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, 2013, 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 * lustre/doc/osd-api.txt.
38 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
41 #define DEBUG_SUBSYSTEM S_MDS
44 #include <obd_class.h>
45 #include <lustre_ver.h>
46 #include <obd_support.h>
47 #include <lprocfs_status.h>
49 #include <lustre_fid.h>
50 #include <lustre_param.h>
51 #include <lustre_fid.h>
52 #include <lustre_lmv.h>
53 #include <md_object.h>
54 #include <lustre_linkea.h>
56 #include "lod_internal.h"
58 static const char dot[] = ".";
59 static const char dotdot[] = "..";
61 extern struct kmem_cache *lod_object_kmem;
62 static const struct dt_body_operations lod_body_lnk_ops;
65 * Implementation of dt_index_operations::dio_lookup
67 * Used with regular (non-striped) objects.
69 * \see dt_index_operations::dio_lookup() in the API description for details.
71 static int lod_index_lookup(const struct lu_env *env, struct dt_object *dt,
72 struct dt_rec *rec, const struct dt_key *key,
73 struct lustre_capa *capa)
75 struct dt_object *next = dt_object_child(dt);
76 return next->do_index_ops->dio_lookup(env, next, rec, key, capa);
80 * Implementation of dt_index_operations::dio_declare_insert.
82 * Used with regular (non-striped) objects.
84 * \see dt_index_operations::dio_declare_insert() in the API description
87 static int lod_declare_index_insert(const struct lu_env *env,
89 const struct dt_rec *rec,
90 const struct dt_key *key,
91 struct thandle *handle)
93 return dt_declare_insert(env, dt_object_child(dt), rec, key, handle);
97 * Implementation of dt_index_operations::dio_insert.
99 * Used with regular (non-striped) objects
101 * \see dt_index_operations::dio_insert() in the API description for details.
103 static int lod_index_insert(const struct lu_env *env,
104 struct dt_object *dt,
105 const struct dt_rec *rec,
106 const struct dt_key *key,
108 struct lustre_capa *capa,
111 return dt_insert(env, dt_object_child(dt), rec, key, th, capa, ign);
115 * Implementation of dt_index_operations::dio_declare_delete.
117 * Used with regular (non-striped) objects.
119 * \see dt_index_operations::dio_declare_delete() in the API description
122 static int lod_declare_index_delete(const struct lu_env *env,
123 struct dt_object *dt,
124 const struct dt_key *key,
127 return dt_declare_delete(env, dt_object_child(dt), key, th);
131 * Implementation of dt_index_operations::dio_delete.
133 * Used with regular (non-striped) objects.
135 * \see dt_index_operations::dio_delete() in the API description for details.
137 static int lod_index_delete(const struct lu_env *env,
138 struct dt_object *dt,
139 const struct dt_key *key,
141 struct lustre_capa *capa)
143 return dt_delete(env, dt_object_child(dt), key, th, capa);
147 * Implementation of dt_it_ops::init.
149 * Used with regular (non-striped) objects.
151 * \see dt_it_ops::init() in the API description for details.
153 static struct dt_it *lod_it_init(const struct lu_env *env,
154 struct dt_object *dt, __u32 attr,
155 struct lustre_capa *capa)
157 struct dt_object *next = dt_object_child(dt);
158 struct lod_it *it = &lod_env_info(env)->lti_it;
159 struct dt_it *it_next;
162 it_next = next->do_index_ops->dio_it.init(env, next, attr, capa);
166 /* currently we do not use more than one iterator per thread
167 * so we store it in thread info. if at some point we need
168 * more active iterators in a single thread, we can allocate
170 LASSERT(it->lit_obj == NULL);
172 it->lit_it = it_next;
175 return (struct dt_it *)it;
178 #define LOD_CHECK_IT(env, it) \
180 LASSERT((it)->lit_obj != NULL); \
181 LASSERT((it)->lit_it != NULL); \
185 * Implementation of dt_index_operations::dio_it.fini.
187 * Used with regular (non-striped) objects.
189 * \see dt_index_operations::dio_it.fini() in the API description for details.
191 void lod_it_fini(const struct lu_env *env, struct dt_it *di)
193 struct lod_it *it = (struct lod_it *)di;
195 LOD_CHECK_IT(env, it);
196 it->lit_obj->do_index_ops->dio_it.fini(env, it->lit_it);
198 /* the iterator not in use any more */
204 * Implementation of dt_it_ops::get.
206 * Used with regular (non-striped) objects.
208 * \see dt_it_ops::get() in the API description for details.
210 int lod_it_get(const struct lu_env *env, struct dt_it *di,
211 const struct dt_key *key)
213 const struct lod_it *it = (const struct lod_it *)di;
215 LOD_CHECK_IT(env, it);
216 return it->lit_obj->do_index_ops->dio_it.get(env, it->lit_it, key);
220 * Implementation of dt_it_ops::put.
222 * Used with regular (non-striped) objects.
224 * \see dt_it_ops::put() in the API description for details.
226 void lod_it_put(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.put(env, it->lit_it);
235 * Implementation of dt_it_ops::next.
237 * Used with regular (non-striped) objects
239 * \see dt_it_ops::next() in the API description for details.
241 int lod_it_next(const struct lu_env *env, struct dt_it *di)
243 struct lod_it *it = (struct lod_it *)di;
245 LOD_CHECK_IT(env, it);
246 return it->lit_obj->do_index_ops->dio_it.next(env, it->lit_it);
250 * Implementation of dt_it_ops::key.
252 * Used with regular (non-striped) objects.
254 * \see dt_it_ops::key() in the API description for details.
256 struct dt_key *lod_it_key(const struct lu_env *env, const struct dt_it *di)
258 const struct lod_it *it = (const struct lod_it *)di;
260 LOD_CHECK_IT(env, it);
261 return it->lit_obj->do_index_ops->dio_it.key(env, it->lit_it);
265 * Implementation of dt_it_ops::key_size.
267 * Used with regular (non-striped) objects.
269 * \see dt_it_ops::key_size() in the API description for details.
271 int lod_it_key_size(const struct lu_env *env, const struct dt_it *di)
273 struct lod_it *it = (struct lod_it *)di;
275 LOD_CHECK_IT(env, it);
276 return it->lit_obj->do_index_ops->dio_it.key_size(env, it->lit_it);
280 * Implementation of dt_it_ops::rec.
282 * Used with regular (non-striped) objects.
284 * \see dt_it_ops::rec() in the API description for details.
286 int lod_it_rec(const struct lu_env *env, const struct dt_it *di,
287 struct dt_rec *rec, __u32 attr)
289 const struct lod_it *it = (const struct lod_it *)di;
291 LOD_CHECK_IT(env, it);
292 return it->lit_obj->do_index_ops->dio_it.rec(env, it->lit_it, rec,
297 * Implementation of dt_it_ops::rec_size.
299 * Used with regular (non-striped) objects.
301 * \see dt_it_ops::rec_size() in the API description for details.
303 int lod_it_rec_size(const struct lu_env *env, const struct dt_it *di,
306 const struct lod_it *it = (const struct lod_it *)di;
308 LOD_CHECK_IT(env, it);
309 return it->lit_obj->do_index_ops->dio_it.rec_size(env, it->lit_it,
314 * Implementation of dt_it_ops::store.
316 * Used with regular (non-striped) objects.
318 * \see dt_it_ops::store() in the API description for details.
320 __u64 lod_it_store(const struct lu_env *env, const struct dt_it *di)
322 const struct lod_it *it = (const struct lod_it *)di;
324 LOD_CHECK_IT(env, it);
325 return it->lit_obj->do_index_ops->dio_it.store(env, it->lit_it);
329 * Implementation of dt_it_ops::load.
331 * Used with regular (non-striped) objects.
333 * \see dt_it_ops::load() in the API description for details.
335 int lod_it_load(const struct lu_env *env, const struct dt_it *di, __u64 hash)
337 const struct lod_it *it = (const struct lod_it *)di;
339 LOD_CHECK_IT(env, it);
340 return it->lit_obj->do_index_ops->dio_it.load(env, it->lit_it, hash);
344 * Implementation of dt_it_ops::key_rec.
346 * Used with regular (non-striped) objects.
348 * \see dt_it_ops::rec() in the API description for details.
350 int lod_it_key_rec(const struct lu_env *env, const struct dt_it *di,
353 const struct lod_it *it = (const struct lod_it *)di;
355 LOD_CHECK_IT(env, it);
356 return it->lit_obj->do_index_ops->dio_it.key_rec(env, it->lit_it,
360 static struct dt_index_operations lod_index_ops = {
361 .dio_lookup = lod_index_lookup,
362 .dio_declare_insert = lod_declare_index_insert,
363 .dio_insert = lod_index_insert,
364 .dio_declare_delete = lod_declare_index_delete,
365 .dio_delete = lod_index_delete,
373 .key_size = lod_it_key_size,
375 .rec_size = lod_it_rec_size,
376 .store = lod_it_store,
378 .key_rec = lod_it_key_rec,
383 * Implementation of dt_it_ops::init.
385 * Used with striped objects. Internally just initializes the iterator
386 * on the first stripe.
388 * \see dt_it_ops::init() in the API description for details.
390 static struct dt_it *lod_striped_it_init(const struct lu_env *env,
391 struct dt_object *dt, __u32 attr,
392 struct lustre_capa *capa)
394 struct lod_object *lo = lod_dt_obj(dt);
395 struct dt_object *next;
396 struct lod_it *it = &lod_env_info(env)->lti_it;
397 struct dt_it *it_next;
400 LASSERT(lo->ldo_stripenr > 0);
401 next = lo->ldo_stripe[0];
402 LASSERT(next != NULL);
403 LASSERT(next->do_index_ops != NULL);
405 it_next = next->do_index_ops->dio_it.init(env, next, attr, capa);
409 /* currently we do not use more than one iterator per thread
410 * so we store it in thread info. if at some point we need
411 * more active iterators in a single thread, we can allocate
413 LASSERT(it->lit_obj == NULL);
415 it->lit_stripe_index = 0;
417 it->lit_it = it_next;
420 return (struct dt_it *)it;
423 #define LOD_CHECK_STRIPED_IT(env, it, lo) \
425 LASSERT((it)->lit_obj != NULL); \
426 LASSERT((it)->lit_it != NULL); \
427 LASSERT((lo)->ldo_stripenr > 0); \
428 LASSERT((it)->lit_stripe_index < (lo)->ldo_stripenr); \
432 * Implementation of dt_it_ops::fini.
434 * Used with striped objects.
436 * \see dt_it_ops::fini() in the API description for details.
438 static void lod_striped_it_fini(const struct lu_env *env, struct dt_it *di)
440 struct lod_it *it = (struct lod_it *)di;
441 struct lod_object *lo = lod_dt_obj(it->lit_obj);
442 struct dt_object *next;
444 LOD_CHECK_STRIPED_IT(env, it, lo);
446 next = lo->ldo_stripe[it->lit_stripe_index];
447 LASSERT(next != NULL);
448 LASSERT(next->do_index_ops != NULL);
450 next->do_index_ops->dio_it.fini(env, it->lit_it);
452 /* the iterator not in use any more */
455 it->lit_stripe_index = 0;
459 * Implementation of dt_it_ops::get.
461 * Right now it's not used widely, only to reset the iterator to the
462 * initial position. It should be possible to implement a full version
463 * which chooses a correct stripe to be able to position with any key.
465 * \see dt_it_ops::get() in the API description for details.
467 static int lod_striped_it_get(const struct lu_env *env, struct dt_it *di,
468 const struct dt_key *key)
470 const struct lod_it *it = (const struct lod_it *)di;
471 struct lod_object *lo = lod_dt_obj(it->lit_obj);
472 struct dt_object *next;
475 LOD_CHECK_STRIPED_IT(env, it, lo);
477 next = lo->ldo_stripe[it->lit_stripe_index];
478 LASSERT(next != NULL);
479 LASSERT(next->do_index_ops != NULL);
481 return next->do_index_ops->dio_it.get(env, it->lit_it, key);
485 * Implementation of dt_it_ops::put.
487 * Used with striped objects.
489 * \see dt_it_ops::put() in the API description for details.
491 static void lod_striped_it_put(const struct lu_env *env, struct dt_it *di)
493 struct lod_it *it = (struct lod_it *)di;
494 struct lod_object *lo = lod_dt_obj(it->lit_obj);
495 struct dt_object *next;
497 LOD_CHECK_STRIPED_IT(env, it, lo);
499 next = lo->ldo_stripe[it->lit_stripe_index];
500 LASSERT(next != NULL);
501 LASSERT(next->do_index_ops != NULL);
503 return next->do_index_ops->dio_it.put(env, it->lit_it);
507 * Implementation of dt_it_ops::next.
509 * Used with striped objects. When the end of the current stripe is
510 * reached, the method takes the next stripe's iterator.
512 * \see dt_it_ops::next() in the API description for details.
514 static int lod_striped_it_next(const struct lu_env *env, struct dt_it *di)
516 struct lod_it *it = (struct lod_it *)di;
517 struct lod_object *lo = lod_dt_obj(it->lit_obj);
518 struct dt_object *next;
519 struct dt_it *it_next;
523 LOD_CHECK_STRIPED_IT(env, it, lo);
525 next = lo->ldo_stripe[it->lit_stripe_index];
526 LASSERT(next != NULL);
527 LASSERT(next->do_index_ops != NULL);
529 rc = next->do_index_ops->dio_it.next(env, it->lit_it);
533 if (rc == 0 && it->lit_stripe_index == 0)
536 if (rc == 0 && it->lit_stripe_index > 0) {
537 struct lu_dirent *ent;
539 ent = (struct lu_dirent *)lod_env_info(env)->lti_key;
541 rc = next->do_index_ops->dio_it.rec(env, it->lit_it,
542 (struct dt_rec *)ent,
547 /* skip . and .. for slave stripe */
548 if ((strncmp(ent->lde_name, ".",
549 le16_to_cpu(ent->lde_namelen)) == 0 &&
550 le16_to_cpu(ent->lde_namelen) == 1) ||
551 (strncmp(ent->lde_name, "..",
552 le16_to_cpu(ent->lde_namelen)) == 0 &&
553 le16_to_cpu(ent->lde_namelen) == 2))
559 /* go to next stripe */
560 if (it->lit_stripe_index + 1 >= lo->ldo_stripenr)
563 it->lit_stripe_index++;
565 next->do_index_ops->dio_it.put(env, it->lit_it);
566 next->do_index_ops->dio_it.fini(env, it->lit_it);
568 rc = next->do_ops->do_index_try(env, next, &dt_directory_features);
572 next = lo->ldo_stripe[it->lit_stripe_index];
573 LASSERT(next != NULL);
574 LASSERT(next->do_index_ops != NULL);
576 it_next = next->do_index_ops->dio_it.init(env, next, it->lit_attr,
578 if (!IS_ERR(it_next)) {
579 it->lit_it = it_next;
582 rc = PTR_ERR(it_next);
589 * Implementation of dt_it_ops::key.
591 * Used with striped objects.
593 * \see dt_it_ops::key() in the API description for details.
595 static struct dt_key *lod_striped_it_key(const struct lu_env *env,
596 const struct dt_it *di)
598 const struct lod_it *it = (const struct lod_it *)di;
599 struct lod_object *lo = lod_dt_obj(it->lit_obj);
600 struct dt_object *next;
602 LOD_CHECK_STRIPED_IT(env, it, lo);
604 next = lo->ldo_stripe[it->lit_stripe_index];
605 LASSERT(next != NULL);
606 LASSERT(next->do_index_ops != NULL);
608 return next->do_index_ops->dio_it.key(env, it->lit_it);
612 * Implementation of dt_it_ops::key_size.
614 * Used with striped objects.
616 * \see dt_it_ops::size() in the API description for details.
618 static int lod_striped_it_key_size(const struct lu_env *env,
619 const struct dt_it *di)
621 struct lod_it *it = (struct lod_it *)di;
622 struct lod_object *lo = lod_dt_obj(it->lit_obj);
623 struct dt_object *next;
625 LOD_CHECK_STRIPED_IT(env, it, lo);
627 next = lo->ldo_stripe[it->lit_stripe_index];
628 LASSERT(next != NULL);
629 LASSERT(next->do_index_ops != NULL);
631 return next->do_index_ops->dio_it.key_size(env, it->lit_it);
635 * Implementation of dt_it_ops::rec.
637 * Used with striped objects.
639 * \see dt_it_ops::rec() in the API description for details.
641 static int lod_striped_it_rec(const struct lu_env *env, const struct dt_it *di,
642 struct dt_rec *rec, __u32 attr)
644 const struct lod_it *it = (const struct lod_it *)di;
645 struct lod_object *lo = lod_dt_obj(it->lit_obj);
646 struct dt_object *next;
648 LOD_CHECK_STRIPED_IT(env, it, lo);
650 next = lo->ldo_stripe[it->lit_stripe_index];
651 LASSERT(next != NULL);
652 LASSERT(next->do_index_ops != NULL);
654 return next->do_index_ops->dio_it.rec(env, it->lit_it, rec, attr);
658 * Implementation of dt_it_ops::rec_size.
660 * Used with striped objects.
662 * \see dt_it_ops::rec_size() in the API description for details.
664 static int lod_striped_it_rec_size(const struct lu_env *env,
665 const struct dt_it *di, __u32 attr)
667 struct lod_it *it = (struct lod_it *)di;
668 struct lod_object *lo = lod_dt_obj(it->lit_obj);
669 struct dt_object *next;
671 LOD_CHECK_STRIPED_IT(env, it, lo);
673 next = lo->ldo_stripe[it->lit_stripe_index];
674 LASSERT(next != NULL);
675 LASSERT(next->do_index_ops != NULL);
677 return next->do_index_ops->dio_it.rec_size(env, it->lit_it, attr);
681 * Implementation of dt_it_ops::store.
683 * Used with striped objects.
685 * \see dt_it_ops::store() in the API description for details.
687 static __u64 lod_striped_it_store(const struct lu_env *env,
688 const struct dt_it *di)
690 const struct lod_it *it = (const struct lod_it *)di;
691 struct lod_object *lo = lod_dt_obj(it->lit_obj);
692 struct dt_object *next;
694 LOD_CHECK_STRIPED_IT(env, it, lo);
696 next = lo->ldo_stripe[it->lit_stripe_index];
697 LASSERT(next != NULL);
698 LASSERT(next->do_index_ops != NULL);
700 return next->do_index_ops->dio_it.store(env, it->lit_it);
704 * Implementation of dt_it_ops::load.
706 * Used with striped objects.
708 * \see dt_it_ops::load() in the API description for details.
710 static int lod_striped_it_load(const struct lu_env *env,
711 const struct dt_it *di, __u64 hash)
713 const struct lod_it *it = (const struct lod_it *)di;
714 struct lod_object *lo = lod_dt_obj(it->lit_obj);
715 struct dt_object *next;
717 LOD_CHECK_STRIPED_IT(env, it, lo);
719 next = lo->ldo_stripe[it->lit_stripe_index];
720 LASSERT(next != NULL);
721 LASSERT(next->do_index_ops != NULL);
723 return next->do_index_ops->dio_it.load(env, it->lit_it, hash);
726 static struct dt_index_operations lod_striped_index_ops = {
727 .dio_lookup = lod_index_lookup,
728 .dio_declare_insert = lod_declare_index_insert,
729 .dio_insert = lod_index_insert,
730 .dio_declare_delete = lod_declare_index_delete,
731 .dio_delete = lod_index_delete,
733 .init = lod_striped_it_init,
734 .fini = lod_striped_it_fini,
735 .get = lod_striped_it_get,
736 .put = lod_striped_it_put,
737 .next = lod_striped_it_next,
738 .key = lod_striped_it_key,
739 .key_size = lod_striped_it_key_size,
740 .rec = lod_striped_it_rec,
741 .rec_size = lod_striped_it_rec_size,
742 .store = lod_striped_it_store,
743 .load = lod_striped_it_load,
748 * Append the FID for each shard of the striped directory after the
749 * given LMV EA header.
751 * To simplify striped directory and the consistency verification,
752 * we only store the LMV EA header on disk, for both master object
753 * and slave objects. When someone wants to know the whole LMV EA,
754 * such as client readdir(), we can build the entrie LMV EA on the
755 * MDT side (in RAM) via iterating the sub-directory entries that
756 * are contained in the master object of the stripe directory.
758 * For the master object of the striped directroy, the valid name
759 * for each shard is composed of the ${shard_FID}:${shard_idx}.
761 * There may be holes in the LMV EA if some shards' name entries
762 * are corrupted or lost.
764 * \param[in] env pointer to the thread context
765 * \param[in] lo pointer to the master object of the striped directory
766 * \param[in] buf pointer to the lu_buf which will hold the LMV EA
767 * \param[in] resize whether re-allocate the buffer if it is not big enough
769 * \retval positive size of the LMV EA
770 * \retval 0 for nothing to be loaded
771 * \retval negative error number on failure
773 int lod_load_lmv_shards(const struct lu_env *env, struct lod_object *lo,
774 struct lu_buf *buf, bool resize)
776 struct lu_dirent *ent =
777 (struct lu_dirent *)lod_env_info(env)->lti_key;
778 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
779 struct dt_object *obj = dt_object_child(&lo->ldo_obj);
780 struct lmv_mds_md_v1 *lmv1 = buf->lb_buf;
782 const struct dt_it_ops *iops;
784 __u32 magic = le32_to_cpu(lmv1->lmv_magic);
789 /* If it is not a striped directory, then load nothing. */
790 if (magic != LMV_MAGIC_V1)
793 /* If it is in migration (or failure), then load nothing. */
794 if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
797 stripes = le32_to_cpu(lmv1->lmv_stripe_count);
801 rc = lmv_mds_md_size(stripes, magic);
805 if (buf->lb_len < lmv1_size) {
814 lu_buf_alloc(buf, lmv1_size);
819 memcpy(buf->lb_buf, tbuf.lb_buf, tbuf.lb_len);
822 if (unlikely(!dt_try_as_dir(env, obj)))
825 memset(&lmv1->lmv_stripe_fids[0], 0, stripes * sizeof(struct lu_fid));
826 iops = &obj->do_index_ops->dio_it;
827 it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA);
831 rc = iops->load(env, it, 0);
833 rc = iops->next(env, it);
838 char name[FID_LEN + 2] = "";
843 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
849 fid_le_to_cpu(&fid, &ent->lde_fid);
850 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
851 if (ent->lde_name[0] == '.') {
852 if (ent->lde_namelen == 1)
855 if (ent->lde_namelen == 2 && ent->lde_name[1] == '.')
859 len = snprintf(name, FID_LEN + 1, DFID":", PFID(&ent->lde_fid));
860 /* The ent->lde_name is composed of ${FID}:${index} */
861 if (ent->lde_namelen < len + 1 ||
862 memcmp(ent->lde_name, name, len) != 0) {
863 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
864 "%s: invalid shard name %.*s with the FID "DFID
865 " for the striped directory "DFID", %s\n",
866 lod2obd(lod)->obd_name, ent->lde_namelen,
867 ent->lde_name, PFID(&fid),
868 PFID(lu_object_fid(&obj->do_lu)),
869 lod->lod_lmv_failout ? "failout" : "skip");
871 if (lod->lod_lmv_failout)
879 if (ent->lde_name[len] < '0' ||
880 ent->lde_name[len] > '9') {
881 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
882 "%s: invalid shard name %.*s with the "
883 "FID "DFID" for the striped directory "
885 lod2obd(lod)->obd_name, ent->lde_namelen,
886 ent->lde_name, PFID(&fid),
887 PFID(lu_object_fid(&obj->do_lu)),
888 lod->lod_lmv_failout ?
891 if (lod->lod_lmv_failout)
897 index = index * 10 + ent->lde_name[len++] - '0';
898 } while (len < ent->lde_namelen);
900 if (len == ent->lde_namelen) {
901 /* Out of LMV EA range. */
902 if (index >= stripes) {
903 CERROR("%s: the shard %.*s for the striped "
904 "directory "DFID" is out of the known "
905 "LMV EA range [0 - %u], failout\n",
906 lod2obd(lod)->obd_name, ent->lde_namelen,
908 PFID(lu_object_fid(&obj->do_lu)),
914 /* The slot has been occupied. */
915 if (!fid_is_zero(&lmv1->lmv_stripe_fids[index])) {
919 &lmv1->lmv_stripe_fids[index]);
920 CERROR("%s: both the shard "DFID" and "DFID
921 " for the striped directory "DFID
922 " claim the same LMV EA slot at the "
923 "index %d, failout\n",
924 lod2obd(lod)->obd_name,
925 PFID(&fid0), PFID(&fid),
926 PFID(lu_object_fid(&obj->do_lu)), index);
931 /* stored as LE mode */
932 lmv1->lmv_stripe_fids[index] = ent->lde_fid;
935 rc = iops->next(env, it);
942 RETURN(rc > 0 ? lmv_mds_md_size(stripes, magic) : rc);
946 * Implementation of dt_object_operations::do_index_try.
948 * \see dt_object_operations::do_index_try() in the API description for details.
950 static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
951 const struct dt_index_features *feat)
953 struct lod_object *lo = lod_dt_obj(dt);
954 struct dt_object *next = dt_object_child(dt);
958 LASSERT(next->do_ops);
959 LASSERT(next->do_ops->do_index_try);
961 rc = lod_load_striping_locked(env, lo);
965 rc = next->do_ops->do_index_try(env, next, feat);
969 if (lo->ldo_stripenr > 0) {
972 for (i = 0; i < lo->ldo_stripenr; i++) {
973 if (dt_object_exists(lo->ldo_stripe[i]) == 0)
975 rc = lo->ldo_stripe[i]->do_ops->do_index_try(env,
976 lo->ldo_stripe[i], feat);
980 dt->do_index_ops = &lod_striped_index_ops;
982 dt->do_index_ops = &lod_index_ops;
989 * Implementation of dt_object_operations::do_read_lock.
991 * \see dt_object_operations::do_read_lock() in the API description for details.
993 static void lod_object_read_lock(const struct lu_env *env,
994 struct dt_object *dt, unsigned role)
996 dt_read_lock(env, dt_object_child(dt), role);
1000 * Implementation of dt_object_operations::do_write_lock.
1002 * \see dt_object_operations::do_write_lock() in the API description for
1005 static void lod_object_write_lock(const struct lu_env *env,
1006 struct dt_object *dt, unsigned role)
1008 dt_write_lock(env, dt_object_child(dt), role);
1012 * Implementation of dt_object_operations::do_read_unlock.
1014 * \see dt_object_operations::do_read_unlock() in the API description for
1017 static void lod_object_read_unlock(const struct lu_env *env,
1018 struct dt_object *dt)
1020 dt_read_unlock(env, dt_object_child(dt));
1024 * Implementation of dt_object_operations::do_write_unlock.
1026 * \see dt_object_operations::do_write_unlock() in the API description for
1029 static void lod_object_write_unlock(const struct lu_env *env,
1030 struct dt_object *dt)
1032 dt_write_unlock(env, dt_object_child(dt));
1036 * Implementation of dt_object_operations::do_write_locked.
1038 * \see dt_object_operations::do_write_locked() in the API description for
1041 static int lod_object_write_locked(const struct lu_env *env,
1042 struct dt_object *dt)
1044 return dt_write_locked(env, dt_object_child(dt));
1048 * Implementation of dt_object_operations::do_attr_get.
1050 * \see dt_object_operations::do_attr_get() in the API description for details.
1052 static int lod_attr_get(const struct lu_env *env,
1053 struct dt_object *dt,
1054 struct lu_attr *attr,
1055 struct lustre_capa *capa)
1057 /* Note: for striped directory, client will merge attributes
1058 * from all of the sub-stripes see lmv_merge_attr(), and there
1059 * no MDD logic depend on directory nlink/size/time, so we can
1060 * always use master inode nlink and size for now. */
1061 return dt_attr_get(env, dt_object_child(dt), attr, capa);
1065 * Mark all of the striped directory sub-stripes dead.
1067 * When a striped object is a subject to removal, we have
1068 * to mark all the stripes to prevent further access to
1069 * them (e.g. create a new file in those). So we mark
1070 * all the stripes with LMV_HASH_FLAG_DEAD. The function
1071 * can be used to declare the changes and to apply them.
1072 * If the object isn't striped, then just return success.
1074 * \param[in] env execution environment
1075 * \param[in] dt the striped object
1076 * \param[in] handle transaction handle
1077 * \param[in] declare whether to declare the change or apply
1079 * \retval 0 on success
1080 * \retval negative if failed
1082 static int lod_mark_dead_object(const struct lu_env *env,
1083 struct dt_object *dt,
1084 struct thandle *handle,
1087 struct lod_object *lo = lod_dt_obj(dt);
1088 struct lmv_mds_md_v1 *lmv;
1089 __u32 dead_hash_type;
1095 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
1098 rc = lod_load_striping_locked(env, lo);
1102 if (lo->ldo_stripenr == 0)
1105 rc = lod_get_lmv_ea(env, lo);
1109 lmv = lod_env_info(env)->lti_ea_store;
1110 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1111 dead_hash_type = le32_to_cpu(lmv->lmv_hash_type) | LMV_HASH_FLAG_DEAD;
1112 lmv->lmv_hash_type = cpu_to_le32(dead_hash_type);
1113 for (i = 0; i < lo->ldo_stripenr; i++) {
1116 lmv->lmv_master_mdt_index = i;
1118 buf.lb_len = sizeof(*lmv);
1120 rc = dt_declare_xattr_set(env, lo->ldo_stripe[i], &buf,
1122 LU_XATTR_REPLACE, handle);
1124 rc = dt_xattr_set(env, lo->ldo_stripe[i], &buf,
1125 XATTR_NAME_LMV, LU_XATTR_REPLACE,
1126 handle, BYPASS_CAPA);
1136 * Implementation of dt_object_operations::do_declare_attr_set.
1138 * If the object is striped, then apply the changes to all the stripes.
1140 * \see dt_object_operations::do_declare_attr_set() in the API description
1143 static int lod_declare_attr_set(const struct lu_env *env,
1144 struct dt_object *dt,
1145 const struct lu_attr *attr,
1146 struct thandle *handle)
1148 struct dt_object *next = dt_object_child(dt);
1149 struct lod_object *lo = lod_dt_obj(dt);
1153 /* Set dead object on all other stripes */
1154 if (attr->la_valid & LA_FLAGS && !(attr->la_valid & ~LA_FLAGS) &&
1155 attr->la_flags & LUSTRE_SLAVE_DEAD_FL) {
1156 rc = lod_mark_dead_object(env, dt, handle, true);
1161 * declare setattr on the local object
1163 rc = dt_declare_attr_set(env, next, attr, handle);
1167 /* osp_declare_attr_set() ignores all attributes other than
1168 * UID, GID, and size, and osp_attr_set() ignores all but UID
1169 * and GID. Declaration of size attr setting happens through
1170 * lod_declare_init_size(), and not through this function.
1171 * Therefore we need not load striping unless ownership is
1172 * changing. This should save memory and (we hope) speed up
1174 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1175 if (!(attr->la_valid & (LA_UID | LA_GID)))
1178 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1181 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE |
1182 LA_ATIME | LA_MTIME | LA_CTIME)))
1186 * load striping information, notice we don't do this when object
1187 * is being initialized as we don't need this information till
1188 * few specific cases like destroy, chown
1190 rc = lod_load_striping(env, lo);
1194 if (lo->ldo_stripenr == 0)
1198 * if object is striped declare changes on the stripes
1200 LASSERT(lo->ldo_stripe);
1201 for (i = 0; i < lo->ldo_stripenr; i++) {
1202 if (likely(lo->ldo_stripe[i] != NULL)) {
1203 rc = dt_declare_attr_set(env, lo->ldo_stripe[i], attr,
1206 CERROR("failed declaration: %d\n", rc);
1212 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE) &&
1213 dt_object_exists(next) != 0 &&
1214 dt_object_remote(next) == 0)
1215 dt_declare_xattr_del(env, next, XATTR_NAME_LOV, handle);
1217 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) &&
1218 dt_object_exists(next) &&
1219 dt_object_remote(next) == 0 && S_ISREG(attr->la_mode)) {
1220 struct lod_thread_info *info = lod_env_info(env);
1221 struct lu_buf *buf = &info->lti_buf;
1223 buf->lb_buf = info->lti_ea_store;
1224 buf->lb_len = info->lti_ea_store_size;
1225 dt_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1226 LU_XATTR_REPLACE, handle);
1233 * Implementation of dt_object_operations::do_attr_set.
1235 * If the object is striped, then apply the changes to all or subset of
1236 * the stripes depending on the object type and specific attributes.
1238 * \see dt_object_operations::do_attr_set() in the API description for details.
1240 static int lod_attr_set(const struct lu_env *env,
1241 struct dt_object *dt,
1242 const struct lu_attr *attr,
1243 struct thandle *handle,
1244 struct lustre_capa *capa)
1246 struct dt_object *next = dt_object_child(dt);
1247 struct lod_object *lo = lod_dt_obj(dt);
1251 /* Set dead object on all other stripes */
1252 if (attr->la_valid & LA_FLAGS && !(attr->la_valid & ~LA_FLAGS) &&
1253 attr->la_flags & LUSTRE_SLAVE_DEAD_FL) {
1254 rc = lod_mark_dead_object(env, dt, handle, false);
1259 * apply changes to the local object
1261 rc = dt_attr_set(env, next, attr, handle, capa);
1265 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1266 if (!(attr->la_valid & (LA_UID | LA_GID)))
1269 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1272 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE |
1273 LA_ATIME | LA_MTIME | LA_CTIME)))
1277 if (lo->ldo_stripenr == 0)
1281 * if object is striped, apply changes to all the stripes
1283 LASSERT(lo->ldo_stripe);
1284 for (i = 0; i < lo->ldo_stripenr; i++) {
1285 if (unlikely(lo->ldo_stripe[i] == NULL))
1287 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
1288 (dt_object_exists(lo->ldo_stripe[i]) == 0))
1291 rc = dt_attr_set(env, lo->ldo_stripe[i], attr, handle, capa);
1293 CERROR("failed declaration: %d\n", rc);
1298 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE) &&
1299 dt_object_exists(next) != 0 &&
1300 dt_object_remote(next) == 0)
1301 dt_xattr_del(env, next, XATTR_NAME_LOV, handle, BYPASS_CAPA);
1303 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) &&
1304 dt_object_exists(next) &&
1305 dt_object_remote(next) == 0 && S_ISREG(attr->la_mode)) {
1306 struct lod_thread_info *info = lod_env_info(env);
1307 struct lu_buf *buf = &info->lti_buf;
1308 struct ost_id *oi = &info->lti_ostid;
1309 struct lu_fid *fid = &info->lti_fid;
1310 struct lov_mds_md_v1 *lmm;
1311 struct lov_ost_data_v1 *objs;
1315 rc1 = lod_get_lov_ea(env, lo);
1319 buf->lb_buf = info->lti_ea_store;
1320 buf->lb_len = info->lti_ea_store_size;
1321 lmm = info->lti_ea_store;
1322 magic = le32_to_cpu(lmm->lmm_magic);
1323 if (magic == LOV_MAGIC_V1)
1324 objs = &(lmm->lmm_objects[0]);
1326 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1327 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1328 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1330 fid_to_ostid(fid, oi);
1331 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1332 dt_xattr_set(env, next, buf, XATTR_NAME_LOV,
1333 LU_XATTR_REPLACE, handle, BYPASS_CAPA);
1340 * Implementation of dt_object_operations::do_xattr_get.
1342 * If LOV EA is requested from the root object and it's not
1343 * found, then return default striping for the filesystem.
1345 * \see dt_object_operations::do_xattr_get() in the API description for details.
1347 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1348 struct lu_buf *buf, const char *name,
1349 struct lustre_capa *capa)
1351 struct lod_thread_info *info = lod_env_info(env);
1352 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1356 rc = dt_xattr_get(env, dt_object_child(dt), buf, name, capa);
1357 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1358 struct lmv_mds_md_v1 *lmv1;
1361 if (rc > (typeof(rc))sizeof(*lmv1))
1364 if (rc < (typeof(rc))sizeof(*lmv1))
1365 RETURN(rc = rc > 0 ? -EINVAL : rc);
1367 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1368 CLASSERT(sizeof(*lmv1) <= sizeof(info->lti_key));
1370 info->lti_buf.lb_buf = info->lti_key;
1371 info->lti_buf.lb_len = sizeof(*lmv1);
1372 rc = dt_xattr_get(env, dt_object_child(dt),
1373 &info->lti_buf, name, capa);
1374 if (unlikely(rc != sizeof(*lmv1)))
1375 RETURN(rc = rc > 0 ? -EINVAL : rc);
1377 lmv1 = info->lti_buf.lb_buf;
1378 /* The on-disk LMV EA only contains header, but the
1379 * returned LMV EA size should contain the space for
1380 * the FIDs of all shards of the striped directory. */
1381 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1382 rc = lmv_mds_md_size(
1383 le32_to_cpu(lmv1->lmv_stripe_count),
1386 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1390 RETURN(rc = rc1 != 0 ? rc1 : rc);
1393 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1397 * lod returns default striping on the real root of the device
1398 * this is like the root stores default striping for the whole
1399 * filesystem. historically we've been using a different approach
1400 * and store it in the config.
1402 dt_root_get(env, dev->lod_child, &info->lti_fid);
1403 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1405 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1406 struct lov_user_md *lum = buf->lb_buf;
1407 struct lov_desc *desc = &dev->lod_desc;
1409 if (buf->lb_buf == NULL) {
1411 } else if (buf->lb_len >= sizeof(*lum)) {
1412 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1413 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1414 lmm_oi_set_id(&lum->lmm_oi, 0);
1415 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1416 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1417 lum->lmm_stripe_size = cpu_to_le32(
1418 desc->ld_default_stripe_size);
1419 lum->lmm_stripe_count = cpu_to_le16(
1420 desc->ld_default_stripe_count);
1421 lum->lmm_stripe_offset = cpu_to_le16(
1422 desc->ld_default_stripe_offset);
1435 * Checks that the magic and the number of the stripes are sane.
1437 * \param[in] lod lod device
1438 * \param[in] lum a buffer storing LMV EA to verify
1440 * \retval 0 if the EA is sane
1441 * \retval negative otherwise
1443 static int lod_verify_md_striping(struct lod_device *lod,
1444 const struct lmv_user_md_v1 *lum)
1449 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC))
1450 GOTO(out, rc = -EINVAL);
1452 if (unlikely(le32_to_cpu(lum->lum_stripe_count) == 0))
1453 GOTO(out, rc = -EINVAL);
1456 CERROR("%s: invalid lmv_user_md: magic = %x, "
1457 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1458 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1459 (int)le32_to_cpu(lum->lum_stripe_offset),
1460 le32_to_cpu(lum->lum_stripe_count), rc);
1465 * Initialize LMV EA for a slave.
1467 * Initialize slave's LMV EA from the master's LMV EA.
1469 * \param[in] master_lmv a buffer containing master's EA
1470 * \param[out] slave_lmv a buffer where slave's EA will be stored
1473 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1474 const struct lmv_mds_md_v1 *master_lmv)
1476 *slave_lmv = *master_lmv;
1477 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1483 * Generate LMV EA from the object passed as \a dt. The object must have
1484 * the stripes created and initialized.
1486 * \param[in] env execution environment
1487 * \param[in] dt object
1488 * \param[out] lmv_buf buffer storing generated LMV EA
1490 * \retval 0 on success
1491 * \retval negative if failed
1493 int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1494 struct lu_buf *lmv_buf)
1496 struct lod_thread_info *info = lod_env_info(env);
1497 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1498 struct lod_object *lo = lod_dt_obj(dt);
1499 struct lmv_mds_md_v1 *lmm1;
1501 int type = LU_SEQ_RANGE_ANY;
1506 LASSERT(lo->ldo_dir_striped != 0);
1507 LASSERT(lo->ldo_stripenr > 0);
1508 stripe_count = lo->ldo_stripenr;
1509 /* Only store the LMV EA heahder on the disk. */
1510 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1511 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1515 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1518 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1519 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1520 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1521 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1522 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1527 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1528 lmv_buf->lb_buf = info->lti_ea_store;
1529 lmv_buf->lb_len = sizeof(*lmm1);
1530 lo->ldo_dir_striping_cached = 1;
1536 * Create in-core represenation for a striped directory.
1538 * Parse the buffer containing LMV EA and instantiate LU objects
1539 * representing the stripe objects. The pointers to the objects are
1540 * stored in ldo_stripe field of \a lo. This function is used when
1541 * we need to access an already created object (i.e. load from a disk).
1543 * \param[in] env execution environment
1544 * \param[in] lo lod object
1545 * \param[in] buf buffer containing LMV EA
1547 * \retval 0 on success
1548 * \retval negative if failed
1550 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1551 const struct lu_buf *buf)
1553 struct lod_thread_info *info = lod_env_info(env);
1554 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1555 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1556 struct dt_object **stripe;
1557 union lmv_mds_md *lmm = buf->lb_buf;
1558 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1559 struct lu_fid *fid = &info->lti_fid;
1564 if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
1567 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1568 lo->ldo_dir_slave_stripe = 1;
1572 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1575 if (le32_to_cpu(lmv1->lmv_stripe_count) < 1)
1578 LASSERT(lo->ldo_stripe == NULL);
1579 OBD_ALLOC(stripe, sizeof(stripe[0]) *
1580 (le32_to_cpu(lmv1->lmv_stripe_count)));
1584 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1585 struct dt_device *tgt_dt;
1586 struct dt_object *dto;
1587 int type = LU_SEQ_RANGE_ANY;
1590 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1591 if (!fid_is_sane(fid))
1592 GOTO(out, rc = -ESTALE);
1594 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1598 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1599 tgt_dt = lod->lod_child;
1601 struct lod_tgt_desc *tgt;
1603 tgt = LTD_TGT(ltd, idx);
1605 GOTO(out, rc = -ESTALE);
1606 tgt_dt = tgt->ltd_tgt;
1609 dto = dt_locate_at(env, tgt_dt, fid,
1610 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1613 GOTO(out, rc = PTR_ERR(dto));
1618 lo->ldo_stripe = stripe;
1619 lo->ldo_stripenr = le32_to_cpu(lmv1->lmv_stripe_count);
1620 lo->ldo_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1622 lod_object_free_striping(env, lo);
1628 * Create a striped directory.
1630 * Create a striped directory with a given stripe pattern on the specified MDTs.
1631 * A striped directory is represented as a regular directory - an index listing
1632 * all the stripes. The stripes point back to the master object with ".." and
1633 * LinkEA. The master object gets LMV EA which identifies it as a striped
1634 * directory. The function allocates FIDs for all the stripes.
1636 * \param[in] env execution environment
1637 * \param[in] dt object
1638 * \param[in] attr attributes to initialize the objects with
1639 * \param[in] lum a pattern specifying the number of stripes and
1641 * \param[in] dof type of objects to be created
1642 * \param[in] th transaction handle
1644 * \retval 0 on success
1645 * \retval negative if failed
1647 static int lod_prep_md_striped_create(const struct lu_env *env,
1648 struct dt_object *dt,
1649 struct lu_attr *attr,
1650 const struct lmv_user_md_v1 *lum,
1651 struct dt_object_format *dof,
1654 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1655 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1656 struct lod_object *lo = lod_dt_obj(dt);
1657 struct lod_thread_info *info = lod_env_info(env);
1658 struct dt_object **stripe;
1659 struct lu_buf lmv_buf;
1660 struct lu_buf slave_lmv_buf;
1661 struct lmv_mds_md_v1 *lmm;
1662 struct lmv_mds_md_v1 *slave_lmm = NULL;
1663 struct dt_insert_rec *rec = &info->lti_dt_rec;
1671 /* The lum has been verifed in lod_verify_md_striping */
1672 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
1673 LASSERT(le32_to_cpu(lum->lum_stripe_count) > 0);
1675 stripe_count = le32_to_cpu(lum->lum_stripe_count);
1677 /* shrink the stripe_count to the avaible MDT count */
1678 if (stripe_count > lod->lod_remote_mdt_count + 1)
1679 stripe_count = lod->lod_remote_mdt_count + 1;
1681 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_count);
1685 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
1686 if (idx_array == NULL)
1687 GOTO(out_free, rc = -ENOMEM);
1689 for (i = 0; i < stripe_count; i++) {
1690 struct lod_tgt_desc *tgt = NULL;
1691 struct dt_object *dto;
1692 struct lu_fid fid = { 0 };
1694 struct lu_object_conf conf = { 0 };
1695 struct dt_device *tgt_dt = NULL;
1698 /* Right now, master stripe and master object are
1699 * on the same MDT */
1700 idx = le32_to_cpu(lum->lum_stripe_offset);
1701 rc = obd_fid_alloc(env, lod->lod_child_exp, &fid,
1705 tgt_dt = lod->lod_child;
1709 idx = (idx_array[i - 1] + 1) % (lod->lod_remote_mdt_count + 1);
1711 for (j = 0; j < lod->lod_remote_mdt_count;
1712 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
1713 bool already_allocated = false;
1716 CDEBUG(D_INFO, "try idx %d, mdt cnt %u,"
1717 " allocated %u, last allocated %d\n", idx,
1718 lod->lod_remote_mdt_count, i, idx_array[i - 1]);
1720 /* Find next available target */
1721 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx))
1724 /* check whether the idx already exists
1725 * in current allocated array */
1726 for (k = 0; k < i; k++) {
1727 if (idx_array[k] == idx) {
1728 already_allocated = true;
1733 if (already_allocated)
1736 /* check the status of the OSP */
1737 tgt = LTD_TGT(ltd, idx);
1741 tgt_dt = tgt->ltd_tgt;
1742 rc = dt_statfs(env, tgt_dt, NULL);
1744 /* this OSP doesn't feel well */
1749 rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
1758 /* Can not allocate more stripes */
1759 if (j == lod->lod_remote_mdt_count) {
1760 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
1761 lod2obd(lod)->obd_name, stripe_count, i - 1);
1765 CDEBUG(D_INFO, "idx %d, mdt cnt %u,"
1766 " allocated %u, last allocated %d\n", idx,
1767 lod->lod_remote_mdt_count, i, idx_array[i - 1]);
1770 /* tgt_dt and fid must be ready after search avaible OSP
1771 * in the above loop */
1772 LASSERT(tgt_dt != NULL);
1773 LASSERT(fid_is_sane(&fid));
1774 conf.loc_flags = LOC_F_NEW;
1775 dto = dt_locate_at(env, tgt_dt, &fid,
1776 dt->do_lu.lo_dev->ld_site->ls_top_dev,
1779 GOTO(out_put, rc = PTR_ERR(dto));
1784 lo->ldo_dir_striped = 1;
1785 lo->ldo_stripe = stripe;
1786 lo->ldo_stripenr = i;
1787 lo->ldo_stripes_allocated = stripe_count;
1789 if (lo->ldo_stripenr == 0)
1790 GOTO(out_put, rc = -ENOSPC);
1792 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1795 lmm = lmv_buf.lb_buf;
1797 OBD_ALLOC_PTR(slave_lmm);
1798 if (slave_lmm == NULL)
1799 GOTO(out_put, rc = -ENOMEM);
1801 lod_prep_slave_lmv_md(slave_lmm, lmm);
1802 slave_lmv_buf.lb_buf = slave_lmm;
1803 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1805 if (!dt_try_as_dir(env, dt_object_child(dt)))
1806 GOTO(out_put, rc = -EINVAL);
1808 rec->rec_type = S_IFDIR;
1809 for (i = 0; i < lo->ldo_stripenr; i++) {
1810 struct dt_object *dto = stripe[i];
1811 char *stripe_name = info->lti_key;
1812 struct lu_name *sname;
1813 struct linkea_data ldata = { 0 };
1814 struct lu_buf linkea_buf;
1816 rc = dt_declare_create(env, dto, attr, NULL, dof, th);
1820 if (!dt_try_as_dir(env, dto))
1821 GOTO(out_put, rc = -EINVAL);
1823 rec->rec_fid = lu_object_fid(&dto->do_lu);
1824 rc = dt_declare_insert(env, dto, (const struct dt_rec *)rec,
1825 (const struct dt_key *)dot, th);
1829 /* master stripe FID will be put to .. */
1830 rec->rec_fid = lu_object_fid(&dt->do_lu);
1831 rc = dt_declare_insert(env, dto, (const struct dt_rec *)rec,
1832 (const struct dt_key *)dotdot, th);
1836 /* probably nothing to inherite */
1837 if (lo->ldo_striping_cached &&
1838 !LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size,
1839 lo->ldo_def_stripenr,
1840 lo->ldo_def_stripe_offset,
1842 struct lov_user_md_v3 *v3;
1844 /* sigh, lti_ea_store has been used for lmv_buf,
1845 * so we have to allocate buffer for default
1849 GOTO(out_put, rc = -ENOMEM);
1851 memset(v3, 0, sizeof(*v3));
1852 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
1853 v3->lmm_stripe_count =
1854 cpu_to_le16(lo->ldo_def_stripenr);
1855 v3->lmm_stripe_offset =
1856 cpu_to_le16(lo->ldo_def_stripe_offset);
1857 v3->lmm_stripe_size =
1858 cpu_to_le32(lo->ldo_def_stripe_size);
1859 if (lo->ldo_pool != NULL)
1860 strlcpy(v3->lmm_pool_name, lo->ldo_pool,
1861 sizeof(v3->lmm_pool_name));
1863 info->lti_buf.lb_buf = v3;
1864 info->lti_buf.lb_len = sizeof(*v3);
1865 rc = dt_declare_xattr_set(env, dto,
1874 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1875 cfs_fail_val != i) {
1876 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1878 slave_lmm->lmv_master_mdt_index =
1881 slave_lmm->lmv_master_mdt_index =
1883 rc = dt_declare_xattr_set(env, dto, &slave_lmv_buf,
1884 XATTR_NAME_LMV, 0, th);
1889 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1891 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1892 PFID(lu_object_fid(&dto->do_lu)), i + 1);
1894 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1895 PFID(lu_object_fid(&dto->do_lu)), i);
1897 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
1898 rc = linkea_data_new(&ldata, &info->lti_linkea_buf);
1902 rc = linkea_add_buf(&ldata, sname, lu_object_fid(&dt->do_lu));
1906 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1907 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1908 rc = dt_declare_xattr_set(env, dto, &linkea_buf,
1909 XATTR_NAME_LINK, 0, th);
1913 rec->rec_fid = lu_object_fid(&dto->do_lu);
1914 rc = dt_declare_insert(env, dt_object_child(dt),
1915 (const struct dt_rec *)rec,
1916 (const struct dt_key *)stripe_name, th);
1920 rc = dt_declare_ref_add(env, dt_object_child(dt), th);
1925 rc = dt_declare_xattr_set(env, dt_object_child(dt), &lmv_buf,
1926 XATTR_NAME_LMV, 0, th);
1932 for (i = 0; i < stripe_count; i++)
1933 if (stripe[i] != NULL)
1934 lu_object_put(env, &stripe[i]->do_lu);
1935 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_count);
1936 lo->ldo_stripenr = 0;
1937 lo->ldo_stripes_allocated = 0;
1938 lo->ldo_stripe = NULL;
1942 if (idx_array != NULL)
1943 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
1944 if (slave_lmm != NULL)
1945 OBD_FREE_PTR(slave_lmm);
1951 * Declare create striped md object.
1953 * The function declares intention to create a striped directory. This is a
1954 * wrapper for lod_prep_md_striped_create(). The only additional functionality
1955 * is to verify pattern \a lum_buf is good. Check that function for the details.
1957 * \param[in] env execution environment
1958 * \param[in] dt object
1959 * \param[in] attr attributes to initialize the objects with
1960 * \param[in] lum_buf a pattern specifying the number of stripes and
1962 * \param[in] dof type of objects to be created
1963 * \param[in] th transaction handle
1965 * \retval 0 on success
1966 * \retval negative if failed
1969 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
1970 struct dt_object *dt,
1971 struct lu_attr *attr,
1972 const struct lu_buf *lum_buf,
1973 struct dt_object_format *dof,
1976 struct lod_object *lo = lod_dt_obj(dt);
1977 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1978 struct lmv_user_md_v1 *lum;
1982 lum = lum_buf->lb_buf;
1983 LASSERT(lum != NULL);
1985 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
1986 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
1987 (int)le32_to_cpu(lum->lum_stripe_offset));
1989 if (le32_to_cpu(lum->lum_stripe_count) == 0)
1992 rc = lod_verify_md_striping(lod, lum);
1996 /* prepare dir striped objects */
1997 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
1999 /* failed to create striping, let's reset
2000 * config so that others don't get confused */
2001 lod_object_free_striping(env, lo);
2010 * Implementation of dt_object_operations::do_declare_xattr_set.
2012 * Used with regular (non-striped) objects. Basically it
2013 * initializes the striping information and applies the
2014 * change to all the stripes.
2016 * \see dt_object_operations::do_declare_xattr_set() in the API description
2019 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2020 struct dt_object *dt,
2021 const struct lu_buf *buf,
2022 const char *name, int fl,
2025 struct dt_object *next = dt_object_child(dt);
2026 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2027 struct lod_object *lo = lod_dt_obj(dt);
2032 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2033 struct lmv_user_md_v1 *lum;
2035 LASSERT(buf != NULL && buf->lb_buf != NULL);
2037 rc = lod_verify_md_striping(d, lum);
2042 rc = dt_declare_xattr_set(env, next, buf, name, fl, th);
2046 /* set xattr to each stripes, if needed */
2047 rc = lod_load_striping(env, lo);
2051 /* Note: Do not set LinkEA on sub-stripes, otherwise
2052 * it will confuse the fid2path process(see mdt_path_current()).
2053 * The linkEA between master and sub-stripes is set in
2054 * lod_xattr_set_lmv(). */
2055 if (lo->ldo_stripenr == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
2058 for (i = 0; i < lo->ldo_stripenr; i++) {
2059 LASSERT(lo->ldo_stripe[i]);
2060 rc = dt_declare_xattr_set(env, lo->ldo_stripe[i], buf,
2070 * Implementation of dt_object_operations::do_declare_xattr_set.
2072 * \see dt_object_operations::do_declare_xattr_set() in the API description
2075 * the extension to the API:
2076 * - declaring LOVEA requests striping creation
2077 * - LU_XATTR_REPLACE means layout swap
2079 static int lod_declare_xattr_set(const struct lu_env *env,
2080 struct dt_object *dt,
2081 const struct lu_buf *buf,
2082 const char *name, int fl,
2085 struct dt_object *next = dt_object_child(dt);
2086 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2092 * allow to declare predefined striping on a new (!mode) object
2093 * which is supposed to be replay of regular file creation
2094 * (when LOV setting is declared)
2095 * LU_XATTR_REPLACE is set to indicate a layout swap
2097 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
2098 if ((S_ISREG(mode) || mode == 0) && strcmp(name, XATTR_NAME_LOV) == 0 &&
2099 !(fl & LU_XATTR_REPLACE)) {
2101 * this is a request to manipulate object's striping
2103 if (dt_object_exists(dt)) {
2104 rc = dt_attr_get(env, next, attr, BYPASS_CAPA);
2108 memset(attr, 0, sizeof(*attr));
2109 attr->la_valid = LA_TYPE | LA_MODE;
2110 attr->la_mode = S_IFREG;
2112 rc = lod_declare_striped_object(env, dt, attr, buf, th);
2113 } else if (S_ISDIR(mode)) {
2114 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
2116 rc = dt_declare_xattr_set(env, next, buf, name, fl, th);
2123 * Resets cached default striping in the object.
2125 * \param[in] lo object
2127 static void lod_lov_stripe_cache_clear(struct lod_object *lo)
2129 lo->ldo_striping_cached = 0;
2130 lo->ldo_def_striping_set = 0;
2131 lod_object_set_pool(lo, NULL);
2132 lo->ldo_def_stripe_size = 0;
2133 lo->ldo_def_stripenr = 0;
2134 if (lo->ldo_dir_stripe != NULL)
2135 lo->ldo_dir_striping_cached = 0;
2139 * Apply xattr changes to the object.
2141 * Applies xattr changes to the object and the stripes if the latter exist.
2143 * \param[in] env execution environment
2144 * \param[in] dt object
2145 * \param[in] buf buffer pointing to the new value of xattr
2146 * \param[in] name name of xattr
2147 * \param[in] fl flags
2148 * \param[in] th transaction handle
2149 * \param[in] capa not used currently
2151 * \retval 0 on success
2152 * \retval negative if failed
2154 static int lod_xattr_set_internal(const struct lu_env *env,
2155 struct dt_object *dt,
2156 const struct lu_buf *buf,
2157 const char *name, int fl, struct thandle *th,
2158 struct lustre_capa *capa)
2160 struct dt_object *next = dt_object_child(dt);
2161 struct lod_object *lo = lod_dt_obj(dt);
2166 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
2167 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2170 /* Note: Do not set LinkEA on sub-stripes, otherwise
2171 * it will confuse the fid2path process(see mdt_path_current()).
2172 * The linkEA between master and sub-stripes is set in
2173 * lod_xattr_set_lmv(). */
2174 if (lo->ldo_stripenr == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
2177 for (i = 0; i < lo->ldo_stripenr; i++) {
2178 LASSERT(lo->ldo_stripe[i]);
2179 rc = dt_xattr_set(env, lo->ldo_stripe[i], buf, name, fl, th,
2189 * Delete an extended attribute.
2191 * Deletes specified xattr from the object and the stripes if the latter exist.
2193 * \param[in] env execution environment
2194 * \param[in] dt object
2195 * \param[in] name name of xattr
2196 * \param[in] th transaction handle
2197 * \param[in] capa not used currently
2199 * \retval 0 on success
2200 * \retval negative if failed
2202 static int lod_xattr_del_internal(const struct lu_env *env,
2203 struct dt_object *dt,
2204 const char *name, struct thandle *th,
2205 struct lustre_capa *capa)
2207 struct dt_object *next = dt_object_child(dt);
2208 struct lod_object *lo = lod_dt_obj(dt);
2213 rc = dt_xattr_del(env, next, name, th, capa);
2214 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2217 if (lo->ldo_stripenr == 0)
2220 for (i = 0; i < lo->ldo_stripenr; i++) {
2221 LASSERT(lo->ldo_stripe[i]);
2222 rc = dt_xattr_del(env, lo->ldo_stripe[i], name, th,
2232 * Set default striping on a directory.
2234 * Sets specified striping on a directory object unless it matches the default
2235 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
2236 * EA. This striping will be used when regular file is being created in this
2239 * \param[in] env execution environment
2240 * \param[in] dt the striped object
2241 * \param[in] buf buffer with the striping
2242 * \param[in] name name of EA
2243 * \param[in] fl xattr flag (see OSD API description)
2244 * \param[in] th transaction handle
2245 * \param[in] capa not used
2247 * \retval 0 on success
2248 * \retval negative if failed
2250 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
2251 struct dt_object *dt,
2252 const struct lu_buf *buf,
2253 const char *name, int fl,
2255 struct lustre_capa *capa)
2257 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2258 struct lod_object *l = lod_dt_obj(dt);
2259 struct lov_user_md_v1 *lum;
2260 struct lov_user_md_v3 *v3 = NULL;
2261 const char *pool_name = NULL;
2265 /* If it is striped dir, we should clear the stripe cache for
2266 * slave stripe as well, but there are no effective way to
2267 * notify the LOD on the slave MDT, so we do not cache stripe
2268 * information for slave stripe for now. XXX*/
2269 lod_lov_stripe_cache_clear(l);
2270 LASSERT(buf != NULL && buf->lb_buf != NULL);
2273 rc = lod_verify_striping(d, buf, false);
2277 if (lum->lmm_magic == LOV_USER_MAGIC_V3) {
2279 if (v3->lmm_pool_name[0] != '\0')
2280 pool_name = v3->lmm_pool_name;
2283 /* if { size, offset, count } = { 0, -1, 0 } and no pool
2284 * (i.e. all default values specified) then delete default
2285 * striping from dir. */
2287 "set default striping: sz %u # %u offset %d %s %s\n",
2288 (unsigned)lum->lmm_stripe_size,
2289 (unsigned)lum->lmm_stripe_count,
2290 (int)lum->lmm_stripe_offset,
2291 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
2293 if (LOVEA_DELETE_VALUES(lum->lmm_stripe_size, lum->lmm_stripe_count,
2294 lum->lmm_stripe_offset, pool_name)) {
2295 rc = lod_xattr_del_internal(env, dt, name, th, capa);
2299 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th, capa);
2306 * Set default striping on a directory object.
2308 * Sets specified striping on a directory object unless it matches the default
2309 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
2310 * EA. This striping will be used when a new directory is being created in the
2313 * \param[in] env execution environment
2314 * \param[in] dt the striped object
2315 * \param[in] buf buffer with the striping
2316 * \param[in] name name of EA
2317 * \param[in] fl xattr flag (see OSD API description)
2318 * \param[in] th transaction handle
2319 * \param[in] capa not used
2321 * \retval 0 on success
2322 * \retval negative if failed
2324 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
2325 struct dt_object *dt,
2326 const struct lu_buf *buf,
2327 const char *name, int fl,
2329 struct lustre_capa *capa)
2331 struct lod_object *l = lod_dt_obj(dt);
2332 struct lmv_user_md_v1 *lum;
2336 LASSERT(buf != NULL && buf->lb_buf != NULL);
2339 CDEBUG(D_OTHER, "set default stripe_count # %u stripe_offset %d\n",
2340 le32_to_cpu(lum->lum_stripe_count),
2341 (int)le32_to_cpu(lum->lum_stripe_offset));
2343 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
2344 le32_to_cpu(lum->lum_stripe_offset)) &&
2345 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
2346 rc = lod_xattr_del_internal(env, dt, name, th, capa);
2350 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th, capa);
2355 /* Update default stripe cache */
2356 if (l->ldo_dir_stripe == NULL) {
2357 OBD_ALLOC_PTR(l->ldo_dir_stripe);
2358 if (l->ldo_dir_stripe == NULL)
2362 l->ldo_dir_striping_cached = 0;
2363 l->ldo_dir_def_striping_set = 1;
2364 l->ldo_dir_def_stripenr = le32_to_cpu(lum->lum_stripe_count);
2370 * Turn directory into a striped directory.
2372 * During replay the client sends the striping created before MDT
2373 * failure, then the layer above LOD sends this defined striping
2374 * using ->do_xattr_set(), so LOD uses this method to replay creation
2375 * of the stripes. Notice the original information for the striping
2376 * (#stripes, FIDs, etc) was transfered in declare path.
2378 * \param[in] env execution environment
2379 * \param[in] dt the striped object
2380 * \param[in] buf not used currently
2381 * \param[in] name not used currently
2382 * \param[in] fl xattr flag (see OSD API description)
2383 * \param[in] th transaction handle
2384 * \param[in] capa not used
2386 * \retval 0 on success
2387 * \retval negative if failed
2389 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
2390 const struct lu_buf *buf, const char *name,
2391 int fl, struct thandle *th,
2392 struct lustre_capa *capa)
2394 struct lod_object *lo = lod_dt_obj(dt);
2395 struct lod_thread_info *info = lod_env_info(env);
2396 struct lu_attr *attr = &info->lti_attr;
2397 struct dt_object_format *dof = &info->lti_format;
2398 struct lu_buf lmv_buf;
2399 struct lu_buf slave_lmv_buf;
2400 struct lmv_mds_md_v1 *lmm;
2401 struct lmv_mds_md_v1 *slave_lmm = NULL;
2402 struct dt_insert_rec *rec = &info->lti_dt_rec;
2407 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
2410 /* The stripes are supposed to be allocated in declare phase,
2411 * if there are no stripes being allocated, it will skip */
2412 if (lo->ldo_stripenr == 0)
2415 rc = dt_attr_get(env, dt_object_child(dt), attr, BYPASS_CAPA);
2419 attr->la_valid = LA_TYPE | LA_MODE;
2420 dof->dof_type = DFT_DIR;
2422 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
2425 lmm = lmv_buf.lb_buf;
2427 OBD_ALLOC_PTR(slave_lmm);
2428 if (slave_lmm == NULL)
2431 lod_prep_slave_lmv_md(slave_lmm, lmm);
2432 slave_lmv_buf.lb_buf = slave_lmm;
2433 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
2435 rec->rec_type = S_IFDIR;
2436 for (i = 0; i < lo->ldo_stripenr; i++) {
2437 struct dt_object *dto;
2438 char *stripe_name = info->lti_key;
2439 struct lu_name *sname;
2440 struct linkea_data ldata = { 0 };
2441 struct lu_buf linkea_buf;
2443 dto = lo->ldo_stripe[i];
2444 dt_write_lock(env, dto, MOR_TGT_CHILD);
2445 rc = dt_create(env, dto, attr, NULL, dof, th);
2446 dt_write_unlock(env, dto);
2450 rec->rec_fid = lu_object_fid(&dto->do_lu);
2451 rc = dt_insert(env, dto, (const struct dt_rec *)rec,
2452 (const struct dt_key *)dot, th, capa, 0);
2456 rec->rec_fid = lu_object_fid(&dt->do_lu);
2457 rc = dt_insert(env, dto, (struct dt_rec *)rec,
2458 (const struct dt_key *)dotdot, th, capa, 0);
2462 if (lo->ldo_striping_cached &&
2463 !LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size,
2464 lo->ldo_def_stripenr,
2465 lo->ldo_def_stripe_offset,
2467 struct lov_user_md_v3 *v3;
2469 /* sigh, lti_ea_store has been used for lmv_buf,
2470 * so we have to allocate buffer for default
2476 memset(v3, 0, sizeof(*v3));
2477 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
2478 v3->lmm_stripe_count =
2479 cpu_to_le16(lo->ldo_def_stripenr);
2480 v3->lmm_stripe_offset =
2481 cpu_to_le16(lo->ldo_def_stripe_offset);
2482 v3->lmm_stripe_size =
2483 cpu_to_le32(lo->ldo_def_stripe_size);
2484 if (lo->ldo_pool != NULL)
2485 strlcpy(v3->lmm_pool_name, lo->ldo_pool,
2486 sizeof(v3->lmm_pool_name));
2488 info->lti_buf.lb_buf = v3;
2489 info->lti_buf.lb_len = sizeof(*v3);
2490 rc = dt_xattr_set(env, dto, &info->lti_buf,
2491 XATTR_NAME_LOV, 0, th, capa);
2497 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
2498 cfs_fail_val != i) {
2499 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
2501 slave_lmm->lmv_master_mdt_index =
2504 slave_lmm->lmv_master_mdt_index =
2506 rc = dt_xattr_set(env, dto, &slave_lmv_buf,
2507 XATTR_NAME_LMV, fl, th, capa);
2512 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
2514 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2515 PFID(lu_object_fid(&dto->do_lu)), i + 1);
2517 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2518 PFID(lu_object_fid(&dto->do_lu)), i);
2520 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
2521 rc = linkea_data_new(&ldata, &info->lti_linkea_buf);
2525 rc = linkea_add_buf(&ldata, sname, lu_object_fid(&dt->do_lu));
2529 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
2530 linkea_buf.lb_len = ldata.ld_leh->leh_len;
2531 rc = dt_xattr_set(env, dto, &linkea_buf, XATTR_NAME_LINK,
2532 0, th, BYPASS_CAPA);
2536 rec->rec_fid = lu_object_fid(&dto->do_lu);
2537 rc = dt_insert(env, dt_object_child(dt),
2538 (const struct dt_rec *)rec,
2539 (const struct dt_key *)stripe_name, th, capa, 0);
2543 rc = dt_ref_add(env, dt_object_child(dt), th);
2548 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
2549 rc = dt_xattr_set(env, dt_object_child(dt), &lmv_buf,
2550 XATTR_NAME_LMV, fl, th, capa);
2553 if (slave_lmm != NULL)
2554 OBD_FREE_PTR(slave_lmm);
2560 * Helper function to declare/execute creation of a striped directory
2562 * Called in declare/create object path, prepare striping for a directory
2563 * and prepare defaults data striping for the objects to be created in
2564 * that directory. Notice the function calls "declaration" or "execution"
2565 * methods depending on \a declare param. This is a consequence of the
2566 * current approach while we don't have natural distributed transactions:
2567 * we basically execute non-local updates in the declare phase. So, the
2568 * arguments for the both phases are the same and this is the reason for
2569 * this function to exist.
2571 * \param[in] env execution environment
2572 * \param[in] dt object
2573 * \param[in] attr attributes the stripes will be created with
2574 * \param[in] dof format of stripes (see OSD API description)
2575 * \param[in] th transaction handle
2576 * \param[in] declare where to call "declare" or "execute" methods
2578 * \retval 0 on success
2579 * \retval negative if failed
2581 int lod_dir_striping_create_internal(const struct lu_env *env,
2582 struct dt_object *dt,
2583 struct lu_attr *attr,
2584 struct dt_object_format *dof,
2588 struct lod_thread_info *info = lod_env_info(env);
2589 struct lod_object *lo = lod_dt_obj(dt);
2593 if (!LMVEA_DELETE_VALUES(lo->ldo_stripenr,
2594 lo->ldo_dir_stripe_offset)) {
2595 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
2596 int stripe_count = lo->ldo_stripenr;
2598 if (info->lti_ea_store_size < sizeof(*v1)) {
2599 rc = lod_ea_store_resize(info, sizeof(*v1));
2602 v1 = info->lti_ea_store;
2605 memset(v1, 0, sizeof(*v1));
2606 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
2607 v1->lum_stripe_count = cpu_to_le32(stripe_count);
2608 v1->lum_stripe_offset =
2609 cpu_to_le32(lo->ldo_dir_stripe_offset);
2611 info->lti_buf.lb_buf = v1;
2612 info->lti_buf.lb_len = sizeof(*v1);
2615 rc = lod_declare_xattr_set_lmv(env, dt, attr,
2616 &info->lti_buf, dof, th);
2618 rc = lod_xattr_set_lmv(env, dt, &info->lti_buf,
2619 XATTR_NAME_LMV, 0, th,
2625 /* Transfer default LMV striping from the parent */
2626 if (lo->ldo_dir_striping_cached &&
2627 !LMVEA_DELETE_VALUES(lo->ldo_dir_def_stripenr,
2628 lo->ldo_dir_def_stripe_offset)) {
2629 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
2630 int def_stripe_count = lo->ldo_dir_def_stripenr;
2632 if (info->lti_ea_store_size < sizeof(*v1)) {
2633 rc = lod_ea_store_resize(info, sizeof(*v1));
2636 v1 = info->lti_ea_store;
2639 memset(v1, 0, sizeof(*v1));
2640 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
2641 v1->lum_stripe_count = cpu_to_le32(def_stripe_count);
2642 v1->lum_stripe_offset =
2643 cpu_to_le32(lo->ldo_dir_def_stripe_offset);
2645 cpu_to_le32(lo->ldo_dir_def_hash_type);
2647 info->lti_buf.lb_buf = v1;
2648 info->lti_buf.lb_len = sizeof(*v1);
2650 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
2651 XATTR_NAME_DEFAULT_LMV,
2654 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
2656 XATTR_NAME_DEFAULT_LMV, 0,
2662 /* Transfer default LOV striping from the parent */
2663 if (lo->ldo_striping_cached &&
2664 !LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size,
2665 lo->ldo_def_stripenr,
2666 lo->ldo_def_stripe_offset,
2668 struct lov_user_md_v3 *v3 = info->lti_ea_store;
2670 if (info->lti_ea_store_size < sizeof(*v3)) {
2671 rc = lod_ea_store_resize(info, sizeof(*v3));
2674 v3 = info->lti_ea_store;
2677 memset(v3, 0, sizeof(*v3));
2678 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
2679 v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
2680 v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
2681 v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
2682 if (lo->ldo_pool != NULL)
2683 strlcpy(v3->lmm_pool_name, lo->ldo_pool,
2684 sizeof(v3->lmm_pool_name));
2686 info->lti_buf.lb_buf = v3;
2687 info->lti_buf.lb_len = sizeof(*v3);
2690 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
2691 XATTR_NAME_LOV, 0, th);
2693 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
2694 XATTR_NAME_LOV, 0, th,
2703 static int lod_declare_dir_striping_create(const struct lu_env *env,
2704 struct dt_object *dt,
2705 struct lu_attr *attr,
2706 struct dt_object_format *dof,
2709 return lod_dir_striping_create_internal(env, dt, attr, dof, th, true);
2712 static int lod_dir_striping_create(const struct lu_env *env,
2713 struct dt_object *dt,
2714 struct lu_attr *attr,
2715 struct dt_object_format *dof,
2718 return lod_dir_striping_create_internal(env, dt, attr, dof, th, false);
2722 * Implementation of dt_object_operations::do_xattr_set.
2724 * Sets specified extended attribute on the object. Three types of EAs are
2726 * LOV EA - stores striping for a regular file or default striping (when set
2728 * LMV EA - stores a marker for the striped directories
2729 * DMV EA - stores default directory striping
2731 * When striping is applied to a non-striped existing object (this is called
2732 * late striping), then LOD notices the caller wants to turn the object into a
2733 * striped one. The stripe objects are created and appropriate EA is set:
2734 * LOV EA storing all the stripes directly or LMV EA storing just a small header
2735 * with striping configuration.
2737 * \see dt_object_operations::do_xattr_set() in the API description for details.
2739 static int lod_xattr_set(const struct lu_env *env,
2740 struct dt_object *dt, const struct lu_buf *buf,
2741 const char *name, int fl, struct thandle *th,
2742 struct lustre_capa *capa)
2744 struct dt_object *next = dt_object_child(dt);
2748 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
2749 strcmp(name, XATTR_NAME_LMV) == 0) {
2750 struct lmv_mds_md_v1 *lmm = buf->lb_buf;
2752 if (lmm != NULL && le32_to_cpu(lmm->lmv_hash_type) &
2753 LMV_HASH_FLAG_MIGRATION)
2754 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
2756 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
2761 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
2762 strcmp(name, XATTR_NAME_LOV) == 0) {
2764 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl, th, capa);
2766 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
2767 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2769 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
2772 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
2773 !strcmp(name, XATTR_NAME_LOV)) {
2774 /* in case of lov EA swap, just set it
2775 * if not, it is a replay so check striping match what we
2776 * already have during req replay, declare_xattr_set()
2777 * defines striping, then create() does the work
2779 if (fl & LU_XATTR_REPLACE) {
2780 /* free stripes, then update disk */
2781 lod_object_free_striping(env, lod_dt_obj(dt));
2782 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
2784 rc = lod_striping_create(env, dt, NULL, NULL, th);
2789 /* then all other xattr */
2790 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th, capa);
2796 * Implementation of dt_object_operations::do_declare_xattr_del.
2798 * \see dt_object_operations::do_declare_xattr_del() in the API description
2801 static int lod_declare_xattr_del(const struct lu_env *env,
2802 struct dt_object *dt, const char *name,
2805 return dt_declare_xattr_del(env, dt_object_child(dt), name, th);
2809 * Implementation of dt_object_operations::do_xattr_del.
2811 * If EA storing a regular striping is being deleted, then release
2812 * all the references to the stripe objects in core.
2814 * \see dt_object_operations::do_xattr_del() in the API description for details.
2816 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
2817 const char *name, struct thandle *th,
2818 struct lustre_capa *capa)
2820 if (!strcmp(name, XATTR_NAME_LOV))
2821 lod_object_free_striping(env, lod_dt_obj(dt));
2822 return dt_xattr_del(env, dt_object_child(dt), name, th, capa);
2826 * Implementation of dt_object_operations::do_xattr_list.
2828 * \see dt_object_operations::do_xattr_list() in the API description
2831 static int lod_xattr_list(const struct lu_env *env,
2832 struct dt_object *dt, const struct lu_buf *buf,
2833 struct lustre_capa *capa)
2835 return dt_xattr_list(env, dt_object_child(dt), buf, capa);
2839 * Initialize a pool the object belongs to.
2841 * When a striped object is being created, striping configuration
2842 * may demand the stripes are allocated on a limited set of the
2843 * targets. These limited sets are known as "pools". So we copy
2844 * a pool name into the object and later actual creation methods
2845 * (like lod_object_create()) will use this information to allocate
2846 * the stripes properly.
2848 * \param[in] o object
2849 * \param[in] pool pool name
2851 int lod_object_set_pool(struct lod_object *o, char *pool)
2856 len = strlen(o->ldo_pool);
2857 OBD_FREE(o->ldo_pool, len + 1);
2862 OBD_ALLOC(o->ldo_pool, len + 1);
2863 if (o->ldo_pool == NULL)
2865 strcpy(o->ldo_pool, pool);
2870 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
2872 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
2877 * Cache default regular striping in the object.
2879 * To improve performance of striped regular object creation we cache
2880 * default LOV striping (if it exists) in the parent directory object.
2882 * \param[in] env execution environment
2883 * \param[in] lp object
2885 * \retval 0 on success
2886 * \retval negative if failed
2888 static int lod_cache_parent_lov_striping(const struct lu_env *env,
2889 struct lod_object *lp)
2891 struct lod_thread_info *info = lod_env_info(env);
2892 struct lov_user_md_v1 *v1 = NULL;
2893 struct lov_user_md_v3 *v3 = NULL;
2897 /* called from MDD without parent being write locked,
2899 dt_write_lock(env, dt_object_child(&lp->ldo_obj), 0);
2900 rc = lod_get_lov_ea(env, lp);
2904 if (rc < (typeof(rc))sizeof(struct lov_user_md)) {
2905 /* don't lookup for non-existing or invalid striping */
2906 lp->ldo_def_striping_set = 0;
2907 lp->ldo_striping_cached = 1;
2908 lp->ldo_def_stripe_size = 0;
2909 lp->ldo_def_stripenr = 0;
2910 lp->ldo_def_stripe_offset = (typeof(v1->lmm_stripe_offset))(-1);
2911 GOTO(unlock, rc = 0);
2915 v1 = info->lti_ea_store;
2916 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
2917 lustre_swab_lov_user_md_v1(v1);
2918 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
2919 v3 = (struct lov_user_md_v3 *)v1;
2920 lustre_swab_lov_user_md_v3(v3);
2923 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1)
2924 GOTO(unlock, rc = 0);
2926 if (v1->lmm_pattern != LOV_PATTERN_RAID0 && v1->lmm_pattern != 0)
2927 GOTO(unlock, rc = 0);
2929 CDEBUG(D_INFO, DFID" stripe_count=%d stripe_size=%d stripe_offset=%d\n",
2930 PFID(lu_object_fid(&lp->ldo_obj.do_lu)),
2931 (int)v1->lmm_stripe_count,
2932 (int)v1->lmm_stripe_size, (int)v1->lmm_stripe_offset);
2934 lp->ldo_def_stripenr = v1->lmm_stripe_count;
2935 lp->ldo_def_stripe_size = v1->lmm_stripe_size;
2936 lp->ldo_def_stripe_offset = v1->lmm_stripe_offset;
2937 lp->ldo_striping_cached = 1;
2938 lp->ldo_def_striping_set = 1;
2939 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2940 /* XXX: sanity check here */
2941 v3 = (struct lov_user_md_v3 *) v1;
2942 if (v3->lmm_pool_name[0])
2943 lod_object_set_pool(lp, v3->lmm_pool_name);
2947 dt_write_unlock(env, dt_object_child(&lp->ldo_obj));
2953 * Cache default directory striping in the object.
2955 * To improve performance of striped directory creation we cache default
2956 * directory striping (if it exists) in the parent directory object.
2958 * \param[in] env execution environment
2959 * \param[in] lp object
2961 * \retval 0 on success
2962 * \retval negative if failed
2964 static int lod_cache_parent_lmv_striping(const struct lu_env *env,
2965 struct lod_object *lp)
2967 struct lod_thread_info *info = lod_env_info(env);
2968 struct lmv_user_md_v1 *v1 = NULL;
2972 /* called from MDD without parent being write locked,
2974 dt_write_lock(env, dt_object_child(&lp->ldo_obj), 0);
2975 rc = lod_get_default_lmv_ea(env, lp);
2979 if (rc < (typeof(rc))sizeof(struct lmv_user_md)) {
2980 /* don't lookup for non-existing or invalid striping */
2981 lp->ldo_dir_def_striping_set = 0;
2982 lp->ldo_dir_striping_cached = 1;
2983 lp->ldo_dir_def_stripenr = 0;
2984 lp->ldo_dir_def_stripe_offset =
2985 (typeof(v1->lum_stripe_offset))(-1);
2986 lp->ldo_dir_def_hash_type = LMV_HASH_TYPE_FNV_1A_64;
2987 GOTO(unlock, rc = 0);
2991 v1 = info->lti_ea_store;
2993 lp->ldo_dir_def_stripenr = le32_to_cpu(v1->lum_stripe_count);
2994 lp->ldo_dir_def_stripe_offset = le32_to_cpu(v1->lum_stripe_offset);
2995 lp->ldo_dir_def_hash_type = le32_to_cpu(v1->lum_hash_type);
2996 lp->ldo_dir_def_striping_set = 1;
2997 lp->ldo_dir_striping_cached = 1;
3001 dt_write_unlock(env, dt_object_child(&lp->ldo_obj));
3006 * Cache default striping in the object.
3008 * To improve performance of striped object creation we cache default striping
3009 * (if it exists) in the parent directory object. We always cache default
3010 * striping for the regular files (stored in LOV EA) and we cache default
3011 * striping for the directories if requested by \a child_mode (when a new
3012 * directory is being created).
3014 * \param[in] env execution environment
3015 * \param[in] lp object
3016 * \param[in] child_mode new object's mode
3018 * \retval 0 on success
3019 * \retval negative if failed
3021 static int lod_cache_parent_striping(const struct lu_env *env,
3022 struct lod_object *lp,
3028 rc = lod_load_striping(env, lp);
3032 if (!lp->ldo_striping_cached) {
3033 /* we haven't tried to get default striping for
3034 * the directory yet, let's cache it in the object */
3035 rc = lod_cache_parent_lov_striping(env, lp);
3040 if (S_ISDIR(child_mode) && !lp->ldo_dir_striping_cached)
3041 rc = lod_cache_parent_lmv_striping(env, lp);
3047 * Implementation of dt_object_operations::do_ah_init.
3049 * This method is used to make a decision on the striping configuration for the
3050 * object being created. It can be taken from the \a parent object if it exists,
3051 * or filesystem's default. The resulting configuration (number of stripes,
3052 * stripe size/offset, pool name, etc) is stored in the object itself and will
3053 * be used by the methods like ->doo_declare_create().
3055 * \see dt_object_operations::do_ah_init() in the API description for details.
3057 static void lod_ah_init(const struct lu_env *env,
3058 struct dt_allocation_hint *ah,
3059 struct dt_object *parent,
3060 struct dt_object *child,
3063 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
3064 struct dt_object *nextp = NULL;
3065 struct dt_object *nextc;
3066 struct lod_object *lp = NULL;
3067 struct lod_object *lc;
3068 struct lov_desc *desc;
3074 if (likely(parent)) {
3075 nextp = dt_object_child(parent);
3076 lp = lod_dt_obj(parent);
3077 rc = lod_load_striping(env, lp);
3082 nextc = dt_object_child(child);
3083 lc = lod_dt_obj(child);
3085 LASSERT(lc->ldo_stripenr == 0);
3086 LASSERT(lc->ldo_stripe == NULL);
3089 * local object may want some hints
3090 * in case of late striping creation, ->ah_init()
3091 * can be called with local object existing
3093 if (!dt_object_exists(nextc) || dt_object_remote(nextc)) {
3094 struct dt_object *obj;
3096 obj = (nextp != NULL && dt_object_remote(nextp)) ? NULL : nextp;
3097 nextc->do_ops->do_ah_init(env, ah, obj, nextc, child_mode);
3100 if (S_ISDIR(child_mode)) {
3101 if (lc->ldo_dir_stripe == NULL) {
3102 OBD_ALLOC_PTR(lc->ldo_dir_stripe);
3103 if (lc->ldo_dir_stripe == NULL)
3107 if (lp->ldo_dir_stripe == NULL) {
3108 OBD_ALLOC_PTR(lp->ldo_dir_stripe);
3109 if (lp->ldo_dir_stripe == NULL)
3113 rc = lod_cache_parent_striping(env, lp, child_mode);
3117 /* transfer defaults to new directory */
3118 if (lp->ldo_striping_cached) {
3120 lod_object_set_pool(lc, lp->ldo_pool);
3121 lc->ldo_def_stripenr = lp->ldo_def_stripenr;
3122 lc->ldo_def_stripe_size = lp->ldo_def_stripe_size;
3123 lc->ldo_def_stripe_offset = lp->ldo_def_stripe_offset;
3124 lc->ldo_striping_cached = 1;
3125 lc->ldo_def_striping_set = 1;
3126 CDEBUG(D_OTHER, "inherite EA sz:%d off:%d nr:%d\n",
3127 (int)lc->ldo_def_stripe_size,
3128 (int)lc->ldo_def_stripe_offset,
3129 (int)lc->ldo_def_stripenr);
3132 /* transfer dir defaults to new directory */
3133 if (lp->ldo_dir_striping_cached) {
3134 lc->ldo_dir_def_stripenr = lp->ldo_dir_def_stripenr;
3135 lc->ldo_dir_def_stripe_offset =
3136 lp->ldo_dir_def_stripe_offset;
3137 lc->ldo_dir_def_hash_type =
3138 lp->ldo_dir_def_hash_type;
3139 lc->ldo_dir_striping_cached = 1;
3140 lc->ldo_dir_def_striping_set = 1;
3141 CDEBUG(D_INFO, "inherit default EA nr:%d off:%d t%u\n",
3142 (int)lc->ldo_dir_def_stripenr,
3143 (int)lc->ldo_dir_def_stripe_offset,
3144 lc->ldo_dir_def_hash_type);
3147 /* It should always honour the specified stripes */
3148 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0) {
3149 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
3151 rc = lod_verify_md_striping(d, lum1);
3153 le32_to_cpu(lum1->lum_stripe_count) > 1) {
3154 /* Directory will be striped only if
3155 * stripe_count > 1 */
3157 le32_to_cpu(lum1->lum_stripe_count);
3158 lc->ldo_dir_stripe_offset =
3159 le32_to_cpu(lum1->lum_stripe_offset);
3160 lc->ldo_dir_hash_type =
3161 le32_to_cpu(lum1->lum_hash_type);
3162 CDEBUG(D_INFO, "set stripe EA nr:%hu off:%d\n",
3164 (int)lc->ldo_dir_stripe_offset);
3166 /* then check whether there is default stripes from parent */
3167 } else if (lp->ldo_dir_def_striping_set) {
3168 /* If there are default dir stripe from parent */
3169 lc->ldo_stripenr = lp->ldo_dir_def_stripenr;
3170 lc->ldo_dir_stripe_offset =
3171 lp->ldo_dir_def_stripe_offset;
3172 lc->ldo_dir_hash_type =
3173 lp->ldo_dir_def_hash_type;
3174 CDEBUG(D_INFO, "inherit EA nr:%hu off:%d\n",
3176 (int)lc->ldo_dir_stripe_offset);
3178 /* set default stripe for this directory */
3179 lc->ldo_stripenr = 0;
3180 lc->ldo_dir_stripe_offset = -1;
3183 CDEBUG(D_INFO, "final striping count:%hu, offset:%d\n",
3184 lc->ldo_stripenr, (int)lc->ldo_dir_stripe_offset);
3190 * if object is going to be striped over OSTs, transfer default
3191 * striping information to the child, so that we can use it
3192 * during declaration and creation
3194 if (!lod_object_will_be_striped(S_ISREG(child_mode),
3195 lu_object_fid(&child->do_lu)))
3198 * try from the parent
3200 if (likely(parent)) {
3201 lod_cache_parent_striping(env, lp, child_mode);
3203 lc->ldo_def_stripe_offset = LOV_OFFSET_DEFAULT;
3205 if (lp->ldo_def_striping_set) {
3207 lod_object_set_pool(lc, lp->ldo_pool);
3208 lc->ldo_stripenr = lp->ldo_def_stripenr;
3209 lc->ldo_stripe_size = lp->ldo_def_stripe_size;
3210 lc->ldo_def_stripe_offset = lp->ldo_def_stripe_offset;
3211 CDEBUG(D_OTHER, "striping from parent: #%d, sz %d %s\n",
3212 lc->ldo_stripenr, lc->ldo_stripe_size,
3213 lp->ldo_pool ? lp->ldo_pool : "");
3218 * if the parent doesn't provide with specific pattern, grab fs-wide one
3220 desc = &d->lod_desc;
3221 if (lc->ldo_stripenr == 0)
3222 lc->ldo_stripenr = desc->ld_default_stripe_count;
3223 if (lc->ldo_stripe_size == 0)
3224 lc->ldo_stripe_size = desc->ld_default_stripe_size;
3225 CDEBUG(D_OTHER, "final striping: # %d stripes, sz %d from %s\n",
3226 lc->ldo_stripenr, lc->ldo_stripe_size,
3227 lc->ldo_pool ? lc->ldo_pool : "");
3230 /* we do not cache stripe information for slave stripe, see
3231 * lod_xattr_set_lov_on_dir */
3232 if (lp != NULL && lp->ldo_dir_slave_stripe)
3233 lod_lov_stripe_cache_clear(lp);
3238 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
3240 * Size initialization on late striping.
3242 * Propagate the size of a truncated object to a deferred striping.
3243 * This function handles a special case when truncate was done on a
3244 * non-striped object and now while the striping is being created
3245 * we can't lose that size, so we have to propagate it to the stripes
3248 * \param[in] env execution environment
3249 * \param[in] dt object
3250 * \param[in] th transaction handle
3252 * \retval 0 on success
3253 * \retval negative if failed
3255 static int lod_declare_init_size(const struct lu_env *env,
3256 struct dt_object *dt, struct thandle *th)
3258 struct dt_object *next = dt_object_child(dt);
3259 struct lod_object *lo = lod_dt_obj(dt);
3260 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3261 uint64_t size, offs;
3265 /* XXX: we support the simplest (RAID0) striping so far */
3266 LASSERT(lo->ldo_stripe || lo->ldo_stripenr == 0);
3267 LASSERT(lo->ldo_stripe_size > 0);
3269 rc = dt_attr_get(env, next, attr, BYPASS_CAPA);
3270 LASSERT(attr->la_valid & LA_SIZE);
3274 size = attr->la_size;
3278 /* ll_do_div64(a, b) returns a % b, and a = a / b */
3279 ll_do_div64(size, (__u64) lo->ldo_stripe_size);
3280 stripe = ll_do_div64(size, (__u64) lo->ldo_stripenr);
3282 size = size * lo->ldo_stripe_size;
3283 offs = attr->la_size;
3284 size += ll_do_div64(offs, lo->ldo_stripe_size);
3286 attr->la_valid = LA_SIZE;
3287 attr->la_size = size;
3289 rc = dt_declare_attr_set(env, lo->ldo_stripe[stripe], attr, th);
3295 * Declare creation of striped object.
3297 * The function declares creation stripes for a regular object. The function
3298 * also declares whether the stripes will be created with non-zero size if
3299 * previously size was set non-zero on the master object. If object \a dt is
3300 * not local, then only fully defined striping can be applied in \a lovea.
3301 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
3304 * \param[in] env execution environment
3305 * \param[in] dt object
3306 * \param[in] attr attributes the stripes will be created with
3307 * \param[in] lovea a buffer containing striping description
3308 * \param[in] th transaction handle
3310 * \retval 0 on success
3311 * \retval negative if failed
3313 int lod_declare_striped_object(const struct lu_env *env, struct dt_object *dt,
3314 struct lu_attr *attr,
3315 const struct lu_buf *lovea, struct thandle *th)
3317 struct lod_thread_info *info = lod_env_info(env);
3318 struct dt_object *next = dt_object_child(dt);
3319 struct lod_object *lo = lod_dt_obj(dt);
3323 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO)) {
3324 /* failed to create striping, let's reset
3325 * config so that others don't get confused */
3326 lod_object_free_striping(env, lo);
3327 GOTO(out, rc = -ENOMEM);
3330 if (!dt_object_remote(next)) {
3331 /* choose OST and generate appropriate objects */
3332 rc = lod_qos_prep_create(env, lo, attr, lovea, th);
3334 /* failed to create striping, let's reset
3335 * config so that others don't get confused */
3336 lod_object_free_striping(env, lo);
3341 * declare storage for striping data
3343 info->lti_buf.lb_len = lov_mds_md_size(lo->ldo_stripenr,
3344 lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1);
3346 /* LOD can not choose OST objects for remote objects, i.e.
3347 * stripes must be ready before that. Right now, it can only
3348 * happen during migrate, i.e. migrate process needs to create
3349 * remote regular file (mdd_migrate_create), then the migrate
3350 * process will provide stripeEA. */
3351 LASSERT(lovea != NULL);
3352 info->lti_buf = *lovea;
3355 rc = dt_declare_xattr_set(env, next, &info->lti_buf,
3356 XATTR_NAME_LOV, 0, th);
3361 * if striping is created with local object's size > 0,
3362 * we have to propagate this size to specific object
3363 * the case is possible only when local object was created previously
3365 if (dt_object_exists(next))
3366 rc = lod_declare_init_size(env, dt, th);
3373 * Implementation of dt_object_operations::do_declare_create.
3375 * The method declares creation of a new object. If the object will be striped,
3376 * then helper functions are called to find FIDs for the stripes, declare
3377 * creation of the stripes and declare initialization of the striping
3378 * information to be stored in the master object.
3380 * \see dt_object_operations::do_declare_create() in the API description
3383 static int lod_declare_object_create(const struct lu_env *env,
3384 struct dt_object *dt,
3385 struct lu_attr *attr,
3386 struct dt_allocation_hint *hint,
3387 struct dt_object_format *dof,
3390 struct dt_object *next = dt_object_child(dt);
3391 struct lod_object *lo = lod_dt_obj(dt);
3400 * first of all, we declare creation of local object
3402 rc = dt_declare_create(env, next, attr, hint, dof, th);
3406 if (dof->dof_type == DFT_SYM)
3407 dt->do_body_ops = &lod_body_lnk_ops;
3410 * it's lod_ah_init() who has decided the object will striped
3412 if (dof->dof_type == DFT_REGULAR) {
3413 /* callers don't want stripes */
3414 /* XXX: all tricky interactions with ->ah_make_hint() decided
3415 * to use striping, then ->declare_create() behaving differently
3416 * should be cleaned */
3417 if (dof->u.dof_reg.striped == 0)
3418 lo->ldo_stripenr = 0;
3419 if (lo->ldo_stripenr > 0)
3420 rc = lod_declare_striped_object(env, dt, attr,
3422 } else if (dof->dof_type == DFT_DIR) {
3423 /* Orphan object (like migrating object) does not have
3424 * lod_dir_stripe, see lod_ah_init */
3425 if (lo->ldo_dir_stripe != NULL)
3426 rc = lod_declare_dir_striping_create(env, dt, attr,
3434 * Creation of a striped regular object.
3436 * The function is called to create the stripe objects for a regular
3437 * striped file. This can happen at the initial object creation or
3438 * when the caller asks LOD to do so using ->do_xattr_set() method
3439 * (so called late striping). Notice all the information are already
3440 * prepared in the form of the list of objects (ldo_stripe field).
3441 * This is done during declare phase.
3443 * \param[in] env execution environment
3444 * \param[in] dt object
3445 * \param[in] attr attributes the stripes will be created with
3446 * \param[in] dof format of stripes (see OSD API description)
3447 * \param[in] th transaction handle
3449 * \retval 0 on success
3450 * \retval negative if failed
3452 int lod_striping_create(const struct lu_env *env, struct dt_object *dt,
3453 struct lu_attr *attr, struct dt_object_format *dof,
3456 struct lod_object *lo = lod_dt_obj(dt);
3460 LASSERT(lo->ldo_striping_cached == 0);
3462 /* create all underlying objects */
3463 for (i = 0; i < lo->ldo_stripenr; i++) {
3464 LASSERT(lo->ldo_stripe[i]);
3465 rc = dt_create(env, lo->ldo_stripe[i], attr, NULL, dof, th);
3471 rc = lod_generate_and_set_lovea(env, lo, th);
3477 * Implementation of dt_object_operations::do_create.
3479 * If any of preceeding methods (like ->do_declare_create(),
3480 * ->do_ah_init(), etc) chose to create a striped object,
3481 * then this method will create the master and the stripes.
3483 * \see dt_object_operations::do_create() in the API description for details.
3485 static int lod_object_create(const struct lu_env *env, struct dt_object *dt,
3486 struct lu_attr *attr,
3487 struct dt_allocation_hint *hint,
3488 struct dt_object_format *dof, struct thandle *th)
3490 struct dt_object *next = dt_object_child(dt);
3491 struct lod_object *lo = lod_dt_obj(dt);
3495 /* create local object */
3496 rc = dt_create(env, next, attr, hint, dof, th);
3500 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
3501 lo->ldo_stripe && dof->u.dof_reg.striped != 0)
3502 rc = lod_striping_create(env, dt, attr, dof, th);
3508 * Implementation of dt_object_operations::do_declare_destroy.
3510 * If the object is a striped directory, then the function declares reference
3511 * removal from the master object (this is an index) to the stripes and declares
3512 * destroy of all the stripes. In all the cases, it declares an intention to
3513 * destroy the object itself.
3515 * \see dt_object_operations::do_declare_destroy() in the API description
3518 static int lod_declare_object_destroy(const struct lu_env *env,
3519 struct dt_object *dt,
3522 struct dt_object *next = dt_object_child(dt);
3523 struct lod_object *lo = lod_dt_obj(dt);
3524 struct lod_thread_info *info = lod_env_info(env);
3525 char *stripe_name = info->lti_key;
3530 * load striping information, notice we don't do this when object
3531 * is being initialized as we don't need this information till
3532 * few specific cases like destroy, chown
3534 rc = lod_load_striping(env, lo);
3538 /* declare destroy for all underlying objects */
3539 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
3540 rc = next->do_ops->do_index_try(env, next,
3541 &dt_directory_features);
3545 for (i = 0; i < lo->ldo_stripenr; i++) {
3546 rc = dt_declare_ref_del(env, next, th);
3549 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3550 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
3552 rc = dt_declare_delete(env, next,
3553 (const struct dt_key *)stripe_name, th);
3559 * we declare destroy for the local object
3561 rc = dt_declare_destroy(env, next, th);
3565 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ))
3568 /* declare destroy all striped objects */
3569 for (i = 0; i < lo->ldo_stripenr; i++) {
3570 if (likely(lo->ldo_stripe[i] != NULL)) {
3571 rc = dt_declare_destroy(env, lo->ldo_stripe[i], th);
3581 * Implementation of dt_object_operations::do_destroy.
3583 * If the object is a striped directory, then the function removes references
3584 * from the master object (this is an index) to the stripes and destroys all
3585 * the stripes. In all the cases, the function destroys the object itself.
3587 * \see dt_object_operations::do_destroy() in the API description for details.
3589 static int lod_object_destroy(const struct lu_env *env,
3590 struct dt_object *dt, struct thandle *th)
3592 struct dt_object *next = dt_object_child(dt);
3593 struct lod_object *lo = lod_dt_obj(dt);
3594 struct lod_thread_info *info = lod_env_info(env);
3595 char *stripe_name = info->lti_key;
3600 /* destroy sub-stripe of master object */
3601 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
3602 rc = next->do_ops->do_index_try(env, next,
3603 &dt_directory_features);
3607 for (i = 0; i < lo->ldo_stripenr; i++) {
3608 rc = dt_ref_del(env, next, th);
3612 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3613 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
3616 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
3617 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
3618 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)));
3620 rc = dt_delete(env, next,
3621 (const struct dt_key *)stripe_name,
3627 rc = dt_destroy(env, next, th);
3631 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ))
3634 /* destroy all striped objects */
3635 for (i = 0; i < lo->ldo_stripenr; i++) {
3636 if (likely(lo->ldo_stripe[i] != NULL) &&
3637 (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
3638 i == cfs_fail_val)) {
3639 rc = dt_destroy(env, lo->ldo_stripe[i], th);
3649 * Implementation of dt_object_operations::do_declare_ref_add.
3651 * \see dt_object_operations::do_declare_ref_add() in the API description
3654 static int lod_declare_ref_add(const struct lu_env *env,
3655 struct dt_object *dt, struct thandle *th)
3657 return dt_declare_ref_add(env, dt_object_child(dt), th);
3661 * Implementation of dt_object_operations::do_ref_add.
3663 * \see dt_object_operations::do_ref_add() in the API description for details.
3665 static int lod_ref_add(const struct lu_env *env,
3666 struct dt_object *dt, struct thandle *th)
3668 return dt_ref_add(env, dt_object_child(dt), th);
3672 * Implementation of dt_object_operations::do_declare_ref_del.
3674 * \see dt_object_operations::do_declare_ref_del() in the API description
3677 static int lod_declare_ref_del(const struct lu_env *env,
3678 struct dt_object *dt, struct thandle *th)
3680 return dt_declare_ref_del(env, dt_object_child(dt), th);
3684 * Implementation of dt_object_operations::do_ref_del
3686 * \see dt_object_operations::do_ref_del() in the API description for details.
3688 static int lod_ref_del(const struct lu_env *env,
3689 struct dt_object *dt, struct thandle *th)
3691 return dt_ref_del(env, dt_object_child(dt), th);
3695 * Implementation of dt_object_operations::do_capa_get.
3697 * \see dt_object_operations::do_capa_get() in the API description for details.
3699 static struct obd_capa *lod_capa_get(const struct lu_env *env,
3700 struct dt_object *dt,
3701 struct lustre_capa *old, __u64 opc)
3703 return dt_capa_get(env, dt_object_child(dt), old, opc);
3707 * Implementation of dt_object_operations::do_object_sync.
3709 * \see dt_object_operations::do_object_sync() in the API description
3712 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
3713 __u64 start, __u64 end)
3715 return dt_object_sync(env, dt_object_child(dt), start, end);
3718 struct lod_slave_locks {
3720 struct lustre_handle lsl_handle[0];
3724 * Release LDLM locks on the stripes of a striped directory.
3726 * Iterates over all the locks taken on the stripe objects and
3727 * release them using ->do_object_unlock() method.
3729 * \param[in] env execution environment
3730 * \param[in] dt striped object
3731 * \param[in] einfo lock description
3732 * \param[in] policy data describing requested lock
3734 * \retval 0 on success
3735 * \retval negative if failed
3737 static int lod_object_unlock_internal(const struct lu_env *env,
3738 struct dt_object *dt,
3739 struct ldlm_enqueue_info *einfo,
3740 ldlm_policy_data_t *policy)
3742 struct lod_object *lo = lod_dt_obj(dt);
3743 struct lod_slave_locks *slave_locks = einfo->ei_cbdata;
3748 if (slave_locks == NULL)
3751 for (i = 1; i < slave_locks->lsl_lock_count; i++) {
3752 if (lustre_handle_is_used(&slave_locks->lsl_handle[i])) {
3755 einfo->ei_cbdata = &slave_locks->lsl_handle[i];
3756 rc1 = dt_object_unlock(env, lo->ldo_stripe[i], einfo,
3759 rc = rc == 0 ? rc1 : rc;
3767 * Implementation of dt_object_operations::do_object_unlock.
3769 * Used to release LDLM lock(s).
3771 * \see dt_object_operations::do_object_unlock() in the API description
3774 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
3775 struct ldlm_enqueue_info *einfo,
3776 union ldlm_policy_data *policy)
3778 struct lod_object *lo = lod_dt_obj(dt);
3779 struct lod_slave_locks *slave_locks = einfo->ei_cbdata;
3780 int slave_locks_size;
3784 if (slave_locks == NULL)
3787 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3790 rc = lod_load_striping(env, lo);
3794 /* Note: for remote lock for single stripe dir, MDT will cancel
3795 * the lock by lockh directly */
3796 if (lo->ldo_stripenr <= 1 && dt_object_remote(dt_object_child(dt)))
3799 /* Only cancel slave lock for striped dir */
3800 rc = lod_object_unlock_internal(env, dt, einfo, policy);
3802 slave_locks_size = sizeof(*slave_locks) + slave_locks->lsl_lock_count *
3803 sizeof(slave_locks->lsl_handle[0]);
3804 OBD_FREE(slave_locks, slave_locks_size);
3805 einfo->ei_cbdata = NULL;
3811 * Implementation of dt_object_operations::do_object_lock.
3813 * Used to get LDLM lock on the non-striped and striped objects.
3815 * \see dt_object_operations::do_object_lock() in the API description
3818 static int lod_object_lock(const struct lu_env *env,
3819 struct dt_object *dt,
3820 struct lustre_handle *lh,
3821 struct ldlm_enqueue_info *einfo,
3822 union ldlm_policy_data *policy)
3824 struct lod_object *lo = lod_dt_obj(dt);
3827 int slave_locks_size;
3828 struct lod_slave_locks *slave_locks = NULL;
3831 /* remote object lock */
3832 if (!einfo->ei_enq_slave) {
3833 LASSERT(dt_object_remote(dt));
3834 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
3838 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3841 rc = lod_load_striping(env, lo);
3846 if (lo->ldo_stripenr <= 1)
3849 slave_locks_size = sizeof(*slave_locks) + lo->ldo_stripenr *
3850 sizeof(slave_locks->lsl_handle[0]);
3851 /* Freed in lod_object_unlock */
3852 OBD_ALLOC(slave_locks, slave_locks_size);
3853 if (slave_locks == NULL)
3855 slave_locks->lsl_lock_count = lo->ldo_stripenr;
3857 /* striped directory lock */
3858 for (i = 1; i < lo->ldo_stripenr; i++) {
3859 struct lustre_handle lockh;
3860 struct ldlm_res_id *res_id;
3862 res_id = &lod_env_info(env)->lti_res_id;
3863 fid_build_reg_res_name(lu_object_fid(&lo->ldo_stripe[i]->do_lu),
3865 einfo->ei_res_id = res_id;
3867 LASSERT(lo->ldo_stripe[i]);
3868 rc = dt_object_lock(env, lo->ldo_stripe[i], &lockh, einfo,
3872 slave_locks->lsl_handle[i] = lockh;
3875 einfo->ei_cbdata = slave_locks;
3878 if (rc != 0 && slave_locks != NULL) {
3879 einfo->ei_cbdata = slave_locks;
3880 lod_object_unlock_internal(env, dt, einfo, policy);
3881 OBD_FREE(slave_locks, slave_locks_size);
3882 einfo->ei_cbdata = NULL;
3888 struct dt_object_operations lod_obj_ops = {
3889 .do_read_lock = lod_object_read_lock,
3890 .do_write_lock = lod_object_write_lock,
3891 .do_read_unlock = lod_object_read_unlock,
3892 .do_write_unlock = lod_object_write_unlock,
3893 .do_write_locked = lod_object_write_locked,
3894 .do_attr_get = lod_attr_get,
3895 .do_declare_attr_set = lod_declare_attr_set,
3896 .do_attr_set = lod_attr_set,
3897 .do_xattr_get = lod_xattr_get,
3898 .do_declare_xattr_set = lod_declare_xattr_set,
3899 .do_xattr_set = lod_xattr_set,
3900 .do_declare_xattr_del = lod_declare_xattr_del,
3901 .do_xattr_del = lod_xattr_del,
3902 .do_xattr_list = lod_xattr_list,
3903 .do_ah_init = lod_ah_init,
3904 .do_declare_create = lod_declare_object_create,
3905 .do_create = lod_object_create,
3906 .do_declare_destroy = lod_declare_object_destroy,
3907 .do_destroy = lod_object_destroy,
3908 .do_index_try = lod_index_try,
3909 .do_declare_ref_add = lod_declare_ref_add,
3910 .do_ref_add = lod_ref_add,
3911 .do_declare_ref_del = lod_declare_ref_del,
3912 .do_ref_del = lod_ref_del,
3913 .do_capa_get = lod_capa_get,
3914 .do_object_sync = lod_object_sync,
3915 .do_object_lock = lod_object_lock,
3916 .do_object_unlock = lod_object_unlock,
3920 * Implementation of dt_body_operations::dbo_read.
3922 * \see dt_body_operations::dbo_read() in the API description for details.
3924 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
3925 struct lu_buf *buf, loff_t *pos,
3926 struct lustre_capa *capa)
3928 struct dt_object *next = dt_object_child(dt);
3929 return next->do_body_ops->dbo_read(env, next, buf, pos, capa);
3933 * Implementation of dt_body_operations::dbo_declare_write.
3935 * \see dt_body_operations::dbo_declare_write() in the API description
3938 static ssize_t lod_declare_write(const struct lu_env *env,
3939 struct dt_object *dt,
3940 const struct lu_buf *buf, loff_t pos,
3943 return dt_declare_record_write(env, dt_object_child(dt),
3948 * Implementation of dt_body_operations::dbo_write.
3950 * \see dt_body_operations::dbo_write() in the API description for details.
3952 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
3953 const struct lu_buf *buf, loff_t *pos,
3954 struct thandle *th, struct lustre_capa *capa, int iq)
3956 struct dt_object *next = dt_object_child(dt);
3958 return next->do_body_ops->dbo_write(env, next, buf, pos, th, capa, iq);
3961 static const struct dt_body_operations lod_body_lnk_ops = {
3962 .dbo_read = lod_read,
3963 .dbo_declare_write = lod_declare_write,
3964 .dbo_write = lod_write
3968 * Implementation of lu_object_operations::loo_object_init.
3970 * The function determines the type and the index of the target device using
3971 * sequence of the object's FID. Then passes control down to the
3972 * corresponding device:
3973 * OSD for the local objects, OSP for remote
3975 * \see lu_object_operations::loo_object_init() in the API description
3978 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
3979 const struct lu_object_conf *conf)
3981 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
3982 struct lu_device *cdev = NULL;
3983 struct lu_object *cobj;
3984 struct lod_tgt_descs *ltd = NULL;
3985 struct lod_tgt_desc *tgt;
3987 int type = LU_SEQ_RANGE_ANY;
3991 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
3993 /* Note: Sometimes, it will Return EAGAIN here, see
3994 * ptrlpc_import_delay_req(), which might confuse
3995 * lu_object_find_at() and make it wait there incorrectly.
3996 * so we convert it to EIO here.*/
4003 if (type == LU_SEQ_RANGE_MDT &&
4004 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
4005 cdev = &lod->lod_child->dd_lu_dev;
4006 } else if (type == LU_SEQ_RANGE_MDT) {
4007 ltd = &lod->lod_mdt_descs;
4009 } else if (type == LU_SEQ_RANGE_OST) {
4010 ltd = &lod->lod_ost_descs;
4017 if (ltd->ltd_tgts_size > idx &&
4018 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
4019 tgt = LTD_TGT(ltd, idx);
4021 LASSERT(tgt != NULL);
4022 LASSERT(tgt->ltd_tgt != NULL);
4024 cdev = &(tgt->ltd_tgt->dd_lu_dev);
4026 lod_putref(lod, ltd);
4029 if (unlikely(cdev == NULL))
4032 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
4033 if (unlikely(cobj == NULL))
4036 lu_object_add(lo, cobj);
4043 * Release resources associated with striping.
4045 * If the object is striped (regular or directory), then release
4046 * the stripe objects references and free the ldo_stripe array.
4048 * \param[in] env execution environment
4049 * \param[in] lo object
4051 void lod_object_free_striping(const struct lu_env *env, struct lod_object *lo)
4055 if (lo->ldo_dir_stripe != NULL) {
4056 OBD_FREE_PTR(lo->ldo_dir_stripe);
4057 lo->ldo_dir_stripe = NULL;
4060 if (lo->ldo_stripe) {
4061 LASSERT(lo->ldo_stripes_allocated > 0);
4063 for (i = 0; i < lo->ldo_stripenr; i++) {
4064 if (lo->ldo_stripe[i])
4065 lu_object_put(env, &lo->ldo_stripe[i]->do_lu);
4068 i = sizeof(struct dt_object *) * lo->ldo_stripes_allocated;
4069 OBD_FREE(lo->ldo_stripe, i);
4070 lo->ldo_stripe = NULL;
4071 lo->ldo_stripes_allocated = 0;
4073 lo->ldo_stripenr = 0;
4074 lo->ldo_pattern = 0;
4078 * Implementation of lu_object_operations::loo_object_start.
4080 * \see lu_object_operations::loo_object_start() in the API description
4083 static int lod_object_start(const struct lu_env *env, struct lu_object *o)
4085 if (S_ISLNK(o->lo_header->loh_attr & S_IFMT))
4086 lu2lod_obj(o)->ldo_obj.do_body_ops = &lod_body_lnk_ops;
4091 * Implementation of lu_object_operations::loo_object_free.
4093 * \see lu_object_operations::loo_object_free() in the API description
4096 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
4098 struct lod_object *mo = lu2lod_obj(o);
4101 * release all underlying object pinned
4104 lod_object_free_striping(env, mo);
4106 lod_object_set_pool(mo, NULL);
4109 OBD_SLAB_FREE_PTR(mo, lod_object_kmem);
4113 * Implementation of lu_object_operations::loo_object_release.
4115 * \see lu_object_operations::loo_object_release() in the API description
4118 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
4120 /* XXX: shouldn't we release everything here in case if object
4121 * creation failed before? */
4125 * Implementation of lu_object_operations::loo_object_print.
4127 * \see lu_object_operations::loo_object_print() in the API description
4130 static int lod_object_print(const struct lu_env *env, void *cookie,
4131 lu_printer_t p, const struct lu_object *l)
4133 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
4135 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
4138 struct lu_object_operations lod_lu_obj_ops = {
4139 .loo_object_init = lod_object_init,
4140 .loo_object_start = lod_object_start,
4141 .loo_object_free = lod_object_free,
4142 .loo_object_release = lod_object_release,
4143 .loo_object_print = lod_object_print,