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, 2015, Intel Corporation.
29 * lustre/lod/lod_object.c
31 * This file contains implementations of methods for the OSD API
32 * for the Logical Object Device (LOD) layer, which provides a virtual
33 * local OSD object interface to the MDD layer, and abstracts the
34 * addressing of local (OSD) and remote (OSP) objects. The API is
35 * described in the file lustre/include/dt_object.h and in
36 * Documentation/osd-api.txt.
38 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
41 #define DEBUG_SUBSYSTEM S_MDS
44 #include <obd_class.h>
45 #include <obd_support.h>
47 #include <lustre_fid.h>
48 #include <lustre_linkea.h>
49 #include <lustre_lmv.h>
50 #include <lustre_param.h>
51 #include <lustre_swab.h>
52 #include <lustre_ver.h>
53 #include <lprocfs_status.h>
54 #include <md_object.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 rc = next->do_ops->do_index_try(env, next, &dt_directory_features);
578 LASSERT(next->do_index_ops != NULL);
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 * Implementation of dt_object_operations::do_declare_attr_set.
1070 * If the object is striped, then apply the changes to all the stripes.
1072 * \see dt_object_operations::do_declare_attr_set() in the API description
1075 static int lod_declare_attr_set(const struct lu_env *env,
1076 struct dt_object *dt,
1077 const struct lu_attr *attr,
1080 struct dt_object *next = dt_object_child(dt);
1081 struct lod_object *lo = lod_dt_obj(dt);
1086 * declare setattr on the local object
1088 rc = lod_sub_object_declare_attr_set(env, next, attr, th);
1092 /* osp_declare_attr_set() ignores all attributes other than
1093 * UID, GID, and size, and osp_attr_set() ignores all but UID
1094 * and GID. Declaration of size attr setting happens through
1095 * lod_declare_init_size(), and not through this function.
1096 * Therefore we need not load striping unless ownership is
1097 * changing. This should save memory and (we hope) speed up
1099 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1100 if (!(attr->la_valid & (LA_UID | LA_GID)))
1103 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1106 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE |
1107 LA_ATIME | LA_MTIME | LA_CTIME |
1112 * load striping information, notice we don't do this when object
1113 * is being initialized as we don't need this information till
1114 * few specific cases like destroy, chown
1116 rc = lod_load_striping(env, lo);
1120 if (lo->ldo_stripenr == 0)
1124 * if object is striped declare changes on the stripes
1126 LASSERT(lo->ldo_stripe);
1127 for (i = 0; i < lo->ldo_stripenr; i++) {
1128 if (lo->ldo_stripe[i] == NULL)
1130 rc = lod_sub_object_declare_attr_set(env,
1131 lo->ldo_stripe[i], attr,
1137 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE) &&
1138 dt_object_exists(next) != 0 &&
1139 dt_object_remote(next) == 0)
1140 lod_sub_object_declare_xattr_del(env, next,
1141 XATTR_NAME_LOV, th);
1143 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) &&
1144 dt_object_exists(next) &&
1145 dt_object_remote(next) == 0 && S_ISREG(attr->la_mode)) {
1146 struct lod_thread_info *info = lod_env_info(env);
1147 struct lu_buf *buf = &info->lti_buf;
1149 buf->lb_buf = info->lti_ea_store;
1150 buf->lb_len = info->lti_ea_store_size;
1151 lod_sub_object_declare_xattr_set(env, next, buf,
1153 LU_XATTR_REPLACE, th);
1160 * Implementation of dt_object_operations::do_attr_set.
1162 * If the object is striped, then apply the changes to all or subset of
1163 * the stripes depending on the object type and specific attributes.
1165 * \see dt_object_operations::do_attr_set() in the API description for details.
1167 static int lod_attr_set(const struct lu_env *env,
1168 struct dt_object *dt,
1169 const struct lu_attr *attr,
1172 struct dt_object *next = dt_object_child(dt);
1173 struct lod_object *lo = lod_dt_obj(dt);
1178 * apply changes to the local object
1180 rc = lod_sub_object_attr_set(env, next, attr, th);
1184 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1185 if (!(attr->la_valid & (LA_UID | LA_GID)))
1188 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1191 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE |
1192 LA_ATIME | LA_MTIME | LA_CTIME |
1197 if (lo->ldo_stripenr == 0)
1201 * if object is striped, apply changes to all the stripes
1203 LASSERT(lo->ldo_stripe);
1204 for (i = 0; i < lo->ldo_stripenr; i++) {
1205 if (unlikely(lo->ldo_stripe[i] == NULL))
1208 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
1209 (dt_object_exists(lo->ldo_stripe[i]) == 0))
1212 rc = lod_sub_object_attr_set(env, lo->ldo_stripe[i], attr, th);
1217 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE) &&
1218 dt_object_exists(next) != 0 &&
1219 dt_object_remote(next) == 0)
1220 rc = lod_sub_object_xattr_del(env, next, 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;
1227 struct ost_id *oi = &info->lti_ostid;
1228 struct lu_fid *fid = &info->lti_fid;
1229 struct lov_mds_md_v1 *lmm;
1230 struct lov_ost_data_v1 *objs;
1234 rc1 = lod_get_lov_ea(env, lo);
1238 buf->lb_buf = info->lti_ea_store;
1239 buf->lb_len = info->lti_ea_store_size;
1240 lmm = info->lti_ea_store;
1241 magic = le32_to_cpu(lmm->lmm_magic);
1242 if (magic == LOV_MAGIC_V1)
1243 objs = &(lmm->lmm_objects[0]);
1245 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1246 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1247 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1249 fid_to_ostid(fid, oi);
1250 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1252 rc = lod_sub_object_xattr_set(env, next, buf, XATTR_NAME_LOV,
1253 LU_XATTR_REPLACE, th);
1260 * Implementation of dt_object_operations::do_xattr_get.
1262 * If LOV EA is requested from the root object and it's not
1263 * found, then return default striping for the filesystem.
1265 * \see dt_object_operations::do_xattr_get() in the API description for details.
1267 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1268 struct lu_buf *buf, const char *name)
1270 struct lod_thread_info *info = lod_env_info(env);
1271 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1275 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1276 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1277 struct lmv_mds_md_v1 *lmv1;
1280 if (rc > (typeof(rc))sizeof(*lmv1))
1283 if (rc < (typeof(rc))sizeof(*lmv1))
1284 RETURN(rc = rc > 0 ? -EINVAL : rc);
1286 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1287 CLASSERT(sizeof(*lmv1) <= sizeof(info->lti_key));
1289 info->lti_buf.lb_buf = info->lti_key;
1290 info->lti_buf.lb_len = sizeof(*lmv1);
1291 rc = dt_xattr_get(env, dt_object_child(dt),
1292 &info->lti_buf, name);
1293 if (unlikely(rc != sizeof(*lmv1)))
1294 RETURN(rc = rc > 0 ? -EINVAL : rc);
1296 lmv1 = info->lti_buf.lb_buf;
1297 /* The on-disk LMV EA only contains header, but the
1298 * returned LMV EA size should contain the space for
1299 * the FIDs of all shards of the striped directory. */
1300 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1301 rc = lmv_mds_md_size(
1302 le32_to_cpu(lmv1->lmv_stripe_count),
1305 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1309 RETURN(rc = rc1 != 0 ? rc1 : rc);
1312 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1316 * lod returns default striping on the real root of the device
1317 * this is like the root stores default striping for the whole
1318 * filesystem. historically we've been using a different approach
1319 * and store it in the config.
1321 dt_root_get(env, dev->lod_child, &info->lti_fid);
1322 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1324 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1325 struct lov_user_md *lum = buf->lb_buf;
1326 struct lov_desc *desc = &dev->lod_desc;
1328 if (buf->lb_buf == NULL) {
1330 } else if (buf->lb_len >= sizeof(*lum)) {
1331 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1332 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1333 lmm_oi_set_id(&lum->lmm_oi, 0);
1334 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1335 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1336 lum->lmm_stripe_size = cpu_to_le32(
1337 desc->ld_default_stripe_size);
1338 lum->lmm_stripe_count = cpu_to_le16(
1339 desc->ld_default_stripe_count);
1340 lum->lmm_stripe_offset = cpu_to_le16(
1341 desc->ld_default_stripe_offset);
1354 * Checks that the magic of the stripe is sane.
1356 * \param[in] lod lod device
1357 * \param[in] lum a buffer storing LMV EA to verify
1359 * \retval 0 if the EA is sane
1360 * \retval negative otherwise
1362 static int lod_verify_md_striping(struct lod_device *lod,
1363 const struct lmv_user_md_v1 *lum)
1365 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1366 CERROR("%s: invalid lmv_user_md: magic = %x, "
1367 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1368 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1369 (int)le32_to_cpu(lum->lum_stripe_offset),
1370 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1378 * Initialize LMV EA for a slave.
1380 * Initialize slave's LMV EA from the master's LMV EA.
1382 * \param[in] master_lmv a buffer containing master's EA
1383 * \param[out] slave_lmv a buffer where slave's EA will be stored
1386 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1387 const struct lmv_mds_md_v1 *master_lmv)
1389 *slave_lmv = *master_lmv;
1390 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1396 * Generate LMV EA from the object passed as \a dt. The object must have
1397 * the stripes created and initialized.
1399 * \param[in] env execution environment
1400 * \param[in] dt object
1401 * \param[out] lmv_buf buffer storing generated LMV EA
1403 * \retval 0 on success
1404 * \retval negative if failed
1406 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1407 struct lu_buf *lmv_buf)
1409 struct lod_thread_info *info = lod_env_info(env);
1410 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1411 struct lod_object *lo = lod_dt_obj(dt);
1412 struct lmv_mds_md_v1 *lmm1;
1414 int type = LU_SEQ_RANGE_ANY;
1419 LASSERT(lo->ldo_dir_striped != 0);
1420 LASSERT(lo->ldo_stripenr > 0);
1421 stripe_count = lo->ldo_stripenr;
1422 /* Only store the LMV EA heahder on the disk. */
1423 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1424 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1428 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1431 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1432 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1433 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1434 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1435 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1440 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1441 lmv_buf->lb_buf = info->lti_ea_store;
1442 lmv_buf->lb_len = sizeof(*lmm1);
1448 * Create in-core represenation for a striped directory.
1450 * Parse the buffer containing LMV EA and instantiate LU objects
1451 * representing the stripe objects. The pointers to the objects are
1452 * stored in ldo_stripe field of \a lo. This function is used when
1453 * we need to access an already created object (i.e. load from a disk).
1455 * \param[in] env execution environment
1456 * \param[in] lo lod object
1457 * \param[in] buf buffer containing LMV EA
1459 * \retval 0 on success
1460 * \retval negative if failed
1462 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1463 const struct lu_buf *buf)
1465 struct lod_thread_info *info = lod_env_info(env);
1466 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1467 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1468 struct dt_object **stripe;
1469 union lmv_mds_md *lmm = buf->lb_buf;
1470 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1471 struct lu_fid *fid = &info->lti_fid;
1476 if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
1479 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1480 lo->ldo_dir_slave_stripe = 1;
1484 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1487 if (le32_to_cpu(lmv1->lmv_stripe_count) < 1)
1490 LASSERT(lo->ldo_stripe == NULL);
1491 OBD_ALLOC(stripe, sizeof(stripe[0]) *
1492 (le32_to_cpu(lmv1->lmv_stripe_count)));
1496 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1497 struct dt_device *tgt_dt;
1498 struct dt_object *dto;
1499 int type = LU_SEQ_RANGE_ANY;
1502 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1503 if (!fid_is_sane(fid))
1504 GOTO(out, rc = -ESTALE);
1506 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1510 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1511 tgt_dt = lod->lod_child;
1513 struct lod_tgt_desc *tgt;
1515 tgt = LTD_TGT(ltd, idx);
1517 GOTO(out, rc = -ESTALE);
1518 tgt_dt = tgt->ltd_tgt;
1521 dto = dt_locate_at(env, tgt_dt, fid,
1522 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1525 GOTO(out, rc = PTR_ERR(dto));
1530 lo->ldo_stripe = stripe;
1531 lo->ldo_stripenr = le32_to_cpu(lmv1->lmv_stripe_count);
1532 lo->ldo_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1534 lod_object_free_striping(env, lo);
1540 * Declare create a striped directory.
1542 * Declare creating a striped directory with a given stripe pattern on the
1543 * specified MDTs. A striped directory is represented as a regular directory
1544 * - an index listing all the stripes. The stripes point back to the master
1545 * object with ".." and LinkEA. The master object gets LMV EA which
1546 * identifies it as a striped directory. The function allocates FIDs
1549 * \param[in] env execution environment
1550 * \param[in] dt object
1551 * \param[in] attr attributes to initialize the objects with
1552 * \param[in] dof type of objects to be created
1553 * \param[in] th transaction handle
1555 * \retval 0 on success
1556 * \retval negative if failed
1558 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1559 struct dt_object *dt,
1560 struct lu_attr *attr,
1561 struct dt_object_format *dof,
1564 struct lod_thread_info *info = lod_env_info(env);
1565 struct lu_buf lmv_buf;
1566 struct lu_buf slave_lmv_buf;
1567 struct lmv_mds_md_v1 *lmm;
1568 struct lmv_mds_md_v1 *slave_lmm = NULL;
1569 struct dt_insert_rec *rec = &info->lti_dt_rec;
1570 struct lod_object *lo = lod_dt_obj(dt);
1575 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1578 lmm = lmv_buf.lb_buf;
1580 OBD_ALLOC_PTR(slave_lmm);
1581 if (slave_lmm == NULL)
1582 GOTO(out, rc = -ENOMEM);
1584 lod_prep_slave_lmv_md(slave_lmm, lmm);
1585 slave_lmv_buf.lb_buf = slave_lmm;
1586 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1588 if (!dt_try_as_dir(env, dt_object_child(dt)))
1589 GOTO(out, rc = -EINVAL);
1591 rec->rec_type = S_IFDIR;
1592 for (i = 0; i < lo->ldo_stripenr; i++) {
1593 struct dt_object *dto = lo->ldo_stripe[i];
1594 char *stripe_name = info->lti_key;
1595 struct lu_name *sname;
1596 struct linkea_data ldata = { NULL };
1597 struct lu_buf linkea_buf;
1599 rc = lod_sub_object_declare_create(env, dto, attr, NULL,
1604 if (!dt_try_as_dir(env, dto))
1605 GOTO(out, rc = -EINVAL);
1607 rc = lod_sub_object_declare_ref_add(env, dto, th);
1611 rec->rec_fid = lu_object_fid(&dto->do_lu);
1612 rc = lod_sub_object_declare_insert(env, dto,
1613 (const struct dt_rec *)rec,
1614 (const struct dt_key *)dot, th);
1618 /* master stripe FID will be put to .. */
1619 rec->rec_fid = lu_object_fid(&dt->do_lu);
1620 rc = lod_sub_object_declare_insert(env, dto,
1621 (const struct dt_rec *)rec,
1622 (const struct dt_key *)dotdot,
1627 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1628 cfs_fail_val != i) {
1629 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1631 slave_lmm->lmv_master_mdt_index =
1634 slave_lmm->lmv_master_mdt_index =
1636 rc = lod_sub_object_declare_xattr_set(env, dto,
1637 &slave_lmv_buf, XATTR_NAME_LMV, 0, th);
1642 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1644 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1645 PFID(lu_object_fid(&dto->do_lu)), i + 1);
1647 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1648 PFID(lu_object_fid(&dto->do_lu)), i);
1650 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
1651 rc = linkea_data_new(&ldata, &info->lti_linkea_buf);
1655 rc = linkea_add_buf(&ldata, sname, lu_object_fid(&dt->do_lu));
1659 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1660 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1661 rc = lod_sub_object_declare_xattr_set(env, dto, &linkea_buf,
1662 XATTR_NAME_LINK, 0, th);
1666 rec->rec_fid = lu_object_fid(&dto->do_lu);
1667 rc = lod_sub_object_declare_insert(env, dt_object_child(dt),
1668 (const struct dt_rec *)rec,
1669 (const struct dt_key *)stripe_name,
1674 rc = lod_sub_object_declare_ref_add(env, dt_object_child(dt),
1680 rc = lod_sub_object_declare_xattr_set(env, dt_object_child(dt),
1681 &lmv_buf, XATTR_NAME_LMV, 0, th);
1685 if (slave_lmm != NULL)
1686 OBD_FREE_PTR(slave_lmm);
1691 static int lod_prep_md_striped_create(const struct lu_env *env,
1692 struct dt_object *dt,
1693 struct lu_attr *attr,
1694 const struct lmv_user_md_v1 *lum,
1695 struct dt_object_format *dof,
1698 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1699 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1700 struct lod_object *lo = lod_dt_obj(dt);
1701 struct dt_object **stripe;
1710 /* The lum has been verifed in lod_verify_md_striping */
1711 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
1712 LASSERT(le32_to_cpu(lum->lum_stripe_count) > 0);
1714 stripe_count = le32_to_cpu(lum->lum_stripe_count);
1716 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
1717 if (idx_array == NULL)
1720 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_count);
1722 GOTO(out_free, rc = -ENOMEM);
1724 /* Start index will be the master MDT */
1725 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1726 idx_array[0] = master_index;
1727 for (i = 0; i < stripe_count; i++) {
1728 struct lod_tgt_desc *tgt = NULL;
1729 struct dt_object *dto;
1730 struct lu_fid fid = { 0 };
1732 struct lu_object_conf conf = { 0 };
1733 struct dt_device *tgt_dt = NULL;
1735 /* Try to find next avaible target */
1737 for (j = 0; j < lod->lod_remote_mdt_count;
1738 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
1739 bool already_allocated = false;
1742 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
1743 idx, lod->lod_remote_mdt_count + 1, i);
1744 if (idx == master_index) {
1745 /* Allocate the FID locally */
1746 rc = obd_fid_alloc(env, lod->lod_child_exp,
1750 tgt_dt = lod->lod_child;
1754 /* Find next available target */
1755 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx))
1758 if (likely(!OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
1759 /* check whether the idx already exists
1760 * in current allocated array */
1761 for (k = 0; k < i; k++) {
1762 if (idx_array[k] == idx) {
1763 already_allocated = true;
1768 if (already_allocated)
1772 /* check the status of the OSP */
1773 tgt = LTD_TGT(ltd, idx);
1777 tgt_dt = tgt->ltd_tgt;
1778 rc = dt_statfs(env, tgt_dt, NULL);
1780 /* this OSP doesn't feel well */
1785 rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
1794 /* Can not allocate more stripes */
1795 if (j == lod->lod_remote_mdt_count) {
1796 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
1797 lod2obd(lod)->obd_name, stripe_count, i - 1);
1801 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
1802 idx, i, PFID(&fid));
1804 /* Set the start index for next stripe allocation */
1805 if (i < stripe_count - 1)
1806 idx_array[i + 1] = (idx + 1) %
1807 (lod->lod_remote_mdt_count + 1);
1808 /* tgt_dt and fid must be ready after search avaible OSP
1809 * in the above loop */
1810 LASSERT(tgt_dt != NULL);
1811 LASSERT(fid_is_sane(&fid));
1812 conf.loc_flags = LOC_F_NEW;
1813 dto = dt_locate_at(env, tgt_dt, &fid,
1814 dt->do_lu.lo_dev->ld_site->ls_top_dev,
1817 GOTO(out_put, rc = PTR_ERR(dto));
1821 lo->ldo_dir_striped = 1;
1822 lo->ldo_stripe = stripe;
1823 lo->ldo_stripenr = i;
1824 lo->ldo_stripes_allocated = stripe_count;
1826 if (lo->ldo_stripenr == 0)
1827 GOTO(out_put, rc = -ENOSPC);
1829 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
1835 for (i = 0; i < stripe_count; i++)
1836 if (stripe[i] != NULL)
1837 lu_object_put(env, &stripe[i]->do_lu);
1838 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_count);
1839 lo->ldo_stripenr = 0;
1840 lo->ldo_stripes_allocated = 0;
1841 lo->ldo_stripe = NULL;
1845 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
1851 * Declare create striped md object.
1853 * The function declares intention to create a striped directory. This is a
1854 * wrapper for lod_prep_md_striped_create(). The only additional functionality
1855 * is to verify pattern \a lum_buf is good. Check that function for the details.
1857 * \param[in] env execution environment
1858 * \param[in] dt object
1859 * \param[in] attr attributes to initialize the objects with
1860 * \param[in] lum_buf a pattern specifying the number of stripes and
1862 * \param[in] dof type of objects to be created
1863 * \param[in] th transaction handle
1865 * \retval 0 on success
1866 * \retval negative if failed
1869 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
1870 struct dt_object *dt,
1871 struct lu_attr *attr,
1872 const struct lu_buf *lum_buf,
1873 struct dt_object_format *dof,
1876 struct lod_object *lo = lod_dt_obj(dt);
1877 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1878 struct lmv_user_md_v1 *lum;
1882 lum = lum_buf->lb_buf;
1883 LASSERT(lum != NULL);
1885 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
1886 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
1887 (int)le32_to_cpu(lum->lum_stripe_offset));
1889 if (le32_to_cpu(lum->lum_stripe_count) == 0)
1892 rc = lod_verify_md_striping(lod, lum);
1896 /* prepare dir striped objects */
1897 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
1899 /* failed to create striping, let's reset
1900 * config so that others don't get confused */
1901 lod_object_free_striping(env, lo);
1909 * Implementation of dt_object_operations::do_declare_xattr_set.
1911 * Used with regular (non-striped) objects. Basically it
1912 * initializes the striping information and applies the
1913 * change to all the stripes.
1915 * \see dt_object_operations::do_declare_xattr_set() in the API description
1918 static int lod_dir_declare_xattr_set(const struct lu_env *env,
1919 struct dt_object *dt,
1920 const struct lu_buf *buf,
1921 const char *name, int fl,
1924 struct dt_object *next = dt_object_child(dt);
1925 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
1926 struct lod_object *lo = lod_dt_obj(dt);
1931 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
1932 struct lmv_user_md_v1 *lum;
1934 LASSERT(buf != NULL && buf->lb_buf != NULL);
1936 rc = lod_verify_md_striping(d, lum);
1941 rc = lod_sub_object_declare_xattr_set(env, next, buf, name, fl, th);
1945 /* Note: Do not set LinkEA on sub-stripes, otherwise
1946 * it will confuse the fid2path process(see mdt_path_current()).
1947 * The linkEA between master and sub-stripes is set in
1948 * lod_xattr_set_lmv(). */
1949 if (strcmp(name, XATTR_NAME_LINK) == 0)
1952 /* set xattr to each stripes, if needed */
1953 rc = lod_load_striping(env, lo);
1957 if (lo->ldo_stripenr == 0)
1960 for (i = 0; i < lo->ldo_stripenr; i++) {
1961 LASSERT(lo->ldo_stripe[i]);
1963 rc = lod_sub_object_declare_xattr_set(env, lo->ldo_stripe[i],
1973 * Reset parent FID on OST object
1975 * Replace parent FID with @dt object FID, which is only called during migration
1976 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
1977 * the FID is changed.
1979 * \param[in] env execution environment
1980 * \param[in] dt dt_object whose stripes's parent FID will be reset
1981 * \parem[in] th thandle
1982 * \param[in] declare if it is declare
1984 * \retval 0 if reset succeeds
1985 * \retval negative errno if reset fais
1987 static int lod_object_replace_parent_fid(const struct lu_env *env,
1988 struct dt_object *dt,
1989 struct thandle *th, bool declare)
1991 struct lod_object *lo = lod_dt_obj(dt);
1992 struct lod_thread_info *info = lod_env_info(env);
1993 struct lu_buf *buf = &info->lti_buf;
1994 struct filter_fid *ff;
1998 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2000 /* set xattr to each stripes, if needed */
2001 rc = lod_load_striping(env, lo);
2005 if (lo->ldo_stripenr == 0)
2008 if (info->lti_ea_store_size < sizeof(*ff)) {
2009 rc = lod_ea_store_resize(info, sizeof(*ff));
2014 buf->lb_buf = info->lti_ea_store;
2015 buf->lb_len = info->lti_ea_store_size;
2017 for (i = 0; i < lo->ldo_stripenr; i++) {
2018 if (lo->ldo_stripe[i] == NULL)
2021 rc = dt_xattr_get(env, lo->ldo_stripe[i], buf,
2029 fid_le_to_cpu(&ff->ff_parent, &ff->ff_parent);
2030 ff->ff_parent.f_seq = lu_object_fid(&dt->do_lu)->f_seq;
2031 ff->ff_parent.f_oid = lu_object_fid(&dt->do_lu)->f_oid;
2032 fid_cpu_to_le(&ff->ff_parent, &ff->ff_parent);
2035 rc = lod_sub_object_declare_xattr_set(env,
2036 lo->ldo_stripe[i], buf,
2038 LU_XATTR_REPLACE, th);
2040 rc = lod_sub_object_xattr_set(env, lo->ldo_stripe[i],
2041 buf, XATTR_NAME_FID,
2042 LU_XATTR_REPLACE, th);
2052 * Implementation of dt_object_operations::do_declare_xattr_set.
2054 * \see dt_object_operations::do_declare_xattr_set() in the API description
2057 * the extension to the API:
2058 * - declaring LOVEA requests striping creation
2059 * - LU_XATTR_REPLACE means layout swap
2061 static int lod_declare_xattr_set(const struct lu_env *env,
2062 struct dt_object *dt,
2063 const struct lu_buf *buf,
2064 const char *name, int fl,
2067 struct dt_object *next = dt_object_child(dt);
2068 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2074 * allow to declare predefined striping on a new (!mode) object
2075 * which is supposed to be replay of regular file creation
2076 * (when LOV setting is declared)
2077 * LU_XATTR_REPLACE is set to indicate a layout swap
2079 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
2080 if ((S_ISREG(mode) || mode == 0) && strcmp(name, XATTR_NAME_LOV) == 0 &&
2081 !(fl & LU_XATTR_REPLACE)) {
2083 * this is a request to manipulate object's striping
2085 if (dt_object_exists(dt)) {
2086 rc = dt_attr_get(env, next, attr);
2090 memset(attr, 0, sizeof(*attr));
2091 attr->la_valid = LA_TYPE | LA_MODE;
2092 attr->la_mode = S_IFREG;
2094 rc = lod_declare_striped_object(env, dt, attr, buf, th);
2095 } else if (S_ISDIR(mode)) {
2096 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
2097 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
2098 rc = lod_object_replace_parent_fid(env, dt, th, true);
2100 rc = lod_sub_object_declare_xattr_set(env, next, buf, name,
2108 * Apply xattr changes to the object.
2110 * Applies xattr changes to the object and the stripes if the latter exist.
2112 * \param[in] env execution environment
2113 * \param[in] dt object
2114 * \param[in] buf buffer pointing to the new value of xattr
2115 * \param[in] name name of xattr
2116 * \param[in] fl flags
2117 * \param[in] th transaction handle
2119 * \retval 0 on success
2120 * \retval negative if failed
2122 static int lod_xattr_set_internal(const struct lu_env *env,
2123 struct dt_object *dt,
2124 const struct lu_buf *buf,
2125 const char *name, int fl,
2128 struct dt_object *next = dt_object_child(dt);
2129 struct lod_object *lo = lod_dt_obj(dt);
2134 rc = lod_sub_object_xattr_set(env, next, buf, name, fl, th);
2135 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2138 /* Note: Do not set LinkEA on sub-stripes, otherwise
2139 * it will confuse the fid2path process(see mdt_path_current()).
2140 * The linkEA between master and sub-stripes is set in
2141 * lod_xattr_set_lmv(). */
2142 if (lo->ldo_stripenr == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
2145 for (i = 0; i < lo->ldo_stripenr; i++) {
2146 LASSERT(lo->ldo_stripe[i]);
2148 rc = lod_sub_object_xattr_set(env, lo->ldo_stripe[i], buf, name,
2158 * Delete an extended attribute.
2160 * Deletes specified xattr from the object and the stripes if the latter exist.
2162 * \param[in] env execution environment
2163 * \param[in] dt object
2164 * \param[in] name name of xattr
2165 * \param[in] th transaction handle
2167 * \retval 0 on success
2168 * \retval negative if failed
2170 static int lod_xattr_del_internal(const struct lu_env *env,
2171 struct dt_object *dt,
2172 const char *name, struct thandle *th)
2174 struct dt_object *next = dt_object_child(dt);
2175 struct lod_object *lo = lod_dt_obj(dt);
2180 rc = lod_sub_object_xattr_del(env, next, name, th);
2181 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2184 if (lo->ldo_stripenr == 0)
2187 for (i = 0; i < lo->ldo_stripenr; i++) {
2188 LASSERT(lo->ldo_stripe[i]);
2190 rc = lod_sub_object_xattr_del(env, lo->ldo_stripe[i], name,
2200 * Set default striping on a directory.
2202 * Sets specified striping on a directory object unless it matches the default
2203 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
2204 * EA. This striping will be used when regular file is being created in this
2207 * \param[in] env execution environment
2208 * \param[in] dt the striped object
2209 * \param[in] buf buffer with the striping
2210 * \param[in] name name of EA
2211 * \param[in] fl xattr flag (see OSD API description)
2212 * \param[in] th transaction handle
2214 * \retval 0 on success
2215 * \retval negative if failed
2217 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
2218 struct dt_object *dt,
2219 const struct lu_buf *buf,
2220 const char *name, int fl,
2223 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2224 struct lov_user_md_v1 *lum;
2225 struct lov_user_md_v3 *v3 = NULL;
2226 const char *pool_name = NULL;
2230 LASSERT(buf != NULL && buf->lb_buf != NULL);
2233 rc = lod_verify_striping(d, buf, false);
2237 if (lum->lmm_magic == LOV_USER_MAGIC_V3) {
2239 if (v3->lmm_pool_name[0] != '\0')
2240 pool_name = v3->lmm_pool_name;
2243 /* if { size, offset, count } = { 0, -1, 0 } and no pool
2244 * (i.e. all default values specified) then delete default
2245 * striping from dir. */
2247 "set default striping: sz %u # %u offset %d %s %s\n",
2248 (unsigned)lum->lmm_stripe_size,
2249 (unsigned)lum->lmm_stripe_count,
2250 (int)lum->lmm_stripe_offset,
2251 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
2253 if (LOVEA_DELETE_VALUES(lum->lmm_stripe_size, lum->lmm_stripe_count,
2254 lum->lmm_stripe_offset, pool_name)) {
2255 rc = lod_xattr_del_internal(env, dt, name, th);
2259 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
2266 * Set default striping on a directory object.
2268 * Sets specified striping on a directory object unless it matches the default
2269 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
2270 * EA. This striping will be used when a new directory is being created in the
2273 * \param[in] env execution environment
2274 * \param[in] dt the striped object
2275 * \param[in] buf buffer with the striping
2276 * \param[in] name name of EA
2277 * \param[in] fl xattr flag (see OSD API description)
2278 * \param[in] th transaction handle
2280 * \retval 0 on success
2281 * \retval negative if failed
2283 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
2284 struct dt_object *dt,
2285 const struct lu_buf *buf,
2286 const char *name, int fl,
2289 struct lmv_user_md_v1 *lum;
2293 LASSERT(buf != NULL && buf->lb_buf != NULL);
2296 CDEBUG(D_OTHER, "set default stripe_count # %u stripe_offset %d\n",
2297 le32_to_cpu(lum->lum_stripe_count),
2298 (int)le32_to_cpu(lum->lum_stripe_offset));
2300 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
2301 le32_to_cpu(lum->lum_stripe_offset)) &&
2302 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
2303 rc = lod_xattr_del_internal(env, dt, name, th);
2307 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
2316 * Turn directory into a striped directory.
2318 * During replay the client sends the striping created before MDT
2319 * failure, then the layer above LOD sends this defined striping
2320 * using ->do_xattr_set(), so LOD uses this method to replay creation
2321 * of the stripes. Notice the original information for the striping
2322 * (#stripes, FIDs, etc) was transferred in declare path.
2324 * \param[in] env execution environment
2325 * \param[in] dt the striped object
2326 * \param[in] buf not used currently
2327 * \param[in] name not used currently
2328 * \param[in] fl xattr flag (see OSD API description)
2329 * \param[in] th transaction handle
2331 * \retval 0 on success
2332 * \retval negative if failed
2334 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
2335 const struct lu_buf *buf, const char *name,
2336 int fl, struct thandle *th)
2338 struct lod_object *lo = lod_dt_obj(dt);
2339 struct lod_thread_info *info = lod_env_info(env);
2340 struct lu_attr *attr = &info->lti_attr;
2341 struct dt_object_format *dof = &info->lti_format;
2342 struct lu_buf lmv_buf;
2343 struct lu_buf slave_lmv_buf;
2344 struct lmv_mds_md_v1 *lmm;
2345 struct lmv_mds_md_v1 *slave_lmm = NULL;
2346 struct dt_insert_rec *rec = &info->lti_dt_rec;
2351 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
2354 /* The stripes are supposed to be allocated in declare phase,
2355 * if there are no stripes being allocated, it will skip */
2356 if (lo->ldo_stripenr == 0)
2359 rc = dt_attr_get(env, dt_object_child(dt), attr);
2363 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME |
2364 LA_MODE | LA_UID | LA_GID | LA_TYPE;
2365 dof->dof_type = DFT_DIR;
2367 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
2370 lmm = lmv_buf.lb_buf;
2372 OBD_ALLOC_PTR(slave_lmm);
2373 if (slave_lmm == NULL)
2376 lod_prep_slave_lmv_md(slave_lmm, lmm);
2377 slave_lmv_buf.lb_buf = slave_lmm;
2378 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
2380 rec->rec_type = S_IFDIR;
2381 for (i = 0; i < lo->ldo_stripenr; i++) {
2382 struct dt_object *dto;
2383 char *stripe_name = info->lti_key;
2384 struct lu_name *sname;
2385 struct linkea_data ldata = { NULL };
2386 struct lu_buf linkea_buf;
2388 dto = lo->ldo_stripe[i];
2390 dt_write_lock(env, dto, MOR_TGT_CHILD);
2391 rc = lod_sub_object_create(env, dto, attr, NULL, dof,
2394 dt_write_unlock(env, dto);
2398 rc = lod_sub_object_ref_add(env, dto, th);
2399 dt_write_unlock(env, dto);
2403 rec->rec_fid = lu_object_fid(&dto->do_lu);
2404 rc = lod_sub_object_index_insert(env, dto,
2405 (const struct dt_rec *)rec,
2406 (const struct dt_key *)dot, th, 0);
2410 rec->rec_fid = lu_object_fid(&dt->do_lu);
2411 rc = lod_sub_object_index_insert(env, dto, (struct dt_rec *)rec,
2412 (const struct dt_key *)dotdot, th, 0);
2416 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
2417 cfs_fail_val != i) {
2418 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
2420 slave_lmm->lmv_master_mdt_index =
2423 slave_lmm->lmv_master_mdt_index =
2426 rc = lod_sub_object_xattr_set(env, dto, &slave_lmv_buf,
2427 XATTR_NAME_LMV, fl, th);
2432 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
2434 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2435 PFID(lu_object_fid(&dto->do_lu)), i + 1);
2437 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2438 PFID(lu_object_fid(&dto->do_lu)), i);
2440 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
2441 rc = linkea_data_new(&ldata, &info->lti_linkea_buf);
2445 rc = linkea_add_buf(&ldata, sname, lu_object_fid(&dt->do_lu));
2449 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
2450 linkea_buf.lb_len = ldata.ld_leh->leh_len;
2451 rc = lod_sub_object_xattr_set(env, dto, &linkea_buf,
2452 XATTR_NAME_LINK, 0, th);
2456 rec->rec_fid = lu_object_fid(&dto->do_lu);
2457 rc = lod_sub_object_index_insert(env, dt_object_child(dt),
2458 (const struct dt_rec *)rec,
2459 (const struct dt_key *)stripe_name, th, 0);
2463 rc = lod_sub_object_ref_add(env, dt_object_child(dt), th);
2468 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
2469 rc = lod_sub_object_xattr_set(env, dt_object_child(dt),
2470 &lmv_buf, XATTR_NAME_LMV, fl, th);
2472 if (slave_lmm != NULL)
2473 OBD_FREE_PTR(slave_lmm);
2479 * Helper function to declare/execute creation of a striped directory
2481 * Called in declare/create object path, prepare striping for a directory
2482 * and prepare defaults data striping for the objects to be created in
2483 * that directory. Notice the function calls "declaration" or "execution"
2484 * methods depending on \a declare param. This is a consequence of the
2485 * current approach while we don't have natural distributed transactions:
2486 * we basically execute non-local updates in the declare phase. So, the
2487 * arguments for the both phases are the same and this is the reason for
2488 * this function to exist.
2490 * \param[in] env execution environment
2491 * \param[in] dt object
2492 * \param[in] attr attributes the stripes will be created with
2493 * \param[in] dof format of stripes (see OSD API description)
2494 * \param[in] th transaction handle
2495 * \param[in] declare where to call "declare" or "execute" methods
2497 * \retval 0 on success
2498 * \retval negative if failed
2500 static int lod_dir_striping_create_internal(const struct lu_env *env,
2501 struct dt_object *dt,
2502 struct lu_attr *attr,
2503 struct dt_object_format *dof,
2507 struct lod_thread_info *info = lod_env_info(env);
2508 struct lod_object *lo = lod_dt_obj(dt);
2509 const struct lod_default_striping *lds = lo->ldo_def_striping;
2510 const char *poolname = NULL;
2514 LASSERT(ergo(lds != NULL,
2515 lds->lds_def_striping_set ||
2516 lds->lds_dir_def_striping_set));
2518 if (!LMVEA_DELETE_VALUES(lo->ldo_stripenr,
2519 lo->ldo_dir_stripe_offset)) {
2520 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
2521 int stripe_count = lo->ldo_stripenr;
2523 if (info->lti_ea_store_size < sizeof(*v1)) {
2524 rc = lod_ea_store_resize(info, sizeof(*v1));
2527 v1 = info->lti_ea_store;
2530 memset(v1, 0, sizeof(*v1));
2531 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
2532 v1->lum_stripe_count = cpu_to_le32(stripe_count);
2533 v1->lum_stripe_offset =
2534 cpu_to_le32(lo->ldo_dir_stripe_offset);
2536 info->lti_buf.lb_buf = v1;
2537 info->lti_buf.lb_len = sizeof(*v1);
2540 rc = lod_declare_xattr_set_lmv(env, dt, attr,
2541 &info->lti_buf, dof, th);
2543 rc = lod_xattr_set_lmv(env, dt, &info->lti_buf,
2544 XATTR_NAME_LMV, 0, th);
2549 /* Transfer default LMV striping from the parent */
2550 if (lds != NULL && lds->lds_dir_def_striping_set &&
2551 !LMVEA_DELETE_VALUES(lds->lds_dir_def_stripenr,
2552 lds->lds_dir_def_stripe_offset)) {
2553 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
2555 if (info->lti_ea_store_size < sizeof(*v1)) {
2556 rc = lod_ea_store_resize(info, sizeof(*v1));
2559 v1 = info->lti_ea_store;
2562 memset(v1, 0, sizeof(*v1));
2563 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
2564 v1->lum_stripe_count = cpu_to_le32(lds->lds_dir_def_stripenr);
2565 v1->lum_stripe_offset =
2566 cpu_to_le32(lds->lds_dir_def_stripe_offset);
2568 cpu_to_le32(lds->lds_dir_def_hash_type);
2570 info->lti_buf.lb_buf = v1;
2571 info->lti_buf.lb_len = sizeof(*v1);
2573 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
2574 XATTR_NAME_DEFAULT_LMV,
2577 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
2579 XATTR_NAME_DEFAULT_LMV, 0,
2585 if (lds != NULL && lds->lds_def_pool[0] != '\0')
2586 poolname = lds->lds_def_pool;
2588 /* Transfer default LOV striping from the parent */
2589 if (lds != NULL && lds->lds_def_striping_set &&
2590 !LOVEA_DELETE_VALUES(lds->lds_def_stripe_size,
2591 lds->lds_def_stripenr,
2592 lds->lds_def_stripe_offset,
2594 struct lov_user_md_v3 *v3 = info->lti_ea_store;
2596 if (info->lti_ea_store_size < sizeof(*v3)) {
2597 rc = lod_ea_store_resize(info, sizeof(*v3));
2600 v3 = info->lti_ea_store;
2603 memset(v3, 0, sizeof(*v3));
2604 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
2605 v3->lmm_stripe_count = cpu_to_le16(lds->lds_def_stripenr);
2606 v3->lmm_stripe_offset = cpu_to_le16(lds->lds_def_stripe_offset);
2607 v3->lmm_stripe_size = cpu_to_le32(lds->lds_def_stripe_size);
2608 if (poolname != NULL)
2609 strlcpy(v3->lmm_pool_name, poolname,
2610 sizeof(v3->lmm_pool_name));
2612 info->lti_buf.lb_buf = v3;
2613 info->lti_buf.lb_len = sizeof(*v3);
2616 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
2617 XATTR_NAME_LOV, 0, th);
2619 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
2620 XATTR_NAME_LOV, 0, th);
2628 static int lod_declare_dir_striping_create(const struct lu_env *env,
2629 struct dt_object *dt,
2630 struct lu_attr *attr,
2631 struct dt_object_format *dof,
2634 return lod_dir_striping_create_internal(env, dt, attr, dof, th, true);
2637 static int lod_dir_striping_create(const struct lu_env *env,
2638 struct dt_object *dt,
2639 struct lu_attr *attr,
2640 struct dt_object_format *dof,
2643 return lod_dir_striping_create_internal(env, dt, attr, dof, th, false);
2647 * Implementation of dt_object_operations::do_xattr_set.
2649 * Sets specified extended attribute on the object. Three types of EAs are
2651 * LOV EA - stores striping for a regular file or default striping (when set
2653 * LMV EA - stores a marker for the striped directories
2654 * DMV EA - stores default directory striping
2656 * When striping is applied to a non-striped existing object (this is called
2657 * late striping), then LOD notices the caller wants to turn the object into a
2658 * striped one. The stripe objects are created and appropriate EA is set:
2659 * LOV EA storing all the stripes directly or LMV EA storing just a small header
2660 * with striping configuration.
2662 * \see dt_object_operations::do_xattr_set() in the API description for details.
2664 static int lod_xattr_set(const struct lu_env *env,
2665 struct dt_object *dt, const struct lu_buf *buf,
2666 const char *name, int fl, struct thandle *th)
2668 struct dt_object *next = dt_object_child(dt);
2672 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
2673 strcmp(name, XATTR_NAME_LMV) == 0) {
2674 struct lmv_mds_md_v1 *lmm = buf->lb_buf;
2676 if (lmm != NULL && le32_to_cpu(lmm->lmv_hash_type) &
2677 LMV_HASH_FLAG_MIGRATION)
2678 rc = lod_sub_object_xattr_set(env, next, buf, name, fl,
2681 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
2686 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
2687 strcmp(name, XATTR_NAME_LOV) == 0) {
2689 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl, th);
2691 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
2692 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2694 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
2697 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
2698 !strcmp(name, XATTR_NAME_LOV)) {
2699 /* in case of lov EA swap, just set it
2700 * if not, it is a replay so check striping match what we
2701 * already have during req replay, declare_xattr_set()
2702 * defines striping, then create() does the work */
2703 if (fl & LU_XATTR_REPLACE) {
2704 /* free stripes, then update disk */
2705 lod_object_free_striping(env, lod_dt_obj(dt));
2707 rc = lod_sub_object_xattr_set(env, next, buf, name,
2709 } else if (dt_object_remote(dt)) {
2710 /* This only happens during migration, see
2711 * mdd_migrate_create(), in which Master MDT will
2712 * create a remote target object, and only set
2713 * (migrating) stripe EA on the remote object,
2714 * and does not need creating each stripes. */
2715 rc = lod_sub_object_xattr_set(env, next, buf, name,
2718 rc = lod_striping_create(env, dt, NULL, NULL, th);
2721 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
2722 rc = lod_object_replace_parent_fid(env, dt, th, false);
2727 /* then all other xattr */
2728 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
2734 * Implementation of dt_object_operations::do_declare_xattr_del.
2736 * \see dt_object_operations::do_declare_xattr_del() in the API description
2739 static int lod_declare_xattr_del(const struct lu_env *env,
2740 struct dt_object *dt, const char *name,
2743 struct lod_object *lo = lod_dt_obj(dt);
2748 rc = lod_sub_object_declare_xattr_del(env, dt_object_child(dt),
2753 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
2756 /* set xattr to each stripes, if needed */
2757 rc = lod_load_striping(env, lo);
2761 if (lo->ldo_stripenr == 0)
2764 for (i = 0; i < lo->ldo_stripenr; i++) {
2765 LASSERT(lo->ldo_stripe[i]);
2766 rc = lod_sub_object_declare_xattr_del(env, lo->ldo_stripe[i],
2776 * Implementation of dt_object_operations::do_xattr_del.
2778 * If EA storing a regular striping is being deleted, then release
2779 * all the references to the stripe objects in core.
2781 * \see dt_object_operations::do_xattr_del() in the API description for details.
2783 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
2784 const char *name, struct thandle *th)
2786 struct dt_object *next = dt_object_child(dt);
2787 struct lod_object *lo = lod_dt_obj(dt);
2792 if (!strcmp(name, XATTR_NAME_LOV))
2793 lod_object_free_striping(env, lod_dt_obj(dt));
2795 rc = lod_sub_object_xattr_del(env, next, name, th);
2796 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2799 if (lo->ldo_stripenr == 0)
2802 for (i = 0; i < lo->ldo_stripenr; i++) {
2803 LASSERT(lo->ldo_stripe[i]);
2805 rc = lod_sub_object_xattr_del(env, lo->ldo_stripe[i], name, th);
2814 * Implementation of dt_object_operations::do_xattr_list.
2816 * \see dt_object_operations::do_xattr_list() in the API description
2819 static int lod_xattr_list(const struct lu_env *env,
2820 struct dt_object *dt, const struct lu_buf *buf)
2822 return dt_xattr_list(env, dt_object_child(dt), buf);
2825 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
2827 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
2832 * Get default striping.
2834 * \param[in] env execution environment
2835 * \param[in] lo object
2836 * \param[out] lds default striping
2838 * \retval 0 on success
2839 * \retval negative if failed
2841 static int lod_get_default_lov_striping(const struct lu_env *env,
2842 struct lod_object *lo,
2843 struct lod_default_striping *lds)
2845 struct lod_thread_info *info = lod_env_info(env);
2846 struct lov_user_md_v1 *v1 = NULL;
2847 struct lov_user_md_v3 *v3 = NULL;
2850 rc = lod_get_lov_ea(env, lo);
2854 if (rc < (typeof(rc))sizeof(struct lov_user_md))
2857 v1 = info->lti_ea_store;
2858 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
2859 lustre_swab_lov_user_md_v1(v1);
2860 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
2861 v3 = (struct lov_user_md_v3 *)v1;
2862 lustre_swab_lov_user_md_v3(v3);
2865 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1)
2868 if (v1->lmm_pattern != LOV_PATTERN_RAID0 && v1->lmm_pattern != 0)
2871 lds->lds_def_stripenr = v1->lmm_stripe_count;
2872 lds->lds_def_stripe_size = v1->lmm_stripe_size;
2873 lds->lds_def_stripe_offset = v1->lmm_stripe_offset;
2874 lds->lds_def_striping_set = 1;
2875 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2876 v3 = (struct lov_user_md_v3 *)v1;
2877 if (v3->lmm_pool_name[0] != '\0')
2878 strlcpy(lds->lds_def_pool, v3->lmm_pool_name,
2879 sizeof(lds->lds_def_pool));
2886 * Get default directory striping.
2888 * \param[in] env execution environment
2889 * \param[in] lo object
2890 * \param[out] lds default striping
2892 * \retval 0 on success
2893 * \retval negative if failed
2895 static int lod_get_default_lmv_striping(const struct lu_env *env,
2896 struct lod_object *lo,
2897 struct lod_default_striping *lds)
2899 struct lod_thread_info *info = lod_env_info(env);
2900 struct lmv_user_md_v1 *v1 = NULL;
2903 rc = lod_get_default_lmv_ea(env, lo);
2907 if (rc < (typeof(rc))sizeof(struct lmv_user_md))
2910 v1 = info->lti_ea_store;
2912 lds->lds_dir_def_stripenr = le32_to_cpu(v1->lum_stripe_count);
2913 lds->lds_dir_def_stripe_offset = le32_to_cpu(v1->lum_stripe_offset);
2914 lds->lds_dir_def_hash_type = le32_to_cpu(v1->lum_hash_type);
2915 lds->lds_dir_def_striping_set = 1;
2921 * Get default striping in the object.
2923 * Get object default striping and default directory striping.
2925 * \param[in] env execution environment
2926 * \param[in] lo object
2927 * \param[out] lds default striping
2929 * \retval 0 on success
2930 * \retval negative if failed
2932 static int lod_get_default_striping(const struct lu_env *env,
2933 struct lod_object *lo,
2934 struct lod_default_striping *lds)
2938 rc = lod_get_default_lov_striping(env, lo, lds);
2940 rc = lod_get_default_lmv_striping(env, lo, lds);
2945 * Apply default striping on object.
2947 * If object striping pattern is not set, set to the one in default striping.
2948 * The default striping is from parent or fs.
2950 * \param[in] lo new object
2951 * \param[in] lds default striping
2952 * \param[in] mode new object's mode
2954 static void lod_striping_from_default(struct lod_object *lo,
2955 const struct lod_default_striping *lds,
2958 if (lds->lds_def_striping_set && S_ISREG(mode)) {
2959 if (lo->ldo_stripenr == 0)
2960 lo->ldo_stripenr = lds->lds_def_stripenr;
2961 if (lo->ldo_stripe_size == 0)
2962 lo->ldo_stripe_size = lds->lds_def_stripe_size;
2963 if (lo->ldo_stripe_offset == LOV_OFFSET_DEFAULT)
2964 lo->ldo_stripe_offset = lds->lds_def_stripe_offset;
2965 if (lo->ldo_pool == NULL && lds->lds_def_pool[0] != '\0')
2966 lod_object_set_pool(lo, lds->lds_def_pool);
2968 CDEBUG(D_INFO, "striping from default: count %hu, size %u, "
2969 "offset %d, pool %s\n",
2970 lo->ldo_stripenr, lo->ldo_stripe_size,
2971 (int)lo->ldo_stripe_offset, lo->ldo_pool ?: "");
2972 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
2973 if (lo->ldo_stripenr == 0)
2974 lo->ldo_stripenr = lds->lds_dir_def_stripenr;
2975 if (lo->ldo_dir_stripe_offset == -1)
2976 lo->ldo_dir_stripe_offset =
2977 lds->lds_dir_def_stripe_offset;
2978 if (lo->ldo_dir_hash_type == 0)
2979 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
2981 CDEBUG(D_INFO, "striping from default: count %hu, offset %d, "
2983 lo->ldo_stripenr, (int)lo->ldo_dir_stripe_offset,
2984 lo->ldo_dir_hash_type);
2989 * Implementation of dt_object_operations::do_ah_init.
2991 * This method is used to make a decision on the striping configuration for the
2992 * object being created. It can be taken from the \a parent object if it exists,
2993 * or filesystem's default. The resulting configuration (number of stripes,
2994 * stripe size/offset, pool name, etc) is stored in the object itself and will
2995 * be used by the methods like ->doo_declare_create().
2997 * \see dt_object_operations::do_ah_init() in the API description for details.
2999 static void lod_ah_init(const struct lu_env *env,
3000 struct dt_allocation_hint *ah,
3001 struct dt_object *parent,
3002 struct dt_object *child,
3005 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
3006 struct lod_thread_info *info = lod_env_info(env);
3007 struct lod_default_striping *lds = &info->lti_def_striping;
3008 struct dt_object *nextp = NULL;
3009 struct dt_object *nextc;
3010 struct lod_object *lp = NULL;
3011 struct lod_object *lc;
3012 struct lov_desc *desc;
3017 if (likely(parent)) {
3018 nextp = dt_object_child(parent);
3019 lp = lod_dt_obj(parent);
3022 nextc = dt_object_child(child);
3023 lc = lod_dt_obj(child);
3025 LASSERT(lc->ldo_stripenr == 0);
3026 LASSERT(lc->ldo_stripe == NULL);
3028 if (!dt_object_exists(nextc))
3029 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
3031 if (S_ISDIR(child_mode)) {
3032 /* other default values are 0 */
3033 lc->ldo_dir_stripe_offset = -1;
3035 memset(lds, 0, sizeof(*lds));
3036 lod_get_default_striping(env, lp, lds);
3038 /* inherit parent default striping */
3039 if (lds->lds_def_striping_set || lds->lds_dir_def_striping_set)
3040 lc->ldo_def_striping = lds;
3042 /* It should always honour the specified stripes */
3043 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
3044 lod_verify_md_striping(d, ah->dah_eadata) == 0) {
3045 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
3047 lc->ldo_stripenr = le32_to_cpu(lum1->lum_stripe_count);
3048 lc->ldo_dir_stripe_offset =
3049 le32_to_cpu(lum1->lum_stripe_offset);
3050 lc->ldo_dir_hash_type =
3051 le32_to_cpu(lum1->lum_hash_type);
3052 CDEBUG(D_INFO, "set dir stripe: count %hu, offset %d, "
3055 (int)lc->ldo_dir_stripe_offset,
3056 lc->ldo_dir_hash_type);
3058 lod_striping_from_default(lc, lds, child_mode);
3061 /* shrink the stripe_count to the avaible MDT count */
3062 if (lc->ldo_stripenr > d->lod_remote_mdt_count + 1 &&
3063 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))
3064 lc->ldo_stripenr = d->lod_remote_mdt_count + 1;
3066 /* Directory will be striped only if stripe_count > 1, if
3067 * stripe_count == 1, let's reset stripenr = 0 to avoid
3068 * create single master stripe and also help to unify the
3069 * stripe handling of directories and files */
3070 if (lc->ldo_stripenr == 1)
3071 lc->ldo_stripenr = 0;
3073 CDEBUG(D_INFO, "final dir stripe [%hu %d %u]\n",
3074 lc->ldo_stripenr, (int)lc->ldo_dir_stripe_offset,
3075 lc->ldo_dir_hash_type);
3081 * if object is going to be striped over OSTs, transfer default
3082 * striping information to the child, so that we can use it
3083 * during declaration and creation
3085 if (!lod_object_will_be_striped(S_ISREG(child_mode),
3086 lu_object_fid(&child->do_lu)))
3089 /* other default values are 0 */
3090 lc->ldo_stripe_offset = LOV_OFFSET_DEFAULT;
3092 /* striping from parent default */
3093 if (likely(parent)) {
3094 memset(lds, 0, sizeof(*lds));
3095 lod_get_default_lov_striping(env, lp, lds);
3096 lod_striping_from_default(lc, lds, child_mode);
3099 if (d->lod_md_root == NULL) {
3100 struct dt_object *root;
3101 struct lod_object *lroot;
3103 lu_root_fid(&info->lti_fid);
3104 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
3105 if (!IS_ERR(root)) {
3106 lroot = lod_dt_obj(root);
3108 spin_lock(&d->lod_lock);
3109 if (d->lod_md_root != NULL)
3111 &d->lod_md_root->ldo_obj.do_lu);
3112 d->lod_md_root = lroot;
3113 spin_unlock(&d->lod_lock);
3117 /* if parent doesn't provide all defaults, striping from fs default */
3118 if (d->lod_md_root != NULL &&
3119 (lc->ldo_stripenr == 0 ||
3120 lc->ldo_stripe_size == 0 ||
3121 lc->ldo_stripe_offset == LOV_OFFSET_DEFAULT ||
3122 lc->ldo_pool == NULL)) {
3123 memset(lds, 0, sizeof(*lds));
3124 lod_get_default_lov_striping(env, d->lod_md_root, lds);
3125 lod_striping_from_default(lc, lds, child_mode);
3129 * fs default striping may not be explicitly set, or historically set
3130 * in config log, check striping sanity here and fix to sane values.
3132 desc = &d->lod_desc;
3133 if (lc->ldo_stripenr == 0)
3134 lc->ldo_stripenr = desc->ld_default_stripe_count;
3135 if (lc->ldo_stripe_size == 0)
3136 lc->ldo_stripe_size = desc->ld_default_stripe_size;
3138 CDEBUG(D_INFO, "final striping [%hu %u %d %s]\n",
3139 lc->ldo_stripenr, lc->ldo_stripe_size,
3140 (int)lc->ldo_stripe_offset, lc->ldo_pool ?: "");
3144 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
3146 * Size initialization on late striping.
3148 * Propagate the size of a truncated object to a deferred striping.
3149 * This function handles a special case when truncate was done on a
3150 * non-striped object and now while the striping is being created
3151 * we can't lose that size, so we have to propagate it to the stripes
3154 * \param[in] env execution environment
3155 * \param[in] dt object
3156 * \param[in] th transaction handle
3158 * \retval 0 on success
3159 * \retval negative if failed
3161 static int lod_declare_init_size(const struct lu_env *env,
3162 struct dt_object *dt, struct thandle *th)
3164 struct dt_object *next = dt_object_child(dt);
3165 struct lod_object *lo = lod_dt_obj(dt);
3166 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3167 uint64_t size, offs;
3171 /* XXX: we support the simplest (RAID0) striping so far */
3172 LASSERT(lo->ldo_stripe || lo->ldo_stripenr == 0);
3173 LASSERT(lo->ldo_stripe_size > 0);
3175 if (lo->ldo_stripenr == 0)
3178 rc = dt_attr_get(env, next, attr);
3179 LASSERT(attr->la_valid & LA_SIZE);
3183 size = attr->la_size;
3187 /* ll_do_div64(a, b) returns a % b, and a = a / b */
3188 ll_do_div64(size, (__u64) lo->ldo_stripe_size);
3189 stripe = ll_do_div64(size, (__u64) lo->ldo_stripenr);
3191 size = size * lo->ldo_stripe_size;
3192 offs = attr->la_size;
3193 size += ll_do_div64(offs, lo->ldo_stripe_size);
3195 attr->la_valid = LA_SIZE;
3196 attr->la_size = size;
3198 rc = lod_sub_object_declare_attr_set(env, lo->ldo_stripe[stripe], attr,
3205 * Declare creation of striped object.
3207 * The function declares creation stripes for a regular object. The function
3208 * also declares whether the stripes will be created with non-zero size if
3209 * previously size was set non-zero on the master object. If object \a dt is
3210 * not local, then only fully defined striping can be applied in \a lovea.
3211 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
3214 * \param[in] env execution environment
3215 * \param[in] dt object
3216 * \param[in] attr attributes the stripes will be created with
3217 * \param[in] lovea a buffer containing striping description
3218 * \param[in] th transaction handle
3220 * \retval 0 on success
3221 * \retval negative if failed
3223 int lod_declare_striped_object(const struct lu_env *env, struct dt_object *dt,
3224 struct lu_attr *attr,
3225 const struct lu_buf *lovea, struct thandle *th)
3227 struct lod_thread_info *info = lod_env_info(env);
3228 struct dt_object *next = dt_object_child(dt);
3229 struct lod_object *lo = lod_dt_obj(dt);
3233 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
3234 GOTO(out, rc = -ENOMEM);
3236 if (!dt_object_remote(next)) {
3237 /* choose OST and generate appropriate objects */
3238 rc = lod_qos_prep_create(env, lo, attr, lovea, th);
3243 * declare storage for striping data
3245 info->lti_buf.lb_len = lov_mds_md_size(lo->ldo_stripenr,
3246 lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1);
3248 /* LOD can not choose OST objects for remote objects, i.e.
3249 * stripes must be ready before that. Right now, it can only
3250 * happen during migrate, i.e. migrate process needs to create
3251 * remote regular file (mdd_migrate_create), then the migrate
3252 * process will provide stripeEA. */
3253 LASSERT(lovea != NULL);
3254 info->lti_buf = *lovea;
3257 rc = lod_sub_object_declare_xattr_set(env, next, &info->lti_buf,
3258 XATTR_NAME_LOV, 0, th);
3263 * if striping is created with local object's size > 0,
3264 * we have to propagate this size to specific object
3265 * the case is possible only when local object was created previously
3267 if (dt_object_exists(next))
3268 rc = lod_declare_init_size(env, dt, th);
3271 /* failed to create striping or to set initial size, let's reset
3272 * config so that others don't get confused */
3274 lod_object_free_striping(env, lo);
3280 * Implementation of dt_object_operations::do_declare_create.
3282 * The method declares creation of a new object. If the object will be striped,
3283 * then helper functions are called to find FIDs for the stripes, declare
3284 * creation of the stripes and declare initialization of the striping
3285 * information to be stored in the master object.
3287 * \see dt_object_operations::do_declare_create() in the API description
3290 static int lod_declare_object_create(const struct lu_env *env,
3291 struct dt_object *dt,
3292 struct lu_attr *attr,
3293 struct dt_allocation_hint *hint,
3294 struct dt_object_format *dof,
3297 struct dt_object *next = dt_object_child(dt);
3298 struct lod_object *lo = lod_dt_obj(dt);
3307 * first of all, we declare creation of local object
3309 rc = lod_sub_object_declare_create(env, next, attr, hint, dof, th);
3313 if (dof->dof_type == DFT_SYM)
3314 dt->do_body_ops = &lod_body_lnk_ops;
3315 else if (dof->dof_type == DFT_REGULAR)
3316 dt->do_body_ops = &lod_body_ops;
3319 * it's lod_ah_init() that has decided the object will be striped
3321 if (dof->dof_type == DFT_REGULAR) {
3322 /* callers don't want stripes */
3323 /* XXX: all tricky interactions with ->ah_make_hint() decided
3324 * to use striping, then ->declare_create() behaving differently
3325 * should be cleaned */
3326 if (dof->u.dof_reg.striped == 0)
3327 lo->ldo_stripenr = 0;
3328 if (lo->ldo_stripenr > 0)
3329 rc = lod_declare_striped_object(env, dt, attr,
3331 } else if (dof->dof_type == DFT_DIR) {
3332 struct seq_server_site *ss;
3334 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
3336 /* If the parent has default stripeEA, and client
3337 * did not find it before sending create request,
3338 * then MDT will return -EREMOTE, and client will
3339 * retrieve the default stripeEA and re-create the
3342 * Note: if dah_eadata != NULL, it means creating the
3343 * striped directory with specified stripeEA, then it
3344 * should ignore the default stripeEA */
3345 if (hint != NULL && hint->dah_eadata == NULL) {
3346 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
3347 GOTO(out, rc = -EREMOTE);
3349 if (lo->ldo_dir_stripe_offset == -1) {
3350 /* child and parent should be in the same MDT */
3351 if (hint->dah_parent != NULL &&
3352 dt_object_remote(hint->dah_parent))
3353 GOTO(out, rc = -EREMOTE);
3354 } else if (lo->ldo_dir_stripe_offset !=
3356 struct lod_device *lod;
3357 struct lod_tgt_descs *ltd;
3358 struct lod_tgt_desc *tgt = NULL;
3359 bool found_mdt = false;
3362 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
3363 ltd = &lod->lod_mdt_descs;
3364 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
3365 tgt = LTD_TGT(ltd, i);
3366 if (tgt->ltd_index ==
3367 lo->ldo_dir_stripe_offset) {
3373 /* If the MDT indicated by stripe_offset can be
3374 * found, then tell client to resend the create
3375 * request to the correct MDT, otherwise return
3376 * error to client */
3378 GOTO(out, rc = -EREMOTE);
3380 GOTO(out, rc = -EINVAL);
3384 rc = lod_declare_dir_striping_create(env, dt, attr, dof, th);
3391 * Creation of a striped regular object.
3393 * The function is called to create the stripe objects for a regular
3394 * striped file. This can happen at the initial object creation or
3395 * when the caller asks LOD to do so using ->do_xattr_set() method
3396 * (so called late striping). Notice all the information are already
3397 * prepared in the form of the list of objects (ldo_stripe field).
3398 * This is done during declare phase.
3400 * \param[in] env execution environment
3401 * \param[in] dt object
3402 * \param[in] attr attributes the stripes will be created with
3403 * \param[in] dof format of stripes (see OSD API description)
3404 * \param[in] th transaction handle
3406 * \retval 0 on success
3407 * \retval negative if failed
3409 int lod_striping_create(const struct lu_env *env, struct dt_object *dt,
3410 struct lu_attr *attr, struct dt_object_format *dof,
3413 struct lod_object *lo = lod_dt_obj(dt);
3417 /* create all underlying objects */
3418 for (i = 0; i < lo->ldo_stripenr; i++) {
3419 LASSERT(lo->ldo_stripe[i]);
3420 rc = lod_sub_object_create(env, lo->ldo_stripe[i], attr, NULL,
3427 rc = lod_generate_and_set_lovea(env, lo, th);
3433 * Implementation of dt_object_operations::do_create.
3435 * If any of preceeding methods (like ->do_declare_create(),
3436 * ->do_ah_init(), etc) chose to create a striped object,
3437 * then this method will create the master and the stripes.
3439 * \see dt_object_operations::do_create() in the API description for details.
3441 static int lod_object_create(const struct lu_env *env, struct dt_object *dt,
3442 struct lu_attr *attr,
3443 struct dt_allocation_hint *hint,
3444 struct dt_object_format *dof, struct thandle *th)
3446 struct lod_object *lo = lod_dt_obj(dt);
3450 /* create local object */
3451 rc = lod_sub_object_create(env, dt_object_child(dt), attr, hint, dof,
3456 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
3457 lo->ldo_stripe && dof->u.dof_reg.striped != 0)
3458 rc = lod_striping_create(env, dt, attr, dof, th);
3464 * Implementation of dt_object_operations::do_declare_destroy.
3466 * If the object is a striped directory, then the function declares reference
3467 * removal from the master object (this is an index) to the stripes and declares
3468 * destroy of all the stripes. In all the cases, it declares an intention to
3469 * destroy the object itself.
3471 * \see dt_object_operations::do_declare_destroy() in the API description
3474 static int lod_declare_object_destroy(const struct lu_env *env,
3475 struct dt_object *dt,
3478 struct dt_object *next = dt_object_child(dt);
3479 struct lod_object *lo = lod_dt_obj(dt);
3480 struct lod_thread_info *info = lod_env_info(env);
3481 char *stripe_name = info->lti_key;
3486 * load striping information, notice we don't do this when object
3487 * is being initialized as we don't need this information till
3488 * few specific cases like destroy, chown
3490 rc = lod_load_striping(env, lo);
3494 /* declare destroy for all underlying objects */
3495 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
3496 rc = next->do_ops->do_index_try(env, next,
3497 &dt_directory_features);
3501 for (i = 0; i < lo->ldo_stripenr; i++) {
3502 rc = lod_sub_object_declare_ref_del(env, next, th);
3506 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3507 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
3509 rc = lod_sub_object_declare_delete(env, next,
3510 (const struct dt_key *)stripe_name, th);
3517 * we declare destroy for the local object
3519 rc = lod_sub_object_declare_destroy(env, next, th);
3523 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ))
3526 /* declare destroy all striped objects */
3527 for (i = 0; i < lo->ldo_stripenr; i++) {
3528 if (lo->ldo_stripe[i] == NULL)
3531 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
3532 rc = lod_sub_object_declare_ref_del(env,
3533 lo->ldo_stripe[i], th);
3535 rc = lod_sub_object_declare_destroy(env, lo->ldo_stripe[i],
3545 * Implementation of dt_object_operations::do_destroy.
3547 * If the object is a striped directory, then the function removes references
3548 * from the master object (this is an index) to the stripes and destroys all
3549 * the stripes. In all the cases, the function destroys the object itself.
3551 * \see dt_object_operations::do_destroy() in the API description for details.
3553 static int lod_object_destroy(const struct lu_env *env,
3554 struct dt_object *dt, struct thandle *th)
3556 struct dt_object *next = dt_object_child(dt);
3557 struct lod_object *lo = lod_dt_obj(dt);
3558 struct lod_thread_info *info = lod_env_info(env);
3559 char *stripe_name = info->lti_key;
3564 /* destroy sub-stripe of master object */
3565 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
3566 rc = next->do_ops->do_index_try(env, next,
3567 &dt_directory_features);
3571 for (i = 0; i < lo->ldo_stripenr; i++) {
3572 rc = lod_sub_object_ref_del(env, next, th);
3576 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3577 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
3580 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
3581 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
3582 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)));
3584 rc = lod_sub_object_delete(env, next,
3585 (const struct dt_key *)stripe_name, th);
3591 rc = lod_sub_object_destroy(env, next, th);
3595 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ))
3598 /* destroy all striped objects */
3599 for (i = 0; i < lo->ldo_stripenr; i++) {
3600 if (likely(lo->ldo_stripe[i] != NULL) &&
3601 (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
3602 i == cfs_fail_val)) {
3603 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
3604 dt_write_lock(env, lo->ldo_stripe[i],
3606 rc = lod_sub_object_ref_del(env,
3607 lo->ldo_stripe[i], th);
3608 dt_write_unlock(env, lo->ldo_stripe[i]);
3613 rc = lod_sub_object_destroy(env, lo->ldo_stripe[i], th);
3623 * Implementation of dt_object_operations::do_declare_ref_add.
3625 * \see dt_object_operations::do_declare_ref_add() in the API description
3628 static int lod_declare_ref_add(const struct lu_env *env,
3629 struct dt_object *dt, struct thandle *th)
3631 return lod_sub_object_declare_ref_add(env, dt_object_child(dt), th);
3635 * Implementation of dt_object_operations::do_ref_add.
3637 * \see dt_object_operations::do_ref_add() in the API description for details.
3639 static int lod_ref_add(const struct lu_env *env,
3640 struct dt_object *dt, struct thandle *th)
3642 return lod_sub_object_ref_add(env, dt_object_child(dt), th);
3646 * Implementation of dt_object_operations::do_declare_ref_del.
3648 * \see dt_object_operations::do_declare_ref_del() in the API description
3651 static int lod_declare_ref_del(const struct lu_env *env,
3652 struct dt_object *dt, struct thandle *th)
3654 return lod_sub_object_declare_ref_del(env, dt_object_child(dt), th);
3658 * Implementation of dt_object_operations::do_ref_del
3660 * \see dt_object_operations::do_ref_del() in the API description for details.
3662 static int lod_ref_del(const struct lu_env *env,
3663 struct dt_object *dt, struct thandle *th)
3665 return lod_sub_object_ref_del(env, dt_object_child(dt), th);
3669 * Implementation of dt_object_operations::do_object_sync.
3671 * \see dt_object_operations::do_object_sync() in the API description
3674 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
3675 __u64 start, __u64 end)
3677 return dt_object_sync(env, dt_object_child(dt), start, end);
3681 * Release LDLM locks on the stripes of a striped directory.
3683 * Iterates over all the locks taken on the stripe objects and
3686 * \param[in] env execution environment
3687 * \param[in] dt striped object
3688 * \param[in] einfo lock description
3689 * \param[in] policy data describing requested lock
3691 * \retval 0 on success
3692 * \retval negative if failed
3694 static int lod_object_unlock_internal(const struct lu_env *env,
3695 struct dt_object *dt,
3696 struct ldlm_enqueue_info *einfo,
3697 union ldlm_policy_data *policy)
3699 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
3704 if (slave_locks == NULL)
3707 for (i = 1; i < slave_locks->count; i++) {
3708 if (lustre_handle_is_used(&slave_locks->handles[i]))
3709 ldlm_lock_decref_and_cancel(&slave_locks->handles[i],
3717 * Implementation of dt_object_operations::do_object_unlock.
3719 * Used to release LDLM lock(s).
3721 * \see dt_object_operations::do_object_unlock() in the API description
3724 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
3725 struct ldlm_enqueue_info *einfo,
3726 union ldlm_policy_data *policy)
3728 struct lod_object *lo = lod_dt_obj(dt);
3729 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
3730 int slave_locks_size;
3734 if (slave_locks == NULL)
3737 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
3738 LASSERT(lo->ldo_stripenr > 1);
3739 /* Note: for remote lock for single stripe dir, MDT will cancel
3740 * the lock by lockh directly */
3741 LASSERT(!dt_object_remote(dt_object_child(dt)));
3743 /* locks were unlocked in MDT layer */
3744 for (i = 1; i < slave_locks->count; i++) {
3745 LASSERT(!lustre_handle_is_used(&slave_locks->handles[i]));
3746 dt_invalidate(env, lo->ldo_stripe[i]);
3749 slave_locks_size = sizeof(*slave_locks) + slave_locks->count *
3750 sizeof(slave_locks->handles[0]);
3751 OBD_FREE(slave_locks, slave_locks_size);
3752 einfo->ei_cbdata = NULL;
3758 * Implementation of dt_object_operations::do_object_lock.
3760 * Used to get LDLM lock on the non-striped and striped objects.
3762 * \see dt_object_operations::do_object_lock() in the API description
3765 static int lod_object_lock(const struct lu_env *env,
3766 struct dt_object *dt,
3767 struct lustre_handle *lh,
3768 struct ldlm_enqueue_info *einfo,
3769 union ldlm_policy_data *policy)
3771 struct lod_object *lo = lod_dt_obj(dt);
3774 int slave_locks_size;
3775 struct lustre_handle_array *slave_locks = NULL;
3778 /* remote object lock */
3779 if (!einfo->ei_enq_slave) {
3780 LASSERT(dt_object_remote(dt));
3781 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
3785 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3786 GOTO(out, rc = -ENOTDIR);
3788 rc = lod_load_striping(env, lo);
3793 if (lo->ldo_stripenr <= 1) {
3795 * NB, ei_cbdata stores pointer to slave locks, if no locks
3796 * taken, make sure it's set to NULL, otherwise MDT will try to
3799 einfo->ei_cbdata = NULL;
3803 slave_locks_size = sizeof(*slave_locks) + lo->ldo_stripenr *
3804 sizeof(slave_locks->handles[0]);
3805 /* Freed in lod_object_unlock */
3806 OBD_ALLOC(slave_locks, slave_locks_size);
3807 if (slave_locks == NULL)
3808 GOTO(out, rc = -ENOMEM);
3809 slave_locks->count = lo->ldo_stripenr;
3811 /* striped directory lock */
3812 for (i = 1; i < lo->ldo_stripenr; i++) {
3813 struct lustre_handle lockh;
3814 struct ldlm_res_id *res_id;
3816 res_id = &lod_env_info(env)->lti_res_id;
3817 fid_build_reg_res_name(lu_object_fid(&lo->ldo_stripe[i]->do_lu),
3819 einfo->ei_res_id = res_id;
3821 LASSERT(lo->ldo_stripe[i] != NULL);
3822 if (likely(dt_object_remote(lo->ldo_stripe[i]))) {
3823 rc = dt_object_lock(env, lo->ldo_stripe[i], &lockh,
3826 struct ldlm_namespace *ns = einfo->ei_namespace;
3827 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
3828 ldlm_completion_callback completion = einfo->ei_cb_cp;
3829 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
3831 if (einfo->ei_mode == LCK_PW ||
3832 einfo->ei_mode == LCK_EX)
3833 dlmflags |= LDLM_FL_COS_INCOMPAT;
3835 /* This only happens if there are mulitple stripes
3836 * on the master MDT, i.e. except stripe0, there are
3837 * other stripes on the Master MDT as well, Only
3838 * happens in the test case right now. */
3839 LASSERT(ns != NULL);
3840 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS,
3841 policy, einfo->ei_mode,
3842 &dlmflags, blocking,
3844 NULL, 0, LVB_T_NONE,
3849 slave_locks->handles[i] = lockh;
3851 einfo->ei_cbdata = slave_locks;
3853 if (rc != 0 && slave_locks != NULL) {
3854 lod_object_unlock_internal(env, dt, einfo, policy);
3855 OBD_FREE(slave_locks, slave_locks_size);
3860 einfo->ei_cbdata = NULL;
3865 * Implementation of dt_object_operations::do_invalidate.
3867 * \see dt_object_operations::do_invalidate() in the API description for details
3869 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
3871 return dt_invalidate(env, dt_object_child(dt));
3874 struct dt_object_operations lod_obj_ops = {
3875 .do_read_lock = lod_object_read_lock,
3876 .do_write_lock = lod_object_write_lock,
3877 .do_read_unlock = lod_object_read_unlock,
3878 .do_write_unlock = lod_object_write_unlock,
3879 .do_write_locked = lod_object_write_locked,
3880 .do_attr_get = lod_attr_get,
3881 .do_declare_attr_set = lod_declare_attr_set,
3882 .do_attr_set = lod_attr_set,
3883 .do_xattr_get = lod_xattr_get,
3884 .do_declare_xattr_set = lod_declare_xattr_set,
3885 .do_xattr_set = lod_xattr_set,
3886 .do_declare_xattr_del = lod_declare_xattr_del,
3887 .do_xattr_del = lod_xattr_del,
3888 .do_xattr_list = lod_xattr_list,
3889 .do_ah_init = lod_ah_init,
3890 .do_declare_create = lod_declare_object_create,
3891 .do_create = lod_object_create,
3892 .do_declare_destroy = lod_declare_object_destroy,
3893 .do_destroy = lod_object_destroy,
3894 .do_index_try = lod_index_try,
3895 .do_declare_ref_add = lod_declare_ref_add,
3896 .do_ref_add = lod_ref_add,
3897 .do_declare_ref_del = lod_declare_ref_del,
3898 .do_ref_del = lod_ref_del,
3899 .do_object_sync = lod_object_sync,
3900 .do_object_lock = lod_object_lock,
3901 .do_object_unlock = lod_object_unlock,
3902 .do_invalidate = lod_invalidate,
3906 * Implementation of dt_body_operations::dbo_read.
3908 * \see dt_body_operations::dbo_read() in the API description for details.
3910 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
3911 struct lu_buf *buf, loff_t *pos)
3913 struct dt_object *next = dt_object_child(dt);
3914 return next->do_body_ops->dbo_read(env, next, buf, pos);
3918 * Implementation of dt_body_operations::dbo_declare_write.
3920 * \see dt_body_operations::dbo_declare_write() in the API description
3923 static ssize_t lod_declare_write(const struct lu_env *env,
3924 struct dt_object *dt,
3925 const struct lu_buf *buf, loff_t pos,
3928 return lod_sub_object_declare_write(env, dt_object_child(dt), buf, pos,
3933 * Implementation of dt_body_operations::dbo_write.
3935 * \see dt_body_operations::dbo_write() in the API description for details.
3937 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
3938 const struct lu_buf *buf, loff_t *pos,
3939 struct thandle *th, int iq)
3941 return lod_sub_object_write(env, dt_object_child(dt), buf, pos, th, iq);
3944 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
3945 __u64 start, __u64 end, struct thandle *th)
3947 if (dt_object_remote(dt))
3950 return lod_sub_object_declare_punch(env, dt_object_child(dt), start,
3954 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
3955 __u64 start, __u64 end, struct thandle *th)
3957 if (dt_object_remote(dt))
3960 return lod_sub_object_punch(env, dt_object_child(dt), start, end, th);
3963 static const struct dt_body_operations lod_body_lnk_ops = {
3964 .dbo_read = lod_read,
3965 .dbo_declare_write = lod_declare_write,
3966 .dbo_write = lod_write
3969 static const struct dt_body_operations lod_body_ops = {
3970 .dbo_read = lod_read,
3971 .dbo_declare_write = lod_declare_write,
3972 .dbo_write = lod_write,
3973 .dbo_declare_punch = lod_declare_punch,
3974 .dbo_punch = lod_punch,
3978 * Implementation of lu_object_operations::loo_object_init.
3980 * The function determines the type and the index of the target device using
3981 * sequence of the object's FID. Then passes control down to the
3982 * corresponding device:
3983 * OSD for the local objects, OSP for remote
3985 * \see lu_object_operations::loo_object_init() in the API description
3988 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
3989 const struct lu_object_conf *conf)
3991 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
3992 struct lu_device *cdev = NULL;
3993 struct lu_object *cobj;
3994 struct lod_tgt_descs *ltd = NULL;
3995 struct lod_tgt_desc *tgt;
3997 int type = LU_SEQ_RANGE_ANY;
4001 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
4003 /* Note: Sometimes, it will Return EAGAIN here, see
4004 * ptrlpc_import_delay_req(), which might confuse
4005 * lu_object_find_at() and make it wait there incorrectly.
4006 * so we convert it to EIO here.*/
4013 if (type == LU_SEQ_RANGE_MDT &&
4014 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
4015 cdev = &lod->lod_child->dd_lu_dev;
4016 } else if (type == LU_SEQ_RANGE_MDT) {
4017 ltd = &lod->lod_mdt_descs;
4019 } else if (type == LU_SEQ_RANGE_OST) {
4020 ltd = &lod->lod_ost_descs;
4027 if (ltd->ltd_tgts_size > idx &&
4028 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
4029 tgt = LTD_TGT(ltd, idx);
4031 LASSERT(tgt != NULL);
4032 LASSERT(tgt->ltd_tgt != NULL);
4034 cdev = &(tgt->ltd_tgt->dd_lu_dev);
4036 lod_putref(lod, ltd);
4039 if (unlikely(cdev == NULL))
4042 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
4043 if (unlikely(cobj == NULL))
4046 lu_object_add(lo, cobj);
4053 * Release resources associated with striping.
4055 * If the object is striped (regular or directory), then release
4056 * the stripe objects references and free the ldo_stripe array.
4058 * \param[in] env execution environment
4059 * \param[in] lo object
4061 void lod_object_free_striping(const struct lu_env *env, struct lod_object *lo)
4065 if (lo->ldo_stripe != NULL) {
4068 LASSERT(lo->ldo_stripes_allocated > 0);
4070 for (i = 0; i < lo->ldo_stripenr; i++) {
4071 if (lo->ldo_stripe[i])
4072 lu_object_put(env, &lo->ldo_stripe[i]->do_lu);
4075 len = sizeof(struct dt_object *) * lo->ldo_stripes_allocated;
4076 OBD_FREE(lo->ldo_stripe, len);
4077 lo->ldo_stripe = NULL;
4078 lo->ldo_stripes_allocated = 0;
4080 lo->ldo_stripenr = 0;
4084 * Implementation of lu_object_operations::loo_object_start.
4086 * \see lu_object_operations::loo_object_start() in the API description
4089 static int lod_object_start(const struct lu_env *env, struct lu_object *o)
4091 if (S_ISLNK(o->lo_header->loh_attr & S_IFMT)) {
4092 lu2lod_obj(o)->ldo_obj.do_body_ops = &lod_body_lnk_ops;
4093 } else if (S_ISREG(o->lo_header->loh_attr & S_IFMT) ||
4094 fid_is_local_file(lu_object_fid(o))) {
4095 /* Note: some local file (like last rcvd) is created
4096 * through bottom layer (OSD), so the object initialization
4097 * comes to lod, it does not set loh_attr yet, so
4098 * set do_body_ops for local file anyway */
4099 lu2lod_obj(o)->ldo_obj.do_body_ops = &lod_body_ops;
4105 * Implementation of lu_object_operations::loo_object_free.
4107 * \see lu_object_operations::loo_object_free() in the API description
4110 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
4112 struct lod_object *lo = lu2lod_obj(o);
4114 lod_object_set_pool(lo, NULL);
4115 /* release all underlying object pinned */
4116 lod_object_free_striping(env, lo);
4118 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
4122 * Implementation of lu_object_operations::loo_object_release.
4124 * \see lu_object_operations::loo_object_release() in the API description
4127 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
4129 /* XXX: shouldn't we release everything here in case if object
4130 * creation failed before? */
4134 * Implementation of lu_object_operations::loo_object_print.
4136 * \see lu_object_operations::loo_object_print() in the API description
4139 static int lod_object_print(const struct lu_env *env, void *cookie,
4140 lu_printer_t p, const struct lu_object *l)
4142 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
4144 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
4147 struct lu_object_operations lod_lu_obj_ops = {
4148 .loo_object_init = lod_object_init,
4149 .loo_object_start = lod_object_start,
4150 .loo_object_free = lod_object_free,
4151 .loo_object_release = lod_object_release,
4152 .loo_object_print = lod_object_print,