)
1972 idx_array[i + 1] = (idx + 1) %
1973 (lod->lod_remote_mdt_count + 1);
1974 /* tgt_dt and fid must be ready after search avaible OSP
1975 * in the above loop */
1976 LASSERT(tgt_dt != NULL);
1977 LASSERT(fid_is_sane(&fid));
1978 conf.loc_flags = LOC_F_NEW;
1979 dto = dt_locate_at(env, tgt_dt, &fid,
1980 dt->do_lu.lo_dev->ld_site->ls_top_dev,
1983 GOTO(out_put, rc = PTR_ERR(dto));
1987 lo->ldo_dir_stripe_loaded = 1;
1988 lo->ldo_dir_striped = 1;
1989 lo->ldo_stripe = stripe;
1990 lo->ldo_dir_stripe_count = i;
1991 lo->ldo_dir_stripes_allocated = stripe_count;
1993 if (lo->ldo_dir_stripe_count == 0)
1994 GOTO(out_put, rc = -ENOSPC);
1996 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2002 for (i = 0; i < stripe_count; i++)
2003 if (stripe[i] != NULL)
2004 dt_object_put(env, stripe[i]);
2005 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_count);
2006 lo->ldo_dir_stripe_count = 0;
2007 lo->ldo_dir_stripes_allocated = 0;
2008 lo->ldo_stripe = NULL;
2012 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
2018 * Declare create striped md object.
2020 * The function declares intention to create a striped directory. This is a
2021 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2022 * is to verify pattern \a lum_buf is good. Check that function for the details.
2024 * \param[in] env execution environment
2025 * \param[in] dt object
2026 * \param[in] attr attributes to initialize the objects with
2027 * \param[in] lum_buf a pattern specifying the number of stripes and
2029 * \param[in] dof type of objects to be created
2030 * \param[in] th transaction handle
2032 * \retval 0 on success
2033 * \retval negative if failed
2036 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2037 struct dt_object *dt,
2038 struct lu_attr *attr,
2039 const struct lu_buf *lum_buf,
2040 struct dt_object_format *dof,
2043 struct lod_object *lo = lod_dt_obj(dt);
2044 struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
2048 LASSERT(lum != NULL);
2050 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
2051 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
2052 (int)le32_to_cpu(lum->lum_stripe_offset));
2054 if (lo->ldo_dir_stripe_count == 0)
2057 /* prepare dir striped objects */
2058 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2060 /* failed to create striping, let's reset
2061 * config so that others don't get confused */
2062 lod_object_free_striping(env, lo);
2070 * Implementation of dt_object_operations::do_declare_xattr_set.
2072 * Used with regular (non-striped) objects. Basically it
2073 * initializes the striping information and applies the
2074 * change to all the stripes.
2076 * \see dt_object_operations::do_declare_xattr_set() in the API description
2079 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2080 struct dt_object *dt,
2081 const struct lu_buf *buf,
2082 const char *name, int fl,
2085 struct dt_object *next = dt_object_child(dt);
2086 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2087 struct lod_object *lo = lod_dt_obj(dt);
2092 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2093 struct lmv_user_md_v1 *lum;
2095 LASSERT(buf != NULL && buf->lb_buf != NULL);
2097 rc = lod_verify_md_striping(d, lum);
2100 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2101 rc = lod_verify_striping(d, lo, buf, false);
2106 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2110 /* Note: Do not set LinkEA on sub-stripes, otherwise
2111 * it will confuse the fid2path process(see mdt_path_current()).
2112 * The linkEA between master and sub-stripes is set in
2113 * lod_xattr_set_lmv(). */
2114 if (strcmp(name, XATTR_NAME_LINK) == 0)
2117 /* set xattr to each stripes, if needed */
2118 rc = lod_load_striping(env, lo);
2122 if (lo->ldo_dir_stripe_count == 0)
2125 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2126 LASSERT(lo->ldo_stripe[i]);
2128 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2138 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2139 struct lod_object *lo,
2140 struct dt_object *dt, struct thandle *th,
2141 int comp_idx, int stripe_idx,
2142 struct lod_obj_stripe_cb_data *data)
2144 struct lod_thread_info *info = lod_env_info(env);
2145 struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
2146 struct filter_fid *ff = &info->lti_ff;
2147 struct lu_buf *buf = &info->lti_buf;
2151 buf->lb_len = sizeof(*ff);
2152 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2159 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
2160 if (lu_fid_eq(lu_object_fid(&lo->ldo_obj.do_lu), &ff->ff_parent) &&
2161 ff->ff_layout.ol_comp_id == comp->llc_id)
2164 /* rewrite filter_fid */
2165 memset(ff, 0, sizeof(*ff));
2166 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2167 ff->ff_parent.f_ver = stripe_idx;
2168 ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
2169 ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
2170 ff->ff_layout.ol_comp_id = comp->llc_id;
2171 ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
2172 ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
2173 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
2175 if (data->locd_declare)
2176 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2177 LU_XATTR_REPLACE, th);
2179 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2180 LU_XATTR_REPLACE, th);
2186 * Reset parent FID on OST object
2188 * Replace parent FID with @dt object FID, which is only called during migration
2189 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2190 * the FID is changed.
2192 * \param[in] env execution environment
2193 * \param[in] dt dt_object whose stripes's parent FID will be reset
2194 * \parem[in] th thandle
2195 * \param[in] declare if it is declare
2197 * \retval 0 if reset succeeds
2198 * \retval negative errno if reset fails
2200 static int lod_replace_parent_fid(const struct lu_env *env,
2201 struct dt_object *dt,
2202 struct thandle *th, bool declare)
2204 struct lod_object *lo = lod_dt_obj(dt);
2205 struct lod_thread_info *info = lod_env_info(env);
2206 struct lu_buf *buf = &info->lti_buf;
2207 struct filter_fid *ff;
2208 struct lod_obj_stripe_cb_data data = { { 0 } };
2212 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2214 /* set xattr to each stripes, if needed */
2215 rc = lod_load_striping(env, lo);
2219 if (!lod_obj_is_striped(dt))
2222 if (info->lti_ea_store_size < sizeof(*ff)) {
2223 rc = lod_ea_store_resize(info, sizeof(*ff));
2228 buf->lb_buf = info->lti_ea_store;
2229 buf->lb_len = info->lti_ea_store_size;
2231 data.locd_declare = declare;
2232 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2233 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2238 inline __u16 lod_comp_entry_stripe_count(struct lod_object *lo,
2239 struct lod_layout_component *entry,
2242 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2246 else if (lod_comp_inited(entry))
2247 return entry->llc_stripe_count;
2248 else if ((__u16)-1 == entry->llc_stripe_count)
2249 return lod->lod_desc.ld_tgt_count;
2251 return lod_get_stripe_count(lod, lo, entry->llc_stripe_count);
2254 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2256 int magic, size = 0, i;
2257 struct lod_layout_component *comp_entries;
2262 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2263 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2265 lo->ldo_def_striping->lds_def_striping_is_composite;
2267 comp_cnt = lo->ldo_comp_cnt;
2268 comp_entries = lo->ldo_comp_entries;
2269 is_composite = lo->ldo_is_composite;
2273 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2275 size = sizeof(struct lov_comp_md_v1) +
2276 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2277 LASSERT(size % sizeof(__u64) == 0);
2280 for (i = 0; i < comp_cnt; i++) {
2283 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2284 stripe_count = lod_comp_entry_stripe_count(lo, &comp_entries[i],
2286 if (!is_dir && is_composite)
2287 lod_comp_shrink_stripe_count(&comp_entries[i],
2290 size += lov_user_md_size(stripe_count, magic);
2291 LASSERT(size % sizeof(__u64) == 0);
2297 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2298 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2301 * \param[in] env execution environment
2302 * \param[in] dt dt_object to add components on
2303 * \param[in] buf buffer contains components to be added
2304 * \parem[in] th thandle
2306 * \retval 0 on success
2307 * \retval negative errno on failure
2309 static int lod_declare_layout_add(const struct lu_env *env,
2310 struct dt_object *dt,
2311 const struct lu_buf *buf,
2314 struct lod_thread_info *info = lod_env_info(env);
2315 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2316 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2317 struct dt_object *next = dt_object_child(dt);
2318 struct lov_desc *desc = &d->lod_desc;
2319 struct lod_object *lo = lod_dt_obj(dt);
2320 struct lov_user_md_v3 *v3;
2321 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2323 int i, rc, array_cnt, old_array_cnt;
2326 LASSERT(lo->ldo_is_composite);
2328 if (lo->ldo_flr_state != LCM_FL_NOT_FLR)
2331 rc = lod_verify_striping(d, lo, buf, false);
2335 magic = comp_v1->lcm_magic;
2336 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2337 lustre_swab_lov_comp_md_v1(comp_v1);
2338 magic = comp_v1->lcm_magic;
2341 if (magic != LOV_USER_MAGIC_COMP_V1)
2344 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2345 OBD_ALLOC(comp_array, sizeof(*comp_array) * array_cnt);
2346 if (comp_array == NULL)
2349 memcpy(comp_array, lo->ldo_comp_entries,
2350 sizeof(*comp_array) * lo->ldo_comp_cnt);
2352 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2353 struct lov_user_md_v1 *v1;
2354 struct lu_extent *ext;
2356 v1 = (struct lov_user_md *)((char *)comp_v1 +
2357 comp_v1->lcm_entries[i].lcme_offset);
2358 ext = &comp_v1->lcm_entries[i].lcme_extent;
2360 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2361 lod_comp->llc_extent.e_start = ext->e_start;
2362 lod_comp->llc_extent.e_end = ext->e_end;
2363 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2364 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2366 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2367 if (!lod_comp->llc_stripe_count ||
2368 lod_comp->llc_stripe_count == (__u16)-1)
2369 lod_comp->llc_stripe_count =
2370 desc->ld_default_stripe_count;
2371 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2372 if (!lod_comp->llc_stripe_size)
2373 lod_comp->llc_stripe_size =
2374 desc->ld_default_stripe_size;
2376 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2377 v3 = (struct lov_user_md_v3 *) v1;
2378 if (v3->lmm_pool_name[0] != '\0') {
2379 rc = lod_set_pool(&lod_comp->llc_pool,
2387 old_array = lo->ldo_comp_entries;
2388 old_array_cnt = lo->ldo_comp_cnt;
2390 lo->ldo_comp_entries = comp_array;
2391 lo->ldo_comp_cnt = array_cnt;
2393 /* No need to increase layout generation here, it will be increased
2394 * later when generating component ID for the new components */
2396 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2397 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2398 XATTR_NAME_LOV, 0, th);
2400 lo->ldo_comp_entries = old_array;
2401 lo->ldo_comp_cnt = old_array_cnt;
2405 OBD_FREE(old_array, sizeof(*lod_comp) * old_array_cnt);
2407 LASSERT(lo->ldo_mirror_count == 1);
2408 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2413 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2414 lod_comp = &comp_array[i];
2415 if (lod_comp->llc_pool != NULL) {
2416 OBD_FREE(lod_comp->llc_pool,
2417 strlen(lod_comp->llc_pool) + 1);
2418 lod_comp->llc_pool = NULL;
2421 OBD_FREE(comp_array, sizeof(*comp_array) * array_cnt);
2426 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2427 * the '$field' can only be 'flags' now. The xattr value is binary
2428 * lov_comp_md_v1 which contains the component ID(s) and the value of
2429 * the field to be modified.
2431 * \param[in] env execution environment
2432 * \param[in] dt dt_object to be modified
2433 * \param[in] op operation string, like "set.flags"
2434 * \param[in] buf buffer contains components to be set
2435 * \parem[in] th thandle
2437 * \retval 0 on success
2438 * \retval negative errno on failure
2440 static int lod_declare_layout_set(const struct lu_env *env,
2441 struct dt_object *dt,
2442 char *op, const struct lu_buf *buf,
2445 struct lod_layout_component *lod_comp;
2446 struct lod_thread_info *info = lod_env_info(env);
2447 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2448 struct lod_object *lo = lod_dt_obj(dt);
2449 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2452 bool changed = false;
2455 if (strcmp(op, "set.flags") != 0) {
2456 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2457 lod2obd(d)->obd_name, op);
2461 magic = comp_v1->lcm_magic;
2462 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2463 lustre_swab_lov_comp_md_v1(comp_v1);
2464 magic = comp_v1->lcm_magic;
2467 if (magic != LOV_USER_MAGIC_COMP_V1)
2470 if (comp_v1->lcm_entry_count == 0) {
2471 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2472 lod2obd(d)->obd_name);
2476 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2477 id = comp_v1->lcm_entries[i].lcme_id;
2479 for (j = 0; j < lo->ldo_comp_cnt; j++) {
2480 lod_comp = &lo->ldo_comp_entries[j];
2481 if (id == lod_comp->llc_id || id == LCME_ID_ALL) {
2482 lod_comp->llc_flags =
2483 comp_v1->lcm_entries[i].lcme_flags;
2490 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
2491 lod2obd(d)->obd_name);
2495 lod_obj_inc_layout_gen(lo);
2497 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2498 rc = lod_sub_declare_xattr_set(env, dt, &info->lti_buf,
2499 XATTR_NAME_LOV, 0, th);
2504 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
2505 * and the xattr value is a unique component ID or a special lcme_id.
2507 * \param[in] env execution environment
2508 * \param[in] dt dt_object to be operated on
2509 * \param[in] buf buffer contains component ID or lcme_id
2510 * \parem[in] th thandle
2512 * \retval 0 on success
2513 * \retval negative errno on failure
2515 static int lod_declare_layout_del(const struct lu_env *env,
2516 struct dt_object *dt,
2517 const struct lu_buf *buf,
2520 struct lod_thread_info *info = lod_env_info(env);
2521 struct dt_object *next = dt_object_child(dt);
2522 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2523 struct lod_object *lo = lod_dt_obj(dt);
2524 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2525 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2526 __u32 magic, id, flags, neg_flags = 0;
2530 LASSERT(lo->ldo_is_composite);
2532 if (lo->ldo_flr_state != LCM_FL_NOT_FLR)
2535 magic = comp_v1->lcm_magic;
2536 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2537 lustre_swab_lov_comp_md_v1(comp_v1);
2538 magic = comp_v1->lcm_magic;
2541 if (magic != LOV_USER_MAGIC_COMP_V1)
2544 id = comp_v1->lcm_entries[0].lcme_id;
2545 flags = comp_v1->lcm_entries[0].lcme_flags;
2547 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
2548 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
2549 lod2obd(d)->obd_name, id, flags);
2553 if (id != LCME_ID_INVAL && flags != 0) {
2554 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
2555 lod2obd(d)->obd_name);
2559 if (flags & LCME_FL_NEG) {
2560 neg_flags = flags & ~LCME_FL_NEG;
2564 left = lo->ldo_comp_cnt;
2568 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
2569 struct lod_layout_component *lod_comp;
2571 lod_comp = &lo->ldo_comp_entries[i];
2573 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
2575 else if (flags && !(flags & lod_comp->llc_flags))
2577 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
2580 if (left != (i + 1)) {
2581 CDEBUG(D_LAYOUT, "%s: this deletion will create "
2582 "a hole.\n", lod2obd(d)->obd_name);
2587 /* Mark the component as deleted */
2588 lod_comp->llc_id = LCME_ID_INVAL;
2590 /* Not instantiated component */
2591 if (lod_comp->llc_stripe == NULL)
2594 LASSERT(lod_comp->llc_stripe_count > 0);
2595 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
2596 struct dt_object *obj = lod_comp->llc_stripe[j];
2600 rc = lod_sub_declare_destroy(env, obj, th);
2606 LASSERTF(left >= 0, "left = %d\n", left);
2607 if (left == lo->ldo_comp_cnt) {
2608 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
2609 lod2obd(d)->obd_name, id);
2613 memset(attr, 0, sizeof(*attr));
2614 attr->la_valid = LA_SIZE;
2615 rc = lod_sub_declare_attr_set(env, next, attr, th);
2620 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2621 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2622 XATTR_NAME_LOV, 0, th);
2624 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
2631 * Declare layout add/set/del operations issued by special xattr names:
2633 * XATTR_LUSTRE_LOV.add add component(s) to existing file
2634 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
2635 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
2637 * \param[in] env execution environment
2638 * \param[in] dt object
2639 * \param[in] name name of xattr
2640 * \param[in] buf lu_buf contains xattr value
2641 * \param[in] th transaction handle
2643 * \retval 0 on success
2644 * \retval negative if failed
2646 static int lod_declare_modify_layout(const struct lu_env *env,
2647 struct dt_object *dt,
2649 const struct lu_buf *buf,
2652 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2653 struct lod_object *lo = lod_dt_obj(dt);
2654 struct dt_object *next = dt_object_child(&lo->ldo_obj);
2656 int rc, len = strlen(XATTR_LUSTRE_LOV);
2659 LASSERT(dt_object_exists(dt));
2661 if (strlen(name) <= len || name[len] != '.') {
2662 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
2663 lod2obd(d)->obd_name, name);
2668 dt_write_lock(env, next, 0);
2669 rc = lod_load_striping_locked(env, lo);
2673 /* the layout to be modified must be a composite layout */
2674 if (!lo->ldo_is_composite) {
2675 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
2676 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
2677 GOTO(unlock, rc = -EINVAL);
2680 op = (char *)name + len;
2681 if (strcmp(op, "add") == 0) {
2682 rc = lod_declare_layout_add(env, dt, buf, th);
2683 } else if (strcmp(op, "del") == 0) {
2684 rc = lod_declare_layout_del(env, dt, buf, th);
2685 } else if (strncmp(op, "set", strlen("set")) == 0) {
2686 rc = lod_declare_layout_set(env, dt, op, buf, th);
2688 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
2689 lod2obd(d)->obd_name, name);
2690 GOTO(unlock, rc = -ENOTSUPP);
2694 lod_object_free_striping(env, lo);
2695 dt_write_unlock(env, next);
2701 * Convert a plain file lov_mds_md to a composite layout.
2703 * \param[in,out] info the thread info::lti_ea_store buffer contains little
2704 * endian plain file layout
2706 * \retval 0 on success, <0 on failure
2708 static int lod_layout_convert(struct lod_thread_info *info)
2710 struct lov_mds_md *lmm = info->lti_ea_store;
2711 struct lov_mds_md *lmm_save;
2712 struct lov_comp_md_v1 *lcm;
2713 struct lov_comp_md_entry_v1 *lcme;
2719 /* realloc buffer to a composite layout which contains one component */
2720 blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
2721 le32_to_cpu(lmm->lmm_magic));
2722 size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
2724 OBD_ALLOC_LARGE(lmm_save, blob_size);
2726 GOTO(out, rc = -ENOMEM);
2728 memcpy(lmm_save, lmm, blob_size);
2730 if (info->lti_ea_store_size < size) {
2731 rc = lod_ea_store_resize(info, size);
2736 lcm = info->lti_ea_store;
2737 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
2738 lcm->lcm_size = cpu_to_le32(size);
2739 lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
2740 lmm_save->lmm_layout_gen));
2741 lcm->lcm_flags = cpu_to_le16(LCM_FL_NOT_FLR);
2742 lcm->lcm_entry_count = cpu_to_le16(1);
2743 lcm->lcm_mirror_count = 0;
2745 lcme = &lcm->lcm_entries[0];
2746 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
2747 lcme->lcme_extent.e_start = 0;
2748 lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
2749 lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
2750 lcme->lcme_size = cpu_to_le32(blob_size);
2752 memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
2757 OBD_FREE_LARGE(lmm_save, blob_size);
2762 * Merge layouts to form a mirrored file.
2764 static int lod_declare_layout_merge(const struct lu_env *env,
2765 struct dt_object *dt, const struct lu_buf *mbuf,
2768 struct lod_thread_info *info = lod_env_info(env);
2769 struct lu_buf *buf = &info->lti_buf;
2770 struct lod_object *lo = lod_dt_obj(dt);
2771 struct lov_comp_md_v1 *lcm;
2772 struct lov_comp_md_v1 *cur_lcm;
2773 struct lov_comp_md_v1 *merge_lcm;
2774 struct lov_comp_md_entry_v1 *lcme;
2777 __u16 cur_entry_count;
2778 __u16 merge_entry_count;
2780 __u16 mirror_id = 0;
2785 merge_lcm = mbuf->lb_buf;
2786 if (mbuf->lb_len < sizeof(*merge_lcm))
2789 /* must be an existing layout from disk */
2790 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
2793 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
2795 /* do not allow to merge two mirrored files */
2796 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
2799 /* verify the target buffer */
2800 rc = lod_get_lov_ea(env, lo);
2802 RETURN(rc ? : -ENODATA);
2804 cur_lcm = info->lti_ea_store;
2805 switch (le32_to_cpu(cur_lcm->lcm_magic)) {
2808 rc = lod_layout_convert(info);
2810 case LOV_MAGIC_COMP_V1:
2819 /* info->lti_ea_store could be reallocated in lod_layout_convert() */
2820 cur_lcm = info->lti_ea_store;
2821 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
2823 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
2824 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
2825 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
2828 /* size of new layout */
2829 size = le32_to_cpu(cur_lcm->lcm_size) +
2830 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
2832 memset(buf, 0, sizeof(*buf));
2833 lu_buf_alloc(buf, size);
2834 if (buf->lb_buf == NULL)
2838 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
2840 offset = sizeof(*lcm) +
2841 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
2842 for (i = 0; i < cur_entry_count; i++) {
2843 struct lov_comp_md_entry_v1 *cur_lcme;
2845 lcme = &lcm->lcm_entries[i];
2846 cur_lcme = &cur_lcm->lcm_entries[i];
2848 lcme->lcme_offset = cpu_to_le32(offset);
2849 memcpy((char *)lcm + offset,
2850 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
2851 le32_to_cpu(lcme->lcme_size));
2853 offset += le32_to_cpu(lcme->lcme_size);
2855 if (mirror_count == 1) {
2856 /* new mirrored file, create new mirror ID */
2857 id = pflr_id(1, i + 1);
2858 lcme->lcme_id = cpu_to_le32(id);
2861 id = MAX(le32_to_cpu(lcme->lcme_id), id);
2864 mirror_id = mirror_id_of(id) + 1;
2865 for (i = 0; i < merge_entry_count; i++) {
2866 struct lov_comp_md_entry_v1 *merge_lcme;
2868 merge_lcme = &merge_lcm->lcm_entries[i];
2869 lcme = &lcm->lcm_entries[cur_entry_count + i];
2871 *lcme = *merge_lcme;
2872 lcme->lcme_offset = cpu_to_le32(offset);
2874 id = pflr_id(mirror_id, i + 1);
2875 lcme->lcme_id = cpu_to_le32(id);
2877 memcpy((char *)lcm + offset,
2878 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
2879 le32_to_cpu(lcme->lcme_size));
2881 offset += le32_to_cpu(lcme->lcme_size);
2884 /* fixup layout information */
2885 lod_obj_inc_layout_gen(lo);
2886 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
2887 lcm->lcm_size = cpu_to_le32(size);
2888 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
2889 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
2890 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NOT_FLR)
2891 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
2893 LASSERT(dt_write_locked(env, dt_object_child(dt)));
2894 lod_object_free_striping(env, lo);
2895 rc = lod_parse_striping(env, lo, buf);
2899 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
2900 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
2908 * Implementation of dt_object_operations::do_declare_xattr_set.
2910 * \see dt_object_operations::do_declare_xattr_set() in the API description
2913 * the extension to the API:
2914 * - declaring LOVEA requests striping creation
2915 * - LU_XATTR_REPLACE means layout swap
2917 static int lod_declare_xattr_set(const struct lu_env *env,
2918 struct dt_object *dt,
2919 const struct lu_buf *buf,
2920 const char *name, int fl,
2923 struct dt_object *next = dt_object_child(dt);
2924 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2929 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
2930 if ((S_ISREG(mode) || mode == 0) &&
2931 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE)) &&
2932 (strcmp(name, XATTR_NAME_LOV) == 0 ||
2933 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
2935 * this is a request to create object's striping.
2937 * allow to declare predefined striping on a new (!mode) object
2938 * which is supposed to be replay of regular file creation
2939 * (when LOV setting is declared)
2941 * LU_XATTR_REPLACE is set to indicate a layout swap
2943 if (dt_object_exists(dt)) {
2944 rc = dt_attr_get(env, next, attr);
2948 memset(attr, 0, sizeof(*attr));
2949 attr->la_valid = LA_TYPE | LA_MODE;
2950 attr->la_mode = S_IFREG;
2952 rc = lod_declare_striped_create(env, dt, attr, buf, th);
2953 } else if (fl & LU_XATTR_MERGE) {
2954 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
2955 strcmp(name, XATTR_LUSTRE_LOV) == 0);
2956 rc = lod_declare_layout_merge(env, dt, buf, th);
2957 } else if (S_ISREG(mode) &&
2958 strlen(name) > strlen(XATTR_LUSTRE_LOV) + 1 &&
2959 strncmp(name, XATTR_LUSTRE_LOV,
2960 strlen(XATTR_LUSTRE_LOV)) == 0) {
2962 * this is a request to modify object's striping.
2963 * add/set/del component(s).
2965 if (!dt_object_exists(dt))
2968 rc = lod_declare_modify_layout(env, dt, name, buf, th);
2969 } else if (S_ISDIR(mode)) {
2970 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
2971 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
2972 rc = lod_replace_parent_fid(env, dt, th, true);
2974 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2981 * Apply xattr changes to the object.
2983 * Applies xattr changes to the object and the stripes if the latter exist.
2985 * \param[in] env execution environment
2986 * \param[in] dt object
2987 * \param[in] buf buffer pointing to the new value of xattr
2988 * \param[in] name name of xattr
2989 * \param[in] fl flags
2990 * \param[in] th transaction handle
2992 * \retval 0 on success
2993 * \retval negative if failed
2995 static int lod_xattr_set_internal(const struct lu_env *env,
2996 struct dt_object *dt,
2997 const struct lu_buf *buf,
2998 const char *name, int fl,
3001 struct dt_object *next = dt_object_child(dt);
3002 struct lod_object *lo = lod_dt_obj(dt);
3007 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3008 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3011 /* Note: Do not set LinkEA on sub-stripes, otherwise
3012 * it will confuse the fid2path process(see mdt_path_current()).
3013 * The linkEA between master and sub-stripes is set in
3014 * lod_xattr_set_lmv(). */
3015 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
3018 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3019 LASSERT(lo->ldo_stripe[i]);
3021 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
3031 * Delete an extended attribute.
3033 * Deletes specified xattr from the object and the stripes if the latter exist.
3035 * \param[in] env execution environment
3036 * \param[in] dt object
3037 * \param[in] name name of xattr
3038 * \param[in] th transaction handle
3040 * \retval 0 on success
3041 * \retval negative if failed
3043 static int lod_xattr_del_internal(const struct lu_env *env,
3044 struct dt_object *dt,
3045 const char *name, struct thandle *th)
3047 struct dt_object *next = dt_object_child(dt);
3048 struct lod_object *lo = lod_dt_obj(dt);
3053 rc = lod_sub_xattr_del(env, next, name, th);
3054 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3057 if (lo->ldo_dir_stripe_count == 0)
3060 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3061 LASSERT(lo->ldo_stripe[i]);
3063 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3072 * Set default striping on a directory.
3074 * Sets specified striping on a directory object unless it matches the default
3075 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3076 * EA. This striping will be used when regular file is being created in this
3079 * \param[in] env execution environment
3080 * \param[in] dt the striped object
3081 * \param[in] buf buffer with the striping
3082 * \param[in] name name of EA
3083 * \param[in] fl xattr flag (see OSD API description)
3084 * \param[in] th transaction handle
3086 * \retval 0 on success
3087 * \retval negative if failed
3089 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
3090 struct dt_object *dt,
3091 const struct lu_buf *buf,
3092 const char *name, int fl,
3095 struct lov_user_md_v1 *lum;
3096 struct lov_user_md_v3 *v3 = NULL;
3097 const char *pool_name = NULL;
3102 LASSERT(buf != NULL && buf->lb_buf != NULL);
3105 switch (lum->lmm_magic) {
3106 case LOV_USER_MAGIC_V3:
3108 if (v3->lmm_pool_name[0] != '\0')
3109 pool_name = v3->lmm_pool_name;
3111 case LOV_USER_MAGIC_V1:
3112 /* if { size, offset, count } = { 0, -1, 0 } and no pool
3113 * (i.e. all default values specified) then delete default
3114 * striping from dir. */
3116 "set default striping: sz %u # %u offset %d %s %s\n",
3117 (unsigned)lum->lmm_stripe_size,
3118 (unsigned)lum->lmm_stripe_count,
3119 (int)lum->lmm_stripe_offset,
3120 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
3122 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
3123 lum->lmm_stripe_count,
3124 lum->lmm_stripe_offset,
3127 case LOV_USER_MAGIC_COMP_V1:
3131 CERROR("Invalid magic %x\n", lum->lmm_magic);
3136 rc = lod_xattr_del_internal(env, dt, name, th);
3140 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3147 * Set default striping on a directory object.
3149 * Sets specified striping on a directory object unless it matches the default
3150 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3151 * EA. This striping will be used when a new directory is being created in the
3154 * \param[in] env execution environment
3155 * \param[in] dt the striped object
3156 * \param[in] buf buffer with the striping
3157 * \param[in] name name of EA
3158 * \param[in] fl xattr flag (see OSD API description)
3159 * \param[in] th transaction handle
3161 * \retval 0 on success
3162 * \retval negative if failed
3164 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
3165 struct dt_object *dt,
3166 const struct lu_buf *buf,
3167 const char *name, int fl,
3170 struct lmv_user_md_v1 *lum;
3174 LASSERT(buf != NULL && buf->lb_buf != NULL);
3177 CDEBUG(D_OTHER, "set default stripe_count # %u stripe_offset %d\n",
3178 le32_to_cpu(lum->lum_stripe_count),
3179 (int)le32_to_cpu(lum->lum_stripe_offset));
3181 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
3182 le32_to_cpu(lum->lum_stripe_offset)) &&
3183 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
3184 rc = lod_xattr_del_internal(env, dt, name, th);
3188 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3197 * Turn directory into a striped directory.
3199 * During replay the client sends the striping created before MDT
3200 * failure, then the layer above LOD sends this defined striping
3201 * using ->do_xattr_set(), so LOD uses this method to replay creation
3202 * of the stripes. Notice the original information for the striping
3203 * (#stripes, FIDs, etc) was transferred in declare path.
3205 * \param[in] env execution environment
3206 * \param[in] dt the striped object
3207 * \param[in] buf not used currently
3208 * \param[in] name not used currently
3209 * \param[in] fl xattr flag (see OSD API description)
3210 * \param[in] th transaction handle
3212 * \retval 0 on success
3213 * \retval negative if failed
3215 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
3216 const struct lu_buf *buf, const char *name,
3217 int fl, struct thandle *th)
3219 struct lod_object *lo = lod_dt_obj(dt);
3220 struct lod_thread_info *info = lod_env_info(env);
3221 struct lu_attr *attr = &info->lti_attr;
3222 struct dt_object_format *dof = &info->lti_format;
3223 struct lu_buf lmv_buf;
3224 struct lu_buf slave_lmv_buf;
3225 struct lmv_mds_md_v1 *lmm;
3226 struct lmv_mds_md_v1 *slave_lmm = NULL;
3227 struct dt_insert_rec *rec = &info->lti_dt_rec;
3232 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3235 /* The stripes are supposed to be allocated in declare phase,
3236 * if there are no stripes being allocated, it will skip */
3237 if (lo->ldo_dir_stripe_count == 0)
3240 rc = dt_attr_get(env, dt_object_child(dt), attr);
3244 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME |
3245 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
3246 dof->dof_type = DFT_DIR;
3248 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
3251 lmm = lmv_buf.lb_buf;
3253 OBD_ALLOC_PTR(slave_lmm);
3254 if (slave_lmm == NULL)
3257 lod_prep_slave_lmv_md(slave_lmm, lmm);
3258 slave_lmv_buf.lb_buf = slave_lmm;
3259 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
3261 rec->rec_type = S_IFDIR;
3262 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3263 struct dt_object *dto;
3264 char *stripe_name = info->lti_key;
3265 struct lu_name *sname;
3266 struct linkea_data ldata = { NULL };
3267 struct lu_buf linkea_buf;
3269 dto = lo->ldo_stripe[i];
3271 dt_write_lock(env, dto, MOR_TGT_CHILD);
3272 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
3274 dt_write_unlock(env, dto);
3278 rc = lod_sub_ref_add(env, dto, th);
3279 dt_write_unlock(env, dto);
3283 rec->rec_fid = lu_object_fid(&dto->do_lu);
3284 rc = lod_sub_insert(env, dto, (const struct dt_rec *)rec,
3285 (const struct dt_key *)dot, th, 0);
3289 rec->rec_fid = lu_object_fid(&dt->do_lu);
3290 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
3291 (const struct dt_key *)dotdot, th, 0);
3295 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
3296 cfs_fail_val != i) {
3297 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
3299 slave_lmm->lmv_master_mdt_index =
3302 slave_lmm->lmv_master_mdt_index =
3305 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
3306 XATTR_NAME_LMV, fl, th);
3311 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
3313 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3314 PFID(lu_object_fid(&dto->do_lu)), i + 1);
3316 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3317 PFID(lu_object_fid(&dto->do_lu)), i);
3319 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
3320 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
3321 sname, lu_object_fid(&dt->do_lu));
3325 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
3326 linkea_buf.lb_len = ldata.ld_leh->leh_len;
3327 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
3328 XATTR_NAME_LINK, 0, th);
3332 rec->rec_fid = lu_object_fid(&dto->do_lu);
3333 rc = lod_sub_insert(env, dt_object_child(dt),
3334 (const struct dt_rec *)rec,
3335 (const struct dt_key *)stripe_name, th, 0);
3339 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
3344 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
3345 rc = lod_sub_xattr_set(env, dt_object_child(dt),
3346 &lmv_buf, XATTR_NAME_LMV, fl, th);
3348 if (slave_lmm != NULL)
3349 OBD_FREE_PTR(slave_lmm);
3355 * Helper function to declare/execute creation of a striped directory
3357 * Called in declare/create object path, prepare striping for a directory
3358 * and prepare defaults data striping for the objects to be created in
3359 * that directory. Notice the function calls "declaration" or "execution"
3360 * methods depending on \a declare param. This is a consequence of the
3361 * current approach while we don't have natural distributed transactions:
3362 * we basically execute non-local updates in the declare phase. So, the
3363 * arguments for the both phases are the same and this is the reason for
3364 * this function to exist.
3366 * \param[in] env execution environment
3367 * \param[in] dt object
3368 * \param[in] attr attributes the stripes will be created with
3369 * \param[in] lmu lmv_user_md if MDT indices are specified
3370 * \param[in] dof format of stripes (see OSD API description)
3371 * \param[in] th transaction handle
3372 * \param[in] declare where to call "declare" or "execute" methods
3374 * \retval 0 on success
3375 * \retval negative if failed
3377 static int lod_dir_striping_create_internal(const struct lu_env *env,
3378 struct dt_object *dt,
3379 struct lu_attr *attr,
3380 const struct lu_buf *lmu,
3381 struct dt_object_format *dof,
3385 struct lod_thread_info *info = lod_env_info(env);
3386 struct lod_object *lo = lod_dt_obj(dt);
3387 const struct lod_default_striping *lds = lo->ldo_def_striping;
3391 LASSERT(ergo(lds != NULL,
3392 lds->lds_def_striping_set ||
3393 lds->lds_dir_def_striping_set));
3395 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
3396 lo->ldo_dir_stripe_offset)) {
3398 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3399 int stripe_count = lo->ldo_dir_stripe_count;
3401 if (info->lti_ea_store_size < sizeof(*v1)) {
3402 rc = lod_ea_store_resize(info, sizeof(*v1));
3405 v1 = info->lti_ea_store;
3408 memset(v1, 0, sizeof(*v1));
3409 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3410 v1->lum_stripe_count = cpu_to_le32(stripe_count);
3411 v1->lum_stripe_offset =
3412 cpu_to_le32(lo->ldo_dir_stripe_offset);
3414 info->lti_buf.lb_buf = v1;
3415 info->lti_buf.lb_len = sizeof(*v1);
3416 lmu = &info->lti_buf;
3420 rc = lod_declare_xattr_set_lmv(env, dt, attr, lmu, dof,
3423 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
3429 /* Transfer default LMV striping from the parent */
3430 if (lds != NULL && lds->lds_dir_def_striping_set &&
3431 !LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
3432 lds->lds_dir_def_stripe_offset)) {
3433 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3435 if (info->lti_ea_store_size < sizeof(*v1)) {
3436 rc = lod_ea_store_resize(info, sizeof(*v1));
3439 v1 = info->lti_ea_store;
3442 memset(v1, 0, sizeof(*v1));
3443 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3444 v1->lum_stripe_count =
3445 cpu_to_le32(lds->lds_dir_def_stripe_count);
3446 v1->lum_stripe_offset =
3447 cpu_to_le32(lds->lds_dir_def_stripe_offset);
3449 cpu_to_le32(lds->lds_dir_def_hash_type);
3451 info->lti_buf.lb_buf = v1;
3452 info->lti_buf.lb_len = sizeof(*v1);
3454 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3455 XATTR_NAME_DEFAULT_LMV,
3458 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
3460 XATTR_NAME_DEFAULT_LMV, 0,
3466 /* Transfer default LOV striping from the parent */
3467 if (lds != NULL && lds->lds_def_striping_set &&
3468 lds->lds_def_comp_cnt != 0) {
3469 struct lov_mds_md *lmm;
3470 int lmm_size = lod_comp_md_size(lo, true);
3472 if (info->lti_ea_store_size < lmm_size) {
3473 rc = lod_ea_store_resize(info, lmm_size);
3477 lmm = info->lti_ea_store;
3479 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
3483 info->lti_buf.lb_buf = lmm;
3484 info->lti_buf.lb_len = lmm_size;
3487 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3488 XATTR_NAME_LOV, 0, th);
3490 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
3491 XATTR_NAME_LOV, 0, th);
3499 static int lod_declare_dir_striping_create(const struct lu_env *env,
3500 struct dt_object *dt,
3501 struct lu_attr *attr,
3503 struct dt_object_format *dof,
3506 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
3510 static int lod_dir_striping_create(const struct lu_env *env,
3511 struct dt_object *dt,
3512 struct lu_attr *attr,
3513 struct dt_object_format *dof,
3516 return lod_dir_striping_create_internal(env, dt, attr, NULL, dof, th,
3521 * Make LOV EA for striped object.
3523 * Generate striping information and store it in the LOV EA of the given
3524 * object. The caller must ensure nobody else is calling the function
3525 * against the object concurrently. The transaction must be started.
3526 * FLDB service must be running as well; it's used to map FID to the target,
3527 * which is stored in LOV EA.
3529 * \param[in] env execution environment for this thread
3530 * \param[in] lo LOD object
3531 * \param[in] th transaction handle
3533 * \retval 0 if LOV EA is stored successfully
3534 * \retval negative error number on failure
3536 static int lod_generate_and_set_lovea(const struct lu_env *env,
3537 struct lod_object *lo,
3540 struct lod_thread_info *info = lod_env_info(env);
3541 struct dt_object *next = dt_object_child(&lo->ldo_obj);
3542 struct lov_mds_md_v1 *lmm;
3548 if (lo->ldo_comp_cnt == 0) {
3549 lod_object_free_striping(env, lo);
3550 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
3554 lmm_size = lod_comp_md_size(lo, false);
3555 if (info->lti_ea_store_size < lmm_size) {
3556 rc = lod_ea_store_resize(info, lmm_size);
3560 lmm = info->lti_ea_store;
3562 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
3566 info->lti_buf.lb_buf = lmm;
3567 info->lti_buf.lb_len = lmm_size;
3568 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
3569 XATTR_NAME_LOV, 0, th);
3574 * Delete layout component(s)
3576 * \param[in] env execution environment for this thread
3577 * \param[in] dt object
3578 * \param[in] th transaction handle
3580 * \retval 0 on success
3581 * \retval negative error number on failure
3583 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
3586 struct lod_layout_component *lod_comp;
3587 struct lod_object *lo = lod_dt_obj(dt);
3588 struct dt_object *next = dt_object_child(dt);
3589 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3592 LASSERT(lo->ldo_is_composite);
3593 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
3595 left = lo->ldo_comp_cnt;
3596 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3597 lod_comp = &lo->ldo_comp_entries[i];
3599 if (lod_comp->llc_id != LCME_ID_INVAL)
3603 /* Not instantiated component */
3604 if (lod_comp->llc_stripe == NULL)
3607 LASSERT(lod_comp->llc_stripe_count > 0);
3608 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
3609 struct dt_object *obj = lod_comp->llc_stripe[j];
3613 rc = lod_sub_destroy(env, obj, th);
3617 lu_object_put(env, &obj->do_lu);
3618 lod_comp->llc_stripe[j] = NULL;
3620 OBD_FREE(lod_comp->llc_stripe, sizeof(struct dt_object *) *
3621 lod_comp->llc_stripes_allocated);
3622 lod_comp->llc_stripe = NULL;
3623 lod_comp->llc_stripes_allocated = 0;
3624 lod_obj_set_pool(lo, i, NULL);
3625 if (lod_comp->llc_ostlist.op_array) {
3626 OBD_FREE(lod_comp->llc_ostlist.op_array,
3627 lod_comp->llc_ostlist.op_size);
3628 lod_comp->llc_ostlist.op_array = NULL;
3629 lod_comp->llc_ostlist.op_size = 0;
3633 LASSERTF(left >= 0 && left < lo->ldo_comp_cnt, "left = %d\n", left);
3635 struct lod_layout_component *comp_array;
3637 OBD_ALLOC(comp_array, sizeof(*comp_array) * left);
3638 if (comp_array == NULL)
3639 GOTO(out, rc = -ENOMEM);
3641 memcpy(&comp_array[0], &lo->ldo_comp_entries[0],
3642 sizeof(*comp_array) * left);
3644 OBD_FREE(lo->ldo_comp_entries,
3645 sizeof(*comp_array) * lo->ldo_comp_cnt);
3646 lo->ldo_comp_entries = comp_array;
3647 lo->ldo_comp_cnt = left;
3649 LASSERT(lo->ldo_mirror_count == 1);
3650 lo->ldo_mirrors[0].lme_end = left - 1;
3651 lod_obj_inc_layout_gen(lo);
3653 lod_free_comp_entries(lo);
3656 LASSERT(dt_object_exists(dt));
3657 rc = dt_attr_get(env, next, attr);
3661 if (attr->la_size > 0) {
3663 attr->la_valid = LA_SIZE;
3664 rc = lod_sub_attr_set(env, next, attr, th);
3669 rc = lod_generate_and_set_lovea(env, lo, th);
3673 lod_object_free_striping(env, lo);
3678 * Implementation of dt_object_operations::do_xattr_set.
3680 * Sets specified extended attribute on the object. Three types of EAs are
3682 * LOV EA - stores striping for a regular file or default striping (when set
3684 * LMV EA - stores a marker for the striped directories
3685 * DMV EA - stores default directory striping
3687 * When striping is applied to a non-striped existing object (this is called
3688 * late striping), then LOD notices the caller wants to turn the object into a
3689 * striped one. The stripe objects are created and appropriate EA is set:
3690 * LOV EA storing all the stripes directly or LMV EA storing just a small header
3691 * with striping configuration.
3693 * \see dt_object_operations::do_xattr_set() in the API description for details.
3695 static int lod_xattr_set(const struct lu_env *env,
3696 struct dt_object *dt, const struct lu_buf *buf,
3697 const char *name, int fl, struct thandle *th)
3699 struct dt_object *next = dt_object_child(dt);
3703 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
3704 strcmp(name, XATTR_NAME_LMV) == 0) {
3705 struct lmv_mds_md_v1 *lmm = buf->lb_buf;
3707 if (lmm != NULL && le32_to_cpu(lmm->lmv_hash_type) &
3708 LMV_HASH_FLAG_MIGRATION)
3709 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3711 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
3716 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
3717 strcmp(name, XATTR_NAME_LOV) == 0) {
3719 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl, th);
3721 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
3722 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
3724 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
3727 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
3728 (!strcmp(name, XATTR_NAME_LOV) ||
3729 !strncmp(name, XATTR_LUSTRE_LOV,
3730 strlen(XATTR_LUSTRE_LOV)))) {
3731 /* in case of lov EA swap, just set it
3732 * if not, it is a replay so check striping match what we
3733 * already have during req replay, declare_xattr_set()
3734 * defines striping, then create() does the work */
3735 if (fl & LU_XATTR_REPLACE) {
3736 /* free stripes, then update disk */
3737 lod_object_free_striping(env, lod_dt_obj(dt));
3739 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3740 } else if (dt_object_remote(dt)) {
3741 /* This only happens during migration, see
3742 * mdd_migrate_create(), in which Master MDT will
3743 * create a remote target object, and only set
3744 * (migrating) stripe EA on the remote object,
3745 * and does not need creating each stripes. */
3746 rc = lod_sub_xattr_set(env, next, buf, name,
3748 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
3749 /* delete component(s) */
3750 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
3751 rc = lod_layout_del(env, dt, th);
3754 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
3755 * it's going to create create file with specified
3756 * component(s), the striping must have not being
3757 * cached in this case;
3759 * Otherwise, it's going to add/change component(s) to
3760 * an existing file, the striping must have been cached
3763 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
3764 !strcmp(name, XATTR_NAME_LOV),
3765 !lod_dt_obj(dt)->ldo_comp_cached));
3767 rc = lod_striped_create(env, dt, NULL, NULL, th);
3770 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3771 rc = lod_replace_parent_fid(env, dt, th, false);
3776 /* then all other xattr */
3777 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3783 * Implementation of dt_object_operations::do_declare_xattr_del.
3785 * \see dt_object_operations::do_declare_xattr_del() in the API description
3788 static int lod_declare_xattr_del(const struct lu_env *env,
3789 struct dt_object *dt, const char *name,
3792 struct lod_object *lo = lod_dt_obj(dt);
3797 rc = lod_sub_declare_xattr_del(env, dt_object_child(dt), name, th);
3801 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3804 /* set xattr to each stripes, if needed */
3805 rc = lod_load_striping(env, lo);
3809 if (lo->ldo_dir_stripe_count == 0)
3812 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3813 LASSERT(lo->ldo_stripe[i]);
3814 rc = lod_sub_declare_xattr_del(env, lo->ldo_stripe[i],
3824 * Implementation of dt_object_operations::do_xattr_del.
3826 * If EA storing a regular striping is being deleted, then release
3827 * all the references to the stripe objects in core.
3829 * \see dt_object_operations::do_xattr_del() in the API description for details.
3831 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
3832 const char *name, struct thandle *th)
3834 struct dt_object *next = dt_object_child(dt);
3835 struct lod_object *lo = lod_dt_obj(dt);
3840 if (!strcmp(name, XATTR_NAME_LOV))
3841 lod_object_free_striping(env, lod_dt_obj(dt));
3843 rc = lod_sub_xattr_del(env, next, name, th);
3844 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3847 if (lo->ldo_dir_stripe_count == 0)
3850 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3851 LASSERT(lo->ldo_stripe[i]);
3853 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3862 * Implementation of dt_object_operations::do_xattr_list.
3864 * \see dt_object_operations::do_xattr_list() in the API description
3867 static int lod_xattr_list(const struct lu_env *env,
3868 struct dt_object *dt, const struct lu_buf *buf)
3870 return dt_xattr_list(env, dt_object_child(dt), buf);
3873 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
3875 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
3880 * Get default striping.
3882 * \param[in] env execution environment
3883 * \param[in] lo object
3884 * \param[out] lds default striping
3886 * \retval 0 on success
3887 * \retval negative if failed
3889 static int lod_get_default_lov_striping(const struct lu_env *env,
3890 struct lod_object *lo,
3891 struct lod_default_striping *lds)
3893 struct lod_thread_info *info = lod_env_info(env);
3894 struct lov_user_md_v1 *v1 = NULL;
3895 struct lov_user_md_v3 *v3 = NULL;
3896 struct lov_comp_md_v1 *comp_v1 = NULL;
3903 lds->lds_def_striping_set = 0;
3905 rc = lod_get_lov_ea(env, lo);
3909 if (rc < (typeof(rc))sizeof(struct lov_user_md))
3912 v1 = info->lti_ea_store;
3913 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
3914 lustre_swab_lov_user_md_v1(v1);
3915 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
3916 v3 = (struct lov_user_md_v3 *)v1;
3917 lustre_swab_lov_user_md_v3(v3);
3918 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
3919 comp_v1 = (struct lov_comp_md_v1 *)v1;
3920 lustre_swab_lov_comp_md_v1(comp_v1);
3923 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1 &&
3924 v1->lmm_magic != LOV_MAGIC_COMP_V1)
3927 if (v1->lmm_magic == LOV_MAGIC_COMP_V1) {
3928 comp_v1 = (struct lov_comp_md_v1 *)v1;
3929 comp_cnt = comp_v1->lcm_entry_count;
3932 mirror_cnt = comp_v1->lcm_mirror_count + 1;
3940 /* realloc default comp entries if necessary */
3941 rc = lod_def_striping_comp_resize(lds, comp_cnt);
3945 lds->lds_def_comp_cnt = comp_cnt;
3946 lds->lds_def_striping_is_composite = composite;
3947 lds->lds_def_mirror_cnt = mirror_cnt;
3949 for (i = 0; i < comp_cnt; i++) {
3950 struct lod_layout_component *lod_comp;
3951 struct lu_extent *ext;
3954 lod_comp = &lds->lds_def_comp_entries[i];
3956 * reset lod_comp values, llc_stripes is always NULL in
3957 * the default striping template, llc_pool will be reset
3960 memset(lod_comp, 0, offsetof(typeof(*lod_comp), llc_pool));
3963 v1 = (struct lov_user_md *)((char *)comp_v1 +
3964 comp_v1->lcm_entries[i].lcme_offset);
3965 ext = &comp_v1->lcm_entries[i].lcme_extent;
3966 lod_comp->llc_extent = *ext;
3969 if (v1->lmm_pattern != LOV_PATTERN_RAID0 &&
3970 v1->lmm_pattern != LOV_PATTERN_MDT &&
3971 v1->lmm_pattern != 0) {
3972 lod_free_def_comp_entries(lds);
3976 CDEBUG(D_LAYOUT, DFID" stripe_count=%d stripe_size=%d "
3977 "stripe_offset=%d\n",
3978 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
3979 (int)v1->lmm_stripe_count, (int)v1->lmm_stripe_size,
3980 (int)v1->lmm_stripe_offset);
3982 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
3983 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
3984 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
3985 lod_comp->llc_pattern = v1->lmm_pattern;
3988 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
3989 /* XXX: sanity check here */
3990 v3 = (struct lov_user_md_v3 *) v1;
3991 if (v3->lmm_pool_name[0] != '\0')
3992 pool = v3->lmm_pool_name;
3994 lod_set_def_pool(lds, i, pool);
3997 lds->lds_def_striping_set = 1;
4002 * Get default directory striping.
4004 * \param[in] env execution environment
4005 * \param[in] lo object
4006 * \param[out] lds default striping
4008 * \retval 0 on success
4009 * \retval negative if failed
4011 static int lod_get_default_lmv_striping(const struct lu_env *env,
4012 struct lod_object *lo,
4013 struct lod_default_striping *lds)
4015 struct lod_thread_info *info = lod_env_info(env);
4016 struct lmv_user_md_v1 *v1 = NULL;
4020 lds->lds_dir_def_striping_set = 0;
4021 rc = lod_get_default_lmv_ea(env, lo);
4025 if (rc < (typeof(rc))sizeof(struct lmv_user_md))
4028 v1 = info->lti_ea_store;
4030 lds->lds_dir_def_stripe_count = le32_to_cpu(v1->lum_stripe_count);
4031 lds->lds_dir_def_stripe_offset = le32_to_cpu(v1->lum_stripe_offset);
4032 lds->lds_dir_def_hash_type = le32_to_cpu(v1->lum_hash_type);
4033 lds->lds_dir_def_striping_set = 1;
4039 * Get default striping in the object.
4041 * Get object default striping and default directory striping.
4043 * \param[in] env execution environment
4044 * \param[in] lo object
4045 * \param[out] lds default striping
4047 * \retval 0 on success
4048 * \retval negative if failed
4050 static int lod_get_default_striping(const struct lu_env *env,
4051 struct lod_object *lo,
4052 struct lod_default_striping *lds)
4056 rc = lod_get_default_lov_striping(env, lo, lds);
4057 rc1 = lod_get_default_lmv_striping(env, lo, lds);
4058 if (rc == 0 && rc1 < 0)
4065 * Apply default striping on object.
4067 * If object striping pattern is not set, set to the one in default striping.
4068 * The default striping is from parent or fs.
4070 * \param[in] lo new object
4071 * \param[in] lds default striping
4072 * \param[in] mode new object's mode
4074 static void lod_striping_from_default(struct lod_object *lo,
4075 const struct lod_default_striping *lds,
4078 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
4079 struct lov_desc *desc = &d->lod_desc;
4082 if (lds->lds_def_striping_set && S_ISREG(mode)) {
4083 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
4084 lds->lds_def_comp_cnt);
4088 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
4089 if (lds->lds_def_mirror_cnt > 1)
4090 lo->ldo_flr_state = LCM_FL_RDONLY;
4092 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4093 struct lod_layout_component *obj_comp =
4094 &lo->ldo_comp_entries[i];
4095 struct lod_layout_component *def_comp =
4096 &lds->lds_def_comp_entries[i];
4098 CDEBUG(D_LAYOUT, "Inherite from default: size:%hu "
4099 "nr:%u offset:%u pattern %#x %s\n",
4100 def_comp->llc_stripe_size,
4101 def_comp->llc_stripe_count,
4102 def_comp->llc_stripe_offset,
4103 def_comp->llc_pattern,
4104 def_comp->llc_pool ?: "");
4106 *obj_comp = *def_comp;
4107 if (def_comp->llc_pool != NULL) {
4108 /* pointer was copied from def_comp */
4109 obj_comp->llc_pool = NULL;
4110 lod_obj_set_pool(lo, i, def_comp->llc_pool);
4114 * Don't initialize these fields for plain layout
4115 * (v1/v3) here, they are inherited in the order of
4116 * 'parent' -> 'fs default (root)' -> 'global default
4117 * values for stripe_count & stripe_size'.
4119 * see lod_ah_init().
4121 if (!lo->ldo_is_composite)
4124 if (obj_comp->llc_stripe_count <= 0 &&
4125 obj_comp->llc_pattern != LOV_PATTERN_MDT)
4126 obj_comp->llc_stripe_count =
4127 desc->ld_default_stripe_count;
4128 if (obj_comp->llc_stripe_size <= 0)
4129 obj_comp->llc_stripe_size =
4130 desc->ld_default_stripe_size;
4132 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
4133 if (lo->ldo_dir_stripe_count == 0)
4134 lo->ldo_dir_stripe_count =
4135 lds->lds_dir_def_stripe_count;
4136 if (lo->ldo_dir_stripe_offset == -1)
4137 lo->ldo_dir_stripe_offset =
4138 lds->lds_dir_def_stripe_offset;
4139 if (lo->ldo_dir_hash_type == 0)
4140 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
4142 CDEBUG(D_LAYOUT, "striping from default dir: count:%hu, "
4143 "offset:%u, hash_type:%u\n",
4144 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
4145 lo->ldo_dir_hash_type);
4149 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root)
4151 struct lod_layout_component *lod_comp;
4153 if (lo->ldo_comp_cnt == 0)
4156 if (lo->ldo_is_composite)
4159 lod_comp = &lo->ldo_comp_entries[0];
4161 if (lod_comp->llc_stripe_count <= 0 ||
4162 lod_comp->llc_stripe_size <= 0)
4165 if (from_root && (lod_comp->llc_pool == NULL ||
4166 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
4173 * Implementation of dt_object_operations::do_ah_init.
4175 * This method is used to make a decision on the striping configuration for the
4176 * object being created. It can be taken from the \a parent object if it exists,
4177 * or filesystem's default. The resulting configuration (number of stripes,
4178 * stripe size/offset, pool name, etc) is stored in the object itself and will
4179 * be used by the methods like ->doo_declare_create().
4181 * \see dt_object_operations::do_ah_init() in the API description for details.
4183 static void lod_ah_init(const struct lu_env *env,
4184 struct dt_allocation_hint *ah,
4185 struct dt_object *parent,
4186 struct dt_object *child,
4189 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
4190 struct lod_thread_info *info = lod_env_info(env);
4191 struct lod_default_striping *lds = &info->lti_def_striping;
4192 struct dt_object *nextp = NULL;
4193 struct dt_object *nextc;
4194 struct lod_object *lp = NULL;
4195 struct lod_object *lc;
4196 struct lov_desc *desc;
4197 struct lod_layout_component *lod_comp;
4203 if (likely(parent)) {
4204 nextp = dt_object_child(parent);
4205 lp = lod_dt_obj(parent);
4208 nextc = dt_object_child(child);
4209 lc = lod_dt_obj(child);
4211 LASSERT(!lod_obj_is_striped(child));
4212 /* default layout template may have been set on the regular file
4213 * when this is called from mdd_create_data() */
4214 if (S_ISREG(child_mode))
4215 lod_free_comp_entries(lc);
4217 if (!dt_object_exists(nextc))
4218 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
4220 if (S_ISDIR(child_mode)) {
4221 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
4223 /* other default values are 0 */
4224 lc->ldo_dir_stripe_offset = -1;
4226 /* get default striping from parent object */
4227 if (likely(lp != NULL))
4228 lod_get_default_striping(env, lp, lds);
4230 /* set child default striping info, default value is NULL */
4231 if (lds->lds_def_striping_set || lds->lds_dir_def_striping_set)
4232 lc->ldo_def_striping = lds;
4234 /* It should always honour the specified stripes */
4235 /* Note: old client (< 2.7)might also do lfs mkdir, whose EA
4236 * will have old magic. In this case, we should ignore the
4237 * stripe count and try to create dir by default stripe.
4239 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
4240 (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
4241 le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC)) {
4242 lc->ldo_dir_stripe_count =
4243 le32_to_cpu(lum1->lum_stripe_count);
4244 lc->ldo_dir_stripe_offset =
4245 le32_to_cpu(lum1->lum_stripe_offset);
4246 lc->ldo_dir_hash_type =
4247 le32_to_cpu(lum1->lum_hash_type);
4249 "set dirstripe: count %hu, offset %d, hash %u\n",
4250 lc->ldo_dir_stripe_count,
4251 (int)lc->ldo_dir_stripe_offset,
4252 lc->ldo_dir_hash_type);
4254 /* transfer defaults LMV to new directory */
4255 lod_striping_from_default(lc, lds, child_mode);
4258 /* shrink the stripe_count to the avaible MDT count */
4259 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
4260 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))
4261 lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
4263 /* Directory will be striped only if stripe_count > 1, if
4264 * stripe_count == 1, let's reset stripe_count = 0 to avoid
4265 * create single master stripe and also help to unify the
4266 * stripe handling of directories and files */
4267 if (lc->ldo_dir_stripe_count == 1)
4268 lc->ldo_dir_stripe_count = 0;
4270 CDEBUG(D_INFO, "final dir stripe [%hu %d %u]\n",
4271 lc->ldo_dir_stripe_count,
4272 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
4277 /* child object regular file*/
4279 if (!lod_object_will_be_striped(S_ISREG(child_mode),
4280 lu_object_fid(&child->do_lu)))
4283 /* If object is going to be striped over OSTs, transfer default
4284 * striping information to the child, so that we can use it
4285 * during declaration and creation.
4287 * Try from the parent first.
4289 if (likely(lp != NULL)) {
4290 rc = lod_get_default_lov_striping(env, lp, lds);
4292 lod_striping_from_default(lc, lds, child_mode);
4295 /* Initialize lod_device::lod_md_root object reference */
4296 if (d->lod_md_root == NULL) {
4297 struct dt_object *root;
4298 struct lod_object *lroot;
4300 lu_root_fid(&info->lti_fid);
4301 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
4302 if (!IS_ERR(root)) {
4303 lroot = lod_dt_obj(root);
4305 spin_lock(&d->lod_lock);
4306 if (d->lod_md_root != NULL)
4307 dt_object_put(env, &d->lod_md_root->ldo_obj);
4308 d->lod_md_root = lroot;
4309 spin_unlock(&d->lod_lock);
4313 /* try inherit layout from the root object (fs default) when:
4314 * - parent does not have default layout; or
4315 * - parent has plain(v1/v3) default layout, and some attributes
4316 * are not specified in the default layout;
4318 if (d->lod_md_root != NULL && lod_need_inherit_more(lc, true)) {
4319 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds);
4322 if (lc->ldo_comp_cnt == 0) {
4323 lod_striping_from_default(lc, lds, child_mode);
4324 } else if (!lds->lds_def_striping_is_composite) {
4325 struct lod_layout_component *def_comp;
4327 LASSERT(!lc->ldo_is_composite);
4328 lod_comp = &lc->ldo_comp_entries[0];
4329 def_comp = &lds->lds_def_comp_entries[0];
4331 if (lod_comp->llc_stripe_count <= 0)
4332 lod_comp->llc_stripe_count =
4333 def_comp->llc_stripe_count;
4334 if (lod_comp->llc_stripe_size <= 0)
4335 lod_comp->llc_stripe_size =
4336 def_comp->llc_stripe_size;
4337 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT)
4338 lod_comp->llc_stripe_offset =
4339 def_comp->llc_stripe_offset;
4340 if (lod_comp->llc_pool == NULL)
4341 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
4346 * fs default striping may not be explicitly set, or historically set
4347 * in config log, use them.
4349 if (lod_need_inherit_more(lc, false)) {
4350 if (lc->ldo_comp_cnt == 0) {
4351 rc = lod_alloc_comp_entries(lc, 0, 1);
4353 /* fail to allocate memory, will create a
4354 * non-striped file. */
4356 lc->ldo_is_composite = 0;
4357 lod_comp = &lc->ldo_comp_entries[0];
4358 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4360 LASSERT(!lc->ldo_is_composite);
4361 lod_comp = &lc->ldo_comp_entries[0];
4362 desc = &d->lod_desc;
4363 if (lod_comp->llc_stripe_count <= 0)
4364 lod_comp->llc_stripe_count =
4365 desc->ld_default_stripe_count;
4366 if (lod_comp->llc_stripe_size <= 0)
4367 lod_comp->llc_stripe_size =
4368 desc->ld_default_stripe_size;
4374 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
4376 * Size initialization on late striping.
4378 * Propagate the size of a truncated object to a deferred striping.
4379 * This function handles a special case when truncate was done on a
4380 * non-striped object and now while the striping is being created
4381 * we can't lose that size, so we have to propagate it to the stripes
4384 * \param[in] env execution environment
4385 * \param[in] dt object
4386 * \param[in] th transaction handle
4388 * \retval 0 on success
4389 * \retval negative if failed
4391 static int lod_declare_init_size(const struct lu_env *env,
4392 struct dt_object *dt, struct thandle *th)
4394 struct dt_object *next = dt_object_child(dt);
4395 struct lod_object *lo = lod_dt_obj(dt);
4396 struct dt_object **objects = NULL;
4397 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4398 uint64_t size, offs;
4399 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
4400 struct lu_extent size_ext;
4403 if (!lod_obj_is_striped(dt))
4406 rc = dt_attr_get(env, next, attr);
4407 LASSERT(attr->la_valid & LA_SIZE);
4411 size = attr->la_size;
4415 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
4416 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4417 struct lod_layout_component *lod_comp;
4418 struct lu_extent *extent;
4420 lod_comp = &lo->ldo_comp_entries[i];
4422 if (lod_comp->llc_stripe == NULL)
4425 extent = &lod_comp->llc_extent;
4426 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
4427 if (!lo->ldo_is_composite ||
4428 lu_extent_is_overlapped(extent, &size_ext)) {
4429 objects = lod_comp->llc_stripe;
4430 stripe_count = lod_comp->llc_stripe_count;
4431 stripe_size = lod_comp->llc_stripe_size;
4434 if (stripe_count == 0)
4437 LASSERT(objects != NULL && stripe_size != 0);
4438 /* ll_do_div64(a, b) returns a % b, and a = a / b */
4439 ll_do_div64(size, (__u64)stripe_size);
4440 stripe = ll_do_div64(size, (__u64)stripe_count);
4441 LASSERT(objects[stripe] != NULL);
4443 size = size * stripe_size;
4444 offs = attr->la_size;
4445 size += ll_do_div64(offs, stripe_size);
4447 attr->la_valid = LA_SIZE;
4448 attr->la_size = size;
4450 rc = lod_sub_declare_attr_set(env, objects[stripe],
4459 * Declare creation of striped object.
4461 * The function declares creation stripes for a regular object. The function
4462 * also declares whether the stripes will be created with non-zero size if
4463 * previously size was set non-zero on the master object. If object \a dt is
4464 * not local, then only fully defined striping can be applied in \a lovea.
4465 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
4468 * \param[in] env execution environment
4469 * \param[in] dt object
4470 * \param[in] attr attributes the stripes will be created with
4471 * \param[in] lovea a buffer containing striping description
4472 * \param[in] th transaction handle
4474 * \retval 0 on success
4475 * \retval negative if failed
4477 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
4478 struct lu_attr *attr,
4479 const struct lu_buf *lovea, struct thandle *th)
4481 struct lod_thread_info *info = lod_env_info(env);
4482 struct dt_object *next = dt_object_child(dt);
4483 struct lod_object *lo = lod_dt_obj(dt);
4487 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
4488 GOTO(out, rc = -ENOMEM);
4490 if (!dt_object_remote(next)) {
4491 /* choose OST and generate appropriate objects */
4492 rc = lod_prepare_create(env, lo, attr, lovea, th);
4497 * declare storage for striping data
4499 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
4501 /* LOD can not choose OST objects for remote objects, i.e.
4502 * stripes must be ready before that. Right now, it can only
4503 * happen during migrate, i.e. migrate process needs to create
4504 * remote regular file (mdd_migrate_create), then the migrate
4505 * process will provide stripeEA. */
4506 LASSERT(lovea != NULL);
4507 info->lti_buf = *lovea;
4510 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
4511 XATTR_NAME_LOV, 0, th);
4516 * if striping is created with local object's size > 0,
4517 * we have to propagate this size to specific object
4518 * the case is possible only when local object was created previously
4520 if (dt_object_exists(next))
4521 rc = lod_declare_init_size(env, dt, th);
4524 /* failed to create striping or to set initial size, let's reset
4525 * config so that others don't get confused */
4527 lod_object_free_striping(env, lo);
4533 * Implementation of dt_object_operations::do_declare_create.
4535 * The method declares creation of a new object. If the object will be striped,
4536 * then helper functions are called to find FIDs for the stripes, declare
4537 * creation of the stripes and declare initialization of the striping
4538 * information to be stored in the master object.
4540 * \see dt_object_operations::do_declare_create() in the API description
4543 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
4544 struct lu_attr *attr,
4545 struct dt_allocation_hint *hint,
4546 struct dt_object_format *dof, struct thandle *th)
4548 struct dt_object *next = dt_object_child(dt);
4549 struct lod_object *lo = lod_dt_obj(dt);
4558 * first of all, we declare creation of local object
4560 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
4565 * it's lod_ah_init() that has decided the object will be striped
4567 if (dof->dof_type == DFT_REGULAR) {
4568 /* callers don't want stripes */
4569 /* XXX: all tricky interactions with ->ah_make_hint() decided
4570 * to use striping, then ->declare_create() behaving differently
4571 * should be cleaned */
4572 if (dof->u.dof_reg.striped != 0)
4573 rc = lod_declare_striped_create(env, dt, attr,
4575 } else if (dof->dof_type == DFT_DIR) {
4576 struct seq_server_site *ss;
4577 struct lu_buf buf = { NULL };
4578 struct lu_buf *lmu = NULL;
4580 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
4582 /* If the parent has default stripeEA, and client
4583 * did not find it before sending create request,
4584 * then MDT will return -EREMOTE, and client will
4585 * retrieve the default stripeEA and re-create the
4588 * Note: if dah_eadata != NULL, it means creating the
4589 * striped directory with specified stripeEA, then it
4590 * should ignore the default stripeEA */
4591 if (hint != NULL && hint->dah_eadata == NULL) {
4592 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
4593 GOTO(out, rc = -EREMOTE);
4595 if (lo->ldo_dir_stripe_offset == -1) {
4596 /* child and parent should be in the same MDT */
4597 if (hint->dah_parent != NULL &&
4598 dt_object_remote(hint->dah_parent))
4599 GOTO(out, rc = -EREMOTE);
4600 } else if (lo->ldo_dir_stripe_offset !=
4602 struct lod_device *lod;
4603 struct lod_tgt_descs *ltd;
4604 struct lod_tgt_desc *tgt = NULL;
4605 bool found_mdt = false;
4608 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
4609 ltd = &lod->lod_mdt_descs;
4610 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
4611 tgt = LTD_TGT(ltd, i);
4612 if (tgt->ltd_index ==
4613 lo->ldo_dir_stripe_offset) {
4619 /* If the MDT indicated by stripe_offset can be
4620 * found, then tell client to resend the create
4621 * request to the correct MDT, otherwise return
4622 * error to client */
4624 GOTO(out, rc = -EREMOTE);
4626 GOTO(out, rc = -EINVAL);
4628 } else if (hint && hint->dah_eadata) {
4630 lmu->lb_buf = (void *)hint->dah_eadata;
4631 lmu->lb_len = hint->dah_eadata_len;
4634 rc = lod_declare_dir_striping_create(env, dt, attr, lmu, dof,
4638 /* failed to create striping or to set initial size, let's reset
4639 * config so that others don't get confused */
4641 lod_object_free_striping(env, lo);
4646 * Generate component ID for new created component.
4648 * \param[in] lo LOD object
4649 * \param[in] comp_idx index of ldo_comp_entries
4651 * \retval component ID on success
4652 * \retval LCME_ID_INVAL on failure
4654 static __u32 lod_gen_component_id(struct lod_object *lo,
4655 int mirror_id, int comp_idx)
4657 struct lod_layout_component *lod_comp;
4658 __u32 id, start, end;
4661 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
4663 lod_obj_inc_layout_gen(lo);
4664 id = lo->ldo_layout_gen;
4665 if (likely(id <= SEQ_ID_MAX))
4666 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
4668 /* Layout generation wraps, need to check collisions. */
4669 start = id & SEQ_ID_MASK;
4672 for (id = start; id <= end; id++) {
4673 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4674 lod_comp = &lo->ldo_comp_entries[i];
4675 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
4678 /* Found the ununsed ID */
4679 if (i == lo->ldo_comp_cnt)
4680 RETURN(pflr_id(mirror_id, id));
4682 if (end == LCME_ID_MAX) {
4684 end = min(lo->ldo_layout_gen & LCME_ID_MASK,
4685 (__u32)(LCME_ID_MAX - 1));
4689 RETURN(LCME_ID_INVAL);
4693 * Creation of a striped regular object.
4695 * The function is called to create the stripe objects for a regular
4696 * striped file. This can happen at the initial object creation or
4697 * when the caller asks LOD to do so using ->do_xattr_set() method
4698 * (so called late striping). Notice all the information are already
4699 * prepared in the form of the list of objects (ldo_stripe field).
4700 * This is done during declare phase.
4702 * \param[in] env execution environment
4703 * \param[in] dt object
4704 * \param[in] attr attributes the stripes will be created with
4705 * \param[in] dof format of stripes (see OSD API description)
4706 * \param[in] th transaction handle
4708 * \retval 0 on success
4709 * \retval negative if failed
4711 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
4712 struct lu_attr *attr, struct dt_object_format *dof,
4715 struct lod_layout_component *lod_comp;
4716 struct lod_object *lo = lod_dt_obj(dt);
4721 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
4723 mirror_id = lo->ldo_mirror_count > 1 ? 1 : 0;
4725 /* create all underlying objects */
4726 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4727 lod_comp = &lo->ldo_comp_entries[i];
4729 if (lod_comp->llc_extent.e_start == 0 && i > 0) /* new mirror */
4732 if (lod_comp->llc_id == LCME_ID_INVAL) {
4733 lod_comp->llc_id = lod_gen_component_id(lo,
4735 if (lod_comp->llc_id == LCME_ID_INVAL)
4736 GOTO(out, rc = -ERANGE);
4739 if (lod_comp_inited(lod_comp))
4742 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
4743 lod_comp_set_init(lod_comp);
4745 if (lov_pattern(lod_comp->llc_pattern) == LOV_PATTERN_MDT)
4746 lod_comp_set_init(lod_comp);
4748 if (lod_comp->llc_stripe == NULL)
4751 LASSERT(lod_comp->llc_stripe_count);
4752 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4753 struct dt_object *object = lod_comp->llc_stripe[j];
4754 LASSERT(object != NULL);
4755 rc = lod_sub_create(env, object, attr, NULL, dof, th);
4759 lod_comp_set_init(lod_comp);
4762 rc = lod_fill_mirrors(lo);
4766 rc = lod_generate_and_set_lovea(env, lo, th);
4770 lo->ldo_comp_cached = 1;
4774 lod_object_free_striping(env, lo);
4779 * Implementation of dt_object_operations::do_create.
4781 * If any of preceeding methods (like ->do_declare_create(),
4782 * ->do_ah_init(), etc) chose to create a striped object,
4783 * then this method will create the master and the stripes.
4785 * \see dt_object_operations::do_create() in the API description for details.
4787 static int lod_create(const struct lu_env *env, struct dt_object *dt,
4788 struct lu_attr *attr, struct dt_allocation_hint *hint,
4789 struct dt_object_format *dof, struct thandle *th)
4794 /* create local object */
4795 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
4799 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
4800 lod_obj_is_striped(dt) && dof->u.dof_reg.striped != 0) {
4801 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
4802 rc = lod_striped_create(env, dt, attr, dof, th);
4809 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
4810 struct dt_object *dt, struct thandle *th,
4811 int comp_idx, int stripe_idx,
4812 struct lod_obj_stripe_cb_data *data)
4814 if (data->locd_declare)
4815 return lod_sub_declare_destroy(env, dt, th);
4816 else if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
4817 stripe_idx == cfs_fail_val)
4818 return lod_sub_destroy(env, dt, th);
4824 * Implementation of dt_object_operations::do_declare_destroy.
4826 * If the object is a striped directory, then the function declares reference
4827 * removal from the master object (this is an index) to the stripes and declares
4828 * destroy of all the stripes. In all the cases, it declares an intention to
4829 * destroy the object itself.
4831 * \see dt_object_operations::do_declare_destroy() in the API description
4834 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
4837 struct dt_object *next = dt_object_child(dt);
4838 struct lod_object *lo = lod_dt_obj(dt);
4839 struct lod_thread_info *info = lod_env_info(env);
4840 char *stripe_name = info->lti_key;
4845 * load striping information, notice we don't do this when object
4846 * is being initialized as we don't need this information till
4847 * few specific cases like destroy, chown
4849 rc = lod_load_striping(env, lo);
4853 /* declare destroy for all underlying objects */
4854 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4855 rc = next->do_ops->do_index_try(env, next,
4856 &dt_directory_features);
4860 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4861 rc = lod_sub_declare_ref_del(env, next, th);
4865 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4866 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
4868 rc = lod_sub_declare_delete(env, next,
4869 (const struct dt_key *)stripe_name, th);
4876 * we declare destroy for the local object
4878 rc = lod_sub_declare_destroy(env, next, th);
4882 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
4883 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
4886 if (!lod_obj_is_striped(dt))
4889 /* declare destroy all striped objects */
4890 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4891 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4892 if (lo->ldo_stripe[i] == NULL)
4895 rc = lod_sub_declare_ref_del(env, lo->ldo_stripe[i],
4898 rc = lod_sub_declare_destroy(env, lo->ldo_stripe[i],
4904 struct lod_obj_stripe_cb_data data = { { 0 } };
4906 data.locd_declare = true;
4907 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
4908 rc = lod_obj_for_each_stripe(env, lo, th, &data);
4915 * Implementation of dt_object_operations::do_destroy.
4917 * If the object is a striped directory, then the function removes references
4918 * from the master object (this is an index) to the stripes and destroys all
4919 * the stripes. In all the cases, the function destroys the object itself.
4921 * \see dt_object_operations::do_destroy() in the API description for details.
4923 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
4926 struct dt_object *next = dt_object_child(dt);
4927 struct lod_object *lo = lod_dt_obj(dt);
4928 struct lod_thread_info *info = lod_env_info(env);
4929 char *stripe_name = info->lti_key;
4934 /* destroy sub-stripe of master object */
4935 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4936 rc = next->do_ops->do_index_try(env, next,
4937 &dt_directory_features);
4941 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4942 rc = lod_sub_ref_del(env, next, th);
4946 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4947 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
4950 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
4951 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
4952 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)));
4954 rc = lod_sub_delete(env, next,
4955 (const struct dt_key *)stripe_name, th);
4961 rc = lod_sub_destroy(env, next, th);
4965 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
4966 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
4969 if (!lod_obj_is_striped(dt))
4972 /* destroy all striped objects */
4973 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4974 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4975 if (lo->ldo_stripe[i] == NULL)
4977 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
4978 i == cfs_fail_val) {
4979 dt_write_lock(env, lo->ldo_stripe[i],
4981 rc = lod_sub_ref_del(env, lo->ldo_stripe[i],
4983 dt_write_unlock(env, lo->ldo_stripe[i]);
4987 rc = lod_sub_destroy(env, lo->ldo_stripe[i],
4994 struct lod_obj_stripe_cb_data data = { { 0 } };
4996 data.locd_declare = false;
4997 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
4998 rc = lod_obj_for_each_stripe(env, lo, th, &data);
5005 * Implementation of dt_object_operations::do_declare_ref_add.
5007 * \see dt_object_operations::do_declare_ref_add() in the API description
5010 static int lod_declare_ref_add(const struct lu_env *env,
5011 struct dt_object *dt, struct thandle *th)
5013 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
5017 * Implementation of dt_object_operations::do_ref_add.
5019 * \see dt_object_operations::do_ref_add() in the API description for details.
5021 static int lod_ref_add(const struct lu_env *env,
5022 struct dt_object *dt, struct thandle *th)
5024 return lod_sub_ref_add(env, dt_object_child(dt), th);
5028 * Implementation of dt_object_operations::do_declare_ref_del.
5030 * \see dt_object_operations::do_declare_ref_del() in the API description
5033 static int lod_declare_ref_del(const struct lu_env *env,
5034 struct dt_object *dt, struct thandle *th)
5036 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
5040 * Implementation of dt_object_operations::do_ref_del
5042 * \see dt_object_operations::do_ref_del() in the API description for details.
5044 static int lod_ref_del(const struct lu_env *env,
5045 struct dt_object *dt, struct thandle *th)
5047 return lod_sub_ref_del(env, dt_object_child(dt), th);
5051 * Implementation of dt_object_operations::do_object_sync.
5053 * \see dt_object_operations::do_object_sync() in the API description
5056 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
5057 __u64 start, __u64 end)
5059 return dt_object_sync(env, dt_object_child(dt), start, end);
5063 * Release LDLM locks on the stripes of a striped directory.
5065 * Iterates over all the locks taken on the stripe objects and
5068 * \param[in] env execution environment
5069 * \param[in] dt striped object
5070 * \param[in] einfo lock description
5071 * \param[in] policy data describing requested lock
5073 * \retval 0 on success
5074 * \retval negative if failed
5076 static int lod_object_unlock_internal(const struct lu_env *env,
5077 struct dt_object *dt,
5078 struct ldlm_enqueue_info *einfo,
5079 union ldlm_policy_data *policy)
5081 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
5086 if (slave_locks == NULL)
5089 for (i = 1; i < slave_locks->count; i++) {
5090 if (lustre_handle_is_used(&slave_locks->handles[i]))
5091 ldlm_lock_decref_and_cancel(&slave_locks->handles[i],
5099 * Implementation of dt_object_operations::do_object_unlock.
5101 * Used to release LDLM lock(s).
5103 * \see dt_object_operations::do_object_unlock() in the API description
5106 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
5107 struct ldlm_enqueue_info *einfo,
5108 union ldlm_policy_data *policy)
5110 struct lod_object *lo = lod_dt_obj(dt);
5111 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
5112 int slave_locks_size;
5116 if (slave_locks == NULL)
5119 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
5120 LASSERT(lo->ldo_dir_stripe_count > 1);
5121 /* Note: for remote lock for single stripe dir, MDT will cancel
5122 * the lock by lockh directly */
5123 LASSERT(!dt_object_remote(dt_object_child(dt)));
5125 /* locks were unlocked in MDT layer */
5126 for (i = 1; i < slave_locks->count; i++) {
5127 LASSERT(!lustre_handle_is_used(&slave_locks->handles[i]));
5128 dt_invalidate(env, lo->ldo_stripe[i]);
5131 slave_locks_size = sizeof(*slave_locks) + slave_locks->count *
5132 sizeof(slave_locks->handles[0]);
5133 OBD_FREE(slave_locks, slave_locks_size);
5134 einfo->ei_cbdata = NULL;
5140 * Implementation of dt_object_operations::do_object_lock.
5142 * Used to get LDLM lock on the non-striped and striped objects.
5144 * \see dt_object_operations::do_object_lock() in the API description
5147 static int lod_object_lock(const struct lu_env *env,
5148 struct dt_object *dt,
5149 struct lustre_handle *lh,
5150 struct ldlm_enqueue_info *einfo,
5151 union ldlm_policy_data *policy)
5153 struct lod_object *lo = lod_dt_obj(dt);
5156 int slave_locks_size;
5157 struct lustre_handle_array *slave_locks = NULL;
5160 /* remote object lock */
5161 if (!einfo->ei_enq_slave) {
5162 LASSERT(dt_object_remote(dt));
5163 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
5167 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
5168 GOTO(out, rc = -ENOTDIR);
5170 rc = lod_load_striping(env, lo);
5175 if (lo->ldo_dir_stripe_count <= 1) {
5177 * NB, ei_cbdata stores pointer to slave locks, if no locks
5178 * taken, make sure it's set to NULL, otherwise MDT will try to
5181 einfo->ei_cbdata = NULL;
5185 slave_locks_size = sizeof(*slave_locks) + lo->ldo_dir_stripe_count *
5186 sizeof(slave_locks->handles[0]);
5187 /* Freed in lod_object_unlock */
5188 OBD_ALLOC(slave_locks, slave_locks_size);
5189 if (slave_locks == NULL)
5190 GOTO(out, rc = -ENOMEM);
5191 slave_locks->count = lo->ldo_dir_stripe_count;
5193 /* striped directory lock */
5194 for (i = 1; i < lo->ldo_dir_stripe_count; i++) {
5195 struct lustre_handle lockh;
5196 struct ldlm_res_id *res_id;
5198 res_id = &lod_env_info(env)->lti_res_id;
5199 fid_build_reg_res_name(lu_object_fid(&lo->ldo_stripe[i]->do_lu),
5201 einfo->ei_res_id = res_id;
5203 LASSERT(lo->ldo_stripe[i] != NULL);
5204 if (likely(dt_object_remote(lo->ldo_stripe[i]))) {
5205 rc = dt_object_lock(env, lo->ldo_stripe[i], &lockh,
5208 struct ldlm_namespace *ns = einfo->ei_namespace;
5209 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
5210 ldlm_completion_callback completion = einfo->ei_cb_cp;
5211 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
5213 if (einfo->ei_mode == LCK_PW ||
5214 einfo->ei_mode == LCK_EX)
5215 dlmflags |= LDLM_FL_COS_INCOMPAT;
5217 /* This only happens if there are mulitple stripes
5218 * on the master MDT, i.e. except stripe0, there are
5219 * other stripes on the Master MDT as well, Only
5220 * happens in the test case right now. */
5221 LASSERT(ns != NULL);
5222 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS,
5223 policy, einfo->ei_mode,
5224 &dlmflags, blocking,
5226 NULL, 0, LVB_T_NONE,
5231 slave_locks->handles[i] = lockh;
5233 einfo->ei_cbdata = slave_locks;
5235 if (rc != 0 && slave_locks != NULL) {
5236 lod_object_unlock_internal(env, dt, einfo, policy);
5237 OBD_FREE(slave_locks, slave_locks_size);
5242 einfo->ei_cbdata = NULL;
5247 * Implementation of dt_object_operations::do_invalidate.
5249 * \see dt_object_operations::do_invalidate() in the API description for details
5251 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
5253 return dt_invalidate(env, dt_object_child(dt));
5256 static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
5260 /* clear memory region that will be used for layout change */
5261 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
5262 info->lti_count = 0;
5264 if (info->lti_comp_size >= comp_cnt)
5267 if (info->lti_comp_size > 0) {
5268 OBD_FREE(info->lti_comp_idx,
5269 info->lti_comp_size * sizeof(__u32));
5270 info->lti_comp_size = 0;
5273 OBD_ALLOC(info->lti_comp_idx, comp_cnt * sizeof(__u32));
5274 if (!info->lti_comp_idx)
5277 info->lti_comp_size = comp_cnt;
5281 static int lod_declare_instantiate_components(const struct lu_env *env,
5282 struct lod_object *lo, struct thandle *th)
5284 struct lod_thread_info *info = lod_env_info(env);
5285 struct ost_pool *inuse = &info->lti_inuse_osts;
5290 LASSERT(info->lti_count < lo->ldo_comp_cnt);
5291 if (info->lti_count > 0) {
5292 /* Prepare inuse array for composite file */
5293 rc = lod_prepare_inuse(env, lo);
5298 for (i = 0; i < info->lti_count; i++) {
5299 rc = lod_qos_prep_create(env, lo, NULL, th,
5300 info->lti_comp_idx[i], inuse);
5306 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
5307 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
5308 &info->lti_buf, XATTR_NAME_LOV, 0, th);
5314 static int lod_declare_update_plain(const struct lu_env *env,
5315 struct lod_object *lo, struct layout_intent *layout,
5316 const struct lu_buf *buf, struct thandle *th)
5318 struct lod_thread_info *info = lod_env_info(env);
5319 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5320 struct lod_layout_component *lod_comp;
5321 struct lov_comp_md_v1 *comp_v1 = NULL;
5322 bool replay = false;
5326 LASSERT(lo->ldo_flr_state == LCM_FL_NOT_FLR);
5329 * In case the client is passing lovea, which only happens during
5330 * the replay of layout intent write RPC for now, we may need to
5331 * parse the lovea and apply new layout configuration.
5333 if (buf && buf->lb_len) {
5334 struct lov_user_md_v1 *v1 = buf->lb_buf;
5336 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
5337 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
5338 LOV_MAGIC_COMP_V1)) {
5339 CERROR("%s: the replay buffer of layout extend "
5340 "(magic %#x) does not contain expected "
5341 "composite layout.\n",
5342 lod2obd(d)->obd_name, v1->lmm_magic);
5343 GOTO(out, rc = -EINVAL);
5346 lod_object_free_striping(env, lo);
5347 rc = lod_use_defined_striping(env, lo, buf);
5351 rc = lod_get_lov_ea(env, lo);
5354 /* old on-disk EA is stored in info->lti_buf */
5355 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
5358 /* non replay path */
5359 rc = lod_load_striping_locked(env, lo);
5364 if (layout->li_opc == LAYOUT_INTENT_TRUNC) {
5366 * trunc transfers [size, eof) in the intent extent, while
5367 * we'd instantiated components covers [0, size).
5369 layout->li_extent.e_end = layout->li_extent.e_start;
5370 layout->li_extent.e_start = 0;
5373 /* Make sure defined layout covers the requested write range. */
5374 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
5375 if (lo->ldo_comp_cnt > 1 &&
5376 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
5377 lod_comp->llc_extent.e_end < layout->li_extent.e_end) {
5378 CDEBUG(replay ? D_ERROR : D_LAYOUT,
5379 "%s: the defined layout [0, %#llx) does not covers "
5380 "the write range "DEXT"\n",
5381 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
5382 PEXT(&layout->li_extent));
5383 GOTO(out, rc = -EINVAL);
5386 CDEBUG(D_LAYOUT, "%s: "DFID": instantiate components "DEXT"\n",
5387 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
5388 PEXT(&layout->li_extent));
5391 * Iterate ld->ldo_comp_entries, find the component whose extent under
5392 * the write range and not instantianted.
5394 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5395 lod_comp = &lo->ldo_comp_entries[i];
5397 if (lod_comp->llc_extent.e_start >= layout->li_extent.e_end)
5401 if (lod_comp_inited(lod_comp))
5405 * In replay path, lod_comp is the EA passed by
5406 * client replay buffer, comp_v1 is the pre-recovery
5407 * on-disk EA, we'd sift out those components which
5408 * were init-ed in the on-disk EA.
5410 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
5415 * this component hasn't instantiated in normal path, or during
5416 * replay it needs replay the instantiation.
5419 /* A released component is being extended */
5420 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
5421 GOTO(out, rc = -EINVAL);
5423 LASSERT(info->lti_comp_idx != NULL);
5424 info->lti_comp_idx[info->lti_count++] = i;
5427 if (info->lti_count == 0)
5430 lod_obj_inc_layout_gen(lo);
5431 rc = lod_declare_instantiate_components(env, lo, th);
5434 lod_object_free_striping(env, lo);
5438 #define lod_foreach_mirror_comp(comp, lo, mirror_idx) \
5439 for (comp = &lo->ldo_comp_entries[lo->ldo_mirrors[mirror_idx].lme_start]; \
5440 comp <= &lo->ldo_comp_entries[lo->ldo_mirrors[mirror_idx].lme_end]; \
5443 static inline int lod_comp_index(struct lod_object *lo,
5444 struct lod_layout_component *lod_comp)
5446 LASSERT(lod_comp >= lo->ldo_comp_entries &&
5447 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
5449 return lod_comp - lo->ldo_comp_entries;
5453 * Stale other mirrors by writing extent.
5455 static void lod_stale_components(struct lod_object *lo, int primary,
5456 struct lu_extent *extent)
5458 struct lod_layout_component *pri_comp, *lod_comp;
5461 /* The writing extent decides which components in the primary
5462 * are affected... */
5463 CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
5464 lod_foreach_mirror_comp(pri_comp, lo, primary) {
5465 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
5468 CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
5469 lod_comp_index(lo, pri_comp),
5470 PEXT(&pri_comp->llc_extent));
5472 for (i = 0; i < lo->ldo_mirror_count; i++) {
5476 /* ... and then stale other components that are
5477 * overlapping with primary components */
5478 lod_foreach_mirror_comp(lod_comp, lo, i) {
5479 if (!lu_extent_is_overlapped(
5480 &pri_comp->llc_extent,
5481 &lod_comp->llc_extent))
5484 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
5485 i, lod_comp_index(lo, lod_comp));
5487 lod_comp->llc_flags |= LCME_FL_STALE;
5488 lo->ldo_mirrors[i].lme_stale = 1;
5494 static int lod_declare_update_rdonly(const struct lu_env *env,
5495 struct lod_object *lo, struct md_layout_change *mlc,
5498 struct lod_thread_info *info = lod_env_info(env);
5499 struct lu_attr *layout_attr = &info->lti_layout_attr;
5500 struct lod_layout_component *lod_comp;
5501 struct layout_intent *layout = mlc->mlc_intent;
5502 struct lu_extent extent = layout->li_extent;
5503 unsigned int seq = 0;
5509 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE);
5510 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
5511 LASSERT(lo->ldo_mirror_count > 0);
5513 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
5514 PFID(lod_object_fid(lo)), PEXT(&extent));
5516 if (OBD_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
5517 get_random_bytes(&seq, sizeof(seq));
5518 seq %= lo->ldo_mirror_count;
5522 * Pick a mirror as the primary.
5523 * Now it only picks the first mirror, this algo can be
5524 * revised later after knowing the topology of cluster or
5525 * the availability of OSTs.
5527 for (picked = -1, i = 0; i < lo->ldo_mirror_count; i++) {
5528 int index = (i + seq) % lo->ldo_mirror_count;
5530 if (!lo->ldo_mirrors[index].lme_stale) {
5535 if (picked < 0) /* failed to pick a primary */
5538 CDEBUG(D_LAYOUT, DFID": picked mirror %u as primary\n",
5539 PFID(lod_object_fid(lo)), lo->ldo_mirrors[picked].lme_id);
5541 /* stale overlapping components from other mirrors */
5542 lod_stale_components(lo, picked, &extent);
5544 /* instantiate components for the picked mirror, start from 0 */
5545 if (layout->li_opc == LAYOUT_INTENT_TRUNC) {
5547 * trunc transfers [size, eof) in the intent extent, we'd
5548 * stale components overlapping [size, eof), while we'd
5549 * instantiated components covers [0, size).
5551 extent.e_end = extent.e_start;
5555 lod_foreach_mirror_comp(lod_comp, lo, picked) {
5556 if (!lu_extent_is_overlapped(&extent,
5557 &lod_comp->llc_extent))
5560 if (lod_comp_inited(lod_comp))
5563 CDEBUG(D_LAYOUT, "instantiate: %u / %u\n",
5564 i, lod_comp_index(lo, lod_comp));
5566 info->lti_comp_idx[info->lti_count++] =
5567 lod_comp_index(lo, lod_comp);
5570 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
5572 /* Reset the layout version once it's becoming too large.
5573 * This way it can make sure that the layout version is
5574 * monotonously increased in this writing era. */
5575 lod_obj_inc_layout_gen(lo);
5576 if (lo->ldo_layout_gen > (LCME_ID_MAX >> 1)) {
5577 __u32 layout_version;
5579 cfs_get_random_bytes(&layout_version, sizeof(layout_version));
5580 lo->ldo_layout_gen = layout_version & 0xffff;
5583 rc = lod_declare_instantiate_components(env, lo, th);
5587 layout_attr->la_valid = LA_LAYOUT_VERSION;
5588 layout_attr->la_layout_version = 0; /* set current version */
5589 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
5595 lod_object_free_striping(env, lo);
5599 static int lod_declare_update_write_pending(const struct lu_env *env,
5600 struct lod_object *lo, struct md_layout_change *mlc,
5603 struct lod_thread_info *info = lod_env_info(env);
5604 struct lu_attr *layout_attr = &info->lti_layout_attr;
5605 struct lod_layout_component *lod_comp;
5606 struct lu_extent extent = { 0 };
5612 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
5613 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
5614 mlc->mlc_opc == MD_LAYOUT_RESYNC);
5616 /* look for the primary mirror */
5617 for (i = 0; i < lo->ldo_mirror_count; i++) {
5618 if (lo->ldo_mirrors[i].lme_stale)
5621 LASSERTF(primary < 0, DFID " has multiple primary: %u / %u",
5622 PFID(lod_object_fid(lo)),
5623 lo->ldo_mirrors[i].lme_id,
5624 lo->ldo_mirrors[primary].lme_id);
5629 CERROR(DFID ": doesn't have a primary mirror\n",
5630 PFID(lod_object_fid(lo)));
5631 GOTO(out, rc = -ENODATA);
5634 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
5635 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
5637 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
5639 /* for LAYOUT_WRITE opc, it has to do the following operations:
5640 * 1. stale overlapping componets from stale mirrors;
5641 * 2. instantiate components of the primary mirror;
5642 * 3. transfter layout version to all objects of the primary;
5644 * for LAYOUT_RESYNC opc, it will do:
5645 * 1. instantiate components of all stale mirrors;
5646 * 2. transfer layout version to all objects to close write era. */
5648 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
5649 LASSERT(mlc->mlc_intent != NULL);
5651 extent = mlc->mlc_intent->li_extent;
5653 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
5654 PFID(lod_object_fid(lo)), PEXT(&extent));
5656 /* 1. stale overlapping components */
5657 lod_stale_components(lo, primary, &extent);
5659 /* 2. find out the components need instantiating.
5660 * instantiate [0, mlc->mlc_intent->e_end) */
5661 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC) {
5663 * trunc transfers [size, eof) in the intent extent,
5664 * we'd stale components overlapping [size, eof),
5665 * while we'd instantiated components covers [0, size).
5667 extent.e_end = extent.e_start;
5671 lod_foreach_mirror_comp(lod_comp, lo, primary) {
5672 if (!lu_extent_is_overlapped(&extent,
5673 &lod_comp->llc_extent))
5676 if (lod_comp_inited(lod_comp))
5679 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
5680 primary, lod_comp_index(lo, lod_comp));
5681 info->lti_comp_idx[info->lti_count++] =
5682 lod_comp_index(lo, lod_comp);
5684 } else { /* MD_LAYOUT_RESYNC */
5685 /* figure out the components that have been instantiated in
5686 * in primary to decide what components should be instantiated
5687 * in stale mirrors */
5688 lod_foreach_mirror_comp(lod_comp, lo, primary) {
5689 if (!lod_comp_inited(lod_comp))
5692 extent.e_end = lod_comp->llc_extent.e_end;
5696 DFID": instantiate all stale components in "DEXT"\n",
5697 PFID(lod_object_fid(lo)), PEXT(&extent));
5699 /* 1. instantiate all components within this extent, even
5700 * non-stale components so that it won't need to instantiate
5701 * those components for mirror truncate later. */
5702 for (i = 0; i < lo->ldo_mirror_count; i++) {
5706 LASSERTF(lo->ldo_mirrors[i].lme_stale,
5707 "both %d and %d are primary\n", i, primary);
5709 lod_foreach_mirror_comp(lod_comp, lo, i) {
5710 if (!lu_extent_is_overlapped(&extent,
5711 &lod_comp->llc_extent))
5714 if (lod_comp_inited(lod_comp))
5717 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
5718 i, lod_comp_index(lo, lod_comp));
5720 info->lti_comp_idx[info->lti_count++] =
5721 lod_comp_index(lo, lod_comp);
5725 /* change the file state to SYNC_PENDING */
5726 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
5729 rc = lod_declare_instantiate_components(env, lo, th);
5733 /* 3. transfer layout version to OST objects.
5734 * transfer new layout version to OST objects so that stale writes
5735 * can be denied. It also ends an era of writing by setting
5736 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
5737 * send write RPC; only resync RPCs could do it. */
5738 layout_attr->la_valid = LA_LAYOUT_VERSION;
5739 layout_attr->la_layout_version = 0; /* set current version */
5740 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
5741 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
5742 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
5746 lod_obj_inc_layout_gen(lo);
5749 lod_object_free_striping(env, lo);
5753 static int lod_declare_update_sync_pending(const struct lu_env *env,
5754 struct lod_object *lo, struct md_layout_change *mlc,
5757 struct lod_thread_info *info = lod_env_info(env);
5758 unsigned sync_components = 0;
5759 unsigned resync_components = 0;
5764 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
5765 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
5766 mlc->mlc_opc == MD_LAYOUT_WRITE);
5768 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
5769 PFID(lod_object_fid(lo)), mlc->mlc_opc);
5771 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
5772 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
5773 PFID(lod_object_fid(lo)));
5775 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
5776 return lod_declare_update_write_pending(env, lo, mlc, th);
5779 /* MD_LAYOUT_RESYNC_DONE */
5781 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5782 struct lod_layout_component *lod_comp;
5785 lod_comp = &lo->ldo_comp_entries[i];
5787 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
5792 for (j = 0; j < mlc->mlc_resync_count; j++) {
5793 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
5796 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
5797 lod_comp->llc_flags &= ~LCME_FL_STALE;
5798 resync_components++;
5804 for (i = 0; i < mlc->mlc_resync_count; i++) {
5805 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
5808 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
5809 "or already synced\n", PFID(lod_object_fid(lo)),
5810 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
5811 GOTO(out, rc = -EINVAL);
5814 if (!sync_components || !resync_components) {
5815 CDEBUG(D_LAYOUT, DFID": no mirror in sync or resync\n",
5816 PFID(lod_object_fid(lo)));
5818 /* tend to return an error code here to prevent
5819 * the MDT from setting SoM attribute */
5820 GOTO(out, rc = -EINVAL);
5823 CDEBUG(D_LAYOUT, DFID": resynced %u/%zu components\n",
5824 PFID(lod_object_fid(lo)),
5825 resync_components, mlc->mlc_resync_count);
5827 lo->ldo_flr_state = LCM_FL_RDONLY;
5828 lod_obj_inc_layout_gen(lo);
5830 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
5831 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
5832 &info->lti_buf, XATTR_NAME_LOV, 0, th);
5837 lod_object_free_striping(env, lo);
5841 static int lod_declare_layout_change(const struct lu_env *env,
5842 struct dt_object *dt, struct md_layout_change *mlc,
5845 struct lod_thread_info *info = lod_env_info(env);
5846 struct lod_object *lo = lod_dt_obj(dt);
5850 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
5851 dt_object_remote(dt_object_child(dt)))
5854 lod_write_lock(env, dt, 0);
5855 rc = lod_load_striping_locked(env, lo);
5859 LASSERT(lo->ldo_comp_cnt > 0);
5861 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
5865 switch (lo->ldo_flr_state) {
5866 case LCM_FL_NOT_FLR:
5867 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
5871 rc = lod_declare_update_rdonly(env, lo, mlc, th);
5873 case LCM_FL_WRITE_PENDING:
5874 rc = lod_declare_update_write_pending(env, lo, mlc, th);
5876 case LCM_FL_SYNC_PENDING:
5877 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
5884 dt_write_unlock(env, dt);
5889 * Instantiate layout component objects which covers the intent write offset.
5891 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
5892 struct md_layout_change *mlc, struct thandle *th)
5894 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
5895 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
5896 struct lod_object *lo = lod_dt_obj(dt);
5899 rc = lod_striped_create(env, dt, attr, NULL, th);
5900 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
5901 layout_attr->la_layout_version |= lo->ldo_layout_gen;
5902 rc = lod_attr_set(env, dt, layout_attr, th);
5908 struct dt_object_operations lod_obj_ops = {
5909 .do_read_lock = lod_read_lock,
5910 .do_write_lock = lod_write_lock,
5911 .do_read_unlock = lod_read_unlock,
5912 .do_write_unlock = lod_write_unlock,
5913 .do_write_locked = lod_write_locked,
5914 .do_attr_get = lod_attr_get,
5915 .do_declare_attr_set = lod_declare_attr_set,
5916 .do_attr_set = lod_attr_set,
5917 .do_xattr_get = lod_xattr_get,
5918 .do_declare_xattr_set = lod_declare_xattr_set,
5919 .do_xattr_set = lod_xattr_set,
5920 .do_declare_xattr_del = lod_declare_xattr_del,
5921 .do_xattr_del = lod_xattr_del,
5922 .do_xattr_list = lod_xattr_list,
5923 .do_ah_init = lod_ah_init,
5924 .do_declare_create = lod_declare_create,
5925 .do_create = lod_create,
5926 .do_declare_destroy = lod_declare_destroy,
5927 .do_destroy = lod_destroy,
5928 .do_index_try = lod_index_try,
5929 .do_declare_ref_add = lod_declare_ref_add,
5930 .do_ref_add = lod_ref_add,
5931 .do_declare_ref_del = lod_declare_ref_del,
5932 .do_ref_del = lod_ref_del,
5933 .do_object_sync = lod_object_sync,
5934 .do_object_lock = lod_object_lock,
5935 .do_object_unlock = lod_object_unlock,
5936 .do_invalidate = lod_invalidate,
5937 .do_declare_layout_change = lod_declare_layout_change,
5938 .do_layout_change = lod_layout_change,
5942 * Implementation of dt_body_operations::dbo_read.
5944 * \see dt_body_operations::dbo_read() in the API description for details.
5946 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
5947 struct lu_buf *buf, loff_t *pos)
5949 struct dt_object *next = dt_object_child(dt);
5951 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
5952 S_ISLNK(dt->do_lu.lo_header->loh_attr));
5953 return next->do_body_ops->dbo_read(env, next, buf, pos);
5957 * Implementation of dt_body_operations::dbo_declare_write.
5959 * \see dt_body_operations::dbo_declare_write() in the API description
5962 static ssize_t lod_declare_write(const struct lu_env *env,
5963 struct dt_object *dt,
5964 const struct lu_buf *buf, loff_t pos,
5967 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
5971 * Implementation of dt_body_operations::dbo_write.
5973 * \see dt_body_operations::dbo_write() in the API description for details.
5975 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
5976 const struct lu_buf *buf, loff_t *pos,
5977 struct thandle *th, int iq)
5979 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
5980 S_ISLNK(dt->do_lu.lo_header->loh_attr));
5981 return lod_sub_write(env, dt_object_child(dt), buf, pos, th, iq);
5984 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
5985 __u64 start, __u64 end, struct thandle *th)
5987 if (dt_object_remote(dt))
5990 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
5993 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
5994 __u64 start, __u64 end, struct thandle *th)
5996 if (dt_object_remote(dt))
5999 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
6000 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
6004 * different type of files use the same body_ops because object may be created
6005 * in OUT, where there is no chance to set correct body_ops for each type, so
6006 * body_ops themselves will check file type inside, see lod_read/write/punch for
6009 const struct dt_body_operations lod_body_ops = {
6010 .dbo_read = lod_read,
6011 .dbo_declare_write = lod_declare_write,
6012 .dbo_write = lod_write,
6013 .dbo_declare_punch = lod_declare_punch,
6014 .dbo_punch = lod_punch,
6018 * Implementation of lu_object_operations::loo_object_init.
6020 * The function determines the type and the index of the target device using
6021 * sequence of the object's FID. Then passes control down to the
6022 * corresponding device:
6023 * OSD for the local objects, OSP for remote
6025 * \see lu_object_operations::loo_object_init() in the API description
6028 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
6029 const struct lu_object_conf *conf)
6031 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
6032 struct lu_device *cdev = NULL;
6033 struct lu_object *cobj;
6034 struct lod_tgt_descs *ltd = NULL;
6035 struct lod_tgt_desc *tgt;
6037 int type = LU_SEQ_RANGE_ANY;
6041 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
6043 /* Note: Sometimes, it will Return EAGAIN here, see
6044 * ptrlpc_import_delay_req(), which might confuse
6045 * lu_object_find_at() and make it wait there incorrectly.
6046 * so we convert it to EIO here.*/
6053 if (type == LU_SEQ_RANGE_MDT &&
6054 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
6055 cdev = &lod->lod_child->dd_lu_dev;
6056 } else if (type == LU_SEQ_RANGE_MDT) {
6057 ltd = &lod->lod_mdt_descs;
6059 } else if (type == LU_SEQ_RANGE_OST) {
6060 ltd = &lod->lod_ost_descs;
6067 if (ltd->ltd_tgts_size > idx &&
6068 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
6069 tgt = LTD_TGT(ltd, idx);
6071 LASSERT(tgt != NULL);
6072 LASSERT(tgt->ltd_tgt != NULL);
6074 cdev = &(tgt->ltd_tgt->dd_lu_dev);
6076 lod_putref(lod, ltd);
6079 if (unlikely(cdev == NULL))
6082 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
6083 if (unlikely(cobj == NULL))
6086 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
6088 lu_object_add(lo, cobj);
6095 * Release resources associated with striping.
6097 * If the object is striped (regular or directory), then release
6098 * the stripe objects references and free the ldo_stripe array.
6100 * \param[in] env execution environment
6101 * \param[in] lo object
6103 void lod_object_free_striping(const struct lu_env *env, struct lod_object *lo)
6105 struct lod_layout_component *lod_comp;
6108 if (lo->ldo_stripe != NULL) {
6109 LASSERT(lo->ldo_comp_entries == NULL);
6110 LASSERT(lo->ldo_dir_stripes_allocated > 0);
6112 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6113 if (lo->ldo_stripe[i])
6114 dt_object_put(env, lo->ldo_stripe[i]);
6117 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
6118 OBD_FREE(lo->ldo_stripe, j);
6119 lo->ldo_stripe = NULL;
6120 lo->ldo_dir_stripes_allocated = 0;
6121 lo->ldo_dir_stripe_loaded = 0;
6122 lo->ldo_dir_stripe_count = 0;
6123 } else if (lo->ldo_comp_entries != NULL) {
6124 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6125 /* free lod_layout_component::llc_stripe array */
6126 lod_comp = &lo->ldo_comp_entries[i];
6128 if (lod_comp->llc_stripe == NULL)
6130 LASSERT(lod_comp->llc_stripes_allocated != 0);
6131 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
6132 if (lod_comp->llc_stripe[j] != NULL)
6134 &lod_comp->llc_stripe[j]->do_lu);
6136 OBD_FREE(lod_comp->llc_stripe,
6137 sizeof(struct dt_object *) *
6138 lod_comp->llc_stripes_allocated);
6139 lod_comp->llc_stripe = NULL;
6140 lod_comp->llc_stripes_allocated = 0;
6142 lod_free_comp_entries(lo);
6143 lo->ldo_comp_cached = 0;
6148 * Implementation of lu_object_operations::loo_object_free.
6150 * \see lu_object_operations::loo_object_free() in the API description
6153 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
6155 struct lod_object *lo = lu2lod_obj(o);
6157 /* release all underlying object pinned */
6158 lod_object_free_striping(env, lo);
6160 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
6164 * Implementation of lu_object_operations::loo_object_release.
6166 * \see lu_object_operations::loo_object_release() in the API description
6169 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
6171 /* XXX: shouldn't we release everything here in case if object
6172 * creation failed before? */
6176 * Implementation of lu_object_operations::loo_object_print.
6178 * \see lu_object_operations::loo_object_print() in the API description
6181 static int lod_object_print(const struct lu_env *env, void *cookie,
6182 lu_printer_t p, const struct lu_object *l)
6184 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
6186 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
6189 struct lu_object_operations lod_lu_obj_ops = {
6190 .loo_object_init = lod_object_init,
6191 .loo_object_free = lod_object_free,
6192 .loo_object_release = lod_object_release,
6193 .loo_object_print = lod_object_print,