4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * lustre/lod/lod_object.c
31 * This file contains implementations of methods for the OSD API
32 * for the Logical Object Device (LOD) layer, which provides a virtual
33 * local OSD object interface to the MDD layer, and abstracts the
34 * addressing of local (OSD) and remote (OSP) objects. The API is
35 * described in the file lustre/include/dt_object.h and in
36 * Documentation/osd-api.txt.
38 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
41 #define DEBUG_SUBSYSTEM S_MDS
43 #include <linux/random.h>
46 #include <obd_class.h>
47 #include <obd_support.h>
49 #include <lustre_fid.h>
50 #include <lustre_linkea.h>
51 #include <lustre_lmv.h>
52 #include <uapi/linux/lustre/lustre_param.h>
53 #include <lustre_swab.h>
54 #include <uapi/linux/lustre/lustre_ver.h>
55 #include <lprocfs_status.h>
56 #include <md_object.h>
58 #include "lod_internal.h"
60 static const char dot[] = ".";
61 static const char dotdot[] = "..";
64 * Implementation of dt_index_operations::dio_lookup
66 * Used with regular (non-striped) objects.
68 * \see dt_index_operations::dio_lookup() in the API description for details.
70 static int lod_lookup(const struct lu_env *env, struct dt_object *dt,
71 struct dt_rec *rec, const struct dt_key *key)
73 struct dt_object *next = dt_object_child(dt);
74 return next->do_index_ops->dio_lookup(env, next, rec, key);
78 * Implementation of dt_index_operations::dio_declare_insert.
80 * Used with regular (non-striped) objects.
82 * \see dt_index_operations::dio_declare_insert() in the API description
85 static int lod_declare_insert(const struct lu_env *env, struct dt_object *dt,
86 const struct dt_rec *rec,
87 const struct dt_key *key, struct thandle *th)
89 return lod_sub_declare_insert(env, dt_object_child(dt), rec, key, th);
93 * Implementation of dt_index_operations::dio_insert.
95 * Used with regular (non-striped) objects
97 * \see dt_index_operations::dio_insert() in the API description for details.
99 static int lod_insert(const struct lu_env *env, struct dt_object *dt,
100 const struct dt_rec *rec, const struct dt_key *key,
103 return lod_sub_insert(env, dt_object_child(dt), rec, key, th);
107 * Implementation of dt_index_operations::dio_declare_delete.
109 * Used with regular (non-striped) objects.
111 * \see dt_index_operations::dio_declare_delete() in the API description
114 static int lod_declare_delete(const struct lu_env *env, struct dt_object *dt,
115 const struct dt_key *key, struct thandle *th)
117 return lod_sub_declare_delete(env, dt_object_child(dt), key, th);
121 * Implementation of dt_index_operations::dio_delete.
123 * Used with regular (non-striped) objects.
125 * \see dt_index_operations::dio_delete() in the API description for details.
127 static int lod_delete(const struct lu_env *env, struct dt_object *dt,
128 const struct dt_key *key, struct thandle *th)
130 return lod_sub_delete(env, dt_object_child(dt), key, th);
134 * Implementation of dt_it_ops::init.
136 * Used with regular (non-striped) objects.
138 * \see dt_it_ops::init() in the API description for details.
140 static struct dt_it *lod_it_init(const struct lu_env *env,
141 struct dt_object *dt, __u32 attr)
143 struct dt_object *next = dt_object_child(dt);
144 struct lod_it *it = &lod_env_info(env)->lti_it;
145 struct dt_it *it_next;
147 it_next = next->do_index_ops->dio_it.init(env, next, attr);
151 /* currently we do not use more than one iterator per thread
152 * so we store it in thread info. if at some point we need
153 * more active iterators in a single thread, we can allocate
155 LASSERT(it->lit_obj == NULL);
157 it->lit_it = it_next;
160 return (struct dt_it *)it;
163 #define LOD_CHECK_IT(env, it) \
165 LASSERT((it)->lit_obj != NULL); \
166 LASSERT((it)->lit_it != NULL); \
170 * Implementation of dt_index_operations::dio_it.fini.
172 * Used with regular (non-striped) objects.
174 * \see dt_index_operations::dio_it.fini() in the API description for details.
176 static void lod_it_fini(const struct lu_env *env, struct dt_it *di)
178 struct lod_it *it = (struct lod_it *)di;
180 LOD_CHECK_IT(env, it);
181 it->lit_obj->do_index_ops->dio_it.fini(env, it->lit_it);
183 /* the iterator not in use any more */
189 * Implementation of dt_it_ops::get.
191 * Used with regular (non-striped) objects.
193 * \see dt_it_ops::get() in the API description for details.
195 static int lod_it_get(const struct lu_env *env, struct dt_it *di,
196 const struct dt_key *key)
198 const struct lod_it *it = (const struct lod_it *)di;
200 LOD_CHECK_IT(env, it);
201 return it->lit_obj->do_index_ops->dio_it.get(env, it->lit_it, key);
205 * Implementation of dt_it_ops::put.
207 * Used with regular (non-striped) objects.
209 * \see dt_it_ops::put() in the API description for details.
211 static void lod_it_put(const struct lu_env *env, struct dt_it *di)
213 struct lod_it *it = (struct lod_it *)di;
215 LOD_CHECK_IT(env, it);
216 return it->lit_obj->do_index_ops->dio_it.put(env, it->lit_it);
220 * Implementation of dt_it_ops::next.
222 * Used with regular (non-striped) objects
224 * \see dt_it_ops::next() in the API description for details.
226 static int lod_it_next(const struct lu_env *env, struct dt_it *di)
228 struct lod_it *it = (struct lod_it *)di;
230 LOD_CHECK_IT(env, it);
231 return it->lit_obj->do_index_ops->dio_it.next(env, it->lit_it);
235 * Implementation of dt_it_ops::key.
237 * Used with regular (non-striped) objects.
239 * \see dt_it_ops::key() in the API description for details.
241 static struct dt_key *lod_it_key(const struct lu_env *env,
242 const struct dt_it *di)
244 const struct lod_it *it = (const struct lod_it *)di;
246 LOD_CHECK_IT(env, it);
247 return it->lit_obj->do_index_ops->dio_it.key(env, it->lit_it);
251 * Implementation of dt_it_ops::key_size.
253 * Used with regular (non-striped) objects.
255 * \see dt_it_ops::key_size() in the API description for details.
257 static int lod_it_key_size(const struct lu_env *env, const struct dt_it *di)
259 struct lod_it *it = (struct lod_it *)di;
261 LOD_CHECK_IT(env, it);
262 return it->lit_obj->do_index_ops->dio_it.key_size(env, it->lit_it);
266 * Implementation of dt_it_ops::rec.
268 * Used with regular (non-striped) objects.
270 * \see dt_it_ops::rec() in the API description for details.
272 static int lod_it_rec(const struct lu_env *env, const struct dt_it *di,
273 struct dt_rec *rec, __u32 attr)
275 const struct lod_it *it = (const struct lod_it *)di;
277 LOD_CHECK_IT(env, it);
278 return it->lit_obj->do_index_ops->dio_it.rec(env, it->lit_it, rec,
283 * Implementation of dt_it_ops::rec_size.
285 * Used with regular (non-striped) objects.
287 * \see dt_it_ops::rec_size() in the API description for details.
289 static int lod_it_rec_size(const struct lu_env *env, const struct dt_it *di,
292 const struct lod_it *it = (const struct lod_it *)di;
294 LOD_CHECK_IT(env, it);
295 return it->lit_obj->do_index_ops->dio_it.rec_size(env, it->lit_it,
300 * Implementation of dt_it_ops::store.
302 * Used with regular (non-striped) objects.
304 * \see dt_it_ops::store() in the API description for details.
306 static __u64 lod_it_store(const struct lu_env *env, const struct dt_it *di)
308 const struct lod_it *it = (const struct lod_it *)di;
310 LOD_CHECK_IT(env, it);
311 return it->lit_obj->do_index_ops->dio_it.store(env, it->lit_it);
315 * Implementation of dt_it_ops::load.
317 * Used with regular (non-striped) objects.
319 * \see dt_it_ops::load() in the API description for details.
321 static int lod_it_load(const struct lu_env *env, const struct dt_it *di,
324 const struct lod_it *it = (const struct lod_it *)di;
326 LOD_CHECK_IT(env, it);
327 return it->lit_obj->do_index_ops->dio_it.load(env, it->lit_it, hash);
331 * Implementation of dt_it_ops::key_rec.
333 * Used with regular (non-striped) objects.
335 * \see dt_it_ops::rec() in the API description for details.
337 static int lod_it_key_rec(const struct lu_env *env, const struct dt_it *di,
340 const struct lod_it *it = (const struct lod_it *)di;
342 LOD_CHECK_IT(env, it);
343 return it->lit_obj->do_index_ops->dio_it.key_rec(env, it->lit_it,
347 static struct dt_index_operations lod_index_ops = {
348 .dio_lookup = lod_lookup,
349 .dio_declare_insert = lod_declare_insert,
350 .dio_insert = lod_insert,
351 .dio_declare_delete = lod_declare_delete,
352 .dio_delete = lod_delete,
360 .key_size = lod_it_key_size,
362 .rec_size = lod_it_rec_size,
363 .store = lod_it_store,
365 .key_rec = lod_it_key_rec,
370 * Implementation of dt_it_ops::init.
372 * Used with striped objects. Internally just initializes the iterator
373 * on the first stripe.
375 * \see dt_it_ops::init() in the API description for details.
377 static struct dt_it *lod_striped_it_init(const struct lu_env *env,
378 struct dt_object *dt, __u32 attr)
380 struct lod_object *lo = lod_dt_obj(dt);
381 struct dt_object *next;
382 struct lod_it *it = &lod_env_info(env)->lti_it;
383 struct dt_it *it_next;
386 LASSERT(lo->ldo_dir_stripe_count > 0);
389 next = lo->ldo_stripe[index];
390 if (next && dt_object_exists(next))
392 } while (++index < lo->ldo_dir_stripe_count);
394 /* no valid stripe */
395 if (!next || !dt_object_exists(next))
396 return ERR_PTR(-ENODEV);
398 LASSERT(next->do_index_ops != NULL);
400 it_next = next->do_index_ops->dio_it.init(env, next, attr);
404 /* currently we do not use more than one iterator per thread
405 * so we store it in thread info. if at some point we need
406 * more active iterators in a single thread, we can allocate
408 LASSERT(it->lit_obj == NULL);
410 it->lit_stripe_index = index;
412 it->lit_it = it_next;
415 return (struct dt_it *)it;
418 #define LOD_CHECK_STRIPED_IT(env, it, lo) \
420 LASSERT((it)->lit_obj != NULL); \
421 LASSERT((it)->lit_it != NULL); \
422 LASSERT((lo)->ldo_dir_stripe_count > 0); \
423 LASSERT((it)->lit_stripe_index < (lo)->ldo_dir_stripe_count); \
427 * Implementation of dt_it_ops::fini.
429 * Used with striped objects.
431 * \see dt_it_ops::fini() in the API description for details.
433 static void lod_striped_it_fini(const struct lu_env *env, struct dt_it *di)
435 struct lod_it *it = (struct lod_it *)di;
436 struct lod_object *lo = lod_dt_obj(it->lit_obj);
437 struct dt_object *next;
439 /* If lit_it == NULL, then it means the sub_it has been finished,
440 * which only happens in failure cases, see lod_striped_it_next() */
441 if (it->lit_it != NULL) {
442 LOD_CHECK_STRIPED_IT(env, it, lo);
444 next = lo->ldo_stripe[it->lit_stripe_index];
446 LASSERT(next->do_index_ops != NULL);
447 next->do_index_ops->dio_it.fini(env, it->lit_it);
451 /* the iterator not in use any more */
454 it->lit_stripe_index = 0;
458 * Implementation of dt_it_ops::get.
460 * Right now it's not used widely, only to reset the iterator to the
461 * initial position. It should be possible to implement a full version
462 * which chooses a correct stripe to be able to position with any key.
464 * \see dt_it_ops::get() in the API description for details.
466 static int lod_striped_it_get(const struct lu_env *env, struct dt_it *di,
467 const struct dt_key *key)
469 const struct lod_it *it = (const struct lod_it *)di;
470 struct lod_object *lo = lod_dt_obj(it->lit_obj);
471 struct dt_object *next;
473 LOD_CHECK_STRIPED_IT(env, it, lo);
475 next = lo->ldo_stripe[it->lit_stripe_index];
476 LASSERT(next != NULL);
477 LASSERT(dt_object_exists(next));
478 LASSERT(next->do_index_ops != NULL);
480 return next->do_index_ops->dio_it.get(env, it->lit_it, key);
484 * Implementation of dt_it_ops::put.
486 * Used with striped objects.
488 * \see dt_it_ops::put() in the API description for details.
490 static void lod_striped_it_put(const struct lu_env *env, struct dt_it *di)
492 struct lod_it *it = (struct lod_it *)di;
493 struct lod_object *lo = lod_dt_obj(it->lit_obj);
494 struct dt_object *next;
497 * If lit_it == NULL, then it means the sub_it has been finished,
498 * which only happens in failure cases, see lod_striped_it_next()
503 LOD_CHECK_STRIPED_IT(env, it, lo);
505 next = lo->ldo_stripe[it->lit_stripe_index];
506 LASSERT(next != NULL);
507 LASSERT(next->do_index_ops != NULL);
509 return next->do_index_ops->dio_it.put(env, it->lit_it);
513 * Implementation of dt_it_ops::next.
515 * Used with striped objects. When the end of the current stripe is
516 * reached, the method takes the next stripe's iterator.
518 * \see dt_it_ops::next() in the API description for details.
520 static int lod_striped_it_next(const struct lu_env *env, struct dt_it *di)
522 struct lod_it *it = (struct lod_it *)di;
523 struct lod_object *lo = lod_dt_obj(it->lit_obj);
524 struct dt_object *next;
525 struct dt_it *it_next;
531 LOD_CHECK_STRIPED_IT(env, it, lo);
533 next = lo->ldo_stripe[it->lit_stripe_index];
534 LASSERT(next != NULL);
535 LASSERT(dt_object_exists(next));
536 LASSERT(next->do_index_ops != NULL);
538 rc = next->do_index_ops->dio_it.next(env, it->lit_it);
542 if (rc == 0 && it->lit_stripe_index == 0)
545 if (rc == 0 && it->lit_stripe_index > 0) {
546 struct lu_dirent *ent;
548 ent = (struct lu_dirent *)lod_env_info(env)->lti_key;
550 rc = next->do_index_ops->dio_it.rec(env, it->lit_it,
551 (struct dt_rec *)ent,
556 /* skip . and .. for slave stripe */
557 if ((strncmp(ent->lde_name, ".",
558 le16_to_cpu(ent->lde_namelen)) == 0 &&
559 le16_to_cpu(ent->lde_namelen) == 1) ||
560 (strncmp(ent->lde_name, "..",
561 le16_to_cpu(ent->lde_namelen)) == 0 &&
562 le16_to_cpu(ent->lde_namelen) == 2))
568 next->do_index_ops->dio_it.put(env, it->lit_it);
569 next->do_index_ops->dio_it.fini(env, it->lit_it);
572 /* go to next stripe */
573 index = it->lit_stripe_index;
574 while (++index < lo->ldo_dir_stripe_count) {
575 next = lo->ldo_stripe[index];
579 if (!dt_object_exists(next))
582 rc = next->do_ops->do_index_try(env, next,
583 &dt_directory_features);
587 LASSERT(next->do_index_ops != NULL);
589 it_next = next->do_index_ops->dio_it.init(env, next,
592 RETURN(PTR_ERR(it_next));
594 rc = next->do_index_ops->dio_it.get(env, it_next,
595 (const struct dt_key *)"");
597 RETURN(rc == 0 ? -EIO : rc);
599 it->lit_it = it_next;
600 it->lit_stripe_index = index;
609 * Implementation of dt_it_ops::key.
611 * Used with striped objects.
613 * \see dt_it_ops::key() in the API description for details.
615 static struct dt_key *lod_striped_it_key(const struct lu_env *env,
616 const struct dt_it *di)
618 const struct lod_it *it = (const struct lod_it *)di;
619 struct lod_object *lo = lod_dt_obj(it->lit_obj);
620 struct dt_object *next;
622 LOD_CHECK_STRIPED_IT(env, it, lo);
624 next = lo->ldo_stripe[it->lit_stripe_index];
625 LASSERT(next != NULL);
626 LASSERT(next->do_index_ops != NULL);
628 return next->do_index_ops->dio_it.key(env, it->lit_it);
632 * Implementation of dt_it_ops::key_size.
634 * Used with striped objects.
636 * \see dt_it_ops::size() in the API description for details.
638 static int lod_striped_it_key_size(const struct lu_env *env,
639 const struct dt_it *di)
641 struct lod_it *it = (struct lod_it *)di;
642 struct lod_object *lo = lod_dt_obj(it->lit_obj);
643 struct dt_object *next;
645 LOD_CHECK_STRIPED_IT(env, it, lo);
647 next = lo->ldo_stripe[it->lit_stripe_index];
648 LASSERT(next != NULL);
649 LASSERT(next->do_index_ops != NULL);
651 return next->do_index_ops->dio_it.key_size(env, it->lit_it);
655 * Implementation of dt_it_ops::rec.
657 * Used with striped objects.
659 * \see dt_it_ops::rec() in the API description for details.
661 static int lod_striped_it_rec(const struct lu_env *env, const struct dt_it *di,
662 struct dt_rec *rec, __u32 attr)
664 const struct lod_it *it = (const struct lod_it *)di;
665 struct lod_object *lo = lod_dt_obj(it->lit_obj);
666 struct dt_object *next;
668 LOD_CHECK_STRIPED_IT(env, it, lo);
670 next = lo->ldo_stripe[it->lit_stripe_index];
671 LASSERT(next != NULL);
672 LASSERT(next->do_index_ops != NULL);
674 return next->do_index_ops->dio_it.rec(env, it->lit_it, rec, attr);
678 * Implementation of dt_it_ops::rec_size.
680 * Used with striped objects.
682 * \see dt_it_ops::rec_size() in the API description for details.
684 static int lod_striped_it_rec_size(const struct lu_env *env,
685 const struct dt_it *di, __u32 attr)
687 struct lod_it *it = (struct lod_it *)di;
688 struct lod_object *lo = lod_dt_obj(it->lit_obj);
689 struct dt_object *next;
691 LOD_CHECK_STRIPED_IT(env, it, lo);
693 next = lo->ldo_stripe[it->lit_stripe_index];
694 LASSERT(next != NULL);
695 LASSERT(next->do_index_ops != NULL);
697 return next->do_index_ops->dio_it.rec_size(env, it->lit_it, attr);
701 * Implementation of dt_it_ops::store.
703 * Used with striped objects.
705 * \see dt_it_ops::store() in the API description for details.
707 static __u64 lod_striped_it_store(const struct lu_env *env,
708 const struct dt_it *di)
710 const struct lod_it *it = (const struct lod_it *)di;
711 struct lod_object *lo = lod_dt_obj(it->lit_obj);
712 struct dt_object *next;
714 LOD_CHECK_STRIPED_IT(env, it, lo);
716 next = lo->ldo_stripe[it->lit_stripe_index];
717 LASSERT(next != NULL);
718 LASSERT(next->do_index_ops != NULL);
720 return next->do_index_ops->dio_it.store(env, it->lit_it);
724 * Implementation of dt_it_ops::load.
726 * Used with striped objects.
728 * \see dt_it_ops::load() in the API description for details.
730 static int lod_striped_it_load(const struct lu_env *env,
731 const struct dt_it *di, __u64 hash)
733 const struct lod_it *it = (const struct lod_it *)di;
734 struct lod_object *lo = lod_dt_obj(it->lit_obj);
735 struct dt_object *next;
737 LOD_CHECK_STRIPED_IT(env, it, lo);
739 next = lo->ldo_stripe[it->lit_stripe_index];
740 LASSERT(next != NULL);
741 LASSERT(next->do_index_ops != NULL);
743 return next->do_index_ops->dio_it.load(env, it->lit_it, hash);
746 static struct dt_index_operations lod_striped_index_ops = {
747 .dio_lookup = lod_lookup,
748 .dio_declare_insert = lod_declare_insert,
749 .dio_insert = lod_insert,
750 .dio_declare_delete = lod_declare_delete,
751 .dio_delete = lod_delete,
753 .init = lod_striped_it_init,
754 .fini = lod_striped_it_fini,
755 .get = lod_striped_it_get,
756 .put = lod_striped_it_put,
757 .next = lod_striped_it_next,
758 .key = lod_striped_it_key,
759 .key_size = lod_striped_it_key_size,
760 .rec = lod_striped_it_rec,
761 .rec_size = lod_striped_it_rec_size,
762 .store = lod_striped_it_store,
763 .load = lod_striped_it_load,
768 * Append the FID for each shard of the striped directory after the
769 * given LMV EA header.
771 * To simplify striped directory and the consistency verification,
772 * we only store the LMV EA header on disk, for both master object
773 * and slave objects. When someone wants to know the whole LMV EA,
774 * such as client readdir(), we can build the entrie LMV EA on the
775 * MDT side (in RAM) via iterating the sub-directory entries that
776 * are contained in the master object of the stripe directory.
778 * For the master object of the striped directroy, the valid name
779 * for each shard is composed of the ${shard_FID}:${shard_idx}.
781 * There may be holes in the LMV EA if some shards' name entries
782 * are corrupted or lost.
784 * \param[in] env pointer to the thread context
785 * \param[in] lo pointer to the master object of the striped directory
786 * \param[in] buf pointer to the lu_buf which will hold the LMV EA
787 * \param[in] resize whether re-allocate the buffer if it is not big enough
789 * \retval positive size of the LMV EA
790 * \retval 0 for nothing to be loaded
791 * \retval negative error number on failure
793 int lod_load_lmv_shards(const struct lu_env *env, struct lod_object *lo,
794 struct lu_buf *buf, bool resize)
796 struct lu_dirent *ent =
797 (struct lu_dirent *)lod_env_info(env)->lti_key;
798 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
799 struct dt_object *obj = dt_object_child(&lo->ldo_obj);
800 struct lmv_mds_md_v1 *lmv1 = buf->lb_buf;
802 const struct dt_it_ops *iops;
804 __u32 magic = le32_to_cpu(lmv1->lmv_magic);
809 if (magic != LMV_MAGIC_V1)
812 stripes = le32_to_cpu(lmv1->lmv_stripe_count);
816 rc = lmv_mds_md_size(stripes, magic);
820 if (buf->lb_len < lmv1_size) {
829 lu_buf_alloc(buf, lmv1_size);
834 memcpy(buf->lb_buf, tbuf.lb_buf, tbuf.lb_len);
837 if (unlikely(!dt_try_as_dir(env, obj)))
840 memset(&lmv1->lmv_stripe_fids[0], 0, stripes * sizeof(struct lu_fid));
841 iops = &obj->do_index_ops->dio_it;
842 it = iops->init(env, obj, LUDA_64BITHASH);
846 rc = iops->load(env, it, 0);
848 rc = iops->next(env, it);
853 char name[FID_LEN + 2] = "";
858 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
864 fid_le_to_cpu(&fid, &ent->lde_fid);
865 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
866 if (ent->lde_name[0] == '.') {
867 if (ent->lde_namelen == 1)
870 if (ent->lde_namelen == 2 && ent->lde_name[1] == '.')
874 len = snprintf(name, sizeof(name),
875 DFID":", PFID(&ent->lde_fid));
876 /* The ent->lde_name is composed of ${FID}:${index} */
877 if (ent->lde_namelen < len + 1 ||
878 memcmp(ent->lde_name, name, len) != 0) {
879 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
880 "%s: invalid shard name %.*s with the FID "DFID
881 " for the striped directory "DFID", %s\n",
882 lod2obd(lod)->obd_name, ent->lde_namelen,
883 ent->lde_name, PFID(&fid),
884 PFID(lu_object_fid(&obj->do_lu)),
885 lod->lod_lmv_failout ? "failout" : "skip");
887 if (lod->lod_lmv_failout)
895 if (ent->lde_name[len] < '0' ||
896 ent->lde_name[len] > '9') {
897 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
898 "%s: invalid shard name %.*s with the "
899 "FID "DFID" for the striped directory "
901 lod2obd(lod)->obd_name, ent->lde_namelen,
902 ent->lde_name, PFID(&fid),
903 PFID(lu_object_fid(&obj->do_lu)),
904 lod->lod_lmv_failout ?
907 if (lod->lod_lmv_failout)
913 index = index * 10 + ent->lde_name[len++] - '0';
914 } while (len < ent->lde_namelen);
916 if (len == ent->lde_namelen) {
917 /* Out of LMV EA range. */
918 if (index >= stripes) {
919 CERROR("%s: the shard %.*s for the striped "
920 "directory "DFID" is out of the known "
921 "LMV EA range [0 - %u], failout\n",
922 lod2obd(lod)->obd_name, ent->lde_namelen,
924 PFID(lu_object_fid(&obj->do_lu)),
930 /* The slot has been occupied. */
931 if (!fid_is_zero(&lmv1->lmv_stripe_fids[index])) {
935 &lmv1->lmv_stripe_fids[index]);
936 CERROR("%s: both the shard "DFID" and "DFID
937 " for the striped directory "DFID
938 " claim the same LMV EA slot at the "
939 "index %d, failout\n",
940 lod2obd(lod)->obd_name,
941 PFID(&fid0), PFID(&fid),
942 PFID(lu_object_fid(&obj->do_lu)), index);
947 /* stored as LE mode */
948 lmv1->lmv_stripe_fids[index] = ent->lde_fid;
951 rc = iops->next(env, it);
958 RETURN(rc > 0 ? lmv_mds_md_size(stripes, magic) : rc);
962 * Implementation of dt_object_operations::do_index_try.
964 * \see dt_object_operations::do_index_try() in the API description for details.
966 static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
967 const struct dt_index_features *feat)
969 struct lod_object *lo = lod_dt_obj(dt);
970 struct dt_object *next = dt_object_child(dt);
974 LASSERT(next->do_ops);
975 LASSERT(next->do_ops->do_index_try);
977 rc = lod_striping_load(env, lo);
981 rc = next->do_ops->do_index_try(env, next, feat);
985 if (lo->ldo_dir_stripe_count > 0) {
988 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
989 if (!lo->ldo_stripe[i])
991 if (!dt_object_exists(lo->ldo_stripe[i]))
993 rc = lo->ldo_stripe[i]->do_ops->do_index_try(env,
994 lo->ldo_stripe[i], feat);
998 dt->do_index_ops = &lod_striped_index_ops;
1000 dt->do_index_ops = &lod_index_ops;
1007 * Implementation of dt_object_operations::do_read_lock.
1009 * \see dt_object_operations::do_read_lock() in the API description for details.
1011 static void lod_read_lock(const struct lu_env *env, struct dt_object *dt,
1014 dt_read_lock(env, dt_object_child(dt), role);
1018 * Implementation of dt_object_operations::do_write_lock.
1020 * \see dt_object_operations::do_write_lock() in the API description for
1023 static void lod_write_lock(const struct lu_env *env, struct dt_object *dt,
1026 dt_write_lock(env, dt_object_child(dt), role);
1030 * Implementation of dt_object_operations::do_read_unlock.
1032 * \see dt_object_operations::do_read_unlock() in the API description for
1035 static void lod_read_unlock(const struct lu_env *env, struct dt_object *dt)
1037 dt_read_unlock(env, dt_object_child(dt));
1041 * Implementation of dt_object_operations::do_write_unlock.
1043 * \see dt_object_operations::do_write_unlock() in the API description for
1046 static void lod_write_unlock(const struct lu_env *env, struct dt_object *dt)
1048 dt_write_unlock(env, dt_object_child(dt));
1052 * Implementation of dt_object_operations::do_write_locked.
1054 * \see dt_object_operations::do_write_locked() in the API description for
1057 static int lod_write_locked(const struct lu_env *env, struct dt_object *dt)
1059 return dt_write_locked(env, dt_object_child(dt));
1063 * Implementation of dt_object_operations::do_attr_get.
1065 * \see dt_object_operations::do_attr_get() in the API description for details.
1067 static int lod_attr_get(const struct lu_env *env,
1068 struct dt_object *dt,
1069 struct lu_attr *attr)
1071 /* Note: for striped directory, client will merge attributes
1072 * from all of the sub-stripes see lmv_merge_attr(), and there
1073 * no MDD logic depend on directory nlink/size/time, so we can
1074 * always use master inode nlink and size for now. */
1075 return dt_attr_get(env, dt_object_child(dt), attr);
1078 static inline void lod_adjust_stripe_info(struct lod_layout_component *comp,
1079 struct lov_desc *desc,
1082 if (comp->llc_pattern != LOV_PATTERN_MDT) {
1083 if (append_stripes) {
1084 comp->llc_stripe_count = append_stripes;
1085 } else if (!comp->llc_stripe_count) {
1086 comp->llc_stripe_count =
1087 desc->ld_default_stripe_count;
1090 if (comp->llc_stripe_size <= 0)
1091 comp->llc_stripe_size = desc->ld_default_stripe_size;
1094 int lod_obj_for_each_stripe(const struct lu_env *env, struct lod_object *lo,
1096 struct lod_obj_stripe_cb_data *data)
1098 struct lod_layout_component *lod_comp;
1102 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1103 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1104 lod_comp = &lo->ldo_comp_entries[i];
1106 if (lod_comp->llc_stripe == NULL)
1109 /* has stripe but not inited yet, this component has been
1110 * declared to be created, but hasn't created yet.
1112 if (!lod_comp_inited(lod_comp))
1115 if (data->locd_comp_skip_cb &&
1116 data->locd_comp_skip_cb(env, lo, i, data))
1119 if (data->locd_comp_cb) {
1120 rc = data->locd_comp_cb(env, lo, i, data);
1125 /* could used just to do sth about component, not each
1128 if (!data->locd_stripe_cb)
1131 LASSERT(lod_comp->llc_stripe_count > 0);
1132 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
1133 struct dt_object *dt = lod_comp->llc_stripe[j];
1137 rc = data->locd_stripe_cb(env, lo, dt, th, i, j, data);
1145 static bool lod_obj_attr_set_comp_skip_cb(const struct lu_env *env,
1146 struct lod_object *lo, int comp_idx,
1147 struct lod_obj_stripe_cb_data *data)
1149 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[comp_idx];
1150 bool skipped = false;
1152 if (!(data->locd_attr->la_valid & LA_LAYOUT_VERSION))
1155 switch (lo->ldo_flr_state) {
1156 case LCM_FL_WRITE_PENDING: {
1159 /* skip stale components */
1160 if (lod_comp->llc_flags & LCME_FL_STALE) {
1165 /* skip valid and overlapping components, therefore any
1166 * attempts to write overlapped components will never succeed
1167 * because client will get EINPROGRESS. */
1168 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1172 if (lo->ldo_comp_entries[i].llc_flags & LCME_FL_STALE)
1175 if (lu_extent_is_overlapped(&lod_comp->llc_extent,
1176 &lo->ldo_comp_entries[i].llc_extent)) {
1184 LASSERTF(0, "impossible: %d\n", lo->ldo_flr_state);
1185 case LCM_FL_SYNC_PENDING:
1189 CDEBUG(D_LAYOUT, DFID": %s to set component %x to version: %u\n",
1190 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
1191 skipped ? "skipped" : "chose", lod_comp->llc_id,
1192 data->locd_attr->la_layout_version);
1198 lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
1199 struct dt_object *dt, struct thandle *th,
1200 int comp_idx, int stripe_idx,
1201 struct lod_obj_stripe_cb_data *data)
1203 if (data->locd_declare)
1204 return lod_sub_declare_attr_set(env, dt, data->locd_attr, th);
1206 if (data->locd_attr->la_valid & LA_LAYOUT_VERSION) {
1207 CDEBUG(D_LAYOUT, DFID": set layout version: %u, comp_idx: %d\n",
1208 PFID(lu_object_fid(&dt->do_lu)),
1209 data->locd_attr->la_layout_version, comp_idx);
1212 return lod_sub_attr_set(env, dt, data->locd_attr, th);
1216 * Implementation of dt_object_operations::do_declare_attr_set.
1218 * If the object is striped, then apply the changes to all the stripes.
1220 * \see dt_object_operations::do_declare_attr_set() in the API description
1223 static int lod_declare_attr_set(const struct lu_env *env,
1224 struct dt_object *dt,
1225 const struct lu_attr *attr,
1228 struct dt_object *next = dt_object_child(dt);
1229 struct lod_object *lo = lod_dt_obj(dt);
1234 * declare setattr on the local object
1236 rc = lod_sub_declare_attr_set(env, next, attr, th);
1240 /* osp_declare_attr_set() ignores all attributes other than
1241 * UID, GID, PROJID, and size, and osp_attr_set() ignores all
1242 * but UID, GID and PROJID. Declaration of size attr setting
1243 * happens through lod_declare_init_size(), and not through
1244 * this function. Therefore we need not load striping unless
1245 * ownership is changing. This should save memory and (we hope)
1246 * speed up rename().
1248 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1249 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1252 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1255 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID | LA_MODE |
1256 LA_ATIME | LA_MTIME | LA_CTIME |
1261 * load striping information, notice we don't do this when object
1262 * is being initialized as we don't need this information till
1263 * few specific cases like destroy, chown
1265 rc = lod_striping_load(env, lo);
1269 if (!lod_obj_is_striped(dt))
1273 * if object is striped declare changes on the stripes
1275 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1276 LASSERT(lo->ldo_stripe);
1277 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1278 if (lo->ldo_stripe[i] == NULL)
1280 if (!dt_object_exists(lo->ldo_stripe[i]))
1282 rc = lod_sub_declare_attr_set(env, lo->ldo_stripe[i],
1288 struct lod_obj_stripe_cb_data data = { { 0 } };
1290 data.locd_attr = attr;
1291 data.locd_declare = true;
1292 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1293 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1299 if (!dt_object_exists(next) || dt_object_remote(next) ||
1300 !S_ISREG(attr->la_mode))
1303 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1304 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
1308 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) ||
1309 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1310 struct lod_thread_info *info = lod_env_info(env);
1311 struct lu_buf *buf = &info->lti_buf;
1313 buf->lb_buf = info->lti_ea_store;
1314 buf->lb_len = info->lti_ea_store_size;
1315 rc = lod_sub_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1316 LU_XATTR_REPLACE, th);
1323 * Implementation of dt_object_operations::do_attr_set.
1325 * If the object is striped, then apply the changes to all or subset of
1326 * the stripes depending on the object type and specific attributes.
1328 * \see dt_object_operations::do_attr_set() in the API description for details.
1330 static int lod_attr_set(const struct lu_env *env,
1331 struct dt_object *dt,
1332 const struct lu_attr *attr,
1335 struct dt_object *next = dt_object_child(dt);
1336 struct lod_object *lo = lod_dt_obj(dt);
1341 * apply changes to the local object
1343 rc = lod_sub_attr_set(env, next, attr, th);
1347 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1348 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1351 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1354 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE | LA_PROJID |
1355 LA_ATIME | LA_MTIME | LA_CTIME |
1360 /* FIXME: a tricky case in the code path of mdd_layout_change():
1361 * the in-memory striping information has been freed in lod_xattr_set()
1362 * due to layout change. It has to load stripe here again. It only
1363 * changes flags of layout so declare_attr_set() is still accurate */
1364 rc = lod_striping_load(env, lo);
1368 if (!lod_obj_is_striped(dt))
1372 * if object is striped, apply changes to all the stripes
1374 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1375 LASSERT(lo->ldo_stripe);
1376 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1377 if (unlikely(lo->ldo_stripe[i] == NULL))
1380 if ((dt_object_exists(lo->ldo_stripe[i]) == 0))
1383 rc = lod_sub_attr_set(env, lo->ldo_stripe[i], attr, th);
1388 struct lod_obj_stripe_cb_data data = { { 0 } };
1390 data.locd_attr = attr;
1391 data.locd_declare = false;
1392 data.locd_comp_skip_cb = lod_obj_attr_set_comp_skip_cb;
1393 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1394 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1400 if (!dt_object_exists(next) || dt_object_remote(next) ||
1401 !S_ISREG(attr->la_mode))
1404 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1405 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
1409 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE)) {
1410 struct lod_thread_info *info = lod_env_info(env);
1411 struct lu_buf *buf = &info->lti_buf;
1412 struct ost_id *oi = &info->lti_ostid;
1413 struct lu_fid *fid = &info->lti_fid;
1414 struct lov_mds_md_v1 *lmm;
1415 struct lov_ost_data_v1 *objs;
1418 rc = lod_get_lov_ea(env, lo);
1422 buf->lb_buf = info->lti_ea_store;
1423 buf->lb_len = info->lti_ea_store_size;
1424 lmm = info->lti_ea_store;
1425 magic = le32_to_cpu(lmm->lmm_magic);
1426 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1427 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1428 struct lov_comp_md_entry_v1 *lcme =
1429 &lcm->lcm_entries[0];
1431 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
1432 magic = le32_to_cpu(lmm->lmm_magic);
1435 if (magic == LOV_MAGIC_V1)
1436 objs = &(lmm->lmm_objects[0]);
1438 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1439 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1440 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1442 fid_to_ostid(fid, oi);
1443 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1445 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1446 LU_XATTR_REPLACE, th);
1447 } else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1448 struct lod_thread_info *info = lod_env_info(env);
1449 struct lu_buf *buf = &info->lti_buf;
1450 struct lov_comp_md_v1 *lcm;
1451 struct lov_comp_md_entry_v1 *lcme;
1453 rc = lod_get_lov_ea(env, lo);
1457 buf->lb_buf = info->lti_ea_store;
1458 buf->lb_len = info->lti_ea_store_size;
1460 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1 &&
1461 le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_SEL)
1464 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1465 lcme = &lcm->lcm_entries[0];
1466 le64_add_cpu(&lcme->lcme_extent.e_start, 1);
1467 le64_add_cpu(&lcme->lcme_extent.e_end, -1);
1469 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1470 LU_XATTR_REPLACE, th);
1477 * Implementation of dt_object_operations::do_xattr_get.
1479 * If LOV EA is requested from the root object and it's not
1480 * found, then return default striping for the filesystem.
1482 * \see dt_object_operations::do_xattr_get() in the API description for details.
1484 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1485 struct lu_buf *buf, const char *name)
1487 struct lod_thread_info *info = lod_env_info(env);
1488 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1493 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1494 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1495 struct lmv_mds_md_v1 *lmv1;
1496 struct lmv_foreign_md *lfm;
1499 if (rc > (typeof(rc))sizeof(*lmv1))
1502 /* short (<= sizeof(struct lmv_mds_md_v1)) foreign LMV case */
1503 /* XXX empty foreign LMV is not allowed */
1504 if (rc <= offsetof(typeof(*lfm), lfm_value))
1505 RETURN(rc = rc > 0 ? -EINVAL : rc);
1507 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1508 CLASSERT(sizeof(*lmv1) <= sizeof(info->lti_key));
1510 /* lti_buf is large enough for *lmv1 or a short
1511 * (<= sizeof(struct lmv_mds_md_v1)) foreign LMV
1513 info->lti_buf.lb_buf = info->lti_key;
1514 info->lti_buf.lb_len = sizeof(*lmv1);
1515 rc = dt_xattr_get(env, dt_object_child(dt),
1516 &info->lti_buf, name);
1517 if (unlikely(rc <= offsetof(typeof(*lfm),
1519 RETURN(rc = rc > 0 ? -EINVAL : rc);
1521 lfm = info->lti_buf.lb_buf;
1522 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN)
1525 if (unlikely(rc != sizeof(*lmv1)))
1526 RETURN(rc = rc > 0 ? -EINVAL : rc);
1528 lmv1 = info->lti_buf.lb_buf;
1529 /* The on-disk LMV EA only contains header, but the
1530 * returned LMV EA size should contain the space for
1531 * the FIDs of all shards of the striped directory. */
1532 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1533 rc = lmv_mds_md_size(
1534 le32_to_cpu(lmv1->lmv_stripe_count),
1538 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN)
1541 if (rc != sizeof(*lmv1))
1542 RETURN(rc = rc > 0 ? -EINVAL : rc);
1544 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1548 RETURN(rc = rc1 != 0 ? rc1 : rc);
1551 if ((rc > 0) && buf->lb_buf && strcmp(name, XATTR_NAME_LOV) == 0) {
1552 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1554 if (lcm->lcm_magic == cpu_to_le32(LOV_MAGIC_SEL))
1555 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
1558 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1562 * XXX: Only used by lfsck
1564 * lod returns default striping on the real root of the device
1565 * this is like the root stores default striping for the whole
1566 * filesystem. historically we've been using a different approach
1567 * and store it in the config.
1569 dt_root_get(env, dev->lod_child, &info->lti_fid);
1570 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1572 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1573 struct lov_user_md *lum = buf->lb_buf;
1574 struct lov_desc *desc = &dev->lod_ost_descs.ltd_lov_desc;
1576 if (buf->lb_buf == NULL) {
1578 } else if (buf->lb_len >= sizeof(*lum)) {
1579 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1580 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1581 lmm_oi_set_id(&lum->lmm_oi, 0);
1582 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1583 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1584 lum->lmm_stripe_size = cpu_to_le32(
1585 desc->ld_default_stripe_size);
1586 lum->lmm_stripe_count = cpu_to_le16(
1587 desc->ld_default_stripe_count);
1588 lum->lmm_stripe_offset = cpu_to_le16(
1589 desc->ld_default_stripe_offset);
1602 * Checks that the magic of the stripe is sane.
1604 * \param[in] lod lod device
1605 * \param[in] lum a buffer storing LMV EA to verify
1607 * \retval 0 if the EA is sane
1608 * \retval negative otherwise
1610 static int lod_verify_md_striping(struct lod_device *lod,
1611 const struct lmv_user_md_v1 *lum)
1613 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1614 CERROR("%s: invalid lmv_user_md: magic = %x, "
1615 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1616 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1617 (int)le32_to_cpu(lum->lum_stripe_offset),
1618 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1626 * Initialize LMV EA for a slave.
1628 * Initialize slave's LMV EA from the master's LMV EA.
1630 * \param[in] master_lmv a buffer containing master's EA
1631 * \param[out] slave_lmv a buffer where slave's EA will be stored
1634 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1635 const struct lmv_mds_md_v1 *master_lmv)
1637 *slave_lmv = *master_lmv;
1638 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1644 * Generate LMV EA from the object passed as \a dt. The object must have
1645 * the stripes created and initialized.
1647 * \param[in] env execution environment
1648 * \param[in] dt object
1649 * \param[out] lmv_buf buffer storing generated LMV EA
1651 * \retval 0 on success
1652 * \retval negative if failed
1654 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1655 struct lu_buf *lmv_buf)
1657 struct lod_thread_info *info = lod_env_info(env);
1658 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1659 struct lod_object *lo = lod_dt_obj(dt);
1660 struct lmv_mds_md_v1 *lmm1;
1662 int type = LU_SEQ_RANGE_ANY;
1667 LASSERT(lo->ldo_dir_striped != 0);
1668 LASSERT(lo->ldo_dir_stripe_count > 0);
1669 stripe_count = lo->ldo_dir_stripe_count;
1670 /* Only store the LMV EA heahder on the disk. */
1671 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1672 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1676 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1679 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1680 memset(lmm1, 0, sizeof(*lmm1));
1681 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1682 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1683 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1684 if (lo->ldo_dir_hash_type & LMV_HASH_FLAG_MIGRATION) {
1685 lmm1->lmv_migrate_hash = cpu_to_le32(lo->ldo_dir_migrate_hash);
1686 lmm1->lmv_migrate_offset =
1687 cpu_to_le32(lo->ldo_dir_migrate_offset);
1689 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1694 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1695 lmv_buf->lb_buf = info->lti_ea_store;
1696 lmv_buf->lb_len = sizeof(*lmm1);
1702 * Create in-core represenation for a striped directory.
1704 * Parse the buffer containing LMV EA and instantiate LU objects
1705 * representing the stripe objects. The pointers to the objects are
1706 * stored in ldo_stripe field of \a lo. This function is used when
1707 * we need to access an already created object (i.e. load from a disk).
1709 * \param[in] env execution environment
1710 * \param[in] lo lod object
1711 * \param[in] buf buffer containing LMV EA
1713 * \retval 0 on success
1714 * \retval negative if failed
1716 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1717 const struct lu_buf *buf)
1719 struct lod_thread_info *info = lod_env_info(env);
1720 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1721 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1722 struct dt_object **stripe;
1723 union lmv_mds_md *lmm = buf->lb_buf;
1724 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1725 struct lu_fid *fid = &info->lti_fid;
1730 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1732 /* XXX may be useless as not called for foreign LMV ?? */
1733 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_FOREIGN)
1736 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1737 lo->ldo_dir_slave_stripe = 1;
1741 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1744 if (le32_to_cpu(lmv1->lmv_stripe_count) < 1)
1747 LASSERT(lo->ldo_stripe == NULL);
1748 OBD_ALLOC(stripe, sizeof(stripe[0]) *
1749 (le32_to_cpu(lmv1->lmv_stripe_count)));
1753 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1754 struct dt_device *tgt_dt;
1755 struct dt_object *dto;
1756 int type = LU_SEQ_RANGE_ANY;
1759 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1760 if (!fid_is_sane(fid)) {
1765 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1769 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1770 tgt_dt = lod->lod_child;
1772 struct lod_tgt_desc *tgt;
1774 tgt = LTD_TGT(ltd, idx);
1776 GOTO(out, rc = -ESTALE);
1777 tgt_dt = tgt->ltd_tgt;
1780 dto = dt_locate_at(env, tgt_dt, fid,
1781 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1784 GOTO(out, rc = PTR_ERR(dto));
1789 lo->ldo_stripe = stripe;
1790 lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1791 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1793 lod_striping_free_nolock(env, lo);
1799 * Declare create a striped directory.
1801 * Declare creating a striped directory with a given stripe pattern on the
1802 * specified MDTs. A striped directory is represented as a regular directory
1803 * - an index listing all the stripes. The stripes point back to the master
1804 * object with ".." and LinkEA. The master object gets LMV EA which
1805 * identifies it as a striped directory. The function allocates FIDs
1808 * \param[in] env execution environment
1809 * \param[in] dt object
1810 * \param[in] attr attributes to initialize the objects with
1811 * \param[in] dof type of objects to be created
1812 * \param[in] th transaction handle
1814 * \retval 0 on success
1815 * \retval negative if failed
1817 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1818 struct dt_object *dt,
1819 struct lu_attr *attr,
1820 struct dt_object_format *dof,
1823 struct lod_thread_info *info = lod_env_info(env);
1824 struct lu_buf lmv_buf;
1825 struct lu_buf slave_lmv_buf;
1826 struct lmv_mds_md_v1 *lmm;
1827 struct lmv_mds_md_v1 *slave_lmm = NULL;
1828 struct dt_insert_rec *rec = &info->lti_dt_rec;
1829 struct lod_object *lo = lod_dt_obj(dt);
1834 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1837 lmm = lmv_buf.lb_buf;
1839 OBD_ALLOC_PTR(slave_lmm);
1840 if (slave_lmm == NULL)
1841 GOTO(out, rc = -ENOMEM);
1843 lod_prep_slave_lmv_md(slave_lmm, lmm);
1844 slave_lmv_buf.lb_buf = slave_lmm;
1845 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1847 if (!dt_try_as_dir(env, dt_object_child(dt)))
1848 GOTO(out, rc = -EINVAL);
1850 rec->rec_type = S_IFDIR;
1851 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1852 struct dt_object *dto = lo->ldo_stripe[i];
1853 char *stripe_name = info->lti_key;
1854 struct lu_name *sname;
1855 struct linkea_data ldata = { NULL };
1856 struct lu_buf linkea_buf;
1858 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
1862 rc = lod_sub_declare_create(env, dto, attr, NULL, dof, th);
1866 if (!dt_try_as_dir(env, dto))
1867 GOTO(out, rc = -EINVAL);
1869 rc = lod_sub_declare_ref_add(env, dto, th);
1873 rec->rec_fid = lu_object_fid(&dto->do_lu);
1874 rc = lod_sub_declare_insert(env, dto,
1875 (const struct dt_rec *)rec,
1876 (const struct dt_key *)dot, th);
1880 /* master stripe FID will be put to .. */
1881 rec->rec_fid = lu_object_fid(&dt->do_lu);
1882 rc = lod_sub_declare_insert(env, dto,
1883 (const struct dt_rec *)rec,
1884 (const struct dt_key *)dotdot, th);
1888 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1889 cfs_fail_val != i) {
1890 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1892 slave_lmm->lmv_master_mdt_index =
1895 slave_lmm->lmv_master_mdt_index =
1897 rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
1898 XATTR_NAME_LMV, 0, th);
1903 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1905 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1906 PFID(lu_object_fid(&dto->do_lu)), i + 1);
1908 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1909 PFID(lu_object_fid(&dto->do_lu)), i);
1911 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
1912 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1913 sname, lu_object_fid(&dt->do_lu));
1917 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1918 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1919 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
1920 XATTR_NAME_LINK, 0, th);
1924 rec->rec_fid = lu_object_fid(&dto->do_lu);
1925 rc = lod_sub_declare_insert(env, dt_object_child(dt),
1926 (const struct dt_rec *)rec,
1927 (const struct dt_key *)stripe_name,
1932 rc = lod_sub_declare_ref_add(env, dt_object_child(dt), th);
1937 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
1938 &lmv_buf, XATTR_NAME_LMV, 0, th);
1942 if (slave_lmm != NULL)
1943 OBD_FREE_PTR(slave_lmm);
1948 static int lod_prep_md_striped_create(const struct lu_env *env,
1949 struct dt_object *dt,
1950 struct lu_attr *attr,
1951 const struct lmv_user_md_v1 *lum,
1952 struct dt_object_format *dof,
1955 struct lod_thread_info *info = lod_env_info(env);
1956 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1957 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1958 struct lod_object *lo = lod_dt_obj(dt);
1959 struct dt_object **stripe;
1966 bool is_specific = false;
1969 /* The lum has been verifed in lod_verify_md_striping */
1970 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1971 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
1973 stripe_count = lo->ldo_dir_stripe_count;
1975 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
1976 if (idx_array == NULL)
1979 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_count);
1981 GOTO(out_free, rc = -ENOMEM);
1983 /* Start index must be the master MDT */
1984 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1985 idx_array[0] = master_index;
1986 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
1988 for (i = 1; i < stripe_count; i++)
1989 idx_array[i] = le32_to_cpu(lum->lum_objects[i].lum_mds);
1992 for (i = 0; i < stripe_count; i++) {
1993 struct lod_tgt_desc *tgt = NULL;
1994 struct dt_object *dto;
1995 struct lu_fid fid = { 0 };
1997 struct lu_object_conf conf = { 0 };
1998 struct dt_device *tgt_dt = NULL;
2000 /* Try to find next avaible target */
2002 for (j = 0; j < lod->lod_remote_mdt_count;
2003 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
2004 bool already_allocated = false;
2007 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
2008 idx, lod->lod_remote_mdt_count + 1, i);
2010 if (likely(!is_specific &&
2011 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
2012 /* check whether the idx already exists
2013 * in current allocated array */
2014 for (k = 0; k < i; k++) {
2015 if (idx_array[k] == idx) {
2016 already_allocated = true;
2021 if (already_allocated)
2025 /* Sigh, this index is not in the bitmap, let's check
2026 * next available target */
2027 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) &&
2028 idx != master_index)
2031 if (idx == master_index) {
2032 /* Allocate the FID locally */
2033 rc = obd_fid_alloc(env, lod->lod_child_exp,
2037 tgt_dt = lod->lod_child;
2041 /* check the status of the OSP */
2042 tgt = LTD_TGT(ltd, idx);
2046 tgt_dt = tgt->ltd_tgt;
2047 rc = dt_statfs(env, tgt_dt, &info->lti_osfs);
2049 /* this OSP doesn't feel well */
2054 rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
2063 /* Can not allocate more stripes */
2064 if (j == lod->lod_remote_mdt_count) {
2065 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
2066 lod2obd(lod)->obd_name, stripe_count, i);
2070 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
2071 idx, i, PFID(&fid));
2073 /* Set the start index for next stripe allocation */
2074 if (!is_specific && i < stripe_count - 1) {
2076 * for large dir test, put all other slaves on one
2077 * remote MDT, otherwise we may save too many local
2078 * slave locks which will exceed RS_MAX_LOCKS.
2080 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)))
2082 idx_array[i + 1] = (idx + 1) %
2083 (lod->lod_remote_mdt_count + 1);
2085 /* tgt_dt and fid must be ready after search avaible OSP
2086 * in the above loop */
2087 LASSERT(tgt_dt != NULL);
2088 LASSERT(fid_is_sane(&fid));
2090 /* fail a remote stripe FID allocation */
2091 if (i && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_FID))
2094 conf.loc_flags = LOC_F_NEW;
2095 dto = dt_locate_at(env, tgt_dt, &fid,
2096 dt->do_lu.lo_dev->ld_site->ls_top_dev,
2099 GOTO(out_put, rc = PTR_ERR(dto));
2103 lo->ldo_dir_striped = 1;
2104 lo->ldo_stripe = stripe;
2105 lo->ldo_dir_stripe_count = i;
2106 lo->ldo_dir_stripes_allocated = stripe_count;
2108 lo->ldo_dir_stripe_loaded = 1;
2110 if (lo->ldo_dir_stripe_count == 0)
2111 GOTO(out_put, rc = -ENOSPC);
2113 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2119 for (i = 0; i < stripe_count; i++)
2120 if (stripe[i] != NULL)
2121 dt_object_put(env, stripe[i]);
2122 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_count);
2123 lo->ldo_dir_stripe_count = 0;
2124 lo->ldo_dir_stripes_allocated = 0;
2125 lo->ldo_stripe = NULL;
2129 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
2136 * Alloc cached foreign LMV
2138 * \param[in] lo object
2139 * \param[in] size size of foreign LMV
2141 * \retval 0 on success
2142 * \retval negative if failed
2144 int lod_alloc_foreign_lmv(struct lod_object *lo, size_t size)
2146 OBD_ALLOC_LARGE(lo->ldo_foreign_lmv, size);
2147 if (lo->ldo_foreign_lmv == NULL)
2149 lo->ldo_foreign_lmv_size = size;
2150 lo->ldo_dir_is_foreign = 1;
2156 * Declare create striped md object.
2158 * The function declares intention to create a striped directory. This is a
2159 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2160 * is to verify pattern \a lum_buf is good. Check that function for the details.
2162 * \param[in] env execution environment
2163 * \param[in] dt object
2164 * \param[in] attr attributes to initialize the objects with
2165 * \param[in] lum_buf a pattern specifying the number of stripes and
2167 * \param[in] dof type of objects to be created
2168 * \param[in] th transaction handle
2170 * \retval 0 on success
2171 * \retval negative if failed
2174 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2175 struct dt_object *dt,
2176 struct lu_attr *attr,
2177 const struct lu_buf *lum_buf,
2178 struct dt_object_format *dof,
2181 struct lod_object *lo = lod_dt_obj(dt);
2182 struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
2186 LASSERT(lum != NULL);
2188 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
2189 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
2190 (int)le32_to_cpu(lum->lum_stripe_offset));
2192 if (lo->ldo_dir_stripe_count == 0) {
2193 if (lo->ldo_dir_is_foreign) {
2194 rc = lod_alloc_foreign_lmv(lo, lum_buf->lb_len);
2197 memcpy(lo->ldo_foreign_lmv, lum, lum_buf->lb_len);
2198 lo->ldo_dir_stripe_loaded = 1;
2203 /* prepare dir striped objects */
2204 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2206 /* failed to create striping, let's reset
2207 * config so that others don't get confused */
2208 lod_striping_free(env, lo);
2216 * Append source stripes after target stripes for migrating directory. NB, we
2217 * only need to declare this, the append is done inside lod_xattr_set_lmv().
2219 * \param[in] env execution environment
2220 * \param[in] dt target object
2221 * \param[in] buf LMV buf which contains source stripe fids
2222 * \param[in] th transaction handle
2224 * \retval 0 on success
2225 * \retval negative if failed
2227 static int lod_dir_declare_layout_add(const struct lu_env *env,
2228 struct dt_object *dt,
2229 const struct lu_buf *buf,
2232 struct lod_thread_info *info = lod_env_info(env);
2233 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
2234 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
2235 struct lod_object *lo = lod_dt_obj(dt);
2236 struct dt_object *next = dt_object_child(dt);
2237 struct dt_object_format *dof = &info->lti_format;
2238 struct lmv_mds_md_v1 *lmv = buf->lb_buf;
2239 struct dt_object **stripe;
2240 __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
2241 struct lu_fid *fid = &info->lti_fid;
2242 struct lod_tgt_desc *tgt;
2243 struct dt_object *dto;
2244 struct dt_device *tgt_dt;
2245 int type = LU_SEQ_RANGE_ANY;
2246 struct dt_insert_rec *rec = &info->lti_dt_rec;
2247 char *stripe_name = info->lti_key;
2248 struct lu_name *sname;
2249 struct linkea_data ldata = { NULL };
2250 struct lu_buf linkea_buf;
2257 if (le32_to_cpu(lmv->lmv_magic) != LMV_MAGIC_V1)
2260 if (stripe_count == 0)
2263 dof->dof_type = DFT_DIR;
2266 sizeof(*stripe) * (lo->ldo_dir_stripe_count + stripe_count));
2270 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
2271 stripe[i] = lo->ldo_stripe[i];
2273 for (i = 0; i < stripe_count; i++) {
2275 &lmv->lmv_stripe_fids[i]);
2276 if (!fid_is_sane(fid))
2279 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
2283 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
2284 tgt_dt = lod->lod_child;
2286 tgt = LTD_TGT(ltd, idx);
2288 GOTO(out, rc = -ESTALE);
2289 tgt_dt = tgt->ltd_tgt;
2292 dto = dt_locate_at(env, tgt_dt, fid,
2293 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
2296 GOTO(out, rc = PTR_ERR(dto));
2298 stripe[i + lo->ldo_dir_stripe_count] = dto;
2300 if (!dt_try_as_dir(env, dto))
2301 GOTO(out, rc = -ENOTDIR);
2303 rc = lod_sub_declare_ref_add(env, dto, th);
2307 rc = lod_sub_declare_insert(env, dto,
2308 (const struct dt_rec *)rec,
2309 (const struct dt_key *)dot, th);
2313 rc = lod_sub_declare_insert(env, dto,
2314 (const struct dt_rec *)rec,
2315 (const struct dt_key *)dotdot, th);
2319 rc = lod_sub_declare_xattr_set(env, dto, buf,
2320 XATTR_NAME_LMV, 0, th);
2324 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
2325 PFID(lu_object_fid(&dto->do_lu)),
2326 i + lo->ldo_dir_stripe_count);
2328 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
2329 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
2330 sname, lu_object_fid(&dt->do_lu));
2334 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
2335 linkea_buf.lb_len = ldata.ld_leh->leh_len;
2336 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
2337 XATTR_NAME_LINK, 0, th);
2341 rc = lod_sub_declare_insert(env, next,
2342 (const struct dt_rec *)rec,
2343 (const struct dt_key *)stripe_name,
2348 rc = lod_sub_declare_ref_add(env, next, th);
2354 OBD_FREE(lo->ldo_stripe,
2355 sizeof(*stripe) * lo->ldo_dir_stripes_allocated);
2356 lo->ldo_stripe = stripe;
2357 lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count;
2358 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type);
2359 lo->ldo_dir_stripe_count += stripe_count;
2360 lo->ldo_dir_stripes_allocated += stripe_count;
2361 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_MIGRATION;
2365 i = lo->ldo_dir_stripe_count;
2366 while (i < lo->ldo_dir_stripe_count + stripe_count && stripe[i])
2367 dt_object_put(env, stripe[i++]);
2370 sizeof(*stripe) * (stripe_count + lo->ldo_dir_stripe_count));
2374 static int lod_dir_declare_layout_delete(const struct lu_env *env,
2375 struct dt_object *dt,
2376 const struct lu_buf *buf,
2379 struct lod_thread_info *info = lod_env_info(env);
2380 struct lod_object *lo = lod_dt_obj(dt);
2381 struct dt_object *next = dt_object_child(dt);
2382 struct lmv_user_md *lmu = buf->lb_buf;
2383 __u32 final_stripe_count;
2384 char *stripe_name = info->lti_key;
2385 struct dt_object *dto;
2392 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
2393 if (final_stripe_count >= lo->ldo_dir_stripe_count)
2396 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
2397 dto = lo->ldo_stripe[i];
2401 if (!dt_try_as_dir(env, dto))
2404 rc = lod_sub_declare_delete(env, dto,
2405 (const struct dt_key *)dot, th);
2409 rc = lod_sub_declare_ref_del(env, dto, th);
2413 rc = lod_sub_declare_delete(env, dto,
2414 (const struct dt_key *)dotdot, th);
2418 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2419 PFID(lu_object_fid(&dto->do_lu)), i);
2421 rc = lod_sub_declare_delete(env, next,
2422 (const struct dt_key *)stripe_name, th);
2426 rc = lod_sub_declare_ref_del(env, next, th);
2435 * delete stripes from dir master object, the lum_stripe_count in argument is
2436 * the final stripe count, the stripes after that will be deleted, NB, they
2437 * are not destroyed, but deleted from it's parent namespace, this function
2438 * will be called in two places:
2439 * 1. mdd_migrate_create() delete stripes from source, and append them to
2441 * 2. mdd_dir_layout_shrink() delete stripes from source, and destroy them.
2443 static int lod_dir_layout_delete(const struct lu_env *env,
2444 struct dt_object *dt,
2445 const struct lu_buf *buf,
2448 struct lod_thread_info *info = lod_env_info(env);
2449 struct lod_object *lo = lod_dt_obj(dt);
2450 struct dt_object *next = dt_object_child(dt);
2451 struct lmv_user_md *lmu = buf->lb_buf;
2452 __u32 final_stripe_count;
2453 char *stripe_name = info->lti_key;
2454 struct dt_object *dto;
2463 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
2464 if (final_stripe_count >= lo->ldo_dir_stripe_count)
2467 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
2468 dto = lo->ldo_stripe[i];
2472 rc = lod_sub_delete(env, dto,
2473 (const struct dt_key *)dotdot, th);
2477 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2478 PFID(lu_object_fid(&dto->do_lu)), i);
2480 rc = lod_sub_delete(env, next,
2481 (const struct dt_key *)stripe_name, th);
2485 rc = lod_sub_ref_del(env, next, th);
2490 lod_striping_free(env, lod_dt_obj(dt));
2496 * Implementation of dt_object_operations::do_declare_xattr_set.
2498 * Used with regular (non-striped) objects. Basically it
2499 * initializes the striping information and applies the
2500 * change to all the stripes.
2502 * \see dt_object_operations::do_declare_xattr_set() in the API description
2505 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2506 struct dt_object *dt,
2507 const struct lu_buf *buf,
2508 const char *name, int fl,
2511 struct dt_object *next = dt_object_child(dt);
2512 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2513 struct lod_object *lo = lod_dt_obj(dt);
2518 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2519 struct lmv_user_md_v1 *lum;
2521 LASSERT(buf != NULL && buf->lb_buf != NULL);
2523 rc = lod_verify_md_striping(d, lum);
2526 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2527 rc = lod_verify_striping(d, lo, buf, false);
2532 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2536 /* Note: Do not set LinkEA on sub-stripes, otherwise
2537 * it will confuse the fid2path process(see mdt_path_current()).
2538 * The linkEA between master and sub-stripes is set in
2539 * lod_xattr_set_lmv(). */
2540 if (strcmp(name, XATTR_NAME_LINK) == 0)
2543 /* set xattr to each stripes, if needed */
2544 rc = lod_striping_load(env, lo);
2548 if (lo->ldo_dir_stripe_count == 0)
2551 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2552 if (!lo->ldo_stripe[i])
2555 if (!dt_object_exists(lo->ldo_stripe[i]))
2558 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2568 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2569 struct lod_object *lo,
2570 struct dt_object *dt, struct thandle *th,
2571 int comp_idx, int stripe_idx,
2572 struct lod_obj_stripe_cb_data *data)
2574 struct lod_thread_info *info = lod_env_info(env);
2575 struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
2576 struct filter_fid *ff = &info->lti_ff;
2577 struct lu_buf *buf = &info->lti_buf;
2581 buf->lb_len = sizeof(*ff);
2582 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2590 * locd_buf is set if it's called by dir migration, which doesn't check
2593 if (data->locd_buf) {
2594 memset(ff, 0, sizeof(*ff));
2595 ff->ff_parent = *(struct lu_fid *)data->locd_buf->lb_buf;
2597 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
2599 if (lu_fid_eq(lod_object_fid(lo), &ff->ff_parent) &&
2600 ff->ff_layout.ol_comp_id == comp->llc_id)
2603 memset(ff, 0, sizeof(*ff));
2604 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2607 /* rewrite filter_fid */
2608 ff->ff_parent.f_ver = stripe_idx;
2609 ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
2610 ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
2611 ff->ff_layout.ol_comp_id = comp->llc_id;
2612 ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
2613 ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
2614 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
2616 if (data->locd_declare)
2617 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2618 LU_XATTR_REPLACE, th);
2620 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2621 LU_XATTR_REPLACE, th);
2627 * Reset parent FID on OST object
2629 * Replace parent FID with @dt object FID, which is only called during migration
2630 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2631 * the FID is changed.
2633 * \param[in] env execution environment
2634 * \param[in] dt dt_object whose stripes's parent FID will be reset
2635 * \parem[in] th thandle
2636 * \param[in] declare if it is declare
2638 * \retval 0 if reset succeeds
2639 * \retval negative errno if reset fails
2641 static int lod_replace_parent_fid(const struct lu_env *env,
2642 struct dt_object *dt,
2643 const struct lu_buf *buf,
2644 struct thandle *th, bool declare)
2646 struct lod_object *lo = lod_dt_obj(dt);
2647 struct lod_thread_info *info = lod_env_info(env);
2648 struct filter_fid *ff;
2649 struct lod_obj_stripe_cb_data data = { { 0 } };
2653 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2655 /* set xattr to each stripes, if needed */
2656 rc = lod_striping_load(env, lo);
2660 if (!lod_obj_is_striped(dt))
2663 if (info->lti_ea_store_size < sizeof(*ff)) {
2664 rc = lod_ea_store_resize(info, sizeof(*ff));
2669 data.locd_declare = declare;
2670 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2671 data.locd_buf = buf;
2672 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2677 inline __u16 lod_comp_entry_stripe_count(struct lod_object *lo,
2678 struct lod_layout_component *entry,
2681 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2685 else if (lod_comp_inited(entry))
2686 return entry->llc_stripe_count;
2687 else if ((__u16)-1 == entry->llc_stripe_count)
2688 return lod->lod_ost_count;
2690 return lod_get_stripe_count(lod, lo,
2691 entry->llc_stripe_count, false);
2694 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2696 int magic, size = 0, i;
2697 struct lod_layout_component *comp_entries;
2699 bool is_composite, is_foreign = false;
2702 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2703 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2705 lo->ldo_def_striping->lds_def_striping_is_composite;
2707 comp_cnt = lo->ldo_comp_cnt;
2708 comp_entries = lo->ldo_comp_entries;
2709 is_composite = lo->ldo_is_composite;
2710 is_foreign = lo->ldo_is_foreign;
2714 return lo->ldo_foreign_lov_size;
2716 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2718 size = sizeof(struct lov_comp_md_v1) +
2719 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2720 LASSERT(size % sizeof(__u64) == 0);
2723 for (i = 0; i < comp_cnt; i++) {
2726 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2727 stripe_count = lod_comp_entry_stripe_count(lo, &comp_entries[i],
2729 if (!is_dir && is_composite)
2730 lod_comp_shrink_stripe_count(&comp_entries[i],
2733 size += lov_user_md_size(stripe_count, magic);
2734 LASSERT(size % sizeof(__u64) == 0);
2740 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2741 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2744 * \param[in] env execution environment
2745 * \param[in] dt dt_object to add components on
2746 * \param[in] buf buffer contains components to be added
2747 * \parem[in] th thandle
2749 * \retval 0 on success
2750 * \retval negative errno on failure
2752 static int lod_declare_layout_add(const struct lu_env *env,
2753 struct dt_object *dt,
2754 const struct lu_buf *buf,
2757 struct lod_thread_info *info = lod_env_info(env);
2758 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2759 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2760 struct dt_object *next = dt_object_child(dt);
2761 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
2762 struct lod_object *lo = lod_dt_obj(dt);
2763 struct lov_user_md_v3 *v3;
2764 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2766 int i, rc, array_cnt, old_array_cnt;
2769 LASSERT(lo->ldo_is_composite);
2771 if (lo->ldo_flr_state != LCM_FL_NONE)
2774 rc = lod_verify_striping(d, lo, buf, false);
2778 magic = comp_v1->lcm_magic;
2779 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2780 lustre_swab_lov_comp_md_v1(comp_v1);
2781 magic = comp_v1->lcm_magic;
2784 if (magic != LOV_USER_MAGIC_COMP_V1)
2787 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2788 OBD_ALLOC(comp_array, sizeof(*comp_array) * array_cnt);
2789 if (comp_array == NULL)
2792 memcpy(comp_array, lo->ldo_comp_entries,
2793 sizeof(*comp_array) * lo->ldo_comp_cnt);
2795 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2796 struct lov_user_md_v1 *v1;
2797 struct lu_extent *ext;
2799 v1 = (struct lov_user_md *)((char *)comp_v1 +
2800 comp_v1->lcm_entries[i].lcme_offset);
2801 ext = &comp_v1->lcm_entries[i].lcme_extent;
2803 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2804 lod_comp->llc_extent.e_start = ext->e_start;
2805 lod_comp->llc_extent.e_end = ext->e_end;
2806 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2807 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2809 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2810 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2811 lod_adjust_stripe_info(lod_comp, desc, 0);
2813 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2814 v3 = (struct lov_user_md_v3 *) v1;
2815 if (v3->lmm_pool_name[0] != '\0') {
2816 rc = lod_set_pool(&lod_comp->llc_pool,
2824 old_array = lo->ldo_comp_entries;
2825 old_array_cnt = lo->ldo_comp_cnt;
2827 lo->ldo_comp_entries = comp_array;
2828 lo->ldo_comp_cnt = array_cnt;
2830 /* No need to increase layout generation here, it will be increased
2831 * later when generating component ID for the new components */
2833 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2834 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2835 XATTR_NAME_LOV, 0, th);
2837 lo->ldo_comp_entries = old_array;
2838 lo->ldo_comp_cnt = old_array_cnt;
2842 OBD_FREE(old_array, sizeof(*lod_comp) * old_array_cnt);
2844 LASSERT(lo->ldo_mirror_count == 1);
2845 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2850 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2851 lod_comp = &comp_array[i];
2852 if (lod_comp->llc_pool != NULL) {
2853 OBD_FREE(lod_comp->llc_pool,
2854 strlen(lod_comp->llc_pool) + 1);
2855 lod_comp->llc_pool = NULL;
2858 OBD_FREE(comp_array, sizeof(*comp_array) * array_cnt);
2863 * lod_last_non_stale_mirror() - Check if a mirror is the last non-stale mirror.
2864 * @mirror_id: Mirror id to be checked.
2867 * This function checks if a mirror with specified @mirror_id is the last
2868 * non-stale mirror of a LOD object @lo.
2870 * Return: true or false.
2873 bool lod_last_non_stale_mirror(__u16 mirror_id, struct lod_object *lo)
2875 struct lod_layout_component *lod_comp;
2876 bool has_stale_flag;
2879 for (i = 0; i < lo->ldo_mirror_count; i++) {
2880 if (lo->ldo_mirrors[i].lme_id == mirror_id ||
2881 lo->ldo_mirrors[i].lme_stale)
2884 has_stale_flag = false;
2885 lod_foreach_mirror_comp(lod_comp, lo, i) {
2886 if (lod_comp->llc_flags & LCME_FL_STALE) {
2887 has_stale_flag = true;
2891 if (!has_stale_flag)
2899 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2900 * the '$field' can only be 'flags' now. The xattr value is binary
2901 * lov_comp_md_v1 which contains the component ID(s) and the value of
2902 * the field to be modified.
2904 * \param[in] env execution environment
2905 * \param[in] dt dt_object to be modified
2906 * \param[in] op operation string, like "set.flags"
2907 * \param[in] buf buffer contains components to be set
2908 * \parem[in] th thandle
2910 * \retval 0 on success
2911 * \retval negative errno on failure
2913 static int lod_declare_layout_set(const struct lu_env *env,
2914 struct dt_object *dt,
2915 char *op, const struct lu_buf *buf,
2918 struct lod_layout_component *lod_comp;
2919 struct lod_thread_info *info = lod_env_info(env);
2920 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2921 struct lod_object *lo = lod_dt_obj(dt);
2922 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2925 bool changed = false;
2928 if (strcmp(op, "set.flags") != 0) {
2929 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2930 lod2obd(d)->obd_name, op);
2934 magic = comp_v1->lcm_magic;
2935 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2936 lustre_swab_lov_comp_md_v1(comp_v1);
2937 magic = comp_v1->lcm_magic;
2940 if (magic != LOV_USER_MAGIC_COMP_V1)
2943 if (comp_v1->lcm_entry_count == 0) {
2944 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2945 lod2obd(d)->obd_name);
2949 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2950 __u32 id = comp_v1->lcm_entries[i].lcme_id;
2951 __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
2952 __u32 mirror_flag = flags & LCME_MIRROR_FLAGS;
2953 __u16 mirror_id = mirror_id_of(id);
2954 bool neg = flags & LCME_FL_NEG;
2956 if (flags & LCME_FL_INIT) {
2958 lod_striping_free(env, lo);
2962 flags &= ~(LCME_MIRROR_FLAGS | LCME_FL_NEG);
2963 for (j = 0; j < lo->ldo_comp_cnt; j++) {
2964 lod_comp = &lo->ldo_comp_entries[j];
2966 /* lfs only put one flag in each entry */
2967 if ((flags && id != lod_comp->llc_id) ||
2968 (mirror_flag && mirror_id !=
2969 mirror_id_of(lod_comp->llc_id)))
2974 lod_comp->llc_flags &= ~flags;
2976 lod_comp->llc_flags &= ~mirror_flag;
2979 if ((flags & LCME_FL_STALE) &&
2980 lod_last_non_stale_mirror(mirror_id,
2983 lod_comp->llc_flags |= flags;
2986 lod_comp->llc_flags |= mirror_flag;
2987 if (mirror_flag & LCME_FL_NOSYNC)
2988 lod_comp->llc_timestamp =
2989 ktime_get_real_seconds();
2997 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
2998 lod2obd(d)->obd_name);
3002 lod_obj_inc_layout_gen(lo);
3004 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3005 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
3006 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3011 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
3012 * and the xattr value is a unique component ID or a special lcme_id.
3014 * \param[in] env execution environment
3015 * \param[in] dt dt_object to be operated on
3016 * \param[in] buf buffer contains component ID or lcme_id
3017 * \parem[in] th thandle
3019 * \retval 0 on success
3020 * \retval negative errno on failure
3022 static int lod_declare_layout_del(const struct lu_env *env,
3023 struct dt_object *dt,
3024 const struct lu_buf *buf,
3027 struct lod_thread_info *info = lod_env_info(env);
3028 struct dt_object *next = dt_object_child(dt);
3029 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3030 struct lod_object *lo = lod_dt_obj(dt);
3031 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3032 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
3033 __u32 magic, id, flags, neg_flags = 0;
3037 LASSERT(lo->ldo_is_composite);
3039 if (lo->ldo_flr_state != LCM_FL_NONE)
3042 magic = comp_v1->lcm_magic;
3043 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
3044 lustre_swab_lov_comp_md_v1(comp_v1);
3045 magic = comp_v1->lcm_magic;
3048 if (magic != LOV_USER_MAGIC_COMP_V1)
3051 id = comp_v1->lcm_entries[0].lcme_id;
3052 flags = comp_v1->lcm_entries[0].lcme_flags;
3054 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
3055 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
3056 lod2obd(d)->obd_name, id, flags);
3060 if (id != LCME_ID_INVAL && flags != 0) {
3061 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
3062 lod2obd(d)->obd_name);
3066 if (id == LCME_ID_INVAL && !flags) {
3067 CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
3068 lod2obd(d)->obd_name);
3072 if (flags & LCME_FL_NEG) {
3073 neg_flags = flags & ~LCME_FL_NEG;
3077 left = lo->ldo_comp_cnt;
3081 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3082 struct lod_layout_component *lod_comp;
3084 lod_comp = &lo->ldo_comp_entries[i];
3086 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
3088 else if (flags && !(flags & lod_comp->llc_flags))
3090 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
3093 if (left != (i + 1)) {
3094 CDEBUG(D_LAYOUT, "%s: this deletion will create "
3095 "a hole.\n", lod2obd(d)->obd_name);
3100 /* Mark the component as deleted */
3101 lod_comp->llc_id = LCME_ID_INVAL;
3103 /* Not instantiated component */
3104 if (lod_comp->llc_stripe == NULL)
3107 LASSERT(lod_comp->llc_stripe_count > 0);
3108 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
3109 struct dt_object *obj = lod_comp->llc_stripe[j];
3113 rc = lod_sub_declare_destroy(env, obj, th);
3119 LASSERTF(left >= 0, "left = %d\n", left);
3120 if (left == lo->ldo_comp_cnt) {
3121 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
3122 lod2obd(d)->obd_name, id);
3126 memset(attr, 0, sizeof(*attr));
3127 attr->la_valid = LA_SIZE;
3128 rc = lod_sub_declare_attr_set(env, next, attr, th);
3133 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3134 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
3135 XATTR_NAME_LOV, 0, th);
3137 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
3144 * Declare layout add/set/del operations issued by special xattr names:
3146 * XATTR_LUSTRE_LOV.add add component(s) to existing file
3147 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
3148 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
3150 * \param[in] env execution environment
3151 * \param[in] dt object
3152 * \param[in] name name of xattr
3153 * \param[in] buf lu_buf contains xattr value
3154 * \param[in] th transaction handle
3156 * \retval 0 on success
3157 * \retval negative if failed
3159 static int lod_declare_modify_layout(const struct lu_env *env,
3160 struct dt_object *dt,
3162 const struct lu_buf *buf,
3165 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3166 struct lod_object *lo = lod_dt_obj(dt);
3168 int rc, len = strlen(XATTR_LUSTRE_LOV);
3171 LASSERT(dt_object_exists(dt));
3173 if (strlen(name) <= len || name[len] != '.') {
3174 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
3175 lod2obd(d)->obd_name, name);
3180 rc = lod_striping_load(env, lo);
3184 /* the layout to be modified must be a composite layout */
3185 if (!lo->ldo_is_composite) {
3186 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
3187 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3188 GOTO(unlock, rc = -EINVAL);
3191 op = (char *)name + len;
3192 if (strcmp(op, "add") == 0) {
3193 rc = lod_declare_layout_add(env, dt, buf, th);
3194 } else if (strcmp(op, "del") == 0) {
3195 rc = lod_declare_layout_del(env, dt, buf, th);
3196 } else if (strncmp(op, "set", strlen("set")) == 0) {
3197 rc = lod_declare_layout_set(env, dt, op, buf, th);
3199 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
3200 lod2obd(d)->obd_name, name);
3201 GOTO(unlock, rc = -ENOTSUPP);
3205 lod_striping_free(env, lo);