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);
387 next = lo->ldo_stripe[0];
388 LASSERT(next != NULL);
389 LASSERT(next->do_index_ops != NULL);
391 it_next = next->do_index_ops->dio_it.init(env, next, attr);
395 /* currently we do not use more than one iterator per thread
396 * so we store it in thread info. if at some point we need
397 * more active iterators in a single thread, we can allocate
399 LASSERT(it->lit_obj == NULL);
401 it->lit_stripe_index = 0;
403 it->lit_it = it_next;
406 return (struct dt_it *)it;
409 #define LOD_CHECK_STRIPED_IT(env, it, lo) \
411 LASSERT((it)->lit_obj != NULL); \
412 LASSERT((it)->lit_it != NULL); \
413 LASSERT((lo)->ldo_dir_stripe_count > 0); \
414 LASSERT((it)->lit_stripe_index < (lo)->ldo_dir_stripe_count); \
418 * Implementation of dt_it_ops::fini.
420 * Used with striped objects.
422 * \see dt_it_ops::fini() in the API description for details.
424 static void lod_striped_it_fini(const struct lu_env *env, struct dt_it *di)
426 struct lod_it *it = (struct lod_it *)di;
427 struct lod_object *lo = lod_dt_obj(it->lit_obj);
428 struct dt_object *next;
430 /* If lit_it == NULL, then it means the sub_it has been finished,
431 * which only happens in failure cases, see lod_striped_it_next() */
432 if (it->lit_it != NULL) {
433 LOD_CHECK_STRIPED_IT(env, it, lo);
435 next = lo->ldo_stripe[it->lit_stripe_index];
436 LASSERT(next != NULL);
437 LASSERT(next->do_index_ops != NULL);
439 next->do_index_ops->dio_it.fini(env, it->lit_it);
442 /* the iterator not in use any more */
445 it->lit_stripe_index = 0;
449 * Implementation of dt_it_ops::get.
451 * Right now it's not used widely, only to reset the iterator to the
452 * initial position. It should be possible to implement a full version
453 * which chooses a correct stripe to be able to position with any key.
455 * \see dt_it_ops::get() in the API description for details.
457 static int lod_striped_it_get(const struct lu_env *env, struct dt_it *di,
458 const struct dt_key *key)
460 const struct lod_it *it = (const struct lod_it *)di;
461 struct lod_object *lo = lod_dt_obj(it->lit_obj);
462 struct dt_object *next;
465 LOD_CHECK_STRIPED_IT(env, it, lo);
467 next = lo->ldo_stripe[it->lit_stripe_index];
468 LASSERT(next != NULL);
469 LASSERT(next->do_index_ops != NULL);
471 return next->do_index_ops->dio_it.get(env, it->lit_it, key);
475 * Implementation of dt_it_ops::put.
477 * Used with striped objects.
479 * \see dt_it_ops::put() in the API description for details.
481 static void lod_striped_it_put(const struct lu_env *env, struct dt_it *di)
483 struct lod_it *it = (struct lod_it *)di;
484 struct lod_object *lo = lod_dt_obj(it->lit_obj);
485 struct dt_object *next;
487 LOD_CHECK_STRIPED_IT(env, it, lo);
489 next = lo->ldo_stripe[it->lit_stripe_index];
490 LASSERT(next != NULL);
491 LASSERT(next->do_index_ops != NULL);
493 return next->do_index_ops->dio_it.put(env, it->lit_it);
497 * Implementation of dt_it_ops::next.
499 * Used with striped objects. When the end of the current stripe is
500 * reached, the method takes the next stripe's iterator.
502 * \see dt_it_ops::next() in the API description for details.
504 static int lod_striped_it_next(const struct lu_env *env, struct dt_it *di)
506 struct lod_it *it = (struct lod_it *)di;
507 struct lod_object *lo = lod_dt_obj(it->lit_obj);
508 struct dt_object *next;
509 struct dt_it *it_next;
513 LOD_CHECK_STRIPED_IT(env, it, lo);
515 next = lo->ldo_stripe[it->lit_stripe_index];
516 LASSERT(next != NULL);
517 LASSERT(next->do_index_ops != NULL);
519 rc = next->do_index_ops->dio_it.next(env, it->lit_it);
523 if (rc == 0 && it->lit_stripe_index == 0)
526 if (rc == 0 && it->lit_stripe_index > 0) {
527 struct lu_dirent *ent;
529 ent = (struct lu_dirent *)lod_env_info(env)->lti_key;
531 rc = next->do_index_ops->dio_it.rec(env, it->lit_it,
532 (struct dt_rec *)ent,
537 /* skip . and .. for slave stripe */
538 if ((strncmp(ent->lde_name, ".",
539 le16_to_cpu(ent->lde_namelen)) == 0 &&
540 le16_to_cpu(ent->lde_namelen) == 1) ||
541 (strncmp(ent->lde_name, "..",
542 le16_to_cpu(ent->lde_namelen)) == 0 &&
543 le16_to_cpu(ent->lde_namelen) == 2))
549 /* go to next stripe */
550 if (it->lit_stripe_index + 1 >= lo->ldo_dir_stripe_count)
553 it->lit_stripe_index++;
555 next->do_index_ops->dio_it.put(env, it->lit_it);
556 next->do_index_ops->dio_it.fini(env, it->lit_it);
559 next = lo->ldo_stripe[it->lit_stripe_index];
560 LASSERT(next != NULL);
561 rc = next->do_ops->do_index_try(env, next, &dt_directory_features);
565 LASSERT(next->do_index_ops != NULL);
567 it_next = next->do_index_ops->dio_it.init(env, next, it->lit_attr);
568 if (!IS_ERR(it_next)) {
569 it->lit_it = it_next;
572 rc = PTR_ERR(it_next);
579 * Implementation of dt_it_ops::key.
581 * Used with striped objects.
583 * \see dt_it_ops::key() in the API description for details.
585 static struct dt_key *lod_striped_it_key(const struct lu_env *env,
586 const struct dt_it *di)
588 const struct lod_it *it = (const struct lod_it *)di;
589 struct lod_object *lo = lod_dt_obj(it->lit_obj);
590 struct dt_object *next;
592 LOD_CHECK_STRIPED_IT(env, it, lo);
594 next = lo->ldo_stripe[it->lit_stripe_index];
595 LASSERT(next != NULL);
596 LASSERT(next->do_index_ops != NULL);
598 return next->do_index_ops->dio_it.key(env, it->lit_it);
602 * Implementation of dt_it_ops::key_size.
604 * Used with striped objects.
606 * \see dt_it_ops::size() in the API description for details.
608 static int lod_striped_it_key_size(const struct lu_env *env,
609 const struct dt_it *di)
611 struct lod_it *it = (struct lod_it *)di;
612 struct lod_object *lo = lod_dt_obj(it->lit_obj);
613 struct dt_object *next;
615 LOD_CHECK_STRIPED_IT(env, it, lo);
617 next = lo->ldo_stripe[it->lit_stripe_index];
618 LASSERT(next != NULL);
619 LASSERT(next->do_index_ops != NULL);
621 return next->do_index_ops->dio_it.key_size(env, it->lit_it);
625 * Implementation of dt_it_ops::rec.
627 * Used with striped objects.
629 * \see dt_it_ops::rec() in the API description for details.
631 static int lod_striped_it_rec(const struct lu_env *env, const struct dt_it *di,
632 struct dt_rec *rec, __u32 attr)
634 const struct lod_it *it = (const struct lod_it *)di;
635 struct lod_object *lo = lod_dt_obj(it->lit_obj);
636 struct dt_object *next;
638 LOD_CHECK_STRIPED_IT(env, it, lo);
640 next = lo->ldo_stripe[it->lit_stripe_index];
641 LASSERT(next != NULL);
642 LASSERT(next->do_index_ops != NULL);
644 return next->do_index_ops->dio_it.rec(env, it->lit_it, rec, attr);
648 * Implementation of dt_it_ops::rec_size.
650 * Used with striped objects.
652 * \see dt_it_ops::rec_size() in the API description for details.
654 static int lod_striped_it_rec_size(const struct lu_env *env,
655 const struct dt_it *di, __u32 attr)
657 struct lod_it *it = (struct lod_it *)di;
658 struct lod_object *lo = lod_dt_obj(it->lit_obj);
659 struct dt_object *next;
661 LOD_CHECK_STRIPED_IT(env, it, lo);
663 next = lo->ldo_stripe[it->lit_stripe_index];
664 LASSERT(next != NULL);
665 LASSERT(next->do_index_ops != NULL);
667 return next->do_index_ops->dio_it.rec_size(env, it->lit_it, attr);
671 * Implementation of dt_it_ops::store.
673 * Used with striped objects.
675 * \see dt_it_ops::store() in the API description for details.
677 static __u64 lod_striped_it_store(const struct lu_env *env,
678 const struct dt_it *di)
680 const struct lod_it *it = (const struct lod_it *)di;
681 struct lod_object *lo = lod_dt_obj(it->lit_obj);
682 struct dt_object *next;
684 LOD_CHECK_STRIPED_IT(env, it, lo);
686 next = lo->ldo_stripe[it->lit_stripe_index];
687 LASSERT(next != NULL);
688 LASSERT(next->do_index_ops != NULL);
690 return next->do_index_ops->dio_it.store(env, it->lit_it);
694 * Implementation of dt_it_ops::load.
696 * Used with striped objects.
698 * \see dt_it_ops::load() in the API description for details.
700 static int lod_striped_it_load(const struct lu_env *env,
701 const struct dt_it *di, __u64 hash)
703 const struct lod_it *it = (const struct lod_it *)di;
704 struct lod_object *lo = lod_dt_obj(it->lit_obj);
705 struct dt_object *next;
707 LOD_CHECK_STRIPED_IT(env, it, lo);
709 next = lo->ldo_stripe[it->lit_stripe_index];
710 LASSERT(next != NULL);
711 LASSERT(next->do_index_ops != NULL);
713 return next->do_index_ops->dio_it.load(env, it->lit_it, hash);
716 static struct dt_index_operations lod_striped_index_ops = {
717 .dio_lookup = lod_lookup,
718 .dio_declare_insert = lod_declare_insert,
719 .dio_insert = lod_insert,
720 .dio_declare_delete = lod_declare_delete,
721 .dio_delete = lod_delete,
723 .init = lod_striped_it_init,
724 .fini = lod_striped_it_fini,
725 .get = lod_striped_it_get,
726 .put = lod_striped_it_put,
727 .next = lod_striped_it_next,
728 .key = lod_striped_it_key,
729 .key_size = lod_striped_it_key_size,
730 .rec = lod_striped_it_rec,
731 .rec_size = lod_striped_it_rec_size,
732 .store = lod_striped_it_store,
733 .load = lod_striped_it_load,
738 * Append the FID for each shard of the striped directory after the
739 * given LMV EA header.
741 * To simplify striped directory and the consistency verification,
742 * we only store the LMV EA header on disk, for both master object
743 * and slave objects. When someone wants to know the whole LMV EA,
744 * such as client readdir(), we can build the entrie LMV EA on the
745 * MDT side (in RAM) via iterating the sub-directory entries that
746 * are contained in the master object of the stripe directory.
748 * For the master object of the striped directroy, the valid name
749 * for each shard is composed of the ${shard_FID}:${shard_idx}.
751 * There may be holes in the LMV EA if some shards' name entries
752 * are corrupted or lost.
754 * \param[in] env pointer to the thread context
755 * \param[in] lo pointer to the master object of the striped directory
756 * \param[in] buf pointer to the lu_buf which will hold the LMV EA
757 * \param[in] resize whether re-allocate the buffer if it is not big enough
759 * \retval positive size of the LMV EA
760 * \retval 0 for nothing to be loaded
761 * \retval negative error number on failure
763 int lod_load_lmv_shards(const struct lu_env *env, struct lod_object *lo,
764 struct lu_buf *buf, bool resize)
766 struct lu_dirent *ent =
767 (struct lu_dirent *)lod_env_info(env)->lti_key;
768 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
769 struct dt_object *obj = dt_object_child(&lo->ldo_obj);
770 struct lmv_mds_md_v1 *lmv1 = buf->lb_buf;
772 const struct dt_it_ops *iops;
774 __u32 magic = le32_to_cpu(lmv1->lmv_magic);
779 if (magic != LMV_MAGIC_V1)
782 stripes = le32_to_cpu(lmv1->lmv_stripe_count);
786 rc = lmv_mds_md_size(stripes, magic);
790 if (buf->lb_len < lmv1_size) {
799 lu_buf_alloc(buf, lmv1_size);
804 memcpy(buf->lb_buf, tbuf.lb_buf, tbuf.lb_len);
807 if (unlikely(!dt_try_as_dir(env, obj)))
810 memset(&lmv1->lmv_stripe_fids[0], 0, stripes * sizeof(struct lu_fid));
811 iops = &obj->do_index_ops->dio_it;
812 it = iops->init(env, obj, LUDA_64BITHASH);
816 rc = iops->load(env, it, 0);
818 rc = iops->next(env, it);
823 char name[FID_LEN + 2] = "";
828 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
834 fid_le_to_cpu(&fid, &ent->lde_fid);
835 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
836 if (ent->lde_name[0] == '.') {
837 if (ent->lde_namelen == 1)
840 if (ent->lde_namelen == 2 && ent->lde_name[1] == '.')
844 len = snprintf(name, sizeof(name),
845 DFID":", PFID(&ent->lde_fid));
846 /* The ent->lde_name is composed of ${FID}:${index} */
847 if (ent->lde_namelen < len + 1 ||
848 memcmp(ent->lde_name, name, len) != 0) {
849 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
850 "%s: invalid shard name %.*s with the FID "DFID
851 " for the striped directory "DFID", %s\n",
852 lod2obd(lod)->obd_name, ent->lde_namelen,
853 ent->lde_name, PFID(&fid),
854 PFID(lu_object_fid(&obj->do_lu)),
855 lod->lod_lmv_failout ? "failout" : "skip");
857 if (lod->lod_lmv_failout)
865 if (ent->lde_name[len] < '0' ||
866 ent->lde_name[len] > '9') {
867 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
868 "%s: invalid shard name %.*s with the "
869 "FID "DFID" for the striped directory "
871 lod2obd(lod)->obd_name, ent->lde_namelen,
872 ent->lde_name, PFID(&fid),
873 PFID(lu_object_fid(&obj->do_lu)),
874 lod->lod_lmv_failout ?
877 if (lod->lod_lmv_failout)
883 index = index * 10 + ent->lde_name[len++] - '0';
884 } while (len < ent->lde_namelen);
886 if (len == ent->lde_namelen) {
887 /* Out of LMV EA range. */
888 if (index >= stripes) {
889 CERROR("%s: the shard %.*s for the striped "
890 "directory "DFID" is out of the known "
891 "LMV EA range [0 - %u], failout\n",
892 lod2obd(lod)->obd_name, ent->lde_namelen,
894 PFID(lu_object_fid(&obj->do_lu)),
900 /* The slot has been occupied. */
901 if (!fid_is_zero(&lmv1->lmv_stripe_fids[index])) {
905 &lmv1->lmv_stripe_fids[index]);
906 CERROR("%s: both the shard "DFID" and "DFID
907 " for the striped directory "DFID
908 " claim the same LMV EA slot at the "
909 "index %d, failout\n",
910 lod2obd(lod)->obd_name,
911 PFID(&fid0), PFID(&fid),
912 PFID(lu_object_fid(&obj->do_lu)), index);
917 /* stored as LE mode */
918 lmv1->lmv_stripe_fids[index] = ent->lde_fid;
921 rc = iops->next(env, it);
928 RETURN(rc > 0 ? lmv_mds_md_size(stripes, magic) : rc);
932 * Implementation of dt_object_operations::do_index_try.
934 * \see dt_object_operations::do_index_try() in the API description for details.
936 static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
937 const struct dt_index_features *feat)
939 struct lod_object *lo = lod_dt_obj(dt);
940 struct dt_object *next = dt_object_child(dt);
944 LASSERT(next->do_ops);
945 LASSERT(next->do_ops->do_index_try);
947 rc = lod_striping_load(env, lo);
951 rc = next->do_ops->do_index_try(env, next, feat);
955 if (lo->ldo_dir_stripe_count > 0) {
958 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
959 if (dt_object_exists(lo->ldo_stripe[i]) == 0)
961 rc = lo->ldo_stripe[i]->do_ops->do_index_try(env,
962 lo->ldo_stripe[i], feat);
966 dt->do_index_ops = &lod_striped_index_ops;
968 dt->do_index_ops = &lod_index_ops;
975 * Implementation of dt_object_operations::do_read_lock.
977 * \see dt_object_operations::do_read_lock() in the API description for details.
979 static void lod_read_lock(const struct lu_env *env, struct dt_object *dt,
982 dt_read_lock(env, dt_object_child(dt), role);
986 * Implementation of dt_object_operations::do_write_lock.
988 * \see dt_object_operations::do_write_lock() in the API description for
991 static void lod_write_lock(const struct lu_env *env, struct dt_object *dt,
994 dt_write_lock(env, dt_object_child(dt), role);
998 * Implementation of dt_object_operations::do_read_unlock.
1000 * \see dt_object_operations::do_read_unlock() in the API description for
1003 static void lod_read_unlock(const struct lu_env *env, struct dt_object *dt)
1005 dt_read_unlock(env, dt_object_child(dt));
1009 * Implementation of dt_object_operations::do_write_unlock.
1011 * \see dt_object_operations::do_write_unlock() in the API description for
1014 static void lod_write_unlock(const struct lu_env *env, struct dt_object *dt)
1016 dt_write_unlock(env, dt_object_child(dt));
1020 * Implementation of dt_object_operations::do_write_locked.
1022 * \see dt_object_operations::do_write_locked() in the API description for
1025 static int lod_write_locked(const struct lu_env *env, struct dt_object *dt)
1027 return dt_write_locked(env, dt_object_child(dt));
1031 * Implementation of dt_object_operations::do_attr_get.
1033 * \see dt_object_operations::do_attr_get() in the API description for details.
1035 static int lod_attr_get(const struct lu_env *env,
1036 struct dt_object *dt,
1037 struct lu_attr *attr)
1039 /* Note: for striped directory, client will merge attributes
1040 * from all of the sub-stripes see lmv_merge_attr(), and there
1041 * no MDD logic depend on directory nlink/size/time, so we can
1042 * always use master inode nlink and size for now. */
1043 return dt_attr_get(env, dt_object_child(dt), attr);
1046 static inline void lod_adjust_stripe_info(struct lod_layout_component *comp,
1047 struct lov_desc *desc)
1049 if (comp->llc_pattern != LOV_PATTERN_MDT) {
1050 if (!comp->llc_stripe_count)
1051 comp->llc_stripe_count =
1052 desc->ld_default_stripe_count;
1054 if (comp->llc_stripe_size <= 0)
1055 comp->llc_stripe_size = desc->ld_default_stripe_size;
1058 int lod_obj_for_each_stripe(const struct lu_env *env, struct lod_object *lo,
1060 struct lod_obj_stripe_cb_data *data)
1062 struct lod_layout_component *lod_comp;
1066 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1067 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1068 lod_comp = &lo->ldo_comp_entries[i];
1070 if (lod_comp->llc_stripe == NULL)
1073 /* has stripe but not inited yet, this component has been
1074 * declared to be created, but hasn't created yet.
1076 if (!lod_comp_inited(lod_comp))
1079 if (data->locd_comp_skip_cb &&
1080 data->locd_comp_skip_cb(env, lo, i, data))
1083 if (data->locd_comp_cb) {
1084 rc = data->locd_comp_cb(env, lo, i, data);
1089 /* could used just to do sth about component, not each
1092 if (!data->locd_stripe_cb)
1095 LASSERT(lod_comp->llc_stripe_count > 0);
1096 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
1097 struct dt_object *dt = lod_comp->llc_stripe[j];
1101 rc = data->locd_stripe_cb(env, lo, dt, th, i, j, data);
1109 static bool lod_obj_attr_set_comp_skip_cb(const struct lu_env *env,
1110 struct lod_object *lo, int comp_idx,
1111 struct lod_obj_stripe_cb_data *data)
1113 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[comp_idx];
1114 bool skipped = false;
1116 if (!(data->locd_attr->la_valid & LA_LAYOUT_VERSION))
1119 switch (lo->ldo_flr_state) {
1120 case LCM_FL_WRITE_PENDING: {
1123 /* skip stale components */
1124 if (lod_comp->llc_flags & LCME_FL_STALE) {
1129 /* skip valid and overlapping components, therefore any
1130 * attempts to write overlapped components will never succeed
1131 * because client will get EINPROGRESS. */
1132 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1136 if (lo->ldo_comp_entries[i].llc_flags & LCME_FL_STALE)
1139 if (lu_extent_is_overlapped(&lod_comp->llc_extent,
1140 &lo->ldo_comp_entries[i].llc_extent)) {
1148 LASSERTF(0, "impossible: %d\n", lo->ldo_flr_state);
1149 case LCM_FL_SYNC_PENDING:
1153 CDEBUG(D_LAYOUT, DFID": %s to set component %x to version: %u\n",
1154 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
1155 skipped ? "skipped" : "chose", lod_comp->llc_id,
1156 data->locd_attr->la_layout_version);
1162 lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
1163 struct dt_object *dt, struct thandle *th,
1164 int comp_idx, int stripe_idx,
1165 struct lod_obj_stripe_cb_data *data)
1167 if (data->locd_declare)
1168 return lod_sub_declare_attr_set(env, dt, data->locd_attr, th);
1170 if (data->locd_attr->la_valid & LA_LAYOUT_VERSION) {
1171 CDEBUG(D_LAYOUT, DFID": set layout version: %u, comp_idx: %d\n",
1172 PFID(lu_object_fid(&dt->do_lu)),
1173 data->locd_attr->la_layout_version, comp_idx);
1176 return lod_sub_attr_set(env, dt, data->locd_attr, th);
1180 * Implementation of dt_object_operations::do_declare_attr_set.
1182 * If the object is striped, then apply the changes to all the stripes.
1184 * \see dt_object_operations::do_declare_attr_set() in the API description
1187 static int lod_declare_attr_set(const struct lu_env *env,
1188 struct dt_object *dt,
1189 const struct lu_attr *attr,
1192 struct dt_object *next = dt_object_child(dt);
1193 struct lod_object *lo = lod_dt_obj(dt);
1198 * declare setattr on the local object
1200 rc = lod_sub_declare_attr_set(env, next, attr, th);
1204 /* osp_declare_attr_set() ignores all attributes other than
1205 * UID, GID, PROJID, and size, and osp_attr_set() ignores all
1206 * but UID, GID and PROJID. Declaration of size attr setting
1207 * happens through lod_declare_init_size(), and not through
1208 * this function. Therefore we need not load striping unless
1209 * ownership is changing. This should save memory and (we hope)
1210 * speed up rename().
1212 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1213 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1216 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1219 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID | LA_MODE |
1220 LA_ATIME | LA_MTIME | LA_CTIME |
1225 * load striping information, notice we don't do this when object
1226 * is being initialized as we don't need this information till
1227 * few specific cases like destroy, chown
1229 rc = lod_striping_load(env, lo);
1233 if (!lod_obj_is_striped(dt))
1237 * if object is striped declare changes on the stripes
1239 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1240 LASSERT(lo->ldo_stripe);
1241 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1242 if (lo->ldo_stripe[i] == NULL)
1244 rc = lod_sub_declare_attr_set(env, lo->ldo_stripe[i],
1250 struct lod_obj_stripe_cb_data data = { { 0 } };
1252 data.locd_attr = attr;
1253 data.locd_declare = true;
1254 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1255 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1261 if (!dt_object_exists(next) || dt_object_remote(next) ||
1262 !S_ISREG(attr->la_mode))
1265 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1266 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
1270 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) ||
1271 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1272 struct lod_thread_info *info = lod_env_info(env);
1273 struct lu_buf *buf = &info->lti_buf;
1275 buf->lb_buf = info->lti_ea_store;
1276 buf->lb_len = info->lti_ea_store_size;
1277 rc = lod_sub_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1278 LU_XATTR_REPLACE, th);
1285 * Implementation of dt_object_operations::do_attr_set.
1287 * If the object is striped, then apply the changes to all or subset of
1288 * the stripes depending on the object type and specific attributes.
1290 * \see dt_object_operations::do_attr_set() in the API description for details.
1292 static int lod_attr_set(const struct lu_env *env,
1293 struct dt_object *dt,
1294 const struct lu_attr *attr,
1297 struct dt_object *next = dt_object_child(dt);
1298 struct lod_object *lo = lod_dt_obj(dt);
1303 * apply changes to the local object
1305 rc = lod_sub_attr_set(env, next, attr, th);
1309 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1310 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1313 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1316 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE | LA_PROJID |
1317 LA_ATIME | LA_MTIME | LA_CTIME |
1322 /* FIXME: a tricky case in the code path of mdd_layout_change():
1323 * the in-memory striping information has been freed in lod_xattr_set()
1324 * due to layout change. It has to load stripe here again. It only
1325 * changes flags of layout so declare_attr_set() is still accurate */
1326 rc = lod_striping_load(env, lo);
1330 if (!lod_obj_is_striped(dt))
1334 * if object is striped, apply changes to all the stripes
1336 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1337 LASSERT(lo->ldo_stripe);
1338 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1339 if (unlikely(lo->ldo_stripe[i] == NULL))
1342 if ((dt_object_exists(lo->ldo_stripe[i]) == 0))
1345 rc = lod_sub_attr_set(env, lo->ldo_stripe[i], attr, th);
1350 struct lod_obj_stripe_cb_data data = { { 0 } };
1352 data.locd_attr = attr;
1353 data.locd_declare = false;
1354 data.locd_comp_skip_cb = lod_obj_attr_set_comp_skip_cb;
1355 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1356 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1362 if (!dt_object_exists(next) || dt_object_remote(next) ||
1363 !S_ISREG(attr->la_mode))
1366 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1367 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
1371 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE)) {
1372 struct lod_thread_info *info = lod_env_info(env);
1373 struct lu_buf *buf = &info->lti_buf;
1374 struct ost_id *oi = &info->lti_ostid;
1375 struct lu_fid *fid = &info->lti_fid;
1376 struct lov_mds_md_v1 *lmm;
1377 struct lov_ost_data_v1 *objs;
1380 rc = lod_get_lov_ea(env, lo);
1384 buf->lb_buf = info->lti_ea_store;
1385 buf->lb_len = info->lti_ea_store_size;
1386 lmm = info->lti_ea_store;
1387 magic = le32_to_cpu(lmm->lmm_magic);
1388 if (magic == LOV_MAGIC_COMP_V1) {
1389 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1390 struct lov_comp_md_entry_v1 *lcme =
1391 &lcm->lcm_entries[0];
1393 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
1394 magic = le32_to_cpu(lmm->lmm_magic);
1397 if (magic == LOV_MAGIC_V1)
1398 objs = &(lmm->lmm_objects[0]);
1400 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1401 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1402 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1404 fid_to_ostid(fid, oi);
1405 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1407 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1408 LU_XATTR_REPLACE, th);
1409 } else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1410 struct lod_thread_info *info = lod_env_info(env);
1411 struct lu_buf *buf = &info->lti_buf;
1412 struct lov_comp_md_v1 *lcm;
1413 struct lov_comp_md_entry_v1 *lcme;
1415 rc = lod_get_lov_ea(env, lo);
1419 buf->lb_buf = info->lti_ea_store;
1420 buf->lb_len = info->lti_ea_store_size;
1422 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
1425 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1426 lcme = &lcm->lcm_entries[0];
1427 le64_add_cpu(&lcme->lcme_extent.e_start, 1);
1428 le64_add_cpu(&lcme->lcme_extent.e_end, -1);
1430 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1431 LU_XATTR_REPLACE, th);
1438 * Implementation of dt_object_operations::do_xattr_get.
1440 * If LOV EA is requested from the root object and it's not
1441 * found, then return default striping for the filesystem.
1443 * \see dt_object_operations::do_xattr_get() in the API description for details.
1445 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1446 struct lu_buf *buf, const char *name)
1448 struct lod_thread_info *info = lod_env_info(env);
1449 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1454 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1455 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1456 struct lmv_mds_md_v1 *lmv1;
1459 if (rc > (typeof(rc))sizeof(*lmv1))
1462 if (rc < (typeof(rc))sizeof(*lmv1))
1463 RETURN(rc = rc > 0 ? -EINVAL : rc);
1465 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1466 CLASSERT(sizeof(*lmv1) <= sizeof(info->lti_key));
1468 info->lti_buf.lb_buf = info->lti_key;
1469 info->lti_buf.lb_len = sizeof(*lmv1);
1470 rc = dt_xattr_get(env, dt_object_child(dt),
1471 &info->lti_buf, name);
1472 if (unlikely(rc != sizeof(*lmv1)))
1473 RETURN(rc = rc > 0 ? -EINVAL : rc);
1475 lmv1 = info->lti_buf.lb_buf;
1476 /* The on-disk LMV EA only contains header, but the
1477 * returned LMV EA size should contain the space for
1478 * the FIDs of all shards of the striped directory. */
1479 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1480 rc = lmv_mds_md_size(
1481 le32_to_cpu(lmv1->lmv_stripe_count),
1484 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1488 RETURN(rc = rc1 != 0 ? rc1 : rc);
1491 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1495 * XXX: Only used by lfsck
1497 * lod returns default striping on the real root of the device
1498 * this is like the root stores default striping for the whole
1499 * filesystem. historically we've been using a different approach
1500 * and store it in the config.
1502 dt_root_get(env, dev->lod_child, &info->lti_fid);
1503 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1505 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1506 struct lov_user_md *lum = buf->lb_buf;
1507 struct lov_desc *desc = &dev->lod_desc;
1509 if (buf->lb_buf == NULL) {
1511 } else if (buf->lb_len >= sizeof(*lum)) {
1512 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1513 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1514 lmm_oi_set_id(&lum->lmm_oi, 0);
1515 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1516 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1517 lum->lmm_stripe_size = cpu_to_le32(
1518 desc->ld_default_stripe_size);
1519 lum->lmm_stripe_count = cpu_to_le16(
1520 desc->ld_default_stripe_count);
1521 lum->lmm_stripe_offset = cpu_to_le16(
1522 desc->ld_default_stripe_offset);
1535 * Checks that the magic of the stripe is sane.
1537 * \param[in] lod lod device
1538 * \param[in] lum a buffer storing LMV EA to verify
1540 * \retval 0 if the EA is sane
1541 * \retval negative otherwise
1543 static int lod_verify_md_striping(struct lod_device *lod,
1544 const struct lmv_user_md_v1 *lum)
1546 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1547 CERROR("%s: invalid lmv_user_md: magic = %x, "
1548 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1549 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1550 (int)le32_to_cpu(lum->lum_stripe_offset),
1551 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1559 * Initialize LMV EA for a slave.
1561 * Initialize slave's LMV EA from the master's LMV EA.
1563 * \param[in] master_lmv a buffer containing master's EA
1564 * \param[out] slave_lmv a buffer where slave's EA will be stored
1567 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1568 const struct lmv_mds_md_v1 *master_lmv)
1570 *slave_lmv = *master_lmv;
1571 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1577 * Generate LMV EA from the object passed as \a dt. The object must have
1578 * the stripes created and initialized.
1580 * \param[in] env execution environment
1581 * \param[in] dt object
1582 * \param[out] lmv_buf buffer storing generated LMV EA
1584 * \retval 0 on success
1585 * \retval negative if failed
1587 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1588 struct lu_buf *lmv_buf)
1590 struct lod_thread_info *info = lod_env_info(env);
1591 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1592 struct lod_object *lo = lod_dt_obj(dt);
1593 struct lmv_mds_md_v1 *lmm1;
1595 int type = LU_SEQ_RANGE_ANY;
1600 LASSERT(lo->ldo_dir_striped != 0);
1601 LASSERT(lo->ldo_dir_stripe_count > 0);
1602 stripe_count = lo->ldo_dir_stripe_count;
1603 /* Only store the LMV EA heahder on the disk. */
1604 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1605 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1609 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1612 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1613 memset(lmm1, 0, sizeof(*lmm1));
1614 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1615 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1616 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1617 if (lo->ldo_dir_hash_type & LMV_HASH_FLAG_MIGRATION) {
1618 lmm1->lmv_migrate_hash = cpu_to_le32(lo->ldo_dir_migrate_hash);
1619 lmm1->lmv_migrate_offset =
1620 cpu_to_le32(lo->ldo_dir_migrate_offset);
1622 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1627 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1628 lmv_buf->lb_buf = info->lti_ea_store;
1629 lmv_buf->lb_len = sizeof(*lmm1);
1635 * Create in-core represenation for a striped directory.
1637 * Parse the buffer containing LMV EA and instantiate LU objects
1638 * representing the stripe objects. The pointers to the objects are
1639 * stored in ldo_stripe field of \a lo. This function is used when
1640 * we need to access an already created object (i.e. load from a disk).
1642 * \param[in] env execution environment
1643 * \param[in] lo lod object
1644 * \param[in] buf buffer containing LMV EA
1646 * \retval 0 on success
1647 * \retval negative if failed
1649 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1650 const struct lu_buf *buf)
1652 struct lod_thread_info *info = lod_env_info(env);
1653 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1654 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1655 struct dt_object **stripe;
1656 union lmv_mds_md *lmm = buf->lb_buf;
1657 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1658 struct lu_fid *fid = &info->lti_fid;
1663 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1665 if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
1668 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1669 lo->ldo_dir_slave_stripe = 1;
1673 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1676 if (le32_to_cpu(lmv1->lmv_stripe_count) < 1)
1679 LASSERT(lo->ldo_stripe == NULL);
1680 OBD_ALLOC(stripe, sizeof(stripe[0]) *
1681 (le32_to_cpu(lmv1->lmv_stripe_count)));
1685 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1686 struct dt_device *tgt_dt;
1687 struct dt_object *dto;
1688 int type = LU_SEQ_RANGE_ANY;
1691 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1692 if (!fid_is_sane(fid))
1693 GOTO(out, rc = -ESTALE);
1695 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1699 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1700 tgt_dt = lod->lod_child;
1702 struct lod_tgt_desc *tgt;
1704 tgt = LTD_TGT(ltd, idx);
1706 GOTO(out, rc = -ESTALE);
1707 tgt_dt = tgt->ltd_tgt;
1710 dto = dt_locate_at(env, tgt_dt, fid,
1711 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1714 GOTO(out, rc = PTR_ERR(dto));
1719 lo->ldo_stripe = stripe;
1720 lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1721 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1723 lod_striping_free_nolock(env, lo);
1729 * Declare create a striped directory.
1731 * Declare creating a striped directory with a given stripe pattern on the
1732 * specified MDTs. A striped directory is represented as a regular directory
1733 * - an index listing all the stripes. The stripes point back to the master
1734 * object with ".." and LinkEA. The master object gets LMV EA which
1735 * identifies it as a striped directory. The function allocates FIDs
1738 * \param[in] env execution environment
1739 * \param[in] dt object
1740 * \param[in] attr attributes to initialize the objects with
1741 * \param[in] dof type of objects to be created
1742 * \param[in] th transaction handle
1744 * \retval 0 on success
1745 * \retval negative if failed
1747 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1748 struct dt_object *dt,
1749 struct lu_attr *attr,
1750 struct dt_object_format *dof,
1753 struct lod_thread_info *info = lod_env_info(env);
1754 struct lu_buf lmv_buf;
1755 struct lu_buf slave_lmv_buf;
1756 struct lmv_mds_md_v1 *lmm;
1757 struct lmv_mds_md_v1 *slave_lmm = NULL;
1758 struct dt_insert_rec *rec = &info->lti_dt_rec;
1759 struct lod_object *lo = lod_dt_obj(dt);
1764 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1767 lmm = lmv_buf.lb_buf;
1769 OBD_ALLOC_PTR(slave_lmm);
1770 if (slave_lmm == NULL)
1771 GOTO(out, rc = -ENOMEM);
1773 lod_prep_slave_lmv_md(slave_lmm, lmm);
1774 slave_lmv_buf.lb_buf = slave_lmm;
1775 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1777 if (!dt_try_as_dir(env, dt_object_child(dt)))
1778 GOTO(out, rc = -EINVAL);
1780 rec->rec_type = S_IFDIR;
1781 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1782 struct dt_object *dto = lo->ldo_stripe[i];
1783 char *stripe_name = info->lti_key;
1784 struct lu_name *sname;
1785 struct linkea_data ldata = { NULL };
1786 struct lu_buf linkea_buf;
1788 rc = lod_sub_declare_create(env, dto, attr, NULL, dof, th);
1792 if (!dt_try_as_dir(env, dto))
1793 GOTO(out, rc = -EINVAL);
1795 rc = lod_sub_declare_ref_add(env, dto, th);
1799 rec->rec_fid = lu_object_fid(&dto->do_lu);
1800 rc = lod_sub_declare_insert(env, dto,
1801 (const struct dt_rec *)rec,
1802 (const struct dt_key *)dot, th);
1806 /* master stripe FID will be put to .. */
1807 rec->rec_fid = lu_object_fid(&dt->do_lu);
1808 rc = lod_sub_declare_insert(env, dto,
1809 (const struct dt_rec *)rec,
1810 (const struct dt_key *)dotdot, th);
1814 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1815 cfs_fail_val != i) {
1816 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1818 slave_lmm->lmv_master_mdt_index =
1821 slave_lmm->lmv_master_mdt_index =
1823 rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
1824 XATTR_NAME_LMV, 0, th);
1829 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1831 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1832 PFID(lu_object_fid(&dto->do_lu)), i + 1);
1834 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1835 PFID(lu_object_fid(&dto->do_lu)), i);
1837 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
1838 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1839 sname, lu_object_fid(&dt->do_lu));
1843 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1844 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1845 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
1846 XATTR_NAME_LINK, 0, th);
1850 rec->rec_fid = lu_object_fid(&dto->do_lu);
1851 rc = lod_sub_declare_insert(env, dt_object_child(dt),
1852 (const struct dt_rec *)rec,
1853 (const struct dt_key *)stripe_name,
1858 rc = lod_sub_declare_ref_add(env, dt_object_child(dt), th);
1863 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
1864 &lmv_buf, XATTR_NAME_LMV, 0, th);
1868 if (slave_lmm != NULL)
1869 OBD_FREE_PTR(slave_lmm);
1874 static int lod_prep_md_striped_create(const struct lu_env *env,
1875 struct dt_object *dt,
1876 struct lu_attr *attr,
1877 const struct lmv_user_md_v1 *lum,
1878 struct dt_object_format *dof,
1881 struct lod_thread_info *info = lod_env_info(env);
1882 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1883 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1884 struct lod_object *lo = lod_dt_obj(dt);
1885 struct dt_object **stripe;
1892 bool is_specific = false;
1895 /* The lum has been verifed in lod_verify_md_striping */
1896 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1897 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
1898 LASSERT(le32_to_cpu(lum->lum_stripe_count) > 0);
1900 stripe_count = le32_to_cpu(lum->lum_stripe_count);
1902 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
1903 if (idx_array == NULL)
1906 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_count);
1908 GOTO(out_free, rc = -ENOMEM);
1910 /* Start index must be the master MDT */
1911 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1912 idx_array[0] = master_index;
1913 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
1915 for (i = 1; i < stripe_count; i++)
1916 idx_array[i] = le32_to_cpu(lum->lum_objects[i].lum_mds);
1919 for (i = 0; i < stripe_count; i++) {
1920 struct lod_tgt_desc *tgt = NULL;
1921 struct dt_object *dto;
1922 struct lu_fid fid = { 0 };
1924 struct lu_object_conf conf = { 0 };
1925 struct dt_device *tgt_dt = NULL;
1927 /* Try to find next avaible target */
1929 for (j = 0; j < lod->lod_remote_mdt_count;
1930 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
1931 bool already_allocated = false;
1934 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
1935 idx, lod->lod_remote_mdt_count + 1, i);
1937 if (likely(!is_specific &&
1938 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
1939 /* check whether the idx already exists
1940 * in current allocated array */
1941 for (k = 0; k < i; k++) {
1942 if (idx_array[k] == idx) {
1943 already_allocated = true;
1948 if (already_allocated)
1952 /* Sigh, this index is not in the bitmap, let's check
1953 * next available target */
1954 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) &&
1955 idx != master_index)
1958 if (idx == master_index) {
1959 /* Allocate the FID locally */
1960 rc = obd_fid_alloc(env, lod->lod_child_exp,
1964 tgt_dt = lod->lod_child;
1968 /* check the status of the OSP */
1969 tgt = LTD_TGT(ltd, idx);
1973 tgt_dt = tgt->ltd_tgt;
1974 rc = dt_statfs(env, tgt_dt, &info->lti_osfs);
1976 /* this OSP doesn't feel well */
1981 rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
1990 /* Can not allocate more stripes */
1991 if (j == lod->lod_remote_mdt_count) {
1992 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
1993 lod2obd(lod)->obd_name, stripe_count, i);
1997 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
1998 idx, i, PFID(&fid));
2000 /* Set the start index for next stripe allocation */
2001 if (!is_specific && i < stripe_count - 1) {
2003 * for large dir test, put all other slaves on one
2004 * remote MDT, otherwise we may save too many local
2005 * slave locks which will exceed RS_MAX_LOCKS.
2007 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)))
2009 idx_array[i + 1] = (idx + 1) %
2010 (lod->lod_remote_mdt_count + 1);
2012 /* tgt_dt and fid must be ready after search avaible OSP
2013 * in the above loop */
2014 LASSERT(tgt_dt != NULL);
2015 LASSERT(fid_is_sane(&fid));
2016 conf.loc_flags = LOC_F_NEW;
2017 dto = dt_locate_at(env, tgt_dt, &fid,
2018 dt->do_lu.lo_dev->ld_site->ls_top_dev,
2021 GOTO(out_put, rc = PTR_ERR(dto));
2025 lo->ldo_dir_striped = 1;
2026 lo->ldo_stripe = stripe;
2027 lo->ldo_dir_stripe_count = i;
2028 lo->ldo_dir_stripes_allocated = stripe_count;
2030 lo->ldo_dir_stripe_loaded = 1;
2032 if (lo->ldo_dir_stripe_count == 0)
2033 GOTO(out_put, rc = -ENOSPC);
2035 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2041 for (i = 0; i < stripe_count; i++)
2042 if (stripe[i] != NULL)
2043 dt_object_put(env, stripe[i]);
2044 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_count);
2045 lo->ldo_dir_stripe_count = 0;
2046 lo->ldo_dir_stripes_allocated = 0;
2047 lo->ldo_stripe = NULL;
2051 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
2057 * Declare create striped md object.
2059 * The function declares intention to create a striped directory. This is a
2060 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2061 * is to verify pattern \a lum_buf is good. Check that function for the details.
2063 * \param[in] env execution environment
2064 * \param[in] dt object
2065 * \param[in] attr attributes to initialize the objects with
2066 * \param[in] lum_buf a pattern specifying the number of stripes and
2068 * \param[in] dof type of objects to be created
2069 * \param[in] th transaction handle
2071 * \retval 0 on success
2072 * \retval negative if failed
2075 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2076 struct dt_object *dt,
2077 struct lu_attr *attr,
2078 const struct lu_buf *lum_buf,
2079 struct dt_object_format *dof,
2082 struct lod_object *lo = lod_dt_obj(dt);
2083 struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
2087 LASSERT(lum != NULL);
2089 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
2090 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
2091 (int)le32_to_cpu(lum->lum_stripe_offset));
2093 if (lo->ldo_dir_stripe_count == 0)
2096 /* prepare dir striped objects */
2097 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2099 /* failed to create striping, let's reset
2100 * config so that others don't get confused */
2101 lod_striping_free(env, lo);
2109 * Append source stripes after target stripes for migrating directory. NB, we
2110 * only need to declare this, the append is done inside lod_xattr_set_lmv().
2112 * \param[in] env execution environment
2113 * \param[in] dt target object
2114 * \param[in] buf LMV buf which contains source stripe fids
2115 * \param[in] th transaction handle
2117 * \retval 0 on success
2118 * \retval negative if failed
2120 static int lod_dir_declare_layout_add(const struct lu_env *env,
2121 struct dt_object *dt,
2122 const struct lu_buf *buf,
2125 struct lod_thread_info *info = lod_env_info(env);
2126 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
2127 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
2128 struct lod_object *lo = lod_dt_obj(dt);
2129 struct dt_object *next = dt_object_child(dt);
2130 struct dt_object_format *dof = &info->lti_format;
2131 struct lmv_mds_md_v1 *lmv = buf->lb_buf;
2132 struct dt_object **stripe;
2133 __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
2134 struct lu_fid *fid = &info->lti_fid;
2135 struct lod_tgt_desc *tgt;
2136 struct dt_object *dto;
2137 struct dt_device *tgt_dt;
2138 int type = LU_SEQ_RANGE_ANY;
2139 struct dt_insert_rec *rec = &info->lti_dt_rec;
2140 char *stripe_name = info->lti_key;
2141 struct lu_name *sname;
2142 struct linkea_data ldata = { NULL };
2143 struct lu_buf linkea_buf;
2150 if (le32_to_cpu(lmv->lmv_magic) != LMV_MAGIC_V1)
2153 if (stripe_count == 0)
2156 dof->dof_type = DFT_DIR;
2159 sizeof(*stripe) * (lo->ldo_dir_stripe_count + stripe_count));
2163 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
2164 stripe[i] = lo->ldo_stripe[i];
2166 for (i = 0; i < stripe_count; i++) {
2168 &lmv->lmv_stripe_fids[i]);
2169 if (!fid_is_sane(fid))
2170 GOTO(out, rc = -ESTALE);
2172 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
2176 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
2177 tgt_dt = lod->lod_child;
2179 tgt = LTD_TGT(ltd, idx);
2181 GOTO(out, rc = -ESTALE);
2182 tgt_dt = tgt->ltd_tgt;
2185 dto = dt_locate_at(env, tgt_dt, fid,
2186 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
2189 GOTO(out, rc = PTR_ERR(dto));
2191 stripe[i + lo->ldo_dir_stripe_count] = dto;
2193 if (!dt_try_as_dir(env, dto))
2194 GOTO(out, rc = -ENOTDIR);
2196 rc = lod_sub_declare_ref_add(env, dto, th);
2200 rc = lod_sub_declare_insert(env, dto,
2201 (const struct dt_rec *)rec,
2202 (const struct dt_key *)dot, th);
2206 rc = lod_sub_declare_insert(env, dto,
2207 (const struct dt_rec *)rec,
2208 (const struct dt_key *)dotdot, th);
2212 rc = lod_sub_declare_xattr_set(env, dto, buf,
2213 XATTR_NAME_LMV, 0, th);
2217 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
2218 PFID(lu_object_fid(&dto->do_lu)),
2219 i + lo->ldo_dir_stripe_count);
2221 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
2222 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
2223 sname, lu_object_fid(&dt->do_lu));
2227 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
2228 linkea_buf.lb_len = ldata.ld_leh->leh_len;
2229 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
2230 XATTR_NAME_LINK, 0, th);
2234 rc = lod_sub_declare_insert(env, next,
2235 (const struct dt_rec *)rec,
2236 (const struct dt_key *)stripe_name,
2241 rc = lod_sub_declare_ref_add(env, next, th);
2247 OBD_FREE(lo->ldo_stripe,
2248 sizeof(*stripe) * lo->ldo_dir_stripes_allocated);
2249 lo->ldo_stripe = stripe;
2250 lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count;
2251 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type);
2252 lo->ldo_dir_stripe_count += stripe_count;
2253 lo->ldo_dir_stripes_allocated += stripe_count;
2254 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_MIGRATION;
2258 i = lo->ldo_dir_stripe_count;
2259 while (i < lo->ldo_dir_stripe_count + stripe_count && stripe[i])
2260 dt_object_put(env, stripe[i++]);
2263 sizeof(*stripe) * (stripe_count + lo->ldo_dir_stripe_count));
2267 static int lod_dir_declare_layout_delete(const struct lu_env *env,
2268 struct dt_object *dt,
2269 const struct lu_buf *buf,
2272 struct lod_thread_info *info = lod_env_info(env);
2273 struct lod_object *lo = lod_dt_obj(dt);
2274 struct dt_object *next = dt_object_child(dt);
2275 struct lmv_user_md *lmu = buf->lb_buf;
2276 __u32 final_stripe_count;
2277 char *stripe_name = info->lti_key;
2278 struct dt_object *dto;
2285 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
2286 if (final_stripe_count >= lo->ldo_dir_stripe_count)
2289 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
2290 dto = lo->ldo_stripe[i];
2293 if (!dt_try_as_dir(env, dto))
2296 rc = lod_sub_declare_delete(env, dto,
2297 (const struct dt_key *)dot, th);
2301 rc = lod_sub_declare_ref_del(env, dto, th);
2305 rc = lod_sub_declare_delete(env, dto,
2306 (const struct dt_key *)dotdot, th);
2310 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2311 PFID(lu_object_fid(&dto->do_lu)), i);
2313 rc = lod_sub_declare_delete(env, next,
2314 (const struct dt_key *)stripe_name, th);
2318 rc = lod_sub_declare_ref_del(env, next, th);
2327 * delete stripes from dir master object, the lum_stripe_count in argument is
2328 * the final stripe count, the stripes after that will be deleted, NB, they
2329 * are not destroyed, but deleted from it's parent namespace, this function
2330 * will be called in two places:
2331 * 1. mdd_migrate_create() delete stripes from source, and append them to
2333 * 2. mdd_dir_layout_shrink() delete stripes from source, and destroy them.
2335 static int lod_dir_layout_delete(const struct lu_env *env,
2336 struct dt_object *dt,
2337 const struct lu_buf *buf,
2340 struct lod_thread_info *info = lod_env_info(env);
2341 struct lod_object *lo = lod_dt_obj(dt);
2342 struct dt_object *next = dt_object_child(dt);
2343 struct lmv_user_md *lmu = buf->lb_buf;
2344 __u32 final_stripe_count;
2345 char *stripe_name = info->lti_key;
2346 struct dt_object *dto;
2355 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
2356 if (final_stripe_count >= lo->ldo_dir_stripe_count)
2359 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
2360 dto = lo->ldo_stripe[i];
2363 rc = lod_sub_delete(env, dto,
2364 (const struct dt_key *)dotdot, th);
2368 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2369 PFID(lu_object_fid(&dto->do_lu)), i);
2371 rc = lod_sub_delete(env, next,
2372 (const struct dt_key *)stripe_name, th);
2376 rc = lod_sub_ref_del(env, next, th);
2381 lod_striping_free(env, lod_dt_obj(dt));
2387 * Implementation of dt_object_operations::do_declare_xattr_set.
2389 * Used with regular (non-striped) objects. Basically it
2390 * initializes the striping information and applies the
2391 * change to all the stripes.
2393 * \see dt_object_operations::do_declare_xattr_set() in the API description
2396 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2397 struct dt_object *dt,
2398 const struct lu_buf *buf,
2399 const char *name, int fl,
2402 struct dt_object *next = dt_object_child(dt);
2403 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2404 struct lod_object *lo = lod_dt_obj(dt);
2409 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2410 struct lmv_user_md_v1 *lum;
2412 LASSERT(buf != NULL && buf->lb_buf != NULL);
2414 rc = lod_verify_md_striping(d, lum);
2417 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2418 rc = lod_verify_striping(d, lo, buf, false);
2423 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2427 /* Note: Do not set LinkEA on sub-stripes, otherwise
2428 * it will confuse the fid2path process(see mdt_path_current()).
2429 * The linkEA between master and sub-stripes is set in
2430 * lod_xattr_set_lmv(). */
2431 if (strcmp(name, XATTR_NAME_LINK) == 0)
2434 /* set xattr to each stripes, if needed */
2435 rc = lod_striping_load(env, lo);
2439 if (lo->ldo_dir_stripe_count == 0)
2442 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2443 LASSERT(lo->ldo_stripe[i]);
2445 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2455 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2456 struct lod_object *lo,
2457 struct dt_object *dt, struct thandle *th,
2458 int comp_idx, int stripe_idx,
2459 struct lod_obj_stripe_cb_data *data)
2461 struct lod_thread_info *info = lod_env_info(env);
2462 struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
2463 struct filter_fid *ff = &info->lti_ff;
2464 struct lu_buf *buf = &info->lti_buf;
2468 buf->lb_len = sizeof(*ff);
2469 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2477 * locd_buf is set if it's called by dir migration, which doesn't check
2480 if (data->locd_buf) {
2481 memset(ff, 0, sizeof(*ff));
2482 ff->ff_parent = *(struct lu_fid *)data->locd_buf->lb_buf;
2484 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
2486 if (lu_fid_eq(lod_object_fid(lo), &ff->ff_parent) &&
2487 ff->ff_layout.ol_comp_id == comp->llc_id)
2490 memset(ff, 0, sizeof(*ff));
2491 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2494 /* rewrite filter_fid */
2495 ff->ff_parent.f_ver = stripe_idx;
2496 ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
2497 ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
2498 ff->ff_layout.ol_comp_id = comp->llc_id;
2499 ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
2500 ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
2501 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
2503 if (data->locd_declare)
2504 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2505 LU_XATTR_REPLACE, th);
2507 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2508 LU_XATTR_REPLACE, th);
2514 * Reset parent FID on OST object
2516 * Replace parent FID with @dt object FID, which is only called during migration
2517 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2518 * the FID is changed.
2520 * \param[in] env execution environment
2521 * \param[in] dt dt_object whose stripes's parent FID will be reset
2522 * \parem[in] th thandle
2523 * \param[in] declare if it is declare
2525 * \retval 0 if reset succeeds
2526 * \retval negative errno if reset fails
2528 static int lod_replace_parent_fid(const struct lu_env *env,
2529 struct dt_object *dt,
2530 const struct lu_buf *buf,
2531 struct thandle *th, bool declare)
2533 struct lod_object *lo = lod_dt_obj(dt);
2534 struct lod_thread_info *info = lod_env_info(env);
2535 struct filter_fid *ff;
2536 struct lod_obj_stripe_cb_data data = { { 0 } };
2540 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2542 /* set xattr to each stripes, if needed */
2543 rc = lod_striping_load(env, lo);
2547 if (!lod_obj_is_striped(dt))
2550 if (info->lti_ea_store_size < sizeof(*ff)) {
2551 rc = lod_ea_store_resize(info, sizeof(*ff));
2556 data.locd_declare = declare;
2557 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2558 data.locd_buf = buf;
2559 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2564 inline __u16 lod_comp_entry_stripe_count(struct lod_object *lo,
2565 struct lod_layout_component *entry,
2568 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2572 else if (lod_comp_inited(entry))
2573 return entry->llc_stripe_count;
2574 else if ((__u16)-1 == entry->llc_stripe_count)
2575 return lod->lod_desc.ld_tgt_count;
2577 return lod_get_stripe_count(lod, lo, entry->llc_stripe_count);
2580 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2582 int magic, size = 0, i;
2583 struct lod_layout_component *comp_entries;
2588 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2589 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2591 lo->ldo_def_striping->lds_def_striping_is_composite;
2593 comp_cnt = lo->ldo_comp_cnt;
2594 comp_entries = lo->ldo_comp_entries;
2595 is_composite = lo->ldo_is_composite;
2599 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2601 size = sizeof(struct lov_comp_md_v1) +
2602 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2603 LASSERT(size % sizeof(__u64) == 0);
2606 for (i = 0; i < comp_cnt; i++) {
2609 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2610 stripe_count = lod_comp_entry_stripe_count(lo, &comp_entries[i],
2612 if (!is_dir && is_composite)
2613 lod_comp_shrink_stripe_count(&comp_entries[i],
2616 size += lov_user_md_size(stripe_count, magic);
2617 LASSERT(size % sizeof(__u64) == 0);
2623 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2624 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2627 * \param[in] env execution environment
2628 * \param[in] dt dt_object to add components on
2629 * \param[in] buf buffer contains components to be added
2630 * \parem[in] th thandle
2632 * \retval 0 on success
2633 * \retval negative errno on failure
2635 static int lod_declare_layout_add(const struct lu_env *env,
2636 struct dt_object *dt,
2637 const struct lu_buf *buf,
2640 struct lod_thread_info *info = lod_env_info(env);
2641 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2642 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2643 struct dt_object *next = dt_object_child(dt);
2644 struct lov_desc *desc = &d->lod_desc;
2645 struct lod_object *lo = lod_dt_obj(dt);
2646 struct lov_user_md_v3 *v3;
2647 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2649 int i, rc, array_cnt, old_array_cnt;
2652 LASSERT(lo->ldo_is_composite);
2654 if (lo->ldo_flr_state != LCM_FL_NONE)
2657 rc = lod_verify_striping(d, lo, buf, false);
2661 magic = comp_v1->lcm_magic;
2662 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2663 lustre_swab_lov_comp_md_v1(comp_v1);
2664 magic = comp_v1->lcm_magic;
2667 if (magic != LOV_USER_MAGIC_COMP_V1)
2670 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2671 OBD_ALLOC(comp_array, sizeof(*comp_array) * array_cnt);
2672 if (comp_array == NULL)
2675 memcpy(comp_array, lo->ldo_comp_entries,
2676 sizeof(*comp_array) * lo->ldo_comp_cnt);
2678 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2679 struct lov_user_md_v1 *v1;
2680 struct lu_extent *ext;
2682 v1 = (struct lov_user_md *)((char *)comp_v1 +
2683 comp_v1->lcm_entries[i].lcme_offset);
2684 ext = &comp_v1->lcm_entries[i].lcme_extent;
2686 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2687 lod_comp->llc_extent.e_start = ext->e_start;
2688 lod_comp->llc_extent.e_end = ext->e_end;
2689 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2690 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2692 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2693 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2694 lod_adjust_stripe_info(lod_comp, desc);
2696 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2697 v3 = (struct lov_user_md_v3 *) v1;
2698 if (v3->lmm_pool_name[0] != '\0') {
2699 rc = lod_set_pool(&lod_comp->llc_pool,
2707 old_array = lo->ldo_comp_entries;
2708 old_array_cnt = lo->ldo_comp_cnt;
2710 lo->ldo_comp_entries = comp_array;
2711 lo->ldo_comp_cnt = array_cnt;
2713 /* No need to increase layout generation here, it will be increased
2714 * later when generating component ID for the new components */
2716 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2717 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2718 XATTR_NAME_LOV, 0, th);
2720 lo->ldo_comp_entries = old_array;
2721 lo->ldo_comp_cnt = old_array_cnt;
2725 OBD_FREE(old_array, sizeof(*lod_comp) * old_array_cnt);
2727 LASSERT(lo->ldo_mirror_count == 1);
2728 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2733 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2734 lod_comp = &comp_array[i];
2735 if (lod_comp->llc_pool != NULL) {
2736 OBD_FREE(lod_comp->llc_pool,
2737 strlen(lod_comp->llc_pool) + 1);
2738 lod_comp->llc_pool = NULL;
2741 OBD_FREE(comp_array, sizeof(*comp_array) * array_cnt);
2746 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2747 * the '$field' can only be 'flags' now. The xattr value is binary
2748 * lov_comp_md_v1 which contains the component ID(s) and the value of
2749 * the field to be modified.
2751 * \param[in] env execution environment
2752 * \param[in] dt dt_object to be modified
2753 * \param[in] op operation string, like "set.flags"
2754 * \param[in] buf buffer contains components to be set
2755 * \parem[in] th thandle
2757 * \retval 0 on success
2758 * \retval negative errno on failure
2760 static int lod_declare_layout_set(const struct lu_env *env,
2761 struct dt_object *dt,
2762 char *op, const struct lu_buf *buf,
2765 struct lod_layout_component *lod_comp;
2766 struct lod_thread_info *info = lod_env_info(env);
2767 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2768 struct lod_object *lo = lod_dt_obj(dt);
2769 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2772 bool changed = false;
2775 if (strcmp(op, "set.flags") != 0) {
2776 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2777 lod2obd(d)->obd_name, op);
2781 magic = comp_v1->lcm_magic;
2782 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2783 lustre_swab_lov_comp_md_v1(comp_v1);
2784 magic = comp_v1->lcm_magic;
2787 if (magic != LOV_USER_MAGIC_COMP_V1)
2790 if (comp_v1->lcm_entry_count == 0) {
2791 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2792 lod2obd(d)->obd_name);
2796 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2797 __u32 id = comp_v1->lcm_entries[i].lcme_id;
2798 __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
2799 __u32 mirror_flag = flags & LCME_MIRROR_FLAGS;
2800 bool neg = flags & LCME_FL_NEG;
2802 if (flags & LCME_FL_INIT) {
2804 lod_striping_free(env, lo);
2808 flags &= ~(LCME_MIRROR_FLAGS | LCME_FL_NEG);
2809 for (j = 0; j < lo->ldo_comp_cnt; j++) {
2810 lod_comp = &lo->ldo_comp_entries[j];
2812 /* lfs only put one flag in each entry */
2813 if ((flags && id != lod_comp->llc_id) ||
2814 (mirror_flag && mirror_id_of(id) !=
2815 mirror_id_of(lod_comp->llc_id)))
2820 lod_comp->llc_flags &= ~flags;
2822 lod_comp->llc_flags &= ~mirror_flag;
2825 lod_comp->llc_flags |= flags;
2827 lod_comp->llc_flags |= mirror_flag;
2828 if (mirror_flag & LCME_FL_NOSYNC)
2829 lod_comp->llc_timestamp =
2830 ktime_get_real_seconds();
2838 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
2839 lod2obd(d)->obd_name);
2843 lod_obj_inc_layout_gen(lo);
2845 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2846 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
2847 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
2852 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
2853 * and the xattr value is a unique component ID or a special lcme_id.
2855 * \param[in] env execution environment
2856 * \param[in] dt dt_object to be operated on
2857 * \param[in] buf buffer contains component ID or lcme_id
2858 * \parem[in] th thandle
2860 * \retval 0 on success
2861 * \retval negative errno on failure
2863 static int lod_declare_layout_del(const struct lu_env *env,
2864 struct dt_object *dt,
2865 const struct lu_buf *buf,
2868 struct lod_thread_info *info = lod_env_info(env);
2869 struct dt_object *next = dt_object_child(dt);
2870 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2871 struct lod_object *lo = lod_dt_obj(dt);
2872 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2873 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2874 __u32 magic, id, flags, neg_flags = 0;
2878 LASSERT(lo->ldo_is_composite);
2880 if (lo->ldo_flr_state != LCM_FL_NONE)
2883 magic = comp_v1->lcm_magic;
2884 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2885 lustre_swab_lov_comp_md_v1(comp_v1);
2886 magic = comp_v1->lcm_magic;
2889 if (magic != LOV_USER_MAGIC_COMP_V1)
2892 id = comp_v1->lcm_entries[0].lcme_id;
2893 flags = comp_v1->lcm_entries[0].lcme_flags;
2895 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
2896 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
2897 lod2obd(d)->obd_name, id, flags);
2901 if (id != LCME_ID_INVAL && flags != 0) {
2902 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
2903 lod2obd(d)->obd_name);
2907 if (id == LCME_ID_INVAL && !flags) {
2908 CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
2909 lod2obd(d)->obd_name);
2913 if (flags & LCME_FL_NEG) {
2914 neg_flags = flags & ~LCME_FL_NEG;
2918 left = lo->ldo_comp_cnt;
2922 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
2923 struct lod_layout_component *lod_comp;
2925 lod_comp = &lo->ldo_comp_entries[i];
2927 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
2929 else if (flags && !(flags & lod_comp->llc_flags))
2931 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
2934 if (left != (i + 1)) {
2935 CDEBUG(D_LAYOUT, "%s: this deletion will create "
2936 "a hole.\n", lod2obd(d)->obd_name);
2941 /* Mark the component as deleted */
2942 lod_comp->llc_id = LCME_ID_INVAL;
2944 /* Not instantiated component */
2945 if (lod_comp->llc_stripe == NULL)
2948 LASSERT(lod_comp->llc_stripe_count > 0);
2949 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
2950 struct dt_object *obj = lod_comp->llc_stripe[j];
2954 rc = lod_sub_declare_destroy(env, obj, th);
2960 LASSERTF(left >= 0, "left = %d\n", left);
2961 if (left == lo->ldo_comp_cnt) {
2962 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
2963 lod2obd(d)->obd_name, id);
2967 memset(attr, 0, sizeof(*attr));
2968 attr->la_valid = LA_SIZE;
2969 rc = lod_sub_declare_attr_set(env, next, attr, th);
2974 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2975 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2976 XATTR_NAME_LOV, 0, th);
2978 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
2985 * Declare layout add/set/del operations issued by special xattr names:
2987 * XATTR_LUSTRE_LOV.add add component(s) to existing file
2988 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
2989 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
2991 * \param[in] env execution environment
2992 * \param[in] dt object
2993 * \param[in] name name of xattr
2994 * \param[in] buf lu_buf contains xattr value
2995 * \param[in] th transaction handle
2997 * \retval 0 on success
2998 * \retval negative if failed
3000 static int lod_declare_modify_layout(const struct lu_env *env,
3001 struct dt_object *dt,
3003 const struct lu_buf *buf,
3006 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3007 struct lod_object *lo = lod_dt_obj(dt);
3009 int rc, len = strlen(XATTR_LUSTRE_LOV);
3012 LASSERT(dt_object_exists(dt));
3014 if (strlen(name) <= len || name[len] != '.') {
3015 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
3016 lod2obd(d)->obd_name, name);
3021 rc = lod_striping_load(env, lo);
3025 /* the layout to be modified must be a composite layout */
3026 if (!lo->ldo_is_composite) {
3027 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
3028 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3029 GOTO(unlock, rc = -EINVAL);
3032 op = (char *)name + len;
3033 if (strcmp(op, "add") == 0) {
3034 rc = lod_declare_layout_add(env, dt, buf, th);
3035 } else if (strcmp(op, "del") == 0) {
3036 rc = lod_declare_layout_del(env, dt, buf, th);
3037 } else if (strncmp(op, "set", strlen("set")) == 0) {
3038 rc = lod_declare_layout_set(env, dt, op, buf, th);
3040 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
3041 lod2obd(d)->obd_name, name);
3042 GOTO(unlock, rc = -ENOTSUPP);
3046 lod_striping_free(env, lo);
3052 * Convert a plain file lov_mds_md to a composite layout.
3054 * \param[in,out] info the thread info::lti_ea_store buffer contains little
3055 * endian plain file layout
3057 * \retval 0 on success, <0 on failure
3059 static int lod_layout_convert(struct lod_thread_info *info)
3061 struct lov_mds_md *lmm = info->lti_ea_store;
3062 struct lov_mds_md *lmm_save;
3063 struct lov_comp_md_v1 *lcm;
3064 struct lov_comp_md_entry_v1 *lcme;
3070 /* realloc buffer to a composite layout which contains one component */
3071 blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3072 le32_to_cpu(lmm->lmm_magic));
3073 size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
3075 OBD_ALLOC_LARGE(lmm_save, blob_size);
3077 GOTO(out, rc = -ENOMEM);
3079 memcpy(lmm_save, lmm, blob_size);
3081 if (info->lti_ea_store_size < size) {
3082 rc = lod_ea_store_resize(info, size);
3087 lcm = info->lti_ea_store;
3088 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
3089 lcm->lcm_size = cpu_to_le32(size);
3090 lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
3091 lmm_save->lmm_layout_gen));
3092 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
3093 lcm->lcm_entry_count = cpu_to_le16(1);
3094 lcm->lcm_mirror_count = 0;
3096 lcme = &lcm->lcm_entries[0];
3097 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
3098 lcme->lcme_extent.e_start = 0;
3099 lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
3100 lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
3101 lcme->lcme_size = cpu_to_le32(blob_size);
3103 memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
3108 OBD_FREE_LARGE(lmm_save, blob_size);
3113 * Merge layouts to form a mirrored file.
3115 static int lod_declare_layout_merge(const struct lu_env *env,
3116 struct dt_object *dt, const struct lu_buf *mbuf,
3119 struct lod_thread_info *info = lod_env_info(env);
3120 struct lu_buf *buf = &info->lti_buf;
3121 struct lod_object *lo = lod_dt_obj(dt);
3122 struct lov_comp_md_v1 *lcm;
3123 struct lov_comp_md_v1 *cur_lcm;
3124 struct lov_comp_md_v1 *merge_lcm;
3125 struct lov_comp_md_entry_v1 *lcme;
3128 __u16 cur_entry_count;
3129 __u16 merge_entry_count;
3131 __u16 mirror_id = 0;
3136 merge_lcm = mbuf->lb_buf;
3137 if (mbuf->lb_len < sizeof(*merge_lcm))
3140 /* must be an existing layout from disk */
3141 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
3144 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
3146 /* do not allow to merge two mirrored files */
3147 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
3150 /* verify the target buffer */
3151 rc = lod_get_lov_ea(env, lo);
3153 RETURN(rc ? : -ENODATA);
3155 cur_lcm = info->lti_ea_store;
3156 switch (le32_to_cpu(cur_lcm->lcm_magic)) {
3159 rc = lod_layout_convert(info);
3161 case LOV_MAGIC_COMP_V1:
3170 /* info->lti_ea_store could be reallocated in lod_layout_convert() */
3171 cur_lcm = info->lti_ea_store;
3172 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
3174 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
3175 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
3176 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
3179 /* size of new layout */
3180 size = le32_to_cpu(cur_lcm->lcm_size) +
3181 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
3183 memset(buf, 0, sizeof(*buf));
3184 lu_buf_alloc(buf, size);
3185 if (buf->lb_buf == NULL)
3189 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
3191 offset = sizeof(*lcm) +
3192 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
3193 for (i = 0; i < cur_entry_count; i++) {
3194 struct lov_comp_md_entry_v1 *cur_lcme;
3196 lcme = &lcm->lcm_entries[i];
3197 cur_lcme = &cur_lcm->lcm_entries[i];
3199 lcme->lcme_offset = cpu_to_le32(offset);
3200 memcpy((char *)lcm + offset,
3201 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
3202 le32_to_cpu(lcme->lcme_size));
3204 offset += le32_to_cpu(lcme->lcme_size);
3206 if (mirror_count == 1 &&
3207 mirror_id_of(le32_to_cpu(lcme->lcme_id)) == 0) {
3208 /* Add mirror from a non-flr file, create new mirror ID.
3209 * Otherwise, keep existing mirror's component ID, used
3210 * for mirror extension.
3212 id = pflr_id(1, i + 1);
3213 lcme->lcme_id = cpu_to_le32(id);
3216 id = MAX(le32_to_cpu(lcme->lcme_id), id);
3219 mirror_id = mirror_id_of(id) + 1;
3220 for (i = 0; i < merge_entry_count; i++) {
3221 struct lov_comp_md_entry_v1 *merge_lcme;
3223 merge_lcme = &merge_lcm->lcm_entries[i];
3224 lcme = &lcm->lcm_entries[cur_entry_count + i];
3226 *lcme = *merge_lcme;
3227 lcme->lcme_offset = cpu_to_le32(offset);
3229 id = pflr_id(mirror_id, i + 1);
3230 lcme->lcme_id = cpu_to_le32(id);
3232 memcpy((char *)lcm + offset,
3233 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
3234 le32_to_cpu(lcme->lcme_size));
3236 offset += le32_to_cpu(lcme->lcme_size);
3239 /* fixup layout information */
3240 lod_obj_inc_layout_gen(lo);
3241 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3242 lcm->lcm_size = cpu_to_le32(size);
3243 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
3244 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
3245 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NONE)
3246 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
3248 rc = lod_striping_reload(env, lo, buf);
3252 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
3253 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3261 * Split layouts, just set the LOVEA with the layout from mbuf.
3263 static int lod_declare_layout_split(const struct lu_env *env,
3264 struct dt_object *dt, const struct lu_buf *mbuf,
3267 struct lod_object *lo = lod_dt_obj(dt);
3268 struct lov_comp_md_v1 *lcm = mbuf->lb_buf;
3272 lod_obj_inc_layout_gen(lo);
3273 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3275 rc = lod_striping_reload(env, lo, mbuf);
3279 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), mbuf,
3280 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3285 * Implementation of dt_object_operations::do_declare_xattr_set.
3287 * \see dt_object_operations::do_declare_xattr_set() in the API description
3290 * the extension to the API:
3291 * - declaring LOVEA requests striping creation
3292 * - LU_XATTR_REPLACE means layout swap
3294 static int lod_declare_xattr_set(const struct lu_env *env,
3295 struct dt_object *dt,
3296 const struct lu_buf *buf,
3297 const char *name, int fl,
3300 struct dt_object *next = dt_object_child(dt);
3301 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3306 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
3307 if ((S_ISREG(mode) || mode == 0) &&
3308 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE | LU_XATTR_SPLIT)) &&
3309 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3310 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
3312 * this is a request to create object's striping.
3314 * allow to declare predefined striping on a new (!mode) object
3315 * which is supposed to be replay of regular file creation
3316 * (when LOV setting is declared)
3318 * LU_XATTR_REPLACE is set to indicate a layout swap
3320 if (dt_object_exists(dt)) {
3321 rc = dt_attr_get(env, next, attr);
3325 memset(attr, 0, sizeof(*attr));
3326 attr->la_valid = LA_TYPE | LA_MODE;
3327 attr->la_mode = S_IFREG;
3329 rc = lod_declare_striped_create(env, dt, attr, buf, th);
3330 } else if (fl & LU_XATTR_MERGE) {
3331 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3332 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3333 rc = lod_declare_layout_merge(env, dt, buf, th);
3334 } else if (fl & LU_XATTR_SPLIT) {
3335 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3336 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3337 rc = lod_declare_layout_split(env, dt, buf, th);
3338 } else if (S_ISREG(mode) &&
3339 strlen(name) > strlen(XATTR_LUSTRE_LOV) + 1 &&
3340 strncmp(name, XATTR_LUSTRE_LOV,
3341 strlen(XATTR_LUSTRE_LOV)) == 0) {
3343 * this is a request to modify object's striping.
3344 * add/set/del component(s).
3346 if (!dt_object_exists(dt))
3349 rc = lod_declare_modify_layout(env, dt, name, buf, th);
3350 } else if (strncmp(name, XATTR_NAME_LMV, strlen(XATTR_NAME_LMV)) == 0 &&
3351 strlen(name) > strlen(XATTR_NAME_LMV) + 1) {
3352 const char *op = name + strlen(XATTR_NAME_LMV) + 1;
3355 if (strcmp(op, "add") == 0)
3356 rc = lod_dir_declare_layout_add(env, dt, buf, th);
3357 else if (strcmp(op, "del") == 0)
3358 rc = lod_dir_declare_layout_delete(env, dt, buf, th);
3359 else if (strcmp(op, "set") == 0)
3360 rc = lod_sub_declare_xattr_set(env, next, buf,
3361 XATTR_NAME_LMV, fl, th);
3364 } else if (S_ISDIR(mode)) {
3365 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
3366 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3367 rc = lod_replace_parent_fid(env, dt, buf, th, true);
3369 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
3376 * Apply xattr changes to the object.
3378 * Applies xattr changes to the object and the stripes if the latter exist.
3380 * \param[in] env execution environment
3381 * \param[in] dt object
3382 * \param[in] buf buffer pointing to the new value of xattr
3383 * \param[in] name name of xattr
3384 * \param[in] fl flags
3385 * \param[in] th transaction handle
3387 * \retval 0 on success
3388 * \retval negative if failed
3390 static int lod_xattr_set_internal(const struct lu_env *env,
3391 struct dt_object *dt,
3392 const struct lu_buf *buf,
3393 const char *name, int fl,
3396 struct dt_object *next = dt_object_child(dt);
3397 struct lod_object *lo = lod_dt_obj(dt);
3402 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3403 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3406 /* Note: Do not set LinkEA on sub-stripes, otherwise
3407 * it will confuse the fid2path process(see mdt_path_current()).
3408 * The linkEA between master and sub-stripes is set in
3409 * lod_xattr_set_lmv(). */
3410 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
3413 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3414 LASSERT(lo->ldo_stripe[i]);
3416 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
3426 * Delete an extended attribute.
3428 * Deletes specified xattr from the object and the stripes if the latter exist.
3430 * \param[in] env execution environment
3431 * \param[in] dt object
3432 * \param[in] name name of xattr
3433 * \param[in] th transaction handle
3435 * \retval 0 on success
3436 * \retval negative if failed
3438 static int lod_xattr_del_internal(const struct lu_env *env,
3439 struct dt_object *dt,
3440 const char *name, struct thandle *th)
3442 struct dt_object *next = dt_object_child(dt);
3443 struct lod_object *lo = lod_dt_obj(dt);
3448 rc = lod_sub_xattr_del(env, next, name, th);
3449 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3452 if (lo->ldo_dir_stripe_count == 0)
3455 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3456 LASSERT(lo->ldo_stripe[i]);
3458 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3467 * Set default striping on a directory.
3469 * Sets specified striping on a directory object unless it matches the default
3470 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3471 * EA. This striping will be used when regular file is being created in this
3474 * \param[in] env execution environment
3475 * \param[in] dt the striped object
3476 * \param[in] buf buffer with the striping
3477 * \param[in] name name of EA
3478 * \param[in] fl xattr flag (see OSD API description)
3479 * \param[in] th transaction handle
3481 * \retval 0 on success
3482 * \retval negative if failed
3484 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
3485 struct dt_object *dt,
3486 const struct lu_buf *buf,
3487 const char *name, int fl,
3490 struct lov_user_md_v1 *lum;
3491 struct lov_user_md_v3 *v3 = NULL;
3492 const char *pool_name = NULL;
3497 LASSERT(buf != NULL && buf->lb_buf != NULL);
3500 switch (lum->lmm_magic) {
3501 case LOV_USER_MAGIC_SPECIFIC:
3502 case LOV_USER_MAGIC_V3:
3504 if (v3->lmm_pool_name[0] != '\0')
3505 pool_name = v3->lmm_pool_name;
3507 case LOV_USER_MAGIC_V1:
3508 /* if { size, offset, count } = { 0, -1, 0 } and no pool
3509 * (i.e. all default values specified) then delete default
3510 * striping from dir. */
3512 "set default striping: sz %u # %u offset %d %s %s\n",
3513 (unsigned)lum->lmm_stripe_size,
3514 (unsigned)lum->lmm_stripe_count,
3515 (int)lum->lmm_stripe_offset,
3516 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
3518 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
3519 lum->lmm_stripe_count,
3520 lum->lmm_stripe_offset,
3523 case LOV_USER_MAGIC_COMP_V1:
3527 CERROR("Invalid magic %x\n", lum->lmm_magic);
3532 rc = lod_xattr_del_internal(env, dt, name, th);
3536 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3543 * Set default striping on a directory object.
3545 * Sets specified striping on a directory object unless it matches the default
3546 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3547 * EA. This striping will be used when a new directory is being created in the
3550 * \param[in] env execution environment
3551 * \param[in] dt the striped object
3552 * \param[in] buf buffer with the striping
3553 * \param[in] name name of EA
3554 * \param[in] fl xattr flag (see OSD API description)
3555 * \param[in] th transaction handle
3557 * \retval 0 on success
3558 * \retval negative if failed
3560 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
3561 struct dt_object *dt,
3562 const struct lu_buf *buf,
3563 const char *name, int fl,
3566 struct lmv_user_md_v1 *lum;
3570 LASSERT(buf != NULL && buf->lb_buf != NULL);
3573 CDEBUG(D_OTHER, "set default stripe_count # %u stripe_offset %d\n",
3574 le32_to_cpu(lum->lum_stripe_count),
3575 (int)le32_to_cpu(lum->lum_stripe_offset));
3577 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
3578 le32_to_cpu(lum->lum_stripe_offset)) &&
3579 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
3580 rc = lod_xattr_del_internal(env, dt, name, th);
3584 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3593 * Turn directory into a striped directory.
3595 * During replay the client sends the striping created before MDT
3596 * failure, then the layer above LOD sends this defined striping
3597 * using ->do_xattr_set(), so LOD uses this method to replay creation
3598 * of the stripes. Notice the original information for the striping
3599 * (#stripes, FIDs, etc) was transferred in declare path.
3601 * \param[in] env execution environment
3602 * \param[in] dt the striped object
3603 * \param[in] buf not used currently
3604 * \param[in] name not used currently
3605 * \param[in] fl xattr flag (see OSD API description)
3606 * \param[in] th transaction handle
3608 * \retval 0 on success
3609 * \retval negative if failed
3611 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
3612 const struct lu_buf *buf, const char *name,
3613 int fl, struct thandle *th)
3615 struct lod_object *lo = lod_dt_obj(dt);
3616 struct lod_thread_info *info = lod_env_info(env);
3617 struct lu_attr *attr = &info->lti_attr;
3618 struct dt_object_format *dof = &info->lti_format;
3619 struct lu_buf lmv_buf;
3620 struct lu_buf slave_lmv_buf;
3621 struct lmv_mds_md_v1 *lmm;
3622 struct lmv_mds_md_v1 *slave_lmm = NULL;
3623 struct dt_insert_rec *rec = &info->lti_dt_rec;
3628 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3631 /* The stripes are supposed to be allocated in declare phase,
3632 * if there are no stripes being allocated, it will skip */
3633 if (lo->ldo_dir_stripe_count == 0)
3636 rc = dt_attr_get(env, dt_object_child(dt), attr);
3640 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME |
3641 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
3642 dof->dof_type = DFT_DIR;
3644 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
3647 lmm = lmv_buf.lb_buf;
3649 OBD_ALLOC_PTR(slave_lmm);
3650 if (slave_lmm == NULL)
3653 lod_prep_slave_lmv_md(slave_lmm, lmm);
3654 slave_lmv_buf.lb_buf = slave_lmm;
3655 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
3657 rec->rec_type = S_IFDIR;
3658 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3659 struct dt_object *dto = lo->ldo_stripe[i];
3660 char *stripe_name = info->lti_key;
3661 struct lu_name *sname;
3662 struct linkea_data ldata = { NULL };
3663 struct lu_buf linkea_buf;
3665 /* if it's source stripe of migrating directory, don't create */
3666 if (!((lo->ldo_dir_hash_type & LMV_HASH_FLAG_MIGRATION) &&
3667 i >= lo->ldo_dir_migrate_offset)) {
3668 dt_write_lock(env, dto, MOR_TGT_CHILD);
3669 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
3671 dt_write_unlock(env, dto);
3675 rc = lod_sub_ref_add(env, dto, th);
3676 dt_write_unlock(env, dto);
3680 rec->rec_fid = lu_object_fid(&dto->do_lu);
3681 rc = lod_sub_insert(env, dto,
3682 (const struct dt_rec *)rec,
3683 (const struct dt_key *)dot, th);
3688 rec->rec_fid = lu_object_fid(&dt->do_lu);
3689 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
3690 (const struct dt_key *)dotdot, th);
3694 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
3695 cfs_fail_val != i) {
3696 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
3698 slave_lmm->lmv_master_mdt_index =
3701 slave_lmm->lmv_master_mdt_index =
3704 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
3705 XATTR_NAME_LMV, fl, th);
3710 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
3712 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3713 PFID(lu_object_fid(&dto->do_lu)), i + 1);
3715 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3716 PFID(lu_object_fid(&dto->do_lu)), i);
3718 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
3719 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
3720 sname, lu_object_fid(&dt->do_lu));
3724 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
3725 linkea_buf.lb_len = ldata.ld_leh->leh_len;
3726 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
3727 XATTR_NAME_LINK, 0, th);
3731 rec->rec_fid = lu_object_fid(&dto->do_lu);
3732 rc = lod_sub_insert(env, dt_object_child(dt),
3733 (const struct dt_rec *)rec,
3734 (const struct dt_key *)stripe_name, th);
3738 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
3743 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
3744 rc = lod_sub_xattr_set(env, dt_object_child(dt),
3745 &lmv_buf, XATTR_NAME_LMV, fl, th);
3747 if (slave_lmm != NULL)
3748 OBD_FREE_PTR(slave_lmm);
3754 * Helper function to declare/execute creation of a striped directory
3756 * Called in declare/create object path, prepare striping for a directory
3757 * and prepare defaults data striping for the objects to be created in
3758 * that directory. Notice the function calls "declaration" or "execution"
3759 * methods depending on \a declare param. This is a consequence of the
3760 * current approach while we don't have natural distributed transactions:
3761 * we basically execute non-local updates in the declare phase. So, the
3762 * arguments for the both phases are the same and this is the reason for
3763 * this function to exist.
3765 * \param[in] env execution environment
3766 * \param[in] dt object
3767 * \param[in] attr attributes the stripes will be created with
3768 * \param[in] lmu lmv_user_md if MDT indices are specified
3769 * \param[in] dof format of stripes (see OSD API description)
3770 * \param[in] th transaction handle
3771 * \param[in] declare where to call "declare" or "execute" methods
3773 * \retval 0 on success
3774 * \retval negative if failed
3776 static int lod_dir_striping_create_internal(const struct lu_env *env,
3777 struct dt_object *dt,
3778 struct lu_attr *attr,
3779 const struct lu_buf *lmu,
3780 struct dt_object_format *dof,
3784 struct lod_thread_info *info = lod_env_info(env);
3785 struct lod_object *lo = lod_dt_obj(dt);
3786 const struct lod_default_striping *lds = lo->ldo_def_striping;
3790 LASSERT(ergo(lds != NULL,
3791 lds->lds_def_striping_set ||
3792 lds->lds_dir_def_striping_set));
3794 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
3795 lo->ldo_dir_stripe_offset)) {
3797 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3798 int stripe_count = lo->ldo_dir_stripe_count;
3800 if (info->lti_ea_store_size < sizeof(*v1)) {
3801 rc = lod_ea_store_resize(info, sizeof(*v1));
3804 v1 = info->lti_ea_store;
3807 memset(v1, 0, sizeof(*v1));
3808 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3809 v1->lum_stripe_count = cpu_to_le32(stripe_count);
3810 v1->lum_stripe_offset =
3811 cpu_to_le32(lo->ldo_dir_stripe_offset);
3813 info->lti_buf.lb_buf = v1;
3814 info->lti_buf.lb_len = sizeof(*v1);
3815 lmu = &info->lti_buf;
3819 rc = lod_declare_xattr_set_lmv(env, dt, attr, lmu, dof,
3822 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
3828 /* Transfer default LMV striping from the parent */
3829 if (lds != NULL && lds->lds_dir_def_striping_set &&
3830 !LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
3831 lds->lds_dir_def_stripe_offset)) {
3832 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3834 if (info->lti_ea_store_size < sizeof(*v1)) {
3835 rc = lod_ea_store_resize(info, sizeof(*v1));
3838 v1 = info->lti_ea_store;
3841 memset(v1, 0, sizeof(*v1));
3842 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3843 v1->lum_stripe_count =
3844 cpu_to_le32(lds->lds_dir_def_stripe_count);
3845 v1->lum_stripe_offset =
3846 cpu_to_le32(lds->lds_dir_def_stripe_offset);
3848 cpu_to_le32(lds->lds_dir_def_hash_type);
3850 info->lti_buf.lb_buf = v1;
3851 info->lti_buf.lb_len = sizeof(*v1);
3853 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3854 XATTR_NAME_DEFAULT_LMV,
3857 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
3859 XATTR_NAME_DEFAULT_LMV, 0,
3865 /* Transfer default LOV striping from the parent */
3866 if (lds != NULL && lds->lds_def_striping_set &&
3867 lds->lds_def_comp_cnt != 0) {
3868 struct lov_mds_md *lmm;
3869 int lmm_size = lod_comp_md_size(lo, true);
3871 if (info->lti_ea_store_size < lmm_size) {
3872 rc = lod_ea_store_resize(info, lmm_size);
3876 lmm = info->lti_ea_store;
3878 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
3882 info->lti_buf.lb_buf = lmm;
3883 info->lti_buf.lb_len = lmm_size;
3886 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3887 XATTR_NAME_LOV, 0, th);
3889 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
3890 XATTR_NAME_LOV, 0, th);
3898 static int lod_declare_dir_striping_create(const struct lu_env *env,
3899 struct dt_object *dt,
3900 struct lu_attr *attr,
3902 struct dt_object_format *dof,
3905 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
3909 static int lod_dir_striping_create(const struct lu_env *env,
3910 struct dt_object *dt,
3911 struct lu_attr *attr,
3912 struct dt_object_format *dof,
3915 return lod_dir_striping_create_internal(env, dt, attr, NULL, dof, th,
3920 * Make LOV EA for striped object.
3922 * Generate striping information and store it in the LOV EA of the given
3923 * object. The caller must ensure nobody else is calling the function
3924 * against the object concurrently. The transaction must be started.
3925 * FLDB service must be running as well; it's used to map FID to the target,
3926 * which is stored in LOV EA.
3928 * \param[in] env execution environment for this thread
3929 * \param[in] lo LOD object
3930 * \param[in] th transaction handle
3932 * \retval 0 if LOV EA is stored successfully
3933 * \retval negative error number on failure
3935 static int lod_generate_and_set_lovea(const struct lu_env *env,
3936 struct lod_object *lo,
3939 struct lod_thread_info *info = lod_env_info(env);
3940 struct dt_object *next = dt_object_child(&lo->ldo_obj);
3941 struct lov_mds_md_v1 *lmm;
3947 if (lo->ldo_comp_cnt == 0) {
3948 lod_striping_free(env, lo);
3949 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
3953 lmm_size = lod_comp_md_size(lo, false);
3954 if (info->lti_ea_store_size < lmm_size) {
3955 rc = lod_ea_store_resize(info, lmm_size);
3959 lmm = info->lti_ea_store;
3961 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
3965 info->lti_buf.lb_buf = lmm;
3966 info->lti_buf.lb_len = lmm_size;
3967 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
3968 XATTR_NAME_LOV, 0, th);
3973 * Delete layout component(s)
3975 * \param[in] env execution environment for this thread
3976 * \param[in] dt object
3977 * \param[in] th transaction handle
3979 * \retval 0 on success
3980 * \retval negative error number on failure
3982 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
3985 struct lod_layout_component *lod_comp;
3986 struct lod_object *lo = lod_dt_obj(dt);
3987 struct dt_object *next = dt_object_child(dt);
3988 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3991 LASSERT(lo->ldo_is_composite);
3992 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
3994 left = lo->ldo_comp_cnt;
3995 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3996 lod_comp = &lo->ldo_comp_entries[i];
3998 if (lod_comp->llc_id != LCME_ID_INVAL)
4002 /* Not instantiated component */
4003 if (lod_comp->llc_stripe == NULL)
4006 LASSERT(lod_comp->llc_stripe_count > 0);
4007 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4008 struct dt_object *obj = lod_comp->llc_stripe[j];
4012 rc = lod_sub_destroy(env, obj, th);
4016 lu_object_put(env, &obj->do_lu);
4017 lod_comp->llc_stripe[j] = NULL;
4019 OBD_FREE(lod_comp->llc_stripe, sizeof(struct dt_object *) *
4020 lod_comp->llc_stripes_allocated);
4021 lod_comp->llc_stripe = NULL;
4022 OBD_FREE(lod_comp->llc_ost_indices,
4023 sizeof(__u32) * lod_comp->llc_stripes_allocated);
4024 lod_comp->llc_ost_indices = NULL;
4025 lod_comp->llc_stripes_allocated = 0;
4026 lod_obj_set_pool(lo, i, NULL);
4027 if (lod_comp->llc_ostlist.op_array) {
4028 OBD_FREE(lod_comp->llc_ostlist.op_array,
4029 lod_comp->llc_ostlist.op_size);
4030 lod_comp->llc_ostlist.op_array = NULL;
4031 lod_comp->llc_ostlist.op_size = 0;
4035 LASSERTF(left >= 0 && left < lo->ldo_comp_cnt, "left = %d\n", left);
4037 struct lod_layout_component *comp_array;
4039 OBD_ALLOC(comp_array, sizeof(*comp_array) * left);
4040 if (comp_array == NULL)
4041 GOTO(out, rc = -ENOMEM);
4043 memcpy(&comp_array[0], &lo->ldo_comp_entries[0],
4044 sizeof(*comp_array) * left);
4046 OBD_FREE(lo->ldo_comp_entries,
4047 sizeof(*comp_array) * lo->ldo_comp_cnt);
4048 lo->ldo_comp_entries = comp_array;
4049 lo->ldo_comp_cnt = left;
4051 LASSERT(lo->ldo_mirror_count == 1);
4052 lo->ldo_mirrors[0].lme_end = left - 1;
4053 lod_obj_inc_layout_gen(lo);
4055 lod_free_comp_entries(lo);
4058 LASSERT(dt_object_exists(dt));
4059 rc = dt_attr_get(env, next, attr);
4063 if (attr->la_size > 0) {
4065 attr->la_valid = LA_SIZE;
4066 rc = lod_sub_attr_set(env, next, attr, th);
4071 rc = lod_generate_and_set_lovea(env, lo, th);
4075 lod_striping_free(env, lo);
4080 static int lod_get_default_lov_striping(const struct lu_env *env,
4081 struct lod_object *lo,
4082 struct lod_default_striping *lds);
4084 * Implementation of dt_object_operations::do_xattr_set.
4086 * Sets specified extended attribute on the object. Three types of EAs are
4088 * LOV EA - stores striping for a regular file or default striping (when set
4090 * LMV EA - stores a marker for the striped directories
4091 * DMV EA - stores default directory striping
4093 * When striping is applied to a non-striped existing object (this is called
4094 * late striping), then LOD notices the caller wants to turn the object into a
4095 * striped one. The stripe objects are created and appropriate EA is set:
4096 * LOV EA storing all the stripes directly or LMV EA storing just a small header
4097 * with striping configuration.
4099 * \see dt_object_operations::do_xattr_set() in the API description for details.
4101 static int lod_xattr_set(const struct lu_env *env,
4102 struct dt_object *dt, const struct lu_buf *buf,
4103 const char *name, int fl, struct thandle *th)
4105 struct dt_object *next = dt_object_child(dt);
4109 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4110 strcmp(name, XATTR_NAME_LMV) == 0) {
4111 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
4113 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4114 strncmp(name, XATTR_NAME_LMV, strlen(XATTR_NAME_LMV)) == 0 &&
4115 strlen(name) > strlen(XATTR_NAME_LMV) + 1) {
4116 const char *op = name + strlen(XATTR_NAME_LMV) + 1;
4120 * XATTR_NAME_LMV".add" is never called, but only declared,
4121 * because lod_xattr_set_lmv() will do the addition.
4123 if (strcmp(op, "del") == 0)
4124 rc = lod_dir_layout_delete(env, dt, buf, th);
4125 else if (strcmp(op, "set") == 0)
4126 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LMV,
4130 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4131 strcmp(name, XATTR_NAME_LOV) == 0) {
4132 struct lod_thread_info *info = lod_env_info(env);
4133 struct lod_default_striping *lds = &info->lti_def_striping;
4134 struct lov_user_md_v1 *v1 = buf->lb_buf;
4135 char pool[LOV_MAXPOOLNAME + 1];
4138 /* get existing striping config */
4139 rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds);
4143 memset(pool, 0, sizeof(pool));
4144 if (lds->lds_def_striping_set == 1)
4145 lod_layout_get_pool(lds->lds_def_comp_entries,
4146 lds->lds_def_comp_cnt, pool,
4149 is_del = LOVEA_DELETE_VALUES(v1->lmm_stripe_size,
4150 v1->lmm_stripe_count,
4151 v1->lmm_stripe_offset,
4154 /* Retain the pool name if it is not given */
4155 if (v1->lmm_magic == LOV_USER_MAGIC_V1 && pool[0] != '\0' &&
4157 struct lod_thread_info *info = lod_env_info(env);
4158 struct lov_user_md_v3 *v3 = info->lti_ea_store;
4160 memset(v3, 0, sizeof(*v3));
4161 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4162 v3->lmm_pattern = cpu_to_le32(v1->lmm_pattern);
4163 v3->lmm_stripe_count =
4164 cpu_to_le32(v1->lmm_stripe_count);
4165 v3->lmm_stripe_offset =
4166 cpu_to_le32(v1->lmm_stripe_offset);
4167 v3->lmm_stripe_size = cpu_to_le32(v1->lmm_stripe_size);
4169 strlcpy(v3->lmm_pool_name, pool,
4170 sizeof(v3->lmm_pool_name));
4172 info->lti_buf.lb_buf = v3;
4173 info->lti_buf.lb_len = sizeof(*v3);
4174 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4177 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name,
4181 if (lds->lds_def_striping_set == 1 &&
4182 lds->lds_def_comp_entries != NULL)
4183 lod_free_def_comp_entries(lds);
4186 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4187 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
4189 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
4192 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
4193 (!strcmp(name, XATTR_NAME_LOV) ||
4194 !strncmp(name, XATTR_LUSTRE_LOV,
4195 strlen(XATTR_LUSTRE_LOV)))) {
4196 /* in case of lov EA swap, just set it
4197 * if not, it is a replay so check striping match what we
4198 * already have during req replay, declare_xattr_set()
4199 * defines striping, then create() does the work */
4200 if (fl & LU_XATTR_REPLACE) {
4201 /* free stripes, then update disk */
4202 lod_striping_free(env, lod_dt_obj(dt));
4204 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
4205 } else if (dt_object_remote(dt)) {
4206 /* This only happens during migration, see
4207 * mdd_migrate_create(), in which Master MDT will
4208 * create a remote target object, and only set
4209 * (migrating) stripe EA on the remote object,
4210 * and does not need creating each stripes. */
4211 rc = lod_sub_xattr_set(env, next, buf, name,
4213 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
4214 /* delete component(s) */
4215 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
4216 rc = lod_layout_del(env, dt, th);
4219 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
4220 * it's going to create create file with specified
4221 * component(s), the striping must have not being
4222 * cached in this case;
4224 * Otherwise, it's going to add/change component(s) to
4225 * an existing file, the striping must have been cached
4228 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
4229 !strcmp(name, XATTR_NAME_LOV),
4230 !lod_dt_obj(dt)->ldo_comp_cached));
4232 rc = lod_striped_create(env, dt, NULL, NULL, th);
4235 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4236 rc = lod_replace_parent_fid(env, dt, buf, th, false);
4241 /* then all other xattr */
4242 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4248 * Implementation of dt_object_operations::do_declare_xattr_del.
4250 * \see dt_object_operations::do_declare_xattr_del() in the API description
4253 static int lod_declare_xattr_del(const struct lu_env *env,
4254 struct dt_object *dt, const char *name,
4257 struct lod_object *lo = lod_dt_obj(dt);
4258 struct dt_object *next = dt_object_child(dt);
4263 rc = lod_sub_declare_xattr_del(env, next, name, th);
4267 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4270 /* set xattr to each stripes, if needed */
4271 rc = lod_striping_load(env, lo);
4275 if (lo->ldo_dir_stripe_count == 0)
4278 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4279 struct dt_object *dto = lo->ldo_stripe[i];
4282 rc = lod_sub_declare_xattr_del(env, dto, name, th);
4291 * Implementation of dt_object_operations::do_xattr_del.
4293 * If EA storing a regular striping is being deleted, then release
4294 * all the references to the stripe objects in core.
4296 * \see dt_object_operations::do_xattr_del() in the API description for details.
4298 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
4299 const char *name, struct thandle *th)
4301 struct dt_object *next = dt_object_child(dt);
4302 struct lod_object *lo = lod_dt_obj(dt);
4307 if (!strcmp(name, XATTR_NAME_LOV) || !strcmp(name, XATTR_NAME_LMV))
4308 lod_striping_free(env, lod_dt_obj(dt));
4310 rc = lod_sub_xattr_del(env, next, name, th);
4311 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
4314 if (lo->ldo_dir_stripe_count == 0)
4317 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4318 struct dt_object *dto = lo->ldo_stripe[i];
4322 rc = lod_sub_xattr_del(env, dto, name, th);
4331 * Implementation of dt_object_operations::do_xattr_list.
4333 * \see dt_object_operations::do_xattr_list() in the API description
4336 static int lod_xattr_list(const struct lu_env *env,
4337 struct dt_object *dt, const struct lu_buf *buf)
4339 return dt_xattr_list(env, dt_object_child(dt), buf);
4342 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
4344 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
4348 * Copy OST list from layout provided by user.
4350 * \param[in] lod_comp layout_component to be filled
4351 * \param[in] v3 LOV EA V3 user data
4353 * \retval 0 on success
4354 * \retval negative if failed
4356 int lod_comp_copy_ost_lists(struct lod_layout_component *lod_comp,
4357 struct lov_user_md_v3 *v3)
4363 if (v3->lmm_stripe_offset == LOV_OFFSET_DEFAULT)
4364 v3->lmm_stripe_offset = v3->lmm_objects[0].l_ost_idx;
4366 if (lod_comp->llc_ostlist.op_array) {
4367 if (lod_comp->llc_ostlist.op_size >=
4368 v3->lmm_stripe_count * sizeof(__u32)) {
4369 lod_comp->llc_ostlist.op_count =
4370 v3->lmm_stripe_count;
4373 OBD_FREE(lod_comp->llc_ostlist.op_array,
4374 lod_comp->llc_ostlist.op_size);
4377 /* copy ost list from lmm */
4378 lod_comp->llc_ostlist.op_count = v3->lmm_stripe_count;
4379 lod_comp->llc_ostlist.op_size = v3->lmm_stripe_count * sizeof(__u32);
4380 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
4381 lod_comp->llc_ostlist.op_size);
4382 if (!lod_comp->llc_ostlist.op_array)
4385 for (j = 0; j < v3->lmm_stripe_count; j++) {
4386 lod_comp->llc_ostlist.op_array[j] =
4387 v3->lmm_objects[j].l_ost_idx;
4395 * Get default striping.
4397 * \param[in] env execution environment
4398 * \param[in] lo object
4399 * \param[out] lds default striping
4401 * \retval 0 on success
4402 * \retval negative if failed
4404 static int lod_get_default_lov_striping(const struct lu_env *env,
4405 struct lod_object *lo,
4406 struct lod_default_striping *lds)
4408 struct lod_thread_info *info = lod_env_info(env);
4409 struct lov_user_md_v1 *v1 = NULL;
4410 struct lov_user_md_v3 *v3 = NULL;
4411 struct lov_comp_md_v1 *comp_v1 = NULL;
4418 lds->lds_def_striping_set = 0;
4420 rc = lod_get_lov_ea(env, lo);
4424 if (rc < (typeof(rc))sizeof(struct lov_user_md))
4427 v1 = info->lti_ea_store;
4428 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
4429 lustre_swab_lov_user_md_v1(v1);
4430 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
4431 v3 = (struct lov_user_md_v3 *)v1;
4432 lustre_swab_lov_user_md_v3(v3);
4433 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_SPECIFIC)) {
4434 v3 = (struct lov_user_md_v3 *)v1;
4435 lustre_swab_lov_user_md_v3(v3);
4436 lustre_swab_lov_user_md_objects(v3->lmm_objects,
4437 v3->lmm_stripe_count);
4438 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
4439 comp_v1 = (struct lov_comp_md_v1 *)v1;
4440 lustre_swab_lov_comp_md_v1(comp_v1);
4443 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1 &&
4444 v1->lmm_magic != LOV_MAGIC_COMP_V1 &&
4445 v1->lmm_magic != LOV_USER_MAGIC_SPECIFIC)
4448 if (v1->lmm_magic == LOV_MAGIC_COMP_V1) {
4449 comp_v1 = (struct lov_comp_md_v1 *)v1;
4450 comp_cnt = comp_v1->lcm_entry_count;
4453 mirror_cnt = comp_v1->lcm_mirror_count + 1;
4461 /* realloc default comp entries if necessary */
4462 rc = lod_def_striping_comp_resize(lds, comp_cnt);
4466 lds->lds_def_comp_cnt = comp_cnt;
4467 lds->lds_def_striping_is_composite = composite;
4468 lds->lds_def_mirror_cnt = mirror_cnt;
4470 for (i = 0; i < comp_cnt; i++) {
4471 struct lod_layout_component *lod_comp;
4474 lod_comp = &lds->lds_def_comp_entries[i];
4476 * reset lod_comp values, llc_stripes is always NULL in
4477 * the default striping template, llc_pool will be reset
4480 memset(lod_comp, 0, offsetof(typeof(*lod_comp), llc_pool));
4483 v1 = (struct lov_user_md *)((char *)comp_v1 +
4484 comp_v1->lcm_entries[i].lcme_offset);
4485 lod_comp->llc_extent =
4486 comp_v1->lcm_entries[i].lcme_extent;
4487 /* We only inherit certain flags from the layout */
4488 lod_comp->llc_flags =
4489 comp_v1->lcm_entries[i].lcme_flags &
4490 LCME_TEMPLATE_FLAGS;
4493 if (v1->lmm_pattern != LOV_PATTERN_RAID0 &&
4494 v1->lmm_pattern != LOV_PATTERN_MDT &&
4495 v1->lmm_pattern != 0) {
4496 lod_free_def_comp_entries(lds);
4500 CDEBUG(D_LAYOUT, DFID" stripe_count=%d stripe_size=%d "
4501 "stripe_offset=%d\n",
4502 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
4503 (int)v1->lmm_stripe_count, (int)v1->lmm_stripe_size,
4504 (int)v1->lmm_stripe_offset);
4506 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
4507 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
4508 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
4509 lod_comp->llc_pattern = v1->lmm_pattern;
4512 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
4513 /* XXX: sanity check here */
4514 v3 = (struct lov_user_md_v3 *) v1;
4515 if (v3->lmm_pool_name[0] != '\0')
4516 pool = v3->lmm_pool_name;
4518 lod_set_def_pool(lds, i, pool);
4519 if (v1->lmm_magic == LOV_USER_MAGIC_SPECIFIC) {
4520 v3 = (struct lov_user_md_v3 *)v1;
4521 rc = lod_comp_copy_ost_lists(lod_comp, v3);
4524 } else if (lod_comp->llc_ostlist.op_array &&
4525 lod_comp->llc_ostlist.op_count) {
4526 for (j = 0; j < lod_comp->llc_ostlist.op_count; j++)
4527 lod_comp->llc_ostlist.op_array[j] = -1;
4528 lod_comp->llc_ostlist.op_count = 0;
4532 lds->lds_def_striping_set = 1;
4537 * Get default directory striping.
4539 * \param[in] env execution environment
4540 * \param[in] lo object
4541 * \param[out] lds default striping
4543 * \retval 0 on success
4544 * \retval negative if failed
4546 static int lod_get_default_lmv_striping(const struct lu_env *env,
4547 struct lod_object *lo,
4548 struct lod_default_striping *lds)
4550 struct lod_thread_info *info = lod_env_info(env);
4551 struct lmv_user_md_v1 *v1 = NULL;
4555 lds->lds_dir_def_striping_set = 0;
4556 rc = lod_get_default_lmv_ea(env, lo);
4560 if (rc < (typeof(rc))sizeof(struct lmv_user_md))
4563 v1 = info->lti_ea_store;
4565 lds->lds_dir_def_stripe_count = le32_to_cpu(v1->lum_stripe_count);
4566 lds->lds_dir_def_stripe_offset = le32_to_cpu(v1->lum_stripe_offset);
4567 lds->lds_dir_def_hash_type = le32_to_cpu(v1->lum_hash_type);
4568 lds->lds_dir_def_striping_set = 1;
4574 * Get default striping in the object.
4576 * Get object default striping and default directory striping.
4578 * \param[in] env execution environment
4579 * \param[in] lo object
4580 * \param[out] lds default striping
4582 * \retval 0 on success
4583 * \retval negative if failed
4585 static int lod_get_default_striping(const struct lu_env *env,
4586 struct lod_object *lo,
4587 struct lod_default_striping *lds)
4591 rc = lod_get_default_lov_striping(env, lo, lds);
4592 rc1 = lod_get_default_lmv_striping(env, lo, lds);
4593 if (rc == 0 && rc1 < 0)
4600 * Apply default striping on object.
4602 * If object striping pattern is not set, set to the one in default striping.
4603 * The default striping is from parent or fs.
4605 * \param[in] lo new object
4606 * \param[in] lds default striping
4607 * \param[in] mode new object's mode
4609 static void lod_striping_from_default(struct lod_object *lo,
4610 const struct lod_default_striping *lds,
4613 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
4614 struct lov_desc *desc = &d->lod_desc;
4617 if (lds->lds_def_striping_set && S_ISREG(mode)) {
4618 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
4619 lds->lds_def_comp_cnt);
4623 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
4624 if (lds->lds_def_mirror_cnt > 1)
4625 lo->ldo_flr_state = LCM_FL_RDONLY;
4627 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4628 struct lod_layout_component *obj_comp =
4629 &lo->ldo_comp_entries[i];
4630 struct lod_layout_component *def_comp =
4631 &lds->lds_def_comp_entries[i];
4633 CDEBUG(D_LAYOUT, "Inherit from default: flags=%#x "
4634 "size=%hu nr=%u offset=%u pattern=%#x pool=%s\n",
4635 def_comp->llc_flags,
4636 def_comp->llc_stripe_size,
4637 def_comp->llc_stripe_count,
4638 def_comp->llc_stripe_offset,
4639 def_comp->llc_pattern,
4640 def_comp->llc_pool ?: "");
4642 *obj_comp = *def_comp;
4643 if (def_comp->llc_pool != NULL) {
4644 /* pointer was copied from def_comp */
4645 obj_comp->llc_pool = NULL;
4646 lod_obj_set_pool(lo, i, def_comp->llc_pool);
4650 if (def_comp->llc_ostlist.op_array &&
4651 def_comp->llc_ostlist.op_count) {
4652 OBD_ALLOC(obj_comp->llc_ostlist.op_array,
4653 obj_comp->llc_ostlist.op_size);
4654 if (!obj_comp->llc_ostlist.op_array)
4656 memcpy(obj_comp->llc_ostlist.op_array,
4657 def_comp->llc_ostlist.op_array,
4658 obj_comp->llc_ostlist.op_size);
4659 } else if (def_comp->llc_ostlist.op_array) {
4660 obj_comp->llc_ostlist.op_array = NULL;
4664 * Don't initialize these fields for plain layout
4665 * (v1/v3) here, they are inherited in the order of
4666 * 'parent' -> 'fs default (root)' -> 'global default
4667 * values for stripe_count & stripe_size'.
4669 * see lod_ah_init().
4671 if (!lo->ldo_is_composite)
4674 lod_adjust_stripe_info(obj_comp, desc);
4676 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
4677 if (lo->ldo_dir_stripe_count == 0)
4678 lo->ldo_dir_stripe_count =
4679 lds->lds_dir_def_stripe_count;
4680 if (lo->ldo_dir_stripe_offset == -1)
4681 lo->ldo_dir_stripe_offset =
4682 lds->lds_dir_def_stripe_offset;
4683 if (lo->ldo_dir_hash_type == 0)
4684 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
4686 CDEBUG(D_LAYOUT, "striping from default dir: count:%hu, "
4687 "offset:%u, hash_type:%u\n",
4688 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
4689 lo->ldo_dir_hash_type);
4693 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root)
4695 struct lod_layout_component *lod_comp;
4697 if (lo->ldo_comp_cnt == 0)
4700 if (lo->ldo_is_composite)
4703 lod_comp = &lo->ldo_comp_entries[0];
4705 if (lod_comp->llc_stripe_count <= 0 ||
4706 lod_comp->llc_stripe_size <= 0)
4709 if (from_root && (lod_comp->llc_pool == NULL ||
4710 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
4717 * Implementation of dt_object_operations::do_ah_init.
4719 * This method is used to make a decision on the striping configuration for the
4720 * object being created. It can be taken from the \a parent object if it exists,
4721 * or filesystem's default. The resulting configuration (number of stripes,
4722 * stripe size/offset, pool name, etc) is stored in the object itself and will
4723 * be used by the methods like ->doo_declare_create().
4725 * \see dt_object_operations::do_ah_init() in the API description for details.
4727 static void lod_ah_init(const struct lu_env *env,
4728 struct dt_allocation_hint *ah,
4729 struct dt_object *parent,
4730 struct dt_object *child,
4733 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
4734 struct lod_thread_info *info = lod_env_info(env);
4735 struct lod_default_striping *lds = &info->lti_def_striping;
4736 struct dt_object *nextp = NULL;
4737 struct dt_object *nextc;
4738 struct lod_object *lp = NULL;
4739 struct lod_object *lc;
4740 struct lov_desc *desc;
4741 struct lod_layout_component *lod_comp;
4747 if (likely(parent)) {
4748 nextp = dt_object_child(parent);
4749 lp = lod_dt_obj(parent);
4752 nextc = dt_object_child(child);
4753 lc = lod_dt_obj(child);
4755 LASSERT(!lod_obj_is_striped(child));
4756 /* default layout template may have been set on the regular file
4757 * when this is called from mdd_create_data() */
4758 if (S_ISREG(child_mode))
4759 lod_free_comp_entries(lc);
4761 if (!dt_object_exists(nextc))
4762 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
4764 if (S_ISDIR(child_mode)) {
4765 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
4767 /* other default values are 0 */
4768 lc->ldo_dir_stripe_offset = -1;
4770 /* get default striping from parent object */
4771 if (likely(lp != NULL))
4772 lod_get_default_striping(env, lp, lds);
4774 /* set child default striping info, default value is NULL */
4775 if (lds->lds_def_striping_set || lds->lds_dir_def_striping_set)
4776 lc->ldo_def_striping = lds;
4778 /* It should always honour the specified stripes */
4779 /* Note: old client (< 2.7)might also do lfs mkdir, whose EA
4780 * will have old magic. In this case, we should ignore the
4781 * stripe count and try to create dir by default stripe.
4783 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
4784 (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
4785 le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC)) {
4786 lc->ldo_dir_stripe_count =
4787 le32_to_cpu(lum1->lum_stripe_count);
4788 lc->ldo_dir_stripe_offset =
4789 le32_to_cpu(lum1->lum_stripe_offset);
4790 lc->ldo_dir_hash_type =
4791 le32_to_cpu(lum1->lum_hash_type);
4793 "set dirstripe: count %hu, offset %d, hash %u\n",
4794 lc->ldo_dir_stripe_count,
4795 (int)lc->ldo_dir_stripe_offset,
4796 lc->ldo_dir_hash_type);
4798 /* transfer defaults LMV to new directory */
4799 lod_striping_from_default(lc, lds, child_mode);
4801 /* set count 0 to create normal directory */
4802 if (lc->ldo_dir_stripe_count == 1)
4803 lc->ldo_dir_stripe_count = 0;
4806 /* shrink the stripe_count to the avaible MDT count */
4807 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
4808 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)) {
4809 lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
4810 if (lc->ldo_dir_stripe_count == 1)
4811 lc->ldo_dir_stripe_count = 0;
4814 CDEBUG(D_INFO, "final dir stripe [%hu %d %u]\n",
4815 lc->ldo_dir_stripe_count,
4816 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
4821 /* child object regular file*/
4823 if (!lod_object_will_be_striped(S_ISREG(child_mode),
4824 lu_object_fid(&child->do_lu)))
4827 /* If object is going to be striped over OSTs, transfer default
4828 * striping information to the child, so that we can use it
4829 * during declaration and creation.
4831 * Try from the parent first.
4833 if (likely(lp != NULL)) {
4834 rc = lod_get_default_lov_striping(env, lp, lds);
4836 lod_striping_from_default(lc, lds, child_mode);
4839 /* Initialize lod_device::lod_md_root object reference */
4840 if (d->lod_md_root == NULL) {
4841 struct dt_object *root;
4842 struct lod_object *lroot;
4844 lu_root_fid(&info->lti_fid);
4845 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
4846 if (!IS_ERR(root)) {
4847 lroot = lod_dt_obj(root);
4849 spin_lock(&d->lod_lock);
4850 if (d->lod_md_root != NULL)
4851 dt_object_put(env, &d->lod_md_root->ldo_obj);
4852 d->lod_md_root = lroot;
4853 spin_unlock(&d->lod_lock);
4857 /* try inherit layout from the root object (fs default) when:
4858 * - parent does not have default layout; or
4859 * - parent has plain(v1/v3) default layout, and some attributes
4860 * are not specified in the default layout;
4862 if (d->lod_md_root != NULL && lod_need_inherit_more(lc, true)) {
4863 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds);
4866 if (lc->ldo_comp_cnt == 0) {
4867 lod_striping_from_default(lc, lds, child_mode);
4868 } else if (!lds->lds_def_striping_is_composite) {
4869 struct lod_layout_component *def_comp;
4871 LASSERT(!lc->ldo_is_composite);
4872 lod_comp = &lc->ldo_comp_entries[0];
4873 def_comp = &lds->lds_def_comp_entries[0];
4875 if (lod_comp->llc_stripe_count <= 0)
4876 lod_comp->llc_stripe_count =
4877 def_comp->llc_stripe_count;
4878 if (lod_comp->llc_stripe_size <= 0)
4879 lod_comp->llc_stripe_size =
4880 def_comp->llc_stripe_size;
4881 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT)
4882 lod_comp->llc_stripe_offset =
4883 def_comp->llc_stripe_offset;
4884 if (lod_comp->llc_pool == NULL)
4885 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
4890 * fs default striping may not be explicitly set, or historically set
4891 * in config log, use them.
4893 if (lod_need_inherit_more(lc, false)) {
4894 if (lc->ldo_comp_cnt == 0) {
4895 rc = lod_alloc_comp_entries(lc, 0, 1);
4897 /* fail to allocate memory, will create a
4898 * non-striped file. */
4900 lc->ldo_is_composite = 0;
4901 lod_comp = &lc->ldo_comp_entries[0];
4902 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4904 LASSERT(!lc->ldo_is_composite);
4905 lod_comp = &lc->ldo_comp_entries[0];
4906 desc = &d->lod_desc;
4907 lod_adjust_stripe_info(lod_comp, desc);
4913 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
4915 * Size initialization on late striping.
4917 * Propagate the size of a truncated object to a deferred striping.
4918 * This function handles a special case when truncate was done on a
4919 * non-striped object and now while the striping is being created
4920 * we can't lose that size, so we have to propagate it to the stripes
4923 * \param[in] env execution environment
4924 * \param[in] dt object
4925 * \param[in] th transaction handle
4927 * \retval 0 on success
4928 * \retval negative if failed
4930 static int lod_declare_init_size(const struct lu_env *env,
4931 struct dt_object *dt, struct thandle *th)
4933 struct dt_object *next = dt_object_child(dt);
4934 struct lod_object *lo = lod_dt_obj(dt);
4935 struct dt_object **objects = NULL;
4936 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4937 uint64_t size, offs;
4938 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
4939 struct lu_extent size_ext;
4942 if (!lod_obj_is_striped(dt))
4945 rc = dt_attr_get(env, next, attr);
4946 LASSERT(attr->la_valid & LA_SIZE);
4950 size = attr->la_size;
4954 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
4955 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4956 struct lod_layout_component *lod_comp;
4957 struct lu_extent *extent;
4959 lod_comp = &lo->ldo_comp_entries[i];
4961 if (lod_comp->llc_stripe == NULL)
4964 extent = &lod_comp->llc_extent;
4965 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
4966 if (!lo->ldo_is_composite ||
4967 lu_extent_is_overlapped(extent, &size_ext)) {
4968 objects = lod_comp->llc_stripe;
4969 stripe_count = lod_comp->llc_stripe_count;
4970 stripe_size = lod_comp->llc_stripe_size;
4973 if (stripe_count == 0)
4976 LASSERT(objects != NULL && stripe_size != 0);
4977 /* ll_do_div64(a, b) returns a % b, and a = a / b */
4978 ll_do_div64(size, (__u64)stripe_size);
4979 stripe = ll_do_div64(size, (__u64)stripe_count);
4980 LASSERT(objects[stripe] != NULL);
4982 size = size * stripe_size;
4983 offs = attr->la_size;
4984 size += ll_do_div64(offs, stripe_size);
4986 attr->la_valid = LA_SIZE;
4987 attr->la_size = size;
4989 rc = lod_sub_declare_attr_set(env, objects[stripe],
4998 * Declare creation of striped object.
5000 * The function declares creation stripes for a regular object. The function
5001 * also declares whether the stripes will be created with non-zero size if
5002 * previously size was set non-zero on the master object. If object \a dt is
5003 * not local, then only fully defined striping can be applied in \a lovea.
5004 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
5007 * \param[in] env execution environment
5008 * \param[in] dt object
5009 * \param[in] attr attributes the stripes will be created with
5010 * \param[in] lovea a buffer containing striping description
5011 * \param[in] th transaction handle
5013 * \retval 0 on success
5014 * \retval negative if failed
5016 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
5017 struct lu_attr *attr,
5018 const struct lu_buf *lovea, struct thandle *th)
5020 struct lod_thread_info *info = lod_env_info(env);
5021 struct dt_object *next = dt_object_child(dt);
5022 struct lod_object *lo = lod_dt_obj(dt);
5026 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
5027 GOTO(out, rc = -ENOMEM);
5029 if (!dt_object_remote(next)) {
5030 /* choose OST and generate appropriate objects */
5031 rc = lod_prepare_create(env, lo, attr, lovea, th);
5036 * declare storage for striping data
5038 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
5040 /* LOD can not choose OST objects for remote objects, i.e.
5041 * stripes must be ready before that. Right now, it can only
5042 * happen during migrate, i.e. migrate process needs to create
5043 * remote regular file (mdd_migrate_create), then the migrate
5044 * process will provide stripeEA. */
5045 LASSERT(lovea != NULL);
5046 info->lti_buf = *lovea;
5049 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
5050 XATTR_NAME_LOV, 0, th);
5055 * if striping is created with local object's size > 0,
5056 * we have to propagate this size to specific object
5057 * the case is possible only when local object was created previously
5059 if (dt_object_exists(next))
5060 rc = lod_declare_init_size(env, dt, th);
5063 /* failed to create striping or to set initial size, let's reset
5064 * config so that others don't get confused */
5066 lod_striping_free(env, lo);
5072 * Implementation of dt_object_operations::do_declare_create.
5074 * The method declares creation of a new object. If the object will be striped,
5075 * then helper functions are called to find FIDs for the stripes, declare
5076 * creation of the stripes and declare initialization of the striping
5077 * information to be stored in the master object.
5079 * \see dt_object_operations::do_declare_create() in the API description
5082 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
5083 struct lu_attr *attr,
5084 struct dt_allocation_hint *hint,
5085 struct dt_object_format *dof, struct thandle *th)
5087 struct dt_object *next = dt_object_child(dt);
5088 struct lod_object *lo = lod_dt_obj(dt);
5097 * first of all, we declare creation of local object
5099 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
5104 * it's lod_ah_init() that has decided the object will be striped
5106 if (dof->dof_type == DFT_REGULAR) {
5107 /* callers don't want stripes */
5108 /* XXX: all tricky interactions with ->ah_make_hint() decided
5109 * to use striping, then ->declare_create() behaving differently
5110 * should be cleaned */
5111 if (dof->u.dof_reg.striped != 0)
5112 rc = lod_declare_striped_create(env, dt, attr,
5114 } else if (dof->dof_type == DFT_DIR) {
5115 struct seq_server_site *ss;
5116 struct lu_buf buf = { NULL };
5117 struct lu_buf *lmu = NULL;
5119 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
5121 /* If the parent has default stripeEA, and client
5122 * did not find it before sending create request,
5123 * then MDT will return -EREMOTE, and client will
5124 * retrieve the default stripeEA and re-create the
5127 * Note: if dah_eadata != NULL, it means creating the
5128 * striped directory with specified stripeEA, then it
5129 * should ignore the default stripeEA */
5130 if (hint != NULL && hint->dah_eadata == NULL) {
5131 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
5132 GOTO(out, rc = -EREMOTE);
5134 if (lo->ldo_dir_stripe_offset == -1) {
5135 /* child and parent should be in the same MDT */
5136 if (hint->dah_parent != NULL &&
5137 dt_object_remote(hint->dah_parent))
5138 GOTO(out, rc = -EREMOTE);
5139 } else if (lo->ldo_dir_stripe_offset !=
5141 struct lod_device *lod;
5142 struct lod_tgt_descs *ltd;
5143 struct lod_tgt_desc *tgt = NULL;
5144 bool found_mdt = false;
5147 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5148 ltd = &lod->lod_mdt_descs;
5149 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
5150 tgt = LTD_TGT(ltd, i);
5151 if (tgt->ltd_index ==
5152 lo->ldo_dir_stripe_offset) {
5158 /* If the MDT indicated by stripe_offset can be
5159 * found, then tell client to resend the create
5160 * request to the correct MDT, otherwise return
5161 * error to client */
5163 GOTO(out, rc = -EREMOTE);
5165 GOTO(out, rc = -EINVAL);
5167 } else if (hint && hint->dah_eadata) {
5169 lmu->lb_buf = (void *)hint->dah_eadata;
5170 lmu->lb_len = hint->dah_eadata_len;
5173 rc = lod_declare_dir_striping_create(env, dt, attr, lmu, dof,
5177 /* failed to create striping or to set initial size, let's reset
5178 * config so that others don't get confused */
5180 lod_striping_free(env, lo);
5185 * Generate component ID for new created component.
5187 * \param[in] lo LOD object
5188 * \param[in] comp_idx index of ldo_comp_entries
5190 * \retval component ID on success
5191 * \retval LCME_ID_INVAL on failure
5193 static __u32 lod_gen_component_id(struct lod_object *lo,
5194 int mirror_id, int comp_idx)
5196 struct lod_layout_component *lod_comp;
5197 __u32 id, start, end;
5200 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
5202 lod_obj_inc_layout_gen(lo);
5203 id = lo->ldo_layout_gen;
5204 if (likely(id <= SEQ_ID_MAX))
5205 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
5207 /* Layout generation wraps, need to check collisions. */
5208 start = id & SEQ_ID_MASK;
5211 for (id = start; id <= end; id++) {
5212 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5213 lod_comp = &lo->ldo_comp_entries[i];
5214 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
5217 /* Found the ununsed ID */
5218 if (i == lo->ldo_comp_cnt)
5219 RETURN(pflr_id(mirror_id, id));
5221 if (end == LCME_ID_MAX) {
5223 end = min(lo->ldo_layout_gen & LCME_ID_MASK,
5224 (__u32)(LCME_ID_MAX - 1));
5228 RETURN(LCME_ID_INVAL);
5232 * Creation of a striped regular object.
5234 * The function is called to create the stripe objects for a regular
5235 * striped file. This can happen at the initial object creation or
5236 * when the caller asks LOD to do so using ->do_xattr_set() method
5237 * (so called late striping). Notice all the information are already
5238 * prepared in the form of the list of objects (ldo_stripe field).
5239 * This is done during declare phase.
5241 * \param[in] env execution environment
5242 * \param[in] dt object
5243 * \param[in] attr attributes the stripes will be created with
5244 * \param[in] dof format of stripes (see OSD API description)
5245 * \param[in] th transaction handle
5247 * \retval 0 on success
5248 * \retval negative if failed
5250 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
5251 struct lu_attr *attr, struct dt_object_format *dof,
5254 struct lod_layout_component *lod_comp;
5255 struct lod_object *lo = lod_dt_obj(dt);
5260 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
5262 mirror_id = 0; /* non-flr file's mirror_id is 0 */
5263 if (lo->ldo_mirror_count > 1) {
5264 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5265 lod_comp = &lo->ldo_comp_entries[i];
5266 if (lod_comp->llc_id != LCME_ID_INVAL &&
5267 mirror_id_of(lod_comp->llc_id) > mirror_id)
5268 mirror_id = mirror_id_of(lod_comp->llc_id);
5272 /* create all underlying objects */
5273 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5274 lod_comp = &lo->ldo_comp_entries[i];
5276 if (lod_comp->llc_id == LCME_ID_INVAL) {
5277 /* only the component of FLR layout with more than 1
5278 * mirror has mirror ID in its component ID.
5280 if (lod_comp->llc_extent.e_start == 0 &&
5281 lo->ldo_mirror_count > 1)
5284 lod_comp->llc_id = lod_gen_component_id(lo,
5286 if (lod_comp->llc_id == LCME_ID_INVAL)
5287 GOTO(out, rc = -ERANGE);
5290 if (lod_comp_inited(lod_comp))
5293 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
5294 lod_comp_set_init(lod_comp);
5296 if (lov_pattern(lod_comp->llc_pattern) == LOV_PATTERN_MDT)
5297 lod_comp_set_init(lod_comp);
5299 if (lod_comp->llc_stripe == NULL)
5302 LASSERT(lod_comp->llc_stripe_count);
5303 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
5304 struct dt_object *object = lod_comp->llc_stripe[j];
5305 LASSERT(object != NULL);
5306 rc = lod_sub_create(env, object, attr, NULL, dof, th);
5310 lod_comp_set_init(lod_comp);
5313 rc = lod_fill_mirrors(lo);
5317 rc = lod_generate_and_set_lovea(env, lo, th);
5321 lo->ldo_comp_cached = 1;
5325 lod_striping_free(env, lo);
5330 * Implementation of dt_object_operations::do_create.
5332 * If any of preceeding methods (like ->do_declare_create(),
5333 * ->do_ah_init(), etc) chose to create a striped object,
5334 * then this method will create the master and the stripes.
5336 * \see dt_object_operations::do_create() in the API description for details.
5338 static int lod_create(const struct lu_env *env, struct dt_object *dt,
5339 struct lu_attr *attr, struct dt_allocation_hint *hint,
5340 struct dt_object_format *dof, struct thandle *th)
5345 /* create local object */
5346 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
5350 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
5351 lod_obj_is_striped(dt) && dof->u.dof_reg.striped != 0) {
5352 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
5353 rc = lod_striped_create(env, dt, attr, dof, th);
5360 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
5361 struct dt_object *dt, struct thandle *th,
5362 int comp_idx, int stripe_idx,
5363 struct lod_obj_stripe_cb_data *data)
5365 if (data->locd_declare)
5366 return lod_sub_declare_destroy(env, dt, th);
5367 else if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
5368 stripe_idx == cfs_fail_val)
5369 return lod_sub_destroy(env, dt, th);
5375 * Implementation of dt_object_operations::do_declare_destroy.
5377 * If the object is a striped directory, then the function declares reference
5378 * removal from the master object (this is an index) to the stripes and declares
5379 * destroy of all the stripes. In all the cases, it declares an intention to
5380 * destroy the object itself.
5382 * \see dt_object_operations::do_declare_destroy() in the API description
5385 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
5388 struct dt_object *next = dt_object_child(dt);
5389 struct lod_object *lo = lod_dt_obj(dt);
5390 struct lod_thread_info *info = lod_env_info(env);
5391 char *stripe_name = info->lti_key;
5396 * load striping information, notice we don't do this when object
5397 * is being initialized as we don't need this information till
5398 * few specific cases like destroy, chown
5400 rc = lod_striping_load(env, lo);
5404 /* declare destroy for all underlying objects */
5405 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5406 rc = next->do_ops->do_index_try(env, next,
5407 &dt_directory_features);
5411 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5412 rc = lod_sub_declare_ref_del(env, next, th);
5416 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
5417 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
5419 rc = lod_sub_declare_delete(env, next,
5420 (const struct dt_key *)stripe_name, th);
5427 * we declare destroy for the local object
5429 rc = lod_sub_declare_destroy(env, next, th);
5433 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
5434 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
5437 if (!lod_obj_is_striped(dt))
5440 /* declare destroy all striped objects */
5441 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5442 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5443 if (lo->ldo_stripe[i] == NULL)
5446 rc = lod_sub_declare_ref_del(env, lo->ldo_stripe[i],
5449 rc = lod_sub_declare_destroy(env, lo->ldo_stripe[i],
5455 struct lod_obj_stripe_cb_data data = { { 0 } };
5457 data.locd_declare = true;
5458 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
5459 rc = lod_obj_for_each_stripe(env, lo, th, &data);
5466 * Implementation of dt_object_operations::do_destroy.
5468 * If the object is a striped directory, then the function removes references
5469 * from the master object (this is an index) to the stripes and destroys all
5470 * the stripes. In all the cases, the function destroys the object itself.
5472 * \see dt_object_operations::do_destroy() in the API description for details.
5474 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
5477 struct dt_object *next = dt_object_child(dt);
5478 struct lod_object *lo = lod_dt_obj(dt);
5479 struct lod_thread_info *info = lod_env_info(env);
5480 char *stripe_name = info->lti_key;
5485 /* destroy sub-stripe of master object */
5486 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5487 rc = next->do_ops->do_index_try(env, next,
5488 &dt_directory_features);
5492 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5493 rc = lod_sub_ref_del(env, next, th);
5497 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
5498 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
5501 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
5502 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
5503 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)));
5505 rc = lod_sub_delete(env, next,
5506 (const struct dt_key *)stripe_name, th);
5512 rc = lod_sub_destroy(env, next, th);
5516 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
5517 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
5520 if (!lod_obj_is_striped(dt))
5523 /* destroy all striped objects */
5524 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5525 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5526 if (lo->ldo_stripe[i] == NULL)
5528 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
5529 i == cfs_fail_val) {
5530 dt_write_lock(env, lo->ldo_stripe[i],
5532 rc = lod_sub_ref_del(env, lo->ldo_stripe[i],
5534 dt_write_unlock(env, lo->ldo_stripe[i]);
5538 rc = lod_sub_destroy(env, lo->ldo_stripe[i],
5545 struct lod_obj_stripe_cb_data data = { { 0 } };
5547 data.locd_declare = false;
5548 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
5549 rc = lod_obj_for_each_stripe(env, lo, th, &data);
5556 * Implementation of dt_object_operations::do_declare_ref_add.
5558 * \see dt_object_operations::do_declare_ref_add() in the API description
5561 static int lod_declare_ref_add(const struct lu_env *env,
5562 struct dt_object *dt, struct thandle *th)
5564 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
5568 * Implementation of dt_object_operations::do_ref_add.
5570 * \see dt_object_operations::do_ref_add() in the API description for details.
5572 static int lod_ref_add(const struct lu_env *env,
5573 struct dt_object *dt, struct thandle *th)
5575 return lod_sub_ref_add(env, dt_object_child(dt), th);
5579 * Implementation of dt_object_operations::do_declare_ref_del.
5581 * \see dt_object_operations::do_declare_ref_del() in the API description
5584 static int lod_declare_ref_del(const struct lu_env *env,
5585 struct dt_object *dt, struct thandle *th)
5587 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
5591 * Implementation of dt_object_operations::do_ref_del
5593 * \see dt_object_operations::do_ref_del() in the API description for details.
5595 static int lod_ref_del(const struct lu_env *env,
5596 struct dt_object *dt, struct thandle *th)
5598 return lod_sub_ref_del(env, dt_object_child(dt), th);
5602 * Implementation of dt_object_operations::do_object_sync.
5604 * \see dt_object_operations::do_object_sync() in the API description
5607 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
5608 __u64 start, __u64 end)
5610 return dt_object_sync(env, dt_object_child(dt), start, end);
5614 * Implementation of dt_object_operations::do_object_unlock.
5616 * Used to release LDLM lock(s).
5618 * \see dt_object_operations::do_object_unlock() in the API description
5621 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
5622 struct ldlm_enqueue_info *einfo,
5623 union ldlm_policy_data *policy)
5625 struct lod_object *lo = lod_dt_obj(dt);
5626 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
5627 int slave_locks_size;
5631 if (slave_locks == NULL)
5634 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
5635 /* Note: for remote lock for single stripe dir, MDT will cancel
5636 * the lock by lockh directly */
5637 LASSERT(!dt_object_remote(dt_object_child(dt)));
5639 /* locks were unlocked in MDT layer */
5640 for (i = 0; i < slave_locks->ha_count; i++)
5641 LASSERT(!lustre_handle_is_used(&slave_locks->ha_handles[i]));
5644 * NB, ha_count may not equal to ldo_dir_stripe_count, because dir
5645 * layout may change, e.g., shrink dir layout after migration.
5647 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
5648 dt_invalidate(env, lo->ldo_stripe[i]);
5650 slave_locks_size = offsetof(typeof(*slave_locks),
5651 ha_handles[slave_locks->ha_count]);
5652 OBD_FREE(slave_locks, slave_locks_size);
5653 einfo->ei_cbdata = NULL;
5659 * Implementation of dt_object_operations::do_object_lock.
5661 * Used to get LDLM lock on the non-striped and striped objects.
5663 * \see dt_object_operations::do_object_lock() in the API description
5666 static int lod_object_lock(const struct lu_env *env,
5667 struct dt_object *dt,
5668 struct lustre_handle *lh,
5669 struct ldlm_enqueue_info *einfo,
5670 union ldlm_policy_data *policy)
5672 struct lod_object *lo = lod_dt_obj(dt);
5673 int slave_locks_size;
5674 struct lustre_handle_array *slave_locks = NULL;
5679 /* remote object lock */
5680 if (!einfo->ei_enq_slave) {
5681 LASSERT(dt_object_remote(dt));
5682 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
5686 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
5689 rc = lod_striping_load(env, lo);
5694 if (lo->ldo_dir_stripe_count <= 1)
5697 slave_locks_size = offsetof(typeof(*slave_locks),
5698 ha_handles[lo->ldo_dir_stripe_count]);
5699 /* Freed in lod_object_unlock */
5700 OBD_ALLOC(slave_locks, slave_locks_size);
5703 slave_locks->ha_count = lo->ldo_dir_stripe_count;
5705 /* striped directory lock */
5706 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5707 struct lustre_handle lockh;
5708 struct ldlm_res_id *res_id;
5710 res_id = &lod_env_info(env)->lti_res_id;
5711 fid_build_reg_res_name(lu_object_fid(&lo->ldo_stripe[i]->do_lu),
5713 einfo->ei_res_id = res_id;
5715 LASSERT(lo->ldo_stripe[i] != NULL);
5716 if (dt_object_remote(lo->ldo_stripe[i])) {
5717 set_bit(i, (void *)slave_locks->ha_map);
5718 rc = dt_object_lock(env, lo->ldo_stripe[i], &lockh,
5721 struct ldlm_namespace *ns = einfo->ei_namespace;
5722 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
5723 ldlm_completion_callback completion = einfo->ei_cb_cp;
5724 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
5726 if (einfo->ei_mode == LCK_PW ||
5727 einfo->ei_mode == LCK_EX)
5728 dlmflags |= LDLM_FL_COS_INCOMPAT;
5730 LASSERT(ns != NULL);
5731 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS,
5732 policy, einfo->ei_mode,
5733 &dlmflags, blocking,
5735 NULL, 0, LVB_T_NONE,
5740 ldlm_lock_decref_and_cancel(
5741 &slave_locks->ha_handles[i],
5743 OBD_FREE(slave_locks, slave_locks_size);
5746 slave_locks->ha_handles[i] = lockh;
5748 einfo->ei_cbdata = slave_locks;
5754 * Implementation of dt_object_operations::do_invalidate.
5756 * \see dt_object_operations::do_invalidate() in the API description for details
5758 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
5760 return dt_invalidate(env, dt_object_child(dt));
5763 static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
5767 /* clear memory region that will be used for layout change */
5768 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
5769 info->lti_count = 0;
5771 if (info->lti_comp_size >= comp_cnt)
5774 if (info->lti_comp_size > 0) {
5775 OBD_FREE(info->lti_comp_idx,
5776 info->lti_comp_size * sizeof(__u32));
5777 info->lti_comp_size = 0;
5780 OBD_ALLOC(info->lti_comp_idx, comp_cnt * sizeof(__u32));
5781 if (!info->lti_comp_idx)
5784 info->lti_comp_size = comp_cnt;
5788 static int lod_declare_instantiate_components(const struct lu_env *env,
5789 struct lod_object *lo, struct thandle *th)
5791 struct lod_thread_info *info = lod_env_info(env);
5796 LASSERT(info->lti_count < lo->ldo_comp_cnt);
5798 for (i = 0; i < info->lti_count; i++) {
5799 rc = lod_qos_prep_create(env, lo, NULL, th,
5800 info->lti_comp_idx[i]);
5806 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
5807 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
5808 &info->lti_buf, XATTR_NAME_LOV, 0, th);
5814 static int lod_declare_update_plain(const struct lu_env *env,
5815 struct lod_object *lo, struct layout_intent *layout,
5816 const struct lu_buf *buf, struct thandle *th)
5818 struct lod_thread_info *info = lod_env_info(env);
5819 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5820 struct lod_layout_component *lod_comp;
5821 struct lov_comp_md_v1 *comp_v1 = NULL;
5822 bool replay = false;
5826 LASSERT(lo->ldo_flr_state == LCM_FL_NONE);
5829 * In case the client is passing lovea, which only happens during
5830 * the replay of layout intent write RPC for now, we may need to
5831 * parse the lovea and apply new layout configuration.
5833 if (buf && buf->lb_len) {
5834 struct lov_user_md_v1 *v1 = buf->lb_buf;
5836 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
5837 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
5838 LOV_MAGIC_COMP_V1)) {
5839 CERROR("%s: the replay buffer of layout extend "
5840 "(magic %#x) does not contain expected "
5841 "composite layout.\n",
5842 lod2obd(d)->obd_name, v1->lmm_magic);
5843 GOTO(out, rc = -EINVAL);
5846 rc = lod_use_defined_striping(env, lo, buf);
5849 lo->ldo_comp_cached = 1;
5851 rc = lod_get_lov_ea(env, lo);
5854 /* old on-disk EA is stored in info->lti_buf */
5855 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
5858 /* non replay path */
5859 rc = lod_striping_load(env, lo);
5864 /* Make sure defined layout covers the requested write range. */
5865 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
5866 if (lo->ldo_comp_cnt > 1 &&
5867 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
5868 lod_comp->llc_extent.e_end < layout->li_extent.e_end) {
5869 CDEBUG(replay ? D_ERROR : D_LAYOUT,
5870 "%s: the defined layout [0, %#llx) does not covers "
5871 "the write range "DEXT"\n",
5872 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
5873 PEXT(&layout->li_extent));
5874 GOTO(out, rc = -EINVAL);
5877 CDEBUG(D_LAYOUT, "%s: "DFID": instantiate components "DEXT"\n",
5878 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
5879 PEXT(&layout->li_extent));
5882 * Iterate ld->ldo_comp_entries, find the component whose extent under
5883 * the write range and not instantianted.
5885 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5886 lod_comp = &lo->ldo_comp_entries[i];
5888 if (lod_comp->llc_extent.e_start >= layout->li_extent.e_end)
5892 if (lod_comp_inited(lod_comp))
5896 * In replay path, lod_comp is the EA passed by
5897 * client replay buffer, comp_v1 is the pre-recovery
5898 * on-disk EA, we'd sift out those components which
5899 * were init-ed in the on-disk EA.
5901 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
5906 * this component hasn't instantiated in normal path, or during
5907 * replay it needs replay the instantiation.
5910 /* A released component is being extended */
5911 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
5912 GOTO(out, rc = -EINVAL);
5914 LASSERT(info->lti_comp_idx != NULL);
5915 info->lti_comp_idx[info->lti_count++] = i;
5918 if (info->lti_count == 0)
5921 lod_obj_inc_layout_gen(lo);
5922 rc = lod_declare_instantiate_components(env, lo, th);
5925 lod_striping_free(env, lo);
5929 static inline int lod_comp_index(struct lod_object *lo,
5930 struct lod_layout_component *lod_comp)
5932 LASSERT(lod_comp >= lo->ldo_comp_entries &&
5933 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
5935 return lod_comp - lo->ldo_comp_entries;
5939 * Stale other mirrors by writing extent.
5941 static void lod_stale_components(struct lod_object *lo, int primary,
5942 struct lu_extent *extent)
5944 struct lod_layout_component *pri_comp, *lod_comp;
5947 /* The writing extent decides which components in the primary
5948 * are affected... */
5949 CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
5950 lod_foreach_mirror_comp(pri_comp, lo, primary) {
5951 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
5954 CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
5955 lod_comp_index(lo, pri_comp),
5956 PEXT(&pri_comp->llc_extent));
5958 for (i = 0; i < lo->ldo_mirror_count; i++) {
5962 /* ... and then stale other components that are
5963 * overlapping with primary components */
5964 lod_foreach_mirror_comp(lod_comp, lo, i) {
5965 if (!lu_extent_is_overlapped(
5966 &pri_comp->llc_extent,
5967 &lod_comp->llc_extent))
5970 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
5971 i, lod_comp_index(lo, lod_comp));
5973 lod_comp->llc_flags |= LCME_FL_STALE;
5974 lo->ldo_mirrors[i].lme_stale = 1;
5981 * check an OST's availability
5982 * \param[in] env execution environment
5983 * \param[in] lo lod object
5984 * \param[in] dt dt object
5985 * \param[in] index mirror index
5987 * \retval negative if failed
5988 * \retval 1 if \a dt is available
5989 * \retval 0 if \a dt is not available
5991 static inline int lod_check_ost_avail(const struct lu_env *env,
5992 struct lod_object *lo,
5993 struct dt_object *dt, int index)
5995 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5996 struct lod_tgt_desc *ost;
5998 int type = LU_SEQ_RANGE_OST;
6001 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &idx, &type);
6003 CERROR("%s: can't locate "DFID":rc = %d\n",
6004 lod2obd(lod)->obd_name, PFID(lu_object_fid(&dt->do_lu)),
6009 ost = OST_TGT(lod, idx);
6010 if (ost->ltd_statfs.os_state &
6011 (OS_STATE_READONLY | OS_STATE_ENOSPC | OS_STATE_ENOINO |
6012 OS_STATE_NOPRECREATE) ||
6013 ost->ltd_active == 0) {
6014 CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail, rc = %d\n",
6015 PFID(lod_object_fid(lo)), index, idx, rc);
6023 * Pick primary mirror for write
6024 * \param[in] env execution environment
6025 * \param[in] lo object
6026 * \param[in] extent write range
6028 static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo,
6029 struct lu_extent *extent)
6031 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6032 unsigned int seq = 0;
6033 struct lod_layout_component *lod_comp;
6035 int picked = -1, second_pick = -1, third_pick = -1;
6038 if (OBD_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
6039 get_random_bytes(&seq, sizeof(seq));
6040 seq %= lo->ldo_mirror_count;
6044 * Pick a mirror as the primary, and check the availability of OSTs.
6046 * This algo can be revised later after knowing the topology of
6049 lod_qos_statfs_update(env, lod);
6050 for (i = 0; i < lo->ldo_mirror_count; i++) {
6051 bool ost_avail = true;
6052 int index = (i + seq) % lo->ldo_mirror_count;
6054 if (lo->ldo_mirrors[index].lme_stale) {
6055 CDEBUG(D_LAYOUT, DFID": mirror %d stale\n",
6056 PFID(lod_object_fid(lo)), index);
6060 /* 2nd pick is for the primary mirror containing unavail OST */
6061 if (lo->ldo_mirrors[index].lme_primary && second_pick < 0)
6062 second_pick = index;
6064 /* 3rd pick is for non-primary mirror containing unavail OST */
6065 if (second_pick < 0 && third_pick < 0)
6069 * we found a non-primary 1st pick, we'd like to find a
6070 * potential pirmary mirror.
6072 if (picked >= 0 && !lo->ldo_mirrors[index].lme_primary)
6075 /* check the availability of OSTs */
6076 lod_foreach_mirror_comp(lod_comp, lo, index) {
6077 if (!lod_comp_inited(lod_comp) || !lod_comp->llc_stripe)
6080 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6081 struct dt_object *dt = lod_comp->llc_stripe[j];
6083 rc = lod_check_ost_avail(env, lo, dt, index);
6090 } /* for all dt object in one component */
6093 } /* for all components in a mirror */
6096 * the OSTs where allocated objects locates in the components
6097 * of the mirror are available.
6102 /* this mirror has all OSTs available */
6106 * primary with all OSTs are available, this is the perfect
6109 if (lo->ldo_mirrors[index].lme_primary)
6111 } /* for all mirrors */
6113 /* failed to pick a sound mirror, lower our expectation */
6115 picked = second_pick;
6117 picked = third_pick;
6125 * figure out the components should be instantiated for resync.
6127 static int lod_prepare_resync(const struct lu_env *env, struct lod_object *lo,
6128 struct lu_extent *extent)
6130 struct lod_thread_info *info = lod_env_info(env);
6131 struct lod_layout_component *lod_comp;
6132 unsigned int need_sync = 0;
6136 DFID": instantiate all stale components in "DEXT"\n",
6137 PFID(lod_object_fid(lo)), PEXT(extent));
6140 * instantiate all components within this extent, even non-stale
6143 for (i = 0; i < lo->ldo_mirror_count; i++) {
6144 if (!lo->ldo_mirrors[i].lme_stale)
6147 lod_foreach_mirror_comp(lod_comp, lo, i) {
6148 if (!lu_extent_is_overlapped(extent,
6149 &lod_comp->llc_extent))
6154 if (lod_comp_inited(lod_comp))
6157 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
6158 i, lod_comp_index(lo, lod_comp));
6159 info->lti_comp_idx[info->lti_count++] =
6160 lod_comp_index(lo, lod_comp);
6164 return need_sync ? 0 : -EALREADY;
6167 static int lod_declare_update_rdonly(const struct lu_env *env,
6168 struct lod_object *lo, struct md_layout_change *mlc,
6171 struct lod_thread_info *info = lod_env_info(env);
6172 struct lu_attr *layout_attr = &info->lti_layout_attr;
6173 struct lod_layout_component *lod_comp;
6174 struct lu_extent extent = { 0 };
6178 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
6179 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
6180 mlc->mlc_opc == MD_LAYOUT_RESYNC);
6181 LASSERT(lo->ldo_mirror_count > 0);
6183 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
6184 struct layout_intent *layout = mlc->mlc_intent;
6187 extent = layout->li_extent;
6188 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
6189 PFID(lod_object_fid(lo)), PEXT(&extent));
6191 picked = lod_primary_pick(env, lo, &extent);
6195 CDEBUG(D_LAYOUT, DFID": picked mirror id %u as primary\n",
6196 PFID(lod_object_fid(lo)),
6197 lo->ldo_mirrors[picked].lme_id);
6199 if (layout->li_opc == LAYOUT_INTENT_TRUNC) {
6201 * trunc transfers [0, size) in the intent extent, we'd
6202 * stale components overlapping [size, eof).
6204 extent.e_start = extent.e_end;
6205 extent.e_end = OBD_OBJECT_EOF;
6208 /* stale overlapping components from other mirrors */
6209 lod_stale_components(lo, picked, &extent);
6211 /* restore truncate intent extent */
6212 if (layout->li_opc == LAYOUT_INTENT_TRUNC)
6213 extent.e_end = extent.e_start;
6215 /* instantiate components for the picked mirror, start from 0 */
6218 lod_foreach_mirror_comp(lod_comp, lo, picked) {
6219 if (!lu_extent_is_overlapped(&extent,
6220 &lod_comp->llc_extent))
6223 if (lod_comp_inited(lod_comp))
6226 info->lti_comp_idx[info->lti_count++] =
6227 lod_comp_index(lo, lod_comp);
6230 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
6231 } else { /* MD_LAYOUT_RESYNC */
6235 * could contain multiple non-stale mirrors, so we need to
6236 * prep uninited all components assuming any non-stale mirror
6237 * could be picked as the primary mirror.
6239 for (i = 0; i < lo->ldo_mirror_count; i++) {
6240 if (lo->ldo_mirrors[i].lme_stale)
6243 lod_foreach_mirror_comp(lod_comp, lo, i) {
6244 if (!lod_comp_inited(lod_comp))
6247 if (extent.e_end < lod_comp->llc_extent.e_end)
6249 lod_comp->llc_extent.e_end;
6253 rc = lod_prepare_resync(env, lo, &extent);
6256 /* change the file state to SYNC_PENDING */
6257 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
6260 /* Reset the layout version once it's becoming too large.
6261 * This way it can make sure that the layout version is
6262 * monotonously increased in this writing era. */
6263 lod_obj_inc_layout_gen(lo);
6264 if (lo->ldo_layout_gen > (LCME_ID_MAX >> 1)) {
6265 __u32 layout_version;
6267 cfs_get_random_bytes(&layout_version, sizeof(layout_version));
6268 lo->ldo_layout_gen = layout_version & 0xffff;
6271 rc = lod_declare_instantiate_components(env, lo, th);
6275 layout_attr->la_valid = LA_LAYOUT_VERSION;
6276 layout_attr->la_layout_version = 0; /* set current version */
6277 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
6278 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
6279 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
6285 lod_striping_free(env, lo);
6289 static int lod_declare_update_write_pending(const struct lu_env *env,
6290 struct lod_object *lo, struct md_layout_change *mlc,
6293 struct lod_thread_info *info = lod_env_info(env);
6294 struct lu_attr *layout_attr = &info->lti_layout_attr;
6295 struct lod_layout_component *lod_comp;
6296 struct lu_extent extent = { 0 };
6302 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
6303 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
6304 mlc->mlc_opc == MD_LAYOUT_RESYNC);
6306 /* look for the primary mirror */
6307 for (i = 0; i < lo->ldo_mirror_count; i++) {
6308 if (lo->ldo_mirrors[i].lme_stale)
6311 LASSERTF(primary < 0, DFID " has multiple primary: %u / %u",
6312 PFID(lod_object_fid(lo)),
6313 lo->ldo_mirrors[i].lme_id,
6314 lo->ldo_mirrors[primary].lme_id);
6319 CERROR(DFID ": doesn't have a primary mirror\n",
6320 PFID(lod_object_fid(lo)));
6321 GOTO(out, rc = -ENODATA);
6324 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
6325 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
6327 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
6329 /* for LAYOUT_WRITE opc, it has to do the following operations:
6330 * 1. stale overlapping componets from stale mirrors;
6331 * 2. instantiate components of the primary mirror;
6332 * 3. transfter layout version to all objects of the primary;
6334 * for LAYOUT_RESYNC opc, it will do:
6335 * 1. instantiate components of all stale mirrors;
6336 * 2. transfer layout version to all objects to close write era. */
6338 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
6339 LASSERT(mlc->mlc_intent != NULL);
6341 extent = mlc->mlc_intent->li_extent;
6343 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
6344 PFID(lod_object_fid(lo)), PEXT(&extent));
6346 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC) {
6348 * trunc transfers [0, size) in the intent extent, we'd
6349 * stale components overlapping [size, eof).
6351 extent.e_start = extent.e_end;
6352 extent.e_end = OBD_OBJECT_EOF;
6354 /* 1. stale overlapping components */
6355 lod_stale_components(lo, primary, &extent);
6357 /* 2. find out the components need instantiating.
6358 * instantiate [0, mlc->mlc_intent->e_end) */
6360 /* restore truncate intent extent */
6361 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC)
6362 extent.e_end = extent.e_start;
6365 lod_foreach_mirror_comp(lod_comp, lo, primary) {
6366 if (!lu_extent_is_overlapped(&extent,
6367 &lod_comp->llc_extent))
6370 if (lod_comp_inited(lod_comp))
6373 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
6374 primary, lod_comp_index(lo, lod_comp));
6375 info->lti_comp_idx[info->lti_count++] =
6376 lod_comp_index(lo, lod_comp);
6378 } else { /* MD_LAYOUT_RESYNC */
6379 lod_foreach_mirror_comp(lod_comp, lo, primary) {
6380 if (!lod_comp_inited(lod_comp))
6383 extent.e_end = lod_comp->llc_extent.e_end;
6386 rc = lod_prepare_resync(env, lo, &extent);
6389 /* change the file state to SYNC_PENDING */
6390 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
6393 rc = lod_declare_instantiate_components(env, lo, th);
6397 /* 3. transfer layout version to OST objects.
6398 * transfer new layout version to OST objects so that stale writes
6399 * can be denied. It also ends an era of writing by setting
6400 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
6401 * send write RPC; only resync RPCs could do it. */
6402 layout_attr->la_valid = LA_LAYOUT_VERSION;
6403 layout_attr->la_layout_version = 0; /* set current version */
6404 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
6405 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
6406 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
6410 lod_obj_inc_layout_gen(lo);
6413 lod_striping_free(env, lo);
6417 static int lod_declare_update_sync_pending(const struct lu_env *env,
6418 struct lod_object *lo, struct md_layout_change *mlc,
6421 struct lod_thread_info *info = lod_env_info(env);
6422 unsigned sync_components = 0;
6423 unsigned resync_components = 0;
6428 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
6429 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
6430 mlc->mlc_opc == MD_LAYOUT_WRITE);
6432 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
6433 PFID(lod_object_fid(lo)), mlc->mlc_opc);
6435 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
6436 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
6437 PFID(lod_object_fid(lo)));
6439 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
6440 return lod_declare_update_write_pending(env, lo, mlc, th);
6443 /* MD_LAYOUT_RESYNC_DONE */
6445 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6446 struct lod_layout_component *lod_comp;
6449 lod_comp = &lo->ldo_comp_entries[i];
6451 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
6456 for (j = 0; j < mlc->mlc_resync_count; j++) {
6457 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
6460 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
6461 lod_comp->llc_flags &= ~LCME_FL_STALE;
6462 resync_components++;
6468 for (i = 0; i < mlc->mlc_resync_count; i++) {
6469 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
6472 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
6473 "or already synced\n", PFID(lod_object_fid(lo)),
6474 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
6475 GOTO(out, rc = -EINVAL);
6478 if (!sync_components || (mlc->mlc_resync_count && !resync_components)) {
6479 CDEBUG(D_LAYOUT, DFID": no mirror in sync\n",
6480 PFID(lod_object_fid(lo)));
6482 /* tend to return an error code here to prevent
6483 * the MDT from setting SoM attribute */
6484 GOTO(out, rc = -EINVAL);
6487 CDEBUG(D_LAYOUT, DFID": resynced %u/%zu components\n",
6488 PFID(lod_object_fid(lo)),
6489 resync_components, mlc->mlc_resync_count);
6491 lo->ldo_flr_state = LCM_FL_RDONLY;
6492 lod_obj_inc_layout_gen(lo);
6494 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6495 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
6496 &info->lti_buf, XATTR_NAME_LOV, 0, th);
6501 lod_striping_free(env, lo);
6505 static int lod_declare_layout_change(const struct lu_env *env,
6506 struct dt_object *dt, struct md_layout_change *mlc,
6509 struct lod_thread_info *info = lod_env_info(env);
6510 struct lod_object *lo = lod_dt_obj(dt);
6514 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
6515 dt_object_remote(dt_object_child(dt)))
6518 rc = lod_striping_load(env, lo);
6522 LASSERT(lo->ldo_comp_cnt > 0);
6524 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
6528 switch (lo->ldo_flr_state) {
6530 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
6534 rc = lod_declare_update_rdonly(env, lo, mlc, th);
6536 case LCM_FL_WRITE_PENDING:
6537 rc = lod_declare_update_write_pending(env, lo, mlc, th);
6539 case LCM_FL_SYNC_PENDING:
6540 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
6551 * Instantiate layout component objects which covers the intent write offset.
6553 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
6554 struct md_layout_change *mlc, struct thandle *th)
6556 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
6557 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
6558 struct lod_object *lo = lod_dt_obj(dt);
6561 rc = lod_striped_create(env, dt, attr, NULL, th);
6562 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
6563 layout_attr->la_layout_version |= lo->ldo_layout_gen;
6564 rc = lod_attr_set(env, dt, layout_attr, th);
6570 struct dt_object_operations lod_obj_ops = {
6571 .do_read_lock = lod_read_lock,
6572 .do_write_lock = lod_write_lock,
6573 .do_read_unlock = lod_read_unlock,
6574 .do_write_unlock = lod_write_unlock,
6575 .do_write_locked = lod_write_locked,
6576 .do_attr_get = lod_attr_get,
6577 .do_declare_attr_set = lod_declare_attr_set,
6578 .do_attr_set = lod_attr_set,
6579 .do_xattr_get = lod_xattr_get,
6580 .do_declare_xattr_set = lod_declare_xattr_set,
6581 .do_xattr_set = lod_xattr_set,
6582 .do_declare_xattr_del = lod_declare_xattr_del,
6583 .do_xattr_del = lod_xattr_del,
6584 .do_xattr_list = lod_xattr_list,
6585 .do_ah_init = lod_ah_init,
6586 .do_declare_create = lod_declare_create,
6587 .do_create = lod_create,
6588 .do_declare_destroy = lod_declare_destroy,
6589 .do_destroy = lod_destroy,
6590 .do_index_try = lod_index_try,
6591 .do_declare_ref_add = lod_declare_ref_add,
6592 .do_ref_add = lod_ref_add,
6593 .do_declare_ref_del = lod_declare_ref_del,
6594 .do_ref_del = lod_ref_del,
6595 .do_object_sync = lod_object_sync,
6596 .do_object_lock = lod_object_lock,
6597 .do_object_unlock = lod_object_unlock,
6598 .do_invalidate = lod_invalidate,
6599 .do_declare_layout_change = lod_declare_layout_change,
6600 .do_layout_change = lod_layout_change,
6604 * Implementation of dt_body_operations::dbo_read.
6606 * \see dt_body_operations::dbo_read() in the API description for details.
6608 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
6609 struct lu_buf *buf, loff_t *pos)
6611 struct dt_object *next = dt_object_child(dt);
6613 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
6614 S_ISLNK(dt->do_lu.lo_header->loh_attr));
6615 return next->do_body_ops->dbo_read(env, next, buf, pos);
6619 * Implementation of dt_body_operations::dbo_declare_write.
6621 * \see dt_body_operations::dbo_declare_write() in the API description
6624 static ssize_t lod_declare_write(const struct lu_env *env,
6625 struct dt_object *dt,
6626 const struct lu_buf *buf, loff_t pos,
6629 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
6633 * Implementation of dt_body_operations::dbo_write.
6635 * \see dt_body_operations::dbo_write() in the API description for details.
6637 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
6638 const struct lu_buf *buf, loff_t *pos,
6641 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
6642 S_ISLNK(dt->do_lu.lo_header->loh_attr));
6643 return lod_sub_write(env, dt_object_child(dt), buf, pos, th);
6646 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
6647 __u64 start, __u64 end, struct thandle *th)
6649 if (dt_object_remote(dt))
6652 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
6655 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
6656 __u64 start, __u64 end, struct thandle *th)
6658 if (dt_object_remote(dt))
6661 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
6662 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
6666 * different type of files use the same body_ops because object may be created
6667 * in OUT, where there is no chance to set correct body_ops for each type, so
6668 * body_ops themselves will check file type inside, see lod_read/write/punch for
6671 const struct dt_body_operations lod_body_ops = {
6672 .dbo_read = lod_read,
6673 .dbo_declare_write = lod_declare_write,
6674 .dbo_write = lod_write,
6675 .dbo_declare_punch = lod_declare_punch,
6676 .dbo_punch = lod_punch,
6680 * Implementation of lu_object_operations::loo_object_init.
6682 * The function determines the type and the index of the target device using
6683 * sequence of the object's FID. Then passes control down to the
6684 * corresponding device:
6685 * OSD for the local objects, OSP for remote
6687 * \see lu_object_operations::loo_object_init() in the API description
6690 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
6691 const struct lu_object_conf *conf)
6693 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
6694 struct lu_device *cdev = NULL;
6695 struct lu_object *cobj;
6696 struct lod_tgt_descs *ltd = NULL;
6697 struct lod_tgt_desc *tgt;
6699 int type = LU_SEQ_RANGE_ANY;
6703 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
6705 /* Note: Sometimes, it will Return EAGAIN here, see
6706 * ptrlpc_import_delay_req(), which might confuse
6707 * lu_object_find_at() and make it wait there incorrectly.
6708 * so we convert it to EIO here.*/
6715 if (type == LU_SEQ_RANGE_MDT &&
6716 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
6717 cdev = &lod->lod_child->dd_lu_dev;
6718 } else if (type == LU_SEQ_RANGE_MDT) {
6719 ltd = &lod->lod_mdt_descs;
6721 } else if (type == LU_SEQ_RANGE_OST) {
6722 ltd = &lod->lod_ost_descs;
6729 if (ltd->ltd_tgts_size > idx &&
6730 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
6731 tgt = LTD_TGT(ltd, idx);
6733 LASSERT(tgt != NULL);
6734 LASSERT(tgt->ltd_tgt != NULL);
6736 cdev = &(tgt->ltd_tgt->dd_lu_dev);
6738 lod_putref(lod, ltd);
6741 if (unlikely(cdev == NULL))
6744 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
6745 if (unlikely(cobj == NULL))
6748 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
6750 lu_object_add(lo, cobj);
6757 * Release resources associated with striping.
6759 * If the object is striped (regular or directory), then release
6760 * the stripe objects references and free the ldo_stripe array.
6762 * \param[in] env execution environment
6763 * \param[in] lo object
6765 void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo)
6767 struct lod_layout_component *lod_comp;
6770 if (lo->ldo_stripe != NULL) {
6771 LASSERT(lo->ldo_comp_entries == NULL);
6772 LASSERT(lo->ldo_dir_stripes_allocated > 0);
6774 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6775 if (lo->ldo_stripe[i])
6776 dt_object_put(env, lo->ldo_stripe[i]);
6779 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
6780 OBD_FREE(lo->ldo_stripe, j);
6781 lo->ldo_stripe = NULL;
6782 lo->ldo_dir_stripes_allocated = 0;
6783 lo->ldo_dir_stripe_loaded = 0;
6784 lo->ldo_dir_stripe_count = 0;
6785 } else if (lo->ldo_comp_entries != NULL) {
6786 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6787 /* free lod_layout_component::llc_stripe array */
6788 lod_comp = &lo->ldo_comp_entries[i];
6790 if (lod_comp->llc_stripe == NULL)
6792 LASSERT(lod_comp->llc_stripes_allocated != 0);
6793 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
6794 if (lod_comp->llc_stripe[j] != NULL)
6796 &lod_comp->llc_stripe[j]->do_lu);
6798 OBD_FREE(lod_comp->llc_stripe,
6799 sizeof(struct dt_object *) *
6800 lod_comp->llc_stripes_allocated);
6801 lod_comp->llc_stripe = NULL;
6802 OBD_FREE(lod_comp->llc_ost_indices,
6804 lod_comp->llc_stripes_allocated);
6805 lod_comp->llc_ost_indices = NULL;
6806 lod_comp->llc_stripes_allocated = 0;
6808 lod_free_comp_entries(lo);
6809 lo->ldo_comp_cached = 0;
6813 void lod_striping_free(const struct lu_env *env, struct lod_object *lo)
6815 mutex_lock(&lo->ldo_layout_mutex);
6816 lod_striping_free_nolock(env, lo);
6817 mutex_unlock(&lo->ldo_layout_mutex);
6821 * Implementation of lu_object_operations::loo_object_free.
6823 * \see lu_object_operations::loo_object_free() in the API description
6826 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
6828 struct lod_object *lo = lu2lod_obj(o);
6830 /* release all underlying object pinned */
6831 lod_striping_free(env, lo);
6833 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
6837 * Implementation of lu_object_operations::loo_object_release.
6839 * \see lu_object_operations::loo_object_release() in the API description
6842 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
6844 /* XXX: shouldn't we release everything here in case if object
6845 * creation failed before? */
6849 * Implementation of lu_object_operations::loo_object_print.
6851 * \see lu_object_operations::loo_object_print() in the API description
6854 static int lod_object_print(const struct lu_env *env, void *cookie,
6855 lu_printer_t p, const struct lu_object *l)
6857 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
6859 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
6862 struct lu_object_operations lod_lu_obj_ops = {
6863 .loo_object_init = lod_object_init,
6864 .loo_object_free = lod_object_free,
6865 .loo_object_release = lod_object_release,
6866 .loo_object_print = lod_object_print,