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_thread_info *info = lod_env_info(env);
1977 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1978 struct lu_tgt_descs *ltd = &lod->lod_mdt_descs;
1979 struct lu_tgt_desc *tgt = NULL;
1980 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
1981 struct dt_device *tgt_dt = NULL;
1982 struct lu_fid fid = { 0 };
1983 struct dt_object *dto;
1985 u32 stripe_count = lo->ldo_dir_stripe_count;
1991 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1992 if (stripe_count > 1)
1993 /* Set the start index for the 2nd stripe allocation */
1994 mdt_indices[1] = (mdt_indices[0] + 1) %
1995 (lod->lod_remote_mdt_count + 1);
1997 for (; stripe_idx < stripe_count; stripe_idx++) {
1998 /* Try to find next avaible target */
1999 idx = mdt_indices[stripe_idx];
2000 for (j = 0; j < lod->lod_remote_mdt_count;
2001 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
2002 bool already_allocated = false;
2005 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
2006 idx, lod->lod_remote_mdt_count + 1, stripe_idx);
2008 if (likely(!is_specific &&
2009 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
2010 /* check whether the idx already exists
2011 * in current allocated array */
2012 for (k = 0; k < stripe_idx; k++) {
2013 if (mdt_indices[k] == idx) {
2014 already_allocated = true;
2019 if (already_allocated)
2023 /* Sigh, this index is not in the bitmap, let's check
2024 * next available target */
2025 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) &&
2026 idx != master_index)
2029 if (idx == master_index) {
2030 /* Allocate the FID locally */
2031 rc = obd_fid_alloc(env, lod->lod_child_exp,
2035 tgt_dt = lod->lod_child;
2039 /* check the status of the OSP */
2040 tgt = LTD_TGT(ltd, idx);
2044 tgt_dt = tgt->ltd_tgt;
2045 rc = dt_statfs(env, tgt_dt, &info->lti_osfs);
2047 /* this OSP doesn't feel well */
2050 rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
2057 /* Can not allocate more stripes */
2058 if (j == lod->lod_remote_mdt_count) {
2059 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
2060 lod2obd(lod)->obd_name, stripe_count,
2065 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
2066 idx, stripe_idx, PFID(&fid));
2067 mdt_indices[stripe_idx] = idx;
2068 /* Set the start index for next stripe allocation */
2069 if (!is_specific && stripe_idx < stripe_count - 1) {
2071 * for large dir test, put all other slaves on one
2072 * remote MDT, otherwise we may save too many local
2073 * slave locks which will exceed RS_MAX_LOCKS.
2075 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)))
2077 mdt_indices[stripe_idx + 1] = (idx + 1) %
2078 (lod->lod_remote_mdt_count + 1);
2080 /* tgt_dt and fid must be ready after search avaible OSP
2081 * in the above loop */
2082 LASSERT(tgt_dt != NULL);
2083 LASSERT(fid_is_sane(&fid));
2085 /* fail a remote stripe FID allocation */
2086 if (stripe_idx && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_FID))
2089 dto = dt_locate_at(env, tgt_dt, &fid,
2090 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
2097 stripes[stripe_idx] = dto;
2103 for (j = 1; j < stripe_idx; j++) {
2104 LASSERT(stripes[j] != NULL);
2105 dt_object_put(env, stripes[j]);
2111 static int lod_prep_md_striped_create(const struct lu_env *env,
2112 struct dt_object *dt,
2113 struct lu_attr *attr,
2114 const struct lmv_user_md_v1 *lum,
2115 struct dt_object_format *dof,
2118 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
2119 struct lod_object *lo = lod_dt_obj(dt);
2120 struct dt_object **stripes;
2121 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2122 struct lu_fid fid = { 0 };
2129 /* The lum has been verifed in lod_verify_md_striping */
2130 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
2131 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
2133 stripe_count = lo->ldo_dir_stripe_count;
2135 OBD_ALLOC(stripes, sizeof(stripes[0]) * stripe_count);
2139 /* Allocate the first stripe locally */
2140 rc = obd_fid_alloc(env, lod->lod_child_exp, &fid, NULL);
2144 stripes[0] = dt_locate_at(env, lod->lod_child, &fid,
2145 dt->do_lu.lo_dev->ld_site->ls_top_dev, &conf);
2146 if (IS_ERR(stripes[0]))
2147 GOTO(out, rc = PTR_ERR(stripes[0]));
2149 if (lo->ldo_dir_stripe_offset == LMV_OFFSET_DEFAULT) {
2150 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
2151 rc = lod_mdt_alloc_qos(env, lo, stripes);
2153 rc = lod_mdt_alloc_rr(env, lo, stripes);
2156 bool is_specific = false;
2158 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
2160 GOTO(out, rc = -ENOMEM);
2162 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
2164 for (i = 0; i < stripe_count; i++)
2166 le32_to_cpu(lum->lum_objects[i].lum_mds);
2169 /* stripe 0 is local */
2171 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
2172 rc = lod_mdt_alloc_specific(env, lo, stripes, idx_array,
2174 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
2182 lo->ldo_dir_striped = 1;
2183 lo->ldo_stripe = stripes;
2184 lo->ldo_dir_stripe_count = rc;
2185 lo->ldo_dir_stripes_allocated = stripe_count;
2187 lo->ldo_dir_stripe_loaded = 1;
2189 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2191 lod_striping_free(env, lo);
2197 if (!IS_ERR_OR_NULL(stripes[0]))
2198 dt_object_put(env, stripes[0]);
2199 for (i = 1; i < stripe_count; i++)
2200 LASSERT(!stripes[i]);
2201 OBD_FREE(stripes, sizeof(stripes[0]) * stripe_count);
2208 * Alloc cached foreign LMV
2210 * \param[in] lo object
2211 * \param[in] size size of foreign LMV
2213 * \retval 0 on success
2214 * \retval negative if failed
2216 int lod_alloc_foreign_lmv(struct lod_object *lo, size_t size)
2218 OBD_ALLOC_LARGE(lo->ldo_foreign_lmv, size);
2219 if (lo->ldo_foreign_lmv == NULL)
2221 lo->ldo_foreign_lmv_size = size;
2222 lo->ldo_dir_is_foreign = 1;
2228 * Declare create striped md object.
2230 * The function declares intention to create a striped directory. This is a
2231 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2232 * is to verify pattern \a lum_buf is good. Check that function for the details.
2234 * \param[in] env execution environment
2235 * \param[in] dt object
2236 * \param[in] attr attributes to initialize the objects with
2237 * \param[in] lum_buf a pattern specifying the number of stripes and
2239 * \param[in] dof type of objects to be created
2240 * \param[in] th transaction handle
2242 * \retval 0 on success
2243 * \retval negative if failed
2246 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2247 struct dt_object *dt,
2248 struct lu_attr *attr,
2249 const struct lu_buf *lum_buf,
2250 struct dt_object_format *dof,
2253 struct lod_object *lo = lod_dt_obj(dt);
2254 struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
2258 LASSERT(lum != NULL);
2260 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
2261 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
2262 (int)le32_to_cpu(lum->lum_stripe_offset));
2264 if (lo->ldo_dir_stripe_count == 0) {
2265 if (lo->ldo_dir_is_foreign) {
2266 rc = lod_alloc_foreign_lmv(lo, lum_buf->lb_len);
2269 memcpy(lo->ldo_foreign_lmv, lum, lum_buf->lb_len);
2270 lo->ldo_dir_stripe_loaded = 1;
2275 /* prepare dir striped objects */
2276 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2278 /* failed to create striping, let's reset
2279 * config so that others don't get confused */
2280 lod_striping_free(env, lo);
2288 * Set or replace striped directory layout, and LFSCK may set layout on a plain
2289 * directory, so don't check stripe count.
2291 * \param[in] env execution environment
2292 * \param[in] dt target object
2293 * \param[in] buf LMV buf which contains source stripe fids
2294 * \param[in] fl set or replace
2295 * \param[in] th transaction handle
2297 * \retval 0 on success
2298 * \retval negative if failed
2300 static int lod_dir_layout_set(const struct lu_env *env,
2301 struct dt_object *dt,
2302 const struct lu_buf *buf,
2306 struct dt_object *next = dt_object_child(dt);
2307 struct lod_object *lo = lod_dt_obj(dt);
2308 struct lmv_mds_md_v1 *lmv = buf->lb_buf;
2309 struct lmv_mds_md_v1 *slave_lmv;
2310 struct lu_buf slave_buf;
2316 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LMV, fl, th);
2320 OBD_ALLOC_PTR(slave_lmv);
2324 lod_prep_slave_lmv_md(slave_lmv, lmv);
2325 slave_buf.lb_buf = slave_lmv;
2326 slave_buf.lb_len = sizeof(*slave_lmv);
2328 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2329 if (!lo->ldo_stripe[i])
2332 if (!dt_object_exists(lo->ldo_stripe[i]))
2335 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], &slave_buf,
2336 XATTR_NAME_LMV, fl, th);
2341 lod_striping_free(env, lod_dt_obj(dt));
2342 OBD_FREE_PTR(slave_lmv);
2348 * Implementation of dt_object_operations::do_declare_xattr_set.
2350 * Used with regular (non-striped) objects. Basically it
2351 * initializes the striping information and applies the
2352 * change to all the stripes.
2354 * \see dt_object_operations::do_declare_xattr_set() in the API description
2357 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2358 struct dt_object *dt,
2359 const struct lu_buf *buf,
2360 const char *name, int fl,
2363 struct dt_object *next = dt_object_child(dt);
2364 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2365 struct lod_object *lo = lod_dt_obj(dt);
2370 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2371 struct lmv_user_md_v1 *lum;
2373 LASSERT(buf != NULL && buf->lb_buf != NULL);
2375 rc = lod_verify_md_striping(d, lum);
2378 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2379 rc = lod_verify_striping(d, lo, buf, false);
2384 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2388 /* Note: Do not set LinkEA on sub-stripes, otherwise
2389 * it will confuse the fid2path process(see mdt_path_current()).
2390 * The linkEA between master and sub-stripes is set in
2391 * lod_xattr_set_lmv(). */
2392 if (strcmp(name, XATTR_NAME_LINK) == 0)
2395 /* set xattr to each stripes, if needed */
2396 rc = lod_striping_load(env, lo);
2400 if (lo->ldo_dir_stripe_count == 0)
2403 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2404 if (!lo->ldo_stripe[i])
2407 if (!dt_object_exists(lo->ldo_stripe[i]))
2410 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2420 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2421 struct lod_object *lo,
2422 struct dt_object *dt, struct thandle *th,
2423 int comp_idx, int stripe_idx,
2424 struct lod_obj_stripe_cb_data *data)
2426 struct lod_thread_info *info = lod_env_info(env);
2427 struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
2428 struct filter_fid *ff = &info->lti_ff;
2429 struct lu_buf *buf = &info->lti_buf;
2433 buf->lb_len = sizeof(*ff);
2434 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2442 * locd_buf is set if it's called by dir migration, which doesn't check
2445 if (data->locd_buf) {
2446 memset(ff, 0, sizeof(*ff));
2447 ff->ff_parent = *(struct lu_fid *)data->locd_buf->lb_buf;
2449 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
2451 if (lu_fid_eq(lod_object_fid(lo), &ff->ff_parent) &&
2452 ff->ff_layout.ol_comp_id == comp->llc_id)
2455 memset(ff, 0, sizeof(*ff));
2456 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2459 /* rewrite filter_fid */
2460 ff->ff_parent.f_ver = stripe_idx;
2461 ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
2462 ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
2463 ff->ff_layout.ol_comp_id = comp->llc_id;
2464 ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
2465 ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
2466 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
2468 if (data->locd_declare)
2469 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2470 LU_XATTR_REPLACE, th);
2472 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2473 LU_XATTR_REPLACE, th);
2479 * Reset parent FID on OST object
2481 * Replace parent FID with @dt object FID, which is only called during migration
2482 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2483 * the FID is changed.
2485 * \param[in] env execution environment
2486 * \param[in] dt dt_object whose stripes's parent FID will be reset
2487 * \parem[in] th thandle
2488 * \param[in] declare if it is declare
2490 * \retval 0 if reset succeeds
2491 * \retval negative errno if reset fails
2493 static int lod_replace_parent_fid(const struct lu_env *env,
2494 struct dt_object *dt,
2495 const struct lu_buf *buf,
2496 struct thandle *th, bool declare)
2498 struct lod_object *lo = lod_dt_obj(dt);
2499 struct lod_thread_info *info = lod_env_info(env);
2500 struct filter_fid *ff;
2501 struct lod_obj_stripe_cb_data data = { { 0 } };
2505 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2507 /* set xattr to each stripes, if needed */
2508 rc = lod_striping_load(env, lo);
2512 if (!lod_obj_is_striped(dt))
2515 if (info->lti_ea_store_size < sizeof(*ff)) {
2516 rc = lod_ea_store_resize(info, sizeof(*ff));
2521 data.locd_declare = declare;
2522 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2523 data.locd_buf = buf;
2524 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2529 inline __u16 lod_comp_entry_stripe_count(struct lod_object *lo,
2530 struct lod_layout_component *entry,
2533 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2537 else if (lod_comp_inited(entry))
2538 return entry->llc_stripe_count;
2539 else if ((__u16)-1 == entry->llc_stripe_count)
2540 return lod->lod_ost_count;
2542 return lod_get_stripe_count(lod, lo,
2543 entry->llc_stripe_count, false);
2546 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2548 int magic, size = 0, i;
2549 struct lod_layout_component *comp_entries;
2551 bool is_composite, is_foreign = false;
2554 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2555 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2557 lo->ldo_def_striping->lds_def_striping_is_composite;
2559 comp_cnt = lo->ldo_comp_cnt;
2560 comp_entries = lo->ldo_comp_entries;
2561 is_composite = lo->ldo_is_composite;
2562 is_foreign = lo->ldo_is_foreign;
2566 return lo->ldo_foreign_lov_size;
2568 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2570 size = sizeof(struct lov_comp_md_v1) +
2571 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2572 LASSERT(size % sizeof(__u64) == 0);
2575 for (i = 0; i < comp_cnt; i++) {
2578 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2579 stripe_count = lod_comp_entry_stripe_count(lo, &comp_entries[i],
2581 if (!is_dir && is_composite)
2582 lod_comp_shrink_stripe_count(&comp_entries[i],
2585 size += lov_user_md_size(stripe_count, magic);
2586 LASSERT(size % sizeof(__u64) == 0);
2592 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2593 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2596 * \param[in] env execution environment
2597 * \param[in] dt dt_object to add components on
2598 * \param[in] buf buffer contains components to be added
2599 * \parem[in] th thandle
2601 * \retval 0 on success
2602 * \retval negative errno on failure
2604 static int lod_declare_layout_add(const struct lu_env *env,
2605 struct dt_object *dt,
2606 const struct lu_buf *buf,
2609 struct lod_thread_info *info = lod_env_info(env);
2610 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2611 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2612 struct dt_object *next = dt_object_child(dt);
2613 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
2614 struct lod_object *lo = lod_dt_obj(dt);
2615 struct lov_user_md_v3 *v3;
2616 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2618 int i, rc, array_cnt, old_array_cnt;
2621 LASSERT(lo->ldo_is_composite);
2623 if (lo->ldo_flr_state != LCM_FL_NONE)
2626 rc = lod_verify_striping(d, lo, buf, false);
2630 magic = comp_v1->lcm_magic;
2631 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2632 lustre_swab_lov_comp_md_v1(comp_v1);
2633 magic = comp_v1->lcm_magic;
2636 if (magic != LOV_USER_MAGIC_COMP_V1)
2639 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2640 OBD_ALLOC(comp_array, sizeof(*comp_array) * array_cnt);
2641 if (comp_array == NULL)
2644 memcpy(comp_array, lo->ldo_comp_entries,
2645 sizeof(*comp_array) * lo->ldo_comp_cnt);
2647 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2648 struct lov_user_md_v1 *v1;
2649 struct lu_extent *ext;
2651 v1 = (struct lov_user_md *)((char *)comp_v1 +
2652 comp_v1->lcm_entries[i].lcme_offset);
2653 ext = &comp_v1->lcm_entries[i].lcme_extent;
2655 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2656 lod_comp->llc_extent.e_start = ext->e_start;
2657 lod_comp->llc_extent.e_end = ext->e_end;
2658 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2659 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2661 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2662 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2663 lod_adjust_stripe_info(lod_comp, desc, 0);
2665 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2666 v3 = (struct lov_user_md_v3 *) v1;
2667 if (v3->lmm_pool_name[0] != '\0') {
2668 rc = lod_set_pool(&lod_comp->llc_pool,
2676 old_array = lo->ldo_comp_entries;
2677 old_array_cnt = lo->ldo_comp_cnt;
2679 lo->ldo_comp_entries = comp_array;
2680 lo->ldo_comp_cnt = array_cnt;
2682 /* No need to increase layout generation here, it will be increased
2683 * later when generating component ID for the new components */
2685 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2686 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2687 XATTR_NAME_LOV, 0, th);
2689 lo->ldo_comp_entries = old_array;
2690 lo->ldo_comp_cnt = old_array_cnt;
2694 OBD_FREE(old_array, sizeof(*lod_comp) * old_array_cnt);
2696 LASSERT(lo->ldo_mirror_count == 1);
2697 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2702 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2703 lod_comp = &comp_array[i];
2704 if (lod_comp->llc_pool != NULL) {
2705 OBD_FREE(lod_comp->llc_pool,
2706 strlen(lod_comp->llc_pool) + 1);
2707 lod_comp->llc_pool = NULL;
2710 OBD_FREE(comp_array, sizeof(*comp_array) * array_cnt);
2715 * lod_last_non_stale_mirror() - Check if a mirror is the last non-stale mirror.
2716 * @mirror_id: Mirror id to be checked.
2719 * This function checks if a mirror with specified @mirror_id is the last
2720 * non-stale mirror of a LOD object @lo.
2722 * Return: true or false.
2725 bool lod_last_non_stale_mirror(__u16 mirror_id, struct lod_object *lo)
2727 struct lod_layout_component *lod_comp;
2728 bool has_stale_flag;
2731 for (i = 0; i < lo->ldo_mirror_count; i++) {
2732 if (lo->ldo_mirrors[i].lme_id == mirror_id ||
2733 lo->ldo_mirrors[i].lme_stale)
2736 has_stale_flag = false;
2737 lod_foreach_mirror_comp(lod_comp, lo, i) {
2738 if (lod_comp->llc_flags & LCME_FL_STALE) {
2739 has_stale_flag = true;
2743 if (!has_stale_flag)
2751 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2752 * the '$field' can only be 'flags' now. The xattr value is binary
2753 * lov_comp_md_v1 which contains the component ID(s) and the value of
2754 * the field to be modified.
2755 * Please update allowed_lustre_lov macro if $field groks more values
2758 * \param[in] env execution environment
2759 * \param[in] dt dt_object to be modified
2760 * \param[in] op operation string, like "set.flags"
2761 * \param[in] buf buffer contains components to be set
2762 * \parem[in] th thandle
2764 * \retval 0 on success
2765 * \retval negative errno on failure
2767 static int lod_declare_layout_set(const struct lu_env *env,
2768 struct dt_object *dt,
2769 char *op, const struct lu_buf *buf,
2772 struct lod_layout_component *lod_comp;
2773 struct lod_thread_info *info = lod_env_info(env);
2774 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2775 struct lod_object *lo = lod_dt_obj(dt);
2776 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2779 bool changed = false;
2782 /* Please update allowed_lustre_lov macro if op
2783 * groks more values in the future
2785 if (strcmp(op, "set.flags") != 0) {
2786 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2787 lod2obd(d)->obd_name, op);
2791 magic = comp_v1->lcm_magic;
2792 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2793 lustre_swab_lov_comp_md_v1(comp_v1);
2794 magic = comp_v1->lcm_magic;
2797 if (magic != LOV_USER_MAGIC_COMP_V1)
2800 if (comp_v1->lcm_entry_count == 0) {
2801 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2802 lod2obd(d)->obd_name);
2806 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2807 __u32 id = comp_v1->lcm_entries[i].lcme_id;
2808 __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
2809 __u32 mirror_flag = flags & LCME_MIRROR_FLAGS;
2810 __u16 mirror_id = mirror_id_of(id);
2811 bool neg = flags & LCME_FL_NEG;
2813 if (flags & LCME_FL_INIT) {
2815 lod_striping_free(env, lo);
2819 flags &= ~(LCME_MIRROR_FLAGS | LCME_FL_NEG);
2820 for (j = 0; j < lo->ldo_comp_cnt; j++) {
2821 lod_comp = &lo->ldo_comp_entries[j];
2823 /* lfs only put one flag in each entry */
2824 if ((flags && id != lod_comp->llc_id) ||
2825 (mirror_flag && mirror_id !=
2826 mirror_id_of(lod_comp->llc_id)))
2831 lod_comp->llc_flags &= ~flags;
2833 lod_comp->llc_flags &= ~mirror_flag;
2836 if ((flags & LCME_FL_STALE) &&
2837 lod_last_non_stale_mirror(mirror_id,
2840 lod_comp->llc_flags |= flags;
2843 lod_comp->llc_flags |= mirror_flag;
2844 if (mirror_flag & LCME_FL_NOSYNC)
2845 lod_comp->llc_timestamp =
2846 ktime_get_real_seconds();
2854 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
2855 lod2obd(d)->obd_name);
2859 lod_obj_inc_layout_gen(lo);
2861 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2862 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
2863 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
2868 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
2869 * and the xattr value is a unique component ID or a special lcme_id.
2871 * \param[in] env execution environment
2872 * \param[in] dt dt_object to be operated on
2873 * \param[in] buf buffer contains component ID or lcme_id
2874 * \parem[in] th thandle
2876 * \retval 0 on success
2877 * \retval negative errno on failure
2879 static int lod_declare_layout_del(const struct lu_env *env,
2880 struct dt_object *dt,
2881 const struct lu_buf *buf,
2884 struct lod_thread_info *info = lod_env_info(env);
2885 struct dt_object *next = dt_object_child(dt);
2886 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2887 struct lod_object *lo = lod_dt_obj(dt);
2888 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2889 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2890 __u32 magic, id, flags, neg_flags = 0;
2894 LASSERT(lo->ldo_is_composite);
2896 if (lo->ldo_flr_state != LCM_FL_NONE)
2899 magic = comp_v1->lcm_magic;
2900 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2901 lustre_swab_lov_comp_md_v1(comp_v1);
2902 magic = comp_v1->lcm_magic;
2905 if (magic != LOV_USER_MAGIC_COMP_V1)
2908 id = comp_v1->lcm_entries[0].lcme_id;
2909 flags = comp_v1->lcm_entries[0].lcme_flags;
2911 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
2912 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
2913 lod2obd(d)->obd_name, id, flags);
2917 if (id != LCME_ID_INVAL && flags != 0) {
2918 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
2919 lod2obd(d)->obd_name);
2923 if (id == LCME_ID_INVAL && !flags) {
2924 CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
2925 lod2obd(d)->obd_name);
2929 if (flags & LCME_FL_NEG) {
2930 neg_flags = flags & ~LCME_FL_NEG;
2934 left = lo->ldo_comp_cnt;
2938 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
2939 struct lod_layout_component *lod_comp;
2941 lod_comp = &lo->ldo_comp_entries[i];
2943 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
2945 else if (flags && !(flags & lod_comp->llc_flags))
2947 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
2950 if (left != (i + 1)) {
2951 CDEBUG(D_LAYOUT, "%s: this deletion will create "
2952 "a hole.\n", lod2obd(d)->obd_name);
2957 /* Mark the component as deleted */
2958 lod_comp->llc_id = LCME_ID_INVAL;
2960 /* Not instantiated component */
2961 if (lod_comp->llc_stripe == NULL)
2964 LASSERT(lod_comp->llc_stripe_count > 0);
2965 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
2966 struct dt_object *obj = lod_comp->llc_stripe[j];
2970 rc = lod_sub_declare_destroy(env, obj, th);
2976 LASSERTF(left >= 0, "left = %d\n", left);
2977 if (left == lo->ldo_comp_cnt) {
2978 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
2979 lod2obd(d)->obd_name, id);
2983 memset(attr, 0, sizeof(*attr));
2984 attr->la_valid = LA_SIZE;
2985 rc = lod_sub_declare_attr_set(env, next, attr, th);
2990 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2991 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2992 XATTR_NAME_LOV, 0, th);
2994 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
3001 * Declare layout add/set/del operations issued by special xattr names:
3003 * XATTR_LUSTRE_LOV.add add component(s) to existing file
3004 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
3005 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
3007 * \param[in] env execution environment
3008 * \param[in] dt object
3009 * \param[in] name name of xattr
3010 * \param[in] buf lu_buf contains xattr value
3011 * \param[in] th transaction handle
3013 * \retval 0 on success
3014 * \retval negative if failed
3016 static int lod_declare_modify_layout(const struct lu_env *env,
3017 struct dt_object *dt,
3019 const struct lu_buf *buf,
3022 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3023 struct lod_object *lo = lod_dt_obj(dt);
3025 int rc, len = strlen(XATTR_LUSTRE_LOV);
3028 LASSERT(dt_object_exists(dt));
3030 if (strlen(name) <= len || name[len] != '.') {
3031 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
3032 lod2obd(d)->obd_name, name);
3037 rc = lod_striping_load(env, lo);
3041 /* the layout to be modified must be a composite layout */
3042 if (!lo->ldo_is_composite) {
3043 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
3044 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3045 GOTO(unlock, rc = -EINVAL);
3048 op = (char *)name + len;
3049 if (strcmp(op, "add") == 0) {
3050 rc = lod_declare_layout_add(env, dt, buf, th);
3051 } else if (strcmp(op, "del") == 0) {
3052 rc = lod_declare_layout_del(env, dt, buf, th);
3053 } else if (strncmp(op, "set", strlen("set")) == 0) {
3054 rc = lod_declare_layout_set(env, dt, op, buf, th);
3056 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
3057 lod2obd(d)->obd_name, name);
3058 GOTO(unlock, rc = -ENOTSUPP);
3062 lod_striping_free(env, lo);
3068 * Convert a plain file lov_mds_md to a composite layout.
3070 * \param[in,out] info the thread info::lti_ea_store buffer contains little
3071 * endian plain file layout
3073 * \retval 0 on success, <0 on failure
3075 static int lod_layout_convert(struct lod_thread_info *info)
3077 struct lov_mds_md *lmm = info->lti_ea_store;
3078 struct lov_mds_md *lmm_save;
3079 struct lov_comp_md_v1 *lcm;
3080 struct lov_comp_md_entry_v1 *lcme;
3086 /* realloc buffer to a composite layout which contains one component */
3087 blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3088 le32_to_cpu(lmm->lmm_magic));
3089 size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
3091 OBD_ALLOC_LARGE(lmm_save, blob_size);
3093 GOTO(out, rc = -ENOMEM);
3095 memcpy(lmm_save, lmm, blob_size);
3097 if (info->lti_ea_store_size < size) {
3098 rc = lod_ea_store_resize(info, size);
3103 lcm = info->lti_ea_store;
3104 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
3105 lcm->lcm_size = cpu_to_le32(size);
3106 lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
3107 lmm_save->lmm_layout_gen));
3108 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
3109 lcm->lcm_entry_count = cpu_to_le16(1);
3110 lcm->lcm_mirror_count = 0;
3112 lcme = &lcm->lcm_entries[0];
3113 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
3114 lcme->lcme_extent.e_start = 0;
3115 lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
3116 lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
3117 lcme->lcme_size = cpu_to_le32(blob_size);
3119 memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
3124 OBD_FREE_LARGE(lmm_save, blob_size);
3129 * Merge layouts to form a mirrored file.
3131 static int lod_declare_layout_merge(const struct lu_env *env,
3132 struct dt_object *dt, const struct lu_buf *mbuf,
3135 struct lod_thread_info *info = lod_env_info(env);
3136 struct lu_buf *buf = &info->lti_buf;
3137 struct lod_object *lo = lod_dt_obj(dt);
3138 struct lov_comp_md_v1 *lcm;
3139 struct lov_comp_md_v1 *cur_lcm;
3140 struct lov_comp_md_v1 *merge_lcm;
3141 struct lov_comp_md_entry_v1 *lcme;
3142 struct lov_mds_md_v1 *lmm;
3145 __u16 cur_entry_count;
3146 __u16 merge_entry_count;
3148 __u16 mirror_id = 0;
3155 merge_lcm = mbuf->lb_buf;
3156 if (mbuf->lb_len < sizeof(*merge_lcm))
3159 /* must be an existing layout from disk */
3160 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
3163 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
3165 /* do not allow to merge two mirrored files */
3166 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
3169 /* verify the target buffer */
3170 rc = lod_get_lov_ea(env, lo);
3172 RETURN(rc ? : -ENODATA);
3174 cur_lcm = info->lti_ea_store;
3175 switch (le32_to_cpu(cur_lcm->lcm_magic)) {
3178 rc = lod_layout_convert(info);
3180 case LOV_MAGIC_COMP_V1:
3190 /* info->lti_ea_store could be reallocated in lod_layout_convert() */
3191 cur_lcm = info->lti_ea_store;
3192 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
3194 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
3195 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
3196 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
3199 /* size of new layout */
3200 size = le32_to_cpu(cur_lcm->lcm_size) +
3201 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
3203 memset(buf, 0, sizeof(*buf));
3204 lu_buf_alloc(buf, size);
3205 if (buf->lb_buf == NULL)
3209 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
3211 offset = sizeof(*lcm) +
3212 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
3213 for (i = 0; i < cur_entry_count; i++) {
3214 struct lov_comp_md_entry_v1 *cur_lcme;
3216 lcme = &lcm->lcm_entries[i];
3217 cur_lcme = &cur_lcm->lcm_entries[i];
3219 lcme->lcme_offset = cpu_to_le32(offset);
3220 memcpy((char *)lcm + offset,
3221 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
3222 le32_to_cpu(lcme->lcme_size));
3224 offset += le32_to_cpu(lcme->lcme_size);
3226 if (mirror_count == 1 &&
3227 mirror_id_of(le32_to_cpu(lcme->lcme_id)) == 0) {
3228 /* Add mirror from a non-flr file, create new mirror ID.
3229 * Otherwise, keep existing mirror's component ID, used
3230 * for mirror extension.
3232 id = pflr_id(1, i + 1);
3233 lcme->lcme_id = cpu_to_le32(id);
3236 id = max(le32_to_cpu(lcme->lcme_id), id);
3239 mirror_id = mirror_id_of(id) + 1;
3241 /* check if first entry in new layout is DOM */
3242 lmm = (struct lov_mds_md_v1 *)((char *)merge_lcm +
3243 merge_lcm->lcm_entries[0].lcme_offset);
3244 merge_has_dom = lov_pattern(le32_to_cpu(lmm->lmm_pattern)) ==
3247 for (i = 0; i < merge_entry_count; i++) {
3248 struct lov_comp_md_entry_v1 *merge_lcme;
3250 merge_lcme = &merge_lcm->lcm_entries[i];
3251 lcme = &lcm->lcm_entries[cur_entry_count + i];
3253 *lcme = *merge_lcme;
3254 lcme->lcme_offset = cpu_to_le32(offset);
3255 if (merge_has_dom && i == 0)
3256 lcme->lcme_flags |= cpu_to_le32(LCME_FL_STALE);
3258 id = pflr_id(mirror_id, i + 1);
3259 lcme->lcme_id = cpu_to_le32(id);
3261 memcpy((char *)lcm + offset,
3262 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
3263 le32_to_cpu(lcme->lcme_size));
3265 offset += le32_to_cpu(lcme->lcme_size);
3268 /* fixup layout information */
3269 lod_obj_inc_layout_gen(lo);
3270 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3271 lcm->lcm_size = cpu_to_le32(size);
3272 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
3273 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
3274 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NONE)
3275 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
3277 rc = lod_striping_reload(env, lo, buf);
3281 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
3282 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3290 * Split layouts, just set the LOVEA with the layout from mbuf.
3292 static int lod_declare_layout_split(const struct lu_env *env,
3293 struct dt_object *dt, const struct lu_buf *mbuf,
3296 struct lod_object *lo = lod_dt_obj(dt);
3297 struct lov_comp_md_v1 *lcm = mbuf->lb_buf;
3301 lod_obj_inc_layout_gen(lo);
3302 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3304 rc = lod_striping_reload(env, lo, mbuf);
3308 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), mbuf,
3309 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3314 * Implementation of dt_object_operations::do_declare_xattr_set.
3316 * \see dt_object_operations::do_declare_xattr_set() in the API description
3319 * the extension to the API:
3320 * - declaring LOVEA requests striping creation
3321 * - LU_XATTR_REPLACE means layout swap
3323 static int lod_declare_xattr_set(const struct lu_env *env,
3324 struct dt_object *dt,
3325 const struct lu_buf *buf,
3326 const char *name, int fl,
3329 struct dt_object *next = dt_object_child(dt);
3330 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3335 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
3336 if ((S_ISREG(mode) || mode == 0) &&
3337 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE | LU_XATTR_SPLIT)) &&
3338 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3339 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
3341 * this is a request to create object's striping.
3343 * allow to declare predefined striping on a new (!mode) object
3344 * which is supposed to be replay of regular file creation
3345 * (when LOV setting is declared)
3347 * LU_XATTR_REPLACE is set to indicate a layout swap
3349 if (dt_object_exists(dt)) {
3350 rc = dt_attr_get(env, next, attr);
3354 memset(attr, 0, sizeof(*attr));
3355 attr->la_valid = LA_TYPE | LA_MODE;
3356 attr->la_mode = S_IFREG;
3358 rc = lod_declare_striped_create(env, dt, attr, buf, th);
3359 } else if (fl & LU_XATTR_MERGE) {
3360 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3361 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3362 rc = lod_declare_layout_merge(env, dt, buf, th);
3363 } else if (fl & LU_XATTR_SPLIT) {
3364 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3365 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3366 rc = lod_declare_layout_split(env, dt, buf, th);
3367 } else if (S_ISREG(mode) &&
3368 strlen(name) >= sizeof(XATTR_LUSTRE_LOV) + 3 &&
3369 allowed_lustre_lov(name)) {
3371 * this is a request to modify object's striping.
3372 * add/set/del component(s).
3374 if (!dt_object_exists(dt))
3377 rc = lod_declare_modify_layout(env, dt, name, buf, th);
3378 } else if (S_ISDIR(mode)) {
3379 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
3380 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3381 rc = lod_replace_parent_fid(env, dt, buf, th, true);
3383 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
3390 * Apply xattr changes to the object.
3392 * Applies xattr changes to the object and the stripes if the latter exist.
3394 * \param[in] env execution environment
3395 * \param[in] dt object
3396 * \param[in] buf buffer pointing to the new value of xattr
3397 * \param[in] name name of xattr
3398 * \param[in] fl flags
3399 * \param[in] th transaction handle
3401 * \retval 0 on success
3402 * \retval negative if failed
3404 static int lod_xattr_set_internal(const struct lu_env *env,
3405 struct dt_object *dt,
3406 const struct lu_buf *buf,
3407 const char *name, int fl,
3410 struct dt_object *next = dt_object_child(dt);
3411 struct lod_object *lo = lod_dt_obj(dt);
3416 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3417 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3420 /* Note: Do not set LinkEA on sub-stripes, otherwise
3421 * it will confuse the fid2path process(see mdt_path_current()).
3422 * The linkEA between master and sub-stripes is set in
3423 * lod_xattr_set_lmv(). */
3424 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
3427 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3428 if (!lo->ldo_stripe[i])
3431 if (!dt_object_exists(lo->ldo_stripe[i]))
3434 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
3444 * Delete an extended attribute.
3446 * Deletes specified xattr from the object and the stripes if the latter exist.
3448 * \param[in] env execution environment
3449 * \param[in] dt object
3450 * \param[in] name name of xattr
3451 * \param[in] th transaction handle
3453 * \retval 0 on success
3454 * \retval negative if failed
3456 static int lod_xattr_del_internal(const struct lu_env *env,
3457 struct dt_object *dt,
3458 const char *name, struct thandle *th)
3460 struct dt_object *next = dt_object_child(dt);
3461 struct lod_object *lo = lod_dt_obj(dt);
3466 rc = lod_sub_xattr_del(env, next, name, th);
3467 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3470 if (lo->ldo_dir_stripe_count == 0)
3473 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3474 LASSERT(lo->ldo_stripe[i]);
3476 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3485 * Set default striping on a directory.
3487 * Sets specified striping on a directory object unless it matches the default
3488 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3489 * EA. This striping will be used when regular file is being created in this
3492 * \param[in] env execution environment
3493 * \param[in] dt the striped object
3494 * \param[in] buf buffer with the striping
3495 * \param[in] name name of EA
3496 * \param[in] fl xattr flag (see OSD API description)
3497 * \param[in] th transaction handle
3499 * \retval 0 on success
3500 * \retval negative if failed
3502 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
3503 struct dt_object *dt,
3504 const struct lu_buf *buf,
3505 const char *name, int fl,
3508 struct lov_user_md_v1 *lum;
3509 struct lov_user_md_v3 *v3 = NULL;
3510 const char *pool_name = NULL;
3515 LASSERT(buf != NULL && buf->lb_buf != NULL);
3518 switch (lum->lmm_magic) {
3519 case LOV_USER_MAGIC_SPECIFIC:
3520 case LOV_USER_MAGIC_V3:
3522 if (v3->lmm_pool_name[0] != '\0')
3523 pool_name = v3->lmm_pool_name;
3525 case LOV_USER_MAGIC_V1:
3526 /* if { size, offset, count } = { 0, -1, 0 } and no pool
3527 * (i.e. all default values specified) then delete default
3528 * striping from dir. */
3530 "set default striping: sz %u # %u offset %d %s %s\n",
3531 (unsigned)lum->lmm_stripe_size,
3532 (unsigned)lum->lmm_stripe_count,
3533 (int)lum->lmm_stripe_offset,
3534 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
3536 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
3537 lum->lmm_stripe_count,
3538 lum->lmm_stripe_offset,
3541 case LOV_USER_MAGIC_COMP_V1:
3543 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lum;
3544 struct lov_comp_md_entry_v1 *lcme;
3547 comp_cnt = le16_to_cpu(lcm->lcm_entry_count);
3548 for (i = 0; i < comp_cnt; i++) {
3549 lcme = &lcm->lcm_entries[i];
3550 if (lcme->lcme_flags & cpu_to_le32(LCME_FL_EXTENSION)) {
3551 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
3560 CERROR("Invalid magic %x\n", lum->lmm_magic);
3565 rc = lod_xattr_del_internal(env, dt, name, th);
3569 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3576 * Set default striping on a directory object.
3578 * Sets specified striping on a directory object unless it matches the default
3579 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3580 * EA. This striping will be used when a new directory is being created in the
3583 * \param[in] env execution environment
3584 * \param[in] dt the striped object
3585 * \param[in] buf buffer with the striping
3586 * \param[in] name name of EA
3587 * \param[in] fl xattr flag (see OSD API description)
3588 * \param[in] th transaction handle
3590 * \retval 0 on success
3591 * \retval negative if failed
3593 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
3594 struct dt_object *dt,
3595 const struct lu_buf *buf,
3596 const char *name, int fl,
3599 struct lmv_user_md_v1 *lum;
3604 LASSERT(buf != NULL && buf->lb_buf != NULL);
3608 "set default stripe_count # %u stripe_offset %d hash %u\n",
3609 le32_to_cpu(lum->lum_stripe_count),
3610 (int)le32_to_cpu(lum->lum_stripe_offset),
3611 le32_to_cpu(lum->lum_hash_type));
3613 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
3614 le32_to_cpu(lum->lum_stripe_offset)) &&
3615 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
3616 rc = lod_xattr_del_internal(env, dt, name, th);
3620 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3629 * Turn directory into a striped directory.
3631 * During replay the client sends the striping created before MDT
3632 * failure, then the layer above LOD sends this defined striping
3633 * using ->do_xattr_set(), so LOD uses this method to replay creation
3634 * of the stripes. Notice the original information for the striping
3635 * (#stripes, FIDs, etc) was transferred in declare path.
3637 * \param[in] env execution environment
3638 * \param[in] dt the striped object
3639 * \param[in] buf not used currently
3640 * \param[in] name not used currently
3641 * \param[in] fl xattr flag (see OSD API description)
3642 * \param[in] th transaction handle
3644 * \retval 0 on success
3645 * \retval negative if failed
3647 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
3648 const struct lu_buf *buf, const char *name,
3649 int fl, struct thandle *th)
3651 struct lod_object *lo = lod_dt_obj(dt);
3652 struct lod_thread_info *info = lod_env_info(env);
3653 struct lu_attr *attr = &info->lti_attr;
3654 struct dt_object_format *dof = &info->lti_format;
3655 struct lu_buf lmv_buf;
3656 struct lu_buf slave_lmv_buf;
3657 struct lmv_mds_md_v1 *lmm;
3658 struct lmv_mds_md_v1 *slave_lmm = NULL;
3659 struct dt_insert_rec *rec = &info->lti_dt_rec;
3664 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3667 /* The stripes are supposed to be allocated in declare phase,
3668 * if there are no stripes being allocated, it will skip */
3669 if (lo->ldo_dir_stripe_count == 0) {
3670 if (lo->ldo_dir_is_foreign) {
3671 rc = lod_sub_xattr_set(env, dt_object_child(dt), buf,
3672 XATTR_NAME_LMV, fl, th);
3679 rc = dt_attr_get(env, dt_object_child(dt), attr);
3683 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME |
3684 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
3685 dof->dof_type = DFT_DIR;
3687 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
3690 lmm = lmv_buf.lb_buf;
3692 OBD_ALLOC_PTR(slave_lmm);
3693 if (slave_lmm == NULL)
3696 lod_prep_slave_lmv_md(slave_lmm, lmm);
3697 slave_lmv_buf.lb_buf = slave_lmm;
3698 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
3700 rec->rec_type = S_IFDIR;
3701 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3702 struct dt_object *dto = lo->ldo_stripe[i];
3703 char *stripe_name = info->lti_key;
3704 struct lu_name *sname;
3705 struct linkea_data ldata = { NULL };
3706 struct lu_buf linkea_buf;
3708 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
3712 /* fail a remote stripe creation */
3713 if (i && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_CREATE))
3716 /* if it's source stripe of migrating directory, don't create */
3717 if (!((lo->ldo_dir_hash_type & LMV_HASH_FLAG_MIGRATION) &&
3718 i >= lo->ldo_dir_migrate_offset)) {
3719 dt_write_lock(env, dto, DT_TGT_CHILD);
3720 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
3722 dt_write_unlock(env, dto);
3726 rc = lod_sub_ref_add(env, dto, th);
3727 dt_write_unlock(env, dto);
3731 rec->rec_fid = lu_object_fid(&dto->do_lu);
3732 rc = lod_sub_insert(env, dto,
3733 (const struct dt_rec *)rec,
3734 (const struct dt_key *)dot, th);
3739 rec->rec_fid = lu_object_fid(&dt->do_lu);
3740 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
3741 (const struct dt_key *)dotdot, th);
3745 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
3746 cfs_fail_val != i) {
3747 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
3749 slave_lmm->lmv_master_mdt_index =
3752 slave_lmm->lmv_master_mdt_index =
3755 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
3756 XATTR_NAME_LMV, 0, th);
3761 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
3763 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3764 PFID(lu_object_fid(&dto->do_lu)), i + 1);
3766 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3767 PFID(lu_object_fid(&dto->do_lu)), i);
3769 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
3770 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
3771 sname, lu_object_fid(&dt->do_lu));
3775 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
3776 linkea_buf.lb_len = ldata.ld_leh->leh_len;
3777 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
3778 XATTR_NAME_LINK, 0, th);
3782 rec->rec_fid = lu_object_fid(&dto->do_lu);
3783 rc = lod_sub_insert(env, dt_object_child(dt),
3784 (const struct dt_rec *)rec,
3785 (const struct dt_key *)stripe_name, th);
3789 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
3794 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
3795 rc = lod_sub_xattr_set(env, dt_object_child(dt),
3796 &lmv_buf, XATTR_NAME_LMV, fl, th);
3798 if (slave_lmm != NULL)
3799 OBD_FREE_PTR(slave_lmm);
3805 * Helper function to declare/execute creation of a striped directory
3807 * Called in declare/create object path, prepare striping for a directory
3808 * and prepare defaults data striping for the objects to be created in
3809 * that directory. Notice the function calls "declaration" or "execution"
3810 * methods depending on \a declare param. This is a consequence of the
3811 * current approach while we don't have natural distributed transactions:
3812 * we basically execute non-local updates in the declare phase. So, the
3813 * arguments for the both phases are the same and this is the reason for
3814 * this function to exist.
3816 * \param[in] env execution environment
3817 * \param[in] dt object
3818 * \param[in] attr attributes the stripes will be created with
3819 * \param[in] lmu lmv_user_md if MDT indices are specified
3820 * \param[in] dof format of stripes (see OSD API description)
3821 * \param[in] th transaction handle
3822 * \param[in] declare where to call "declare" or "execute" methods
3824 * \retval 0 on success
3825 * \retval negative if failed
3827 static int lod_dir_striping_create_internal(const struct lu_env *env,
3828 struct dt_object *dt,
3829 struct lu_attr *attr,
3830 const struct lu_buf *lmu,
3831 struct dt_object_format *dof,
3835 struct lod_thread_info *info = lod_env_info(env);
3836 struct lod_object *lo = lod_dt_obj(dt);
3837 const struct lod_default_striping *lds = lo->ldo_def_striping;
3841 LASSERT(ergo(lds != NULL,
3842 lds->lds_def_striping_set ||
3843 lds->lds_dir_def_striping_set));
3845 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
3846 lo->ldo_dir_stripe_offset)) {
3848 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3849 int stripe_count = lo->ldo_dir_stripe_count;
3851 if (info->lti_ea_store_size < sizeof(*v1)) {
3852 rc = lod_ea_store_resize(info, sizeof(*v1));
3855 v1 = info->lti_ea_store;
3858 memset(v1, 0, sizeof(*v1));
3859 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3860 v1->lum_stripe_count = cpu_to_le32(stripe_count);
3861 v1->lum_stripe_offset =
3862 cpu_to_le32(lo->ldo_dir_stripe_offset);
3864 info->lti_buf.lb_buf = v1;
3865 info->lti_buf.lb_len = sizeof(*v1);
3866 lmu = &info->lti_buf;
3870 rc = lod_declare_xattr_set_lmv(env, dt, attr, lmu, dof,
3873 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
3878 /* foreign LMV EA case */
3880 struct lmv_foreign_md *lfm = lmu->lb_buf;
3882 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
3883 rc = lod_declare_xattr_set_lmv(env, dt, attr,
3887 if (lo->ldo_dir_is_foreign) {
3888 LASSERT(lo->ldo_foreign_lmv != NULL &&
3889 lo->ldo_foreign_lmv_size > 0);
3890 info->lti_buf.lb_buf = lo->ldo_foreign_lmv;
3891 info->lti_buf.lb_len = lo->ldo_foreign_lmv_size;
3892 lmu = &info->lti_buf;
3893 rc = lod_xattr_set_lmv(env, dt, lmu,
3894 XATTR_NAME_LMV, 0, th);
3899 /* Transfer default LMV striping from the parent */
3900 if (lds != NULL && lds->lds_dir_def_striping_set &&
3901 !(LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
3902 lds->lds_dir_def_stripe_offset) &&
3903 le32_to_cpu(lds->lds_dir_def_hash_type) !=
3904 LMV_HASH_TYPE_UNKNOWN)) {
3905 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3907 if (info->lti_ea_store_size < sizeof(*v1)) {
3908 rc = lod_ea_store_resize(info, sizeof(*v1));
3911 v1 = info->lti_ea_store;
3914 memset(v1, 0, sizeof(*v1));
3915 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3916 v1->lum_stripe_count =
3917 cpu_to_le32(lds->lds_dir_def_stripe_count);
3918 v1->lum_stripe_offset =
3919 cpu_to_le32(lds->lds_dir_def_stripe_offset);
3921 cpu_to_le32(lds->lds_dir_def_hash_type);
3923 info->lti_buf.lb_buf = v1;
3924 info->lti_buf.lb_len = sizeof(*v1);
3926 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3927 XATTR_NAME_DEFAULT_LMV,
3930 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
3932 XATTR_NAME_DEFAULT_LMV, 0,
3938 /* Transfer default LOV striping from the parent */
3939 if (lds != NULL && lds->lds_def_striping_set &&
3940 lds->lds_def_comp_cnt != 0) {
3941 struct lov_mds_md *lmm;
3942 int lmm_size = lod_comp_md_size(lo, true);
3944 if (info->lti_ea_store_size < lmm_size) {
3945 rc = lod_ea_store_resize(info, lmm_size);
3949 lmm = info->lti_ea_store;
3951 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
3955 info->lti_buf.lb_buf = lmm;
3956 info->lti_buf.lb_len = lmm_size;
3959 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3960 XATTR_NAME_LOV, 0, th);
3962 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
3963 XATTR_NAME_LOV, 0, th);
3971 static int lod_declare_dir_striping_create(const struct lu_env *env,
3972 struct dt_object *dt,
3973 struct lu_attr *attr,
3975 struct dt_object_format *dof,
3978 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
3982 static int lod_dir_striping_create(const struct lu_env *env,
3983 struct dt_object *dt,
3984 struct lu_attr *attr,
3985 struct dt_object_format *dof,
3988 return lod_dir_striping_create_internal(env, dt, attr, NULL, dof, th,
3993 * Make LOV EA for striped object.
3995 * Generate striping information and store it in the LOV EA of the given
3996 * object. The caller must ensure nobody else is calling the function
3997 * against the object concurrently. The transaction must be started.
3998 * FLDB service must be running as well; it's used to map FID to the target,
3999 * which is stored in LOV EA.
4001 * \param[in] env execution environment for this thread
4002 * \param[in] lo LOD object
4003 * \param[in] th transaction handle
4005 * \retval 0 if LOV EA is stored successfully
4006 * \retval negative error number on failure
4008 static int lod_generate_and_set_lovea(const struct lu_env *env,
4009 struct lod_object *lo,
4012 struct lod_thread_info *info = lod_env_info(env);
4013 struct dt_object *next = dt_object_child(&lo->ldo_obj);
4014 struct lov_mds_md_v1 *lmm;
4020 if (lo->ldo_comp_cnt == 0 && !lo->ldo_is_foreign) {
4021 lod_striping_free(env, lo);
4022 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
4026 lmm_size = lod_comp_md_size(lo, false);
4027 if (info->lti_ea_store_size < lmm_size) {
4028 rc = lod_ea_store_resize(info, lmm_size);
4032 lmm = info->lti_ea_store;
4034 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
4038 info->lti_buf.lb_buf = lmm;
4039 info->lti_buf.lb_len = lmm_size;
4040 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
4041 XATTR_NAME_LOV, 0, th);
4045 static __u32 lod_gen_component_id(struct lod_object *lo,
4046 int mirror_id, int comp_idx);
4049 * Repeat an existing component
4051 * Creates a new layout by replicating an existing component. Uses striping
4052 * policy from previous component as a template for the striping for the new
4055 * New component starts with zero length, will be extended (or removed) before
4056 * returning layout to client.
4058 * NB: Reallocates layout components array (lo->ldo_comp_entries), invalidating
4059 * any pre-existing pointers to components. Handle with care.
4061 * \param[in] env execution environment for this thread
4062 * \param[in,out] lo object to update the layout of
4063 * \param[in] index index of component to copy
4065 * \retval 0 on success
4066 * \retval negative errno on error
4068 static int lod_layout_repeat_comp(const struct lu_env *env,
4069 struct lod_object *lo, int index)
4071 struct lod_layout_component *lod_comp;
4072 struct lod_layout_component *new_comp = NULL;
4073 struct lod_layout_component *comp_array;
4074 int rc = 0, i, new_cnt = lo->ldo_comp_cnt + 1;
4079 lod_comp = &lo->ldo_comp_entries[index];
4080 LASSERT(lod_comp_inited(lod_comp) && lod_comp->llc_id != LCME_ID_INVAL);
4082 CDEBUG(D_LAYOUT, "repeating component %d\n", index);
4084 OBD_ALLOC(comp_array, sizeof(*comp_array) * new_cnt);
4085 if (comp_array == NULL)
4086 GOTO(out, rc = -ENOMEM);
4088 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4089 memcpy(&comp_array[i + offset], &lo->ldo_comp_entries[i],
4090 sizeof(*comp_array));
4092 /* Duplicate this component in to the next slot */
4094 new_comp = &comp_array[i + 1];
4095 memcpy(&comp_array[i + 1], &lo->ldo_comp_entries[i],
4096 sizeof(*comp_array));
4097 /* We must now skip this new component when copying */
4102 /* Set up copied component */
4103 new_comp->llc_flags &= ~LCME_FL_INIT;
4104 new_comp->llc_stripe = NULL;
4105 new_comp->llc_stripes_allocated = 0;
4106 new_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4107 /* for uninstantiated components, layout gen stores default stripe
4109 new_comp->llc_layout_gen = lod_comp->llc_stripe_offset;
4110 /* This makes the repeated component zero-length, placed at the end of
4111 * the preceding component */
4112 new_comp->llc_extent.e_start = new_comp->llc_extent.e_end;
4113 new_comp->llc_timestamp = lod_comp->llc_timestamp;
4114 new_comp->llc_pool = NULL;
4116 rc = lod_set_pool(&new_comp->llc_pool, lod_comp->llc_pool);
4120 if (new_comp->llc_ostlist.op_array) {
4121 __u32 *op_array = NULL;
4123 OBD_ALLOC(op_array, new_comp->llc_ostlist.op_size);
4125 GOTO(out, rc = -ENOMEM);
4126 memcpy(op_array, &new_comp->llc_ostlist.op_array,
4127 new_comp->llc_ostlist.op_size);
4128 new_comp->llc_ostlist.op_array = op_array;
4131 OBD_FREE(lo->ldo_comp_entries,
4132 sizeof(*comp_array) * lo->ldo_comp_cnt);
4133 lo->ldo_comp_entries = comp_array;
4134 lo->ldo_comp_cnt = new_cnt;
4136 /* Generate an id for the new component */
4137 mirror_id = mirror_id_of(new_comp->llc_id);
4138 new_comp->llc_id = LCME_ID_INVAL;
4139 new_comp->llc_id = lod_gen_component_id(lo, mirror_id, index + 1);
4140 if (new_comp->llc_id == LCME_ID_INVAL)
4141 GOTO(out, rc = -ERANGE);
4146 OBD_FREE(comp_array, sizeof(*comp_array) * new_cnt);
4151 static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
4155 /* clear memory region that will be used for layout change */
4156 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
4157 info->lti_count = 0;
4159 if (info->lti_comp_size >= comp_cnt)
4162 if (info->lti_comp_size > 0) {
4163 OBD_FREE(info->lti_comp_idx,
4164 info->lti_comp_size * sizeof(__u32));
4165 info->lti_comp_size = 0;
4168 OBD_ALLOC(info->lti_comp_idx, comp_cnt * sizeof(__u32));
4169 if (!info->lti_comp_idx)
4172 info->lti_comp_size = comp_cnt;
4177 * Prepare new layout minus deleted components
4179 * Removes components marked for deletion (LCME_ID_INVAL) by copying to a new
4180 * layout and skipping those components. Removes stripe objects if any exist.
4183 * Reallocates layout components array (lo->ldo_comp_entries), invalidating
4184 * any pre-existing pointers to components.
4186 * Caller is responsible for updating mirror end (ldo_mirror[].lme_end).
4188 * \param[in] env execution environment for this thread
4189 * \param[in,out] lo object to update the layout of
4190 * \param[in] th transaction handle for this operation
4192 * \retval # of components deleted
4193 * \retval negative errno on error
4195 static int lod_layout_del_prep_layout(const struct lu_env *env,
4196 struct lod_object *lo,
4199 struct lod_layout_component *lod_comp;
4200 struct lod_thread_info *info = lod_env_info(env);
4201 int rc = 0, i, j, deleted = 0;
4205 LASSERT(lo->ldo_is_composite);
4206 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
4208 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
4212 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4213 lod_comp = &lo->ldo_comp_entries[i];
4215 if (lod_comp->llc_id != LCME_ID_INVAL) {
4216 /* Build array of things to keep */
4217 info->lti_comp_idx[info->lti_count++] = i;
4221 lod_obj_set_pool(lo, i, NULL);
4222 if (lod_comp->llc_ostlist.op_array) {
4223 OBD_FREE(lod_comp->llc_ostlist.op_array,
4224 lod_comp->llc_ostlist.op_size);
4225 lod_comp->llc_ostlist.op_array = NULL;
4226 lod_comp->llc_ostlist.op_size = 0;
4230 CDEBUG(D_LAYOUT, "deleting comp %d, left %d\n", i,
4231 lo->ldo_comp_cnt - deleted);
4233 /* No striping info for this component */
4234 if (lod_comp->llc_stripe == NULL)
4237 LASSERT(lod_comp->llc_stripe_count > 0);
4238 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4239 struct dt_object *obj = lod_comp->llc_stripe[j];
4244 /* components which are not init have no sub objects
4246 if (lod_comp_inited(lod_comp)) {
4247 rc = lod_sub_destroy(env, obj, th);
4252 lu_object_put(env, &obj->do_lu);
4253 lod_comp->llc_stripe[j] = NULL;
4255 OBD_FREE(lod_comp->llc_stripe, sizeof(*lod_comp->llc_stripe) *
4256 lod_comp->llc_stripes_allocated);
4257 lod_comp->llc_stripe = NULL;
4258 OBD_FREE(lod_comp->llc_ost_indices,
4259 sizeof(__u32) * lod_comp->llc_stripes_allocated);
4260 lod_comp->llc_ost_indices = NULL;
4261 lod_comp->llc_stripes_allocated = 0;
4264 /* info->lti_count has the amount of left components */
4265 LASSERTF(info->lti_count >= 0 && info->lti_count < lo->ldo_comp_cnt,
4266 "left = %d, lo->ldo_comp_cnt %d\n", (int)info->lti_count,
4267 (int)lo->ldo_comp_cnt);
4269 if (info->lti_count > 0) {
4270 struct lod_layout_component *comp_array;
4272 OBD_ALLOC(comp_array, sizeof(*comp_array) * info->lti_count);
4273 if (comp_array == NULL)
4274 GOTO(out, rc = -ENOMEM);
4276 for (i = 0; i < info->lti_count; i++) {
4277 memcpy(&comp_array[i],
4278 &lo->ldo_comp_entries[info->lti_comp_idx[i]],
4279 sizeof(*comp_array));
4282 OBD_FREE(lo->ldo_comp_entries,
4283 sizeof(*comp_array) * lo->ldo_comp_cnt);
4284 lo->ldo_comp_entries = comp_array;
4285 lo->ldo_comp_cnt = info->lti_count;
4287 lod_free_comp_entries(lo);
4292 return rc ? rc : deleted;
4296 * Delete layout component(s)
4298 * This function sets up the layout data in the env and does the setattrs
4299 * required to write out the new layout. The layout itself is modified in
4300 * lod_layout_del_prep_layout.
4302 * \param[in] env execution environment for this thread
4303 * \param[in] dt object
4304 * \param[in] th transaction handle
4306 * \retval 0 on success
4307 * \retval negative error number on failure
4309 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
4312 struct lod_object *lo = lod_dt_obj(dt);
4313 struct dt_object *next = dt_object_child(dt);
4314 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4317 LASSERT(lo->ldo_mirror_count == 1);
4319 rc = lod_layout_del_prep_layout(env, lo, th);
4323 /* Only do this if we didn't delete all components */
4324 if (lo->ldo_comp_cnt > 0) {
4325 lo->ldo_mirrors[0].lme_end = lo->ldo_comp_cnt - 1;
4326 lod_obj_inc_layout_gen(lo);
4329 LASSERT(dt_object_exists(dt));
4330 rc = dt_attr_get(env, next, attr);
4334 if (attr->la_size > 0) {
4336 attr->la_valid = LA_SIZE;
4337 rc = lod_sub_attr_set(env, next, attr, th);
4342 rc = lod_generate_and_set_lovea(env, lo, th);
4346 lod_striping_free(env, lo);
4351 static int lod_get_default_lov_striping(const struct lu_env *env,
4352 struct lod_object *lo,
4353 struct lod_default_striping *lds,
4354 struct dt_allocation_hint *ah);
4356 * Implementation of dt_object_operations::do_xattr_set.
4358 * Sets specified extended attribute on the object. Three types of EAs are
4360 * LOV EA - stores striping for a regular file or default striping (when set
4362 * LMV EA - stores a marker for the striped directories
4363 * DMV EA - stores default directory striping
4365 * When striping is applied to a non-striped existing object (this is called
4366 * late striping), then LOD notices the caller wants to turn the object into a
4367 * striped one. The stripe objects are created and appropriate EA is set:
4368 * LOV EA storing all the stripes directly or LMV EA storing just a small header
4369 * with striping configuration.
4371 * \see dt_object_operations::do_xattr_set() in the API description for details.
4373 static int lod_xattr_set(const struct lu_env *env,
4374 struct dt_object *dt, const struct lu_buf *buf,
4375 const char *name, int fl, struct thandle *th)
4377 struct dt_object *next = dt_object_child(dt);
4382 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4383 !strcmp(name, XATTR_NAME_LMV)) {
4385 case LU_XATTR_CREATE:
4386 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
4389 case LU_XATTR_REPLACE:
4390 rc = lod_dir_layout_set(env, dt, buf, fl, th);
4397 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4398 strcmp(name, XATTR_NAME_LOV) == 0) {
4399 struct lod_default_striping *lds = lod_lds_buf_get(env);
4400 struct lov_user_md_v1 *v1 = buf->lb_buf;
4401 char pool[LOV_MAXPOOLNAME + 1];
4404 /* get existing striping config */
4405 rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds,
4410 memset(pool, 0, sizeof(pool));
4411 if (lds->lds_def_striping_set == 1)
4412 lod_layout_get_pool(lds->lds_def_comp_entries,
4413 lds->lds_def_comp_cnt, pool,
4416 is_del = LOVEA_DELETE_VALUES(v1->lmm_stripe_size,
4417 v1->lmm_stripe_count,
4418 v1->lmm_stripe_offset,
4421 /* Retain the pool name if it is not given */
4422 if (v1->lmm_magic == LOV_USER_MAGIC_V1 && pool[0] != '\0' &&
4424 struct lod_thread_info *info = lod_env_info(env);
4425 struct lov_user_md_v3 *v3 = info->lti_ea_store;
4427 memset(v3, 0, sizeof(*v3));
4428 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4429 v3->lmm_pattern = cpu_to_le32(v1->lmm_pattern);
4430 v3->lmm_stripe_count =
4431 cpu_to_le32(v1->lmm_stripe_count);
4432 v3->lmm_stripe_offset =
4433 cpu_to_le32(v1->lmm_stripe_offset);
4434 v3->lmm_stripe_size = cpu_to_le32(v1->lmm_stripe_size);
4436 strlcpy(v3->lmm_pool_name, pool,
4437 sizeof(v3->lmm_pool_name));
4439 info->lti_buf.lb_buf = v3;
4440 info->lti_buf.lb_len = sizeof(*v3);
4441 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4444 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name,
4448 if (lds->lds_def_striping_set == 1 &&
4449 lds->lds_def_comp_entries != NULL)
4450 lod_free_def_comp_entries(lds);
4453 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4454 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
4456 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
4459 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
4460 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4461 strcmp(name, XATTR_LUSTRE_LOV) == 0 ||
4462 allowed_lustre_lov(name))) {
4463 /* in case of lov EA swap, just set it
4464 * if not, it is a replay so check striping match what we
4465 * already have during req replay, declare_xattr_set()
4466 * defines striping, then create() does the work */
4467 if (fl & LU_XATTR_REPLACE) {
4468 /* free stripes, then update disk */
4469 lod_striping_free(env, lod_dt_obj(dt));
4471 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
4472 } else if (dt_object_remote(dt)) {
4473 /* This only happens during migration, see
4474 * mdd_migrate_create(), in which Master MDT will
4475 * create a remote target object, and only set
4476 * (migrating) stripe EA on the remote object,
4477 * and does not need creating each stripes. */
4478 rc = lod_sub_xattr_set(env, next, buf, name,
4480 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
4481 /* delete component(s) */
4482 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
4483 rc = lod_layout_del(env, dt, th);
4486 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
4487 * it's going to create create file with specified
4488 * component(s), the striping must have not being
4489 * cached in this case;
4491 * Otherwise, it's going to add/change component(s) to
4492 * an existing file, the striping must have been cached
4495 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
4496 !strcmp(name, XATTR_NAME_LOV),
4497 !lod_dt_obj(dt)->ldo_comp_cached));
4499 rc = lod_striped_create(env, dt, NULL, NULL, th);
4502 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4503 rc = lod_replace_parent_fid(env, dt, buf, th, false);
4508 /* then all other xattr */
4509 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4515 * Implementation of dt_object_operations::do_declare_xattr_del.
4517 * \see dt_object_operations::do_declare_xattr_del() in the API description
4520 static int lod_declare_xattr_del(const struct lu_env *env,
4521 struct dt_object *dt, const char *name,
4524 struct lod_object *lo = lod_dt_obj(dt);
4525 struct dt_object *next = dt_object_child(dt);
4530 rc = lod_sub_declare_xattr_del(env, next, name, th);
4534 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4537 /* NB: don't delete stripe LMV, because when we do this, normally we
4538 * will remove stripes, besides, if directory LMV is corrupt, this will
4539 * prevent deleting its LMV and fixing it (via LFSCK).
4541 if (!strcmp(name, XATTR_NAME_LMV))
4544 rc = lod_striping_load(env, lo);
4548 if (lo->ldo_dir_stripe_count == 0)
4551 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4552 struct dt_object *dto = lo->ldo_stripe[i];
4557 rc = lod_sub_declare_xattr_del(env, dto, name, th);
4566 * Implementation of dt_object_operations::do_xattr_del.
4568 * If EA storing a regular striping is being deleted, then release
4569 * all the references to the stripe objects in core.
4571 * \see dt_object_operations::do_xattr_del() in the API description for details.
4573 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
4574 const char *name, struct thandle *th)
4576 struct dt_object *next = dt_object_child(dt);
4577 struct lod_object *lo = lod_dt_obj(dt);
4582 if (!strcmp(name, XATTR_NAME_LOV) || !strcmp(name, XATTR_NAME_LMV))
4583 lod_striping_free(env, lod_dt_obj(dt));
4585 rc = lod_sub_xattr_del(env, next, name, th);
4586 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
4589 if (!strcmp(name, XATTR_NAME_LMV))
4592 if (lo->ldo_dir_stripe_count == 0)
4595 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4596 struct dt_object *dto = lo->ldo_stripe[i];
4601 rc = lod_sub_xattr_del(env, dto, name, th);
4610 * Implementation of dt_object_operations::do_xattr_list.
4612 * \see dt_object_operations::do_xattr_list() in the API description
4615 static int lod_xattr_list(const struct lu_env *env,
4616 struct dt_object *dt, const struct lu_buf *buf)
4618 return dt_xattr_list(env, dt_object_child(dt), buf);
4621 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
4623 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
4627 * Copy OST list from layout provided by user.
4629 * \param[in] lod_comp layout_component to be filled
4630 * \param[in] v3 LOV EA V3 user data
4632 * \retval 0 on success
4633 * \retval negative if failed
4635 int lod_comp_copy_ost_lists(struct lod_layout_component *lod_comp,
4636 struct lov_user_md_v3 *v3)
4642 if (v3->lmm_stripe_offset == LOV_OFFSET_DEFAULT)
4643 v3->lmm_stripe_offset = v3->lmm_objects[0].l_ost_idx;
4645 if (lod_comp->llc_ostlist.op_array) {
4646 if (lod_comp->llc_ostlist.op_size >=
4647 v3->lmm_stripe_count * sizeof(__u32)) {
4648 lod_comp->llc_ostlist.op_count =
4649 v3->lmm_stripe_count;
4652 OBD_FREE(lod_comp->llc_ostlist.op_array,
4653 lod_comp->llc_ostlist.op_size);
4656 /* copy ost list from lmm */
4657 lod_comp->llc_ostlist.op_count = v3->lmm_stripe_count;
4658 lod_comp->llc_ostlist.op_size = v3->lmm_stripe_count * sizeof(__u32);
4659 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
4660 lod_comp->llc_ostlist.op_size);
4661 if (!lod_comp->llc_ostlist.op_array)
4664 for (j = 0; j < v3->lmm_stripe_count; j++) {
4665 lod_comp->llc_ostlist.op_array[j] =
4666 v3->lmm_objects[j].l_ost_idx;
4674 * Get default striping.
4676 * \param[in] env execution environment
4677 * \param[in] lo object
4678 * \param[out] lds default striping
4680 * \retval 0 on success
4681 * \retval negative if failed
4683 static int lod_get_default_lov_striping(const struct lu_env *env,
4684 struct lod_object *lo,
4685 struct lod_default_striping *lds,
4686 struct dt_allocation_hint *ah)
4688 struct lod_thread_info *info = lod_env_info(env);
4689 struct lov_user_md_v1 *v1 = NULL;
4690 struct lov_user_md_v3 *v3 = NULL;
4691 struct lov_comp_md_v1 *comp_v1 = NULL;
4699 rc = lod_get_lov_ea(env, lo);
4703 if (rc < (typeof(rc))sizeof(struct lov_user_md))
4706 v1 = info->lti_ea_store;
4707 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
4708 lustre_swab_lov_user_md_v1(v1);
4709 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
4710 v3 = (struct lov_user_md_v3 *)v1;
4711 lustre_swab_lov_user_md_v3(v3);
4712 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_SPECIFIC)) {
4713 v3 = (struct lov_user_md_v3 *)v1;
4714 lustre_swab_lov_user_md_v3(v3);
4715 lustre_swab_lov_user_md_objects(v3->lmm_objects,
4716 v3->lmm_stripe_count);
4717 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_COMP_V1) ||
4718 v1->lmm_magic == __swab32(LOV_USER_MAGIC_SEL)) {
4719 comp_v1 = (struct lov_comp_md_v1 *)v1;
4720 lustre_swab_lov_comp_md_v1(comp_v1);
4723 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1 &&
4724 v1->lmm_magic != LOV_MAGIC_COMP_V1 &&
4725 v1->lmm_magic != LOV_MAGIC_SEL &&
4726 v1->lmm_magic != LOV_USER_MAGIC_SPECIFIC)
4729 if ((v1->lmm_magic == LOV_MAGIC_COMP_V1 ||
4730 v1->lmm_magic == LOV_MAGIC_SEL) &&
4731 !(ah && ah->dah_append_stripes)) {
4732 comp_v1 = (struct lov_comp_md_v1 *)v1;
4733 comp_cnt = comp_v1->lcm_entry_count;
4736 mirror_cnt = comp_v1->lcm_mirror_count + 1;
4744 /* realloc default comp entries if necessary */
4745 rc = lod_def_striping_comp_resize(lds, comp_cnt);
4749 lds->lds_def_comp_cnt = comp_cnt;
4750 lds->lds_def_striping_is_composite = composite;
4751 lds->lds_def_mirror_cnt = mirror_cnt;
4753 for (i = 0; i < comp_cnt; i++) {
4754 struct lod_layout_component *lod_comp;
4757 lod_comp = &lds->lds_def_comp_entries[i];
4759 * reset lod_comp values, llc_stripes is always NULL in
4760 * the default striping template, llc_pool will be reset
4763 memset(lod_comp, 0, offsetof(typeof(*lod_comp), llc_pool));
4766 v1 = (struct lov_user_md *)((char *)comp_v1 +
4767 comp_v1->lcm_entries[i].lcme_offset);
4768 lod_comp->llc_extent =
4769 comp_v1->lcm_entries[i].lcme_extent;
4770 /* We only inherit certain flags from the layout */
4771 lod_comp->llc_flags =
4772 comp_v1->lcm_entries[i].lcme_flags &
4773 LCME_TEMPLATE_FLAGS;
4776 if (!lov_pattern_supported(v1->lmm_pattern) &&
4777 !(v1->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
4778 lod_free_def_comp_entries(lds);
4782 CDEBUG(D_LAYOUT, DFID" stripe_count=%d stripe_size=%d stripe_offset=%d append_stripes=%d\n",
4783 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
4784 (int)v1->lmm_stripe_count, (int)v1->lmm_stripe_size,
4785 (int)v1->lmm_stripe_offset,
4786 ah ? ah->dah_append_stripes : 0);
4788 if (ah && ah->dah_append_stripes)
4789 lod_comp->llc_stripe_count = ah->dah_append_stripes;
4791 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
4792 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
4793 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
4794 lod_comp->llc_pattern = v1->lmm_pattern;
4797 if (ah && ah->dah_append_pool && ah->dah_append_pool[0]) {
4798 pool = ah->dah_append_pool;
4799 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
4800 /* XXX: sanity check here */
4801 v3 = (struct lov_user_md_v3 *) v1;
4802 if (v3->lmm_pool_name[0] != '\0')
4803 pool = v3->lmm_pool_name;
4805 lod_set_def_pool(lds, i, pool);
4806 if (v1->lmm_magic == LOV_USER_MAGIC_SPECIFIC) {
4807 v3 = (struct lov_user_md_v3 *)v1;
4808 rc = lod_comp_copy_ost_lists(lod_comp, v3);
4811 } else if (lod_comp->llc_ostlist.op_array &&
4812 lod_comp->llc_ostlist.op_count) {
4813 for (j = 0; j < lod_comp->llc_ostlist.op_count; j++)
4814 lod_comp->llc_ostlist.op_array[j] = -1;
4815 lod_comp->llc_ostlist.op_count = 0;
4819 lds->lds_def_striping_set = 1;
4824 * Get default directory striping.
4826 * \param[in] env execution environment
4827 * \param[in] lo object
4828 * \param[out] lds default striping
4830 * \retval 0 on success
4831 * \retval negative if failed
4833 static int lod_get_default_lmv_striping(const struct lu_env *env,
4834 struct lod_object *lo,
4835 struct lod_default_striping *lds)
4837 struct lmv_user_md *lmu;
4840 lds->lds_dir_def_striping_set = 0;
4842 rc = lod_get_default_lmv_ea(env, lo);
4846 if (rc >= (int)sizeof(*lmu)) {
4847 struct lod_thread_info *info = lod_env_info(env);
4849 lmu = info->lti_ea_store;
4851 lds->lds_dir_def_stripe_count =
4852 le32_to_cpu(lmu->lum_stripe_count);
4853 lds->lds_dir_def_stripe_offset =
4854 le32_to_cpu(lmu->lum_stripe_offset);
4855 lds->lds_dir_def_hash_type =
4856 le32_to_cpu(lmu->lum_hash_type);
4857 lds->lds_dir_def_striping_set = 1;
4864 * Get default striping in the object.
4866 * Get object default striping and default directory striping.
4868 * \param[in] env execution environment
4869 * \param[in] lo object
4870 * \param[out] lds default striping
4872 * \retval 0 on success
4873 * \retval negative if failed
4875 static int lod_get_default_striping(const struct lu_env *env,
4876 struct lod_object *lo,
4877 struct lod_default_striping *lds)
4881 rc = lod_get_default_lov_striping(env, lo, lds, NULL);
4882 rc1 = lod_get_default_lmv_striping(env, lo, lds);
4883 if (rc == 0 && rc1 < 0)
4890 * Apply default striping on object.
4892 * If object striping pattern is not set, set to the one in default striping.
4893 * The default striping is from parent or fs.
4895 * \param[in] lo new object
4896 * \param[in] lds default striping
4897 * \param[in] mode new object's mode
4899 static void lod_striping_from_default(struct lod_object *lo,
4900 const struct lod_default_striping *lds,
4903 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
4906 if (lds->lds_def_striping_set && S_ISREG(mode)) {
4907 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
4909 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
4910 lds->lds_def_comp_cnt);
4914 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
4915 if (lds->lds_def_mirror_cnt > 1)
4916 lo->ldo_flr_state = LCM_FL_RDONLY;
4918 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4919 struct lod_layout_component *obj_comp =
4920 &lo->ldo_comp_entries[i];
4921 struct lod_layout_component *def_comp =
4922 &lds->lds_def_comp_entries[i];
4924 CDEBUG(D_LAYOUT, "Inherit from default: flags=%#x "
4925 "size=%hu nr=%u offset=%u pattern=%#x pool=%s\n",
4926 def_comp->llc_flags,
4927 def_comp->llc_stripe_size,
4928 def_comp->llc_stripe_count,
4929 def_comp->llc_stripe_offset,
4930 def_comp->llc_pattern,
4931 def_comp->llc_pool ?: "");
4933 *obj_comp = *def_comp;
4934 if (def_comp->llc_pool != NULL) {
4935 /* pointer was copied from def_comp */
4936 obj_comp->llc_pool = NULL;
4937 lod_obj_set_pool(lo, i, def_comp->llc_pool);
4941 if (def_comp->llc_ostlist.op_array &&
4942 def_comp->llc_ostlist.op_count) {
4943 OBD_ALLOC(obj_comp->llc_ostlist.op_array,
4944 obj_comp->llc_ostlist.op_size);
4945 if (!obj_comp->llc_ostlist.op_array)
4947 memcpy(obj_comp->llc_ostlist.op_array,
4948 def_comp->llc_ostlist.op_array,
4949 obj_comp->llc_ostlist.op_size);
4950 } else if (def_comp->llc_ostlist.op_array) {
4951 obj_comp->llc_ostlist.op_array = NULL;
4955 * Don't initialize these fields for plain layout
4956 * (v1/v3) here, they are inherited in the order of
4957 * 'parent' -> 'fs default (root)' -> 'global default
4958 * values for stripe_count & stripe_size'.
4960 * see lod_ah_init().
4962 if (!lo->ldo_is_composite)
4965 lod_adjust_stripe_info(obj_comp, desc, 0);
4967 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
4968 if (lo->ldo_dir_stripe_count == 0)
4969 lo->ldo_dir_stripe_count =
4970 lds->lds_dir_def_stripe_count;
4971 if (lo->ldo_dir_stripe_offset == -1)
4972 lo->ldo_dir_stripe_offset =
4973 lds->lds_dir_def_stripe_offset;
4974 if (lo->ldo_dir_hash_type == 0)
4975 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
4977 CDEBUG(D_LAYOUT, "striping from default dir: count:%hu, "
4978 "offset:%u, hash_type:%u\n",
4979 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
4980 lo->ldo_dir_hash_type);
4984 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root,
4987 struct lod_layout_component *lod_comp;
4989 if (lo->ldo_comp_cnt == 0)
4992 if (lo->ldo_is_composite)
4995 lod_comp = &lo->ldo_comp_entries[0];
4997 if (lod_comp->llc_stripe_count <= 0 ||
4998 lod_comp->llc_stripe_size <= 0)
5001 if (from_root && (lod_comp->llc_pool == NULL ||
5002 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
5005 if (append_pool && append_pool[0])
5012 * Implementation of dt_object_operations::do_ah_init.
5014 * This method is used to make a decision on the striping configuration for the
5015 * object being created. It can be taken from the \a parent object if it exists,
5016 * or filesystem's default. The resulting configuration (number of stripes,
5017 * stripe size/offset, pool name, etc) is stored in the object itself and will
5018 * be used by the methods like ->doo_declare_create().
5020 * \see dt_object_operations::do_ah_init() in the API description for details.
5022 static void lod_ah_init(const struct lu_env *env,
5023 struct dt_allocation_hint *ah,
5024 struct dt_object *parent,
5025 struct dt_object *child,
5028 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
5029 struct lod_thread_info *info = lod_env_info(env);
5030 struct lod_default_striping *lds = lod_lds_buf_get(env);
5031 struct dt_object *nextp = NULL;
5032 struct dt_object *nextc;
5033 struct lod_object *lp = NULL;
5034 struct lod_object *lc;
5035 struct lov_desc *desc;
5036 struct lod_layout_component *lod_comp;
5042 if (ah->dah_append_stripes == -1)
5043 ah->dah_append_stripes =
5044 d->lod_ost_descs.ltd_lov_desc.ld_tgt_count;
5046 if (likely(parent)) {
5047 nextp = dt_object_child(parent);
5048 lp = lod_dt_obj(parent);
5051 nextc = dt_object_child(child);
5052 lc = lod_dt_obj(child);
5054 LASSERT(!lod_obj_is_striped(child));
5055 /* default layout template may have been set on the regular file
5056 * when this is called from mdd_create_data() */
5057 if (S_ISREG(child_mode))
5058 lod_free_comp_entries(lc);
5060 if (!dt_object_exists(nextc))
5061 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
5063 if (S_ISDIR(child_mode)) {
5064 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
5066 /* other default values are 0 */
5067 lc->ldo_dir_stripe_offset = -1;
5069 /* no default striping configuration is needed for
5072 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5073 le32_to_cpu(lum1->lum_magic) == LMV_MAGIC_FOREIGN) {
5074 lc->ldo_dir_is_foreign = true;
5075 /* keep stripe_count 0 and stripe_offset -1 */
5076 CDEBUG(D_INFO, "no default striping for foreign dir\n");
5081 * If parent object is not root directory,
5082 * then get default striping from parent object.
5084 if (likely(lp != NULL)) {
5085 lod_get_default_striping(env, lp, lds);
5087 /* inherit default striping except ROOT */
5088 if ((lds->lds_def_striping_set ||
5089 lds->lds_dir_def_striping_set) &&
5090 !fid_is_root(lod_object_fid(lp)))
5091 lc->ldo_def_striping = lds;
5094 /* It should always honour the specified stripes */
5095 /* Note: old client (< 2.7)might also do lfs mkdir, whose EA
5096 * will have old magic. In this case, we should ignore the
5097 * stripe count and try to create dir by default stripe.
5099 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5100 (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
5101 le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC)) {
5102 lc->ldo_dir_stripe_count =
5103 le32_to_cpu(lum1->lum_stripe_count);
5104 lc->ldo_dir_stripe_offset =
5105 le32_to_cpu(lum1->lum_stripe_offset);
5106 lc->ldo_dir_hash_type =
5107 le32_to_cpu(lum1->lum_hash_type);
5109 "set dirstripe: count %hu, offset %d, hash %u\n",
5110 lc->ldo_dir_stripe_count,
5111 (int)lc->ldo_dir_stripe_offset,
5112 lc->ldo_dir_hash_type);
5114 /* transfer defaults LMV to new directory */
5115 lod_striping_from_default(lc, lds, child_mode);
5117 /* set count 0 to create normal directory */
5118 if (lc->ldo_dir_stripe_count == 1)
5119 lc->ldo_dir_stripe_count = 0;
5122 /* shrink the stripe_count to the avaible MDT count */
5123 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
5124 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)) {
5125 lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
5126 if (lc->ldo_dir_stripe_count == 1)
5127 lc->ldo_dir_stripe_count = 0;
5130 if (lc->ldo_dir_hash_type == LMV_HASH_TYPE_UNKNOWN)
5131 lc->ldo_dir_hash_type =
5132 d->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
5134 CDEBUG(D_INFO, "final dir stripe [%hu %d %u]\n",
5135 lc->ldo_dir_stripe_count,
5136 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
5141 /* child object regular file*/
5143 if (!lod_object_will_be_striped(S_ISREG(child_mode),
5144 lu_object_fid(&child->do_lu)))
5147 /* If object is going to be striped over OSTs, transfer default
5148 * striping information to the child, so that we can use it
5149 * during declaration and creation.
5151 * Try from the parent first.
5153 if (likely(lp != NULL)) {
5154 rc = lod_get_default_lov_striping(env, lp, lds, ah);
5156 lod_striping_from_default(lc, lds, child_mode);
5159 /* Initialize lod_device::lod_md_root object reference */
5160 if (d->lod_md_root == NULL) {
5161 struct dt_object *root;
5162 struct lod_object *lroot;
5164 lu_root_fid(&info->lti_fid);
5165 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
5166 if (!IS_ERR(root)) {
5167 lroot = lod_dt_obj(root);
5169 spin_lock(&d->lod_lock);
5170 if (d->lod_md_root != NULL)
5171 dt_object_put(env, &d->lod_md_root->ldo_obj);
5172 d->lod_md_root = lroot;
5173 spin_unlock(&d->lod_lock);
5177 /* try inherit layout from the root object (fs default) when:
5178 * - parent does not have default layout; or
5179 * - parent has plain(v1/v3) default layout, and some attributes
5180 * are not specified in the default layout;
5182 if (d->lod_md_root != NULL &&
5183 lod_need_inherit_more(lc, true, ah->dah_append_pool)) {
5184 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds,
5188 if (lc->ldo_comp_cnt == 0) {
5189 lod_striping_from_default(lc, lds, child_mode);
5190 } else if (!lds->lds_def_striping_is_composite) {
5191 struct lod_layout_component *def_comp;
5193 LASSERT(!lc->ldo_is_composite);
5194 lod_comp = &lc->ldo_comp_entries[0];
5195 def_comp = &lds->lds_def_comp_entries[0];
5197 if (lod_comp->llc_stripe_count <= 0)
5198 lod_comp->llc_stripe_count =
5199 def_comp->llc_stripe_count;
5200 if (lod_comp->llc_stripe_size <= 0)
5201 lod_comp->llc_stripe_size =
5202 def_comp->llc_stripe_size;
5203 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT &&
5204 (!lod_comp->llc_pool || !lod_comp->llc_pool[0]))
5205 lod_comp->llc_stripe_offset =
5206 def_comp->llc_stripe_offset;
5207 if (lod_comp->llc_pool == NULL)
5208 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
5213 * fs default striping may not be explicitly set, or historically set
5214 * in config log, use them.
5216 if (lod_need_inherit_more(lc, false, ah->dah_append_pool)) {
5217 if (lc->ldo_comp_cnt == 0) {
5218 rc = lod_alloc_comp_entries(lc, 0, 1);
5220 /* fail to allocate memory, will create a
5221 * non-striped file. */
5223 lc->ldo_is_composite = 0;
5224 lod_comp = &lc->ldo_comp_entries[0];
5225 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
5227 LASSERT(!lc->ldo_is_composite);
5228 lod_comp = &lc->ldo_comp_entries[0];
5229 desc = &d->lod_ost_descs.ltd_lov_desc;
5230 lod_adjust_stripe_info(lod_comp, desc, ah->dah_append_stripes);
5231 if (ah->dah_append_pool && ah->dah_append_pool[0])
5232 lod_obj_set_pool(lc, 0, ah->dah_append_pool);
5238 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
5240 * Size initialization on late striping.
5242 * Propagate the size of a truncated object to a deferred striping.
5243 * This function handles a special case when truncate was done on a
5244 * non-striped object and now while the striping is being created
5245 * we can't lose that size, so we have to propagate it to the stripes
5248 * \param[in] env execution environment
5249 * \param[in] dt object
5250 * \param[in] th transaction handle
5252 * \retval 0 on success
5253 * \retval negative if failed
5255 static int lod_declare_init_size(const struct lu_env *env,
5256 struct dt_object *dt, struct thandle *th)
5258 struct dt_object *next = dt_object_child(dt);
5259 struct lod_object *lo = lod_dt_obj(dt);
5260 struct dt_object **objects = NULL;
5261 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
5262 uint64_t size, offs;
5263 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
5264 struct lu_extent size_ext;
5267 if (!lod_obj_is_striped(dt))
5270 rc = dt_attr_get(env, next, attr);
5271 LASSERT(attr->la_valid & LA_SIZE);
5275 size = attr->la_size;
5279 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
5280 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5281 struct lod_layout_component *lod_comp;
5282 struct lu_extent *extent;
5284 lod_comp = &lo->ldo_comp_entries[i];
5286 if (lod_comp->llc_stripe == NULL)
5289 extent = &lod_comp->llc_extent;
5290 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
5291 if (!lo->ldo_is_composite ||
5292 lu_extent_is_overlapped(extent, &size_ext)) {
5293 objects = lod_comp->llc_stripe;
5294 stripe_count = lod_comp->llc_stripe_count;
5295 stripe_size = lod_comp->llc_stripe_size;
5298 if (stripe_count == 0)
5301 LASSERT(objects != NULL && stripe_size != 0);
5302 /* ll_do_div64(a, b) returns a % b, and a = a / b */
5303 ll_do_div64(size, (__u64)stripe_size);
5304 stripe = ll_do_div64(size, (__u64)stripe_count);
5305 LASSERT(objects[stripe] != NULL);
5307 size = size * stripe_size;
5308 offs = attr->la_size;
5309 size += ll_do_div64(offs, stripe_size);
5311 attr->la_valid = LA_SIZE;
5312 attr->la_size = size;
5314 rc = lod_sub_declare_attr_set(env, objects[stripe],
5323 * Declare creation of striped object.
5325 * The function declares creation stripes for a regular object. The function
5326 * also declares whether the stripes will be created with non-zero size if
5327 * previously size was set non-zero on the master object. If object \a dt is
5328 * not local, then only fully defined striping can be applied in \a lovea.
5329 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
5332 * \param[in] env execution environment
5333 * \param[in] dt object
5334 * \param[in] attr attributes the stripes will be created with
5335 * \param[in] lovea a buffer containing striping description
5336 * \param[in] th transaction handle
5338 * \retval 0 on success
5339 * \retval negative if failed
5341 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
5342 struct lu_attr *attr,
5343 const struct lu_buf *lovea, struct thandle *th)
5345 struct lod_thread_info *info = lod_env_info(env);
5346 struct dt_object *next = dt_object_child(dt);
5347 struct lod_object *lo = lod_dt_obj(dt);
5351 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
5352 GOTO(out, rc = -ENOMEM);
5354 if (!dt_object_remote(next)) {
5355 /* choose OST and generate appropriate objects */
5356 rc = lod_prepare_create(env, lo, attr, lovea, th);
5361 * declare storage for striping data
5363 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
5365 /* LOD can not choose OST objects for remote objects, i.e.
5366 * stripes must be ready before that. Right now, it can only
5367 * happen during migrate, i.e. migrate process needs to create
5368 * remote regular file (mdd_migrate_create), then the migrate
5369 * process will provide stripeEA. */
5370 LASSERT(lovea != NULL);
5371 info->lti_buf = *lovea;
5374 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
5375 XATTR_NAME_LOV, 0, th);
5380 * if striping is created with local object's size > 0,
5381 * we have to propagate this size to specific object
5382 * the case is possible only when local object was created previously
5384 if (dt_object_exists(next))
5385 rc = lod_declare_init_size(env, dt, th);
5388 /* failed to create striping or to set initial size, let's reset
5389 * config so that others don't get confused */
5391 lod_striping_free(env, lo);
5397 * Whether subdirectories under \a dt should be created on MDTs by space QoS
5399 * If LMV_HASH_FLAG_SPACE is set on directory default layout, its subdirectories
5400 * should be created on MDT by space QoS.
5402 * \param[in] env execution environment
5403 * \param[in] dev lu device
5404 * \param[in] dt object
5406 * \retval 1 if directory should create subdir by space usage
5408 * \retval -ev if failed
5410 static inline int dt_object_qos_mkdir(const struct lu_env *env,
5411 struct lu_device *dev,
5412 struct dt_object *dt)
5414 struct lod_thread_info *info = lod_env_info(env);
5415 struct lu_object *obj;
5416 struct lod_object *lo;
5417 struct lmv_user_md *lmu;
5420 obj = lu_object_find_slice(env, dev, lu_object_fid(&dt->do_lu), NULL);
5422 return PTR_ERR(obj);
5424 lo = lu2lod_obj(obj);
5426 rc = lod_get_default_lmv_ea(env, lo);
5427 dt_object_put(env, dt);
5431 if (rc < (int)sizeof(*lmu))
5434 lmu = info->lti_ea_store;
5435 return le32_to_cpu(lmu->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
5439 * Implementation of dt_object_operations::do_declare_create.
5441 * The method declares creation of a new object. If the object will be striped,
5442 * then helper functions are called to find FIDs for the stripes, declare
5443 * creation of the stripes and declare initialization of the striping
5444 * information to be stored in the master object.
5446 * \see dt_object_operations::do_declare_create() in the API description
5449 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
5450 struct lu_attr *attr,
5451 struct dt_allocation_hint *hint,
5452 struct dt_object_format *dof, struct thandle *th)
5454 struct dt_object *next = dt_object_child(dt);
5455 struct lod_object *lo = lod_dt_obj(dt);
5464 * first of all, we declare creation of local object
5466 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
5471 * it's lod_ah_init() that has decided the object will be striped
5473 if (dof->dof_type == DFT_REGULAR) {
5474 /* callers don't want stripes */
5475 /* XXX: all tricky interactions with ->ah_make_hint() decided
5476 * to use striping, then ->declare_create() behaving differently
5477 * should be cleaned */
5478 if (dof->u.dof_reg.striped != 0)
5479 rc = lod_declare_striped_create(env, dt, attr,
5481 } else if (dof->dof_type == DFT_DIR) {
5482 struct seq_server_site *ss;
5483 struct lu_buf buf = { NULL };
5484 struct lu_buf *lmu = NULL;
5486 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
5488 /* If the parent has default stripeEA, and client
5489 * did not find it before sending create request,
5490 * then MDT will return -EREMOTE, and client will
5491 * retrieve the default stripeEA and re-create the
5494 * Note: if dah_eadata != NULL, it means creating the
5495 * striped directory with specified stripeEA, then it
5496 * should ignore the default stripeEA */
5497 if (hint != NULL && hint->dah_eadata == NULL) {
5498 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
5499 GOTO(out, rc = -EREMOTE);
5501 if (lo->ldo_dir_stripe_offset == LMV_OFFSET_DEFAULT) {
5502 struct lod_default_striping *lds;
5504 lds = lo->ldo_def_striping;
5506 * child and parent should be on the same MDT,
5507 * but if parent has default LMV, and the start
5508 * MDT offset is -1, it's allowed. This check
5509 * is not necessary after 2.12.22 because client
5510 * follows this already, but old client may not.
5512 if (hint->dah_parent &&
5513 dt_object_remote(hint->dah_parent) && lds &&
5514 lds->lds_dir_def_stripe_offset !=
5516 GOTO(out, rc = -EREMOTE);
5517 } else if (lo->ldo_dir_stripe_offset !=
5519 struct lod_device *lod;
5520 struct lu_tgt_desc *mdt = NULL;
5521 bool found_mdt = false;
5523 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5524 lod_foreach_mdt(lod, mdt) {
5525 if (mdt->ltd_index ==
5526 lo->ldo_dir_stripe_offset) {
5532 /* If the MDT indicated by stripe_offset can be
5533 * found, then tell client to resend the create
5534 * request to the correct MDT, otherwise return
5535 * error to client */
5537 GOTO(out, rc = -EREMOTE);
5539 GOTO(out, rc = -EINVAL);
5541 } else if (hint && hint->dah_eadata) {
5543 lmu->lb_buf = (void *)hint->dah_eadata;
5544 lmu->lb_len = hint->dah_eadata_len;
5547 rc = lod_declare_dir_striping_create(env, dt, attr, lmu, dof,
5551 /* failed to create striping or to set initial size, let's reset
5552 * config so that others don't get confused */
5554 lod_striping_free(env, lo);
5559 * Generate component ID for new created component.
5561 * \param[in] lo LOD object
5562 * \param[in] comp_idx index of ldo_comp_entries
5564 * \retval component ID on success
5565 * \retval LCME_ID_INVAL on failure
5567 static __u32 lod_gen_component_id(struct lod_object *lo,
5568 int mirror_id, int comp_idx)
5570 struct lod_layout_component *lod_comp;
5571 __u32 id, start, end;
5574 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
5576 lod_obj_inc_layout_gen(lo);
5577 id = lo->ldo_layout_gen;
5578 if (likely(id <= SEQ_ID_MAX))
5579 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
5581 /* Layout generation wraps, need to check collisions. */
5582 start = id & SEQ_ID_MASK;
5585 for (id = start; id <= end; id++) {
5586 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5587 lod_comp = &lo->ldo_comp_entries[i];
5588 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
5591 /* Found the ununsed ID */
5592 if (i == lo->ldo_comp_cnt)
5593 RETURN(pflr_id(mirror_id, id));
5595 if (end == LCME_ID_MAX) {
5597 end = min(lo->ldo_layout_gen & LCME_ID_MASK,
5598 (__u32)(LCME_ID_MAX - 1));
5602 RETURN(LCME_ID_INVAL);
5606 * Creation of a striped regular object.
5608 * The function is called to create the stripe objects for a regular
5609 * striped file. This can happen at the initial object creation or
5610 * when the caller asks LOD to do so using ->do_xattr_set() method
5611 * (so called late striping). Notice all the information are already
5612 * prepared in the form of the list of objects (ldo_stripe field).
5613 * This is done during declare phase.
5615 * \param[in] env execution environment
5616 * \param[in] dt object
5617 * \param[in] attr attributes the stripes will be created with
5618 * \param[in] dof format of stripes (see OSD API description)
5619 * \param[in] th transaction handle
5621 * \retval 0 on success
5622 * \retval negative if failed
5624 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
5625 struct lu_attr *attr, struct dt_object_format *dof,
5628 struct lod_layout_component *lod_comp;
5629 struct lod_object *lo = lod_dt_obj(dt);
5634 LASSERT((lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL) ||
5635 lo->ldo_is_foreign);
5637 mirror_id = 0; /* non-flr file's mirror_id is 0 */
5638 if (lo->ldo_mirror_count > 1) {
5639 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5640 lod_comp = &lo->ldo_comp_entries[i];
5641 if (lod_comp->llc_id != LCME_ID_INVAL &&
5642 mirror_id_of(lod_comp->llc_id) > mirror_id)
5643 mirror_id = mirror_id_of(lod_comp->llc_id);
5647 /* create all underlying objects */
5648 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5649 lod_comp = &lo->ldo_comp_entries[i];
5651 if (lod_comp->llc_id == LCME_ID_INVAL) {
5652 /* only the component of FLR layout with more than 1
5653 * mirror has mirror ID in its component ID.
5655 if (lod_comp->llc_extent.e_start == 0 &&
5656 lo->ldo_mirror_count > 1)
5659 lod_comp->llc_id = lod_gen_component_id(lo,
5661 if (lod_comp->llc_id == LCME_ID_INVAL)
5662 GOTO(out, rc = -ERANGE);
5665 if (lod_comp_inited(lod_comp))
5668 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
5669 lod_comp_set_init(lod_comp);
5671 if (lov_pattern(lod_comp->llc_pattern) == LOV_PATTERN_MDT)
5672 lod_comp_set_init(lod_comp);
5674 if (lod_comp->llc_stripe == NULL)
5677 LASSERT(lod_comp->llc_stripe_count);
5678 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
5679 struct dt_object *object = lod_comp->llc_stripe[j];
5680 LASSERT(object != NULL);
5681 rc = lod_sub_create(env, object, attr, NULL, dof, th);
5685 lod_comp_set_init(lod_comp);
5688 rc = lod_fill_mirrors(lo);
5692 rc = lod_generate_and_set_lovea(env, lo, th);
5696 lo->ldo_comp_cached = 1;
5700 lod_striping_free(env, lo);
5704 static inline bool lod_obj_is_dom(struct dt_object *dt)
5706 struct lod_object *lo = lod_dt_obj(dt);
5708 if (!dt_object_exists(dt_object_child(dt)))
5711 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
5714 if (!lo->ldo_comp_cnt)
5717 return (lov_pattern(lo->ldo_comp_entries[0].llc_pattern) ==
5722 * Implementation of dt_object_operations::do_create.
5724 * If any of preceeding methods (like ->do_declare_create(),
5725 * ->do_ah_init(), etc) chose to create a striped object,
5726 * then this method will create the master and the stripes.
5728 * \see dt_object_operations::do_create() in the API description for details.
5730 static int lod_create(const struct lu_env *env, struct dt_object *dt,
5731 struct lu_attr *attr, struct dt_allocation_hint *hint,
5732 struct dt_object_format *dof, struct thandle *th)
5737 /* create local object */
5738 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
5742 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
5743 (lod_obj_is_striped(dt) || lod_obj_is_dom(dt)) &&
5744 dof->u.dof_reg.striped != 0) {
5745 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
5746 rc = lod_striped_create(env, dt, attr, dof, th);
5753 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
5754 struct dt_object *dt, struct thandle *th,
5755 int comp_idx, int stripe_idx,
5756 struct lod_obj_stripe_cb_data *data)
5758 if (data->locd_declare)
5759 return lod_sub_declare_destroy(env, dt, th);
5760 else if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
5761 stripe_idx == cfs_fail_val)
5762 return lod_sub_destroy(env, dt, th);
5768 * Implementation of dt_object_operations::do_declare_destroy.
5770 * If the object is a striped directory, then the function declares reference
5771 * removal from the master object (this is an index) to the stripes and declares
5772 * destroy of all the stripes. In all the cases, it declares an intention to
5773 * destroy the object itself.
5775 * \see dt_object_operations::do_declare_destroy() in the API description
5778 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
5781 struct dt_object *next = dt_object_child(dt);
5782 struct lod_object *lo = lod_dt_obj(dt);
5783 struct lod_thread_info *info = lod_env_info(env);
5784 struct dt_object *stripe;
5785 char *stripe_name = info->lti_key;
5791 * load striping information, notice we don't do this when object
5792 * is being initialized as we don't need this information till
5793 * few specific cases like destroy, chown
5795 rc = lod_striping_load(env, lo);
5799 /* declare destroy for all underlying objects */
5800 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5801 rc = next->do_ops->do_index_try(env, next,
5802 &dt_directory_features);
5806 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5807 stripe = lo->ldo_stripe[i];
5811 rc = lod_sub_declare_ref_del(env, next, th);
5815 snprintf(stripe_name, sizeof(info->lti_key),
5817 PFID(lu_object_fid(&stripe->do_lu)), i);
5818 rc = lod_sub_declare_delete(env, next,
5819 (const struct dt_key *)stripe_name, th);
5826 * we declare destroy for the local object
5828 rc = lod_sub_declare_destroy(env, next, th);
5832 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
5833 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
5836 if (!lod_obj_is_striped(dt))
5839 /* declare destroy all striped objects */
5840 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5841 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5842 stripe = lo->ldo_stripe[i];
5846 if (!dt_object_exists(stripe))
5849 rc = lod_sub_declare_ref_del(env, stripe, th);
5853 rc = lod_sub_declare_destroy(env, stripe, th);
5858 struct lod_obj_stripe_cb_data data = { { 0 } };
5860 data.locd_declare = true;
5861 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
5862 rc = lod_obj_for_each_stripe(env, lo, th, &data);
5869 * Implementation of dt_object_operations::do_destroy.
5871 * If the object is a striped directory, then the function removes references
5872 * from the master object (this is an index) to the stripes and destroys all
5873 * the stripes. In all the cases, the function destroys the object itself.
5875 * \see dt_object_operations::do_destroy() in the API description for details.
5877 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
5880 struct dt_object *next = dt_object_child(dt);
5881 struct lod_object *lo = lod_dt_obj(dt);
5882 struct lod_thread_info *info = lod_env_info(env);
5883 char *stripe_name = info->lti_key;
5884 struct dt_object *stripe;
5890 /* destroy sub-stripe of master object */
5891 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5892 rc = next->do_ops->do_index_try(env, next,
5893 &dt_directory_features);
5897 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5898 stripe = lo->ldo_stripe[i];
5902 rc = lod_sub_ref_del(env, next, th);
5906 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
5907 PFID(lu_object_fid(&stripe->do_lu)), i);
5909 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
5910 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
5911 PFID(lu_object_fid(&stripe->do_lu)));
5913 rc = lod_sub_delete(env, next,
5914 (const struct dt_key *)stripe_name, th);
5920 rc = lod_sub_destroy(env, next, th);
5924 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
5925 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
5928 if (!lod_obj_is_striped(dt))
5931 /* destroy all striped objects */
5932 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5933 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5934 stripe = lo->ldo_stripe[i];
5938 if (!dt_object_exists(stripe))
5941 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
5942 i == cfs_fail_val) {
5943 dt_write_lock(env, stripe, DT_TGT_CHILD);
5944 rc = lod_sub_ref_del(env, stripe, th);
5945 dt_write_unlock(env, stripe);
5949 rc = lod_sub_destroy(env, stripe, th);
5955 struct lod_obj_stripe_cb_data data = { { 0 } };
5957 data.locd_declare = false;
5958 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
5959 rc = lod_obj_for_each_stripe(env, lo, th, &data);
5966 * Implementation of dt_object_operations::do_declare_ref_add.
5968 * \see dt_object_operations::do_declare_ref_add() in the API description
5971 static int lod_declare_ref_add(const struct lu_env *env,
5972 struct dt_object *dt, struct thandle *th)
5974 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
5978 * Implementation of dt_object_operations::do_ref_add.
5980 * \see dt_object_operations::do_ref_add() in the API description for details.
5982 static int lod_ref_add(const struct lu_env *env,
5983 struct dt_object *dt, struct thandle *th)
5985 return lod_sub_ref_add(env, dt_object_child(dt), th);
5989 * Implementation of dt_object_operations::do_declare_ref_del.
5991 * \see dt_object_operations::do_declare_ref_del() in the API description
5994 static int lod_declare_ref_del(const struct lu_env *env,
5995 struct dt_object *dt, struct thandle *th)
5997 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
6001 * Implementation of dt_object_operations::do_ref_del
6003 * \see dt_object_operations::do_ref_del() in the API description for details.
6005 static int lod_ref_del(const struct lu_env *env,
6006 struct dt_object *dt, struct thandle *th)
6008 return lod_sub_ref_del(env, dt_object_child(dt), th);
6012 * Implementation of dt_object_operations::do_object_sync.
6014 * \see dt_object_operations::do_object_sync() in the API description
6017 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
6018 __u64 start, __u64 end)
6020 return dt_object_sync(env, dt_object_child(dt), start, end);
6024 * Implementation of dt_object_operations::do_object_unlock.
6026 * Used to release LDLM lock(s).
6028 * \see dt_object_operations::do_object_unlock() in the API description
6031 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
6032 struct ldlm_enqueue_info *einfo,
6033 union ldlm_policy_data *policy)
6035 struct lod_object *lo = lod_dt_obj(dt);
6036 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
6037 int slave_locks_size;
6041 if (slave_locks == NULL)
6044 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
6045 /* Note: for remote lock for single stripe dir, MDT will cancel
6046 * the lock by lockh directly */
6047 LASSERT(!dt_object_remote(dt_object_child(dt)));
6049 /* locks were unlocked in MDT layer */
6050 for (i = 0; i < slave_locks->ha_count; i++)
6051 LASSERT(!lustre_handle_is_used(&slave_locks->ha_handles[i]));
6054 * NB, ha_count may not equal to ldo_dir_stripe_count, because dir
6055 * layout may change, e.g., shrink dir layout after migration.
6057 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6058 if (lo->ldo_stripe[i])
6059 dt_invalidate(env, lo->ldo_stripe[i]);
6062 slave_locks_size = offsetof(typeof(*slave_locks),
6063 ha_handles[slave_locks->ha_count]);
6064 OBD_FREE(slave_locks, slave_locks_size);
6065 einfo->ei_cbdata = NULL;
6071 * Implementation of dt_object_operations::do_object_lock.
6073 * Used to get LDLM lock on the non-striped and striped objects.
6075 * \see dt_object_operations::do_object_lock() in the API description
6078 static int lod_object_lock(const struct lu_env *env,
6079 struct dt_object *dt,
6080 struct lustre_handle *lh,
6081 struct ldlm_enqueue_info *einfo,
6082 union ldlm_policy_data *policy)
6084 struct lod_object *lo = lod_dt_obj(dt);
6085 int slave_locks_size;
6086 struct lustre_handle_array *slave_locks = NULL;
6091 /* remote object lock */
6092 if (!einfo->ei_enq_slave) {
6093 LASSERT(dt_object_remote(dt));
6094 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
6098 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
6101 rc = lod_striping_load(env, lo);
6106 if (lo->ldo_dir_stripe_count <= 1)
6109 slave_locks_size = offsetof(typeof(*slave_locks),
6110 ha_handles[lo->ldo_dir_stripe_count]);
6111 /* Freed in lod_object_unlock */
6112 OBD_ALLOC(slave_locks, slave_locks_size);
6115 slave_locks->ha_count = lo->ldo_dir_stripe_count;
6117 /* striped directory lock */
6118 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6119 struct lustre_handle lockh;
6120 struct ldlm_res_id *res_id;
6121 struct dt_object *stripe;
6123 stripe = lo->ldo_stripe[i];
6127 res_id = &lod_env_info(env)->lti_res_id;
6128 fid_build_reg_res_name(lu_object_fid(&stripe->do_lu), res_id);
6129 einfo->ei_res_id = res_id;
6131 if (dt_object_remote(stripe)) {
6132 set_bit(i, (void *)slave_locks->ha_map);
6133 rc = dt_object_lock(env, stripe, &lockh, einfo, policy);
6135 struct ldlm_namespace *ns = einfo->ei_namespace;
6136 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
6137 ldlm_completion_callback completion = einfo->ei_cb_cp;
6138 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
6140 if (einfo->ei_mode == LCK_PW ||
6141 einfo->ei_mode == LCK_EX)
6142 dlmflags |= LDLM_FL_COS_INCOMPAT;
6144 LASSERT(ns != NULL);
6145 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS,
6146 policy, einfo->ei_mode,
6147 &dlmflags, blocking,
6149 NULL, 0, LVB_T_NONE,
6154 ldlm_lock_decref_and_cancel(
6155 &slave_locks->ha_handles[i],
6157 OBD_FREE(slave_locks, slave_locks_size);
6160 slave_locks->ha_handles[i] = lockh;
6162 einfo->ei_cbdata = slave_locks;
6168 * Implementation of dt_object_operations::do_invalidate.
6170 * \see dt_object_operations::do_invalidate() in the API description for details
6172 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
6174 return dt_invalidate(env, dt_object_child(dt));
6177 static int lod_declare_instantiate_components(const struct lu_env *env,
6178 struct lod_object *lo, struct thandle *th)
6180 struct lod_thread_info *info = lod_env_info(env);
6185 LASSERT(info->lti_count < lo->ldo_comp_cnt);
6187 for (i = 0; i < info->lti_count; i++) {
6188 rc = lod_qos_prep_create(env, lo, NULL, th,
6189 info->lti_comp_idx[i]);
6195 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6196 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
6197 &info->lti_buf, XATTR_NAME_LOV, 0, th);
6204 * Check OSTs for an existing component for further extension
6206 * Checks if OSTs are still healthy and not out of space. Gets free space
6207 * on OSTs (relative to allocation watermark rmb_low) and compares to
6208 * the proposed new_end for this component.
6210 * Decides whether or not to extend a component on its current OSTs.
6212 * \param[in] env execution environment for this thread
6213 * \param[in] lo object we're checking
6214 * \param[in] index index of this component
6215 * \param[in] extension_size extension size for this component
6216 * \param[in] extent layout extent for requested operation
6217 * \param[in] comp_extent extension component extent
6218 * \param[in] write if this is write operation
6220 * \retval true - OK to extend on current OSTs
6221 * \retval false - do not extend on current OSTs
6223 static bool lod_sel_osts_allowed(const struct lu_env *env,
6224 struct lod_object *lo,
6225 int index, __u64 extension_size,
6226 struct lu_extent *extent,
6227 struct lu_extent *comp_extent, int write)
6229 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[index];
6230 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6231 struct obd_statfs *sfs = &lod_env_info(env)->lti_osfs;
6232 __u64 available = 0;
6239 LASSERT(lod_comp->llc_stripe_count != 0);
6242 (extent->e_start == 0 && extent->e_end == OBD_OBJECT_EOF)) {
6243 /* truncate or append */
6244 size = extension_size;
6246 /* In case of write op, check the real write extent,
6247 * it may be larger than the extension_size */
6248 size = roundup(min(extent->e_end, comp_extent->e_end) -
6249 max(extent->e_start, comp_extent->e_start),
6252 /* extension_size is file level, so we must divide by stripe count to
6253 * compare it to available space on a single OST */
6254 size /= lod_comp->llc_stripe_count;
6256 lod_getref(&lod->lod_ost_descs);
6257 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
6258 int index = lod_comp->llc_ost_indices[i];
6259 struct lod_tgt_desc *ost = OST_TGT(lod, index);
6260 struct obd_statfs_info info = { 0 };
6261 int j, repeated = 0;
6265 /* Get the number of times this OST repeats in this component.
6266 * Note: inter-component repeats are not counted as this is
6267 * considered as a rare case: we try to not repeat OST in other
6268 * components if possible. */
6269 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6270 if (index != lod_comp->llc_ost_indices[j])
6273 /* already handled */
6279 if (j < lod_comp->llc_stripe_count)
6282 if (!cfs_bitmap_check(lod->lod_ost_bitmap, index)) {
6283 CDEBUG(D_LAYOUT, "ost %d no longer present\n", index);
6288 rc = dt_statfs_info(env, ost->ltd_tgt, sfs, &info);
6290 CDEBUG(D_LAYOUT, "statfs failed for ost %d, error %d\n",
6296 if (sfs->os_state & OS_STATE_ENOSPC ||
6297 sfs->os_state & OS_STATE_READONLY ||
6298 sfs->os_state & OS_STATE_DEGRADED) {
6299 CDEBUG(D_LAYOUT, "ost %d is not availble for SEL "
6300 "extension, state %u\n", index, sfs->os_state);
6306 available = sfs->os_bavail * sfs->os_bsize;
6307 /* 'available' is relative to the allocation threshold */
6308 available -= (__u64) info.os_reserved_mb_low << 20;
6310 CDEBUG(D_LAYOUT, "ost %d lowwm: %d highwm: %d, "
6311 "%llu %% blocks available, %llu %% blocks free\n",
6312 index, info.os_reserved_mb_low, info.os_reserved_mb_high,
6313 (100ull * sfs->os_bavail) / sfs->os_blocks,
6314 (100ull * sfs->os_bfree) / sfs->os_blocks);
6316 if (size * repeated > available) {
6318 CDEBUG(D_LAYOUT, "low space on ost %d, available %llu "
6319 "< extension size %llu\n", index, available,
6324 lod_putref(lod, &lod->lod_ost_descs);
6330 * Adjust extents after component removal
6332 * When we remove an extension component, we move the start of the next
6333 * component to match the start of the extension component, so no space is left
6336 * \param[in] env execution environment for this thread
6337 * \param[in] lo object
6338 * \param[in] max_comp layout component
6339 * \param[in] index index of this component
6341 * \retval 0 on success
6342 * \retval negative errno on error
6344 static void lod_sel_adjust_extents(const struct lu_env *env,
6345 struct lod_object *lo,
6346 int max_comp, int index)
6348 struct lod_layout_component *lod_comp = NULL;
6349 struct lod_layout_component *next = NULL;
6350 struct lod_layout_component *prev = NULL;
6351 __u64 new_start = 0;
6355 /* Extension space component */
6356 lod_comp = &lo->ldo_comp_entries[index];
6357 next = &lo->ldo_comp_entries[index + 1];
6358 prev = &lo->ldo_comp_entries[index - 1];
6360 LASSERT(lod_comp != NULL && prev != NULL && next != NULL);
6361 LASSERT(lod_comp->llc_flags & LCME_FL_EXTENSION);
6363 /* Previous is being removed */
6364 if (prev && prev->llc_id == LCME_ID_INVAL)
6365 new_start = prev->llc_extent.e_start;
6367 new_start = lod_comp->llc_extent.e_start;
6369 for (i = index + 1; i < max_comp; i++) {
6370 lod_comp = &lo->ldo_comp_entries[i];
6372 start = lod_comp->llc_extent.e_start;
6373 lod_comp->llc_extent.e_start = new_start;
6375 /* We only move zero length extendable components */
6376 if (!(start == lod_comp->llc_extent.e_end))
6379 LASSERT(!(lod_comp->llc_flags & LCME_FL_INIT));
6381 lod_comp->llc_extent.e_end = new_start;
6385 /* Calculate the proposed 'new end' for a component we're extending */
6386 static __u64 lod_extension_new_end(__u64 extension_size, __u64 extent_end,
6387 __u32 stripe_size, __u64 component_end,
6388 __u64 extension_end)
6392 LASSERT(extension_size != 0 && stripe_size != 0);
6394 /* Round up to extension size */
6395 if (extent_end == OBD_OBJECT_EOF) {
6396 new_end = OBD_OBJECT_EOF;
6398 /* Add at least extension_size to the previous component_end,
6399 * covering the req layout extent */
6400 new_end = max(extent_end - component_end, extension_size);
6401 new_end = roundup(new_end, extension_size);
6402 new_end += component_end;
6404 /* Component end must be min stripe size aligned */
6405 if (new_end % stripe_size) {
6406 CDEBUG(D_LAYOUT, "new component end is not aligned "
6407 "by the stripe size %u: [%llu, %llu) ext size "
6408 "%llu new end %llu, aligning\n",
6409 stripe_size, component_end, extent_end,
6410 extension_size, new_end);
6411 new_end = roundup(new_end, stripe_size);
6415 if (new_end < extent_end)
6416 new_end = OBD_OBJECT_EOF;
6419 /* Don't extend past the end of the extension component */
6420 if (new_end > extension_end)
6421 new_end = extension_end;
6426 /* As lod_sel_handler() could be re-entered for the same component several
6427 * times, this is the data for the next call. Fields could be changed to
6428 * component indexes when needed, (e.g. if there is no need to instantiate
6429 * all the previous components up to the current position) to tell the caller
6430 * where to start over from. */
6437 * Process extent updates for a particular layout component
6439 * Handle layout updates for a particular extension space component touched by
6440 * a layout update operation. Core function of self-extending PFL feature.
6442 * In general, this function processes exactly *one* stage of an extension
6443 * operation, modifying the layout accordingly, then returns to the caller.
6444 * The caller is responsible for restarting processing with the new layout,
6445 * which may repeatedly return to this function until the extension updates
6448 * This function does one of a few things to the layout:
6449 * 1. Extends the component before the current extension space component to
6450 * allow it to accomodate the requested operation (if space/policy permit that
6451 * component to continue on its current OSTs)
6453 * 2. If extension of the existing component fails, we do one of two things:
6454 * a. If there is a component after the extension space, we remove the
6455 * extension space component, move the start of the next component down
6456 * accordingly, then notify the caller to restart processing w/the new
6458 * b. If there is no following component, we try repeating the current
6459 * component, creating a new component using the current one as a
6460 * template (keeping its stripe properties but not specific striping),
6461 * and try assigning striping for this component. If there is sufficient
6462 * free space on the OSTs chosen for this component, it is instantiated
6463 * and i/o continues there.
6465 * If there is not sufficient space on the new OSTs, we remove this new
6466 * component & extend the current component.
6468 * Note further that uninited components followed by extension space can be zero
6469 * length meaning that we will try to extend them before initializing them, and
6470 * if that fails, they will be removed without initialization.
6472 * 3. If we extend to/beyond the end of an extension space component, that
6473 * component is exhausted (all of its range has been given to real components),
6474 * so we remove it and restart processing.
6476 * \param[in] env execution environment for this thread
6477 * \param[in,out] lo object to update the layout of
6478 * \param[in] extent layout extent for requested operation, update
6479 * layout to fit this operation
6480 * \param[in] th transaction handle for this operation
6481 * \param[in,out] max_comp the highest comp for the portion of the layout
6482 * we are operating on (For FLR, the chosen
6483 * replica). Updated because we may remove
6485 * \param[in] index index of the extension space component we're
6487 * \param[in] write if this is write op
6488 * \param[in,out] force if the extension is to be forced; set here
6489 to force it on the 2nd call for the same
6492 * \retval 0 on success
6493 * \retval negative errno on error
6495 static int lod_sel_handler(const struct lu_env *env,
6496 struct lod_object *lo,
6497 struct lu_extent *extent,
6498 struct thandle *th, int *max_comp,
6499 int index, int write,
6500 struct sel_data *sd)
6502 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6503 struct lod_thread_info *info = lod_env_info(env);
6504 struct lod_layout_component *lod_comp;
6505 struct lod_layout_component *prev;
6506 struct lod_layout_component *next = NULL;
6507 __u64 extension_size;
6514 /* First component cannot be extension space */
6516 CERROR("%s: "DFID" first component cannot be extension space\n",
6517 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
6521 lod_comp = &lo->ldo_comp_entries[index];
6522 prev = &lo->ldo_comp_entries[index - 1];
6523 if ((index + 1) < *max_comp)
6524 next = &lo->ldo_comp_entries[index + 1];
6526 /* extension size uses the stripe size field as KiB */
6527 extension_size = lod_comp->llc_stripe_size * SEL_UNIT_SIZE;
6529 CDEBUG(D_LAYOUT, "prev start %llu, extension start %llu, extension end"
6530 " %llu, extension size %llu\n", prev->llc_extent.e_start,
6531 lod_comp->llc_extent.e_start, lod_comp->llc_extent.e_end,
6534 /* Two extension space components cannot be adjacent & extension space
6535 * components cannot be init */
6536 if ((prev->llc_flags & LCME_FL_EXTENSION) ||
6537 !(ergo(next, !(next->llc_flags & LCME_FL_EXTENSION))) ||
6538 lod_comp_inited(lod_comp)) {
6539 CERROR("%s: "DFID" invalid extension space components\n",
6540 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
6544 if (!prev->llc_stripe) {
6545 CDEBUG(D_LAYOUT, "Previous component not inited\n");
6546 info->lti_count = 1;
6547 info->lti_comp_idx[0] = index - 1;
6548 rc = lod_declare_instantiate_components(env, lo, th);
6549 /* ENOSPC tells us we can't use this component. If there is
6550 * a next or we are repeating, we either spill over (next) or
6551 * extend the original comp (repeat). Otherwise, return the
6552 * error to the user. */
6553 if (rc == -ENOSPC && (next || sd->sd_repeat))
6559 if (sd->sd_force == 0 && rc == 0)
6560 rc = !lod_sel_osts_allowed(env, lo, index - 1,
6561 extension_size, extent,
6562 &lod_comp->llc_extent, write);
6564 repeated = !!(sd->sd_repeat);
6568 /* Extend previous component */
6570 new_end = lod_extension_new_end(extension_size, extent->e_end,
6571 prev->llc_stripe_size,
6572 prev->llc_extent.e_end,
6573 lod_comp->llc_extent.e_end);
6575 CDEBUG(D_LAYOUT, "new end %llu\n", new_end);
6576 lod_comp->llc_extent.e_start = new_end;
6577 prev->llc_extent.e_end = new_end;
6579 if (prev->llc_extent.e_end == lod_comp->llc_extent.e_end) {
6580 CDEBUG(D_LAYOUT, "Extension component exhausted\n");
6581 lod_comp->llc_id = LCME_ID_INVAL;
6585 /* rc == 1, failed to extend current component */
6588 /* Normal 'spillover' case - Remove the extension
6589 * space component & bring down the start of the next
6591 lod_comp->llc_id = LCME_ID_INVAL;
6593 if (!(prev->llc_flags & LCME_FL_INIT)) {
6594 prev->llc_id = LCME_ID_INVAL;
6597 lod_sel_adjust_extents(env, lo, *max_comp, index);
6598 } else if (lod_comp_inited(prev)) {
6599 /* If there is no next, and the previous component is
6600 * INIT'ed, try repeating the previous component. */
6601 LASSERT(repeated == 0);
6602 rc = lod_layout_repeat_comp(env, lo, index - 1);
6606 /* The previous component is a repeated component.
6607 * Record this so we don't keep trying to repeat it. */
6610 /* If the previous component is not INIT'ed, this may
6611 * be a component we have just instantiated but failed
6612 * to extend. Or even a repeated component we failed
6613 * to prepare a striping for. Do not repeat but instead
6614 * remove the repeated component & force the extention
6615 * of the original one */
6618 prev->llc_id = LCME_ID_INVAL;
6625 rc = lod_layout_del_prep_layout(env, lo, NULL);
6628 LASSERTF(-rc == change,
6629 "number deleted %d != requested %d\n", -rc,
6632 *max_comp = *max_comp + change;
6634 /* lod_del_prep_layout reallocates ldo_comp_entries, so we must
6635 * refresh these pointers before using them */
6636 lod_comp = &lo->ldo_comp_entries[index];
6637 prev = &lo->ldo_comp_entries[index - 1];
6638 CDEBUG(D_LAYOUT, "After extent updates: prev start %llu, current start "
6639 "%llu, current end %llu max_comp %d ldo_comp_cnt %d\n",
6640 prev->llc_extent.e_start, lod_comp->llc_extent.e_start,
6641 lod_comp->llc_extent.e_end, *max_comp, lo->ldo_comp_cnt);
6643 /* Layout changed successfully */
6648 * Declare layout extent updates
6650 * Handles extensions. Identifies extension components touched by current
6651 * operation and passes them to processing function.
6653 * Restarts with updated layouts from the processing function until the current
6654 * operation no longer touches an extension space component.
6656 * \param[in] env execution environment for this thread
6657 * \param[in,out] lo object to update the layout of
6658 * \param[in] extent layout extent for requested operation, update layout to
6659 * fit this operation
6660 * \param[in] th transaction handle for this operation
6661 * \param[in] pick identifies chosen mirror for FLR layouts
6662 * \param[in] write if this is write op
6664 * \retval 1 on layout changed, 0 on no change
6665 * \retval negative errno on error
6667 static int lod_declare_update_extents(const struct lu_env *env,
6668 struct lod_object *lo, struct lu_extent *extent,
6669 struct thandle *th, int pick, int write)
6671 struct lod_thread_info *info = lod_env_info(env);
6672 struct lod_layout_component *lod_comp;
6673 bool layout_changed = false;
6674 struct sel_data sd = { 0 };
6682 /* This makes us work on the components of the chosen mirror */
6683 start_index = lo->ldo_mirrors[pick].lme_start;
6684 max_comp = lo->ldo_mirrors[pick].lme_end + 1;
6685 if (lo->ldo_flr_state == LCM_FL_NONE)
6686 LASSERT(start_index == 0 && max_comp == lo->ldo_comp_cnt);
6688 CDEBUG(D_LAYOUT, "extent->e_start %llu, extent->e_end %llu\n",
6689 extent->e_start, extent->e_end);
6690 for (i = start_index; i < max_comp; i++) {
6691 lod_comp = &lo->ldo_comp_entries[i];
6693 /* We've passed all components of interest */
6694 if (lod_comp->llc_extent.e_start >= extent->e_end)
6697 if (lod_comp->llc_flags & LCME_FL_EXTENSION) {
6698 layout_changed = true;
6699 rc = lod_sel_handler(env, lo, extent, th, &max_comp,
6704 /* Nothing has changed behind the prev one */
6710 /* We may have added or removed components. If so, we must update the
6711 * start & ends of all the mirrors after the current one, and the end
6712 * of the current mirror. */
6713 change = max_comp - 1 - lo->ldo_mirrors[pick].lme_end;
6715 lo->ldo_mirrors[pick].lme_end += change;
6716 for (i = pick + 1; i < lo->ldo_mirror_count; i++) {
6717 lo->ldo_mirrors[i].lme_start += change;
6718 lo->ldo_mirrors[i].lme_end += change;
6724 /* The amount of components has changed, adjust the lti_comp_idx */
6725 rc2 = lod_layout_data_init(info, lo->ldo_comp_cnt);
6727 return rc < 0 ? rc : rc2 < 0 ? rc2 : layout_changed;
6730 /* If striping is already instantiated or INIT'ed DOM? */
6731 static bool lod_is_instantiation_needed(struct lod_layout_component *comp)
6733 return !(((lov_pattern(comp->llc_pattern) == LOV_PATTERN_MDT) &&
6734 lod_comp_inited(comp)) || comp->llc_stripe);
6738 * Declare layout update for a non-FLR layout.
6740 * \param[in] env execution environment for this thread
6741 * \param[in,out] lo object to update the layout of
6742 * \param[in] layout layout intent for requested operation, "update" is
6743 * a process of reacting to this
6744 * \param[in] buf buffer containing lov ea (see comment on usage inline)
6745 * \param[in] th transaction handle for this operation
6747 * \retval 0 on success
6748 * \retval negative errno on error
6750 static int lod_declare_update_plain(const struct lu_env *env,
6751 struct lod_object *lo, struct layout_intent *layout,
6752 const struct lu_buf *buf, struct thandle *th)
6754 struct lod_thread_info *info = lod_env_info(env);
6755 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6756 struct lod_layout_component *lod_comp;
6757 struct lov_comp_md_v1 *comp_v1 = NULL;
6758 bool layout_changed = false;
6759 bool replay = false;
6763 LASSERT(lo->ldo_flr_state == LCM_FL_NONE);
6766 * In case the client is passing lovea, which only happens during
6767 * the replay of layout intent write RPC for now, we may need to
6768 * parse the lovea and apply new layout configuration.
6770 if (buf && buf->lb_len) {
6771 struct lov_user_md_v1 *v1 = buf->lb_buf;
6773 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
6774 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
6775 LOV_MAGIC_COMP_V1)) {
6776 CERROR("%s: the replay buffer of layout extend "
6777 "(magic %#x) does not contain expected "
6778 "composite layout.\n",
6779 lod2obd(d)->obd_name, v1->lmm_magic);
6780 GOTO(out, rc = -EINVAL);
6783 rc = lod_use_defined_striping(env, lo, buf);
6786 lo->ldo_comp_cached = 1;
6788 rc = lod_get_lov_ea(env, lo);
6791 /* old on-disk EA is stored in info->lti_buf */
6792 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
6794 layout_changed = true;
6796 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
6800 /* non replay path */
6801 rc = lod_striping_load(env, lo);
6806 /* Make sure defined layout covers the requested write range. */
6807 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
6808 if (lo->ldo_comp_cnt > 1 &&
6809 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
6810 lod_comp->llc_extent.e_end < layout->li_extent.e_end) {
6811 CDEBUG_LIMIT(replay ? D_ERROR : D_LAYOUT,
6812 "%s: the defined layout [0, %#llx) does not "
6813 "covers the write range "DEXT"\n",
6814 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
6815 PEXT(&layout->li_extent));
6816 GOTO(out, rc = -EINVAL);
6819 CDEBUG(D_LAYOUT, "%s: "DFID": update components "DEXT"\n",
6820 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
6821 PEXT(&layout->li_extent));
6824 rc = lod_declare_update_extents(env, lo, &layout->li_extent,
6825 th, 0, layout->li_opc == LAYOUT_INTENT_WRITE);
6829 layout_changed = true;
6833 * Iterate ld->ldo_comp_entries, find the component whose extent under
6834 * the write range and not instantianted.
6836 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6837 lod_comp = &lo->ldo_comp_entries[i];
6839 if (lod_comp->llc_extent.e_start >= layout->li_extent.e_end)
6843 /* If striping is instantiated or INIT'ed DOM skip */
6844 if (!lod_is_instantiation_needed(lod_comp))
6848 * In replay path, lod_comp is the EA passed by
6849 * client replay buffer, comp_v1 is the pre-recovery
6850 * on-disk EA, we'd sift out those components which
6851 * were init-ed in the on-disk EA.
6853 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
6858 * this component hasn't instantiated in normal path, or during
6859 * replay it needs replay the instantiation.
6862 /* A released component is being extended */
6863 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
6864 GOTO(out, rc = -EINVAL);
6866 LASSERT(info->lti_comp_idx != NULL);
6867 info->lti_comp_idx[info->lti_count++] = i;
6868 layout_changed = true;
6871 if (!layout_changed)
6874 lod_obj_inc_layout_gen(lo);
6875 rc = lod_declare_instantiate_components(env, lo, th);
6879 lod_striping_free(env, lo);
6883 static inline int lod_comp_index(struct lod_object *lo,
6884 struct lod_layout_component *lod_comp)
6886 LASSERT(lod_comp >= lo->ldo_comp_entries &&
6887 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
6889 return lod_comp - lo->ldo_comp_entries;
6893 * Stale other mirrors by writing extent.
6895 static int lod_stale_components(const struct lu_env *env, struct lod_object *lo,
6896 int primary, struct lu_extent *extent,
6899 struct lod_layout_component *pri_comp, *lod_comp;
6900 struct lu_extent pri_extent;
6905 /* The writing extent decides which components in the primary
6906 * are affected... */
6907 CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
6910 lod_foreach_mirror_comp(pri_comp, lo, primary) {
6911 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
6914 CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
6915 lod_comp_index(lo, pri_comp),
6916 PEXT(&pri_comp->llc_extent));
6918 pri_extent.e_start = pri_comp->llc_extent.e_start;
6919 pri_extent.e_end = pri_comp->llc_extent.e_end;
6921 for (i = 0; i < lo->ldo_mirror_count; i++) {
6924 rc = lod_declare_update_extents(env, lo, &pri_extent,
6926 /* if update_extents changed the layout, it may have
6927 * reallocated the component array, so start over to
6928 * avoid using stale pointers */
6934 /* ... and then stale other components that are
6935 * overlapping with primary components */
6936 lod_foreach_mirror_comp(lod_comp, lo, i) {
6937 if (!lu_extent_is_overlapped(
6939 &lod_comp->llc_extent))
6942 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
6943 i, lod_comp_index(lo, lod_comp));
6945 lod_comp->llc_flags |= LCME_FL_STALE;
6946 lo->ldo_mirrors[i].lme_stale = 1;
6955 * check an OST's availability
6956 * \param[in] env execution environment
6957 * \param[in] lo lod object
6958 * \param[in] dt dt object
6959 * \param[in] index mirror index
6961 * \retval negative if failed
6962 * \retval 1 if \a dt is available
6963 * \retval 0 if \a dt is not available
6965 static inline int lod_check_ost_avail(const struct lu_env *env,
6966 struct lod_object *lo,
6967 struct dt_object *dt, int index)
6969 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6970 struct lod_tgt_desc *ost;
6972 int type = LU_SEQ_RANGE_OST;
6975 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &idx, &type);
6977 CERROR("%s: can't locate "DFID":rc = %d\n",
6978 lod2obd(lod)->obd_name, PFID(lu_object_fid(&dt->do_lu)),
6983 ost = OST_TGT(lod, idx);
6984 if (ost->ltd_statfs.os_state &
6985 (OS_STATE_READONLY | OS_STATE_ENOSPC | OS_STATE_ENOINO |
6986 OS_STATE_NOPRECREATE) ||
6987 ost->ltd_active == 0) {
6988 CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail, rc = %d\n",
6989 PFID(lod_object_fid(lo)), index, idx, rc);
6997 * Pick primary mirror for write
6998 * \param[in] env execution environment
6999 * \param[in] lo object
7000 * \param[in] extent write range
7002 static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo,
7003 struct lu_extent *extent)
7005 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7006 unsigned int seq = 0;
7007 struct lod_layout_component *lod_comp;
7009 int picked = -1, second_pick = -1, third_pick = -1;
7012 if (OBD_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
7013 get_random_bytes(&seq, sizeof(seq));
7014 seq %= lo->ldo_mirror_count;
7018 * Pick a mirror as the primary, and check the availability of OSTs.
7020 * This algo can be revised later after knowing the topology of
7023 lod_qos_statfs_update(env, lod, &lod->lod_ost_descs);
7024 for (i = 0; i < lo->ldo_mirror_count; i++) {
7025 bool ost_avail = true;
7026 int index = (i + seq) % lo->ldo_mirror_count;
7028 if (lo->ldo_mirrors[index].lme_stale) {
7029 CDEBUG(D_LAYOUT, DFID": mirror %d stale\n",
7030 PFID(lod_object_fid(lo)), index);
7034 /* 2nd pick is for the primary mirror containing unavail OST */
7035 if (lo->ldo_mirrors[index].lme_primary && second_pick < 0)
7036 second_pick = index;
7038 /* 3rd pick is for non-primary mirror containing unavail OST */
7039 if (second_pick < 0 && third_pick < 0)
7043 * we found a non-primary 1st pick, we'd like to find a
7044 * potential pirmary mirror.
7046 if (picked >= 0 && !lo->ldo_mirrors[index].lme_primary)
7049 /* check the availability of OSTs */
7050 lod_foreach_mirror_comp(lod_comp, lo, index) {
7051 if (!lod_comp_inited(lod_comp) || !lod_comp->llc_stripe)
7054 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
7055 struct dt_object *dt = lod_comp->llc_stripe[j];
7057 rc = lod_check_ost_avail(env, lo, dt, index);
7064 } /* for all dt object in one component */
7067 } /* for all components in a mirror */
7070 * the OSTs where allocated objects locates in the components
7071 * of the mirror are available.
7076 /* this mirror has all OSTs available */
7080 * primary with all OSTs are available, this is the perfect
7083 if (lo->ldo_mirrors[index].lme_primary)
7085 } /* for all mirrors */
7087 /* failed to pick a sound mirror, lower our expectation */
7089 picked = second_pick;
7091 picked = third_pick;
7098 static int lod_prepare_resync_mirror(const struct lu_env *env,
7099 struct lod_object *lo,
7102 struct lod_thread_info *info = lod_env_info(env);
7103 struct lod_layout_component *lod_comp;
7104 bool neg = !!(MIRROR_ID_NEG & mirror_id);
7107 mirror_id &= ~MIRROR_ID_NEG;
7109 for (i = 0; i < lo->ldo_mirror_count; i++) {
7110 if ((!neg && lo->ldo_mirrors[i].lme_id != mirror_id) ||
7111 (neg && lo->ldo_mirrors[i].lme_id == mirror_id))
7114 lod_foreach_mirror_comp(lod_comp, lo, i) {
7115 if (lod_comp_inited(lod_comp))
7118 info->lti_comp_idx[info->lti_count++] =
7119 lod_comp_index(lo, lod_comp);
7127 * figure out the components should be instantiated for resync.
7129 static int lod_prepare_resync(const struct lu_env *env, struct lod_object *lo,
7130 struct lu_extent *extent)
7132 struct lod_thread_info *info = lod_env_info(env);
7133 struct lod_layout_component *lod_comp;
7134 unsigned int need_sync = 0;
7138 DFID": instantiate all stale components in "DEXT"\n",
7139 PFID(lod_object_fid(lo)), PEXT(extent));
7142 * instantiate all components within this extent, even non-stale
7145 for (i = 0; i < lo->ldo_mirror_count; i++) {
7146 if (!lo->ldo_mirrors[i].lme_stale)
7149 lod_foreach_mirror_comp(lod_comp, lo, i) {
7150 if (!lu_extent_is_overlapped(extent,
7151 &lod_comp->llc_extent))
7156 if (lod_comp_inited(lod_comp))
7159 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
7160 i, lod_comp_index(lo, lod_comp));
7161 info->lti_comp_idx[info->lti_count++] =
7162 lod_comp_index(lo, lod_comp);
7166 return need_sync ? 0 : -EALREADY;
7169 static int lod_declare_update_rdonly(const struct lu_env *env,
7170 struct lod_object *lo, struct md_layout_change *mlc,
7173 struct lod_thread_info *info = lod_env_info(env);
7174 struct lu_attr *layout_attr = &info->lti_layout_attr;
7175 struct lod_layout_component *lod_comp;
7176 struct lu_extent extent = { 0 };
7180 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
7181 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
7182 mlc->mlc_opc == MD_LAYOUT_RESYNC);
7183 LASSERT(lo->ldo_mirror_count > 0);
7185 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7186 struct layout_intent *layout = mlc->mlc_intent;
7187 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
7190 extent = layout->li_extent;
7191 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
7192 PFID(lod_object_fid(lo)), PEXT(&extent));
7194 picked = lod_primary_pick(env, lo, &extent);
7198 CDEBUG(D_LAYOUT, DFID": picked mirror id %u as primary\n",
7199 PFID(lod_object_fid(lo)),
7200 lo->ldo_mirrors[picked].lme_id);
7202 /* Update extents of primary before staling */
7203 rc = lod_declare_update_extents(env, lo, &extent, th, picked,
7208 if (layout->li_opc == LAYOUT_INTENT_TRUNC) {
7210 * trunc transfers [0, size) in the intent extent, we'd
7211 * stale components overlapping [size, eof).
7213 extent.e_start = extent.e_end;
7214 extent.e_end = OBD_OBJECT_EOF;
7217 /* stale overlapping components from other mirrors */
7218 rc = lod_stale_components(env, lo, picked, &extent, th);
7222 /* restore truncate intent extent */
7223 if (layout->li_opc == LAYOUT_INTENT_TRUNC)
7224 extent.e_end = extent.e_start;
7226 /* instantiate components for the picked mirror, start from 0 */
7229 lod_foreach_mirror_comp(lod_comp, lo, picked) {
7230 if (!lu_extent_is_overlapped(&extent,
7231 &lod_comp->llc_extent))
7234 if (!lod_is_instantiation_needed(lod_comp))
7237 info->lti_comp_idx[info->lti_count++] =
7238 lod_comp_index(lo, lod_comp);
7241 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
7242 } else { /* MD_LAYOUT_RESYNC */
7246 * could contain multiple non-stale mirrors, so we need to
7247 * prep uninited all components assuming any non-stale mirror
7248 * could be picked as the primary mirror.
7250 if (mlc->mlc_mirror_id == 0) {
7252 for (i = 0; i < lo->ldo_mirror_count; i++) {
7253 if (lo->ldo_mirrors[i].lme_stale)
7256 lod_foreach_mirror_comp(lod_comp, lo, i) {
7257 if (!lod_comp_inited(lod_comp))
7261 lod_comp->llc_extent.e_end)
7263 lod_comp->llc_extent.e_end;
7266 rc = lod_prepare_resync(env, lo, &extent);
7270 /* mirror write, try to init its all components */
7271 rc = lod_prepare_resync_mirror(env, lo,
7272 mlc->mlc_mirror_id);
7277 /* change the file state to SYNC_PENDING */
7278 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
7281 /* Reset the layout version once it's becoming too large.
7282 * This way it can make sure that the layout version is
7283 * monotonously increased in this writing era. */
7284 lod_obj_inc_layout_gen(lo);
7285 if (lo->ldo_layout_gen > (LCME_ID_MAX >> 1)) {
7286 __u32 layout_version;
7288 get_random_bytes(&layout_version, sizeof(layout_version));
7289 lo->ldo_layout_gen = layout_version & 0xffff;
7292 rc = lod_declare_instantiate_components(env, lo, th);
7296 layout_attr->la_valid = LA_LAYOUT_VERSION;
7297 layout_attr->la_layout_version = 0; /* set current version */
7298 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
7299 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
7300 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
7306 lod_striping_free(env, lo);
7310 static int lod_declare_update_write_pending(const struct lu_env *env,
7311 struct lod_object *lo, struct md_layout_change *mlc,
7314 struct lod_thread_info *info = lod_env_info(env);
7315 struct lu_attr *layout_attr = &info->lti_layout_attr;
7316 struct lod_layout_component *lod_comp;
7317 struct lu_extent extent = { 0 };
7323 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
7324 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
7325 mlc->mlc_opc == MD_LAYOUT_RESYNC);
7327 /* look for the primary mirror */
7328 for (i = 0; i < lo->ldo_mirror_count; i++) {
7329 if (lo->ldo_mirrors[i].lme_stale)
7332 LASSERTF(primary < 0, DFID " has multiple primary: %u / %u",
7333 PFID(lod_object_fid(lo)),
7334 lo->ldo_mirrors[i].lme_id,
7335 lo->ldo_mirrors[primary].lme_id);
7340 CERROR(DFID ": doesn't have a primary mirror\n",
7341 PFID(lod_object_fid(lo)));
7342 GOTO(out, rc = -ENODATA);
7345 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
7346 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
7348 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
7350 /* for LAYOUT_WRITE opc, it has to do the following operations:
7351 * 1. stale overlapping componets from stale mirrors;
7352 * 2. instantiate components of the primary mirror;
7353 * 3. transfter layout version to all objects of the primary;
7355 * for LAYOUT_RESYNC opc, it will do:
7356 * 1. instantiate components of all stale mirrors;
7357 * 2. transfer layout version to all objects to close write era. */
7359 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7360 struct layout_intent *layout = mlc->mlc_intent;
7361 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
7363 LASSERT(mlc->mlc_intent != NULL);
7365 extent = mlc->mlc_intent->li_extent;
7367 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
7368 PFID(lod_object_fid(lo)), PEXT(&extent));
7370 /* 1. Update extents of primary before staling */
7371 rc = lod_declare_update_extents(env, lo, &extent, th, primary,
7376 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC) {
7378 * trunc transfers [0, size) in the intent extent, we'd
7379 * stale components overlapping [size, eof).
7381 extent.e_start = extent.e_end;
7382 extent.e_end = OBD_OBJECT_EOF;
7385 /* 2. stale overlapping components */
7386 rc = lod_stale_components(env, lo, primary, &extent, th);
7390 /* 3. find the components which need instantiating.
7391 * instantiate [0, mlc->mlc_intent->e_end) */
7393 /* restore truncate intent extent */
7394 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC)
7395 extent.e_end = extent.e_start;
7398 lod_foreach_mirror_comp(lod_comp, lo, primary) {
7399 if (!lu_extent_is_overlapped(&extent,
7400 &lod_comp->llc_extent))
7403 if (!lod_is_instantiation_needed(lod_comp))
7406 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
7407 primary, lod_comp_index(lo, lod_comp));
7408 info->lti_comp_idx[info->lti_count++] =
7409 lod_comp_index(lo, lod_comp);
7411 } else { /* MD_LAYOUT_RESYNC */
7412 if (mlc->mlc_mirror_id == 0) {
7414 lod_foreach_mirror_comp(lod_comp, lo, primary) {
7415 if (!lod_comp_inited(lod_comp))
7418 extent.e_end = lod_comp->llc_extent.e_end;
7421 rc = lod_prepare_resync(env, lo, &extent);
7425 /* mirror write, try to init its all components */
7426 rc = lod_prepare_resync_mirror(env, lo,
7427 mlc->mlc_mirror_id);
7432 /* change the file state to SYNC_PENDING */
7433 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
7436 rc = lod_declare_instantiate_components(env, lo, th);
7440 /* 3. transfer layout version to OST objects.
7441 * transfer new layout version to OST objects so that stale writes
7442 * can be denied. It also ends an era of writing by setting
7443 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
7444 * send write RPC; only resync RPCs could do it. */
7445 layout_attr->la_valid = LA_LAYOUT_VERSION;
7446 layout_attr->la_layout_version = 0; /* set current version */
7447 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
7448 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
7449 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
7453 lod_obj_inc_layout_gen(lo);
7456 lod_striping_free(env, lo);
7460 static int lod_declare_update_sync_pending(const struct lu_env *env,
7461 struct lod_object *lo, struct md_layout_change *mlc,
7464 struct lod_thread_info *info = lod_env_info(env);
7465 unsigned sync_components = 0;
7466 unsigned resync_components = 0;
7471 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
7472 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
7473 mlc->mlc_opc == MD_LAYOUT_WRITE);
7475 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
7476 PFID(lod_object_fid(lo)), mlc->mlc_opc);
7478 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7479 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
7480 PFID(lod_object_fid(lo)));
7482 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
7483 return lod_declare_update_write_pending(env, lo, mlc, th);
7486 /* MD_LAYOUT_RESYNC_DONE */
7488 for (i = 0; i < lo->ldo_comp_cnt; i++) {
7489 struct lod_layout_component *lod_comp;
7492 lod_comp = &lo->ldo_comp_entries[i];
7494 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
7499 for (j = 0; j < mlc->mlc_resync_count; j++) {
7500 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
7503 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
7504 lod_comp->llc_flags &= ~LCME_FL_STALE;
7505 resync_components++;
7511 for (i = 0; i < mlc->mlc_resync_count; i++) {
7512 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
7515 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
7516 "or already synced\n", PFID(lod_object_fid(lo)),
7517 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
7518 GOTO(out, rc = -EINVAL);
7521 if (!sync_components || (mlc->mlc_resync_count && !resync_components)) {
7522 CDEBUG(D_LAYOUT, DFID": no mirror in sync\n",
7523 PFID(lod_object_fid(lo)));
7525 /* tend to return an error code here to prevent
7526 * the MDT from setting SoM attribute */
7527 GOTO(out, rc = -EINVAL);
7530 CDEBUG(D_LAYOUT, DFID": synced %u resynced %u/%zu components\n",
7531 PFID(lod_object_fid(lo)),
7532 sync_components, resync_components, mlc->mlc_resync_count);
7534 lo->ldo_flr_state = LCM_FL_RDONLY;
7535 lod_obj_inc_layout_gen(lo);
7537 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
7538 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
7539 &info->lti_buf, XATTR_NAME_LOV, 0, th);
7544 lod_striping_free(env, lo);
7548 typedef int (*mlc_handler)(const struct lu_env *env, struct dt_object *dt,
7549 const struct md_layout_change *mlc,
7550 struct thandle *th);
7553 * Attach stripes after target's for migrating directory. NB, we
7554 * only need to declare this, the actual work is done inside
7555 * lod_xattr_set_lmv().
7557 * \param[in] env execution environment
7558 * \param[in] dt target object
7559 * \param[in] mlc layout change data
7560 * \param[in] th transaction handle
7562 * \retval 0 on success
7563 * \retval negative if failed
7565 static int lod_dir_declare_layout_attach(const struct lu_env *env,
7566 struct dt_object *dt,
7567 const struct md_layout_change *mlc,
7570 struct lod_thread_info *info = lod_env_info(env);
7571 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
7572 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
7573 struct lod_object *lo = lod_dt_obj(dt);
7574 struct dt_object *next = dt_object_child(dt);
7575 struct dt_object_format *dof = &info->lti_format;
7576 struct lmv_mds_md_v1 *lmv = mlc->mlc_buf.lb_buf;
7577 struct dt_object **stripes;
7578 __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
7579 struct lu_fid *fid = &info->lti_fid;
7580 struct lod_tgt_desc *tgt;
7581 struct dt_object *dto;
7582 struct dt_device *tgt_dt;
7583 int type = LU_SEQ_RANGE_ANY;
7584 struct dt_insert_rec *rec = &info->lti_dt_rec;
7585 char *stripe_name = info->lti_key;
7586 struct lu_name *sname;
7587 struct linkea_data ldata = { NULL };
7588 struct lu_buf linkea_buf;
7595 if (!lmv_is_sane(lmv))
7598 if (!dt_try_as_dir(env, dt))
7601 dof->dof_type = DFT_DIR;
7604 sizeof(*stripes) * (lo->ldo_dir_stripe_count + stripe_count));
7608 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
7609 stripes[i] = lo->ldo_stripe[i];
7611 rec->rec_type = S_IFDIR;
7613 for (i = 0; i < stripe_count; i++) {
7615 &lmv->lmv_stripe_fids[i]);
7616 if (!fid_is_sane(fid))
7619 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
7623 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
7624 tgt_dt = lod->lod_child;
7626 tgt = LTD_TGT(ltd, idx);
7628 GOTO(out, rc = -ESTALE);
7629 tgt_dt = tgt->ltd_tgt;
7632 dto = dt_locate_at(env, tgt_dt, fid,
7633 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
7636 GOTO(out, rc = PTR_ERR(dto));
7638 stripes[i + lo->ldo_dir_stripe_count] = dto;
7640 if (!dt_try_as_dir(env, dto))
7641 GOTO(out, rc = -ENOTDIR);
7643 rc = lod_sub_declare_ref_add(env, dto, th);
7647 rec->rec_fid = lu_object_fid(&dto->do_lu);
7648 rc = lod_sub_declare_insert(env, dto,
7649 (const struct dt_rec *)rec,
7650 (const struct dt_key *)dot, th);
7654 rc = lod_sub_declare_insert(env, dto,
7655 (const struct dt_rec *)rec,
7656 (const struct dt_key *)dotdot, th);
7660 rc = lod_sub_declare_xattr_set(env, dto, &mlc->mlc_buf,
7661 XATTR_NAME_LMV, 0, th);
7665 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
7666 PFID(lu_object_fid(&dto->do_lu)),
7667 i + lo->ldo_dir_stripe_count);
7669 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
7670 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
7671 sname, lu_object_fid(&dt->do_lu));
7675 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
7676 linkea_buf.lb_len = ldata.ld_leh->leh_len;
7677 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
7678 XATTR_NAME_LINK, 0, th);
7682 rc = lod_sub_declare_insert(env, next,
7683 (const struct dt_rec *)rec,
7684 (const struct dt_key *)stripe_name,
7689 rc = lod_sub_declare_ref_add(env, next, th);
7695 OBD_FREE(lo->ldo_stripe,
7696 sizeof(*stripes) * lo->ldo_dir_stripes_allocated);
7697 lo->ldo_stripe = stripes;
7698 lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count;
7699 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type);
7700 lo->ldo_dir_stripe_count += stripe_count;
7701 lo->ldo_dir_stripes_allocated += stripe_count;
7702 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_MIGRATION;
7706 i = lo->ldo_dir_stripe_count;
7707 while (i < lo->ldo_dir_stripe_count + stripe_count && stripes[i])
7708 dt_object_put(env, stripes[i++]);
7711 sizeof(*stripes) * (stripe_count + lo->ldo_dir_stripe_count));
7715 static int lod_dir_declare_layout_detach(const struct lu_env *env,
7716 struct dt_object *dt,
7717 const struct md_layout_change *unused,
7720 struct lod_thread_info *info = lod_env_info(env);
7721 struct lod_object *lo = lod_dt_obj(dt);
7722 struct dt_object *next = dt_object_child(dt);
7723 char *stripe_name = info->lti_key;
7724 struct dt_object *dto;
7728 if (!dt_try_as_dir(env, dt))
7731 if (!lo->ldo_dir_stripe_count)
7732 return lod_sub_declare_delete(env, next,
7733 (const struct dt_key *)dotdot, th);
7735 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
7736 dto = lo->ldo_stripe[i];
7740 if (!dt_try_as_dir(env, dto))
7743 rc = lod_sub_declare_delete(env, dto,
7744 (const struct dt_key *)dotdot, th);
7748 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
7749 PFID(lu_object_fid(&dto->do_lu)), i);
7751 rc = lod_sub_declare_delete(env, next,
7752 (const struct dt_key *)stripe_name, th);
7756 rc = lod_sub_declare_ref_del(env, next, th);
7764 static int dt_dir_is_empty(const struct lu_env *env,
7765 struct dt_object *obj)
7768 const struct dt_it_ops *iops;
7773 if (!dt_try_as_dir(env, obj))
7776 iops = &obj->do_index_ops->dio_it;
7777 it = iops->init(env, obj, LUDA_64BITHASH);
7779 RETURN(PTR_ERR(it));
7781 rc = iops->get(env, it, (const struct dt_key *)"");
7785 for (rc = 0, i = 0; rc == 0 && i < 3; ++i)
7786 rc = iops->next(env, it);
7792 /* Huh? Index contains no zero key? */
7797 iops->fini(env, it);
7802 static int lod_dir_declare_layout_shrink(const struct lu_env *env,
7803 struct dt_object *dt,
7804 const struct md_layout_change *mlc,
7807 struct lod_thread_info *info = lod_env_info(env);
7808 struct lod_object *lo = lod_dt_obj(dt);
7809 struct dt_object *next = dt_object_child(dt);
7810 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
7811 __u32 final_stripe_count;
7812 char *stripe_name = info->lti_key;
7813 struct lu_buf *lmv_buf = &info->lti_buf;
7814 struct dt_object *dto;
7820 if (!dt_try_as_dir(env, dt))
7823 /* shouldn't be called on plain directory */
7824 LASSERT(lo->ldo_dir_stripe_count);
7826 lmv_buf->lb_buf = &info->lti_lmv.lmv_md_v1;
7827 lmv_buf->lb_len = sizeof(info->lti_lmv.lmv_md_v1);
7829 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
7830 LASSERT(final_stripe_count &&
7831 final_stripe_count < lo->ldo_dir_stripe_count);
7833 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
7834 dto = lo->ldo_stripe[i];
7838 if (i < final_stripe_count) {
7839 if (final_stripe_count == 1)
7842 rc = lod_sub_declare_xattr_set(env, dto, lmv_buf,
7844 LU_XATTR_REPLACE, th);
7851 rc = dt_dir_is_empty(env, dto);
7855 rc = lod_sub_declare_ref_del(env, dto, th);
7859 rc = lod_sub_declare_destroy(env, dto, th);
7863 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
7864 PFID(lu_object_fid(&dto->do_lu)), i);
7866 rc = lod_sub_declare_delete(env, next,
7867 (const struct dt_key *)stripe_name, th);
7871 rc = lod_sub_declare_ref_del(env, next, th);
7876 rc = lod_sub_declare_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
7877 LU_XATTR_REPLACE, th);
7882 * detach all stripes from dir master object, NB, stripes are not destroyed, but
7883 * deleted from it's parent namespace, this function is called in two places:
7884 * 1. mdd_migrate_mdt() detach stripes from source, and attach them to
7886 * 2. mdd_dir_layout_update() detach stripe before turning 1-stripe directory to
7887 * a plain directory.
7889 * \param[in] env execution environment
7890 * \param[in] dt target object
7891 * \param[in] mlc layout change data
7892 * \param[in] th transaction handle
7894 * \retval 0 on success
7895 * \retval negative if failed
7897 static int lod_dir_layout_detach(const struct lu_env *env,
7898 struct dt_object *dt,
7899 const struct md_layout_change *mlc,
7902 struct lod_thread_info *info = lod_env_info(env);
7903 struct lod_object *lo = lod_dt_obj(dt);
7904 struct dt_object *next = dt_object_child(dt);
7905 char *stripe_name = info->lti_key;
7906 struct dt_object *dto;
7912 if (!lo->ldo_dir_stripe_count) {
7913 /* plain directory delete .. */
7914 rc = lod_sub_delete(env, next,
7915 (const struct dt_key *)dotdot, th);
7919 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
7920 dto = lo->ldo_stripe[i];
7924 rc = lod_sub_delete(env, dto,
7925 (const struct dt_key *)dotdot, th);
7929 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
7930 PFID(lu_object_fid(&dto->do_lu)), i);
7932 rc = lod_sub_delete(env, next,
7933 (const struct dt_key *)stripe_name, th);
7937 rc = lod_sub_ref_del(env, next, th);
7942 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
7943 dto = lo->ldo_stripe[i];
7945 dt_object_put(env, dto);
7947 OBD_FREE(lo->ldo_stripe,
7948 sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated);
7949 lo->ldo_stripe = NULL;
7950 lo->ldo_dir_stripes_allocated = 0;
7951 lo->ldo_dir_stripe_count = 0;
7956 static int lod_dir_layout_shrink(const struct lu_env *env,
7957 struct dt_object *dt,
7958 const struct md_layout_change *mlc,
7961 struct lod_thread_info *info = lod_env_info(env);
7962 struct lod_object *lo = lod_dt_obj(dt);
7963 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7964 struct dt_object *next = dt_object_child(dt);
7965 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
7966 __u32 final_stripe_count;
7967 char *stripe_name = info->lti_key;
7968 struct dt_object *dto;
7969 struct lu_buf *lmv_buf = &info->lti_buf;
7970 struct lmv_mds_md_v1 *lmv = &info->lti_lmv.lmv_md_v1;
7972 int type = LU_SEQ_RANGE_ANY;
7978 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
7980 lmv_buf->lb_buf = lmv;
7981 lmv_buf->lb_len = sizeof(*lmv);
7982 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
7983 lmv->lmv_stripe_count = cpu_to_le32(final_stripe_count);
7984 lmv->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type) &
7985 cpu_to_le32(LMV_HASH_TYPE_MASK);
7986 lmv->lmv_layout_version =
7987 cpu_to_le32(lo->ldo_dir_layout_version + 1);
7989 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
7990 dto = lo->ldo_stripe[i];
7994 if (i < final_stripe_count) {
7995 /* if only one stripe left, no need to update
7996 * LMV because this stripe will replace master
7997 * object and act as a plain directory.
7999 if (final_stripe_count == 1)
8003 rc = lod_fld_lookup(env, lod,
8004 lu_object_fid(&dto->do_lu),
8009 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
8010 rc = lod_sub_xattr_set(env, dto, lmv_buf,
8012 LU_XATTR_REPLACE, th);
8019 dt_write_lock(env, dto, DT_TGT_CHILD);
8020 rc = lod_sub_ref_del(env, dto, th);
8021 dt_write_unlock(env, dto);
8025 rc = lod_sub_destroy(env, dto, th);
8029 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8030 PFID(lu_object_fid(&dto->do_lu)), i);
8032 rc = lod_sub_delete(env, next,
8033 (const struct dt_key *)stripe_name, th);
8037 rc = lod_sub_ref_del(env, next, th);
8042 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &mdtidx,
8047 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
8048 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
8049 rc = lod_sub_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
8050 LU_XATTR_REPLACE, th);
8054 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
8055 dto = lo->ldo_stripe[i];
8057 dt_object_put(env, dto);
8059 lo->ldo_dir_stripe_count = final_stripe_count;
8064 static mlc_handler dir_mlc_declare_ops[MD_LAYOUT_MAX] = {
8065 [MD_LAYOUT_ATTACH] = lod_dir_declare_layout_attach,
8066 [MD_LAYOUT_DETACH] = lod_dir_declare_layout_detach,
8067 [MD_LAYOUT_SHRINK] = lod_dir_declare_layout_shrink,
8070 static mlc_handler dir_mlc_ops[MD_LAYOUT_MAX] = {
8071 [MD_LAYOUT_DETACH] = lod_dir_layout_detach,
8072 [MD_LAYOUT_SHRINK] = lod_dir_layout_shrink,
8075 static int lod_declare_layout_change(const struct lu_env *env,
8076 struct dt_object *dt, struct md_layout_change *mlc,
8079 struct lod_thread_info *info = lod_env_info(env);
8080 struct lod_object *lo = lod_dt_obj(dt);
8085 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
8086 LASSERT(dir_mlc_declare_ops[mlc->mlc_opc]);
8087 rc = dir_mlc_declare_ops[mlc->mlc_opc](env, dt, mlc, th);
8091 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
8092 dt_object_remote(dt_object_child(dt)))
8095 rc = lod_striping_load(env, lo);
8099 LASSERT(lo->ldo_comp_cnt > 0);
8101 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
8105 switch (lo->ldo_flr_state) {
8107 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
8111 rc = lod_declare_update_rdonly(env, lo, mlc, th);
8113 case LCM_FL_WRITE_PENDING:
8114 rc = lod_declare_update_write_pending(env, lo, mlc, th);
8116 case LCM_FL_SYNC_PENDING:
8117 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
8128 * Instantiate layout component objects which covers the intent write offset.
8130 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
8131 struct md_layout_change *mlc, struct thandle *th)
8133 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
8134 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
8135 struct lod_object *lo = lod_dt_obj(dt);
8140 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
8141 LASSERT(dir_mlc_ops[mlc->mlc_opc]);
8142 rc = dir_mlc_ops[mlc->mlc_opc](env, dt, mlc, th);
8146 rc = lod_striped_create(env, dt, attr, NULL, th);
8147 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
8148 layout_attr->la_layout_version |= lo->ldo_layout_gen;
8149 rc = lod_attr_set(env, dt, layout_attr, th);
8155 struct dt_object_operations lod_obj_ops = {
8156 .do_read_lock = lod_read_lock,
8157 .do_write_lock = lod_write_lock,
8158 .do_read_unlock = lod_read_unlock,
8159 .do_write_unlock = lod_write_unlock,
8160 .do_write_locked = lod_write_locked,
8161 .do_attr_get = lod_attr_get,
8162 .do_declare_attr_set = lod_declare_attr_set,
8163 .do_attr_set = lod_attr_set,
8164 .do_xattr_get = lod_xattr_get,
8165 .do_declare_xattr_set = lod_declare_xattr_set,
8166 .do_xattr_set = lod_xattr_set,
8167 .do_declare_xattr_del = lod_declare_xattr_del,
8168 .do_xattr_del = lod_xattr_del,
8169 .do_xattr_list = lod_xattr_list,
8170 .do_ah_init = lod_ah_init,
8171 .do_declare_create = lod_declare_create,
8172 .do_create = lod_create,
8173 .do_declare_destroy = lod_declare_destroy,
8174 .do_destroy = lod_destroy,
8175 .do_index_try = lod_index_try,
8176 .do_declare_ref_add = lod_declare_ref_add,
8177 .do_ref_add = lod_ref_add,
8178 .do_declare_ref_del = lod_declare_ref_del,
8179 .do_ref_del = lod_ref_del,
8180 .do_object_sync = lod_object_sync,
8181 .do_object_lock = lod_object_lock,
8182 .do_object_unlock = lod_object_unlock,
8183 .do_invalidate = lod_invalidate,
8184 .do_declare_layout_change = lod_declare_layout_change,
8185 .do_layout_change = lod_layout_change,
8189 * Implementation of dt_body_operations::dbo_read.
8191 * \see dt_body_operations::dbo_read() in the API description for details.
8193 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
8194 struct lu_buf *buf, loff_t *pos)
8196 struct dt_object *next = dt_object_child(dt);
8198 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
8199 S_ISLNK(dt->do_lu.lo_header->loh_attr));
8200 return next->do_body_ops->dbo_read(env, next, buf, pos);
8204 * Implementation of dt_body_operations::dbo_declare_write.
8206 * \see dt_body_operations::dbo_declare_write() in the API description
8209 static ssize_t lod_declare_write(const struct lu_env *env,
8210 struct dt_object *dt,
8211 const struct lu_buf *buf, loff_t pos,
8214 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
8218 * Implementation of dt_body_operations::dbo_write.
8220 * \see dt_body_operations::dbo_write() in the API description for details.
8222 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
8223 const struct lu_buf *buf, loff_t *pos,
8226 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
8227 S_ISLNK(dt->do_lu.lo_header->loh_attr));
8228 return lod_sub_write(env, dt_object_child(dt), buf, pos, th);
8231 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
8232 __u64 start, __u64 end, struct thandle *th)
8234 if (dt_object_remote(dt))
8237 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
8240 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
8241 __u64 start, __u64 end, struct thandle *th)
8243 if (dt_object_remote(dt))
8246 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
8247 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
8251 * different type of files use the same body_ops because object may be created
8252 * in OUT, where there is no chance to set correct body_ops for each type, so
8253 * body_ops themselves will check file type inside, see lod_read/write/punch for
8256 const struct dt_body_operations lod_body_ops = {
8257 .dbo_read = lod_read,
8258 .dbo_declare_write = lod_declare_write,
8259 .dbo_write = lod_write,
8260 .dbo_declare_punch = lod_declare_punch,
8261 .dbo_punch = lod_punch,
8265 * Implementation of lu_object_operations::loo_object_init.
8267 * The function determines the type and the index of the target device using
8268 * sequence of the object's FID. Then passes control down to the
8269 * corresponding device:
8270 * OSD for the local objects, OSP for remote
8272 * \see lu_object_operations::loo_object_init() in the API description
8275 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
8276 const struct lu_object_conf *conf)
8278 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
8279 struct lu_device *cdev = NULL;
8280 struct lu_object *cobj;
8281 struct lod_tgt_descs *ltd = NULL;
8282 struct lod_tgt_desc *tgt;
8284 int type = LU_SEQ_RANGE_ANY;
8288 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
8292 if (type == LU_SEQ_RANGE_MDT &&
8293 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
8294 cdev = &lod->lod_child->dd_lu_dev;
8295 } else if (type == LU_SEQ_RANGE_MDT) {
8296 ltd = &lod->lod_mdt_descs;
8298 } else if (type == LU_SEQ_RANGE_OST) {
8299 ltd = &lod->lod_ost_descs;
8306 if (ltd->ltd_tgts_size > idx &&
8307 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
8308 tgt = LTD_TGT(ltd, idx);
8310 LASSERT(tgt != NULL);
8311 LASSERT(tgt->ltd_tgt != NULL);
8313 cdev = &(tgt->ltd_tgt->dd_lu_dev);
8315 lod_putref(lod, ltd);
8318 if (unlikely(cdev == NULL))
8321 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
8322 if (unlikely(cobj == NULL))
8325 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
8327 lu_object_add(lo, cobj);
8334 * Alloc cached foreign LOV
8336 * \param[in] lo object
8337 * \param[in] size size of foreign LOV
8339 * \retval 0 on success
8340 * \retval negative if failed
8342 int lod_alloc_foreign_lov(struct lod_object *lo, size_t size)
8344 OBD_ALLOC_LARGE(lo->ldo_foreign_lov, size);
8345 if (lo->ldo_foreign_lov == NULL)
8347 lo->ldo_foreign_lov_size = size;
8348 lo->ldo_is_foreign = 1;
8354 * Free cached foreign LOV
8356 * \param[in] lo object
8358 void lod_free_foreign_lov(struct lod_object *lo)
8360 if (lo->ldo_foreign_lov != NULL)
8361 OBD_FREE_LARGE(lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
8362 lo->ldo_foreign_lov = NULL;
8363 lo->ldo_foreign_lov_size = 0;
8364 lo->ldo_is_foreign = 0;
8369 * Free cached foreign LMV
8371 * \param[in] lo object
8373 void lod_free_foreign_lmv(struct lod_object *lo)
8375 if (lo->ldo_foreign_lmv != NULL)
8376 OBD_FREE_LARGE(lo->ldo_foreign_lmv, lo->ldo_foreign_lmv_size);
8377 lo->ldo_foreign_lmv = NULL;
8378 lo->ldo_foreign_lmv_size = 0;
8379 lo->ldo_dir_is_foreign = 0;
8384 * Release resources associated with striping.
8386 * If the object is striped (regular or directory), then release
8387 * the stripe objects references and free the ldo_stripe array.
8389 * \param[in] env execution environment
8390 * \param[in] lo object
8392 void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo)
8394 struct lod_layout_component *lod_comp;
8397 if (unlikely(lo->ldo_is_foreign)) {
8398 lod_free_foreign_lov(lo);
8399 lo->ldo_comp_cached = 0;
8400 } else if (unlikely(lo->ldo_dir_is_foreign)) {
8401 lod_free_foreign_lmv(lo);
8402 lo->ldo_dir_stripe_loaded = 0;
8403 } else if (lo->ldo_stripe != NULL) {
8404 LASSERT(lo->ldo_comp_entries == NULL);
8405 LASSERT(lo->ldo_dir_stripes_allocated > 0);
8407 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8408 if (lo->ldo_stripe[i])
8409 dt_object_put(env, lo->ldo_stripe[i]);
8412 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
8413 OBD_FREE(lo->ldo_stripe, j);
8414 lo->ldo_stripe = NULL;
8415 lo->ldo_dir_stripes_allocated = 0;
8416 lo->ldo_dir_stripe_loaded = 0;
8417 lo->ldo_dir_stripe_count = 0;
8418 } else if (lo->ldo_comp_entries != NULL) {
8419 for (i = 0; i < lo->ldo_comp_cnt; i++) {
8420 /* free lod_layout_component::llc_stripe array */
8421 lod_comp = &lo->ldo_comp_entries[i];
8423 if (lod_comp->llc_stripe == NULL)
8425 LASSERT(lod_comp->llc_stripes_allocated != 0);
8426 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
8427 if (lod_comp->llc_stripe[j] != NULL)
8429 &lod_comp->llc_stripe[j]->do_lu);
8431 OBD_FREE(lod_comp->llc_stripe,
8432 sizeof(struct dt_object *) *
8433 lod_comp->llc_stripes_allocated);
8434 lod_comp->llc_stripe = NULL;
8435 OBD_FREE(lod_comp->llc_ost_indices,
8437 lod_comp->llc_stripes_allocated);
8438 lod_comp->llc_ost_indices = NULL;
8439 lod_comp->llc_stripes_allocated = 0;
8441 lod_free_comp_entries(lo);
8442 lo->ldo_comp_cached = 0;
8446 void lod_striping_free(const struct lu_env *env, struct lod_object *lo)
8448 mutex_lock(&lo->ldo_layout_mutex);
8449 lod_striping_free_nolock(env, lo);
8450 mutex_unlock(&lo->ldo_layout_mutex);
8454 * Implementation of lu_object_operations::loo_object_free.
8456 * \see lu_object_operations::loo_object_free() in the API description
8459 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
8461 struct lod_object *lo = lu2lod_obj(o);
8463 /* release all underlying object pinned */
8464 lod_striping_free(env, lo);
8466 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
8470 * Implementation of lu_object_operations::loo_object_release.
8472 * \see lu_object_operations::loo_object_release() in the API description
8475 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
8477 /* XXX: shouldn't we release everything here in case if object
8478 * creation failed before? */
8482 * Implementation of lu_object_operations::loo_object_print.
8484 * \see lu_object_operations::loo_object_print() in the API description
8487 static int lod_object_print(const struct lu_env *env, void *cookie,
8488 lu_printer_t p, const struct lu_object *l)
8490 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
8492 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
8495 struct lu_object_operations lod_lu_obj_ops = {
8496 .loo_object_init = lod_object_init,
8497 .loo_object_free = lod_object_free,
8498 .loo_object_release = lod_object_release,
8499 .loo_object_print = lod_object_print,