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, 2014, 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 static const struct dt_body_operations lod_body_lnk_ops;
62 static const struct dt_body_operations lod_body_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)
74 struct dt_object *next = dt_object_child(dt);
75 return next->do_index_ops->dio_lookup(env, next, rec, key);
79 * Implementation of dt_index_operations::dio_declare_insert.
81 * Used with regular (non-striped) objects.
83 * \see dt_index_operations::dio_declare_insert() in the API description
86 static int lod_declare_index_insert(const struct lu_env *env,
88 const struct dt_rec *rec,
89 const struct dt_key *key,
92 return lod_sub_object_declare_insert(env, dt_object_child(dt),
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,
110 return lod_sub_object_index_insert(env, dt_object_child(dt), rec, key,
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 lod_sub_object_declare_delete(env, dt_object_child(dt), key,
132 * Implementation of dt_index_operations::dio_delete.
134 * Used with regular (non-striped) objects.
136 * \see dt_index_operations::dio_delete() in the API description for details.
138 static int lod_index_delete(const struct lu_env *env,
139 struct dt_object *dt,
140 const struct dt_key *key,
143 return lod_sub_object_delete(env, dt_object_child(dt), key, th);
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)
156 struct dt_object *next = dt_object_child(dt);
157 struct lod_it *it = &lod_env_info(env)->lti_it;
158 struct dt_it *it_next;
160 it_next = next->do_index_ops->dio_it.init(env, next, attr);
164 /* currently we do not use more than one iterator per thread
165 * so we store it in thread info. if at some point we need
166 * more active iterators in a single thread, we can allocate
168 LASSERT(it->lit_obj == NULL);
170 it->lit_it = it_next;
173 return (struct dt_it *)it;
176 #define LOD_CHECK_IT(env, it) \
178 LASSERT((it)->lit_obj != NULL); \
179 LASSERT((it)->lit_it != NULL); \
183 * Implementation of dt_index_operations::dio_it.fini.
185 * Used with regular (non-striped) objects.
187 * \see dt_index_operations::dio_it.fini() in the API description for details.
189 static void lod_it_fini(const struct lu_env *env, struct dt_it *di)
191 struct lod_it *it = (struct lod_it *)di;
193 LOD_CHECK_IT(env, it);
194 it->lit_obj->do_index_ops->dio_it.fini(env, it->lit_it);
196 /* the iterator not in use any more */
202 * Implementation of dt_it_ops::get.
204 * Used with regular (non-striped) objects.
206 * \see dt_it_ops::get() in the API description for details.
208 static int lod_it_get(const struct lu_env *env, struct dt_it *di,
209 const struct dt_key *key)
211 const struct lod_it *it = (const struct lod_it *)di;
213 LOD_CHECK_IT(env, it);
214 return it->lit_obj->do_index_ops->dio_it.get(env, it->lit_it, key);
218 * Implementation of dt_it_ops::put.
220 * Used with regular (non-striped) objects.
222 * \see dt_it_ops::put() in the API description for details.
224 static void lod_it_put(const struct lu_env *env, struct dt_it *di)
226 struct lod_it *it = (struct lod_it *)di;
228 LOD_CHECK_IT(env, it);
229 return it->lit_obj->do_index_ops->dio_it.put(env, it->lit_it);
233 * Implementation of dt_it_ops::next.
235 * Used with regular (non-striped) objects
237 * \see dt_it_ops::next() in the API description for details.
239 static int lod_it_next(const struct lu_env *env, struct dt_it *di)
241 struct lod_it *it = (struct lod_it *)di;
243 LOD_CHECK_IT(env, it);
244 return it->lit_obj->do_index_ops->dio_it.next(env, it->lit_it);
248 * Implementation of dt_it_ops::key.
250 * Used with regular (non-striped) objects.
252 * \see dt_it_ops::key() in the API description for details.
254 static struct dt_key *lod_it_key(const struct lu_env *env,
255 const struct dt_it *di)
257 const struct lod_it *it = (const struct lod_it *)di;
259 LOD_CHECK_IT(env, it);
260 return it->lit_obj->do_index_ops->dio_it.key(env, it->lit_it);
264 * Implementation of dt_it_ops::key_size.
266 * Used with regular (non-striped) objects.
268 * \see dt_it_ops::key_size() in the API description for details.
270 static int lod_it_key_size(const struct lu_env *env, const struct dt_it *di)
272 struct lod_it *it = (struct lod_it *)di;
274 LOD_CHECK_IT(env, it);
275 return it->lit_obj->do_index_ops->dio_it.key_size(env, it->lit_it);
279 * Implementation of dt_it_ops::rec.
281 * Used with regular (non-striped) objects.
283 * \see dt_it_ops::rec() in the API description for details.
285 static int lod_it_rec(const struct lu_env *env, const struct dt_it *di,
286 struct dt_rec *rec, __u32 attr)
288 const struct lod_it *it = (const struct lod_it *)di;
290 LOD_CHECK_IT(env, it);
291 return it->lit_obj->do_index_ops->dio_it.rec(env, it->lit_it, rec,
296 * Implementation of dt_it_ops::rec_size.
298 * Used with regular (non-striped) objects.
300 * \see dt_it_ops::rec_size() in the API description for details.
302 static int lod_it_rec_size(const struct lu_env *env, const struct dt_it *di,
305 const struct lod_it *it = (const struct lod_it *)di;
307 LOD_CHECK_IT(env, it);
308 return it->lit_obj->do_index_ops->dio_it.rec_size(env, it->lit_it,
313 * Implementation of dt_it_ops::store.
315 * Used with regular (non-striped) objects.
317 * \see dt_it_ops::store() in the API description for details.
319 static __u64 lod_it_store(const struct lu_env *env, const struct dt_it *di)
321 const struct lod_it *it = (const struct lod_it *)di;
323 LOD_CHECK_IT(env, it);
324 return it->lit_obj->do_index_ops->dio_it.store(env, it->lit_it);
328 * Implementation of dt_it_ops::load.
330 * Used with regular (non-striped) objects.
332 * \see dt_it_ops::load() in the API description for details.
334 static int lod_it_load(const struct lu_env *env, const struct dt_it *di,
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 static 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)
393 struct lod_object *lo = lod_dt_obj(dt);
394 struct dt_object *next;
395 struct lod_it *it = &lod_env_info(env)->lti_it;
396 struct dt_it *it_next;
399 LASSERT(lo->ldo_stripenr > 0);
400 next = lo->ldo_stripe[0];
401 LASSERT(next != NULL);
402 LASSERT(next->do_index_ops != NULL);
404 it_next = next->do_index_ops->dio_it.init(env, next, attr);
408 /* currently we do not use more than one iterator per thread
409 * so we store it in thread info. if at some point we need
410 * more active iterators in a single thread, we can allocate
412 LASSERT(it->lit_obj == NULL);
414 it->lit_stripe_index = 0;
416 it->lit_it = it_next;
419 return (struct dt_it *)it;
422 #define LOD_CHECK_STRIPED_IT(env, it, lo) \
424 LASSERT((it)->lit_obj != NULL); \
425 LASSERT((it)->lit_it != NULL); \
426 LASSERT((lo)->ldo_stripenr > 0); \
427 LASSERT((it)->lit_stripe_index < (lo)->ldo_stripenr); \
431 * Implementation of dt_it_ops::fini.
433 * Used with striped objects.
435 * \see dt_it_ops::fini() in the API description for details.
437 static void lod_striped_it_fini(const struct lu_env *env, struct dt_it *di)
439 struct lod_it *it = (struct lod_it *)di;
440 struct lod_object *lo = lod_dt_obj(it->lit_obj);
441 struct dt_object *next;
443 /* If lit_it == NULL, then it means the sub_it has been finished,
444 * which only happens in failure cases, see lod_striped_it_next() */
445 if (it->lit_it != NULL) {
446 LOD_CHECK_STRIPED_IT(env, it, lo);
448 next = lo->ldo_stripe[it->lit_stripe_index];
449 LASSERT(next != NULL);
450 LASSERT(next->do_index_ops != NULL);
452 next->do_index_ops->dio_it.fini(env, it->lit_it);
455 /* the iterator not in use any more */
458 it->lit_stripe_index = 0;
462 * Implementation of dt_it_ops::get.
464 * Right now it's not used widely, only to reset the iterator to the
465 * initial position. It should be possible to implement a full version
466 * which chooses a correct stripe to be able to position with any key.
468 * \see dt_it_ops::get() in the API description for details.
470 static int lod_striped_it_get(const struct lu_env *env, struct dt_it *di,
471 const struct dt_key *key)
473 const struct lod_it *it = (const struct lod_it *)di;
474 struct lod_object *lo = lod_dt_obj(it->lit_obj);
475 struct dt_object *next;
478 LOD_CHECK_STRIPED_IT(env, it, lo);
480 next = lo->ldo_stripe[it->lit_stripe_index];
481 LASSERT(next != NULL);
482 LASSERT(next->do_index_ops != NULL);
484 return next->do_index_ops->dio_it.get(env, it->lit_it, key);
488 * Implementation of dt_it_ops::put.
490 * Used with striped objects.
492 * \see dt_it_ops::put() in the API description for details.
494 static void lod_striped_it_put(const struct lu_env *env, struct dt_it *di)
496 struct lod_it *it = (struct lod_it *)di;
497 struct lod_object *lo = lod_dt_obj(it->lit_obj);
498 struct dt_object *next;
500 LOD_CHECK_STRIPED_IT(env, it, lo);
502 next = lo->ldo_stripe[it->lit_stripe_index];
503 LASSERT(next != NULL);
504 LASSERT(next->do_index_ops != NULL);
506 return next->do_index_ops->dio_it.put(env, it->lit_it);
510 * Implementation of dt_it_ops::next.
512 * Used with striped objects. When the end of the current stripe is
513 * reached, the method takes the next stripe's iterator.
515 * \see dt_it_ops::next() in the API description for details.
517 static int lod_striped_it_next(const struct lu_env *env, struct dt_it *di)
519 struct lod_it *it = (struct lod_it *)di;
520 struct lod_object *lo = lod_dt_obj(it->lit_obj);
521 struct dt_object *next;
522 struct dt_it *it_next;
526 LOD_CHECK_STRIPED_IT(env, it, lo);
528 next = lo->ldo_stripe[it->lit_stripe_index];
529 LASSERT(next != NULL);
530 LASSERT(next->do_index_ops != NULL);
532 rc = next->do_index_ops->dio_it.next(env, it->lit_it);
536 if (rc == 0 && it->lit_stripe_index == 0)
539 if (rc == 0 && it->lit_stripe_index > 0) {
540 struct lu_dirent *ent;
542 ent = (struct lu_dirent *)lod_env_info(env)->lti_key;
544 rc = next->do_index_ops->dio_it.rec(env, it->lit_it,
545 (struct dt_rec *)ent,
550 /* skip . and .. for slave stripe */
551 if ((strncmp(ent->lde_name, ".",
552 le16_to_cpu(ent->lde_namelen)) == 0 &&
553 le16_to_cpu(ent->lde_namelen) == 1) ||
554 (strncmp(ent->lde_name, "..",
555 le16_to_cpu(ent->lde_namelen)) == 0 &&
556 le16_to_cpu(ent->lde_namelen) == 2))
562 /* go to next stripe */
563 if (it->lit_stripe_index + 1 >= lo->ldo_stripenr)
566 it->lit_stripe_index++;
568 next->do_index_ops->dio_it.put(env, it->lit_it);
569 next->do_index_ops->dio_it.fini(env, it->lit_it);
572 next = lo->ldo_stripe[it->lit_stripe_index];
573 LASSERT(next != NULL);
574 LASSERT(next->do_index_ops != NULL);
576 rc = next->do_ops->do_index_try(env, next, &dt_directory_features);
580 it_next = next->do_index_ops->dio_it.init(env, next, it->lit_attr);
581 if (!IS_ERR(it_next)) {
582 it->lit_it = it_next;
585 rc = PTR_ERR(it_next);
592 * Implementation of dt_it_ops::key.
594 * Used with striped objects.
596 * \see dt_it_ops::key() in the API description for details.
598 static struct dt_key *lod_striped_it_key(const struct lu_env *env,
599 const struct dt_it *di)
601 const struct lod_it *it = (const struct lod_it *)di;
602 struct lod_object *lo = lod_dt_obj(it->lit_obj);
603 struct dt_object *next;
605 LOD_CHECK_STRIPED_IT(env, it, lo);
607 next = lo->ldo_stripe[it->lit_stripe_index];
608 LASSERT(next != NULL);
609 LASSERT(next->do_index_ops != NULL);
611 return next->do_index_ops->dio_it.key(env, it->lit_it);
615 * Implementation of dt_it_ops::key_size.
617 * Used with striped objects.
619 * \see dt_it_ops::size() in the API description for details.
621 static int lod_striped_it_key_size(const struct lu_env *env,
622 const struct dt_it *di)
624 struct lod_it *it = (struct lod_it *)di;
625 struct lod_object *lo = lod_dt_obj(it->lit_obj);
626 struct dt_object *next;
628 LOD_CHECK_STRIPED_IT(env, it, lo);
630 next = lo->ldo_stripe[it->lit_stripe_index];
631 LASSERT(next != NULL);
632 LASSERT(next->do_index_ops != NULL);
634 return next->do_index_ops->dio_it.key_size(env, it->lit_it);
638 * Implementation of dt_it_ops::rec.
640 * Used with striped objects.
642 * \see dt_it_ops::rec() in the API description for details.
644 static int lod_striped_it_rec(const struct lu_env *env, const struct dt_it *di,
645 struct dt_rec *rec, __u32 attr)
647 const struct lod_it *it = (const struct lod_it *)di;
648 struct lod_object *lo = lod_dt_obj(it->lit_obj);
649 struct dt_object *next;
651 LOD_CHECK_STRIPED_IT(env, it, lo);
653 next = lo->ldo_stripe[it->lit_stripe_index];
654 LASSERT(next != NULL);
655 LASSERT(next->do_index_ops != NULL);
657 return next->do_index_ops->dio_it.rec(env, it->lit_it, rec, attr);
661 * Implementation of dt_it_ops::rec_size.
663 * Used with striped objects.
665 * \see dt_it_ops::rec_size() in the API description for details.
667 static int lod_striped_it_rec_size(const struct lu_env *env,
668 const struct dt_it *di, __u32 attr)
670 struct lod_it *it = (struct lod_it *)di;
671 struct lod_object *lo = lod_dt_obj(it->lit_obj);
672 struct dt_object *next;
674 LOD_CHECK_STRIPED_IT(env, it, lo);
676 next = lo->ldo_stripe[it->lit_stripe_index];
677 LASSERT(next != NULL);
678 LASSERT(next->do_index_ops != NULL);
680 return next->do_index_ops->dio_it.rec_size(env, it->lit_it, attr);
684 * Implementation of dt_it_ops::store.
686 * Used with striped objects.
688 * \see dt_it_ops::store() in the API description for details.
690 static __u64 lod_striped_it_store(const struct lu_env *env,
691 const struct dt_it *di)
693 const struct lod_it *it = (const struct lod_it *)di;
694 struct lod_object *lo = lod_dt_obj(it->lit_obj);
695 struct dt_object *next;
697 LOD_CHECK_STRIPED_IT(env, it, lo);
699 next = lo->ldo_stripe[it->lit_stripe_index];
700 LASSERT(next != NULL);
701 LASSERT(next->do_index_ops != NULL);
703 return next->do_index_ops->dio_it.store(env, it->lit_it);
707 * Implementation of dt_it_ops::load.
709 * Used with striped objects.
711 * \see dt_it_ops::load() in the API description for details.
713 static int lod_striped_it_load(const struct lu_env *env,
714 const struct dt_it *di, __u64 hash)
716 const struct lod_it *it = (const struct lod_it *)di;
717 struct lod_object *lo = lod_dt_obj(it->lit_obj);
718 struct dt_object *next;
720 LOD_CHECK_STRIPED_IT(env, it, lo);
722 next = lo->ldo_stripe[it->lit_stripe_index];
723 LASSERT(next != NULL);
724 LASSERT(next->do_index_ops != NULL);
726 return next->do_index_ops->dio_it.load(env, it->lit_it, hash);
729 static struct dt_index_operations lod_striped_index_ops = {
730 .dio_lookup = lod_index_lookup,
731 .dio_declare_insert = lod_declare_index_insert,
732 .dio_insert = lod_index_insert,
733 .dio_declare_delete = lod_declare_index_delete,
734 .dio_delete = lod_index_delete,
736 .init = lod_striped_it_init,
737 .fini = lod_striped_it_fini,
738 .get = lod_striped_it_get,
739 .put = lod_striped_it_put,
740 .next = lod_striped_it_next,
741 .key = lod_striped_it_key,
742 .key_size = lod_striped_it_key_size,
743 .rec = lod_striped_it_rec,
744 .rec_size = lod_striped_it_rec_size,
745 .store = lod_striped_it_store,
746 .load = lod_striped_it_load,
751 * Append the FID for each shard of the striped directory after the
752 * given LMV EA header.
754 * To simplify striped directory and the consistency verification,
755 * we only store the LMV EA header on disk, for both master object
756 * and slave objects. When someone wants to know the whole LMV EA,
757 * such as client readdir(), we can build the entrie LMV EA on the
758 * MDT side (in RAM) via iterating the sub-directory entries that
759 * are contained in the master object of the stripe directory.
761 * For the master object of the striped directroy, the valid name
762 * for each shard is composed of the ${shard_FID}:${shard_idx}.
764 * There may be holes in the LMV EA if some shards' name entries
765 * are corrupted or lost.
767 * \param[in] env pointer to the thread context
768 * \param[in] lo pointer to the master object of the striped directory
769 * \param[in] buf pointer to the lu_buf which will hold the LMV EA
770 * \param[in] resize whether re-allocate the buffer if it is not big enough
772 * \retval positive size of the LMV EA
773 * \retval 0 for nothing to be loaded
774 * \retval negative error number on failure
776 int lod_load_lmv_shards(const struct lu_env *env, struct lod_object *lo,
777 struct lu_buf *buf, bool resize)
779 struct lu_dirent *ent =
780 (struct lu_dirent *)lod_env_info(env)->lti_key;
781 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
782 struct dt_object *obj = dt_object_child(&lo->ldo_obj);
783 struct lmv_mds_md_v1 *lmv1 = buf->lb_buf;
785 const struct dt_it_ops *iops;
787 __u32 magic = le32_to_cpu(lmv1->lmv_magic);
792 /* If it is not a striped directory, then load nothing. */
793 if (magic != LMV_MAGIC_V1)
796 /* If it is in migration (or failure), then load nothing. */
797 if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
800 stripes = le32_to_cpu(lmv1->lmv_stripe_count);
804 rc = lmv_mds_md_size(stripes, magic);
808 if (buf->lb_len < lmv1_size) {
817 lu_buf_alloc(buf, lmv1_size);
822 memcpy(buf->lb_buf, tbuf.lb_buf, tbuf.lb_len);
825 if (unlikely(!dt_try_as_dir(env, obj)))
828 memset(&lmv1->lmv_stripe_fids[0], 0, stripes * sizeof(struct lu_fid));
829 iops = &obj->do_index_ops->dio_it;
830 it = iops->init(env, obj, LUDA_64BITHASH);
834 rc = iops->load(env, it, 0);
836 rc = iops->next(env, it);
841 char name[FID_LEN + 2] = "";
846 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
852 fid_le_to_cpu(&fid, &ent->lde_fid);
853 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
854 if (ent->lde_name[0] == '.') {
855 if (ent->lde_namelen == 1)
858 if (ent->lde_namelen == 2 && ent->lde_name[1] == '.')
862 len = snprintf(name, sizeof(name),
863 DFID":", PFID(&ent->lde_fid));
864 /* The ent->lde_name is composed of ${FID}:${index} */
865 if (ent->lde_namelen < len + 1 ||
866 memcmp(ent->lde_name, name, len) != 0) {
867 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
868 "%s: invalid shard name %.*s with the FID "DFID
869 " for the striped directory "DFID", %s\n",
870 lod2obd(lod)->obd_name, ent->lde_namelen,
871 ent->lde_name, PFID(&fid),
872 PFID(lu_object_fid(&obj->do_lu)),
873 lod->lod_lmv_failout ? "failout" : "skip");
875 if (lod->lod_lmv_failout)
883 if (ent->lde_name[len] < '0' ||
884 ent->lde_name[len] > '9') {
885 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
886 "%s: invalid shard name %.*s with the "
887 "FID "DFID" for the striped directory "
889 lod2obd(lod)->obd_name, ent->lde_namelen,
890 ent->lde_name, PFID(&fid),
891 PFID(lu_object_fid(&obj->do_lu)),
892 lod->lod_lmv_failout ?
895 if (lod->lod_lmv_failout)
901 index = index * 10 + ent->lde_name[len++] - '0';
902 } while (len < ent->lde_namelen);
904 if (len == ent->lde_namelen) {
905 /* Out of LMV EA range. */
906 if (index >= stripes) {
907 CERROR("%s: the shard %.*s for the striped "
908 "directory "DFID" is out of the known "
909 "LMV EA range [0 - %u], failout\n",
910 lod2obd(lod)->obd_name, ent->lde_namelen,
912 PFID(lu_object_fid(&obj->do_lu)),
918 /* The slot has been occupied. */
919 if (!fid_is_zero(&lmv1->lmv_stripe_fids[index])) {
923 &lmv1->lmv_stripe_fids[index]);
924 CERROR("%s: both the shard "DFID" and "DFID
925 " for the striped directory "DFID
926 " claim the same LMV EA slot at the "
927 "index %d, failout\n",
928 lod2obd(lod)->obd_name,
929 PFID(&fid0), PFID(&fid),
930 PFID(lu_object_fid(&obj->do_lu)), index);
935 /* stored as LE mode */
936 lmv1->lmv_stripe_fids[index] = ent->lde_fid;
939 rc = iops->next(env, it);
946 RETURN(rc > 0 ? lmv_mds_md_size(stripes, magic) : rc);
950 * Implementation of dt_object_operations::do_index_try.
952 * \see dt_object_operations::do_index_try() in the API description for details.
954 static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
955 const struct dt_index_features *feat)
957 struct lod_object *lo = lod_dt_obj(dt);
958 struct dt_object *next = dt_object_child(dt);
962 LASSERT(next->do_ops);
963 LASSERT(next->do_ops->do_index_try);
965 rc = lod_load_striping_locked(env, lo);
969 rc = next->do_ops->do_index_try(env, next, feat);
973 if (lo->ldo_stripenr > 0) {
976 for (i = 0; i < lo->ldo_stripenr; i++) {
977 if (dt_object_exists(lo->ldo_stripe[i]) == 0)
979 rc = lo->ldo_stripe[i]->do_ops->do_index_try(env,
980 lo->ldo_stripe[i], feat);
984 dt->do_index_ops = &lod_striped_index_ops;
986 dt->do_index_ops = &lod_index_ops;
993 * Implementation of dt_object_operations::do_read_lock.
995 * \see dt_object_operations::do_read_lock() in the API description for details.
997 static void lod_object_read_lock(const struct lu_env *env,
998 struct dt_object *dt, unsigned role)
1000 dt_read_lock(env, dt_object_child(dt), role);
1004 * Implementation of dt_object_operations::do_write_lock.
1006 * \see dt_object_operations::do_write_lock() in the API description for
1009 static void lod_object_write_lock(const struct lu_env *env,
1010 struct dt_object *dt, unsigned role)
1012 dt_write_lock(env, dt_object_child(dt), role);
1016 * Implementation of dt_object_operations::do_read_unlock.
1018 * \see dt_object_operations::do_read_unlock() in the API description for
1021 static void lod_object_read_unlock(const struct lu_env *env,
1022 struct dt_object *dt)
1024 dt_read_unlock(env, dt_object_child(dt));
1028 * Implementation of dt_object_operations::do_write_unlock.
1030 * \see dt_object_operations::do_write_unlock() in the API description for
1033 static void lod_object_write_unlock(const struct lu_env *env,
1034 struct dt_object *dt)
1036 dt_write_unlock(env, dt_object_child(dt));
1040 * Implementation of dt_object_operations::do_write_locked.
1042 * \see dt_object_operations::do_write_locked() in the API description for
1045 static int lod_object_write_locked(const struct lu_env *env,
1046 struct dt_object *dt)
1048 return dt_write_locked(env, dt_object_child(dt));
1052 * Implementation of dt_object_operations::do_attr_get.
1054 * \see dt_object_operations::do_attr_get() in the API description for details.
1056 static int lod_attr_get(const struct lu_env *env,
1057 struct dt_object *dt,
1058 struct lu_attr *attr)
1060 /* Note: for striped directory, client will merge attributes
1061 * from all of the sub-stripes see lmv_merge_attr(), and there
1062 * no MDD logic depend on directory nlink/size/time, so we can
1063 * always use master inode nlink and size for now. */
1064 return dt_attr_get(env, dt_object_child(dt), attr);
1068 * Mark all of the striped directory sub-stripes dead.
1070 * When a striped object is a subject to removal, we have
1071 * to mark all the stripes to prevent further access to
1072 * them (e.g. create a new file in those). So we mark
1073 * all the stripes with LMV_HASH_FLAG_DEAD. The function
1074 * can be used to declare the changes and to apply them.
1075 * If the object isn't striped, then just return success.
1077 * \param[in] env execution environment
1078 * \param[in] dt the striped object
1079 * \param[in] handle transaction handle
1080 * \param[in] declare whether to declare the change or apply
1082 * \retval 0 on success
1083 * \retval negative if failed
1085 static int lod_mark_dead_object(const struct lu_env *env,
1086 struct dt_object *dt,
1090 struct lod_object *lo = lod_dt_obj(dt);
1091 struct lmv_mds_md_v1 *lmv;
1092 __u32 dead_hash_type;
1098 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
1101 rc = lod_load_striping_locked(env, lo);
1105 if (lo->ldo_stripenr == 0)
1108 rc = lod_get_lmv_ea(env, lo);
1112 lmv = lod_env_info(env)->lti_ea_store;
1113 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1114 dead_hash_type = le32_to_cpu(lmv->lmv_hash_type) | LMV_HASH_FLAG_DEAD;
1115 lmv->lmv_hash_type = cpu_to_le32(dead_hash_type);
1116 for (i = 0; i < lo->ldo_stripenr; i++) {
1119 lmv->lmv_master_mdt_index = i;
1121 buf.lb_len = sizeof(*lmv);
1123 rc = lod_sub_object_declare_xattr_set(env,
1124 lo->ldo_stripe[i], &buf,
1126 LU_XATTR_REPLACE, th);
1128 rc = lod_sub_object_xattr_set(env, lo->ldo_stripe[i],
1129 &buf, XATTR_NAME_LMV,
1130 LU_XATTR_REPLACE, th);
1140 * Implementation of dt_object_operations::do_declare_attr_set.
1142 * If the object is striped, then apply the changes to all the stripes.
1144 * \see dt_object_operations::do_declare_attr_set() in the API description
1147 static int lod_declare_attr_set(const struct lu_env *env,
1148 struct dt_object *dt,
1149 const struct lu_attr *attr,
1152 struct dt_object *next = dt_object_child(dt);
1153 struct lod_object *lo = lod_dt_obj(dt);
1157 /* Set dead object on all other stripes */
1158 if (attr->la_valid & LA_FLAGS && !(attr->la_valid & ~LA_FLAGS) &&
1159 attr->la_flags & LUSTRE_SLAVE_DEAD_FL) {
1160 rc = lod_mark_dead_object(env, dt, th, true);
1165 * declare setattr on the local object
1167 rc = lod_sub_object_declare_attr_set(env, next, attr, th);
1171 /* osp_declare_attr_set() ignores all attributes other than
1172 * UID, GID, and size, and osp_attr_set() ignores all but UID
1173 * and GID. Declaration of size attr setting happens through
1174 * lod_declare_init_size(), and not through this function.
1175 * Therefore we need not load striping unless ownership is
1176 * changing. This should save memory and (we hope) speed up
1178 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1179 if (!(attr->la_valid & (LA_UID | LA_GID)))
1182 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1185 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE |
1186 LA_ATIME | LA_MTIME | LA_CTIME |
1191 * load striping information, notice we don't do this when object
1192 * is being initialized as we don't need this information till
1193 * few specific cases like destroy, chown
1195 rc = lod_load_striping(env, lo);
1199 if (lo->ldo_stripenr == 0)
1203 * if object is striped declare changes on the stripes
1205 LASSERT(lo->ldo_stripe);
1206 for (i = 0; i < lo->ldo_stripenr; i++) {
1207 if (lo->ldo_stripe[i] == NULL)
1209 rc = lod_sub_object_declare_attr_set(env,
1210 lo->ldo_stripe[i], attr,
1216 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE) &&
1217 dt_object_exists(next) != 0 &&
1218 dt_object_remote(next) == 0)
1219 lod_sub_object_declare_xattr_del(env, next,
1220 XATTR_NAME_LOV, th);
1222 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) &&
1223 dt_object_exists(next) &&
1224 dt_object_remote(next) == 0 && S_ISREG(attr->la_mode)) {
1225 struct lod_thread_info *info = lod_env_info(env);
1226 struct lu_buf *buf = &info->lti_buf;
1228 buf->lb_buf = info->lti_ea_store;
1229 buf->lb_len = info->lti_ea_store_size;
1230 lod_sub_object_declare_xattr_set(env, next, buf,
1232 LU_XATTR_REPLACE, th);
1239 * Implementation of dt_object_operations::do_attr_set.
1241 * If the object is striped, then apply the changes to all or subset of
1242 * the stripes depending on the object type and specific attributes.
1244 * \see dt_object_operations::do_attr_set() in the API description for details.
1246 static int lod_attr_set(const struct lu_env *env,
1247 struct dt_object *dt,
1248 const struct lu_attr *attr,
1251 struct dt_object *next = dt_object_child(dt);
1252 struct lod_object *lo = lod_dt_obj(dt);
1256 /* Set dead object on all other stripes */
1257 if (attr->la_valid & LA_FLAGS && !(attr->la_valid & ~LA_FLAGS) &&
1258 attr->la_flags & LUSTRE_SLAVE_DEAD_FL) {
1259 rc = lod_mark_dead_object(env, dt, th, false);
1264 * apply changes to the local object
1266 rc = lod_sub_object_attr_set(env, next, attr, th);
1270 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1271 if (!(attr->la_valid & (LA_UID | LA_GID)))
1274 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1277 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE |
1278 LA_ATIME | LA_MTIME | LA_CTIME |
1283 if (lo->ldo_stripenr == 0)
1287 * if object is striped, apply changes to all the stripes
1289 LASSERT(lo->ldo_stripe);
1290 for (i = 0; i < lo->ldo_stripenr; i++) {
1291 if (unlikely(lo->ldo_stripe[i] == NULL))
1294 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
1295 (dt_object_exists(lo->ldo_stripe[i]) == 0))
1298 rc = lod_sub_object_attr_set(env, lo->ldo_stripe[i], attr, th);
1303 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE) &&
1304 dt_object_exists(next) != 0 &&
1305 dt_object_remote(next) == 0)
1306 rc = lod_sub_object_xattr_del(env, next, XATTR_NAME_LOV, th);
1308 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) &&
1309 dt_object_exists(next) &&
1310 dt_object_remote(next) == 0 && S_ISREG(attr->la_mode)) {
1311 struct lod_thread_info *info = lod_env_info(env);
1312 struct lu_buf *buf = &info->lti_buf;
1313 struct ost_id *oi = &info->lti_ostid;
1314 struct lu_fid *fid = &info->lti_fid;
1315 struct lov_mds_md_v1 *lmm;
1316 struct lov_ost_data_v1 *objs;
1320 rc1 = lod_get_lov_ea(env, lo);
1324 buf->lb_buf = info->lti_ea_store;
1325 buf->lb_len = info->lti_ea_store_size;
1326 lmm = info->lti_ea_store;
1327 magic = le32_to_cpu(lmm->lmm_magic);
1328 if (magic == LOV_MAGIC_V1)
1329 objs = &(lmm->lmm_objects[0]);
1331 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1332 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1333 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1335 fid_to_ostid(fid, oi);
1336 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1338 rc = lod_sub_object_xattr_set(env, next, buf, XATTR_NAME_LOV,
1339 LU_XATTR_REPLACE, th);
1346 * Implementation of dt_object_operations::do_xattr_get.
1348 * If LOV EA is requested from the root object and it's not
1349 * found, then return default striping for the filesystem.
1351 * \see dt_object_operations::do_xattr_get() in the API description for details.
1353 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1354 struct lu_buf *buf, const char *name)
1356 struct lod_thread_info *info = lod_env_info(env);
1357 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1361 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1362 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1363 struct lmv_mds_md_v1 *lmv1;
1366 if (rc > (typeof(rc))sizeof(*lmv1))
1369 if (rc < (typeof(rc))sizeof(*lmv1))
1370 RETURN(rc = rc > 0 ? -EINVAL : rc);
1372 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1373 CLASSERT(sizeof(*lmv1) <= sizeof(info->lti_key));
1375 info->lti_buf.lb_buf = info->lti_key;
1376 info->lti_buf.lb_len = sizeof(*lmv1);
1377 rc = dt_xattr_get(env, dt_object_child(dt),
1378 &info->lti_buf, name);
1379 if (unlikely(rc != sizeof(*lmv1)))
1380 RETURN(rc = rc > 0 ? -EINVAL : rc);
1382 lmv1 = info->lti_buf.lb_buf;
1383 /* The on-disk LMV EA only contains header, but the
1384 * returned LMV EA size should contain the space for
1385 * the FIDs of all shards of the striped directory. */
1386 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1387 rc = lmv_mds_md_size(
1388 le32_to_cpu(lmv1->lmv_stripe_count),
1391 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1395 RETURN(rc = rc1 != 0 ? rc1 : rc);
1398 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1402 * lod returns default striping on the real root of the device
1403 * this is like the root stores default striping for the whole
1404 * filesystem. historically we've been using a different approach
1405 * and store it in the config.
1407 dt_root_get(env, dev->lod_child, &info->lti_fid);
1408 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1410 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1411 struct lov_user_md *lum = buf->lb_buf;
1412 struct lov_desc *desc = &dev->lod_desc;
1414 if (buf->lb_buf == NULL) {
1416 } else if (buf->lb_len >= sizeof(*lum)) {
1417 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1418 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1419 lmm_oi_set_id(&lum->lmm_oi, 0);
1420 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1421 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1422 lum->lmm_stripe_size = cpu_to_le32(
1423 desc->ld_default_stripe_size);
1424 lum->lmm_stripe_count = cpu_to_le16(
1425 desc->ld_default_stripe_count);
1426 lum->lmm_stripe_offset = cpu_to_le16(
1427 desc->ld_default_stripe_offset);
1440 * Checks that the magic of the stripe is sane.
1442 * \param[in] lod lod device
1443 * \param[in] lum a buffer storing LMV EA to verify
1445 * \retval 0 if the EA is sane
1446 * \retval negative otherwise
1448 static int lod_verify_md_striping(struct lod_device *lod,
1449 const struct lmv_user_md_v1 *lum)
1451 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1452 CERROR("%s: invalid lmv_user_md: magic = %x, "
1453 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1454 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1455 (int)le32_to_cpu(lum->lum_stripe_offset),
1456 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1464 * Initialize LMV EA for a slave.
1466 * Initialize slave's LMV EA from the master's LMV EA.
1468 * \param[in] master_lmv a buffer containing master's EA
1469 * \param[out] slave_lmv a buffer where slave's EA will be stored
1472 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1473 const struct lmv_mds_md_v1 *master_lmv)
1475 *slave_lmv = *master_lmv;
1476 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1482 * Generate LMV EA from the object passed as \a dt. The object must have
1483 * the stripes created and initialized.
1485 * \param[in] env execution environment
1486 * \param[in] dt object
1487 * \param[out] lmv_buf buffer storing generated LMV EA
1489 * \retval 0 on success
1490 * \retval negative if failed
1492 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1493 struct lu_buf *lmv_buf)
1495 struct lod_thread_info *info = lod_env_info(env);
1496 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1497 struct lod_object *lo = lod_dt_obj(dt);
1498 struct lmv_mds_md_v1 *lmm1;
1500 int type = LU_SEQ_RANGE_ANY;
1505 LASSERT(lo->ldo_dir_striped != 0);
1506 LASSERT(lo->ldo_stripenr > 0);
1507 stripe_count = lo->ldo_stripenr;
1508 /* Only store the LMV EA heahder on the disk. */
1509 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1510 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1514 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1517 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1518 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1519 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1520 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1521 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1526 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1527 lmv_buf->lb_buf = info->lti_ea_store;
1528 lmv_buf->lb_len = sizeof(*lmm1);
1534 * Create in-core represenation for a striped directory.
1536 * Parse the buffer containing LMV EA and instantiate LU objects
1537 * representing the stripe objects. The pointers to the objects are
1538 * stored in ldo_stripe field of \a lo. This function is used when
1539 * we need to access an already created object (i.e. load from a disk).
1541 * \param[in] env execution environment
1542 * \param[in] lo lod object
1543 * \param[in] buf buffer containing LMV EA
1545 * \retval 0 on success
1546 * \retval negative if failed
1548 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1549 const struct lu_buf *buf)
1551 struct lod_thread_info *info = lod_env_info(env);
1552 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1553 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1554 struct dt_object **stripe;
1555 union lmv_mds_md *lmm = buf->lb_buf;
1556 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1557 struct lu_fid *fid = &info->lti_fid;
1562 if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
1565 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1566 lo->ldo_dir_slave_stripe = 1;
1570 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1573 if (le32_to_cpu(lmv1->lmv_stripe_count) < 1)
1576 LASSERT(lo->ldo_stripe == NULL);
1577 OBD_ALLOC(stripe, sizeof(stripe[0]) *
1578 (le32_to_cpu(lmv1->lmv_stripe_count)));
1582 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1583 struct dt_device *tgt_dt;
1584 struct dt_object *dto;
1585 int type = LU_SEQ_RANGE_ANY;
1588 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1589 if (!fid_is_sane(fid))
1590 GOTO(out, rc = -ESTALE);
1592 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1596 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1597 tgt_dt = lod->lod_child;
1599 struct lod_tgt_desc *tgt;
1601 tgt = LTD_TGT(ltd, idx);
1603 GOTO(out, rc = -ESTALE);
1604 tgt_dt = tgt->ltd_tgt;
1607 dto = dt_locate_at(env, tgt_dt, fid,
1608 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1611 GOTO(out, rc = PTR_ERR(dto));
1616 lo->ldo_stripe = stripe;
1617 lo->ldo_stripenr = le32_to_cpu(lmv1->lmv_stripe_count);
1618 lo->ldo_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1620 lod_object_free_striping(env, lo);
1626 * Declare create a striped directory.
1628 * Declare creating a striped directory with a given stripe pattern on the
1629 * specified MDTs. A striped directory is represented as a regular directory
1630 * - an index listing all the stripes. The stripes point back to the master
1631 * object with ".." and LinkEA. The master object gets LMV EA which
1632 * identifies it as a striped directory. The function allocates FIDs
1635 * \param[in] env execution environment
1636 * \param[in] dt object
1637 * \param[in] attr attributes to initialize the objects with
1638 * \param[in] dof type of objects to be created
1639 * \param[in] th transaction handle
1641 * \retval 0 on success
1642 * \retval negative if failed
1644 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1645 struct dt_object *dt,
1646 struct lu_attr *attr,
1647 struct dt_object_format *dof,
1650 struct lod_thread_info *info = lod_env_info(env);
1651 struct lu_buf lmv_buf;
1652 struct lu_buf slave_lmv_buf;
1653 struct lmv_mds_md_v1 *lmm;
1654 struct lmv_mds_md_v1 *slave_lmm = NULL;
1655 struct dt_insert_rec *rec = &info->lti_dt_rec;
1656 struct lod_object *lo = lod_dt_obj(dt);
1661 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1664 lmm = lmv_buf.lb_buf;
1666 OBD_ALLOC_PTR(slave_lmm);
1667 if (slave_lmm == NULL)
1668 GOTO(out, rc = -ENOMEM);
1670 lod_prep_slave_lmv_md(slave_lmm, lmm);
1671 slave_lmv_buf.lb_buf = slave_lmm;
1672 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1674 if (!dt_try_as_dir(env, dt_object_child(dt)))
1675 GOTO(out, rc = -EINVAL);
1677 rec->rec_type = S_IFDIR;
1678 for (i = 0; i < lo->ldo_stripenr; i++) {
1679 struct dt_object *dto = lo->ldo_stripe[i];
1680 char *stripe_name = info->lti_key;
1681 struct lu_name *sname;
1682 struct linkea_data ldata = { NULL };
1683 struct lu_buf linkea_buf;
1685 rc = lod_sub_object_declare_create(env, dto, attr, NULL,
1690 if (!dt_try_as_dir(env, dto))
1691 GOTO(out, rc = -EINVAL);
1693 rc = lod_sub_object_declare_ref_add(env, dto, th);
1697 rec->rec_fid = lu_object_fid(&dto->do_lu);
1698 rc = lod_sub_object_declare_insert(env, dto,
1699 (const struct dt_rec *)rec,
1700 (const struct dt_key *)dot, th);
1704 /* master stripe FID will be put to .. */
1705 rec->rec_fid = lu_object_fid(&dt->do_lu);
1706 rc = lod_sub_object_declare_insert(env, dto,
1707 (const struct dt_rec *)rec,
1708 (const struct dt_key *)dotdot,
1713 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1714 cfs_fail_val != i) {
1715 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1717 slave_lmm->lmv_master_mdt_index =
1720 slave_lmm->lmv_master_mdt_index =
1722 rc = lod_sub_object_declare_xattr_set(env, dto,
1723 &slave_lmv_buf, XATTR_NAME_LMV, 0, th);
1728 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1730 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1731 PFID(lu_object_fid(&dto->do_lu)), i + 1);
1733 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1734 PFID(lu_object_fid(&dto->do_lu)), i);
1736 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
1737 rc = linkea_data_new(&ldata, &info->lti_linkea_buf);
1741 rc = linkea_add_buf(&ldata, sname, lu_object_fid(&dt->do_lu));
1745 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1746 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1747 rc = lod_sub_object_declare_xattr_set(env, dto, &linkea_buf,
1748 XATTR_NAME_LINK, 0, th);
1752 rec->rec_fid = lu_object_fid(&dto->do_lu);
1753 rc = lod_sub_object_declare_insert(env, dt_object_child(dt),
1754 (const struct dt_rec *)rec,
1755 (const struct dt_key *)stripe_name,
1760 rc = lod_sub_object_declare_ref_add(env, dt_object_child(dt),
1766 rc = lod_sub_object_declare_xattr_set(env, dt_object_child(dt),
1767 &lmv_buf, XATTR_NAME_LMV, 0, th);
1771 if (slave_lmm != NULL)
1772 OBD_FREE_PTR(slave_lmm);
1777 static int lod_prep_md_striped_create(const struct lu_env *env,
1778 struct dt_object *dt,
1779 struct lu_attr *attr,
1780 const struct lmv_user_md_v1 *lum,
1781 struct dt_object_format *dof,
1784 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1785 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1786 struct lod_object *lo = lod_dt_obj(dt);
1787 struct dt_object **stripe;
1795 /* The lum has been verifed in lod_verify_md_striping */
1796 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
1797 LASSERT(le32_to_cpu(lum->lum_stripe_count) > 0);
1799 stripe_count = le32_to_cpu(lum->lum_stripe_count);
1801 /* shrink the stripe_count to the avaible MDT count */
1802 if (stripe_count > lod->lod_remote_mdt_count + 1)
1803 stripe_count = lod->lod_remote_mdt_count + 1;
1805 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_count);
1809 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
1810 if (idx_array == NULL)
1811 GOTO(out_free, rc = -ENOMEM);
1813 for (i = 0; i < stripe_count; i++) {
1814 struct lod_tgt_desc *tgt = NULL;
1815 struct dt_object *dto;
1816 struct lu_fid fid = { 0 };
1818 struct lu_object_conf conf = { 0 };
1819 struct dt_device *tgt_dt = NULL;
1822 /* Right now, master stripe and master object are
1823 * on the same MDT */
1824 idx = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1825 rc = obd_fid_alloc(env, lod->lod_child_exp, &fid,
1829 tgt_dt = lod->lod_child;
1833 idx = (idx_array[i - 1] + 1) % (lod->lod_remote_mdt_count + 1);
1835 for (j = 0; j < lod->lod_remote_mdt_count;
1836 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
1837 bool already_allocated = false;
1840 CDEBUG(D_INFO, "try idx %d, mdt cnt %u,"
1841 " allocated %u, last allocated %d\n", idx,
1842 lod->lod_remote_mdt_count, i, idx_array[i - 1]);
1844 /* Find next available target */
1845 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx))
1848 /* check whether the idx already exists
1849 * in current allocated array */
1850 for (k = 0; k < i; k++) {
1851 if (idx_array[k] == idx) {
1852 already_allocated = true;
1857 if (already_allocated)
1860 /* check the status of the OSP */
1861 tgt = LTD_TGT(ltd, idx);
1865 tgt_dt = tgt->ltd_tgt;
1866 rc = dt_statfs(env, tgt_dt, NULL);
1868 /* this OSP doesn't feel well */
1873 rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
1882 /* Can not allocate more stripes */
1883 if (j == lod->lod_remote_mdt_count) {
1884 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
1885 lod2obd(lod)->obd_name, stripe_count, i - 1);
1889 CDEBUG(D_INFO, "idx %d, mdt cnt %u,"
1890 " allocated %u, last allocated %d\n", idx,
1891 lod->lod_remote_mdt_count, i, idx_array[i - 1]);
1894 /* tgt_dt and fid must be ready after search avaible OSP
1895 * in the above loop */
1896 LASSERT(tgt_dt != NULL);
1897 LASSERT(fid_is_sane(&fid));
1898 conf.loc_flags = LOC_F_NEW;
1899 dto = dt_locate_at(env, tgt_dt, &fid,
1900 dt->do_lu.lo_dev->ld_site->ls_top_dev,
1903 GOTO(out_put, rc = PTR_ERR(dto));
1908 lo->ldo_dir_striped = 1;
1909 lo->ldo_stripe = stripe;
1910 lo->ldo_stripenr = i;
1911 lo->ldo_stripes_allocated = stripe_count;
1913 if (lo->ldo_stripenr == 0)
1914 GOTO(out_put, rc = -ENOSPC);
1916 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
1922 for (i = 0; i < stripe_count; i++)
1923 if (stripe[i] != NULL)
1924 lu_object_put(env, &stripe[i]->do_lu);
1925 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_count);
1926 lo->ldo_stripenr = 0;
1927 lo->ldo_stripes_allocated = 0;
1928 lo->ldo_stripe = NULL;
1932 if (idx_array != NULL)
1933 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
1939 * Declare create striped md object.
1941 * The function declares intention to create a striped directory. This is a
1942 * wrapper for lod_prep_md_striped_create(). The only additional functionality
1943 * is to verify pattern \a lum_buf is good. Check that function for the details.
1945 * \param[in] env execution environment
1946 * \param[in] dt object
1947 * \param[in] attr attributes to initialize the objects with
1948 * \param[in] lum_buf a pattern specifying the number of stripes and
1950 * \param[in] dof type of objects to be created
1951 * \param[in] th transaction handle
1953 * \retval 0 on success
1954 * \retval negative if failed
1957 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
1958 struct dt_object *dt,
1959 struct lu_attr *attr,
1960 const struct lu_buf *lum_buf,
1961 struct dt_object_format *dof,
1964 struct lod_object *lo = lod_dt_obj(dt);
1965 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1966 struct lmv_user_md_v1 *lum;
1970 lum = lum_buf->lb_buf;
1971 LASSERT(lum != NULL);
1973 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
1974 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
1975 (int)le32_to_cpu(lum->lum_stripe_offset));
1977 if (le32_to_cpu(lum->lum_stripe_count) == 0)
1980 rc = lod_verify_md_striping(lod, lum);
1984 /* prepare dir striped objects */
1985 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
1987 /* failed to create striping, let's reset
1988 * config so that others don't get confused */
1989 lod_object_free_striping(env, lo);
1998 * Implementation of dt_object_operations::do_declare_xattr_set.
2000 * Used with regular (non-striped) objects. Basically it
2001 * initializes the striping information and applies the
2002 * change to all the stripes.
2004 * \see dt_object_operations::do_declare_xattr_set() in the API description
2007 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2008 struct dt_object *dt,
2009 const struct lu_buf *buf,
2010 const char *name, int fl,
2013 struct dt_object *next = dt_object_child(dt);
2014 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2015 struct lod_object *lo = lod_dt_obj(dt);
2020 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2021 struct lmv_user_md_v1 *lum;
2023 LASSERT(buf != NULL && buf->lb_buf != NULL);
2025 rc = lod_verify_md_striping(d, lum);
2030 rc = lod_sub_object_declare_xattr_set(env, next, buf, name, fl, th);
2034 /* set xattr to each stripes, if needed */
2035 rc = lod_load_striping(env, lo);
2039 /* Note: Do not set LinkEA on sub-stripes, otherwise
2040 * it will confuse the fid2path process(see mdt_path_current()).
2041 * The linkEA between master and sub-stripes is set in
2042 * lod_xattr_set_lmv(). */
2043 if (lo->ldo_stripenr == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
2046 for (i = 0; i < lo->ldo_stripenr; i++) {
2047 LASSERT(lo->ldo_stripe[i]);
2049 rc = lod_sub_object_declare_xattr_set(env, lo->ldo_stripe[i],
2059 * Implementation of dt_object_operations::do_declare_xattr_set.
2061 * \see dt_object_operations::do_declare_xattr_set() in the API description
2064 * the extension to the API:
2065 * - declaring LOVEA requests striping creation
2066 * - LU_XATTR_REPLACE means layout swap
2068 static int lod_declare_xattr_set(const struct lu_env *env,
2069 struct dt_object *dt,
2070 const struct lu_buf *buf,
2071 const char *name, int fl,
2074 struct dt_object *next = dt_object_child(dt);
2075 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2081 * allow to declare predefined striping on a new (!mode) object
2082 * which is supposed to be replay of regular file creation
2083 * (when LOV setting is declared)
2084 * LU_XATTR_REPLACE is set to indicate a layout swap
2086 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
2087 if ((S_ISREG(mode) || mode == 0) && strcmp(name, XATTR_NAME_LOV) == 0 &&
2088 !(fl & LU_XATTR_REPLACE)) {
2090 * this is a request to manipulate object's striping
2092 if (dt_object_exists(dt)) {
2093 rc = dt_attr_get(env, next, attr);
2097 memset(attr, 0, sizeof(*attr));
2098 attr->la_valid = LA_TYPE | LA_MODE;
2099 attr->la_mode = S_IFREG;
2101 rc = lod_declare_striped_object(env, dt, attr, buf, th);
2102 } else if (S_ISDIR(mode)) {
2103 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
2105 rc = lod_sub_object_declare_xattr_set(env, next, buf, name,
2113 * Resets cached default striping in the object.
2115 * \param[in] lo object
2117 static void lod_lov_stripe_cache_clear(struct lod_object *lo)
2119 lo->ldo_def_striping_set = 0;
2120 lo->ldo_def_striping_cached = 0;
2121 lod_object_set_pool(lo, NULL);
2122 lo->ldo_def_stripe_size = 0;
2123 lo->ldo_def_stripenr = 0;
2124 if (lo->ldo_dir_stripe != NULL)
2125 lo->ldo_dir_def_striping_cached = 0;
2129 * Apply xattr changes to the object.
2131 * Applies xattr changes to the object and the stripes if the latter exist.
2133 * \param[in] env execution environment
2134 * \param[in] dt object
2135 * \param[in] buf buffer pointing to the new value of xattr
2136 * \param[in] name name of xattr
2137 * \param[in] fl flags
2138 * \param[in] th transaction handle
2140 * \retval 0 on success
2141 * \retval negative if failed
2143 static int lod_xattr_set_internal(const struct lu_env *env,
2144 struct dt_object *dt,
2145 const struct lu_buf *buf,
2146 const char *name, int fl,
2149 struct dt_object *next = dt_object_child(dt);
2150 struct lod_object *lo = lod_dt_obj(dt);
2155 rc = lod_sub_object_xattr_set(env, next, buf, name, fl, th);
2156 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2159 /* Note: Do not set LinkEA on sub-stripes, otherwise
2160 * it will confuse the fid2path process(see mdt_path_current()).
2161 * The linkEA between master and sub-stripes is set in
2162 * lod_xattr_set_lmv(). */
2163 if (lo->ldo_stripenr == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
2166 for (i = 0; i < lo->ldo_stripenr; i++) {
2167 LASSERT(lo->ldo_stripe[i]);
2169 rc = lod_sub_object_xattr_set(env, lo->ldo_stripe[i], buf, name,
2179 * Delete an extended attribute.
2181 * Deletes specified xattr from the object and the stripes if the latter exist.
2183 * \param[in] env execution environment
2184 * \param[in] dt object
2185 * \param[in] name name of xattr
2186 * \param[in] th transaction handle
2188 * \retval 0 on success
2189 * \retval negative if failed
2191 static int lod_xattr_del_internal(const struct lu_env *env,
2192 struct dt_object *dt,
2193 const char *name, struct thandle *th)
2195 struct dt_object *next = dt_object_child(dt);
2196 struct lod_object *lo = lod_dt_obj(dt);
2201 rc = lod_sub_object_xattr_del(env, next, name, th);
2202 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2205 if (lo->ldo_stripenr == 0)
2208 for (i = 0; i < lo->ldo_stripenr; i++) {
2209 LASSERT(lo->ldo_stripe[i]);
2211 rc = lod_sub_object_xattr_del(env, lo->ldo_stripe[i], name,
2221 * Set default striping on a directory.
2223 * Sets specified striping on a directory object unless it matches the default
2224 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
2225 * EA. This striping will be used when regular file is being created in this
2228 * \param[in] env execution environment
2229 * \param[in] dt the striped object
2230 * \param[in] buf buffer with the striping
2231 * \param[in] name name of EA
2232 * \param[in] fl xattr flag (see OSD API description)
2233 * \param[in] th transaction handle
2235 * \retval 0 on success
2236 * \retval negative if failed
2238 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
2239 struct dt_object *dt,
2240 const struct lu_buf *buf,
2241 const char *name, int fl,
2244 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2245 struct lod_object *l = lod_dt_obj(dt);
2246 struct lov_user_md_v1 *lum;
2247 struct lov_user_md_v3 *v3 = NULL;
2248 const char *pool_name = NULL;
2252 /* If it is striped dir, we should clear the stripe cache for
2253 * slave stripe as well, but there are no effective way to
2254 * notify the LOD on the slave MDT, so we do not cache stripe
2255 * information for slave stripe for now. XXX*/
2256 lod_lov_stripe_cache_clear(l);
2257 LASSERT(buf != NULL && buf->lb_buf != NULL);
2260 rc = lod_verify_striping(d, buf, false);
2264 if (lum->lmm_magic == LOV_USER_MAGIC_V3) {
2266 if (v3->lmm_pool_name[0] != '\0')
2267 pool_name = v3->lmm_pool_name;
2270 /* if { size, offset, count } = { 0, -1, 0 } and no pool
2271 * (i.e. all default values specified) then delete default
2272 * striping from dir. */
2274 "set default striping: sz %u # %u offset %d %s %s\n",
2275 (unsigned)lum->lmm_stripe_size,
2276 (unsigned)lum->lmm_stripe_count,
2277 (int)lum->lmm_stripe_offset,
2278 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
2280 if (LOVEA_DELETE_VALUES(lum->lmm_stripe_size, lum->lmm_stripe_count,
2281 lum->lmm_stripe_offset, pool_name)) {
2282 rc = lod_xattr_del_internal(env, dt, name, th);
2286 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
2293 * Set default striping on a directory object.
2295 * Sets specified striping on a directory object unless it matches the default
2296 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
2297 * EA. This striping will be used when a new directory is being created in the
2300 * \param[in] env execution environment
2301 * \param[in] dt the striped object
2302 * \param[in] buf buffer with the striping
2303 * \param[in] name name of EA
2304 * \param[in] fl xattr flag (see OSD API description)
2305 * \param[in] th transaction handle
2307 * \retval 0 on success
2308 * \retval negative if failed
2310 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
2311 struct dt_object *dt,
2312 const struct lu_buf *buf,
2313 const char *name, int fl,
2316 struct lod_object *l = lod_dt_obj(dt);
2317 struct lmv_user_md_v1 *lum;
2321 LASSERT(buf != NULL && buf->lb_buf != NULL);
2324 CDEBUG(D_OTHER, "set default stripe_count # %u stripe_offset %d\n",
2325 le32_to_cpu(lum->lum_stripe_count),
2326 (int)le32_to_cpu(lum->lum_stripe_offset));
2328 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
2329 le32_to_cpu(lum->lum_stripe_offset)) &&
2330 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
2331 rc = lod_xattr_del_internal(env, dt, name, th);
2335 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
2340 /* Update default stripe cache */
2341 if (l->ldo_dir_stripe == NULL) {
2342 OBD_ALLOC_PTR(l->ldo_dir_stripe);
2343 if (l->ldo_dir_stripe == NULL)
2347 l->ldo_dir_def_striping_cached = 0;
2352 * Turn directory into a striped directory.
2354 * During replay the client sends the striping created before MDT
2355 * failure, then the layer above LOD sends this defined striping
2356 * using ->do_xattr_set(), so LOD uses this method to replay creation
2357 * of the stripes. Notice the original information for the striping
2358 * (#stripes, FIDs, etc) was transferred in declare path.
2360 * \param[in] env execution environment
2361 * \param[in] dt the striped object
2362 * \param[in] buf not used currently
2363 * \param[in] name not used currently
2364 * \param[in] fl xattr flag (see OSD API description)
2365 * \param[in] th transaction handle
2367 * \retval 0 on success
2368 * \retval negative if failed
2370 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
2371 const struct lu_buf *buf, const char *name,
2372 int fl, struct thandle *th)
2374 struct lod_object *lo = lod_dt_obj(dt);
2375 struct lod_thread_info *info = lod_env_info(env);
2376 struct lu_attr *attr = &info->lti_attr;
2377 struct dt_object_format *dof = &info->lti_format;
2378 struct lu_buf lmv_buf;
2379 struct lu_buf slave_lmv_buf;
2380 struct lmv_mds_md_v1 *lmm;
2381 struct lmv_mds_md_v1 *slave_lmm = NULL;
2382 struct dt_insert_rec *rec = &info->lti_dt_rec;
2387 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
2390 /* The stripes are supposed to be allocated in declare phase,
2391 * if there are no stripes being allocated, it will skip */
2392 if (lo->ldo_stripenr == 0)
2395 rc = dt_attr_get(env, dt_object_child(dt), attr);
2399 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME |
2400 LA_MODE | LA_UID | LA_GID | LA_TYPE;
2401 dof->dof_type = DFT_DIR;
2403 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
2406 lmm = lmv_buf.lb_buf;
2408 OBD_ALLOC_PTR(slave_lmm);
2409 if (slave_lmm == NULL)
2412 lod_prep_slave_lmv_md(slave_lmm, lmm);
2413 slave_lmv_buf.lb_buf = slave_lmm;
2414 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
2416 rec->rec_type = S_IFDIR;
2417 for (i = 0; i < lo->ldo_stripenr; i++) {
2418 struct dt_object *dto;
2419 char *stripe_name = info->lti_key;
2420 struct lu_name *sname;
2421 struct linkea_data ldata = { NULL };
2422 struct lu_buf linkea_buf;
2424 dto = lo->ldo_stripe[i];
2426 dt_write_lock(env, dto, MOR_TGT_CHILD);
2427 rc = lod_sub_object_create(env, dto, attr, NULL, dof,
2430 dt_write_unlock(env, dto);
2434 rc = lod_sub_object_ref_add(env, dto, th);
2435 dt_write_unlock(env, dto);
2439 rec->rec_fid = lu_object_fid(&dto->do_lu);
2440 rc = lod_sub_object_index_insert(env, dto,
2441 (const struct dt_rec *)rec,
2442 (const struct dt_key *)dot, th, 0);
2446 rec->rec_fid = lu_object_fid(&dt->do_lu);
2447 rc = lod_sub_object_index_insert(env, dto, (struct dt_rec *)rec,
2448 (const struct dt_key *)dotdot, th, 0);
2452 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
2453 cfs_fail_val != i) {
2454 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
2456 slave_lmm->lmv_master_mdt_index =
2459 slave_lmm->lmv_master_mdt_index =
2462 rc = lod_sub_object_xattr_set(env, dto, &slave_lmv_buf,
2463 XATTR_NAME_LMV, fl, th);
2468 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
2470 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2471 PFID(lu_object_fid(&dto->do_lu)), i + 1);
2473 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2474 PFID(lu_object_fid(&dto->do_lu)), i);
2476 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
2477 rc = linkea_data_new(&ldata, &info->lti_linkea_buf);
2481 rc = linkea_add_buf(&ldata, sname, lu_object_fid(&dt->do_lu));
2485 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
2486 linkea_buf.lb_len = ldata.ld_leh->leh_len;
2487 rc = lod_sub_object_xattr_set(env, dto, &linkea_buf,
2488 XATTR_NAME_LINK, 0, th);
2492 rec->rec_fid = lu_object_fid(&dto->do_lu);
2493 rc = lod_sub_object_index_insert(env, dt_object_child(dt),
2494 (const struct dt_rec *)rec,
2495 (const struct dt_key *)stripe_name, th, 0);
2499 rc = lod_sub_object_ref_add(env, dt_object_child(dt), th);
2504 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
2505 rc = lod_sub_object_xattr_set(env, dt_object_child(dt),
2506 &lmv_buf, XATTR_NAME_LMV, fl, th);
2508 if (slave_lmm != NULL)
2509 OBD_FREE_PTR(slave_lmm);
2515 * Helper function to declare/execute creation of a striped directory
2517 * Called in declare/create object path, prepare striping for a directory
2518 * and prepare defaults data striping for the objects to be created in
2519 * that directory. Notice the function calls "declaration" or "execution"
2520 * methods depending on \a declare param. This is a consequence of the
2521 * current approach while we don't have natural distributed transactions:
2522 * we basically execute non-local updates in the declare phase. So, the
2523 * arguments for the both phases are the same and this is the reason for
2524 * this function to exist.
2526 * \param[in] env execution environment
2527 * \param[in] dt object
2528 * \param[in] attr attributes the stripes will be created with
2529 * \param[in] dof format of stripes (see OSD API description)
2530 * \param[in] th transaction handle
2531 * \param[in] declare where to call "declare" or "execute" methods
2533 * \retval 0 on success
2534 * \retval negative if failed
2536 static int lod_dir_striping_create_internal(const struct lu_env *env,
2537 struct dt_object *dt,
2538 struct lu_attr *attr,
2539 struct dt_object_format *dof,
2543 struct lod_thread_info *info = lod_env_info(env);
2544 struct lod_object *lo = lod_dt_obj(dt);
2548 if (!LMVEA_DELETE_VALUES(lo->ldo_stripenr,
2549 lo->ldo_dir_stripe_offset)) {
2550 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
2551 int stripe_count = lo->ldo_stripenr;
2553 if (info->lti_ea_store_size < sizeof(*v1)) {
2554 rc = lod_ea_store_resize(info, sizeof(*v1));
2557 v1 = info->lti_ea_store;
2560 memset(v1, 0, sizeof(*v1));
2561 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
2562 v1->lum_stripe_count = cpu_to_le32(stripe_count);
2563 v1->lum_stripe_offset =
2564 cpu_to_le32(lo->ldo_dir_stripe_offset);
2566 info->lti_buf.lb_buf = v1;
2567 info->lti_buf.lb_len = sizeof(*v1);
2570 rc = lod_declare_xattr_set_lmv(env, dt, attr,
2571 &info->lti_buf, dof, th);
2573 rc = lod_xattr_set_lmv(env, dt, &info->lti_buf,
2574 XATTR_NAME_LMV, 0, th);
2579 /* Transfer default LMV striping from the parent */
2580 if (lo->ldo_dir_def_striping_set &&
2581 !LMVEA_DELETE_VALUES(lo->ldo_dir_def_stripenr,
2582 lo->ldo_dir_def_stripe_offset)) {
2583 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
2584 int def_stripe_count = lo->ldo_dir_def_stripenr;
2586 if (info->lti_ea_store_size < sizeof(*v1)) {
2587 rc = lod_ea_store_resize(info, sizeof(*v1));
2590 v1 = info->lti_ea_store;
2593 memset(v1, 0, sizeof(*v1));
2594 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
2595 v1->lum_stripe_count = cpu_to_le32(def_stripe_count);
2596 v1->lum_stripe_offset =
2597 cpu_to_le32(lo->ldo_dir_def_stripe_offset);
2599 cpu_to_le32(lo->ldo_dir_def_hash_type);
2601 info->lti_buf.lb_buf = v1;
2602 info->lti_buf.lb_len = sizeof(*v1);
2604 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
2605 XATTR_NAME_DEFAULT_LMV,
2608 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
2610 XATTR_NAME_DEFAULT_LMV, 0,
2616 /* Transfer default LOV striping from the parent */
2617 if (lo->ldo_def_striping_set &&
2618 !LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size,
2619 lo->ldo_def_stripenr,
2620 lo->ldo_def_stripe_offset,
2622 struct lov_user_md_v3 *v3 = info->lti_ea_store;
2624 if (info->lti_ea_store_size < sizeof(*v3)) {
2625 rc = lod_ea_store_resize(info, sizeof(*v3));
2628 v3 = info->lti_ea_store;
2631 memset(v3, 0, sizeof(*v3));
2632 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
2633 v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
2634 v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
2635 v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
2636 if (lo->ldo_pool != NULL)
2637 strlcpy(v3->lmm_pool_name, lo->ldo_pool,
2638 sizeof(v3->lmm_pool_name));
2640 info->lti_buf.lb_buf = v3;
2641 info->lti_buf.lb_len = sizeof(*v3);
2644 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
2645 XATTR_NAME_LOV, 0, th);
2647 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
2648 XATTR_NAME_LOV, 0, th);
2656 static int lod_declare_dir_striping_create(const struct lu_env *env,
2657 struct dt_object *dt,
2658 struct lu_attr *attr,
2659 struct dt_object_format *dof,
2662 return lod_dir_striping_create_internal(env, dt, attr, dof, th, true);
2665 static int lod_dir_striping_create(const struct lu_env *env,
2666 struct dt_object *dt,
2667 struct lu_attr *attr,
2668 struct dt_object_format *dof,
2671 struct lod_object *lo = lod_dt_obj(dt);
2674 rc = lod_dir_striping_create_internal(env, dt, attr, dof, th, false);
2676 lo->ldo_striping_cached = 1;
2682 * Implementation of dt_object_operations::do_xattr_set.
2684 * Sets specified extended attribute on the object. Three types of EAs are
2686 * LOV EA - stores striping for a regular file or default striping (when set
2688 * LMV EA - stores a marker for the striped directories
2689 * DMV EA - stores default directory striping
2691 * When striping is applied to a non-striped existing object (this is called
2692 * late striping), then LOD notices the caller wants to turn the object into a
2693 * striped one. The stripe objects are created and appropriate EA is set:
2694 * LOV EA storing all the stripes directly or LMV EA storing just a small header
2695 * with striping configuration.
2697 * \see dt_object_operations::do_xattr_set() in the API description for details.
2699 static int lod_xattr_set(const struct lu_env *env,
2700 struct dt_object *dt, const struct lu_buf *buf,
2701 const char *name, int fl, struct thandle *th)
2703 struct dt_object *next = dt_object_child(dt);
2707 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
2708 strcmp(name, XATTR_NAME_LMV) == 0) {
2709 struct lmv_mds_md_v1 *lmm = buf->lb_buf;
2711 if (lmm != NULL && le32_to_cpu(lmm->lmv_hash_type) &
2712 LMV_HASH_FLAG_MIGRATION)
2713 rc = lod_sub_object_xattr_set(env, next, buf, name, fl,
2716 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
2721 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
2722 strcmp(name, XATTR_NAME_LOV) == 0) {
2724 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl, th);
2726 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
2727 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2729 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
2732 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
2733 !strcmp(name, XATTR_NAME_LOV)) {
2734 /* in case of lov EA swap, just set it
2735 * if not, it is a replay so check striping match what we
2736 * already have during req replay, declare_xattr_set()
2737 * defines striping, then create() does the work */
2738 if (fl & LU_XATTR_REPLACE) {
2739 /* free stripes, then update disk */
2740 lod_object_free_striping(env, lod_dt_obj(dt));
2742 rc = lod_sub_object_xattr_set(env, next, buf, name,
2744 } else if (dt_object_remote(dt)) {
2745 /* This only happens during migration, see
2746 * mdd_migrate_create(), in which Master MDT will
2747 * create a remote target object, and only set
2748 * (migrating) stripe EA on the remote object,
2749 * and does not need creating each stripes. */
2750 rc = lod_sub_object_xattr_set(env, next, buf, name,
2753 rc = lod_striping_create(env, dt, NULL, NULL, th);
2758 /* then all other xattr */
2759 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
2765 * Implementation of dt_object_operations::do_declare_xattr_del.
2767 * \see dt_object_operations::do_declare_xattr_del() in the API description
2770 static int lod_declare_xattr_del(const struct lu_env *env,
2771 struct dt_object *dt, const char *name,
2774 struct lod_object *lo = lod_dt_obj(dt);
2779 rc = lod_sub_object_declare_xattr_del(env, dt_object_child(dt),
2784 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
2787 /* set xattr to each stripes, if needed */
2788 rc = lod_load_striping(env, lo);
2792 if (lo->ldo_stripenr == 0)
2795 for (i = 0; i < lo->ldo_stripenr; i++) {
2796 LASSERT(lo->ldo_stripe[i]);
2797 rc = lod_sub_object_declare_xattr_del(env, lo->ldo_stripe[i],
2807 * Implementation of dt_object_operations::do_xattr_del.
2809 * If EA storing a regular striping is being deleted, then release
2810 * all the references to the stripe objects in core.
2812 * \see dt_object_operations::do_xattr_del() in the API description for details.
2814 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
2815 const char *name, struct thandle *th)
2817 struct dt_object *next = dt_object_child(dt);
2818 struct lod_object *lo = lod_dt_obj(dt);
2823 if (!strcmp(name, XATTR_NAME_LOV))
2824 lod_object_free_striping(env, lod_dt_obj(dt));
2826 rc = lod_sub_object_xattr_del(env, next, name, th);
2827 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2830 if (lo->ldo_stripenr == 0)
2833 for (i = 0; i < lo->ldo_stripenr; i++) {
2834 LASSERT(lo->ldo_stripe[i]);
2836 rc = lod_sub_object_xattr_del(env, lo->ldo_stripe[i], name, th);
2845 * Implementation of dt_object_operations::do_xattr_list.
2847 * \see dt_object_operations::do_xattr_list() in the API description
2850 static int lod_xattr_list(const struct lu_env *env,
2851 struct dt_object *dt, const struct lu_buf *buf)
2853 return dt_xattr_list(env, dt_object_child(dt), buf);
2857 * Initialize a pool the object belongs to.
2859 * When a striped object is being created, striping configuration
2860 * may demand the stripes are allocated on a limited set of the
2861 * targets. These limited sets are known as "pools". So we copy
2862 * a pool name into the object and later actual creation methods
2863 * (like lod_object_create()) will use this information to allocate
2864 * the stripes properly.
2866 * \param[in] o object
2867 * \param[in] pool pool name
2869 int lod_object_set_pool(struct lod_object *o, char *pool)
2874 len = strlen(o->ldo_pool);
2875 OBD_FREE(o->ldo_pool, len + 1);
2880 OBD_ALLOC(o->ldo_pool, len + 1);
2881 if (o->ldo_pool == NULL)
2883 strcpy(o->ldo_pool, pool);
2888 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
2890 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
2895 * Cache default regular striping in the object.
2897 * To improve performance of striped regular object creation we cache
2898 * default LOV striping (if it exists) in the parent directory object.
2900 * \param[in] env execution environment
2901 * \param[in] lp object
2903 * \retval 0 on success
2904 * \retval negative if failed
2906 static int lod_cache_parent_lov_striping(const struct lu_env *env,
2907 struct lod_object *lp)
2909 struct lod_thread_info *info = lod_env_info(env);
2910 struct lov_user_md_v1 *v1 = NULL;
2911 struct lov_user_md_v3 *v3 = NULL;
2915 /* called from MDD without parent being write locked,
2917 dt_write_lock(env, dt_object_child(&lp->ldo_obj), 0);
2918 rc = lod_get_lov_ea(env, lp);
2922 if (rc < (typeof(rc))sizeof(struct lov_user_md)) {
2923 /* don't lookup for non-existing or invalid striping */
2924 lp->ldo_def_striping_set = 0;
2925 lp->ldo_def_striping_cached = 1;
2926 lp->ldo_def_stripe_size = 0;
2927 lp->ldo_def_stripenr = 0;
2928 lp->ldo_def_stripe_offset = (typeof(v1->lmm_stripe_offset))(-1);
2929 GOTO(unlock, rc = 0);
2933 v1 = info->lti_ea_store;
2934 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
2935 lustre_swab_lov_user_md_v1(v1);
2936 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
2937 v3 = (struct lov_user_md_v3 *)v1;
2938 lustre_swab_lov_user_md_v3(v3);
2941 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1)
2942 GOTO(unlock, rc = 0);
2944 if (v1->lmm_pattern != LOV_PATTERN_RAID0 && v1->lmm_pattern != 0)
2945 GOTO(unlock, rc = 0);
2947 CDEBUG(D_INFO, DFID" stripe_count=%d stripe_size=%d stripe_offset=%d\n",
2948 PFID(lu_object_fid(&lp->ldo_obj.do_lu)),
2949 (int)v1->lmm_stripe_count,
2950 (int)v1->lmm_stripe_size, (int)v1->lmm_stripe_offset);
2952 lp->ldo_def_stripenr = v1->lmm_stripe_count;
2953 lp->ldo_def_stripe_size = v1->lmm_stripe_size;
2954 lp->ldo_def_stripe_offset = v1->lmm_stripe_offset;
2955 lp->ldo_def_striping_cached = 1;
2956 lp->ldo_def_striping_set = 1;
2957 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2958 /* XXX: sanity check here */
2959 v3 = (struct lov_user_md_v3 *) v1;
2960 if (v3->lmm_pool_name[0])
2961 lod_object_set_pool(lp, v3->lmm_pool_name);
2965 dt_write_unlock(env, dt_object_child(&lp->ldo_obj));
2971 * Cache default directory striping in the object.
2973 * To improve performance of striped directory creation we cache default
2974 * directory striping (if it exists) in the parent directory object.
2976 * \param[in] env execution environment
2977 * \param[in] lp object
2979 * \retval 0 on success
2980 * \retval negative if failed
2982 static int lod_cache_parent_lmv_striping(const struct lu_env *env,
2983 struct lod_object *lp)
2985 struct lod_thread_info *info = lod_env_info(env);
2986 struct lmv_user_md_v1 *v1 = NULL;
2990 /* called from MDD without parent being write locked,
2992 dt_write_lock(env, dt_object_child(&lp->ldo_obj), 0);
2993 rc = lod_get_default_lmv_ea(env, lp);
2997 if (rc < (typeof(rc))sizeof(struct lmv_user_md)) {
2998 /* don't lookup for non-existing or invalid striping */
2999 lp->ldo_dir_def_striping_set = 0;
3000 lp->ldo_dir_def_striping_cached = 1;
3001 lp->ldo_dir_def_stripenr = 0;
3002 lp->ldo_dir_def_stripe_offset =
3003 (typeof(v1->lum_stripe_offset))(-1);
3004 lp->ldo_dir_def_hash_type = LMV_HASH_TYPE_FNV_1A_64;
3005 GOTO(unlock, rc = 0);
3009 v1 = info->lti_ea_store;
3011 lp->ldo_dir_def_stripenr = le32_to_cpu(v1->lum_stripe_count);
3012 lp->ldo_dir_def_stripe_offset = le32_to_cpu(v1->lum_stripe_offset);
3013 lp->ldo_dir_def_hash_type = le32_to_cpu(v1->lum_hash_type);
3014 lp->ldo_dir_def_striping_set = 1;
3015 lp->ldo_dir_def_striping_cached = 1;
3019 dt_write_unlock(env, dt_object_child(&lp->ldo_obj));
3024 * Cache default striping in the object.
3026 * To improve performance of striped object creation we cache default striping
3027 * (if it exists) in the parent directory object. We always cache default
3028 * striping for the regular files (stored in LOV EA) and we cache default
3029 * striping for the directories if requested by \a child_mode (when a new
3030 * directory is being created).
3032 * \param[in] env execution environment
3033 * \param[in] lp object
3034 * \param[in] child_mode new object's mode
3036 * \retval 0 on success
3037 * \retval negative if failed
3039 static int lod_cache_parent_striping(const struct lu_env *env,
3040 struct lod_object *lp,
3046 if (!lp->ldo_def_striping_cached) {
3047 /* we haven't tried to get default striping for
3048 * the directory yet, let's cache it in the object */
3049 rc = lod_cache_parent_lov_striping(env, lp);
3054 /* If the parent is on the remote MDT, we should always
3055 * try to refresh the default stripeEA cache, because we
3056 * do not cache default striping information for remote
3058 if (S_ISDIR(child_mode) && (!lp->ldo_dir_def_striping_cached ||
3059 dt_object_remote(&lp->ldo_obj)))
3060 rc = lod_cache_parent_lmv_striping(env, lp);
3066 * Implementation of dt_object_operations::do_ah_init.
3068 * This method is used to make a decision on the striping configuration for the
3069 * object being created. It can be taken from the \a parent object if it exists,
3070 * or filesystem's default. The resulting configuration (number of stripes,
3071 * stripe size/offset, pool name, etc) is stored in the object itself and will
3072 * be used by the methods like ->doo_declare_create().
3074 * \see dt_object_operations::do_ah_init() in the API description for details.
3076 static void lod_ah_init(const struct lu_env *env,
3077 struct dt_allocation_hint *ah,
3078 struct dt_object *parent,
3079 struct dt_object *child,
3082 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
3083 struct dt_object *nextp = NULL;
3084 struct dt_object *nextc;
3085 struct lod_object *lp = NULL;
3086 struct lod_object *lc;
3087 struct lov_desc *desc;
3093 if (likely(parent)) {
3094 nextp = dt_object_child(parent);
3095 lp = lod_dt_obj(parent);
3096 rc = lod_load_striping(env, lp);
3101 nextc = dt_object_child(child);
3102 lc = lod_dt_obj(child);
3104 LASSERT(lc->ldo_stripenr == 0);
3105 LASSERT(lc->ldo_stripe == NULL);
3108 * local object may want some hints
3109 * in case of late striping creation, ->ah_init()
3110 * can be called with local object existing
3112 if (!dt_object_exists(nextc) || dt_object_remote(nextc)) {
3113 struct dt_object *obj;
3115 obj = (nextp != NULL && dt_object_remote(nextp)) ? NULL : nextp;
3116 nextc->do_ops->do_ah_init(env, ah, obj, nextc, child_mode);
3119 if (S_ISDIR(child_mode)) {
3120 if (lc->ldo_dir_stripe == NULL) {
3121 OBD_ALLOC_PTR(lc->ldo_dir_stripe);
3122 if (lc->ldo_dir_stripe == NULL)
3126 LASSERT(lp != NULL);
3127 if (lp->ldo_dir_stripe == NULL) {
3128 OBD_ALLOC_PTR(lp->ldo_dir_stripe);
3129 if (lp->ldo_dir_stripe == NULL)
3133 rc = lod_cache_parent_striping(env, lp, child_mode);
3137 /* transfer defaults to new directory */
3138 if (lp->ldo_def_striping_set) {
3140 lod_object_set_pool(lc, lp->ldo_pool);
3141 lc->ldo_def_stripenr = lp->ldo_def_stripenr;
3142 lc->ldo_def_stripe_size = lp->ldo_def_stripe_size;
3143 lc->ldo_def_stripe_offset = lp->ldo_def_stripe_offset;
3144 lc->ldo_def_striping_set = 1;
3145 lc->ldo_def_striping_cached = 1;
3146 CDEBUG(D_OTHER, "inherite EA sz:%d off:%d nr:%d\n",
3147 (int)lc->ldo_def_stripe_size,
3148 (int)lc->ldo_def_stripe_offset,
3149 (int)lc->ldo_def_stripenr);
3152 /* transfer dir defaults to new directory */
3153 if (lp->ldo_dir_def_striping_set) {
3154 lc->ldo_dir_def_stripenr = lp->ldo_dir_def_stripenr;
3155 lc->ldo_dir_def_stripe_offset =
3156 lp->ldo_dir_def_stripe_offset;
3157 lc->ldo_dir_def_hash_type =
3158 lp->ldo_dir_def_hash_type;
3159 lc->ldo_dir_def_striping_set = 1;
3160 lc->ldo_dir_def_striping_cached = 1;
3161 CDEBUG(D_INFO, "inherit default EA nr:%d off:%d t%u\n",
3162 (int)lc->ldo_dir_def_stripenr,
3163 (int)lc->ldo_dir_def_stripe_offset,
3164 lc->ldo_dir_def_hash_type);
3167 /* It should always honour the specified stripes */
3168 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0) {
3169 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
3171 rc = lod_verify_md_striping(d, lum1);
3173 le32_to_cpu(lum1->lum_stripe_count) > 1) {
3174 /* Directory will be striped only if
3175 * stripe_count > 1 */
3177 le32_to_cpu(lum1->lum_stripe_count);
3178 lc->ldo_dir_stripe_offset =
3179 le32_to_cpu(lum1->lum_stripe_offset);
3180 lc->ldo_dir_hash_type =
3181 le32_to_cpu(lum1->lum_hash_type);
3182 CDEBUG(D_INFO, "set stripe EA nr:%hu off:%d\n",
3184 (int)lc->ldo_dir_stripe_offset);
3186 /* then check whether there is default stripes from parent */
3187 } else if (lp->ldo_dir_def_striping_set) {
3188 /* If there are default dir stripe from parent */
3189 lc->ldo_stripenr = lp->ldo_dir_def_stripenr;
3190 lc->ldo_dir_stripe_offset =
3191 lp->ldo_dir_def_stripe_offset;
3192 lc->ldo_dir_hash_type =
3193 lp->ldo_dir_def_hash_type;
3194 CDEBUG(D_INFO, "inherit EA nr:%hu off:%d\n",
3196 (int)lc->ldo_dir_stripe_offset);
3198 /* set default stripe for this directory */
3199 lc->ldo_stripenr = 0;
3200 lc->ldo_dir_stripe_offset = -1;
3203 CDEBUG(D_INFO, "final striping count:%hu, offset:%d\n",
3204 lc->ldo_stripenr, (int)lc->ldo_dir_stripe_offset);
3210 * if object is going to be striped over OSTs, transfer default
3211 * striping information to the child, so that we can use it
3212 * during declaration and creation
3214 if (!lod_object_will_be_striped(S_ISREG(child_mode),
3215 lu_object_fid(&child->do_lu)))
3218 * try from the parent
3220 if (likely(parent)) {
3221 lod_cache_parent_striping(env, lp, child_mode);
3223 lc->ldo_def_stripe_offset = LOV_OFFSET_DEFAULT;
3225 if (lp->ldo_def_striping_set) {
3227 lod_object_set_pool(lc, lp->ldo_pool);
3228 lc->ldo_stripenr = lp->ldo_def_stripenr;
3229 lc->ldo_stripe_size = lp->ldo_def_stripe_size;
3230 lc->ldo_def_stripe_offset = lp->ldo_def_stripe_offset;
3231 CDEBUG(D_OTHER, "striping from parent: #%d, sz %d %s\n",
3232 lc->ldo_stripenr, lc->ldo_stripe_size,
3233 lp->ldo_pool ? lp->ldo_pool : "");
3238 * if the parent doesn't provide with specific pattern, grab fs-wide one
3240 desc = &d->lod_desc;
3241 if (lc->ldo_stripenr == 0)
3242 lc->ldo_stripenr = desc->ld_default_stripe_count;
3243 if (lc->ldo_stripe_size == 0)
3244 lc->ldo_stripe_size = desc->ld_default_stripe_size;
3245 CDEBUG(D_OTHER, "final striping: # %d stripes, sz %d from %s\n",
3246 lc->ldo_stripenr, lc->ldo_stripe_size,
3247 lc->ldo_pool ? lc->ldo_pool : "");
3250 /* we do not cache stripe information for slave stripe, see
3251 * lod_xattr_set_lov_on_dir */
3252 if (lp != NULL && lp->ldo_dir_slave_stripe)
3253 lod_lov_stripe_cache_clear(lp);
3258 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
3260 * Size initialization on late striping.
3262 * Propagate the size of a truncated object to a deferred striping.
3263 * This function handles a special case when truncate was done on a
3264 * non-striped object and now while the striping is being created
3265 * we can't lose that size, so we have to propagate it to the stripes
3268 * \param[in] env execution environment
3269 * \param[in] dt object
3270 * \param[in] th transaction handle
3272 * \retval 0 on success
3273 * \retval negative if failed
3275 static int lod_declare_init_size(const struct lu_env *env,
3276 struct dt_object *dt, struct thandle *th)
3278 struct dt_object *next = dt_object_child(dt);
3279 struct lod_object *lo = lod_dt_obj(dt);
3280 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3281 uint64_t size, offs;
3285 /* XXX: we support the simplest (RAID0) striping so far */
3286 LASSERT(lo->ldo_stripe || lo->ldo_stripenr == 0);
3287 LASSERT(lo->ldo_stripe_size > 0);
3289 rc = dt_attr_get(env, next, attr);
3290 LASSERT(attr->la_valid & LA_SIZE);
3294 size = attr->la_size;
3298 /* ll_do_div64(a, b) returns a % b, and a = a / b */
3299 ll_do_div64(size, (__u64) lo->ldo_stripe_size);
3300 stripe = ll_do_div64(size, (__u64) lo->ldo_stripenr);
3302 size = size * lo->ldo_stripe_size;
3303 offs = attr->la_size;
3304 size += ll_do_div64(offs, lo->ldo_stripe_size);
3306 attr->la_valid = LA_SIZE;
3307 attr->la_size = size;
3309 rc = lod_sub_object_declare_attr_set(env, lo->ldo_stripe[stripe], attr,
3316 * Declare creation of striped object.
3318 * The function declares creation stripes for a regular object. The function
3319 * also declares whether the stripes will be created with non-zero size if
3320 * previously size was set non-zero on the master object. If object \a dt is
3321 * not local, then only fully defined striping can be applied in \a lovea.
3322 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
3325 * \param[in] env execution environment
3326 * \param[in] dt object
3327 * \param[in] attr attributes the stripes will be created with
3328 * \param[in] lovea a buffer containing striping description
3329 * \param[in] th transaction handle
3331 * \retval 0 on success
3332 * \retval negative if failed
3334 int lod_declare_striped_object(const struct lu_env *env, struct dt_object *dt,
3335 struct lu_attr *attr,
3336 const struct lu_buf *lovea, struct thandle *th)
3338 struct lod_thread_info *info = lod_env_info(env);
3339 struct dt_object *next = dt_object_child(dt);
3340 struct lod_object *lo = lod_dt_obj(dt);
3344 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO)) {
3345 /* failed to create striping, let's reset
3346 * config so that others don't get confused */
3347 lod_object_free_striping(env, lo);
3348 GOTO(out, rc = -ENOMEM);
3351 if (!dt_object_remote(next)) {
3352 /* choose OST and generate appropriate objects */
3353 rc = lod_qos_prep_create(env, lo, attr, lovea, th);
3355 /* failed to create striping, let's reset
3356 * config so that others don't get confused */
3357 lod_object_free_striping(env, lo);
3362 * declare storage for striping data
3364 info->lti_buf.lb_len = lov_mds_md_size(lo->ldo_stripenr,
3365 lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1);
3367 /* LOD can not choose OST objects for remote objects, i.e.
3368 * stripes must be ready before that. Right now, it can only
3369 * happen during migrate, i.e. migrate process needs to create
3370 * remote regular file (mdd_migrate_create), then the migrate
3371 * process will provide stripeEA. */
3372 LASSERT(lovea != NULL);
3373 info->lti_buf = *lovea;
3376 rc = lod_sub_object_declare_xattr_set(env, next, &info->lti_buf,
3377 XATTR_NAME_LOV, 0, th);
3382 * if striping is created with local object's size > 0,
3383 * we have to propagate this size to specific object
3384 * the case is possible only when local object was created previously
3386 if (dt_object_exists(next))
3387 rc = lod_declare_init_size(env, dt, th);
3394 * Implementation of dt_object_operations::do_declare_create.
3396 * The method declares creation of a new object. If the object will be striped,
3397 * then helper functions are called to find FIDs for the stripes, declare
3398 * creation of the stripes and declare initialization of the striping
3399 * information to be stored in the master object.
3401 * \see dt_object_operations::do_declare_create() in the API description
3404 static int lod_declare_object_create(const struct lu_env *env,
3405 struct dt_object *dt,
3406 struct lu_attr *attr,
3407 struct dt_allocation_hint *hint,
3408 struct dt_object_format *dof,
3411 struct dt_object *next = dt_object_child(dt);
3412 struct lod_object *lo = lod_dt_obj(dt);
3421 * first of all, we declare creation of local object
3423 rc = lod_sub_object_declare_create(env, next, attr, hint, dof, th);
3427 if (dof->dof_type == DFT_SYM)
3428 dt->do_body_ops = &lod_body_lnk_ops;
3429 else if (dof->dof_type == DFT_REGULAR)
3430 dt->do_body_ops = &lod_body_ops;
3433 * it's lod_ah_init() that has decided the object will be striped
3435 if (dof->dof_type == DFT_REGULAR) {
3436 /* callers don't want stripes */
3437 /* XXX: all tricky interactions with ->ah_make_hint() decided
3438 * to use striping, then ->declare_create() behaving differently
3439 * should be cleaned */
3440 if (dof->u.dof_reg.striped == 0)
3441 lo->ldo_stripenr = 0;
3442 if (lo->ldo_stripenr > 0)
3443 rc = lod_declare_striped_object(env, dt, attr,
3445 } else if (dof->dof_type == DFT_DIR) {
3446 struct seq_server_site *ss;
3448 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
3450 /* If the parent has default stripeEA, and client
3451 * did not find it before sending create request,
3452 * then MDT will return -EREMOTE, and client will
3453 * retrieve the default stripeEA and re-create the
3456 * Note: if dah_eadata != NULL, it means creating the
3457 * striped directory with specified stripeEA, then it
3458 * should ignore the default stripeEA */
3459 if ((hint == NULL || hint->dah_eadata == NULL) &&
3460 lo->ldo_dir_stripe_offset != -1 &&
3461 lo->ldo_dir_stripe_offset != ss->ss_node_id)
3462 GOTO(out, rc = -EREMOTE);
3464 /* Orphan object (like migrating object) does not have
3465 * lod_dir_stripe, see lod_ah_init */
3466 if (lo->ldo_dir_stripe != NULL)
3467 rc = lod_declare_dir_striping_create(env, dt, attr,
3475 * Creation of a striped regular object.
3477 * The function is called to create the stripe objects for a regular
3478 * striped file. This can happen at the initial object creation or
3479 * when the caller asks LOD to do so using ->do_xattr_set() method
3480 * (so called late striping). Notice all the information are already
3481 * prepared in the form of the list of objects (ldo_stripe field).
3482 * This is done during declare phase.
3484 * \param[in] env execution environment
3485 * \param[in] dt object
3486 * \param[in] attr attributes the stripes will be created with
3487 * \param[in] dof format of stripes (see OSD API description)
3488 * \param[in] th transaction handle
3490 * \retval 0 on success
3491 * \retval negative if failed
3493 int lod_striping_create(const struct lu_env *env, struct dt_object *dt,
3494 struct lu_attr *attr, struct dt_object_format *dof,
3497 struct lod_object *lo = lod_dt_obj(dt);
3501 LASSERT(lo->ldo_striping_cached == 0);
3503 /* create all underlying objects */
3504 for (i = 0; i < lo->ldo_stripenr; i++) {
3505 LASSERT(lo->ldo_stripe[i]);
3506 rc = lod_sub_object_create(env, lo->ldo_stripe[i], attr, NULL,
3513 rc = lod_generate_and_set_lovea(env, lo, th);
3515 lo->ldo_striping_cached = 1;
3522 * Implementation of dt_object_operations::do_create.
3524 * If any of preceeding methods (like ->do_declare_create(),
3525 * ->do_ah_init(), etc) chose to create a striped object,
3526 * then this method will create the master and the stripes.
3528 * \see dt_object_operations::do_create() in the API description for details.
3530 static int lod_object_create(const struct lu_env *env, struct dt_object *dt,
3531 struct lu_attr *attr,
3532 struct dt_allocation_hint *hint,
3533 struct dt_object_format *dof, struct thandle *th)
3535 struct lod_object *lo = lod_dt_obj(dt);
3539 /* create local object */
3540 rc = lod_sub_object_create(env, dt_object_child(dt), attr, hint, dof,
3545 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
3546 lo->ldo_stripe && dof->u.dof_reg.striped != 0)
3547 rc = lod_striping_create(env, dt, attr, dof, th);
3553 * Implementation of dt_object_operations::do_declare_destroy.
3555 * If the object is a striped directory, then the function declares reference
3556 * removal from the master object (this is an index) to the stripes and declares
3557 * destroy of all the stripes. In all the cases, it declares an intention to
3558 * destroy the object itself.
3560 * \see dt_object_operations::do_declare_destroy() in the API description
3563 static int lod_declare_object_destroy(const struct lu_env *env,
3564 struct dt_object *dt,
3567 struct dt_object *next = dt_object_child(dt);
3568 struct lod_object *lo = lod_dt_obj(dt);
3569 struct lod_thread_info *info = lod_env_info(env);
3570 char *stripe_name = info->lti_key;
3575 * load striping information, notice we don't do this when object
3576 * is being initialized as we don't need this information till
3577 * few specific cases like destroy, chown
3579 rc = lod_load_striping(env, lo);
3583 /* declare destroy for all underlying objects */
3584 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
3585 rc = next->do_ops->do_index_try(env, next,
3586 &dt_directory_features);
3590 for (i = 0; i < lo->ldo_stripenr; i++) {
3591 rc = lod_sub_object_declare_ref_del(env, next, th);
3595 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3596 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
3598 rc = lod_sub_object_declare_delete(env, next,
3599 (const struct dt_key *)stripe_name, th);
3606 * we declare destroy for the local object
3608 rc = lod_sub_object_declare_destroy(env, next, th);
3612 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ))
3615 /* declare destroy all striped objects */
3616 for (i = 0; i < lo->ldo_stripenr; i++) {
3617 if (lo->ldo_stripe[i] == NULL)
3620 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
3621 rc = lod_sub_object_declare_ref_del(env,
3622 lo->ldo_stripe[i], th);
3624 rc = lod_sub_object_declare_destroy(env, lo->ldo_stripe[i],
3634 * Implementation of dt_object_operations::do_destroy.
3636 * If the object is a striped directory, then the function removes references
3637 * from the master object (this is an index) to the stripes and destroys all
3638 * the stripes. In all the cases, the function destroys the object itself.
3640 * \see dt_object_operations::do_destroy() in the API description for details.
3642 static int lod_object_destroy(const struct lu_env *env,
3643 struct dt_object *dt, struct thandle *th)
3645 struct dt_object *next = dt_object_child(dt);
3646 struct lod_object *lo = lod_dt_obj(dt);
3647 struct lod_thread_info *info = lod_env_info(env);
3648 char *stripe_name = info->lti_key;
3653 /* destroy sub-stripe of master object */
3654 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
3655 rc = next->do_ops->do_index_try(env, next,
3656 &dt_directory_features);
3660 for (i = 0; i < lo->ldo_stripenr; i++) {
3661 rc = lod_sub_object_ref_del(env, next, th);
3665 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3666 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
3669 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
3670 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
3671 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)));
3673 rc = lod_sub_object_delete(env, next,
3674 (const struct dt_key *)stripe_name, th);
3680 rc = lod_sub_object_destroy(env, next, th);
3684 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ))
3687 /* destroy all striped objects */
3688 for (i = 0; i < lo->ldo_stripenr; i++) {
3689 if (likely(lo->ldo_stripe[i] != NULL) &&
3690 (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
3691 i == cfs_fail_val)) {
3692 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
3693 dt_write_lock(env, lo->ldo_stripe[i],
3695 rc = lod_sub_object_ref_del(env,
3696 lo->ldo_stripe[i], th);
3697 dt_write_unlock(env, lo->ldo_stripe[i]);
3702 rc = lod_sub_object_destroy(env, lo->ldo_stripe[i], th);
3712 * Implementation of dt_object_operations::do_declare_ref_add.
3714 * \see dt_object_operations::do_declare_ref_add() in the API description
3717 static int lod_declare_ref_add(const struct lu_env *env,
3718 struct dt_object *dt, struct thandle *th)
3720 return lod_sub_object_declare_ref_add(env, dt_object_child(dt), th);
3724 * Implementation of dt_object_operations::do_ref_add.
3726 * \see dt_object_operations::do_ref_add() in the API description for details.
3728 static int lod_ref_add(const struct lu_env *env,
3729 struct dt_object *dt, struct thandle *th)
3731 return lod_sub_object_ref_add(env, dt_object_child(dt), th);
3735 * Implementation of dt_object_operations::do_declare_ref_del.
3737 * \see dt_object_operations::do_declare_ref_del() in the API description
3740 static int lod_declare_ref_del(const struct lu_env *env,
3741 struct dt_object *dt, struct thandle *th)
3743 return lod_sub_object_declare_ref_del(env, dt_object_child(dt), th);
3747 * Implementation of dt_object_operations::do_ref_del
3749 * \see dt_object_operations::do_ref_del() in the API description for details.
3751 static int lod_ref_del(const struct lu_env *env,
3752 struct dt_object *dt, struct thandle *th)
3754 return lod_sub_object_ref_del(env, dt_object_child(dt), th);
3758 * Implementation of dt_object_operations::do_object_sync.
3760 * \see dt_object_operations::do_object_sync() in the API description
3763 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
3764 __u64 start, __u64 end)
3766 return dt_object_sync(env, dt_object_child(dt), start, end);
3769 struct lod_slave_locks {
3771 struct lustre_handle lsl_handle[0];
3775 * Release LDLM locks on the stripes of a striped directory.
3777 * Iterates over all the locks taken on the stripe objects and
3778 * release them using ->do_object_unlock() method.
3780 * \param[in] env execution environment
3781 * \param[in] dt striped object
3782 * \param[in] einfo lock description
3783 * \param[in] policy data describing requested lock
3785 * \retval 0 on success
3786 * \retval negative if failed
3788 static int lod_object_unlock_internal(const struct lu_env *env,
3789 struct dt_object *dt,
3790 struct ldlm_enqueue_info *einfo,
3791 ldlm_policy_data_t *policy)
3793 struct lod_slave_locks *slave_locks = einfo->ei_cbdata;
3798 if (slave_locks == NULL)
3801 for (i = 1; i < slave_locks->lsl_lock_count; i++) {
3802 if (lustre_handle_is_used(&slave_locks->lsl_handle[i]))
3803 ldlm_lock_decref(&slave_locks->lsl_handle[i],
3811 * Implementation of dt_object_operations::do_object_unlock.
3813 * Used to release LDLM lock(s).
3815 * \see dt_object_operations::do_object_unlock() in the API description
3818 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
3819 struct ldlm_enqueue_info *einfo,
3820 union ldlm_policy_data *policy)
3822 struct lod_object *lo = lod_dt_obj(dt);
3823 struct lod_slave_locks *slave_locks = einfo->ei_cbdata;
3824 int slave_locks_size;
3828 if (slave_locks == NULL)
3831 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3834 /* Note: for remote lock for single stripe dir, MDT will cancel
3835 * the lock by lockh directly */
3836 if (lo->ldo_stripenr <= 1 && dt_object_remote(dt_object_child(dt)))
3839 /* Only cancel slave lock for striped dir */
3840 rc = lod_object_unlock_internal(env, dt, einfo, policy);
3842 slave_locks_size = sizeof(*slave_locks) + slave_locks->lsl_lock_count *
3843 sizeof(slave_locks->lsl_handle[0]);
3844 OBD_FREE(slave_locks, slave_locks_size);
3845 einfo->ei_cbdata = NULL;
3851 * Implementation of dt_object_operations::do_object_lock.
3853 * Used to get LDLM lock on the non-striped and striped objects.
3855 * \see dt_object_operations::do_object_lock() in the API description
3858 static int lod_object_lock(const struct lu_env *env,
3859 struct dt_object *dt,
3860 struct lustre_handle *lh,
3861 struct ldlm_enqueue_info *einfo,
3862 union ldlm_policy_data *policy)
3864 struct lod_object *lo = lod_dt_obj(dt);
3867 int slave_locks_size;
3868 struct lod_slave_locks *slave_locks = NULL;
3871 /* remote object lock */
3872 if (!einfo->ei_enq_slave) {
3873 LASSERT(dt_object_remote(dt));
3874 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
3878 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3881 rc = lod_load_striping(env, lo);
3886 if (lo->ldo_stripenr <= 1)
3889 slave_locks_size = sizeof(*slave_locks) + lo->ldo_stripenr *
3890 sizeof(slave_locks->lsl_handle[0]);
3891 /* Freed in lod_object_unlock */
3892 OBD_ALLOC(slave_locks, slave_locks_size);
3893 if (slave_locks == NULL)
3895 slave_locks->lsl_lock_count = lo->ldo_stripenr;
3897 /* striped directory lock */
3898 for (i = 1; i < lo->ldo_stripenr; i++) {
3899 struct lustre_handle lockh;
3900 struct ldlm_res_id *res_id;
3902 res_id = &lod_env_info(env)->lti_res_id;
3903 fid_build_reg_res_name(lu_object_fid(&lo->ldo_stripe[i]->do_lu),
3905 einfo->ei_res_id = res_id;
3907 LASSERT(lo->ldo_stripe[i]);
3908 rc = dt_object_lock(env, lo->ldo_stripe[i], &lockh, einfo,
3912 slave_locks->lsl_handle[i] = lockh;
3915 einfo->ei_cbdata = slave_locks;
3918 if (rc != 0 && slave_locks != NULL) {
3919 einfo->ei_cbdata = slave_locks;
3920 lod_object_unlock_internal(env, dt, einfo, policy);
3921 OBD_FREE(slave_locks, slave_locks_size);
3922 einfo->ei_cbdata = NULL;
3928 struct dt_object_operations lod_obj_ops = {
3929 .do_read_lock = lod_object_read_lock,
3930 .do_write_lock = lod_object_write_lock,
3931 .do_read_unlock = lod_object_read_unlock,
3932 .do_write_unlock = lod_object_write_unlock,
3933 .do_write_locked = lod_object_write_locked,
3934 .do_attr_get = lod_attr_get,
3935 .do_declare_attr_set = lod_declare_attr_set,
3936 .do_attr_set = lod_attr_set,
3937 .do_xattr_get = lod_xattr_get,
3938 .do_declare_xattr_set = lod_declare_xattr_set,
3939 .do_xattr_set = lod_xattr_set,
3940 .do_declare_xattr_del = lod_declare_xattr_del,
3941 .do_xattr_del = lod_xattr_del,
3942 .do_xattr_list = lod_xattr_list,
3943 .do_ah_init = lod_ah_init,
3944 .do_declare_create = lod_declare_object_create,
3945 .do_create = lod_object_create,
3946 .do_declare_destroy = lod_declare_object_destroy,
3947 .do_destroy = lod_object_destroy,
3948 .do_index_try = lod_index_try,
3949 .do_declare_ref_add = lod_declare_ref_add,
3950 .do_ref_add = lod_ref_add,
3951 .do_declare_ref_del = lod_declare_ref_del,
3952 .do_ref_del = lod_ref_del,
3953 .do_object_sync = lod_object_sync,
3954 .do_object_lock = lod_object_lock,
3955 .do_object_unlock = lod_object_unlock,
3959 * Implementation of dt_body_operations::dbo_read.
3961 * \see dt_body_operations::dbo_read() in the API description for details.
3963 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
3964 struct lu_buf *buf, loff_t *pos)
3966 struct dt_object *next = dt_object_child(dt);
3967 return next->do_body_ops->dbo_read(env, next, buf, pos);
3971 * Implementation of dt_body_operations::dbo_declare_write.
3973 * \see dt_body_operations::dbo_declare_write() in the API description
3976 static ssize_t lod_declare_write(const struct lu_env *env,
3977 struct dt_object *dt,
3978 const struct lu_buf *buf, loff_t pos,
3981 return lod_sub_object_declare_write(env, dt_object_child(dt), buf, pos,
3986 * Implementation of dt_body_operations::dbo_write.
3988 * \see dt_body_operations::dbo_write() in the API description for details.
3990 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
3991 const struct lu_buf *buf, loff_t *pos,
3992 struct thandle *th, int iq)
3994 return lod_sub_object_write(env, dt_object_child(dt), buf, pos, th, iq);
3997 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
3998 __u64 start, __u64 end, struct thandle *th)
4000 if (dt_object_remote(dt))
4003 return lod_sub_object_declare_punch(env, dt_object_child(dt), start,
4007 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
4008 __u64 start, __u64 end, struct thandle *th)
4010 if (dt_object_remote(dt))
4013 return lod_sub_object_punch(env, dt_object_child(dt), start, end, th);
4016 static const struct dt_body_operations lod_body_lnk_ops = {
4017 .dbo_read = lod_read,
4018 .dbo_declare_write = lod_declare_write,
4019 .dbo_write = lod_write
4022 static const struct dt_body_operations lod_body_ops = {
4023 .dbo_read = lod_read,
4024 .dbo_declare_write = lod_declare_write,
4025 .dbo_write = lod_write,
4026 .dbo_declare_punch = lod_declare_punch,
4027 .dbo_punch = lod_punch,
4031 * Implementation of lu_object_operations::loo_object_init.
4033 * The function determines the type and the index of the target device using
4034 * sequence of the object's FID. Then passes control down to the
4035 * corresponding device:
4036 * OSD for the local objects, OSP for remote
4038 * \see lu_object_operations::loo_object_init() in the API description
4041 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
4042 const struct lu_object_conf *conf)
4044 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
4045 struct lu_device *cdev = NULL;
4046 struct lu_object *cobj;
4047 struct lod_tgt_descs *ltd = NULL;
4048 struct lod_tgt_desc *tgt;
4050 int type = LU_SEQ_RANGE_ANY;
4054 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
4056 /* Note: Sometimes, it will Return EAGAIN here, see
4057 * ptrlpc_import_delay_req(), which might confuse
4058 * lu_object_find_at() and make it wait there incorrectly.
4059 * so we convert it to EIO here.*/
4066 if (type == LU_SEQ_RANGE_MDT &&
4067 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
4068 cdev = &lod->lod_child->dd_lu_dev;
4069 } else if (type == LU_SEQ_RANGE_MDT) {
4070 ltd = &lod->lod_mdt_descs;
4072 } else if (type == LU_SEQ_RANGE_OST) {
4073 ltd = &lod->lod_ost_descs;
4080 if (ltd->ltd_tgts_size > idx &&
4081 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
4082 tgt = LTD_TGT(ltd, idx);
4084 LASSERT(tgt != NULL);
4085 LASSERT(tgt->ltd_tgt != NULL);
4087 cdev = &(tgt->ltd_tgt->dd_lu_dev);
4089 lod_putref(lod, ltd);
4092 if (unlikely(cdev == NULL))
4095 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
4096 if (unlikely(cobj == NULL))
4099 lu_object_add(lo, cobj);
4106 * Release resources associated with striping.
4108 * If the object is striped (regular or directory), then release
4109 * the stripe objects references and free the ldo_stripe array.
4111 * \param[in] env execution environment
4112 * \param[in] lo object
4114 void lod_object_free_striping(const struct lu_env *env, struct lod_object *lo)
4118 if (lo->ldo_dir_stripe != NULL) {
4119 OBD_FREE_PTR(lo->ldo_dir_stripe);
4120 lo->ldo_dir_stripe = NULL;
4123 if (lo->ldo_stripe) {
4124 LASSERT(lo->ldo_stripes_allocated > 0);
4126 for (i = 0; i < lo->ldo_stripenr; i++) {
4127 if (lo->ldo_stripe[i])
4128 lu_object_put(env, &lo->ldo_stripe[i]->do_lu);
4131 i = sizeof(struct dt_object *) * lo->ldo_stripes_allocated;
4132 OBD_FREE(lo->ldo_stripe, i);
4133 lo->ldo_stripe = NULL;
4134 lo->ldo_stripes_allocated = 0;
4136 lo->ldo_striping_cached = 0;
4137 lo->ldo_stripenr = 0;
4138 lo->ldo_pattern = 0;
4142 * Implementation of lu_object_operations::loo_object_start.
4144 * \see lu_object_operations::loo_object_start() in the API description
4147 static int lod_object_start(const struct lu_env *env, struct lu_object *o)
4149 if (S_ISLNK(o->lo_header->loh_attr & S_IFMT)) {
4150 lu2lod_obj(o)->ldo_obj.do_body_ops = &lod_body_lnk_ops;
4151 } else if (S_ISREG(o->lo_header->loh_attr & S_IFMT) ||
4152 fid_is_local_file(lu_object_fid(o))) {
4153 /* Note: some local file (like last rcvd) is created
4154 * through bottom layer (OSD), so the object initialization
4155 * comes to lod, it does not set loh_attr yet, so
4156 * set do_body_ops for local file anyway */
4157 lu2lod_obj(o)->ldo_obj.do_body_ops = &lod_body_ops;
4163 * Implementation of lu_object_operations::loo_object_free.
4165 * \see lu_object_operations::loo_object_free() in the API description
4168 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
4170 struct lod_object *mo = lu2lod_obj(o);
4173 * release all underlying object pinned
4176 lod_object_free_striping(env, mo);
4178 lod_object_set_pool(mo, NULL);
4181 OBD_SLAB_FREE_PTR(mo, lod_object_kmem);
4185 * Implementation of lu_object_operations::loo_object_release.
4187 * \see lu_object_operations::loo_object_release() in the API description
4190 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
4192 /* XXX: shouldn't we release everything here in case if object
4193 * creation failed before? */
4197 * Implementation of lu_object_operations::loo_object_print.
4199 * \see lu_object_operations::loo_object_print() in the API description
4202 static int lod_object_print(const struct lu_env *env, void *cookie,
4203 lu_printer_t p, const struct lu_object *l)
4205 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
4207 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
4210 struct lu_object_operations lod_lu_obj_ops = {
4211 .loo_object_init = lod_object_init,
4212 .loo_object_start = lod_object_start,
4213 .loo_object_free = lod_object_free,
4214 .loo_object_release = lod_object_release,
4215 .loo_object_print = lod_object_print,