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 = scnprintf(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 BUILD_BUG_ON(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 (lmv_is_sane(lmv1))
1533 rc = lmv_mds_md_size(
1534 le32_to_cpu(lmv1->lmv_stripe_count),
1535 le32_to_cpu(lmv1->lmv_magic));
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 (!lmv_is_sane(lmv1))
1744 LASSERT(lo->ldo_stripe == NULL);
1745 OBD_ALLOC(stripe, sizeof(stripe[0]) *
1746 (le32_to_cpu(lmv1->lmv_stripe_count)));
1750 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1751 struct dt_device *tgt_dt;
1752 struct dt_object *dto;
1753 int type = LU_SEQ_RANGE_ANY;
1756 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1757 if (!fid_is_sane(fid)) {
1762 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1766 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1767 tgt_dt = lod->lod_child;
1769 struct lod_tgt_desc *tgt;
1771 tgt = LTD_TGT(ltd, idx);
1773 GOTO(out, rc = -ESTALE);
1774 tgt_dt = tgt->ltd_tgt;
1777 dto = dt_locate_at(env, tgt_dt, fid,
1778 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1781 GOTO(out, rc = PTR_ERR(dto));
1786 lo->ldo_stripe = stripe;
1787 lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1788 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1789 lo->ldo_dir_layout_version = le32_to_cpu(lmv1->lmv_layout_version);
1790 lo->ldo_dir_hash_type = le32_to_cpu(lmv1->lmv_hash_type);
1792 lod_striping_free_nolock(env, lo);
1798 * Declare create a striped directory.
1800 * Declare creating a striped directory with a given stripe pattern on the
1801 * specified MDTs. A striped directory is represented as a regular directory
1802 * - an index listing all the stripes. The stripes point back to the master
1803 * object with ".." and LinkEA. The master object gets LMV EA which
1804 * identifies it as a striped directory. The function allocates FIDs
1807 * \param[in] env execution environment
1808 * \param[in] dt object
1809 * \param[in] attr attributes to initialize the objects with
1810 * \param[in] dof type of objects to be created
1811 * \param[in] th transaction handle
1813 * \retval 0 on success
1814 * \retval negative if failed
1816 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1817 struct dt_object *dt,
1818 struct lu_attr *attr,
1819 struct dt_object_format *dof,
1822 struct lod_thread_info *info = lod_env_info(env);
1823 struct lu_buf lmv_buf;
1824 struct lu_buf slave_lmv_buf;
1825 struct lmv_mds_md_v1 *lmm;
1826 struct lmv_mds_md_v1 *slave_lmm = NULL;
1827 struct dt_insert_rec *rec = &info->lti_dt_rec;
1828 struct lod_object *lo = lod_dt_obj(dt);
1833 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1836 lmm = lmv_buf.lb_buf;
1838 OBD_ALLOC_PTR(slave_lmm);
1839 if (slave_lmm == NULL)
1840 GOTO(out, rc = -ENOMEM);
1842 lod_prep_slave_lmv_md(slave_lmm, lmm);
1843 slave_lmv_buf.lb_buf = slave_lmm;
1844 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1846 if (!dt_try_as_dir(env, dt_object_child(dt)))
1847 GOTO(out, rc = -EINVAL);
1849 rec->rec_type = S_IFDIR;
1850 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1851 struct dt_object *dto = lo->ldo_stripe[i];
1852 char *stripe_name = info->lti_key;
1853 struct lu_name *sname;
1854 struct linkea_data ldata = { NULL };
1855 struct lu_buf linkea_buf;
1857 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
1861 rc = lod_sub_declare_create(env, dto, attr, NULL, dof, th);
1865 if (!dt_try_as_dir(env, dto))
1866 GOTO(out, rc = -EINVAL);
1868 rc = lod_sub_declare_ref_add(env, dto, th);
1872 rec->rec_fid = lu_object_fid(&dto->do_lu);
1873 rc = lod_sub_declare_insert(env, dto,
1874 (const struct dt_rec *)rec,
1875 (const struct dt_key *)dot, th);
1879 /* master stripe FID will be put to .. */
1880 rec->rec_fid = lu_object_fid(&dt->do_lu);
1881 rc = lod_sub_declare_insert(env, dto,
1882 (const struct dt_rec *)rec,
1883 (const struct dt_key *)dotdot, th);
1887 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1888 cfs_fail_val != i) {
1889 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1891 slave_lmm->lmv_master_mdt_index =
1894 slave_lmm->lmv_master_mdt_index =
1896 rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
1897 XATTR_NAME_LMV, 0, th);
1902 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1904 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1905 PFID(lu_object_fid(&dto->do_lu)), i + 1);
1907 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1908 PFID(lu_object_fid(&dto->do_lu)), i);
1910 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
1911 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1912 sname, lu_object_fid(&dt->do_lu));
1916 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1917 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1918 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
1919 XATTR_NAME_LINK, 0, th);
1923 rec->rec_fid = lu_object_fid(&dto->do_lu);
1924 rc = lod_sub_declare_insert(env, dt_object_child(dt),
1925 (const struct dt_rec *)rec,
1926 (const struct dt_key *)stripe_name,
1931 rc = lod_sub_declare_ref_add(env, dt_object_child(dt), th);
1936 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
1937 &lmv_buf, XATTR_NAME_LMV, 0, th);
1941 if (slave_lmm != NULL)
1942 OBD_FREE_PTR(slave_lmm);
1948 * Allocate a striping on a predefined set of MDTs.
1950 * Allocates new striping using the MDT index range provided by the data from
1951 * the lum_obejcts contained in the lmv_user_md passed to this method if
1952 * \a is_specific is true; or allocates new layout starting from MDT index in
1953 * lo->ldo_dir_stripe_offset. The exact order of MDTs is not important and
1954 * varies depending on MDT status. The number of stripes needed and stripe
1955 * offset are taken from the object. If that number cannot be met, then the
1956 * function returns an error and then it's the caller's responsibility to
1957 * release the stripes allocated. All the internal structures are protected,
1958 * but no concurrent allocation is allowed on the same objects.
1960 * \param[in] env execution environment for this thread
1961 * \param[in] lo LOD object
1962 * \param[out] stripes striping created
1963 * \param[out] mdt_indices MDT indices of striping created
1964 * \param[in] is_specific true if the MDTs are provided by lum; false if
1965 * only the starting MDT index is provided
1967 * \retval positive stripes allocated, including the first stripe allocated
1969 * \retval negative errno on failure
1971 static int lod_mdt_alloc_specific(const struct lu_env *env,
1972 struct lod_object *lo,
1973 struct dt_object **stripes,
1974 __u32 *mdt_indices, bool is_specific)
1976 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1977 struct lu_tgt_descs *ltd = &lod->lod_mdt_descs;
1978 struct lu_tgt_desc *tgt = NULL;
1979 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
1980 struct dt_device *tgt_dt = NULL;
1981 struct lu_fid fid = { 0 };
1982 struct dt_object *dto;
1984 u32 stripe_count = lo->ldo_dir_stripe_count;
1990 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1991 if (stripe_count > 1)
1992 /* Set the start index for the 2nd stripe allocation */
1993 mdt_indices[1] = (mdt_indices[0] + 1) %
1994 (lod->lod_remote_mdt_count + 1);
1996 for (; stripe_idx < stripe_count; stripe_idx++) {
1997 /* Try to find next avaible target */
1998 idx = mdt_indices[stripe_idx];
1999 for (j = 0; j < lod->lod_remote_mdt_count;
2000 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
2001 bool already_allocated = false;
2004 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
2005 idx, lod->lod_remote_mdt_count + 1, stripe_idx);
2007 if (likely(!is_specific &&
2008 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
2009 /* check whether the idx already exists
2010 * in current allocated array */
2011 for (k = 0; k < stripe_idx; k++) {
2012 if (mdt_indices[k] == idx) {
2013 already_allocated = true;
2018 if (already_allocated)
2022 /* Sigh, this index is not in the bitmap, let's check
2023 * next available target */
2024 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) &&
2025 idx != master_index)
2028 if (idx == master_index) {
2029 /* Allocate the FID locally */
2030 rc = obd_fid_alloc(env, lod->lod_child_exp,
2034 tgt_dt = lod->lod_child;
2038 /* check the status of the OSP */
2039 tgt = LTD_TGT(ltd, idx);
2043 tgt_dt = tgt->ltd_tgt;
2044 if (!tgt->ltd_active)
2045 /* this OSP doesn't feel well */
2048 rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
2055 /* Can not allocate more stripes */
2056 if (j == lod->lod_remote_mdt_count) {
2057 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
2058 lod2obd(lod)->obd_name, stripe_count,
2063 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
2064 idx, stripe_idx, PFID(&fid));
2065 mdt_indices[stripe_idx] = idx;
2066 /* Set the start index for next stripe allocation */
2067 if (!is_specific && stripe_idx < stripe_count - 1) {
2069 * for large dir test, put all other slaves on one
2070 * remote MDT, otherwise we may save too many local
2071 * slave locks which will exceed RS_MAX_LOCKS.
2073 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)))
2075 mdt_indices[stripe_idx + 1] = (idx + 1) %
2076 (lod->lod_remote_mdt_count + 1);
2078 /* tgt_dt and fid must be ready after search avaible OSP
2079 * in the above loop */
2080 LASSERT(tgt_dt != NULL);
2081 LASSERT(fid_is_sane(&fid));
2083 /* fail a remote stripe FID allocation */
2084 if (stripe_idx && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_FID))
2087 dto = dt_locate_at(env, tgt_dt, &fid,
2088 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
2095 stripes[stripe_idx] = dto;
2101 for (j = 1; j < stripe_idx; j++) {
2102 LASSERT(stripes[j] != NULL);
2103 dt_object_put(env, stripes[j]);
2109 static int lod_prep_md_striped_create(const struct lu_env *env,
2110 struct dt_object *dt,
2111 struct lu_attr *attr,
2112 const struct lmv_user_md_v1 *lum,
2113 struct dt_object_format *dof,
2116 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
2117 struct lod_object *lo = lod_dt_obj(dt);
2118 struct dt_object **stripes;
2119 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2120 struct lu_fid fid = { 0 };
2127 /* The lum has been verifed in lod_verify_md_striping */
2128 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
2129 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
2131 stripe_count = lo->ldo_dir_stripe_count;
2133 OBD_ALLOC(stripes, sizeof(stripes[0]) * stripe_count);
2137 /* Allocate the first stripe locally */
2138 rc = obd_fid_alloc(env, lod->lod_child_exp, &fid, NULL);
2142 stripes[0] = dt_locate_at(env, lod->lod_child, &fid,
2143 dt->do_lu.lo_dev->ld_site->ls_top_dev, &conf);
2144 if (IS_ERR(stripes[0]))
2145 GOTO(out, rc = PTR_ERR(stripes[0]));
2147 if (lo->ldo_dir_stripe_offset == LMV_OFFSET_DEFAULT) {
2148 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
2149 rc = lod_mdt_alloc_qos(env, lo, stripes);
2151 rc = lod_mdt_alloc_rr(env, lo, stripes);
2154 bool is_specific = false;
2156 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
2158 GOTO(out, rc = -ENOMEM);
2160 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
2162 for (i = 0; i < stripe_count; i++)
2164 le32_to_cpu(lum->lum_objects[i].lum_mds);
2167 /* stripe 0 is local */
2169 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
2170 rc = lod_mdt_alloc_specific(env, lo, stripes, idx_array,
2172 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
2180 lo->ldo_dir_striped = 1;
2181 lo->ldo_stripe = stripes;
2182 lo->ldo_dir_stripe_count = rc;
2183 lo->ldo_dir_stripes_allocated = stripe_count;
2185 lo->ldo_dir_stripe_loaded = 1;
2187 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2189 lod_striping_free(env, lo);
2195 if (!IS_ERR_OR_NULL(stripes[0]))
2196 dt_object_put(env, stripes[0]);
2197 for (i = 1; i < stripe_count; i++)
2198 LASSERT(!stripes[i]);
2199 OBD_FREE(stripes, sizeof(stripes[0]) * stripe_count);
2206 * Alloc cached foreign LMV
2208 * \param[in] lo object
2209 * \param[in] size size of foreign LMV
2211 * \retval 0 on success
2212 * \retval negative if failed
2214 int lod_alloc_foreign_lmv(struct lod_object *lo, size_t size)
2216 OBD_ALLOC_LARGE(lo->ldo_foreign_lmv, size);
2217 if (lo->ldo_foreign_lmv == NULL)
2219 lo->ldo_foreign_lmv_size = size;
2220 lo->ldo_dir_is_foreign = 1;
2226 * Declare create striped md object.
2228 * The function declares intention to create a striped directory. This is a
2229 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2230 * is to verify pattern \a lum_buf is good. Check that function for the details.
2232 * \param[in] env execution environment
2233 * \param[in] dt object
2234 * \param[in] attr attributes to initialize the objects with
2235 * \param[in] lum_buf a pattern specifying the number of stripes and
2237 * \param[in] dof type of objects to be created
2238 * \param[in] th transaction handle
2240 * \retval 0 on success
2241 * \retval negative if failed
2244 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2245 struct dt_object *dt,
2246 struct lu_attr *attr,
2247 const struct lu_buf *lum_buf,
2248 struct dt_object_format *dof,
2251 struct lod_object *lo = lod_dt_obj(dt);
2252 struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
2256 LASSERT(lum != NULL);
2258 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
2259 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
2260 (int)le32_to_cpu(lum->lum_stripe_offset));
2262 if (lo->ldo_dir_stripe_count == 0) {
2263 if (lo->ldo_dir_is_foreign) {
2264 rc = lod_alloc_foreign_lmv(lo, lum_buf->lb_len);
2267 memcpy(lo->ldo_foreign_lmv, lum, lum_buf->lb_len);
2268 lo->ldo_dir_stripe_loaded = 1;
2273 /* prepare dir striped objects */
2274 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2276 /* failed to create striping, let's reset
2277 * config so that others don't get confused */
2278 lod_striping_free(env, lo);
2286 * Set or replace striped directory layout, and LFSCK may set layout on a plain
2287 * directory, so don't check stripe count.
2289 * \param[in] env execution environment
2290 * \param[in] dt target object
2291 * \param[in] buf LMV buf which contains source stripe fids
2292 * \param[in] fl set or replace
2293 * \param[in] th transaction handle
2295 * \retval 0 on success
2296 * \retval negative if failed
2298 static int lod_dir_layout_set(const struct lu_env *env,
2299 struct dt_object *dt,
2300 const struct lu_buf *buf,
2304 struct dt_object *next = dt_object_child(dt);
2305 struct lod_object *lo = lod_dt_obj(dt);
2306 struct lmv_mds_md_v1 *lmv = buf->lb_buf;
2307 struct lmv_mds_md_v1 *slave_lmv;
2308 struct lu_buf slave_buf;
2314 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LMV, fl, th);
2318 OBD_ALLOC_PTR(slave_lmv);
2322 lod_prep_slave_lmv_md(slave_lmv, lmv);
2323 slave_buf.lb_buf = slave_lmv;
2324 slave_buf.lb_len = sizeof(*slave_lmv);
2326 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2327 if (!lo->ldo_stripe[i])
2330 if (!dt_object_exists(lo->ldo_stripe[i]))
2333 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], &slave_buf,
2334 XATTR_NAME_LMV, fl, th);
2339 lod_striping_free(env, lod_dt_obj(dt));
2340 OBD_FREE_PTR(slave_lmv);
2346 * Implementation of dt_object_operations::do_declare_xattr_set.
2348 * Used with regular (non-striped) objects. Basically it
2349 * initializes the striping information and applies the
2350 * change to all the stripes.
2352 * \see dt_object_operations::do_declare_xattr_set() in the API description
2355 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2356 struct dt_object *dt,
2357 const struct lu_buf *buf,
2358 const char *name, int fl,
2361 struct dt_object *next = dt_object_child(dt);
2362 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2363 struct lod_object *lo = lod_dt_obj(dt);
2368 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2369 struct lmv_user_md_v1 *lum;
2371 LASSERT(buf != NULL && buf->lb_buf != NULL);
2373 rc = lod_verify_md_striping(d, lum);
2376 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2377 rc = lod_verify_striping(d, lo, buf, false);
2382 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2386 /* Note: Do not set LinkEA on sub-stripes, otherwise
2387 * it will confuse the fid2path process(see mdt_path_current()).
2388 * The linkEA between master and sub-stripes is set in
2389 * lod_xattr_set_lmv(). */
2390 if (strcmp(name, XATTR_NAME_LINK) == 0)
2393 /* set xattr to each stripes, if needed */
2394 rc = lod_striping_load(env, lo);
2398 if (lo->ldo_dir_stripe_count == 0)
2401 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2402 if (!lo->ldo_stripe[i])
2405 if (!dt_object_exists(lo->ldo_stripe[i]))
2408 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2418 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2419 struct lod_object *lo,
2420 struct dt_object *dt, struct thandle *th,
2421 int comp_idx, int stripe_idx,
2422 struct lod_obj_stripe_cb_data *data)
2424 struct lod_thread_info *info = lod_env_info(env);
2425 struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
2426 struct filter_fid *ff = &info->lti_ff;
2427 struct lu_buf *buf = &info->lti_buf;
2431 buf->lb_len = sizeof(*ff);
2432 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2440 * locd_buf is set if it's called by dir migration, which doesn't check
2443 if (data->locd_buf) {
2444 memset(ff, 0, sizeof(*ff));
2445 ff->ff_parent = *(struct lu_fid *)data->locd_buf->lb_buf;
2447 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
2449 if (lu_fid_eq(lod_object_fid(lo), &ff->ff_parent) &&
2450 ff->ff_layout.ol_comp_id == comp->llc_id)
2453 memset(ff, 0, sizeof(*ff));
2454 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2457 /* rewrite filter_fid */
2458 ff->ff_parent.f_ver = stripe_idx;
2459 ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
2460 ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
2461 ff->ff_layout.ol_comp_id = comp->llc_id;
2462 ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
2463 ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
2464 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
2466 if (data->locd_declare)
2467 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2468 LU_XATTR_REPLACE, th);
2470 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2471 LU_XATTR_REPLACE, th);
2477 * Reset parent FID on OST object
2479 * Replace parent FID with @dt object FID, which is only called during migration
2480 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2481 * the FID is changed.
2483 * \param[in] env execution environment
2484 * \param[in] dt dt_object whose stripes's parent FID will be reset
2485 * \parem[in] th thandle
2486 * \param[in] declare if it is declare
2488 * \retval 0 if reset succeeds
2489 * \retval negative errno if reset fails
2491 static int lod_replace_parent_fid(const struct lu_env *env,
2492 struct dt_object *dt,
2493 const struct lu_buf *buf,
2494 struct thandle *th, bool declare)
2496 struct lod_object *lo = lod_dt_obj(dt);
2497 struct lod_thread_info *info = lod_env_info(env);
2498 struct filter_fid *ff;
2499 struct lod_obj_stripe_cb_data data = { { 0 } };
2503 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2505 /* set xattr to each stripes, if needed */
2506 rc = lod_striping_load(env, lo);
2510 if (!lod_obj_is_striped(dt))
2513 if (info->lti_ea_store_size < sizeof(*ff)) {
2514 rc = lod_ea_store_resize(info, sizeof(*ff));
2519 data.locd_declare = declare;
2520 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2521 data.locd_buf = buf;
2522 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2527 inline __u16 lod_comp_entry_stripe_count(struct lod_object *lo,
2528 struct lod_layout_component *entry,
2531 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2535 else if (lod_comp_inited(entry))
2536 return entry->llc_stripe_count;
2537 else if ((__u16)-1 == entry->llc_stripe_count)
2538 return lod->lod_ost_count;
2540 return lod_get_stripe_count(lod, lo,
2541 entry->llc_stripe_count, false);
2544 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2546 int magic, size = 0, i;
2547 struct lod_layout_component *comp_entries;
2549 bool is_composite, is_foreign = false;
2552 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2553 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2555 lo->ldo_def_striping->lds_def_striping_is_composite;
2557 comp_cnt = lo->ldo_comp_cnt;
2558 comp_entries = lo->ldo_comp_entries;
2559 is_composite = lo->ldo_is_composite;
2560 is_foreign = lo->ldo_is_foreign;
2564 return lo->ldo_foreign_lov_size;
2566 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2568 size = sizeof(struct lov_comp_md_v1) +
2569 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2570 LASSERT(size % sizeof(__u64) == 0);
2573 for (i = 0; i < comp_cnt; i++) {
2576 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2577 stripe_count = lod_comp_entry_stripe_count(lo, &comp_entries[i],
2579 if (!is_dir && is_composite)
2580 lod_comp_shrink_stripe_count(&comp_entries[i],
2583 size += lov_user_md_size(stripe_count, magic);
2584 LASSERT(size % sizeof(__u64) == 0);
2590 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2591 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2594 * \param[in] env execution environment
2595 * \param[in] dt dt_object to add components on
2596 * \param[in] buf buffer contains components to be added
2597 * \parem[in] th thandle
2599 * \retval 0 on success
2600 * \retval negative errno on failure
2602 static int lod_declare_layout_add(const struct lu_env *env,
2603 struct dt_object *dt,
2604 const struct lu_buf *buf,
2607 struct lod_thread_info *info = lod_env_info(env);
2608 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2609 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2610 struct dt_object *next = dt_object_child(dt);
2611 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
2612 struct lod_object *lo = lod_dt_obj(dt);
2613 struct lov_user_md_v3 *v3;
2614 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2616 int i, rc, array_cnt, old_array_cnt;
2619 LASSERT(lo->ldo_is_composite);
2621 if (lo->ldo_flr_state != LCM_FL_NONE)
2624 rc = lod_verify_striping(d, lo, buf, false);
2628 magic = comp_v1->lcm_magic;
2629 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2630 lustre_swab_lov_comp_md_v1(comp_v1);
2631 magic = comp_v1->lcm_magic;
2634 if (magic != LOV_USER_MAGIC_COMP_V1)
2637 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2638 OBD_ALLOC(comp_array, sizeof(*comp_array) * array_cnt);
2639 if (comp_array == NULL)
2642 memcpy(comp_array, lo->ldo_comp_entries,
2643 sizeof(*comp_array) * lo->ldo_comp_cnt);
2645 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2646 struct lov_user_md_v1 *v1;
2647 struct lu_extent *ext;
2649 v1 = (struct lov_user_md *)((char *)comp_v1 +
2650 comp_v1->lcm_entries[i].lcme_offset);
2651 ext = &comp_v1->lcm_entries[i].lcme_extent;
2653 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2654 lod_comp->llc_extent.e_start = ext->e_start;
2655 lod_comp->llc_extent.e_end = ext->e_end;
2656 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2657 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2659 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2660 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2661 lod_adjust_stripe_info(lod_comp, desc, 0);
2663 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2664 v3 = (struct lov_user_md_v3 *) v1;
2665 if (v3->lmm_pool_name[0] != '\0') {
2666 rc = lod_set_pool(&lod_comp->llc_pool,
2674 old_array = lo->ldo_comp_entries;
2675 old_array_cnt = lo->ldo_comp_cnt;
2677 lo->ldo_comp_entries = comp_array;
2678 lo->ldo_comp_cnt = array_cnt;
2680 /* No need to increase layout generation here, it will be increased
2681 * later when generating component ID for the new components */
2683 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2684 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2685 XATTR_NAME_LOV, 0, th);
2687 lo->ldo_comp_entries = old_array;
2688 lo->ldo_comp_cnt = old_array_cnt;
2692 OBD_FREE(old_array, sizeof(*lod_comp) * old_array_cnt);
2694 LASSERT(lo->ldo_mirror_count == 1);
2695 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2700 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2701 lod_comp = &comp_array[i];
2702 if (lod_comp->llc_pool != NULL) {
2703 OBD_FREE(lod_comp->llc_pool,
2704 strlen(lod_comp->llc_pool) + 1);
2705 lod_comp->llc_pool = NULL;
2708 OBD_FREE(comp_array, sizeof(*comp_array) * array_cnt);
2713 * lod_last_non_stale_mirror() - Check if a mirror is the last non-stale mirror.
2714 * @mirror_id: Mirror id to be checked.
2717 * This function checks if a mirror with specified @mirror_id is the last
2718 * non-stale mirror of a LOD object @lo.
2720 * Return: true or false.
2723 bool lod_last_non_stale_mirror(__u16 mirror_id, struct lod_object *lo)
2725 struct lod_layout_component *lod_comp;
2726 bool has_stale_flag;
2729 for (i = 0; i < lo->ldo_mirror_count; i++) {
2730 if (lo->ldo_mirrors[i].lme_id == mirror_id ||
2731 lo->ldo_mirrors[i].lme_stale)
2734 has_stale_flag = false;
2735 lod_foreach_mirror_comp(lod_comp, lo, i) {
2736 if (lod_comp->llc_flags & LCME_FL_STALE) {
2737 has_stale_flag = true;
2741 if (!has_stale_flag)
2749 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2750 * the '$field' can only be 'flags' now. The xattr value is binary
2751 * lov_comp_md_v1 which contains the component ID(s) and the value of
2752 * the field to be modified.
2753 * Please update allowed_lustre_lov macro if $field groks more values
2756 * \param[in] env execution environment
2757 * \param[in] dt dt_object to be modified
2758 * \param[in] op operation string, like "set.flags"
2759 * \param[in] buf buffer contains components to be set
2760 * \parem[in] th thandle
2762 * \retval 0 on success
2763 * \retval negative errno on failure
2765 static int lod_declare_layout_set(const struct lu_env *env,
2766 struct dt_object *dt,
2767 char *op, const struct lu_buf *buf,
2770 struct lod_layout_component *lod_comp;
2771 struct lod_thread_info *info = lod_env_info(env);
2772 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2773 struct lod_object *lo = lod_dt_obj(dt);
2774 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2777 bool changed = false;
2780 /* Please update allowed_lustre_lov macro if op
2781 * groks more values in the future
2783 if (strcmp(op, "set.flags") != 0) {
2784 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2785 lod2obd(d)->obd_name, op);
2789 magic = comp_v1->lcm_magic;
2790 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2791 lustre_swab_lov_comp_md_v1(comp_v1);
2792 magic = comp_v1->lcm_magic;
2795 if (magic != LOV_USER_MAGIC_COMP_V1)
2798 if (comp_v1->lcm_entry_count == 0) {
2799 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2800 lod2obd(d)->obd_name);
2804 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2805 __u32 id = comp_v1->lcm_entries[i].lcme_id;
2806 __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
2807 __u32 mirror_flag = flags & LCME_MIRROR_FLAGS;
2808 __u16 mirror_id = mirror_id_of(id);
2809 bool neg = flags & LCME_FL_NEG;
2811 if (flags & LCME_FL_INIT) {
2813 lod_striping_free(env, lo);
2817 flags &= ~(LCME_MIRROR_FLAGS | LCME_FL_NEG);
2818 for (j = 0; j < lo->ldo_comp_cnt; j++) {
2819 lod_comp = &lo->ldo_comp_entries[j];
2821 /* lfs only put one flag in each entry */
2822 if ((flags && id != lod_comp->llc_id) ||
2823 (mirror_flag && mirror_id !=
2824 mirror_id_of(lod_comp->llc_id)))
2829 lod_comp->llc_flags &= ~flags;
2831 lod_comp->llc_flags &= ~mirror_flag;
2834 if ((flags & LCME_FL_STALE) &&
2835 lod_last_non_stale_mirror(mirror_id,
2838 lod_comp->llc_flags |= flags;
2841 lod_comp->llc_flags |= mirror_flag;
2842 if (mirror_flag & LCME_FL_NOSYNC)
2843 lod_comp->llc_timestamp =
2844 ktime_get_real_seconds();
2852 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
2853 lod2obd(d)->obd_name);
2857 lod_obj_inc_layout_gen(lo);
2859 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2860 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
2861 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
2866 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
2867 * and the xattr value is a unique component ID or a special lcme_id.
2869 * \param[in] env execution environment
2870 * \param[in] dt dt_object to be operated on
2871 * \param[in] buf buffer contains component ID or lcme_id
2872 * \parem[in] th thandle
2874 * \retval 0 on success
2875 * \retval negative errno on failure
2877 static int lod_declare_layout_del(const struct lu_env *env,
2878 struct dt_object *dt,
2879 const struct lu_buf *buf,
2882 struct lod_thread_info *info = lod_env_info(env);
2883 struct dt_object *next = dt_object_child(dt);
2884 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2885 struct lod_object *lo = lod_dt_obj(dt);
2886 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2887 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2888 __u32 magic, id, flags, neg_flags = 0;
2892 LASSERT(lo->ldo_is_composite);
2894 if (lo->ldo_flr_state != LCM_FL_NONE)
2897 magic = comp_v1->lcm_magic;
2898 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2899 lustre_swab_lov_comp_md_v1(comp_v1);
2900 magic = comp_v1->lcm_magic;
2903 if (magic != LOV_USER_MAGIC_COMP_V1)
2906 id = comp_v1->lcm_entries[0].lcme_id;
2907 flags = comp_v1->lcm_entries[0].lcme_flags;
2909 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
2910 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
2911 lod2obd(d)->obd_name, id, flags);
2915 if (id != LCME_ID_INVAL && flags != 0) {
2916 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
2917 lod2obd(d)->obd_name);
2921 if (id == LCME_ID_INVAL && !flags) {
2922 CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
2923 lod2obd(d)->obd_name);
2927 if (flags & LCME_FL_NEG) {
2928 neg_flags = flags & ~LCME_FL_NEG;
2932 left = lo->ldo_comp_cnt;
2936 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
2937 struct lod_layout_component *lod_comp;
2939 lod_comp = &lo->ldo_comp_entries[i];
2941 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
2943 else if (flags && !(flags & lod_comp->llc_flags))
2945 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
2948 if (left != (i + 1)) {
2949 CDEBUG(D_LAYOUT, "%s: this deletion will create "
2950 "a hole.\n", lod2obd(d)->obd_name);
2955 /* Mark the component as deleted */
2956 lod_comp->llc_id = LCME_ID_INVAL;
2958 /* Not instantiated component */
2959 if (lod_comp->llc_stripe == NULL)
2962 LASSERT(lod_comp->llc_stripe_count > 0);
2963 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
2964 struct dt_object *obj = lod_comp->llc_stripe[j];
2968 rc = lod_sub_declare_destroy(env, obj, th);
2974 LASSERTF(left >= 0, "left = %d\n", left);
2975 if (left == lo->ldo_comp_cnt) {
2976 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
2977 lod2obd(d)->obd_name, id);
2981 memset(attr, 0, sizeof(*attr));
2982 attr->la_valid = LA_SIZE;
2983 rc = lod_sub_declare_attr_set(env, next, attr, th);
2988 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2989 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2990 XATTR_NAME_LOV, 0, th);
2992 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
2999 * Declare layout add/set/del operations issued by special xattr names:
3001 * XATTR_LUSTRE_LOV.add add component(s) to existing file
3002 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
3003 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
3005 * \param[in] env execution environment
3006 * \param[in] dt object
3007 * \param[in] name name of xattr
3008 * \param[in] buf lu_buf contains xattr value
3009 * \param[in] th transaction handle
3011 * \retval 0 on success
3012 * \retval negative if failed
3014 static int lod_declare_modify_layout(const struct lu_env *env,
3015 struct dt_object *dt,
3017 const struct lu_buf *buf,
3020 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3021 struct lod_object *lo = lod_dt_obj(dt);
3023 int rc, len = strlen(XATTR_LUSTRE_LOV);
3026 LASSERT(dt_object_exists(dt));
3028 if (strlen(name) <= len || name[len] != '.') {
3029 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
3030 lod2obd(d)->obd_name, name);
3035 rc = lod_striping_load(env, lo);
3039 /* the layout to be modified must be a composite layout */
3040 if (!lo->ldo_is_composite) {
3041 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
3042 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3043 GOTO(unlock, rc = -EINVAL);
3046 op = (char *)name + len;
3047 if (strcmp(op, "add") == 0) {
3048 rc = lod_declare_layout_add(env, dt, buf, th);
3049 } else if (strcmp(op, "del") == 0) {
3050 rc = lod_declare_layout_del(env, dt, buf, th);
3051 } else if (strncmp(op, "set", strlen("set")) == 0) {
3052 rc = lod_declare_layout_set(env, dt, op, buf, th);
3054 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
3055 lod2obd(d)->obd_name, name);
3056 GOTO(unlock, rc = -ENOTSUPP);
3060 lod_striping_free(env, lo);
3066 * Convert a plain file lov_mds_md to a composite layout.
3068 * \param[in,out] info the thread info::lti_ea_store buffer contains little
3069 * endian plain file layout
3071 * \retval 0 on success, <0 on failure
3073 static int lod_layout_convert(struct lod_thread_info *info)
3075 struct lov_mds_md *lmm = info->lti_ea_store;
3076 struct lov_mds_md *lmm_save;
3077 struct lov_comp_md_v1 *lcm;
3078 struct lov_comp_md_entry_v1 *lcme;
3084 /* realloc buffer to a composite layout which contains one component */
3085 blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3086 le32_to_cpu(lmm->lmm_magic));
3087 size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
3089 OBD_ALLOC_LARGE(lmm_save, blob_size);
3091 GOTO(out, rc = -ENOMEM);
3093 memcpy(lmm_save, lmm, blob_size);
3095 if (info->lti_ea_store_size < size) {
3096 rc = lod_ea_store_resize(info, size);
3101 lcm = info->lti_ea_store;
3102 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
3103 lcm->lcm_size = cpu_to_le32(size);
3104 lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
3105 lmm_save->lmm_layout_gen));
3106 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
3107 lcm->lcm_entry_count = cpu_to_le16(1);
3108 lcm->lcm_mirror_count = 0;
3110 lcme = &lcm->lcm_entries[0];
3111 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
3112 lcme->lcme_extent.e_start = 0;
3113 lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
3114 lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
3115 lcme->lcme_size = cpu_to_le32(blob_size);
3117 memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
3122 OBD_FREE_LARGE(lmm_save, blob_size);
3127 * Merge layouts to form a mirrored file.
3129 static int lod_declare_layout_merge(const struct lu_env *env,
3130 struct dt_object *dt, const struct lu_buf *mbuf,
3133 struct lod_thread_info *info = lod_env_info(env);
3134 struct lu_buf *buf = &info->lti_buf;
3135 struct lod_object *lo = lod_dt_obj(dt);
3136 struct lov_comp_md_v1 *lcm;
3137 struct lov_comp_md_v1 *cur_lcm;
3138 struct lov_comp_md_v1 *merge_lcm;
3139 struct lov_comp_md_entry_v1 *lcme;
3140 struct lov_mds_md_v1 *lmm;
3143 __u16 cur_entry_count;
3144 __u16 merge_entry_count;
3146 __u16 mirror_id = 0;
3153 merge_lcm = mbuf->lb_buf;
3154 if (mbuf->lb_len < sizeof(*merge_lcm))
3157 /* must be an existing layout from disk */
3158 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
3161 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
3163 /* do not allow to merge two mirrored files */
3164 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
3167 /* verify the target buffer */
3168 rc = lod_get_lov_ea(env, lo);
3170 RETURN(rc ? : -ENODATA);
3172 cur_lcm = info->lti_ea_store;
3173 switch (le32_to_cpu(cur_lcm->lcm_magic)) {
3176 rc = lod_layout_convert(info);
3178 case LOV_MAGIC_COMP_V1:
3188 /* info->lti_ea_store could be reallocated in lod_layout_convert() */
3189 cur_lcm = info->lti_ea_store;
3190 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
3192 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
3193 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
3194 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
3197 /* size of new layout */
3198 size = le32_to_cpu(cur_lcm->lcm_size) +
3199 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
3201 memset(buf, 0, sizeof(*buf));
3202 lu_buf_alloc(buf, size);
3203 if (buf->lb_buf == NULL)
3207 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
3209 offset = sizeof(*lcm) +
3210 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
3211 for (i = 0; i < cur_entry_count; i++) {
3212 struct lov_comp_md_entry_v1 *cur_lcme;
3214 lcme = &lcm->lcm_entries[i];
3215 cur_lcme = &cur_lcm->lcm_entries[i];
3217 lcme->lcme_offset = cpu_to_le32(offset);
3218 memcpy((char *)lcm + offset,
3219 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
3220 le32_to_cpu(lcme->lcme_size));
3222 offset += le32_to_cpu(lcme->lcme_size);
3224 if (mirror_count == 1 &&
3225 mirror_id_of(le32_to_cpu(lcme->lcme_id)) == 0) {
3226 /* Add mirror from a non-flr file, create new mirror ID.
3227 * Otherwise, keep existing mirror's component ID, used
3228 * for mirror extension.
3230 id = pflr_id(1, i + 1);
3231 lcme->lcme_id = cpu_to_le32(id);
3234 id = max(le32_to_cpu(lcme->lcme_id), id);
3237 mirror_id = mirror_id_of(id) + 1;
3239 /* check if first entry in new layout is DOM */
3240 lmm = (struct lov_mds_md_v1 *)((char *)merge_lcm +
3241 merge_lcm->lcm_entries[0].lcme_offset);
3242 merge_has_dom = lov_pattern(le32_to_cpu(lmm->lmm_pattern)) ==
3245 for (i = 0; i < merge_entry_count; i++) {
3246 struct lov_comp_md_entry_v1 *merge_lcme;
3248 merge_lcme = &merge_lcm->lcm_entries[i];
3249 lcme = &lcm->lcm_entries[cur_entry_count + i];
3251 *lcme = *merge_lcme;
3252 lcme->lcme_offset = cpu_to_le32(offset);
3253 if (merge_has_dom && i == 0)
3254 lcme->lcme_flags |= cpu_to_le32(LCME_FL_STALE);
3256 id = pflr_id(mirror_id, i + 1);
3257 lcme->lcme_id = cpu_to_le32(id);
3259 memcpy((char *)lcm + offset,
3260 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
3261 le32_to_cpu(lcme->lcme_size));
3263 offset += le32_to_cpu(lcme->lcme_size);
3266 /* fixup layout information */
3267 lod_obj_inc_layout_gen(lo);
3268 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3269 lcm->lcm_size = cpu_to_le32(size);
3270 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
3271 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
3272 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NONE)
3273 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
3275 rc = lod_striping_reload(env, lo, buf);
3279 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
3280 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3288 * Split layouts, just set the LOVEA with the layout from mbuf.
3290 static int lod_declare_layout_split(const struct lu_env *env,
3291 struct dt_object *dt, const struct lu_buf *mbuf,
3294 struct lod_object *lo = lod_dt_obj(dt);
3295 struct lov_comp_md_v1 *lcm = mbuf->lb_buf;
3299 lod_obj_inc_layout_gen(lo);
3300 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3302 rc = lod_striping_reload(env, lo, mbuf);
3306 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), mbuf,
3307 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3312 * Implementation of dt_object_operations::do_declare_xattr_set.
3314 * \see dt_object_operations::do_declare_xattr_set() in the API description
3317 * the extension to the API:
3318 * - declaring LOVEA requests striping creation
3319 * - LU_XATTR_REPLACE means layout swap
3321 static int lod_declare_xattr_set(const struct lu_env *env,
3322 struct dt_object *dt,
3323 const struct lu_buf *buf,
3324 const char *name, int fl,
3327 struct dt_object *next = dt_object_child(dt);
3328 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3333 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
3334 if ((S_ISREG(mode) || mode == 0) &&
3335 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE | LU_XATTR_SPLIT)) &&
3336 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3337 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
3339 * this is a request to create object's striping.
3341 * allow to declare predefined striping on a new (!mode) object
3342 * which is supposed to be replay of regular file creation
3343 * (when LOV setting is declared)
3345 * LU_XATTR_REPLACE is set to indicate a layout swap
3347 if (dt_object_exists(dt)) {
3348 rc = dt_attr_get(env, next, attr);
3352 memset(attr, 0, sizeof(*attr));
3353 attr->la_valid = LA_TYPE | LA_MODE;
3354 attr->la_mode = S_IFREG;
3356 rc = lod_declare_striped_create(env, dt, attr, buf, th);
3357 } else if (fl & LU_XATTR_MERGE) {
3358 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3359 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3360 rc = lod_declare_layout_merge(env, dt, buf, th);
3361 } else if (fl & LU_XATTR_SPLIT) {
3362 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3363 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3364 rc = lod_declare_layout_split(env, dt, buf, th);
3365 } else if (S_ISREG(mode) &&
3366 strlen(name) >= sizeof(XATTR_LUSTRE_LOV) + 3 &&
3367 allowed_lustre_lov(name)) {
3369 * this is a request to modify object's striping.
3370 * add/set/del component(s).
3372 if (!dt_object_exists(dt))
3375 rc = lod_declare_modify_layout(env, dt, name, buf, th);
3376 } else if (S_ISDIR(mode)) {
3377 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
3378 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3379 rc = lod_replace_parent_fid(env, dt, buf, th, true);
3381 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
3388 * Apply xattr changes to the object.
3390 * Applies xattr changes to the object and the stripes if the latter exist.
3392 * \param[in] env execution environment
3393 * \param[in] dt object
3394 * \param[in] buf buffer pointing to the new value of xattr
3395 * \param[in] name name of xattr
3396 * \param[in] fl flags
3397 * \param[in] th transaction handle
3399 * \retval 0 on success
3400 * \retval negative if failed
3402 static int lod_xattr_set_internal(const struct lu_env *env,
3403 struct dt_object *dt,
3404 const struct lu_buf *buf,
3405 const char *name, int fl,
3408 struct dt_object *next = dt_object_child(dt);
3409 struct lod_object *lo = lod_dt_obj(dt);
3414 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3415 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3418 /* Note: Do not set LinkEA on sub-stripes, otherwise
3419 * it will confuse the fid2path process(see mdt_path_current()).
3420 * The linkEA between master and sub-stripes is set in
3421 * lod_xattr_set_lmv(). */
3422 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
3425 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3426 if (!lo->ldo_stripe[i])
3429 if (!dt_object_exists(lo->ldo_stripe[i]))
3432 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
3442 * Delete an extended attribute.
3444 * Deletes specified xattr from the object and the stripes if the latter exist.
3446 * \param[in] env execution environment
3447 * \param[in] dt object
3448 * \param[in] name name of xattr
3449 * \param[in] th transaction handle
3451 * \retval 0 on success
3452 * \retval negative if failed
3454 static int lod_xattr_del_internal(const struct lu_env *env,
3455 struct dt_object *dt,
3456 const char *name, struct thandle *th)
3458 struct dt_object *next = dt_object_child(dt);
3459 struct lod_object *lo = lod_dt_obj(dt);
3464 rc = lod_sub_xattr_del(env, next, name, th);
3465 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3468 if (lo->ldo_dir_stripe_count == 0)
3471 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3472 LASSERT(lo->ldo_stripe[i]);
3474 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3483 * Set default striping on a directory.
3485 * Sets specified striping on a directory object unless it matches the default
3486 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3487 * EA. This striping will be used when regular file is being created in this
3490 * \param[in] env execution environment
3491 * \param[in] dt the striped object
3492 * \param[in] buf buffer with the striping
3493 * \param[in] name name of EA
3494 * \param[in] fl xattr flag (see OSD API description)
3495 * \param[in] th transaction handle
3497 * \retval 0 on success
3498 * \retval negative if failed
3500 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
3501 struct dt_object *dt,
3502 const struct lu_buf *buf,
3503 const char *name, int fl,
3506 struct lov_user_md_v1 *lum;
3507 struct lov_user_md_v3 *v3 = NULL;
3508 const char *pool_name = NULL;
3513 LASSERT(buf != NULL && buf->lb_buf != NULL);
3516 switch (lum->lmm_magic) {
3517 case LOV_USER_MAGIC_SPECIFIC:
3518 case LOV_USER_MAGIC_V3:
3520 if (v3->lmm_pool_name[0] != '\0')
3521 pool_name = v3->lmm_pool_name;
3523 case LOV_USER_MAGIC_V1:
3524 /* if { size, offset, count } = { 0, -1, 0 } and no pool
3525 * (i.e. all default values specified) then delete default
3526 * striping from dir. */
3528 "set default striping: sz %u # %u offset %d %s %s\n",
3529 (unsigned)lum->lmm_stripe_size,
3530 (unsigned)lum->lmm_stripe_count,
3531 (int)lum->lmm_stripe_offset,
3532 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
3534 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
3535 lum->lmm_stripe_count,
3536 lum->lmm_stripe_offset,
3539 case LOV_USER_MAGIC_COMP_V1:
3541 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lum;
3542 struct lov_comp_md_entry_v1 *lcme;
3545 comp_cnt = le16_to_cpu(lcm->lcm_entry_count);
3546 for (i = 0; i < comp_cnt; i++) {
3547 lcme = &lcm->lcm_entries[i];
3548 if (lcme->lcme_flags & cpu_to_le32(LCME_FL_EXTENSION)) {
3549 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
3558 CERROR("Invalid magic %x\n", lum->lmm_magic);
3563 rc = lod_xattr_del_internal(env, dt, name, th);
3567 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3574 * Set default striping on a directory object.
3576 * Sets specified striping on a directory object unless it matches the default
3577 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3578 * EA. This striping will be used when a new directory is being created in the
3581 * \param[in] env execution environment
3582 * \param[in] dt the striped object
3583 * \param[in] buf buffer with the striping
3584 * \param[in] name name of EA
3585 * \param[in] fl xattr flag (see OSD API description)
3586 * \param[in] th transaction handle
3588 * \retval 0 on success
3589 * \retval negative if failed
3591 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
3592 struct dt_object *dt,
3593 const struct lu_buf *buf,
3594 const char *name, int fl,
3597 struct lmv_user_md_v1 *lum;
3602 LASSERT(buf != NULL && buf->lb_buf != NULL);
3606 "set default stripe_count # %u stripe_offset %d hash %u\n",
3607 le32_to_cpu(lum->lum_stripe_count),
3608 (int)le32_to_cpu(lum->lum_stripe_offset),
3609 le32_to_cpu(lum->lum_hash_type));
3611 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
3612 le32_to_cpu(lum->lum_stripe_offset)) &&
3613 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
3614 rc = lod_xattr_del_internal(env, dt, name, th);
3618 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3627 * Turn directory into a striped directory.
3629 * During replay the client sends the striping created before MDT
3630 * failure, then the layer above LOD sends this defined striping
3631 * using ->do_xattr_set(), so LOD uses this method to replay creation
3632 * of the stripes. Notice the original information for the striping
3633 * (#stripes, FIDs, etc) was transferred in declare path.
3635 * \param[in] env execution environment
3636 * \param[in] dt the striped object
3637 * \param[in] buf not used currently
3638 * \param[in] name not used currently
3639 * \param[in] fl xattr flag (see OSD API description)
3640 * \param[in] th transaction handle
3642 * \retval 0 on success
3643 * \retval negative if failed
3645 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
3646 const struct lu_buf *buf, const char *name,
3647 int fl, struct thandle *th)
3649 struct lod_object *lo = lod_dt_obj(dt);
3650 struct lod_thread_info *info = lod_env_info(env);
3651 struct lu_attr *attr = &info->lti_attr;
3652 struct dt_object_format *dof = &info->lti_format;
3653 struct lu_buf lmv_buf;
3654 struct lu_buf slave_lmv_buf;
3655 struct lmv_mds_md_v1 *lmm;
3656 struct lmv_mds_md_v1 *slave_lmm = NULL;
3657 struct dt_insert_rec *rec = &info->lti_dt_rec;
3662 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3665 /* The stripes are supposed to be allocated in declare phase,
3666 * if there are no stripes being allocated, it will skip */
3667 if (lo->ldo_dir_stripe_count == 0) {
3668 if (lo->ldo_dir_is_foreign) {
3669 rc = lod_sub_xattr_set(env, dt_object_child(dt), buf,
3670 XATTR_NAME_LMV, fl, th);
3677 rc = dt_attr_get(env, dt_object_child(dt), attr);
3681 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME |
3682 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
3683 dof->dof_type = DFT_DIR;
3685 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
3688 lmm = lmv_buf.lb_buf;
3690 OBD_ALLOC_PTR(slave_lmm);
3691 if (slave_lmm == NULL)
3694 lod_prep_slave_lmv_md(slave_lmm, lmm);
3695 slave_lmv_buf.lb_buf = slave_lmm;
3696 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
3698 rec->rec_type = S_IFDIR;
3699 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3700 struct dt_object *dto = lo->ldo_stripe[i];
3701 char *stripe_name = info->lti_key;
3702 struct lu_name *sname;
3703 struct linkea_data ldata = { NULL };
3704 struct lu_buf linkea_buf;
3706 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
3710 /* fail a remote stripe creation */
3711 if (i && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_CREATE))
3714 /* if it's source stripe of migrating directory, don't create */
3715 if (!((lo->ldo_dir_hash_type & LMV_HASH_FLAG_MIGRATION) &&
3716 i >= lo->ldo_dir_migrate_offset)) {
3717 dt_write_lock(env, dto, DT_TGT_CHILD);
3718 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
3720 dt_write_unlock(env, dto);
3724 rc = lod_sub_ref_add(env, dto, th);
3725 dt_write_unlock(env, dto);
3729 rec->rec_fid = lu_object_fid(&dto->do_lu);
3730 rc = lod_sub_insert(env, dto,
3731 (const struct dt_rec *)rec,
3732 (const struct dt_key *)dot, th);
3737 rec->rec_fid = lu_object_fid(&dt->do_lu);
3738 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
3739 (const struct dt_key *)dotdot, th);
3743 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
3744 cfs_fail_val != i) {
3745 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
3747 slave_lmm->lmv_master_mdt_index =
3750 slave_lmm->lmv_master_mdt_index =
3753 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
3754 XATTR_NAME_LMV, 0, th);
3759 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
3761 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3762 PFID(lu_object_fid(&dto->do_lu)), i + 1);
3764 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3765 PFID(lu_object_fid(&dto->do_lu)), i);
3767 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
3768 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
3769 sname, lu_object_fid(&dt->do_lu));
3773 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
3774 linkea_buf.lb_len = ldata.ld_leh->leh_len;
3775 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
3776 XATTR_NAME_LINK, 0, th);
3780 rec->rec_fid = lu_object_fid(&dto->do_lu);
3781 rc = lod_sub_insert(env, dt_object_child(dt),
3782 (const struct dt_rec *)rec,
3783 (const struct dt_key *)stripe_name, th);
3787 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
3792 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
3793 rc = lod_sub_xattr_set(env, dt_object_child(dt),
3794 &lmv_buf, XATTR_NAME_LMV, fl, th);
3796 if (slave_lmm != NULL)
3797 OBD_FREE_PTR(slave_lmm);
3803 * Helper function to declare/execute creation of a striped directory
3805 * Called in declare/create object path, prepare striping for a directory
3806 * and prepare defaults data striping for the objects to be created in
3807 * that directory. Notice the function calls "declaration" or "execution"
3808 * methods depending on \a declare param. This is a consequence of the
3809 * current approach while we don't have natural distributed transactions:
3810 * we basically execute non-local updates in the declare phase. So, the
3811 * arguments for the both phases are the same and this is the reason for
3812 * this function to exist.
3814 * \param[in] env execution environment
3815 * \param[in] dt object
3816 * \param[in] attr attributes the stripes will be created with
3817 * \param[in] lmu lmv_user_md if MDT indices are specified
3818 * \param[in] dof format of stripes (see OSD API description)
3819 * \param[in] th transaction handle
3820 * \param[in] declare where to call "declare" or "execute" methods
3822 * \retval 0 on success
3823 * \retval negative if failed
3825 static int lod_dir_striping_create_internal(const struct lu_env *env,
3826 struct dt_object *dt,
3827 struct lu_attr *attr,
3828 const struct lu_buf *lmu,
3829 struct dt_object_format *dof,
3833 struct lod_thread_info *info = lod_env_info(env);
3834 struct lod_object *lo = lod_dt_obj(dt);
3835 const struct lod_default_striping *lds = lo->ldo_def_striping;
3839 LASSERT(ergo(lds != NULL,
3840 lds->lds_def_striping_set ||
3841 lds->lds_dir_def_striping_set));
3843 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
3844 lo->ldo_dir_stripe_offset)) {
3846 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3847 int stripe_count = lo->ldo_dir_stripe_count;
3849 if (info->lti_ea_store_size < sizeof(*v1)) {
3850 rc = lod_ea_store_resize(info, sizeof(*v1));
3853 v1 = info->lti_ea_store;
3856 memset(v1, 0, sizeof(*v1));
3857 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3858 v1->lum_stripe_count = cpu_to_le32(stripe_count);
3859 v1->lum_stripe_offset =
3860 cpu_to_le32(lo->ldo_dir_stripe_offset);
3862 info->lti_buf.lb_buf = v1;
3863 info->lti_buf.lb_len = sizeof(*v1);
3864 lmu = &info->lti_buf;
3868 rc = lod_declare_xattr_set_lmv(env, dt, attr, lmu, dof,
3871 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
3876 /* foreign LMV EA case */
3878 struct lmv_foreign_md *lfm = lmu->lb_buf;
3880 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
3881 rc = lod_declare_xattr_set_lmv(env, dt, attr,
3885 if (lo->ldo_dir_is_foreign) {
3886 LASSERT(lo->ldo_foreign_lmv != NULL &&
3887 lo->ldo_foreign_lmv_size > 0);
3888 info->lti_buf.lb_buf = lo->ldo_foreign_lmv;
3889 info->lti_buf.lb_len = lo->ldo_foreign_lmv_size;
3890 lmu = &info->lti_buf;
3891 rc = lod_xattr_set_lmv(env, dt, lmu,
3892 XATTR_NAME_LMV, 0, th);
3897 /* Transfer default LMV striping from the parent */
3898 if (lds != NULL && lds->lds_dir_def_striping_set &&
3899 !(LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
3900 lds->lds_dir_def_stripe_offset) &&
3901 le32_to_cpu(lds->lds_dir_def_hash_type) !=
3902 LMV_HASH_TYPE_UNKNOWN)) {
3903 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3905 if (info->lti_ea_store_size < sizeof(*v1)) {
3906 rc = lod_ea_store_resize(info, sizeof(*v1));
3909 v1 = info->lti_ea_store;
3912 memset(v1, 0, sizeof(*v1));
3913 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3914 v1->lum_stripe_count =
3915 cpu_to_le32(lds->lds_dir_def_stripe_count);
3916 v1->lum_stripe_offset =
3917 cpu_to_le32(lds->lds_dir_def_stripe_offset);
3919 cpu_to_le32(lds->lds_dir_def_hash_type);
3921 info->lti_buf.lb_buf = v1;
3922 info->lti_buf.lb_len = sizeof(*v1);
3924 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3925 XATTR_NAME_DEFAULT_LMV,
3928 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
3930 XATTR_NAME_DEFAULT_LMV, 0,
3936 /* Transfer default LOV striping from the parent */
3937 if (lds != NULL && lds->lds_def_striping_set &&
3938 lds->lds_def_comp_cnt != 0) {
3939 struct lov_mds_md *lmm;
3940 int lmm_size = lod_comp_md_size(lo, true);
3942 if (info->lti_ea_store_size < lmm_size) {
3943 rc = lod_ea_store_resize(info, lmm_size);
3947 lmm = info->lti_ea_store;
3949 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
3953 info->lti_buf.lb_buf = lmm;
3954 info->lti_buf.lb_len = lmm_size;
3957 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3958 XATTR_NAME_LOV, 0, th);
3960 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
3961 XATTR_NAME_LOV, 0, th);
3969 static int lod_declare_dir_striping_create(const struct lu_env *env,
3970 struct dt_object *dt,
3971 struct lu_attr *attr,
3973 struct dt_object_format *dof,
3976 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
3980 static int lod_dir_striping_create(const struct lu_env *env,
3981 struct dt_object *dt,
3982 struct lu_attr *attr,
3983 struct dt_object_format *dof,
3986 return lod_dir_striping_create_internal(env, dt, attr, NULL, dof, th,
3991 * Make LOV EA for striped object.
3993 * Generate striping information and store it in the LOV EA of the given
3994 * object. The caller must ensure nobody else is calling the function
3995 * against the object concurrently. The transaction must be started.
3996 * FLDB service must be running as well; it's used to map FID to the target,
3997 * which is stored in LOV EA.
3999 * \param[in] env execution environment for this thread
4000 * \param[in] lo LOD object
4001 * \param[in] th transaction handle
4003 * \retval 0 if LOV EA is stored successfully
4004 * \retval negative error number on failure
4006 static int lod_generate_and_set_lovea(const struct lu_env *env,
4007 struct lod_object *lo,
4010 struct lod_thread_info *info = lod_env_info(env);
4011 struct dt_object *next = dt_object_child(&lo->ldo_obj);
4012 struct lov_mds_md_v1 *lmm;
4018 if (lo->ldo_comp_cnt == 0 && !lo->ldo_is_foreign) {
4019 lod_striping_free(env, lo);
4020 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
4024 lmm_size = lod_comp_md_size(lo, false);
4025 if (info->lti_ea_store_size < lmm_size) {
4026 rc = lod_ea_store_resize(info, lmm_size);
4030 lmm = info->lti_ea_store;
4032 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
4036 info->lti_buf.lb_buf = lmm;
4037 info->lti_buf.lb_len = lmm_size;
4038 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
4039 XATTR_NAME_LOV, 0, th);
4043 static __u32 lod_gen_component_id(struct lod_object *lo,
4044 int mirror_id, int comp_idx);
4047 * Repeat an existing component
4049 * Creates a new layout by replicating an existing component. Uses striping
4050 * policy from previous component as a template for the striping for the new
4053 * New component starts with zero length, will be extended (or removed) before
4054 * returning layout to client.
4056 * NB: Reallocates layout components array (lo->ldo_comp_entries), invalidating
4057 * any pre-existing pointers to components. Handle with care.
4059 * \param[in] env execution environment for this thread
4060 * \param[in,out] lo object to update the layout of
4061 * \param[in] index index of component to copy
4063 * \retval 0 on success
4064 * \retval negative errno on error
4066 static int lod_layout_repeat_comp(const struct lu_env *env,
4067 struct lod_object *lo, int index)
4069 struct lod_layout_component *lod_comp;
4070 struct lod_layout_component *new_comp = NULL;
4071 struct lod_layout_component *comp_array;
4072 int rc = 0, i, new_cnt = lo->ldo_comp_cnt + 1;
4077 lod_comp = &lo->ldo_comp_entries[index];
4078 LASSERT(lod_comp_inited(lod_comp) && lod_comp->llc_id != LCME_ID_INVAL);
4080 CDEBUG(D_LAYOUT, "repeating component %d\n", index);
4082 OBD_ALLOC(comp_array, sizeof(*comp_array) * new_cnt);
4083 if (comp_array == NULL)
4084 GOTO(out, rc = -ENOMEM);
4086 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4087 memcpy(&comp_array[i + offset], &lo->ldo_comp_entries[i],
4088 sizeof(*comp_array));
4090 /* Duplicate this component in to the next slot */
4092 new_comp = &comp_array[i + 1];
4093 memcpy(&comp_array[i + 1], &lo->ldo_comp_entries[i],
4094 sizeof(*comp_array));
4095 /* We must now skip this new component when copying */
4100 /* Set up copied component */
4101 new_comp->llc_flags &= ~LCME_FL_INIT;
4102 new_comp->llc_stripe = NULL;
4103 new_comp->llc_stripes_allocated = 0;
4104 new_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4105 /* for uninstantiated components, layout gen stores default stripe
4107 new_comp->llc_layout_gen = lod_comp->llc_stripe_offset;
4108 /* This makes the repeated component zero-length, placed at the end of
4109 * the preceding component */
4110 new_comp->llc_extent.e_start = new_comp->llc_extent.e_end;
4111 new_comp->llc_timestamp = lod_comp->llc_timestamp;
4112 new_comp->llc_pool = NULL;
4114 rc = lod_set_pool(&new_comp->llc_pool, lod_comp->llc_pool);
4118 if (new_comp->llc_ostlist.op_array) {
4119 __u32 *op_array = NULL;
4121 OBD_ALLOC(op_array, new_comp->llc_ostlist.op_size);
4123 GOTO(out, rc = -ENOMEM);
4124 memcpy(op_array, &new_comp->llc_ostlist.op_array,
4125 new_comp->llc_ostlist.op_size);
4126 new_comp->llc_ostlist.op_array = op_array;
4129 OBD_FREE(lo->ldo_comp_entries,
4130 sizeof(*comp_array) * lo->ldo_comp_cnt);
4131 lo->ldo_comp_entries = comp_array;
4132 lo->ldo_comp_cnt = new_cnt;
4134 /* Generate an id for the new component */
4135 mirror_id = mirror_id_of(new_comp->llc_id);
4136 new_comp->llc_id = LCME_ID_INVAL;
4137 new_comp->llc_id = lod_gen_component_id(lo, mirror_id, index + 1);
4138 if (new_comp->llc_id == LCME_ID_INVAL)
4139 GOTO(out, rc = -ERANGE);
4144 OBD_FREE(comp_array, sizeof(*comp_array) * new_cnt);
4149 static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
4153 /* clear memory region that will be used for layout change */
4154 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
4155 info->lti_count = 0;
4157 if (info->lti_comp_size >= comp_cnt)
4160 if (info->lti_comp_size > 0) {
4161 OBD_FREE(info->lti_comp_idx,
4162 info->lti_comp_size * sizeof(__u32));
4163 info->lti_comp_size = 0;
4166 OBD_ALLOC(info->lti_comp_idx, comp_cnt * sizeof(__u32));
4167 if (!info->lti_comp_idx)
4170 info->lti_comp_size = comp_cnt;
4175 * Prepare new layout minus deleted components
4177 * Removes components marked for deletion (LCME_ID_INVAL) by copying to a new
4178 * layout and skipping those components. Removes stripe objects if any exist.
4181 * Reallocates layout components array (lo->ldo_comp_entries), invalidating
4182 * any pre-existing pointers to components.
4184 * Caller is responsible for updating mirror end (ldo_mirror[].lme_end).
4186 * \param[in] env execution environment for this thread
4187 * \param[in,out] lo object to update the layout of
4188 * \param[in] th transaction handle for this operation
4190 * \retval # of components deleted
4191 * \retval negative errno on error
4193 static int lod_layout_del_prep_layout(const struct lu_env *env,
4194 struct lod_object *lo,
4197 struct lod_layout_component *lod_comp;
4198 struct lod_thread_info *info = lod_env_info(env);
4199 int rc = 0, i, j, deleted = 0;
4203 LASSERT(lo->ldo_is_composite);
4204 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
4206 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
4210 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4211 lod_comp = &lo->ldo_comp_entries[i];
4213 if (lod_comp->llc_id != LCME_ID_INVAL) {
4214 /* Build array of things to keep */
4215 info->lti_comp_idx[info->lti_count++] = i;
4219 lod_obj_set_pool(lo, i, NULL);
4220 if (lod_comp->llc_ostlist.op_array) {
4221 OBD_FREE(lod_comp->llc_ostlist.op_array,
4222 lod_comp->llc_ostlist.op_size);
4223 lod_comp->llc_ostlist.op_array = NULL;
4224 lod_comp->llc_ostlist.op_size = 0;
4228 CDEBUG(D_LAYOUT, "deleting comp %d, left %d\n", i,
4229 lo->ldo_comp_cnt - deleted);
4231 /* No striping info for this component */
4232 if (lod_comp->llc_stripe == NULL)
4235 LASSERT(lod_comp->llc_stripe_count > 0);
4236 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4237 struct dt_object *obj = lod_comp->llc_stripe[j];
4242 /* components which are not init have no sub objects
4244 if (lod_comp_inited(lod_comp)) {
4245 rc = lod_sub_destroy(env, obj, th);
4250 lu_object_put(env, &obj->do_lu);
4251 lod_comp->llc_stripe[j] = NULL;
4253 OBD_FREE(lod_comp->llc_stripe, sizeof(*lod_comp->llc_stripe) *
4254 lod_comp->llc_stripes_allocated);
4255 lod_comp->llc_stripe = NULL;
4256 OBD_FREE(lod_comp->llc_ost_indices,
4257 sizeof(__u32) * lod_comp->llc_stripes_allocated);
4258 lod_comp->llc_ost_indices = NULL;
4259 lod_comp->llc_stripes_allocated = 0;
4262 /* info->lti_count has the amount of left components */
4263 LASSERTF(info->lti_count >= 0 && info->lti_count < lo->ldo_comp_cnt,
4264 "left = %d, lo->ldo_comp_cnt %d\n", (int)info->lti_count,
4265 (int)lo->ldo_comp_cnt);
4267 if (info->lti_count > 0) {
4268 struct lod_layout_component *comp_array;
4270 OBD_ALLOC(comp_array, sizeof(*comp_array) * info->lti_count);
4271 if (comp_array == NULL)
4272 GOTO(out, rc = -ENOMEM);
4274 for (i = 0; i < info->lti_count; i++) {
4275 memcpy(&comp_array[i],
4276 &lo->ldo_comp_entries[info->lti_comp_idx[i]],
4277 sizeof(*comp_array));
4280 OBD_FREE(lo->ldo_comp_entries,
4281 sizeof(*comp_array) * lo->ldo_comp_cnt);
4282 lo->ldo_comp_entries = comp_array;
4283 lo->ldo_comp_cnt = info->lti_count;
4285 lod_free_comp_entries(lo);
4290 return rc ? rc : deleted;
4294 * Delete layout component(s)
4296 * This function sets up the layout data in the env and does the setattrs
4297 * required to write out the new layout. The layout itself is modified in
4298 * lod_layout_del_prep_layout.
4300 * \param[in] env execution environment for this thread
4301 * \param[in] dt object
4302 * \param[in] th transaction handle
4304 * \retval 0 on success
4305 * \retval negative error number on failure
4307 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
4310 struct lod_object *lo = lod_dt_obj(dt);
4311 struct dt_object *next = dt_object_child(dt);
4312 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4315 LASSERT(lo->ldo_mirror_count == 1);
4317 rc = lod_layout_del_prep_layout(env, lo, th);
4321 /* Only do this if we didn't delete all components */
4322 if (lo->ldo_comp_cnt > 0) {
4323 lo->ldo_mirrors[0].lme_end = lo->ldo_comp_cnt - 1;
4324 lod_obj_inc_layout_gen(lo);
4327 LASSERT(dt_object_exists(dt));
4328 rc = dt_attr_get(env, next, attr);
4332 if (attr->la_size > 0) {
4334 attr->la_valid = LA_SIZE;
4335 rc = lod_sub_attr_set(env, next, attr, th);
4340 rc = lod_generate_and_set_lovea(env, lo, th);
4344 lod_striping_free(env, lo);
4349 static int lod_get_default_lov_striping(const struct lu_env *env,
4350 struct lod_object *lo,
4351 struct lod_default_striping *lds,
4352 struct dt_allocation_hint *ah);
4354 * Implementation of dt_object_operations::do_xattr_set.
4356 * Sets specified extended attribute on the object. Three types of EAs are
4358 * LOV EA - stores striping for a regular file or default striping (when set
4360 * LMV EA - stores a marker for the striped directories
4361 * DMV EA - stores default directory striping
4363 * When striping is applied to a non-striped existing object (this is called
4364 * late striping), then LOD notices the caller wants to turn the object into a
4365 * striped one. The stripe objects are created and appropriate EA is set:
4366 * LOV EA storing all the stripes directly or LMV EA storing just a small header
4367 * with striping configuration.
4369 * \see dt_object_operations::do_xattr_set() in the API description for details.
4371 static int lod_xattr_set(const struct lu_env *env,
4372 struct dt_object *dt, const struct lu_buf *buf,
4373 const char *name, int fl, struct thandle *th)
4375 struct dt_object *next = dt_object_child(dt);
4380 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4381 !strcmp(name, XATTR_NAME_LMV)) {
4383 case LU_XATTR_CREATE:
4384 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
4387 case LU_XATTR_REPLACE:
4388 rc = lod_dir_layout_set(env, dt, buf, fl, th);
4395 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4396 strcmp(name, XATTR_NAME_LOV) == 0) {
4397 struct lod_default_striping *lds = lod_lds_buf_get(env);
4398 struct lov_user_md_v1 *v1 = buf->lb_buf;
4399 char pool[LOV_MAXPOOLNAME + 1];
4402 /* get existing striping config */
4403 rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds,
4408 memset(pool, 0, sizeof(pool));
4409 if (lds->lds_def_striping_set == 1)
4410 lod_layout_get_pool(lds->lds_def_comp_entries,
4411 lds->lds_def_comp_cnt, pool,
4414 is_del = LOVEA_DELETE_VALUES(v1->lmm_stripe_size,
4415 v1->lmm_stripe_count,
4416 v1->lmm_stripe_offset,
4419 /* Retain the pool name if it is not given */
4420 if (v1->lmm_magic == LOV_USER_MAGIC_V1 && pool[0] != '\0' &&
4422 struct lod_thread_info *info = lod_env_info(env);
4423 struct lov_user_md_v3 *v3 = info->lti_ea_store;
4425 memset(v3, 0, sizeof(*v3));
4426 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4427 v3->lmm_pattern = cpu_to_le32(v1->lmm_pattern);
4428 v3->lmm_stripe_count =
4429 cpu_to_le32(v1->lmm_stripe_count);
4430 v3->lmm_stripe_offset =
4431 cpu_to_le32(v1->lmm_stripe_offset);
4432 v3->lmm_stripe_size = cpu_to_le32(v1->lmm_stripe_size);
4434 strlcpy(v3->lmm_pool_name, pool,
4435 sizeof(v3->lmm_pool_name));
4437 info->lti_buf.lb_buf = v3;
4438 info->lti_buf.lb_len = sizeof(*v3);
4439 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4442 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name,
4446 if (lds->lds_def_striping_set == 1 &&
4447 lds->lds_def_comp_entries != NULL)
4448 lod_free_def_comp_entries(lds);
4451 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4452 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
4454 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
4457 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
4458 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4459 strcmp(name, XATTR_LUSTRE_LOV) == 0 ||
4460 allowed_lustre_lov(name))) {
4461 /* in case of lov EA swap, just set it
4462 * if not, it is a replay so check striping match what we
4463 * already have during req replay, declare_xattr_set()
4464 * defines striping, then create() does the work */
4465 if (fl & LU_XATTR_REPLACE) {
4466 /* free stripes, then update disk */
4467 lod_striping_free(env, lod_dt_obj(dt));
4469 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
4470 } else if (dt_object_remote(dt)) {
4471 /* This only happens during migration, see
4472 * mdd_migrate_create(), in which Master MDT will
4473 * create a remote target object, and only set
4474 * (migrating) stripe EA on the remote object,
4475 * and does not need creating each stripes. */
4476 rc = lod_sub_xattr_set(env, next, buf, name,
4478 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
4479 /* delete component(s) */
4480 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
4481 rc = lod_layout_del(env, dt, th);
4484 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
4485 * it's going to create create file with specified
4486 * component(s), the striping must have not being
4487 * cached in this case;
4489 * Otherwise, it's going to add/change component(s) to
4490 * an existing file, the striping must have been cached
4493 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
4494 !strcmp(name, XATTR_NAME_LOV),
4495 !lod_dt_obj(dt)->ldo_comp_cached));
4497 rc = lod_striped_create(env, dt, NULL, NULL, th);
4500 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4501 rc = lod_replace_parent_fid(env, dt, buf, th, false);
4506 /* then all other xattr */
4507 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4513 * Implementation of dt_object_operations::do_declare_xattr_del.
4515 * \see dt_object_operations::do_declare_xattr_del() in the API description
4518 static int lod_declare_xattr_del(const struct lu_env *env,
4519 struct dt_object *dt, const char *name,
4522 struct lod_object *lo = lod_dt_obj(dt);
4523 struct dt_object *next = dt_object_child(dt);
4528 rc = lod_sub_declare_xattr_del(env, next, name, th);
4532 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4535 /* NB: don't delete stripe LMV, because when we do this, normally we
4536 * will remove stripes, besides, if directory LMV is corrupt, this will
4537 * prevent deleting its LMV and fixing it (via LFSCK).
4539 if (!strcmp(name, XATTR_NAME_LMV))
4542 rc = lod_striping_load(env, lo);
4546 if (lo->ldo_dir_stripe_count == 0)
4549 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4550 struct dt_object *dto = lo->ldo_stripe[i];
4555 rc = lod_sub_declare_xattr_del(env, dto, name, th);
4564 * Implementation of dt_object_operations::do_xattr_del.
4566 * If EA storing a regular striping is being deleted, then release
4567 * all the references to the stripe objects in core.
4569 * \see dt_object_operations::do_xattr_del() in the API description for details.
4571 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
4572 const char *name, struct thandle *th)
4574 struct dt_object *next = dt_object_child(dt);
4575 struct lod_object *lo = lod_dt_obj(dt);
4580 if (!strcmp(name, XATTR_NAME_LOV) || !strcmp(name, XATTR_NAME_LMV))
4581 lod_striping_free(env, lod_dt_obj(dt));
4583 rc = lod_sub_xattr_del(env, next, name, th);
4584 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
4587 if (!strcmp(name, XATTR_NAME_LMV))
4590 if (lo->ldo_dir_stripe_count == 0)
4593 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4594 struct dt_object *dto = lo->ldo_stripe[i];
4599 rc = lod_sub_xattr_del(env, dto, name, th);
4608 * Implementation of dt_object_operations::do_xattr_list.
4610 * \see dt_object_operations::do_xattr_list() in the API description
4613 static int lod_xattr_list(const struct lu_env *env,
4614 struct dt_object *dt, const struct lu_buf *buf)
4616 return dt_xattr_list(env, dt_object_child(dt), buf);
4619 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
4621 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
4625 * Copy OST list from layout provided by user.
4627 * \param[in] lod_comp layout_component to be filled
4628 * \param[in] v3 LOV EA V3 user data
4630 * \retval 0 on success
4631 * \retval negative if failed
4633 int lod_comp_copy_ost_lists(struct lod_layout_component *lod_comp,
4634 struct lov_user_md_v3 *v3)
4640 if (v3->lmm_stripe_offset == LOV_OFFSET_DEFAULT)
4641 v3->lmm_stripe_offset = v3->lmm_objects[0].l_ost_idx;
4643 if (lod_comp->llc_ostlist.op_array) {
4644 if (lod_comp->llc_ostlist.op_size >=
4645 v3->lmm_stripe_count * sizeof(__u32)) {
4646 lod_comp->llc_ostlist.op_count =
4647 v3->lmm_stripe_count;
4650 OBD_FREE(lod_comp->llc_ostlist.op_array,
4651 lod_comp->llc_ostlist.op_size);
4654 /* copy ost list from lmm */
4655 lod_comp->llc_ostlist.op_count = v3->lmm_stripe_count;
4656 lod_comp->llc_ostlist.op_size = v3->lmm_stripe_count * sizeof(__u32);
4657 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
4658 lod_comp->llc_ostlist.op_size);
4659 if (!lod_comp->llc_ostlist.op_array)
4662 for (j = 0; j < v3->lmm_stripe_count; j++) {
4663 lod_comp->llc_ostlist.op_array[j] =
4664 v3->lmm_objects[j].l_ost_idx;
4672 * Get default striping.
4674 * \param[in] env execution environment
4675 * \param[in] lo object
4676 * \param[out] lds default striping
4678 * \retval 0 on success
4679 * \retval negative if failed
4681 static int lod_get_default_lov_striping(const struct lu_env *env,
4682 struct lod_object *lo,
4683 struct lod_default_striping *lds,
4684 struct dt_allocation_hint *ah)
4686 struct lod_thread_info *info = lod_env_info(env);
4687 struct lov_user_md_v1 *v1 = NULL;
4688 struct lov_user_md_v3 *v3 = NULL;
4689 struct lov_comp_md_v1 *comp_v1 = NULL;
4697 rc = lod_get_lov_ea(env, lo);
4701 if (rc < (typeof(rc))sizeof(struct lov_user_md))
4704 v1 = info->lti_ea_store;
4705 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
4706 lustre_swab_lov_user_md_v1(v1);
4707 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
4708 v3 = (struct lov_user_md_v3 *)v1;
4709 lustre_swab_lov_user_md_v3(v3);
4710 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_SPECIFIC)) {
4711 v3 = (struct lov_user_md_v3 *)v1;
4712 lustre_swab_lov_user_md_v3(v3);
4713 lustre_swab_lov_user_md_objects(v3->lmm_objects,
4714 v3->lmm_stripe_count);
4715 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_COMP_V1) ||
4716 v1->lmm_magic == __swab32(LOV_USER_MAGIC_SEL)) {
4717 comp_v1 = (struct lov_comp_md_v1 *)v1;
4718 lustre_swab_lov_comp_md_v1(comp_v1);
4721 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1 &&
4722 v1->lmm_magic != LOV_MAGIC_COMP_V1 &&
4723 v1->lmm_magic != LOV_MAGIC_SEL &&
4724 v1->lmm_magic != LOV_USER_MAGIC_SPECIFIC)
4727 if ((v1->lmm_magic == LOV_MAGIC_COMP_V1 ||
4728 v1->lmm_magic == LOV_MAGIC_SEL) &&
4729 !(ah && ah->dah_append_stripes)) {
4730 comp_v1 = (struct lov_comp_md_v1 *)v1;
4731 comp_cnt = comp_v1->lcm_entry_count;
4734 mirror_cnt = comp_v1->lcm_mirror_count + 1;
4742 /* realloc default comp entries if necessary */
4743 rc = lod_def_striping_comp_resize(lds, comp_cnt);
4747 lds->lds_def_comp_cnt = comp_cnt;
4748 lds->lds_def_striping_is_composite = composite;
4749 lds->lds_def_mirror_cnt = mirror_cnt;
4751 for (i = 0; i < comp_cnt; i++) {
4752 struct lod_layout_component *lod_comp;
4755 lod_comp = &lds->lds_def_comp_entries[i];
4757 * reset lod_comp values, llc_stripes is always NULL in
4758 * the default striping template, llc_pool will be reset
4761 memset(lod_comp, 0, offsetof(typeof(*lod_comp), llc_pool));
4764 v1 = (struct lov_user_md *)((char *)comp_v1 +
4765 comp_v1->lcm_entries[i].lcme_offset);
4766 lod_comp->llc_extent =
4767 comp_v1->lcm_entries[i].lcme_extent;
4768 /* We only inherit certain flags from the layout */
4769 lod_comp->llc_flags =
4770 comp_v1->lcm_entries[i].lcme_flags &
4771 LCME_TEMPLATE_FLAGS;
4774 if (!lov_pattern_supported(v1->lmm_pattern) &&
4775 !(v1->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
4776 lod_free_def_comp_entries(lds);
4780 CDEBUG(D_LAYOUT, DFID" stripe_count=%d stripe_size=%d stripe_offset=%d append_stripes=%d\n",
4781 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
4782 (int)v1->lmm_stripe_count, (int)v1->lmm_stripe_size,
4783 (int)v1->lmm_stripe_offset,
4784 ah ? ah->dah_append_stripes : 0);
4786 if (ah && ah->dah_append_stripes)
4787 lod_comp->llc_stripe_count = ah->dah_append_stripes;
4789 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
4790 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
4791 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
4792 lod_comp->llc_pattern = v1->lmm_pattern;
4795 if (ah && ah->dah_append_pool && ah->dah_append_pool[0]) {
4796 pool = ah->dah_append_pool;
4797 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
4798 /* XXX: sanity check here */
4799 v3 = (struct lov_user_md_v3 *) v1;
4800 if (v3->lmm_pool_name[0] != '\0')
4801 pool = v3->lmm_pool_name;
4803 lod_set_def_pool(lds, i, pool);
4804 if (v1->lmm_magic == LOV_USER_MAGIC_SPECIFIC) {
4805 v3 = (struct lov_user_md_v3 *)v1;
4806 rc = lod_comp_copy_ost_lists(lod_comp, v3);
4809 } else if (lod_comp->llc_ostlist.op_array &&
4810 lod_comp->llc_ostlist.op_count) {
4811 for (j = 0; j < lod_comp->llc_ostlist.op_count; j++)
4812 lod_comp->llc_ostlist.op_array[j] = -1;
4813 lod_comp->llc_ostlist.op_count = 0;
4817 lds->lds_def_striping_set = 1;
4822 * Get default directory striping.
4824 * \param[in] env execution environment
4825 * \param[in] lo object
4826 * \param[out] lds default striping
4828 * \retval 0 on success
4829 * \retval negative if failed
4831 static int lod_get_default_lmv_striping(const struct lu_env *env,
4832 struct lod_object *lo,
4833 struct lod_default_striping *lds)
4835 struct lmv_user_md *lmu;
4838 lds->lds_dir_def_striping_set = 0;
4840 rc = lod_get_default_lmv_ea(env, lo);
4844 if (rc >= (int)sizeof(*lmu)) {
4845 struct lod_thread_info *info = lod_env_info(env);
4847 lmu = info->lti_ea_store;
4849 lds->lds_dir_def_stripe_count =
4850 le32_to_cpu(lmu->lum_stripe_count);
4851 lds->lds_dir_def_stripe_offset =
4852 le32_to_cpu(lmu->lum_stripe_offset);
4853 lds->lds_dir_def_hash_type =
4854 le32_to_cpu(lmu->lum_hash_type);
4855 lds->lds_dir_def_striping_set = 1;
4862 * Get default striping in the object.
4864 * Get object default striping and default directory striping.
4866 * \param[in] env execution environment
4867 * \param[in] lo object
4868 * \param[out] lds default striping
4870 * \retval 0 on success
4871 * \retval negative if failed
4873 static int lod_get_default_striping(const struct lu_env *env,
4874 struct lod_object *lo,
4875 struct lod_default_striping *lds)
4879 rc = lod_get_default_lov_striping(env, lo, lds, NULL);
4880 rc1 = lod_get_default_lmv_striping(env, lo, lds);
4881 if (rc == 0 && rc1 < 0)
4888 * Apply default striping on object.
4890 * If object striping pattern is not set, set to the one in default striping.
4891 * The default striping is from parent or fs.
4893 * \param[in] lo new object
4894 * \param[in] lds default striping
4895 * \param[in] mode new object's mode
4897 static void lod_striping_from_default(struct lod_object *lo,
4898 const struct lod_default_striping *lds,
4901 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
4904 if (lds->lds_def_striping_set && S_ISREG(mode)) {
4905 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
4907 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
4908 lds->lds_def_comp_cnt);
4912 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
4913 if (lds->lds_def_mirror_cnt > 1)
4914 lo->ldo_flr_state = LCM_FL_RDONLY;
4916 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4917 struct lod_layout_component *obj_comp =
4918 &lo->ldo_comp_entries[i];
4919 struct lod_layout_component *def_comp =
4920 &lds->lds_def_comp_entries[i];
4922 CDEBUG(D_LAYOUT, "Inherit from default: flags=%#x "
4923 "size=%hu nr=%u offset=%u pattern=%#x pool=%s\n",
4924 def_comp->llc_flags,
4925 def_comp->llc_stripe_size,
4926 def_comp->llc_stripe_count,
4927 def_comp->llc_stripe_offset,
4928 def_comp->llc_pattern,
4929 def_comp->llc_pool ?: "");
4931 *obj_comp = *def_comp;
4932 if (def_comp->llc_pool != NULL) {
4933 /* pointer was copied from def_comp */
4934 obj_comp->llc_pool = NULL;
4935 lod_obj_set_pool(lo, i, def_comp->llc_pool);
4939 if (def_comp->llc_ostlist.op_array &&
4940 def_comp->llc_ostlist.op_count) {
4941 OBD_ALLOC(obj_comp->llc_ostlist.op_array,
4942 obj_comp->llc_ostlist.op_size);
4943 if (!obj_comp->llc_ostlist.op_array)
4945 memcpy(obj_comp->llc_ostlist.op_array,
4946 def_comp->llc_ostlist.op_array,
4947 obj_comp->llc_ostlist.op_size);
4948 } else if (def_comp->llc_ostlist.op_array) {
4949 obj_comp->llc_ostlist.op_array = NULL;
4953 * Don't initialize these fields for plain layout
4954 * (v1/v3) here, they are inherited in the order of
4955 * 'parent' -> 'fs default (root)' -> 'global default
4956 * values for stripe_count & stripe_size'.
4958 * see lod_ah_init().
4960 if (!lo->ldo_is_composite)
4963 lod_adjust_stripe_info(obj_comp, desc, 0);
4965 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
4966 if (lo->ldo_dir_stripe_count == 0)
4967 lo->ldo_dir_stripe_count =
4968 lds->lds_dir_def_stripe_count;
4969 if (lo->ldo_dir_stripe_offset == -1)
4970 lo->ldo_dir_stripe_offset =
4971 lds->lds_dir_def_stripe_offset;
4972 if (lo->ldo_dir_hash_type == 0)
4973 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
4975 CDEBUG(D_LAYOUT, "striping from default dir: count:%hu, "
4976 "offset:%u, hash_type:%u\n",
4977 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
4978 lo->ldo_dir_hash_type);
4982 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root,
4985 struct lod_layout_component *lod_comp;
4987 if (lo->ldo_comp_cnt == 0)
4990 if (lo->ldo_is_composite)
4993 lod_comp = &lo->ldo_comp_entries[0];
4995 if (lod_comp->llc_stripe_count <= 0 ||
4996 lod_comp->llc_stripe_size <= 0)
4999 if (from_root && (lod_comp->llc_pool == NULL ||
5000 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
5003 if (append_pool && append_pool[0])
5010 * Implementation of dt_object_operations::do_ah_init.
5012 * This method is used to make a decision on the striping configuration for the
5013 * object being created. It can be taken from the \a parent object if it exists,
5014 * or filesystem's default. The resulting configuration (number of stripes,
5015 * stripe size/offset, pool name, etc) is stored in the object itself and will
5016 * be used by the methods like ->doo_declare_create().
5018 * \see dt_object_operations::do_ah_init() in the API description for details.
5020 static void lod_ah_init(const struct lu_env *env,
5021 struct dt_allocation_hint *ah,
5022 struct dt_object *parent,
5023 struct dt_object *child,
5026 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
5027 struct lod_thread_info *info = lod_env_info(env);
5028 struct lod_default_striping *lds = lod_lds_buf_get(env);
5029 struct dt_object *nextp = NULL;
5030 struct dt_object *nextc;
5031 struct lod_object *lp = NULL;
5032 struct lod_object *lc;
5033 struct lov_desc *desc;
5034 struct lod_layout_component *lod_comp;
5040 if (ah->dah_append_stripes == -1)
5041 ah->dah_append_stripes =
5042 d->lod_ost_descs.ltd_lov_desc.ld_tgt_count;
5044 if (likely(parent)) {
5045 nextp = dt_object_child(parent);
5046 lp = lod_dt_obj(parent);
5049 nextc = dt_object_child(child);
5050 lc = lod_dt_obj(child);
5052 LASSERT(!lod_obj_is_striped(child));
5053 /* default layout template may have been set on the regular file
5054 * when this is called from mdd_create_data() */
5055 if (S_ISREG(child_mode))
5056 lod_free_comp_entries(lc);
5058 if (!dt_object_exists(nextc))
5059 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
5061 if (S_ISDIR(child_mode)) {
5062 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
5064 /* other default values are 0 */
5065 lc->ldo_dir_stripe_offset = -1;
5067 /* no default striping configuration is needed for
5070 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5071 le32_to_cpu(lum1->lum_magic) == LMV_MAGIC_FOREIGN) {
5072 lc->ldo_dir_is_foreign = true;
5073 /* keep stripe_count 0 and stripe_offset -1 */
5074 CDEBUG(D_INFO, "no default striping for foreign dir\n");
5079 * If parent object is not root directory,
5080 * then get default striping from parent object.
5082 if (likely(lp != NULL)) {
5083 lod_get_default_striping(env, lp, lds);
5085 /* inherit default striping except ROOT */
5086 if ((lds->lds_def_striping_set ||
5087 lds->lds_dir_def_striping_set) &&
5088 !fid_is_root(lod_object_fid(lp)))
5089 lc->ldo_def_striping = lds;
5092 /* It should always honour the specified stripes */
5093 /* Note: old client (< 2.7)might also do lfs mkdir, whose EA
5094 * will have old magic. In this case, we should ignore the
5095 * stripe count and try to create dir by default stripe.
5097 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5098 (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
5099 le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC)) {
5100 lc->ldo_dir_stripe_count =
5101 le32_to_cpu(lum1->lum_stripe_count);
5102 lc->ldo_dir_stripe_offset =
5103 le32_to_cpu(lum1->lum_stripe_offset);
5104 lc->ldo_dir_hash_type =
5105 le32_to_cpu(lum1->lum_hash_type);
5107 "set dirstripe: count %hu, offset %d, hash %u\n",
5108 lc->ldo_dir_stripe_count,
5109 (int)lc->ldo_dir_stripe_offset,
5110 lc->ldo_dir_hash_type);
5112 /* transfer defaults LMV to new directory */
5113 lod_striping_from_default(lc, lds, child_mode);
5115 /* set count 0 to create normal directory */
5116 if (lc->ldo_dir_stripe_count == 1)
5117 lc->ldo_dir_stripe_count = 0;
5120 /* shrink the stripe_count to the avaible MDT count */
5121 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
5122 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)) {
5123 lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
5124 if (lc->ldo_dir_stripe_count == 1)
5125 lc->ldo_dir_stripe_count = 0;
5128 if (!(lc->ldo_dir_hash_type & LMV_HASH_TYPE_MASK))
5129 lc->ldo_dir_hash_type |=
5130 d->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
5132 CDEBUG(D_INFO, "final dir stripe [%hu %d %u]\n",
5133 lc->ldo_dir_stripe_count,
5134 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
5139 /* child object regular file*/
5141 if (!lod_object_will_be_striped(S_ISREG(child_mode),
5142 lu_object_fid(&child->do_lu)))
5145 /* If object is going to be striped over OSTs, transfer default
5146 * striping information to the child, so that we can use it
5147 * during declaration and creation.
5149 * Try from the parent first.
5151 if (likely(lp != NULL)) {
5152 rc = lod_get_default_lov_striping(env, lp, lds, ah);
5154 lod_striping_from_default(lc, lds, child_mode);
5157 /* Initialize lod_device::lod_md_root object reference */
5158 if (d->lod_md_root == NULL) {
5159 struct dt_object *root;
5160 struct lod_object *lroot;
5162 lu_root_fid(&info->lti_fid);
5163 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
5164 if (!IS_ERR(root)) {
5165 lroot = lod_dt_obj(root);
5167 spin_lock(&d->lod_lock);
5168 if (d->lod_md_root != NULL)
5169 dt_object_put(env, &d->lod_md_root->ldo_obj);
5170 d->lod_md_root = lroot;
5171 spin_unlock(&d->lod_lock);
5175 /* try inherit layout from the root object (fs default) when:
5176 * - parent does not have default layout; or
5177 * - parent has plain(v1/v3) default layout, and some attributes
5178 * are not specified in the default layout;
5180 if (d->lod_md_root != NULL &&
5181 lod_need_inherit_more(lc, true, ah->dah_append_pool)) {
5182 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds,
5186 if (lc->ldo_comp_cnt == 0) {
5187 lod_striping_from_default(lc, lds, child_mode);
5188 } else if (!lds->lds_def_striping_is_composite) {
5189 struct lod_layout_component *def_comp;
5191 LASSERT(!lc->ldo_is_composite);
5192 lod_comp = &lc->ldo_comp_entries[0];
5193 def_comp = &lds->lds_def_comp_entries[0];
5195 if (lod_comp->llc_stripe_count <= 0)
5196 lod_comp->llc_stripe_count =
5197 def_comp->llc_stripe_count;
5198 if (lod_comp->llc_stripe_size <= 0)
5199 lod_comp->llc_stripe_size =
5200 def_comp->llc_stripe_size;
5201 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT &&
5202 (!lod_comp->llc_pool || !lod_comp->llc_pool[0]))
5203 lod_comp->llc_stripe_offset =
5204 def_comp->llc_stripe_offset;
5205 if (lod_comp->llc_pool == NULL)
5206 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
5211 * fs default striping may not be explicitly set, or historically set
5212 * in config log, use them.
5214 if (lod_need_inherit_more(lc, false, ah->dah_append_pool)) {
5215 if (lc->ldo_comp_cnt == 0) {
5216 rc = lod_alloc_comp_entries(lc, 0, 1);
5218 /* fail to allocate memory, will create a
5219 * non-striped file. */
5221 lc->ldo_is_composite = 0;
5222 lod_comp = &lc->ldo_comp_entries[0];
5223 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
5225 LASSERT(!lc->ldo_is_composite);
5226 lod_comp = &lc->ldo_comp_entries[0];
5227 desc = &d->lod_ost_descs.ltd_lov_desc;
5228 lod_adjust_stripe_info(lod_comp, desc, ah->dah_append_stripes);
5229 if (ah->dah_append_pool && ah->dah_append_pool[0])
5230 lod_obj_set_pool(lc, 0, ah->dah_append_pool);
5237 * Size initialization on late striping.
5239 * Propagate the size of a truncated object to a deferred striping.
5240 * This function handles a special case when truncate was done on a
5241 * non-striped object and now while the striping is being created
5242 * we can't lose that size, so we have to propagate it to the stripes
5245 * \param[in] env execution environment
5246 * \param[in] dt object
5247 * \param[in] th transaction handle
5249 * \retval 0 on success
5250 * \retval negative if failed
5252 static int lod_declare_init_size(const struct lu_env *env,
5253 struct dt_object *dt, struct thandle *th)
5255 struct dt_object *next = dt_object_child(dt);
5256 struct lod_object *lo = lod_dt_obj(dt);
5257 struct dt_object **objects = NULL;
5258 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
5259 uint64_t size, offs;
5260 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
5261 struct lu_extent size_ext;
5264 if (!lod_obj_is_striped(dt))
5267 rc = dt_attr_get(env, next, attr);
5268 LASSERT(attr->la_valid & LA_SIZE);
5272 size = attr->la_size;
5276 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
5277 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5278 struct lod_layout_component *lod_comp;
5279 struct lu_extent *extent;
5281 lod_comp = &lo->ldo_comp_entries[i];
5283 if (lod_comp->llc_stripe == NULL)
5286 extent = &lod_comp->llc_extent;
5287 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
5288 if (!lo->ldo_is_composite ||
5289 lu_extent_is_overlapped(extent, &size_ext)) {
5290 objects = lod_comp->llc_stripe;
5291 stripe_count = lod_comp->llc_stripe_count;
5292 stripe_size = lod_comp->llc_stripe_size;
5295 if (stripe_count == 0)
5298 LASSERT(objects != NULL && stripe_size != 0);
5299 do_div(size, stripe_size);
5300 stripe = do_div(size, stripe_count);
5301 LASSERT(objects[stripe] != NULL);
5303 size = size * stripe_size;
5304 offs = attr->la_size;
5305 size += do_div(offs, stripe_size);
5307 attr->la_valid = LA_SIZE;
5308 attr->la_size = size;
5310 rc = lod_sub_declare_attr_set(env, objects[stripe],
5319 * Declare creation of striped object.
5321 * The function declares creation stripes for a regular object. The function
5322 * also declares whether the stripes will be created with non-zero size if
5323 * previously size was set non-zero on the master object. If object \a dt is
5324 * not local, then only fully defined striping can be applied in \a lovea.
5325 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
5328 * \param[in] env execution environment
5329 * \param[in] dt object
5330 * \param[in] attr attributes the stripes will be created with
5331 * \param[in] lovea a buffer containing striping description
5332 * \param[in] th transaction handle
5334 * \retval 0 on success
5335 * \retval negative if failed
5337 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
5338 struct lu_attr *attr,
5339 const struct lu_buf *lovea, struct thandle *th)
5341 struct lod_thread_info *info = lod_env_info(env);
5342 struct dt_object *next = dt_object_child(dt);
5343 struct lod_object *lo = lod_dt_obj(dt);
5347 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
5348 GOTO(out, rc = -ENOMEM);
5350 if (!dt_object_remote(next)) {
5351 /* choose OST and generate appropriate objects */
5352 rc = lod_prepare_create(env, lo, attr, lovea, th);
5357 * declare storage for striping data
5359 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
5361 /* LOD can not choose OST objects for remote objects, i.e.
5362 * stripes must be ready before that. Right now, it can only
5363 * happen during migrate, i.e. migrate process needs to create
5364 * remote regular file (mdd_migrate_create), then the migrate
5365 * process will provide stripeEA. */
5366 LASSERT(lovea != NULL);
5367 info->lti_buf = *lovea;
5370 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
5371 XATTR_NAME_LOV, 0, th);
5376 * if striping is created with local object's size > 0,
5377 * we have to propagate this size to specific object
5378 * the case is possible only when local object was created previously
5380 if (dt_object_exists(next))
5381 rc = lod_declare_init_size(env, dt, th);
5384 /* failed to create striping or to set initial size, let's reset
5385 * config so that others don't get confused */
5387 lod_striping_free(env, lo);
5393 * Whether subdirectories under \a dt should be created on MDTs by space QoS
5395 * If LMV_HASH_FLAG_SPACE is set on directory default layout, its subdirectories
5396 * should be created on MDT by space QoS.
5398 * \param[in] env execution environment
5399 * \param[in] dev lu device
5400 * \param[in] dt object
5402 * \retval 1 if directory should create subdir by space usage
5404 * \retval -ev if failed
5406 static inline int dt_object_qos_mkdir(const struct lu_env *env,
5407 struct lu_device *dev,
5408 struct dt_object *dt)
5410 struct lod_thread_info *info = lod_env_info(env);
5411 struct lu_object *obj;
5412 struct lod_object *lo;
5413 struct lmv_user_md *lmu;
5416 obj = lu_object_find_slice(env, dev, lu_object_fid(&dt->do_lu), NULL);
5418 return PTR_ERR(obj);
5420 lo = lu2lod_obj(obj);
5422 rc = lod_get_default_lmv_ea(env, lo);
5423 dt_object_put(env, dt);
5427 if (rc < (int)sizeof(*lmu))
5430 lmu = info->lti_ea_store;
5431 return le32_to_cpu(lmu->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
5435 * Implementation of dt_object_operations::do_declare_create.
5437 * The method declares creation of a new object. If the object will be striped,
5438 * then helper functions are called to find FIDs for the stripes, declare
5439 * creation of the stripes and declare initialization of the striping
5440 * information to be stored in the master object.
5442 * \see dt_object_operations::do_declare_create() in the API description
5445 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
5446 struct lu_attr *attr,
5447 struct dt_allocation_hint *hint,
5448 struct dt_object_format *dof, struct thandle *th)
5450 struct dt_object *next = dt_object_child(dt);
5451 struct lod_object *lo = lod_dt_obj(dt);
5460 * first of all, we declare creation of local object
5462 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
5467 * it's lod_ah_init() that has decided the object will be striped
5469 if (dof->dof_type == DFT_REGULAR) {
5470 /* callers don't want stripes */
5471 /* XXX: all tricky interactions with ->ah_make_hint() decided
5472 * to use striping, then ->declare_create() behaving differently
5473 * should be cleaned */
5474 if (dof->u.dof_reg.striped != 0)
5475 rc = lod_declare_striped_create(env, dt, attr,
5477 } else if (dof->dof_type == DFT_DIR) {
5478 struct seq_server_site *ss;
5479 struct lu_buf buf = { NULL };
5480 struct lu_buf *lmu = NULL;
5482 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
5484 /* If the parent has default stripeEA, and client
5485 * did not find it before sending create request,
5486 * then MDT will return -EREMOTE, and client will
5487 * retrieve the default stripeEA and re-create the
5490 * Note: if dah_eadata != NULL, it means creating the
5491 * striped directory with specified stripeEA, then it
5492 * should ignore the default stripeEA */
5493 if (hint != NULL && hint->dah_eadata == NULL) {
5494 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
5495 GOTO(out, rc = -EREMOTE);
5497 if (lo->ldo_dir_stripe_offset == LMV_OFFSET_DEFAULT) {
5498 struct lod_default_striping *lds;
5500 lds = lo->ldo_def_striping;
5502 * child and parent should be on the same MDT,
5503 * but if parent has default LMV, and the start
5504 * MDT offset is -1, it's allowed. This check
5505 * is not necessary after 2.12.22 because client
5506 * follows this already, but old client may not.
5508 if (hint->dah_parent &&
5509 dt_object_remote(hint->dah_parent) && lds &&
5510 lds->lds_dir_def_stripe_offset !=
5512 GOTO(out, rc = -EREMOTE);
5513 } else if (lo->ldo_dir_stripe_offset !=
5515 struct lod_device *lod;
5516 struct lu_tgt_desc *mdt = NULL;
5517 bool found_mdt = false;
5519 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5520 lod_foreach_mdt(lod, mdt) {
5521 if (mdt->ltd_index ==
5522 lo->ldo_dir_stripe_offset) {
5528 /* If the MDT indicated by stripe_offset can be
5529 * found, then tell client to resend the create
5530 * request to the correct MDT, otherwise return
5531 * error to client */
5533 GOTO(out, rc = -EREMOTE);
5535 GOTO(out, rc = -EINVAL);
5537 } else if (hint && hint->dah_eadata) {
5539 lmu->lb_buf = (void *)hint->dah_eadata;
5540 lmu->lb_len = hint->dah_eadata_len;
5543 rc = lod_declare_dir_striping_create(env, dt, attr, lmu, dof,
5547 /* failed to create striping or to set initial size, let's reset
5548 * config so that others don't get confused */
5550 lod_striping_free(env, lo);
5555 * Generate component ID for new created component.
5557 * \param[in] lo LOD object
5558 * \param[in] comp_idx index of ldo_comp_entries
5560 * \retval component ID on success
5561 * \retval LCME_ID_INVAL on failure
5563 static __u32 lod_gen_component_id(struct lod_object *lo,
5564 int mirror_id, int comp_idx)
5566 struct lod_layout_component *lod_comp;
5567 __u32 id, start, end;
5570 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
5572 lod_obj_inc_layout_gen(lo);
5573 id = lo->ldo_layout_gen;
5574 if (likely(id <= SEQ_ID_MAX))
5575 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
5577 /* Layout generation wraps, need to check collisions. */
5578 start = id & SEQ_ID_MASK;
5581 for (id = start; id <= end; id++) {
5582 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5583 lod_comp = &lo->ldo_comp_entries[i];
5584 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
5587 /* Found the ununsed ID */
5588 if (i == lo->ldo_comp_cnt)
5589 RETURN(pflr_id(mirror_id, id));
5591 if (end == LCME_ID_MAX) {
5593 end = min(lo->ldo_layout_gen & LCME_ID_MASK,
5594 (__u32)(LCME_ID_MAX - 1));
5598 RETURN(LCME_ID_INVAL);
5602 * Creation of a striped regular object.
5604 * The function is called to create the stripe objects for a regular
5605 * striped file. This can happen at the initial object creation or
5606 * when the caller asks LOD to do so using ->do_xattr_set() method
5607 * (so called late striping). Notice all the information are already
5608 * prepared in the form of the list of objects (ldo_stripe field).
5609 * This is done during declare phase.
5611 * \param[in] env execution environment
5612 * \param[in] dt object
5613 * \param[in] attr attributes the stripes will be created with
5614 * \param[in] dof format of stripes (see OSD API description)
5615 * \param[in] th transaction handle
5617 * \retval 0 on success
5618 * \retval negative if failed
5620 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
5621 struct lu_attr *attr, struct dt_object_format *dof,
5624 struct lod_layout_component *lod_comp;
5625 struct lod_object *lo = lod_dt_obj(dt);
5630 LASSERT((lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL) ||
5631 lo->ldo_is_foreign);
5633 mirror_id = 0; /* non-flr file's mirror_id is 0 */
5634 if (lo->ldo_mirror_count > 1) {
5635 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5636 lod_comp = &lo->ldo_comp_entries[i];
5637 if (lod_comp->llc_id != LCME_ID_INVAL &&
5638 mirror_id_of(lod_comp->llc_id) > mirror_id)
5639 mirror_id = mirror_id_of(lod_comp->llc_id);
5643 /* create all underlying objects */
5644 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5645 lod_comp = &lo->ldo_comp_entries[i];
5647 if (lod_comp->llc_id == LCME_ID_INVAL) {
5648 /* only the component of FLR layout with more than 1
5649 * mirror has mirror ID in its component ID.
5651 if (lod_comp->llc_extent.e_start == 0 &&
5652 lo->ldo_mirror_count > 1)
5655 lod_comp->llc_id = lod_gen_component_id(lo,
5657 if (lod_comp->llc_id == LCME_ID_INVAL)
5658 GOTO(out, rc = -ERANGE);
5661 if (lod_comp_inited(lod_comp))
5664 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
5665 lod_comp_set_init(lod_comp);
5667 if (lov_pattern(lod_comp->llc_pattern) == LOV_PATTERN_MDT)
5668 lod_comp_set_init(lod_comp);
5670 if (lod_comp->llc_stripe == NULL)
5673 LASSERT(lod_comp->llc_stripe_count);
5674 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
5675 struct dt_object *object = lod_comp->llc_stripe[j];
5676 LASSERT(object != NULL);
5677 rc = lod_sub_create(env, object, attr, NULL, dof, th);
5681 lod_comp_set_init(lod_comp);
5684 rc = lod_fill_mirrors(lo);
5688 rc = lod_generate_and_set_lovea(env, lo, th);
5692 lo->ldo_comp_cached = 1;
5696 lod_striping_free(env, lo);
5700 static inline bool lod_obj_is_dom(struct dt_object *dt)
5702 struct lod_object *lo = lod_dt_obj(dt);
5704 if (!dt_object_exists(dt_object_child(dt)))
5707 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
5710 if (!lo->ldo_comp_cnt)
5713 return (lov_pattern(lo->ldo_comp_entries[0].llc_pattern) ==
5718 * Implementation of dt_object_operations::do_create.
5720 * If any of preceeding methods (like ->do_declare_create(),
5721 * ->do_ah_init(), etc) chose to create a striped object,
5722 * then this method will create the master and the stripes.
5724 * \see dt_object_operations::do_create() in the API description for details.
5726 static int lod_create(const struct lu_env *env, struct dt_object *dt,
5727 struct lu_attr *attr, struct dt_allocation_hint *hint,
5728 struct dt_object_format *dof, struct thandle *th)
5733 /* create local object */
5734 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
5738 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
5739 (lod_obj_is_striped(dt) || lod_obj_is_dom(dt)) &&
5740 dof->u.dof_reg.striped != 0) {
5741 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
5742 rc = lod_striped_create(env, dt, attr, dof, th);
5749 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
5750 struct dt_object *dt, struct thandle *th,
5751 int comp_idx, int stripe_idx,
5752 struct lod_obj_stripe_cb_data *data)
5754 if (data->locd_declare)
5755 return lod_sub_declare_destroy(env, dt, th);
5756 else if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
5757 stripe_idx == cfs_fail_val)
5758 return lod_sub_destroy(env, dt, th);
5764 * Implementation of dt_object_operations::do_declare_destroy.
5766 * If the object is a striped directory, then the function declares reference
5767 * removal from the master object (this is an index) to the stripes and declares
5768 * destroy of all the stripes. In all the cases, it declares an intention to
5769 * destroy the object itself.
5771 * \see dt_object_operations::do_declare_destroy() in the API description
5774 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
5777 struct dt_object *next = dt_object_child(dt);
5778 struct lod_object *lo = lod_dt_obj(dt);
5779 struct lod_thread_info *info = lod_env_info(env);
5780 struct dt_object *stripe;
5781 char *stripe_name = info->lti_key;
5787 * load striping information, notice we don't do this when object
5788 * is being initialized as we don't need this information till
5789 * few specific cases like destroy, chown
5791 rc = lod_striping_load(env, lo);
5795 /* declare destroy for all underlying objects */
5796 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5797 rc = next->do_ops->do_index_try(env, next,
5798 &dt_directory_features);
5802 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5803 stripe = lo->ldo_stripe[i];
5807 rc = lod_sub_declare_ref_del(env, next, th);
5811 snprintf(stripe_name, sizeof(info->lti_key),
5813 PFID(lu_object_fid(&stripe->do_lu)), i);
5814 rc = lod_sub_declare_delete(env, next,
5815 (const struct dt_key *)stripe_name, th);
5822 * we declare destroy for the local object
5824 rc = lod_sub_declare_destroy(env, next, th);
5828 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
5829 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
5832 if (!lod_obj_is_striped(dt))
5835 /* declare destroy all striped objects */
5836 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5837 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5838 stripe = lo->ldo_stripe[i];
5842 if (!dt_object_exists(stripe))
5845 rc = lod_sub_declare_ref_del(env, stripe, th);
5849 rc = lod_sub_declare_destroy(env, stripe, th);
5854 struct lod_obj_stripe_cb_data data = { { 0 } };
5856 data.locd_declare = true;
5857 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
5858 rc = lod_obj_for_each_stripe(env, lo, th, &data);
5865 * Implementation of dt_object_operations::do_destroy.
5867 * If the object is a striped directory, then the function removes references
5868 * from the master object (this is an index) to the stripes and destroys all
5869 * the stripes. In all the cases, the function destroys the object itself.
5871 * \see dt_object_operations::do_destroy() in the API description for details.
5873 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
5876 struct dt_object *next = dt_object_child(dt);
5877 struct lod_object *lo = lod_dt_obj(dt);
5878 struct lod_thread_info *info = lod_env_info(env);
5879 char *stripe_name = info->lti_key;
5880 struct dt_object *stripe;
5886 /* destroy sub-stripe of master object */
5887 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5888 rc = next->do_ops->do_index_try(env, next,
5889 &dt_directory_features);
5893 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5894 stripe = lo->ldo_stripe[i];
5898 rc = lod_sub_ref_del(env, next, th);
5902 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
5903 PFID(lu_object_fid(&stripe->do_lu)), i);
5905 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
5906 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
5907 PFID(lu_object_fid(&stripe->do_lu)));
5909 rc = lod_sub_delete(env, next,
5910 (const struct dt_key *)stripe_name, th);
5916 rc = lod_sub_destroy(env, next, th);
5920 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
5921 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
5924 if (!lod_obj_is_striped(dt))
5927 /* destroy all striped objects */
5928 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5929 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5930 stripe = lo->ldo_stripe[i];
5934 if (!dt_object_exists(stripe))
5937 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
5938 i == cfs_fail_val) {
5939 dt_write_lock(env, stripe, DT_TGT_CHILD);
5940 rc = lod_sub_ref_del(env, stripe, th);
5941 dt_write_unlock(env, stripe);
5945 rc = lod_sub_destroy(env, stripe, th);
5951 struct lod_obj_stripe_cb_data data = { { 0 } };
5953 data.locd_declare = false;
5954 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
5955 rc = lod_obj_for_each_stripe(env, lo, th, &data);
5962 * Implementation of dt_object_operations::do_declare_ref_add.
5964 * \see dt_object_operations::do_declare_ref_add() in the API description
5967 static int lod_declare_ref_add(const struct lu_env *env,
5968 struct dt_object *dt, struct thandle *th)
5970 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
5974 * Implementation of dt_object_operations::do_ref_add.
5976 * \see dt_object_operations::do_ref_add() in the API description for details.
5978 static int lod_ref_add(const struct lu_env *env,
5979 struct dt_object *dt, struct thandle *th)
5981 return lod_sub_ref_add(env, dt_object_child(dt), th);
5985 * Implementation of dt_object_operations::do_declare_ref_del.
5987 * \see dt_object_operations::do_declare_ref_del() in the API description
5990 static int lod_declare_ref_del(const struct lu_env *env,
5991 struct dt_object *dt, struct thandle *th)
5993 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
5997 * Implementation of dt_object_operations::do_ref_del
5999 * \see dt_object_operations::do_ref_del() in the API description for details.
6001 static int lod_ref_del(const struct lu_env *env,
6002 struct dt_object *dt, struct thandle *th)
6004 return lod_sub_ref_del(env, dt_object_child(dt), th);
6008 * Implementation of dt_object_operations::do_object_sync.
6010 * \see dt_object_operations::do_object_sync() in the API description
6013 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
6014 __u64 start, __u64 end)
6016 return dt_object_sync(env, dt_object_child(dt), start, end);
6020 * Implementation of dt_object_operations::do_object_unlock.
6022 * Used to release LDLM lock(s).
6024 * \see dt_object_operations::do_object_unlock() in the API description
6027 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
6028 struct ldlm_enqueue_info *einfo,
6029 union ldlm_policy_data *policy)
6031 struct lod_object *lo = lod_dt_obj(dt);
6032 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
6033 int slave_locks_size;
6037 if (slave_locks == NULL)
6040 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
6041 /* Note: for remote lock for single stripe dir, MDT will cancel
6042 * the lock by lockh directly */
6043 LASSERT(!dt_object_remote(dt_object_child(dt)));
6045 /* locks were unlocked in MDT layer */
6046 for (i = 0; i < slave_locks->ha_count; i++)
6047 LASSERT(!lustre_handle_is_used(&slave_locks->ha_handles[i]));
6050 * NB, ha_count may not equal to ldo_dir_stripe_count, because dir
6051 * layout may change, e.g., shrink dir layout after migration.
6053 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6054 if (lo->ldo_stripe[i])
6055 dt_invalidate(env, lo->ldo_stripe[i]);
6058 slave_locks_size = offsetof(typeof(*slave_locks),
6059 ha_handles[slave_locks->ha_count]);
6060 OBD_FREE(slave_locks, slave_locks_size);
6061 einfo->ei_cbdata = NULL;
6067 * Implementation of dt_object_operations::do_object_lock.
6069 * Used to get LDLM lock on the non-striped and striped objects.
6071 * \see dt_object_operations::do_object_lock() in the API description
6074 static int lod_object_lock(const struct lu_env *env,
6075 struct dt_object *dt,
6076 struct lustre_handle *lh,
6077 struct ldlm_enqueue_info *einfo,
6078 union ldlm_policy_data *policy)
6080 struct lod_object *lo = lod_dt_obj(dt);
6081 int slave_locks_size;
6082 struct lustre_handle_array *slave_locks = NULL;
6087 /* remote object lock */
6088 if (!einfo->ei_enq_slave) {
6089 LASSERT(dt_object_remote(dt));
6090 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
6094 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
6097 rc = lod_striping_load(env, lo);
6102 if (lo->ldo_dir_stripe_count <= 1)
6105 slave_locks_size = offsetof(typeof(*slave_locks),
6106 ha_handles[lo->ldo_dir_stripe_count]);
6107 /* Freed in lod_object_unlock */
6108 OBD_ALLOC(slave_locks, slave_locks_size);
6111 slave_locks->ha_count = lo->ldo_dir_stripe_count;
6113 /* striped directory lock */
6114 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6115 struct lustre_handle lockh;
6116 struct ldlm_res_id *res_id;
6117 struct dt_object *stripe;
6119 stripe = lo->ldo_stripe[i];
6123 res_id = &lod_env_info(env)->lti_res_id;
6124 fid_build_reg_res_name(lu_object_fid(&stripe->do_lu), res_id);
6125 einfo->ei_res_id = res_id;
6127 if (dt_object_remote(stripe)) {
6128 set_bit(i, (void *)slave_locks->ha_map);
6129 rc = dt_object_lock(env, stripe, &lockh, einfo, policy);
6131 struct ldlm_namespace *ns = einfo->ei_namespace;
6132 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
6133 ldlm_completion_callback completion = einfo->ei_cb_cp;
6134 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
6136 if (einfo->ei_mode == LCK_PW ||
6137 einfo->ei_mode == LCK_EX)
6138 dlmflags |= LDLM_FL_COS_INCOMPAT;
6140 LASSERT(ns != NULL);
6141 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS,
6142 policy, einfo->ei_mode,
6143 &dlmflags, blocking,
6145 NULL, 0, LVB_T_NONE,
6150 ldlm_lock_decref_and_cancel(
6151 &slave_locks->ha_handles[i],
6153 OBD_FREE(slave_locks, slave_locks_size);
6156 slave_locks->ha_handles[i] = lockh;
6158 einfo->ei_cbdata = slave_locks;
6164 * Implementation of dt_object_operations::do_invalidate.
6166 * \see dt_object_operations::do_invalidate() in the API description for details
6168 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
6170 return dt_invalidate(env, dt_object_child(dt));
6173 static int lod_declare_instantiate_components(const struct lu_env *env,
6174 struct lod_object *lo, struct thandle *th)
6176 struct lod_thread_info *info = lod_env_info(env);
6181 LASSERT(info->lti_count < lo->ldo_comp_cnt);
6183 for (i = 0; i < info->lti_count; i++) {
6184 rc = lod_qos_prep_create(env, lo, NULL, th,
6185 info->lti_comp_idx[i]);
6191 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6192 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
6193 &info->lti_buf, XATTR_NAME_LOV, 0, th);
6200 * Check OSTs for an existing component for further extension
6202 * Checks if OSTs are still healthy and not out of space. Gets free space
6203 * on OSTs (relative to allocation watermark rmb_low) and compares to
6204 * the proposed new_end for this component.
6206 * Decides whether or not to extend a component on its current OSTs.
6208 * \param[in] env execution environment for this thread
6209 * \param[in] lo object we're checking
6210 * \param[in] index index of this component
6211 * \param[in] extension_size extension size for this component
6212 * \param[in] extent layout extent for requested operation
6213 * \param[in] comp_extent extension component extent
6214 * \param[in] write if this is write operation
6216 * \retval true - OK to extend on current OSTs
6217 * \retval false - do not extend on current OSTs
6219 static bool lod_sel_osts_allowed(const struct lu_env *env,
6220 struct lod_object *lo,
6221 int index, __u64 extension_size,
6222 struct lu_extent *extent,
6223 struct lu_extent *comp_extent, int write)
6225 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[index];
6226 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6227 struct obd_statfs *sfs = &lod_env_info(env)->lti_osfs;
6228 __u64 available = 0;
6235 LASSERT(lod_comp->llc_stripe_count != 0);
6238 (extent->e_start == 0 && extent->e_end == OBD_OBJECT_EOF)) {
6239 /* truncate or append */
6240 size = extension_size;
6242 /* In case of write op, check the real write extent,
6243 * it may be larger than the extension_size */
6244 size = roundup(min(extent->e_end, comp_extent->e_end) -
6245 max(extent->e_start, comp_extent->e_start),
6248 /* extension_size is file level, so we must divide by stripe count to
6249 * compare it to available space on a single OST */
6250 size /= lod_comp->llc_stripe_count;
6252 lod_getref(&lod->lod_ost_descs);
6253 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
6254 int index = lod_comp->llc_ost_indices[i];
6255 struct lod_tgt_desc *ost = OST_TGT(lod, index);
6256 struct obd_statfs_info info = { 0 };
6257 int j, repeated = 0;
6261 /* Get the number of times this OST repeats in this component.
6262 * Note: inter-component repeats are not counted as this is
6263 * considered as a rare case: we try to not repeat OST in other
6264 * components if possible. */
6265 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6266 if (index != lod_comp->llc_ost_indices[j])
6269 /* already handled */
6275 if (j < lod_comp->llc_stripe_count)
6278 if (!cfs_bitmap_check(lod->lod_ost_bitmap, index)) {
6279 CDEBUG(D_LAYOUT, "ost %d no longer present\n", index);
6284 rc = dt_statfs_info(env, ost->ltd_tgt, sfs, &info);
6286 CDEBUG(D_LAYOUT, "statfs failed for ost %d, error %d\n",
6292 if (sfs->os_state & OS_STATE_ENOSPC ||
6293 sfs->os_state & OS_STATE_READONLY ||
6294 sfs->os_state & OS_STATE_DEGRADED) {
6295 CDEBUG(D_LAYOUT, "ost %d is not availble for SEL "
6296 "extension, state %u\n", index, sfs->os_state);
6302 available = sfs->os_bavail * sfs->os_bsize;
6303 /* 'available' is relative to the allocation threshold */
6304 available -= (__u64) info.os_reserved_mb_low << 20;
6306 CDEBUG(D_LAYOUT, "ost %d lowwm: %d highwm: %d, "
6307 "%llu %% blocks available, %llu %% blocks free\n",
6308 index, info.os_reserved_mb_low, info.os_reserved_mb_high,
6309 (100ull * sfs->os_bavail) / sfs->os_blocks,
6310 (100ull * sfs->os_bfree) / sfs->os_blocks);
6312 if (size * repeated > available) {
6314 CDEBUG(D_LAYOUT, "low space on ost %d, available %llu "
6315 "< extension size %llu\n", index, available,
6320 lod_putref(lod, &lod->lod_ost_descs);
6326 * Adjust extents after component removal
6328 * When we remove an extension component, we move the start of the next
6329 * component to match the start of the extension component, so no space is left
6332 * \param[in] env execution environment for this thread
6333 * \param[in] lo object
6334 * \param[in] max_comp layout component
6335 * \param[in] index index of this component
6337 * \retval 0 on success
6338 * \retval negative errno on error
6340 static void lod_sel_adjust_extents(const struct lu_env *env,
6341 struct lod_object *lo,
6342 int max_comp, int index)
6344 struct lod_layout_component *lod_comp = NULL;
6345 struct lod_layout_component *next = NULL;
6346 struct lod_layout_component *prev = NULL;
6347 __u64 new_start = 0;
6351 /* Extension space component */
6352 lod_comp = &lo->ldo_comp_entries[index];
6353 next = &lo->ldo_comp_entries[index + 1];
6354 prev = &lo->ldo_comp_entries[index - 1];
6356 LASSERT(lod_comp != NULL && prev != NULL && next != NULL);
6357 LASSERT(lod_comp->llc_flags & LCME_FL_EXTENSION);
6359 /* Previous is being removed */
6360 if (prev && prev->llc_id == LCME_ID_INVAL)
6361 new_start = prev->llc_extent.e_start;
6363 new_start = lod_comp->llc_extent.e_start;
6365 for (i = index + 1; i < max_comp; i++) {
6366 lod_comp = &lo->ldo_comp_entries[i];
6368 start = lod_comp->llc_extent.e_start;
6369 lod_comp->llc_extent.e_start = new_start;
6371 /* We only move zero length extendable components */
6372 if (!(start == lod_comp->llc_extent.e_end))
6375 LASSERT(!(lod_comp->llc_flags & LCME_FL_INIT));
6377 lod_comp->llc_extent.e_end = new_start;
6381 /* Calculate the proposed 'new end' for a component we're extending */
6382 static __u64 lod_extension_new_end(__u64 extension_size, __u64 extent_end,
6383 __u32 stripe_size, __u64 component_end,
6384 __u64 extension_end)
6388 LASSERT(extension_size != 0 && stripe_size != 0);
6390 /* Round up to extension size */
6391 if (extent_end == OBD_OBJECT_EOF) {
6392 new_end = OBD_OBJECT_EOF;
6394 /* Add at least extension_size to the previous component_end,
6395 * covering the req layout extent */
6396 new_end = max(extent_end - component_end, extension_size);
6397 new_end = roundup(new_end, extension_size);
6398 new_end += component_end;
6400 /* Component end must be min stripe size aligned */
6401 if (new_end % stripe_size) {
6402 CDEBUG(D_LAYOUT, "new component end is not aligned "
6403 "by the stripe size %u: [%llu, %llu) ext size "
6404 "%llu new end %llu, aligning\n",
6405 stripe_size, component_end, extent_end,
6406 extension_size, new_end);
6407 new_end = roundup(new_end, stripe_size);
6411 if (new_end < extent_end)
6412 new_end = OBD_OBJECT_EOF;
6415 /* Don't extend past the end of the extension component */
6416 if (new_end > extension_end)
6417 new_end = extension_end;
6422 /* As lod_sel_handler() could be re-entered for the same component several
6423 * times, this is the data for the next call. Fields could be changed to
6424 * component indexes when needed, (e.g. if there is no need to instantiate
6425 * all the previous components up to the current position) to tell the caller
6426 * where to start over from. */
6433 * Process extent updates for a particular layout component
6435 * Handle layout updates for a particular extension space component touched by
6436 * a layout update operation. Core function of self-extending PFL feature.
6438 * In general, this function processes exactly *one* stage of an extension
6439 * operation, modifying the layout accordingly, then returns to the caller.
6440 * The caller is responsible for restarting processing with the new layout,
6441 * which may repeatedly return to this function until the extension updates
6444 * This function does one of a few things to the layout:
6445 * 1. Extends the component before the current extension space component to
6446 * allow it to accomodate the requested operation (if space/policy permit that
6447 * component to continue on its current OSTs)
6449 * 2. If extension of the existing component fails, we do one of two things:
6450 * a. If there is a component after the extension space, we remove the
6451 * extension space component, move the start of the next component down
6452 * accordingly, then notify the caller to restart processing w/the new
6454 * b. If there is no following component, we try repeating the current
6455 * component, creating a new component using the current one as a
6456 * template (keeping its stripe properties but not specific striping),
6457 * and try assigning striping for this component. If there is sufficient
6458 * free space on the OSTs chosen for this component, it is instantiated
6459 * and i/o continues there.
6461 * If there is not sufficient space on the new OSTs, we remove this new
6462 * component & extend the current component.
6464 * Note further that uninited components followed by extension space can be zero
6465 * length meaning that we will try to extend them before initializing them, and
6466 * if that fails, they will be removed without initialization.
6468 * 3. If we extend to/beyond the end of an extension space component, that
6469 * component is exhausted (all of its range has been given to real components),
6470 * so we remove it and restart processing.
6472 * \param[in] env execution environment for this thread
6473 * \param[in,out] lo object to update the layout of
6474 * \param[in] extent layout extent for requested operation, update
6475 * layout to fit this operation
6476 * \param[in] th transaction handle for this operation
6477 * \param[in,out] max_comp the highest comp for the portion of the layout
6478 * we are operating on (For FLR, the chosen
6479 * replica). Updated because we may remove
6481 * \param[in] index index of the extension space component we're
6483 * \param[in] write if this is write op
6484 * \param[in,out] force if the extension is to be forced; set here
6485 to force it on the 2nd call for the same
6488 * \retval 0 on success
6489 * \retval negative errno on error
6491 static int lod_sel_handler(const struct lu_env *env,
6492 struct lod_object *lo,
6493 struct lu_extent *extent,
6494 struct thandle *th, int *max_comp,
6495 int index, int write,
6496 struct sel_data *sd)
6498 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6499 struct lod_thread_info *info = lod_env_info(env);
6500 struct lod_layout_component *lod_comp;
6501 struct lod_layout_component *prev;
6502 struct lod_layout_component *next = NULL;
6503 __u64 extension_size;
6510 /* First component cannot be extension space */
6512 CERROR("%s: "DFID" first component cannot be extension space\n",
6513 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
6517 lod_comp = &lo->ldo_comp_entries[index];
6518 prev = &lo->ldo_comp_entries[index - 1];
6519 if ((index + 1) < *max_comp)
6520 next = &lo->ldo_comp_entries[index + 1];
6522 /* extension size uses the stripe size field as KiB */
6523 extension_size = lod_comp->llc_stripe_size * SEL_UNIT_SIZE;
6525 CDEBUG(D_LAYOUT, "prev start %llu, extension start %llu, extension end"
6526 " %llu, extension size %llu\n", prev->llc_extent.e_start,
6527 lod_comp->llc_extent.e_start, lod_comp->llc_extent.e_end,
6530 /* Two extension space components cannot be adjacent & extension space
6531 * components cannot be init */
6532 if ((prev->llc_flags & LCME_FL_EXTENSION) ||
6533 !(ergo(next, !(next->llc_flags & LCME_FL_EXTENSION))) ||
6534 lod_comp_inited(lod_comp)) {
6535 CERROR("%s: "DFID" invalid extension space components\n",
6536 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
6540 if (!prev->llc_stripe) {
6541 CDEBUG(D_LAYOUT, "Previous component not inited\n");
6542 info->lti_count = 1;
6543 info->lti_comp_idx[0] = index - 1;
6544 rc = lod_declare_instantiate_components(env, lo, th);
6545 /* ENOSPC tells us we can't use this component. If there is
6546 * a next or we are repeating, we either spill over (next) or
6547 * extend the original comp (repeat). Otherwise, return the
6548 * error to the user. */
6549 if (rc == -ENOSPC && (next || sd->sd_repeat))
6555 if (sd->sd_force == 0 && rc == 0)
6556 rc = !lod_sel_osts_allowed(env, lo, index - 1,
6557 extension_size, extent,
6558 &lod_comp->llc_extent, write);
6560 repeated = !!(sd->sd_repeat);
6564 /* Extend previous component */
6566 new_end = lod_extension_new_end(extension_size, extent->e_end,
6567 prev->llc_stripe_size,
6568 prev->llc_extent.e_end,
6569 lod_comp->llc_extent.e_end);
6571 CDEBUG(D_LAYOUT, "new end %llu\n", new_end);
6572 lod_comp->llc_extent.e_start = new_end;
6573 prev->llc_extent.e_end = new_end;
6575 if (prev->llc_extent.e_end == lod_comp->llc_extent.e_end) {
6576 CDEBUG(D_LAYOUT, "Extension component exhausted\n");
6577 lod_comp->llc_id = LCME_ID_INVAL;
6581 /* rc == 1, failed to extend current component */
6584 /* Normal 'spillover' case - Remove the extension
6585 * space component & bring down the start of the next
6587 lod_comp->llc_id = LCME_ID_INVAL;
6589 if (!(prev->llc_flags & LCME_FL_INIT)) {
6590 prev->llc_id = LCME_ID_INVAL;
6593 lod_sel_adjust_extents(env, lo, *max_comp, index);
6594 } else if (lod_comp_inited(prev)) {
6595 /* If there is no next, and the previous component is
6596 * INIT'ed, try repeating the previous component. */
6597 LASSERT(repeated == 0);
6598 rc = lod_layout_repeat_comp(env, lo, index - 1);
6602 /* The previous component is a repeated component.
6603 * Record this so we don't keep trying to repeat it. */
6606 /* If the previous component is not INIT'ed, this may
6607 * be a component we have just instantiated but failed
6608 * to extend. Or even a repeated component we failed
6609 * to prepare a striping for. Do not repeat but instead
6610 * remove the repeated component & force the extention
6611 * of the original one */
6614 prev->llc_id = LCME_ID_INVAL;
6621 rc = lod_layout_del_prep_layout(env, lo, NULL);
6624 LASSERTF(-rc == change,
6625 "number deleted %d != requested %d\n", -rc,
6628 *max_comp = *max_comp + change;
6630 /* lod_del_prep_layout reallocates ldo_comp_entries, so we must
6631 * refresh these pointers before using them */
6632 lod_comp = &lo->ldo_comp_entries[index];
6633 prev = &lo->ldo_comp_entries[index - 1];
6634 CDEBUG(D_LAYOUT, "After extent updates: prev start %llu, current start "
6635 "%llu, current end %llu max_comp %d ldo_comp_cnt %d\n",
6636 prev->llc_extent.e_start, lod_comp->llc_extent.e_start,
6637 lod_comp->llc_extent.e_end, *max_comp, lo->ldo_comp_cnt);
6639 /* Layout changed successfully */
6644 * Declare layout extent updates
6646 * Handles extensions. Identifies extension components touched by current
6647 * operation and passes them to processing function.
6649 * Restarts with updated layouts from the processing function until the current
6650 * operation no longer touches an extension space component.
6652 * \param[in] env execution environment for this thread
6653 * \param[in,out] lo object to update the layout of
6654 * \param[in] extent layout extent for requested operation, update layout to
6655 * fit this operation
6656 * \param[in] th transaction handle for this operation
6657 * \param[in] pick identifies chosen mirror for FLR layouts
6658 * \param[in] write if this is write op
6660 * \retval 1 on layout changed, 0 on no change
6661 * \retval negative errno on error
6663 static int lod_declare_update_extents(const struct lu_env *env,
6664 struct lod_object *lo, struct lu_extent *extent,
6665 struct thandle *th, int pick, int write)
6667 struct lod_thread_info *info = lod_env_info(env);
6668 struct lod_layout_component *lod_comp;
6669 bool layout_changed = false;
6670 struct sel_data sd = { 0 };
6678 /* This makes us work on the components of the chosen mirror */
6679 start_index = lo->ldo_mirrors[pick].lme_start;
6680 max_comp = lo->ldo_mirrors[pick].lme_end + 1;
6681 if (lo->ldo_flr_state == LCM_FL_NONE)
6682 LASSERT(start_index == 0 && max_comp == lo->ldo_comp_cnt);
6684 CDEBUG(D_LAYOUT, "extent->e_start %llu, extent->e_end %llu\n",
6685 extent->e_start, extent->e_end);
6686 for (i = start_index; i < max_comp; i++) {
6687 lod_comp = &lo->ldo_comp_entries[i];
6689 /* We've passed all components of interest */
6690 if (lod_comp->llc_extent.e_start >= extent->e_end)
6693 if (lod_comp->llc_flags & LCME_FL_EXTENSION) {
6694 layout_changed = true;
6695 rc = lod_sel_handler(env, lo, extent, th, &max_comp,
6700 /* Nothing has changed behind the prev one */
6706 /* We may have added or removed components. If so, we must update the
6707 * start & ends of all the mirrors after the current one, and the end
6708 * of the current mirror. */
6709 change = max_comp - 1 - lo->ldo_mirrors[pick].lme_end;
6711 lo->ldo_mirrors[pick].lme_end += change;
6712 for (i = pick + 1; i < lo->ldo_mirror_count; i++) {
6713 lo->ldo_mirrors[i].lme_start += change;
6714 lo->ldo_mirrors[i].lme_end += change;
6720 /* The amount of components has changed, adjust the lti_comp_idx */
6721 rc2 = lod_layout_data_init(info, lo->ldo_comp_cnt);
6723 return rc < 0 ? rc : rc2 < 0 ? rc2 : layout_changed;
6726 /* If striping is already instantiated or INIT'ed DOM? */
6727 static bool lod_is_instantiation_needed(struct lod_layout_component *comp)
6729 return !(((lov_pattern(comp->llc_pattern) == LOV_PATTERN_MDT) &&
6730 lod_comp_inited(comp)) || comp->llc_stripe);
6734 * Declare layout update for a non-FLR layout.
6736 * \param[in] env execution environment for this thread
6737 * \param[in,out] lo object to update the layout of
6738 * \param[in] layout layout intent for requested operation, "update" is
6739 * a process of reacting to this
6740 * \param[in] buf buffer containing lov ea (see comment on usage inline)
6741 * \param[in] th transaction handle for this operation
6743 * \retval 0 on success
6744 * \retval negative errno on error
6746 static int lod_declare_update_plain(const struct lu_env *env,
6747 struct lod_object *lo, struct layout_intent *layout,
6748 const struct lu_buf *buf, struct thandle *th)
6750 struct lod_thread_info *info = lod_env_info(env);
6751 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6752 struct lod_layout_component *lod_comp;
6753 struct lov_comp_md_v1 *comp_v1 = NULL;
6754 bool layout_changed = false;
6755 bool replay = false;
6759 LASSERT(lo->ldo_flr_state == LCM_FL_NONE);
6762 * In case the client is passing lovea, which only happens during
6763 * the replay of layout intent write RPC for now, we may need to
6764 * parse the lovea and apply new layout configuration.
6766 if (buf && buf->lb_len) {
6767 struct lov_user_md_v1 *v1 = buf->lb_buf;
6769 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
6770 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
6771 LOV_MAGIC_COMP_V1)) {
6772 CERROR("%s: the replay buffer of layout extend "
6773 "(magic %#x) does not contain expected "
6774 "composite layout.\n",
6775 lod2obd(d)->obd_name, v1->lmm_magic);
6776 GOTO(out, rc = -EINVAL);
6779 rc = lod_use_defined_striping(env, lo, buf);
6782 lo->ldo_comp_cached = 1;
6784 rc = lod_get_lov_ea(env, lo);
6787 /* old on-disk EA is stored in info->lti_buf */
6788 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
6790 layout_changed = true;
6792 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
6796 /* non replay path */
6797 rc = lod_striping_load(env, lo);
6802 /* Make sure defined layout covers the requested write range. */
6803 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
6804 if (lo->ldo_comp_cnt > 1 &&
6805 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
6806 lod_comp->llc_extent.e_end < layout->li_extent.e_end) {
6807 CDEBUG_LIMIT(replay ? D_ERROR : D_LAYOUT,
6808 "%s: the defined layout [0, %#llx) does not "
6809 "covers the write range "DEXT"\n",
6810 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
6811 PEXT(&layout->li_extent));
6812 GOTO(out, rc = -EINVAL);
6815 CDEBUG(D_LAYOUT, "%s: "DFID": update components "DEXT"\n",
6816 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
6817 PEXT(&layout->li_extent));
6820 rc = lod_declare_update_extents(env, lo, &layout->li_extent,
6821 th, 0, layout->li_opc == LAYOUT_INTENT_WRITE);
6825 layout_changed = true;
6829 * Iterate ld->ldo_comp_entries, find the component whose extent under
6830 * the write range and not instantianted.
6832 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6833 lod_comp = &lo->ldo_comp_entries[i];
6835 if (lod_comp->llc_extent.e_start >= layout->li_extent.e_end)
6839 /* If striping is instantiated or INIT'ed DOM skip */
6840 if (!lod_is_instantiation_needed(lod_comp))
6844 * In replay path, lod_comp is the EA passed by
6845 * client replay buffer, comp_v1 is the pre-recovery
6846 * on-disk EA, we'd sift out those components which
6847 * were init-ed in the on-disk EA.
6849 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
6854 * this component hasn't instantiated in normal path, or during
6855 * replay it needs replay the instantiation.
6858 /* A released component is being extended */
6859 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
6860 GOTO(out, rc = -EINVAL);
6862 LASSERT(info->lti_comp_idx != NULL);
6863 info->lti_comp_idx[info->lti_count++] = i;
6864 layout_changed = true;
6867 if (!layout_changed)
6870 lod_obj_inc_layout_gen(lo);
6871 rc = lod_declare_instantiate_components(env, lo, th);
6875 lod_striping_free(env, lo);
6879 static inline int lod_comp_index(struct lod_object *lo,
6880 struct lod_layout_component *lod_comp)
6882 LASSERT(lod_comp >= lo->ldo_comp_entries &&
6883 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
6885 return lod_comp - lo->ldo_comp_entries;
6889 * Stale other mirrors by writing extent.
6891 static int lod_stale_components(const struct lu_env *env, struct lod_object *lo,
6892 int primary, struct lu_extent *extent,
6895 struct lod_layout_component *pri_comp, *lod_comp;
6896 struct lu_extent pri_extent;
6901 /* The writing extent decides which components in the primary
6902 * are affected... */
6903 CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
6906 lod_foreach_mirror_comp(pri_comp, lo, primary) {
6907 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
6910 CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
6911 lod_comp_index(lo, pri_comp),
6912 PEXT(&pri_comp->llc_extent));
6914 pri_extent.e_start = pri_comp->llc_extent.e_start;
6915 pri_extent.e_end = pri_comp->llc_extent.e_end;
6917 for (i = 0; i < lo->ldo_mirror_count; i++) {
6920 rc = lod_declare_update_extents(env, lo, &pri_extent,
6922 /* if update_extents changed the layout, it may have
6923 * reallocated the component array, so start over to
6924 * avoid using stale pointers */
6930 /* ... and then stale other components that are
6931 * overlapping with primary components */
6932 lod_foreach_mirror_comp(lod_comp, lo, i) {
6933 if (!lu_extent_is_overlapped(
6935 &lod_comp->llc_extent))
6938 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
6939 i, lod_comp_index(lo, lod_comp));
6941 lod_comp->llc_flags |= LCME_FL_STALE;
6942 lo->ldo_mirrors[i].lme_stale = 1;
6951 * check an OST's availability
6952 * \param[in] env execution environment
6953 * \param[in] lo lod object
6954 * \param[in] dt dt object
6955 * \param[in] index mirror index
6957 * \retval negative if failed
6958 * \retval 1 if \a dt is available
6959 * \retval 0 if \a dt is not available
6961 static inline int lod_check_ost_avail(const struct lu_env *env,
6962 struct lod_object *lo,
6963 struct dt_object *dt, int index)
6965 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6966 struct lod_tgt_desc *ost;
6968 int type = LU_SEQ_RANGE_OST;
6971 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &idx, &type);
6973 CERROR("%s: can't locate "DFID":rc = %d\n",
6974 lod2obd(lod)->obd_name, PFID(lu_object_fid(&dt->do_lu)),
6979 ost = OST_TGT(lod, idx);
6980 if (ost->ltd_statfs.os_state &
6981 (OS_STATE_READONLY | OS_STATE_ENOSPC | OS_STATE_ENOINO |
6982 OS_STATE_NOPRECREATE) ||
6983 ost->ltd_active == 0) {
6984 CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail, rc = %d\n",
6985 PFID(lod_object_fid(lo)), index, idx, rc);
6993 * Pick primary mirror for write
6994 * \param[in] env execution environment
6995 * \param[in] lo object
6996 * \param[in] extent write range
6998 static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo,
6999 struct lu_extent *extent)
7001 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7002 unsigned int seq = 0;
7003 struct lod_layout_component *lod_comp;
7005 int picked = -1, second_pick = -1, third_pick = -1;
7008 if (OBD_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
7009 get_random_bytes(&seq, sizeof(seq));
7010 seq %= lo->ldo_mirror_count;
7014 * Pick a mirror as the primary, and check the availability of OSTs.
7016 * This algo can be revised later after knowing the topology of
7019 lod_qos_statfs_update(env, lod, &lod->lod_ost_descs);
7020 for (i = 0; i < lo->ldo_mirror_count; i++) {
7021 bool ost_avail = true;
7022 int index = (i + seq) % lo->ldo_mirror_count;
7024 if (lo->ldo_mirrors[index].lme_stale) {
7025 CDEBUG(D_LAYOUT, DFID": mirror %d stale\n",
7026 PFID(lod_object_fid(lo)), index);
7030 /* 2nd pick is for the primary mirror containing unavail OST */
7031 if (lo->ldo_mirrors[index].lme_primary && second_pick < 0)
7032 second_pick = index;
7034 /* 3rd pick is for non-primary mirror containing unavail OST */
7035 if (second_pick < 0 && third_pick < 0)
7039 * we found a non-primary 1st pick, we'd like to find a
7040 * potential pirmary mirror.
7042 if (picked >= 0 && !lo->ldo_mirrors[index].lme_primary)
7045 /* check the availability of OSTs */
7046 lod_foreach_mirror_comp(lod_comp, lo, index) {
7047 if (!lod_comp_inited(lod_comp) || !lod_comp->llc_stripe)
7050 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
7051 struct dt_object *dt = lod_comp->llc_stripe[j];
7053 rc = lod_check_ost_avail(env, lo, dt, index);
7060 } /* for all dt object in one component */
7063 } /* for all components in a mirror */
7066 * the OSTs where allocated objects locates in the components
7067 * of the mirror are available.
7072 /* this mirror has all OSTs available */
7076 * primary with all OSTs are available, this is the perfect
7079 if (lo->ldo_mirrors[index].lme_primary)
7081 } /* for all mirrors */
7083 /* failed to pick a sound mirror, lower our expectation */
7085 picked = second_pick;
7087 picked = third_pick;
7094 static int lod_prepare_resync_mirror(const struct lu_env *env,
7095 struct lod_object *lo,
7098 struct lod_thread_info *info = lod_env_info(env);
7099 struct lod_layout_component *lod_comp;
7100 bool neg = !!(MIRROR_ID_NEG & mirror_id);
7103 mirror_id &= ~MIRROR_ID_NEG;
7105 for (i = 0; i < lo->ldo_mirror_count; i++) {
7106 if ((!neg && lo->ldo_mirrors[i].lme_id != mirror_id) ||
7107 (neg && lo->ldo_mirrors[i].lme_id == mirror_id))
7110 lod_foreach_mirror_comp(lod_comp, lo, i) {
7111 if (lod_comp_inited(lod_comp))
7114 info->lti_comp_idx[info->lti_count++] =
7115 lod_comp_index(lo, lod_comp);
7123 * figure out the components should be instantiated for resync.
7125 static int lod_prepare_resync(const struct lu_env *env, struct lod_object *lo,
7126 struct lu_extent *extent)
7128 struct lod_thread_info *info = lod_env_info(env);
7129 struct lod_layout_component *lod_comp;
7130 unsigned int need_sync = 0;
7134 DFID": instantiate all stale components in "DEXT"\n",
7135 PFID(lod_object_fid(lo)), PEXT(extent));
7138 * instantiate all components within this extent, even non-stale
7141 for (i = 0; i < lo->ldo_mirror_count; i++) {
7142 if (!lo->ldo_mirrors[i].lme_stale)
7145 lod_foreach_mirror_comp(lod_comp, lo, i) {
7146 if (!lu_extent_is_overlapped(extent,
7147 &lod_comp->llc_extent))
7152 if (lod_comp_inited(lod_comp))
7155 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
7156 i, lod_comp_index(lo, lod_comp));
7157 info->lti_comp_idx[info->lti_count++] =
7158 lod_comp_index(lo, lod_comp);
7162 return need_sync ? 0 : -EALREADY;
7165 static int lod_declare_update_rdonly(const struct lu_env *env,
7166 struct lod_object *lo, struct md_layout_change *mlc,
7169 struct lod_thread_info *info = lod_env_info(env);
7170 struct lu_attr *layout_attr = &info->lti_layout_attr;
7171 struct lod_layout_component *lod_comp;
7172 struct lu_extent extent = { 0 };
7176 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
7177 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
7178 mlc->mlc_opc == MD_LAYOUT_RESYNC);
7179 LASSERT(lo->ldo_mirror_count > 0);
7181 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7182 struct layout_intent *layout = mlc->mlc_intent;
7183 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
7186 extent = layout->li_extent;
7187 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
7188 PFID(lod_object_fid(lo)), PEXT(&extent));
7190 picked = lod_primary_pick(env, lo, &extent);
7194 CDEBUG(D_LAYOUT, DFID": picked mirror id %u as primary\n",
7195 PFID(lod_object_fid(lo)),
7196 lo->ldo_mirrors[picked].lme_id);
7198 /* Update extents of primary before staling */
7199 rc = lod_declare_update_extents(env, lo, &extent, th, picked,
7204 if (layout->li_opc == LAYOUT_INTENT_TRUNC) {
7206 * trunc transfers [0, size) in the intent extent, we'd
7207 * stale components overlapping [size, eof).
7209 extent.e_start = extent.e_end;
7210 extent.e_end = OBD_OBJECT_EOF;
7213 /* stale overlapping components from other mirrors */
7214 rc = lod_stale_components(env, lo, picked, &extent, th);
7218 /* restore truncate intent extent */
7219 if (layout->li_opc == LAYOUT_INTENT_TRUNC)
7220 extent.e_end = extent.e_start;
7222 /* instantiate components for the picked mirror, start from 0 */
7225 lod_foreach_mirror_comp(lod_comp, lo, picked) {
7226 if (!lu_extent_is_overlapped(&extent,
7227 &lod_comp->llc_extent))
7230 if (!lod_is_instantiation_needed(lod_comp))
7233 info->lti_comp_idx[info->lti_count++] =
7234 lod_comp_index(lo, lod_comp);
7237 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
7238 } else { /* MD_LAYOUT_RESYNC */
7242 * could contain multiple non-stale mirrors, so we need to
7243 * prep uninited all components assuming any non-stale mirror
7244 * could be picked as the primary mirror.
7246 if (mlc->mlc_mirror_id == 0) {
7248 for (i = 0; i < lo->ldo_mirror_count; i++) {
7249 if (lo->ldo_mirrors[i].lme_stale)
7252 lod_foreach_mirror_comp(lod_comp, lo, i) {
7253 if (!lod_comp_inited(lod_comp))
7257 lod_comp->llc_extent.e_end)
7259 lod_comp->llc_extent.e_end;
7262 rc = lod_prepare_resync(env, lo, &extent);
7266 /* mirror write, try to init its all components */
7267 rc = lod_prepare_resync_mirror(env, lo,
7268 mlc->mlc_mirror_id);
7273 /* change the file state to SYNC_PENDING */
7274 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
7277 /* Reset the layout version once it's becoming too large.
7278 * This way it can make sure that the layout version is
7279 * monotonously increased in this writing era. */
7280 lod_obj_inc_layout_gen(lo);
7281 if (lo->ldo_layout_gen > (LCME_ID_MAX >> 1)) {
7282 __u32 layout_version;
7284 get_random_bytes(&layout_version, sizeof(layout_version));
7285 lo->ldo_layout_gen = layout_version & 0xffff;
7288 rc = lod_declare_instantiate_components(env, lo, th);
7292 layout_attr->la_valid = LA_LAYOUT_VERSION;
7293 layout_attr->la_layout_version = 0; /* set current version */
7294 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
7295 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
7296 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
7302 lod_striping_free(env, lo);
7306 static int lod_declare_update_write_pending(const struct lu_env *env,
7307 struct lod_object *lo, struct md_layout_change *mlc,
7310 struct lod_thread_info *info = lod_env_info(env);
7311 struct lu_attr *layout_attr = &info->lti_layout_attr;
7312 struct lod_layout_component *lod_comp;
7313 struct lu_extent extent = { 0 };
7319 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
7320 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
7321 mlc->mlc_opc == MD_LAYOUT_RESYNC);
7323 /* look for the primary mirror */
7324 for (i = 0; i < lo->ldo_mirror_count; i++) {
7325 if (lo->ldo_mirrors[i].lme_stale)
7328 LASSERTF(primary < 0, DFID " has multiple primary: %u / %u",
7329 PFID(lod_object_fid(lo)),
7330 lo->ldo_mirrors[i].lme_id,
7331 lo->ldo_mirrors[primary].lme_id);
7336 CERROR(DFID ": doesn't have a primary mirror\n",
7337 PFID(lod_object_fid(lo)));
7338 GOTO(out, rc = -ENODATA);
7341 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
7342 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
7344 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
7346 /* for LAYOUT_WRITE opc, it has to do the following operations:
7347 * 1. stale overlapping componets from stale mirrors;
7348 * 2. instantiate components of the primary mirror;
7349 * 3. transfter layout version to all objects of the primary;
7351 * for LAYOUT_RESYNC opc, it will do:
7352 * 1. instantiate components of all stale mirrors;
7353 * 2. transfer layout version to all objects to close write era. */
7355 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7356 struct layout_intent *layout = mlc->mlc_intent;
7357 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
7359 LASSERT(mlc->mlc_intent != NULL);
7361 extent = mlc->mlc_intent->li_extent;
7363 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
7364 PFID(lod_object_fid(lo)), PEXT(&extent));
7366 /* 1. Update extents of primary before staling */
7367 rc = lod_declare_update_extents(env, lo, &extent, th, primary,
7372 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC) {
7374 * trunc transfers [0, size) in the intent extent, we'd
7375 * stale components overlapping [size, eof).
7377 extent.e_start = extent.e_end;
7378 extent.e_end = OBD_OBJECT_EOF;
7381 /* 2. stale overlapping components */
7382 rc = lod_stale_components(env, lo, primary, &extent, th);
7386 /* 3. find the components which need instantiating.
7387 * instantiate [0, mlc->mlc_intent->e_end) */
7389 /* restore truncate intent extent */
7390 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC)
7391 extent.e_end = extent.e_start;
7394 lod_foreach_mirror_comp(lod_comp, lo, primary) {
7395 if (!lu_extent_is_overlapped(&extent,
7396 &lod_comp->llc_extent))
7399 if (!lod_is_instantiation_needed(lod_comp))
7402 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
7403 primary, lod_comp_index(lo, lod_comp));
7404 info->lti_comp_idx[info->lti_count++] =
7405 lod_comp_index(lo, lod_comp);
7407 } else { /* MD_LAYOUT_RESYNC */
7408 if (mlc->mlc_mirror_id == 0) {
7410 lod_foreach_mirror_comp(lod_comp, lo, primary) {
7411 if (!lod_comp_inited(lod_comp))
7414 extent.e_end = lod_comp->llc_extent.e_end;
7417 rc = lod_prepare_resync(env, lo, &extent);
7421 /* mirror write, try to init its all components */
7422 rc = lod_prepare_resync_mirror(env, lo,
7423 mlc->mlc_mirror_id);
7428 /* change the file state to SYNC_PENDING */
7429 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
7432 rc = lod_declare_instantiate_components(env, lo, th);
7436 /* 3. transfer layout version to OST objects.
7437 * transfer new layout version to OST objects so that stale writes
7438 * can be denied. It also ends an era of writing by setting
7439 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
7440 * send write RPC; only resync RPCs could do it. */
7441 layout_attr->la_valid = LA_LAYOUT_VERSION;
7442 layout_attr->la_layout_version = 0; /* set current version */
7443 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
7444 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
7445 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
7449 lod_obj_inc_layout_gen(lo);
7452 lod_striping_free(env, lo);
7456 static int lod_declare_update_sync_pending(const struct lu_env *env,
7457 struct lod_object *lo, struct md_layout_change *mlc,
7460 struct lod_thread_info *info = lod_env_info(env);
7461 unsigned sync_components = 0;
7462 unsigned resync_components = 0;
7467 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
7468 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
7469 mlc->mlc_opc == MD_LAYOUT_WRITE);
7471 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
7472 PFID(lod_object_fid(lo)), mlc->mlc_opc);
7474 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7475 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
7476 PFID(lod_object_fid(lo)));
7478 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
7479 return lod_declare_update_write_pending(env, lo, mlc, th);
7482 /* MD_LAYOUT_RESYNC_DONE */
7484 for (i = 0; i < lo->ldo_comp_cnt; i++) {
7485 struct lod_layout_component *lod_comp;
7488 lod_comp = &lo->ldo_comp_entries[i];
7490 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
7495 for (j = 0; j < mlc->mlc_resync_count; j++) {
7496 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
7499 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
7500 lod_comp->llc_flags &= ~LCME_FL_STALE;
7501 resync_components++;
7507 for (i = 0; i < mlc->mlc_resync_count; i++) {
7508 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
7511 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
7512 "or already synced\n", PFID(lod_object_fid(lo)),
7513 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
7514 GOTO(out, rc = -EINVAL);
7517 if (!sync_components || (mlc->mlc_resync_count && !resync_components)) {
7518 CDEBUG(D_LAYOUT, DFID": no mirror in sync\n",
7519 PFID(lod_object_fid(lo)));
7521 /* tend to return an error code here to prevent
7522 * the MDT from setting SoM attribute */
7523 GOTO(out, rc = -EINVAL);
7526 CDEBUG(D_LAYOUT, DFID": synced %u resynced %u/%zu components\n",
7527 PFID(lod_object_fid(lo)),
7528 sync_components, resync_components, mlc->mlc_resync_count);
7530 lo->ldo_flr_state = LCM_FL_RDONLY;
7531 lod_obj_inc_layout_gen(lo);
7533 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
7534 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
7535 &info->lti_buf, XATTR_NAME_LOV, 0, th);
7540 lod_striping_free(env, lo);
7544 typedef int (*mlc_handler)(const struct lu_env *env, struct dt_object *dt,
7545 const struct md_layout_change *mlc,
7546 struct thandle *th);
7549 * Attach stripes after target's for migrating directory. NB, we
7550 * only need to declare this, the actual work is done inside
7551 * lod_xattr_set_lmv().
7553 * \param[in] env execution environment
7554 * \param[in] dt target object
7555 * \param[in] mlc layout change data
7556 * \param[in] th transaction handle
7558 * \retval 0 on success
7559 * \retval negative if failed
7561 static int lod_dir_declare_layout_attach(const struct lu_env *env,
7562 struct dt_object *dt,
7563 const struct md_layout_change *mlc,
7566 struct lod_thread_info *info = lod_env_info(env);
7567 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
7568 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
7569 struct lod_object *lo = lod_dt_obj(dt);
7570 struct dt_object *next = dt_object_child(dt);
7571 struct dt_object_format *dof = &info->lti_format;
7572 struct lmv_mds_md_v1 *lmv = mlc->mlc_buf.lb_buf;
7573 struct dt_object **stripes;
7574 __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
7575 struct lu_fid *fid = &info->lti_fid;
7576 struct lod_tgt_desc *tgt;
7577 struct dt_object *dto;
7578 struct dt_device *tgt_dt;
7579 int type = LU_SEQ_RANGE_ANY;
7580 struct dt_insert_rec *rec = &info->lti_dt_rec;
7581 char *stripe_name = info->lti_key;
7582 struct lu_name *sname;
7583 struct linkea_data ldata = { NULL };
7584 struct lu_buf linkea_buf;
7591 if (!lmv_is_sane(lmv))
7594 if (!dt_try_as_dir(env, dt))
7597 dof->dof_type = DFT_DIR;
7600 sizeof(*stripes) * (lo->ldo_dir_stripe_count + stripe_count));
7604 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
7605 stripes[i] = lo->ldo_stripe[i];
7607 rec->rec_type = S_IFDIR;
7609 for (i = 0; i < stripe_count; i++) {
7611 &lmv->lmv_stripe_fids[i]);
7612 if (!fid_is_sane(fid))
7615 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
7619 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
7620 tgt_dt = lod->lod_child;
7622 tgt = LTD_TGT(ltd, idx);
7624 GOTO(out, rc = -ESTALE);
7625 tgt_dt = tgt->ltd_tgt;
7628 dto = dt_locate_at(env, tgt_dt, fid,
7629 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
7632 GOTO(out, rc = PTR_ERR(dto));
7634 stripes[i + lo->ldo_dir_stripe_count] = dto;
7636 if (!dt_try_as_dir(env, dto))
7637 GOTO(out, rc = -ENOTDIR);
7639 rc = lod_sub_declare_ref_add(env, dto, th);
7643 rec->rec_fid = lu_object_fid(&dto->do_lu);
7644 rc = lod_sub_declare_insert(env, dto,
7645 (const struct dt_rec *)rec,
7646 (const struct dt_key *)dot, th);
7650 rc = lod_sub_declare_insert(env, dto,
7651 (const struct dt_rec *)rec,
7652 (const struct dt_key *)dotdot, th);
7656 rc = lod_sub_declare_xattr_set(env, dto, &mlc->mlc_buf,
7657 XATTR_NAME_LMV, 0, th);
7661 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
7662 PFID(lu_object_fid(&dto->do_lu)),
7663 i + lo->ldo_dir_stripe_count);
7665 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
7666 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
7667 sname, lu_object_fid(&dt->do_lu));
7671 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
7672 linkea_buf.lb_len = ldata.ld_leh->leh_len;
7673 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
7674 XATTR_NAME_LINK, 0, th);
7678 rc = lod_sub_declare_insert(env, next,
7679 (const struct dt_rec *)rec,
7680 (const struct dt_key *)stripe_name,
7685 rc = lod_sub_declare_ref_add(env, next, th);
7691 OBD_FREE(lo->ldo_stripe,
7692 sizeof(*stripes) * lo->ldo_dir_stripes_allocated);
7693 lo->ldo_stripe = stripes;
7694 lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count;
7695 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type);
7696 lo->ldo_dir_stripe_count += stripe_count;
7697 lo->ldo_dir_stripes_allocated += stripe_count;
7698 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_MIGRATION;
7702 i = lo->ldo_dir_stripe_count;
7703 while (i < lo->ldo_dir_stripe_count + stripe_count && stripes[i])
7704 dt_object_put(env, stripes[i++]);
7707 sizeof(*stripes) * (stripe_count + lo->ldo_dir_stripe_count));
7711 static int lod_dir_declare_layout_detach(const struct lu_env *env,
7712 struct dt_object *dt,
7713 const struct md_layout_change *unused,
7716 struct lod_thread_info *info = lod_env_info(env);
7717 struct lod_object *lo = lod_dt_obj(dt);
7718 struct dt_object *next = dt_object_child(dt);
7719 char *stripe_name = info->lti_key;
7720 struct dt_object *dto;
7724 if (!dt_try_as_dir(env, dt))
7727 if (!lo->ldo_dir_stripe_count)
7728 return lod_sub_declare_delete(env, next,
7729 (const struct dt_key *)dotdot, th);
7731 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
7732 dto = lo->ldo_stripe[i];
7736 if (!dt_try_as_dir(env, dto))
7739 rc = lod_sub_declare_delete(env, dto,
7740 (const struct dt_key *)dotdot, th);
7744 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
7745 PFID(lu_object_fid(&dto->do_lu)), i);
7747 rc = lod_sub_declare_delete(env, next,
7748 (const struct dt_key *)stripe_name, th);
7752 rc = lod_sub_declare_ref_del(env, next, th);
7760 static int dt_dir_is_empty(const struct lu_env *env,
7761 struct dt_object *obj)
7764 const struct dt_it_ops *iops;
7769 if (!dt_try_as_dir(env, obj))
7772 iops = &obj->do_index_ops->dio_it;
7773 it = iops->init(env, obj, LUDA_64BITHASH);
7775 RETURN(PTR_ERR(it));
7777 rc = iops->get(env, it, (const struct dt_key *)"");
7781 for (rc = 0, i = 0; rc == 0 && i < 3; ++i)
7782 rc = iops->next(env, it);
7788 /* Huh? Index contains no zero key? */
7793 iops->fini(env, it);
7798 static int lod_dir_declare_layout_shrink(const struct lu_env *env,
7799 struct dt_object *dt,
7800 const struct md_layout_change *mlc,
7803 struct lod_thread_info *info = lod_env_info(env);
7804 struct lod_object *lo = lod_dt_obj(dt);
7805 struct dt_object *next = dt_object_child(dt);
7806 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
7807 __u32 final_stripe_count;
7808 char *stripe_name = info->lti_key;
7809 struct lu_buf *lmv_buf = &info->lti_buf;
7810 struct dt_object *dto;
7816 if (!dt_try_as_dir(env, dt))
7819 /* shouldn't be called on plain directory */
7820 LASSERT(lo->ldo_dir_stripe_count);
7822 lmv_buf->lb_buf = &info->lti_lmv.lmv_md_v1;
7823 lmv_buf->lb_len = sizeof(info->lti_lmv.lmv_md_v1);
7825 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
7826 LASSERT(final_stripe_count &&
7827 final_stripe_count < lo->ldo_dir_stripe_count);
7829 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
7830 dto = lo->ldo_stripe[i];
7834 if (i < final_stripe_count) {
7835 if (final_stripe_count == 1)
7838 rc = lod_sub_declare_xattr_set(env, dto, lmv_buf,
7840 LU_XATTR_REPLACE, th);
7847 rc = dt_dir_is_empty(env, dto);
7851 rc = lod_sub_declare_ref_del(env, dto, th);
7855 rc = lod_sub_declare_destroy(env, dto, th);
7859 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
7860 PFID(lu_object_fid(&dto->do_lu)), i);
7862 rc = lod_sub_declare_delete(env, next,
7863 (const struct dt_key *)stripe_name, th);
7867 rc = lod_sub_declare_ref_del(env, next, th);
7872 rc = lod_sub_declare_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
7873 LU_XATTR_REPLACE, th);
7878 * detach all stripes from dir master object, NB, stripes are not destroyed, but
7879 * deleted from it's parent namespace, this function is called in two places:
7880 * 1. mdd_migrate_mdt() detach stripes from source, and attach them to
7882 * 2. mdd_dir_layout_update() detach stripe before turning 1-stripe directory to
7883 * a plain directory.
7885 * \param[in] env execution environment
7886 * \param[in] dt target object
7887 * \param[in] mlc layout change data
7888 * \param[in] th transaction handle
7890 * \retval 0 on success
7891 * \retval negative if failed
7893 static int lod_dir_layout_detach(const struct lu_env *env,
7894 struct dt_object *dt,
7895 const struct md_layout_change *mlc,
7898 struct lod_thread_info *info = lod_env_info(env);
7899 struct lod_object *lo = lod_dt_obj(dt);
7900 struct dt_object *next = dt_object_child(dt);
7901 char *stripe_name = info->lti_key;
7902 struct dt_object *dto;
7908 if (!lo->ldo_dir_stripe_count) {
7909 /* plain directory delete .. */
7910 rc = lod_sub_delete(env, next,
7911 (const struct dt_key *)dotdot, th);
7915 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
7916 dto = lo->ldo_stripe[i];
7920 rc = lod_sub_delete(env, dto,
7921 (const struct dt_key *)dotdot, th);
7925 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
7926 PFID(lu_object_fid(&dto->do_lu)), i);
7928 rc = lod_sub_delete(env, next,
7929 (const struct dt_key *)stripe_name, th);
7933 rc = lod_sub_ref_del(env, next, th);
7938 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
7939 dto = lo->ldo_stripe[i];
7941 dt_object_put(env, dto);
7943 OBD_FREE(lo->ldo_stripe,
7944 sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated);
7945 lo->ldo_stripe = NULL;
7946 lo->ldo_dir_stripes_allocated = 0;
7947 lo->ldo_dir_stripe_count = 0;
7952 static int lod_dir_layout_shrink(const struct lu_env *env,
7953 struct dt_object *dt,
7954 const struct md_layout_change *mlc,
7957 struct lod_thread_info *info = lod_env_info(env);
7958 struct lod_object *lo = lod_dt_obj(dt);
7959 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7960 struct dt_object *next = dt_object_child(dt);
7961 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
7962 __u32 final_stripe_count;
7963 char *stripe_name = info->lti_key;
7964 struct dt_object *dto;
7965 struct lu_buf *lmv_buf = &info->lti_buf;
7966 struct lmv_mds_md_v1 *lmv = &info->lti_lmv.lmv_md_v1;
7968 int type = LU_SEQ_RANGE_ANY;
7974 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
7976 lmv_buf->lb_buf = lmv;
7977 lmv_buf->lb_len = sizeof(*lmv);
7978 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
7979 lmv->lmv_stripe_count = cpu_to_le32(final_stripe_count);
7980 lmv->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type) &
7981 cpu_to_le32(LMV_HASH_TYPE_MASK);
7982 lmv->lmv_layout_version =
7983 cpu_to_le32(lo->ldo_dir_layout_version + 1);
7985 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
7986 dto = lo->ldo_stripe[i];
7990 if (i < final_stripe_count) {
7991 /* if only one stripe left, no need to update
7992 * LMV because this stripe will replace master
7993 * object and act as a plain directory.
7995 if (final_stripe_count == 1)
7999 rc = lod_fld_lookup(env, lod,
8000 lu_object_fid(&dto->do_lu),
8005 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
8006 rc = lod_sub_xattr_set(env, dto, lmv_buf,
8008 LU_XATTR_REPLACE, th);
8015 dt_write_lock(env, dto, DT_TGT_CHILD);
8016 rc = lod_sub_ref_del(env, dto, th);
8017 dt_write_unlock(env, dto);
8021 rc = lod_sub_destroy(env, dto, th);
8025 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8026 PFID(lu_object_fid(&dto->do_lu)), i);
8028 rc = lod_sub_delete(env, next,
8029 (const struct dt_key *)stripe_name, th);
8033 rc = lod_sub_ref_del(env, next, th);
8038 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &mdtidx,
8043 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
8044 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
8045 rc = lod_sub_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
8046 LU_XATTR_REPLACE, th);
8050 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
8051 dto = lo->ldo_stripe[i];
8053 dt_object_put(env, dto);
8055 lo->ldo_dir_stripe_count = final_stripe_count;
8060 static mlc_handler dir_mlc_declare_ops[MD_LAYOUT_MAX] = {
8061 [MD_LAYOUT_ATTACH] = lod_dir_declare_layout_attach,
8062 [MD_LAYOUT_DETACH] = lod_dir_declare_layout_detach,
8063 [MD_LAYOUT_SHRINK] = lod_dir_declare_layout_shrink,
8066 static mlc_handler dir_mlc_ops[MD_LAYOUT_MAX] = {
8067 [MD_LAYOUT_DETACH] = lod_dir_layout_detach,
8068 [MD_LAYOUT_SHRINK] = lod_dir_layout_shrink,
8071 static int lod_declare_layout_change(const struct lu_env *env,
8072 struct dt_object *dt, struct md_layout_change *mlc,
8075 struct lod_thread_info *info = lod_env_info(env);
8076 struct lod_object *lo = lod_dt_obj(dt);
8081 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
8082 LASSERT(dir_mlc_declare_ops[mlc->mlc_opc]);
8083 rc = dir_mlc_declare_ops[mlc->mlc_opc](env, dt, mlc, th);
8087 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
8088 dt_object_remote(dt_object_child(dt)))
8091 rc = lod_striping_load(env, lo);
8095 LASSERT(lo->ldo_comp_cnt > 0);
8097 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
8101 switch (lo->ldo_flr_state) {
8103 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
8107 rc = lod_declare_update_rdonly(env, lo, mlc, th);
8109 case LCM_FL_WRITE_PENDING:
8110 rc = lod_declare_update_write_pending(env, lo, mlc, th);
8112 case LCM_FL_SYNC_PENDING:
8113 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
8124 * Instantiate layout component objects which covers the intent write offset.
8126 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
8127 struct md_layout_change *mlc, struct thandle *th)
8129 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
8130 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
8131 struct lod_object *lo = lod_dt_obj(dt);
8136 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
8137 LASSERT(dir_mlc_ops[mlc->mlc_opc]);
8138 rc = dir_mlc_ops[mlc->mlc_opc](env, dt, mlc, th);
8142 rc = lod_striped_create(env, dt, attr, NULL, th);
8143 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
8144 layout_attr->la_layout_version |= lo->ldo_layout_gen;
8145 rc = lod_attr_set(env, dt, layout_attr, th);
8151 struct dt_object_operations lod_obj_ops = {
8152 .do_read_lock = lod_read_lock,
8153 .do_write_lock = lod_write_lock,
8154 .do_read_unlock = lod_read_unlock,
8155 .do_write_unlock = lod_write_unlock,
8156 .do_write_locked = lod_write_locked,
8157 .do_attr_get = lod_attr_get,
8158 .do_declare_attr_set = lod_declare_attr_set,
8159 .do_attr_set = lod_attr_set,
8160 .do_xattr_get = lod_xattr_get,
8161 .do_declare_xattr_set = lod_declare_xattr_set,
8162 .do_xattr_set = lod_xattr_set,
8163 .do_declare_xattr_del = lod_declare_xattr_del,
8164 .do_xattr_del = lod_xattr_del,
8165 .do_xattr_list = lod_xattr_list,
8166 .do_ah_init = lod_ah_init,
8167 .do_declare_create = lod_declare_create,
8168 .do_create = lod_create,
8169 .do_declare_destroy = lod_declare_destroy,
8170 .do_destroy = lod_destroy,
8171 .do_index_try = lod_index_try,
8172 .do_declare_ref_add = lod_declare_ref_add,
8173 .do_ref_add = lod_ref_add,
8174 .do_declare_ref_del = lod_declare_ref_del,
8175 .do_ref_del = lod_ref_del,
8176 .do_object_sync = lod_object_sync,
8177 .do_object_lock = lod_object_lock,
8178 .do_object_unlock = lod_object_unlock,
8179 .do_invalidate = lod_invalidate,
8180 .do_declare_layout_change = lod_declare_layout_change,
8181 .do_layout_change = lod_layout_change,
8185 * Implementation of dt_body_operations::dbo_read.
8187 * \see dt_body_operations::dbo_read() in the API description for details.
8189 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
8190 struct lu_buf *buf, loff_t *pos)
8192 struct dt_object *next = dt_object_child(dt);
8194 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
8195 S_ISLNK(dt->do_lu.lo_header->loh_attr));
8196 return next->do_body_ops->dbo_read(env, next, buf, pos);
8200 * Implementation of dt_body_operations::dbo_declare_write.
8202 * \see dt_body_operations::dbo_declare_write() in the API description
8205 static ssize_t lod_declare_write(const struct lu_env *env,
8206 struct dt_object *dt,
8207 const struct lu_buf *buf, loff_t pos,
8210 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
8214 * Implementation of dt_body_operations::dbo_write.
8216 * \see dt_body_operations::dbo_write() in the API description for details.
8218 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
8219 const struct lu_buf *buf, loff_t *pos,
8222 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
8223 S_ISLNK(dt->do_lu.lo_header->loh_attr));
8224 return lod_sub_write(env, dt_object_child(dt), buf, pos, th);
8227 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
8228 __u64 start, __u64 end, struct thandle *th)
8230 if (dt_object_remote(dt))
8233 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
8236 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
8237 __u64 start, __u64 end, struct thandle *th)
8239 if (dt_object_remote(dt))
8242 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
8243 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
8247 * different type of files use the same body_ops because object may be created
8248 * in OUT, where there is no chance to set correct body_ops for each type, so
8249 * body_ops themselves will check file type inside, see lod_read/write/punch for
8252 const struct dt_body_operations lod_body_ops = {
8253 .dbo_read = lod_read,
8254 .dbo_declare_write = lod_declare_write,
8255 .dbo_write = lod_write,
8256 .dbo_declare_punch = lod_declare_punch,
8257 .dbo_punch = lod_punch,
8261 * Implementation of lu_object_operations::loo_object_init.
8263 * The function determines the type and the index of the target device using
8264 * sequence of the object's FID. Then passes control down to the
8265 * corresponding device:
8266 * OSD for the local objects, OSP for remote
8268 * \see lu_object_operations::loo_object_init() in the API description
8271 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
8272 const struct lu_object_conf *conf)
8274 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
8275 struct lu_device *cdev = NULL;
8276 struct lu_object *cobj;
8277 struct lod_tgt_descs *ltd = NULL;
8278 struct lod_tgt_desc *tgt;
8280 int type = LU_SEQ_RANGE_ANY;
8284 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
8288 if (type == LU_SEQ_RANGE_MDT &&
8289 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
8290 cdev = &lod->lod_child->dd_lu_dev;
8291 } else if (type == LU_SEQ_RANGE_MDT) {
8292 ltd = &lod->lod_mdt_descs;
8294 } else if (type == LU_SEQ_RANGE_OST) {
8295 ltd = &lod->lod_ost_descs;
8302 if (ltd->ltd_tgts_size > idx &&
8303 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
8304 tgt = LTD_TGT(ltd, idx);
8306 LASSERT(tgt != NULL);
8307 LASSERT(tgt->ltd_tgt != NULL);
8309 cdev = &(tgt->ltd_tgt->dd_lu_dev);
8311 lod_putref(lod, ltd);
8314 if (unlikely(cdev == NULL))
8317 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
8318 if (unlikely(cobj == NULL))
8321 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
8323 lu_object_add(lo, cobj);
8330 * Alloc cached foreign LOV
8332 * \param[in] lo object
8333 * \param[in] size size of foreign LOV
8335 * \retval 0 on success
8336 * \retval negative if failed
8338 int lod_alloc_foreign_lov(struct lod_object *lo, size_t size)
8340 OBD_ALLOC_LARGE(lo->ldo_foreign_lov, size);
8341 if (lo->ldo_foreign_lov == NULL)
8343 lo->ldo_foreign_lov_size = size;
8344 lo->ldo_is_foreign = 1;
8350 * Free cached foreign LOV
8352 * \param[in] lo object
8354 void lod_free_foreign_lov(struct lod_object *lo)
8356 if (lo->ldo_foreign_lov != NULL)
8357 OBD_FREE_LARGE(lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
8358 lo->ldo_foreign_lov = NULL;
8359 lo->ldo_foreign_lov_size = 0;
8360 lo->ldo_is_foreign = 0;
8365 * Free cached foreign LMV
8367 * \param[in] lo object
8369 void lod_free_foreign_lmv(struct lod_object *lo)
8371 if (lo->ldo_foreign_lmv != NULL)
8372 OBD_FREE_LARGE(lo->ldo_foreign_lmv, lo->ldo_foreign_lmv_size);
8373 lo->ldo_foreign_lmv = NULL;
8374 lo->ldo_foreign_lmv_size = 0;
8375 lo->ldo_dir_is_foreign = 0;
8380 * Release resources associated with striping.
8382 * If the object is striped (regular or directory), then release
8383 * the stripe objects references and free the ldo_stripe array.
8385 * \param[in] env execution environment
8386 * \param[in] lo object
8388 void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo)
8390 struct lod_layout_component *lod_comp;
8393 if (unlikely(lo->ldo_is_foreign)) {
8394 lod_free_foreign_lov(lo);
8395 lo->ldo_comp_cached = 0;
8396 } else if (unlikely(lo->ldo_dir_is_foreign)) {
8397 lod_free_foreign_lmv(lo);
8398 lo->ldo_dir_stripe_loaded = 0;
8399 } else if (lo->ldo_stripe != NULL) {
8400 LASSERT(lo->ldo_comp_entries == NULL);
8401 LASSERT(lo->ldo_dir_stripes_allocated > 0);
8403 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8404 if (lo->ldo_stripe[i])
8405 dt_object_put(env, lo->ldo_stripe[i]);
8408 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
8409 OBD_FREE(lo->ldo_stripe, j);
8410 lo->ldo_stripe = NULL;
8411 lo->ldo_dir_stripes_allocated = 0;
8412 lo->ldo_dir_stripe_loaded = 0;
8413 lo->ldo_dir_stripe_count = 0;
8414 } else if (lo->ldo_comp_entries != NULL) {
8415 for (i = 0; i < lo->ldo_comp_cnt; i++) {
8416 /* free lod_layout_component::llc_stripe array */
8417 lod_comp = &lo->ldo_comp_entries[i];
8419 if (lod_comp->llc_stripe == NULL)
8421 LASSERT(lod_comp->llc_stripes_allocated != 0);
8422 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
8423 if (lod_comp->llc_stripe[j] != NULL)
8425 &lod_comp->llc_stripe[j]->do_lu);
8427 OBD_FREE(lod_comp->llc_stripe,
8428 sizeof(struct dt_object *) *
8429 lod_comp->llc_stripes_allocated);
8430 lod_comp->llc_stripe = NULL;
8431 OBD_FREE(lod_comp->llc_ost_indices,
8433 lod_comp->llc_stripes_allocated);
8434 lod_comp->llc_ost_indices = NULL;
8435 lod_comp->llc_stripes_allocated = 0;
8437 lod_free_comp_entries(lo);
8438 lo->ldo_comp_cached = 0;
8442 void lod_striping_free(const struct lu_env *env, struct lod_object *lo)
8444 mutex_lock(&lo->ldo_layout_mutex);
8445 lod_striping_free_nolock(env, lo);
8446 mutex_unlock(&lo->ldo_layout_mutex);
8450 * Implementation of lu_object_operations::loo_object_free.
8452 * \see lu_object_operations::loo_object_free() in the API description
8455 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
8457 struct lod_object *lo = lu2lod_obj(o);
8459 /* release all underlying object pinned */
8460 lod_striping_free(env, lo);
8462 /* lo doesn't contain a lu_object_header, so we don't need call_rcu */
8463 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
8467 * Implementation of lu_object_operations::loo_object_release.
8469 * \see lu_object_operations::loo_object_release() in the API description
8472 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
8474 /* XXX: shouldn't we release everything here in case if object
8475 * creation failed before? */
8479 * Implementation of lu_object_operations::loo_object_print.
8481 * \see lu_object_operations::loo_object_print() in the API description
8484 static int lod_object_print(const struct lu_env *env, void *cookie,
8485 lu_printer_t p, const struct lu_object *l)
8487 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
8489 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
8492 struct lu_object_operations lod_lu_obj_ops = {
8493 .loo_object_init = lod_object_init,
8494 .loo_object_free = lod_object_free,
8495 .loo_object_release = lod_object_release,
8496 .loo_object_print = lod_object_print,