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 * lustre/doc/osd-api.txt.
38 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
41 #define DEBUG_SUBSYSTEM S_MDS
44 #include <obd_class.h>
45 #include <lustre_ver.h>
46 #include <obd_support.h>
47 #include <lprocfs_status.h>
49 #include <lustre_fid.h>
50 #include <lustre_param.h>
51 #include <lustre_fid.h>
52 #include <lustre_lmv.h>
53 #include <md_object.h>
54 #include <lustre_linkea.h>
56 #include "lod_internal.h"
58 static const char dot[] = ".";
59 static const char dotdot[] = "..";
61 static const struct dt_body_operations lod_body_lnk_ops;
62 static const struct dt_body_operations lod_body_ops;
65 * Implementation of dt_index_operations::dio_lookup
67 * Used with regular (non-striped) objects.
69 * \see dt_index_operations::dio_lookup() in the API description for details.
71 static int lod_index_lookup(const struct lu_env *env, struct dt_object *dt,
72 struct dt_rec *rec, const struct dt_key *key)
74 struct dt_object *next = dt_object_child(dt);
75 return next->do_index_ops->dio_lookup(env, next, rec, key);
79 * Implementation of dt_index_operations::dio_declare_insert.
81 * Used with regular (non-striped) objects.
83 * \see dt_index_operations::dio_declare_insert() in the API description
86 static int lod_declare_index_insert(const struct lu_env *env,
88 const struct dt_rec *rec,
89 const struct dt_key *key,
92 return lod_sub_object_declare_insert(env, dt_object_child(dt),
97 * Implementation of dt_index_operations::dio_insert.
99 * Used with regular (non-striped) objects
101 * \see dt_index_operations::dio_insert() in the API description for details.
103 static int lod_index_insert(const struct lu_env *env,
104 struct dt_object *dt,
105 const struct dt_rec *rec,
106 const struct dt_key *key,
110 return lod_sub_object_index_insert(env, dt_object_child(dt), rec, key,
115 * Implementation of dt_index_operations::dio_declare_delete.
117 * Used with regular (non-striped) objects.
119 * \see dt_index_operations::dio_declare_delete() in the API description
122 static int lod_declare_index_delete(const struct lu_env *env,
123 struct dt_object *dt,
124 const struct dt_key *key,
127 return lod_sub_object_declare_delete(env, dt_object_child(dt), key,
132 * Implementation of dt_index_operations::dio_delete.
134 * Used with regular (non-striped) objects.
136 * \see dt_index_operations::dio_delete() in the API description for details.
138 static int lod_index_delete(const struct lu_env *env,
139 struct dt_object *dt,
140 const struct dt_key *key,
143 return lod_sub_object_delete(env, dt_object_child(dt), key, th);
147 * Implementation of dt_it_ops::init.
149 * Used with regular (non-striped) objects.
151 * \see dt_it_ops::init() in the API description for details.
153 static struct dt_it *lod_it_init(const struct lu_env *env,
154 struct dt_object *dt, __u32 attr)
156 struct dt_object *next = dt_object_child(dt);
157 struct lod_it *it = &lod_env_info(env)->lti_it;
158 struct dt_it *it_next;
160 it_next = next->do_index_ops->dio_it.init(env, next, attr);
164 /* currently we do not use more than one iterator per thread
165 * so we store it in thread info. if at some point we need
166 * more active iterators in a single thread, we can allocate
168 LASSERT(it->lit_obj == NULL);
170 it->lit_it = it_next;
173 return (struct dt_it *)it;
176 #define LOD_CHECK_IT(env, it) \
178 LASSERT((it)->lit_obj != NULL); \
179 LASSERT((it)->lit_it != NULL); \
183 * Implementation of dt_index_operations::dio_it.fini.
185 * Used with regular (non-striped) objects.
187 * \see dt_index_operations::dio_it.fini() in the API description for details.
189 static void lod_it_fini(const struct lu_env *env, struct dt_it *di)
191 struct lod_it *it = (struct lod_it *)di;
193 LOD_CHECK_IT(env, it);
194 it->lit_obj->do_index_ops->dio_it.fini(env, it->lit_it);
196 /* the iterator not in use any more */
202 * Implementation of dt_it_ops::get.
204 * Used with regular (non-striped) objects.
206 * \see dt_it_ops::get() in the API description for details.
208 static int lod_it_get(const struct lu_env *env, struct dt_it *di,
209 const struct dt_key *key)
211 const struct lod_it *it = (const struct lod_it *)di;
213 LOD_CHECK_IT(env, it);
214 return it->lit_obj->do_index_ops->dio_it.get(env, it->lit_it, key);
218 * Implementation of dt_it_ops::put.
220 * Used with regular (non-striped) objects.
222 * \see dt_it_ops::put() in the API description for details.
224 static void lod_it_put(const struct lu_env *env, struct dt_it *di)
226 struct lod_it *it = (struct lod_it *)di;
228 LOD_CHECK_IT(env, it);
229 return it->lit_obj->do_index_ops->dio_it.put(env, it->lit_it);
233 * Implementation of dt_it_ops::next.
235 * Used with regular (non-striped) objects
237 * \see dt_it_ops::next() in the API description for details.
239 static int lod_it_next(const struct lu_env *env, struct dt_it *di)
241 struct lod_it *it = (struct lod_it *)di;
243 LOD_CHECK_IT(env, it);
244 return it->lit_obj->do_index_ops->dio_it.next(env, it->lit_it);
248 * Implementation of dt_it_ops::key.
250 * Used with regular (non-striped) objects.
252 * \see dt_it_ops::key() in the API description for details.
254 static struct dt_key *lod_it_key(const struct lu_env *env,
255 const struct dt_it *di)
257 const struct lod_it *it = (const struct lod_it *)di;
259 LOD_CHECK_IT(env, it);
260 return it->lit_obj->do_index_ops->dio_it.key(env, it->lit_it);
264 * Implementation of dt_it_ops::key_size.
266 * Used with regular (non-striped) objects.
268 * \see dt_it_ops::key_size() in the API description for details.
270 static int lod_it_key_size(const struct lu_env *env, const struct dt_it *di)
272 struct lod_it *it = (struct lod_it *)di;
274 LOD_CHECK_IT(env, it);
275 return it->lit_obj->do_index_ops->dio_it.key_size(env, it->lit_it);
279 * Implementation of dt_it_ops::rec.
281 * Used with regular (non-striped) objects.
283 * \see dt_it_ops::rec() in the API description for details.
285 static int lod_it_rec(const struct lu_env *env, const struct dt_it *di,
286 struct dt_rec *rec, __u32 attr)
288 const struct lod_it *it = (const struct lod_it *)di;
290 LOD_CHECK_IT(env, it);
291 return it->lit_obj->do_index_ops->dio_it.rec(env, it->lit_it, rec,
296 * Implementation of dt_it_ops::rec_size.
298 * Used with regular (non-striped) objects.
300 * \see dt_it_ops::rec_size() in the API description for details.
302 static int lod_it_rec_size(const struct lu_env *env, const struct dt_it *di,
305 const struct lod_it *it = (const struct lod_it *)di;
307 LOD_CHECK_IT(env, it);
308 return it->lit_obj->do_index_ops->dio_it.rec_size(env, it->lit_it,
313 * Implementation of dt_it_ops::store.
315 * Used with regular (non-striped) objects.
317 * \see dt_it_ops::store() in the API description for details.
319 static __u64 lod_it_store(const struct lu_env *env, const struct dt_it *di)
321 const struct lod_it *it = (const struct lod_it *)di;
323 LOD_CHECK_IT(env, it);
324 return it->lit_obj->do_index_ops->dio_it.store(env, it->lit_it);
328 * Implementation of dt_it_ops::load.
330 * Used with regular (non-striped) objects.
332 * \see dt_it_ops::load() in the API description for details.
334 static int lod_it_load(const struct lu_env *env, const struct dt_it *di,
337 const struct lod_it *it = (const struct lod_it *)di;
339 LOD_CHECK_IT(env, it);
340 return it->lit_obj->do_index_ops->dio_it.load(env, it->lit_it, hash);
344 * Implementation of dt_it_ops::key_rec.
346 * Used with regular (non-striped) objects.
348 * \see dt_it_ops::rec() in the API description for details.
350 static int lod_it_key_rec(const struct lu_env *env, const struct dt_it *di,
353 const struct lod_it *it = (const struct lod_it *)di;
355 LOD_CHECK_IT(env, it);
356 return it->lit_obj->do_index_ops->dio_it.key_rec(env, it->lit_it,
360 static struct dt_index_operations lod_index_ops = {
361 .dio_lookup = lod_index_lookup,
362 .dio_declare_insert = lod_declare_index_insert,
363 .dio_insert = lod_index_insert,
364 .dio_declare_delete = lod_declare_index_delete,
365 .dio_delete = lod_index_delete,
373 .key_size = lod_it_key_size,
375 .rec_size = lod_it_rec_size,
376 .store = lod_it_store,
378 .key_rec = lod_it_key_rec,
383 * Implementation of dt_it_ops::init.
385 * Used with striped objects. Internally just initializes the iterator
386 * on the first stripe.
388 * \see dt_it_ops::init() in the API description for details.
390 static struct dt_it *lod_striped_it_init(const struct lu_env *env,
391 struct dt_object *dt, __u32 attr)
393 struct lod_object *lo = lod_dt_obj(dt);
394 struct dt_object *next;
395 struct lod_it *it = &lod_env_info(env)->lti_it;
396 struct dt_it *it_next;
399 LASSERT(lo->ldo_stripenr > 0);
400 next = lo->ldo_stripe[0];
401 LASSERT(next != NULL);
402 LASSERT(next->do_index_ops != NULL);
404 it_next = next->do_index_ops->dio_it.init(env, next, attr);
408 /* currently we do not use more than one iterator per thread
409 * so we store it in thread info. if at some point we need
410 * more active iterators in a single thread, we can allocate
412 LASSERT(it->lit_obj == NULL);
414 it->lit_stripe_index = 0;
416 it->lit_it = it_next;
419 return (struct dt_it *)it;
422 #define LOD_CHECK_STRIPED_IT(env, it, lo) \
424 LASSERT((it)->lit_obj != NULL); \
425 LASSERT((it)->lit_it != NULL); \
426 LASSERT((lo)->ldo_stripenr > 0); \
427 LASSERT((it)->lit_stripe_index < (lo)->ldo_stripenr); \
431 * Implementation of dt_it_ops::fini.
433 * Used with striped objects.
435 * \see dt_it_ops::fini() in the API description for details.
437 static void lod_striped_it_fini(const struct lu_env *env, struct dt_it *di)
439 struct lod_it *it = (struct lod_it *)di;
440 struct lod_object *lo = lod_dt_obj(it->lit_obj);
441 struct dt_object *next;
443 /* If lit_it == NULL, then it means the sub_it has been finished,
444 * which only happens in failure cases, see lod_striped_it_next() */
445 if (it->lit_it != NULL) {
446 LOD_CHECK_STRIPED_IT(env, it, lo);
448 next = lo->ldo_stripe[it->lit_stripe_index];
449 LASSERT(next != NULL);
450 LASSERT(next->do_index_ops != NULL);
452 next->do_index_ops->dio_it.fini(env, it->lit_it);
455 /* the iterator not in use any more */
458 it->lit_stripe_index = 0;
462 * Implementation of dt_it_ops::get.
464 * Right now it's not used widely, only to reset the iterator to the
465 * initial position. It should be possible to implement a full version
466 * which chooses a correct stripe to be able to position with any key.
468 * \see dt_it_ops::get() in the API description for details.
470 static int lod_striped_it_get(const struct lu_env *env, struct dt_it *di,
471 const struct dt_key *key)
473 const struct lod_it *it = (const struct lod_it *)di;
474 struct lod_object *lo = lod_dt_obj(it->lit_obj);
475 struct dt_object *next;
478 LOD_CHECK_STRIPED_IT(env, it, lo);
480 next = lo->ldo_stripe[it->lit_stripe_index];
481 LASSERT(next != NULL);
482 LASSERT(next->do_index_ops != NULL);
484 return next->do_index_ops->dio_it.get(env, it->lit_it, key);
488 * Implementation of dt_it_ops::put.
490 * Used with striped objects.
492 * \see dt_it_ops::put() in the API description for details.
494 static void lod_striped_it_put(const struct lu_env *env, struct dt_it *di)
496 struct lod_it *it = (struct lod_it *)di;
497 struct lod_object *lo = lod_dt_obj(it->lit_obj);
498 struct dt_object *next;
500 LOD_CHECK_STRIPED_IT(env, it, lo);
502 next = lo->ldo_stripe[it->lit_stripe_index];
503 LASSERT(next != NULL);
504 LASSERT(next->do_index_ops != NULL);
506 return next->do_index_ops->dio_it.put(env, it->lit_it);
510 * Implementation of dt_it_ops::next.
512 * Used with striped objects. When the end of the current stripe is
513 * reached, the method takes the next stripe's iterator.
515 * \see dt_it_ops::next() in the API description for details.
517 static int lod_striped_it_next(const struct lu_env *env, struct dt_it *di)
519 struct lod_it *it = (struct lod_it *)di;
520 struct lod_object *lo = lod_dt_obj(it->lit_obj);
521 struct dt_object *next;
522 struct dt_it *it_next;
526 LOD_CHECK_STRIPED_IT(env, it, lo);
528 next = lo->ldo_stripe[it->lit_stripe_index];
529 LASSERT(next != NULL);
530 LASSERT(next->do_index_ops != NULL);
532 rc = next->do_index_ops->dio_it.next(env, it->lit_it);
536 if (rc == 0 && it->lit_stripe_index == 0)
539 if (rc == 0 && it->lit_stripe_index > 0) {
540 struct lu_dirent *ent;
542 ent = (struct lu_dirent *)lod_env_info(env)->lti_key;
544 rc = next->do_index_ops->dio_it.rec(env, it->lit_it,
545 (struct dt_rec *)ent,
550 /* skip . and .. for slave stripe */
551 if ((strncmp(ent->lde_name, ".",
552 le16_to_cpu(ent->lde_namelen)) == 0 &&
553 le16_to_cpu(ent->lde_namelen) == 1) ||
554 (strncmp(ent->lde_name, "..",
555 le16_to_cpu(ent->lde_namelen)) == 0 &&
556 le16_to_cpu(ent->lde_namelen) == 2))
562 /* go to next stripe */
563 if (it->lit_stripe_index + 1 >= lo->ldo_stripenr)
566 it->lit_stripe_index++;
568 next->do_index_ops->dio_it.put(env, it->lit_it);
569 next->do_index_ops->dio_it.fini(env, it->lit_it);
572 next = lo->ldo_stripe[it->lit_stripe_index];
573 LASSERT(next != NULL);
574 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 * Mark all of the striped directory sub-stripes dead.
1070 * When a striped object is a subject to removal, we have
1071 * to mark all the stripes to prevent further access to
1072 * them (e.g. create a new file in those). So we mark
1073 * all the stripes with LMV_HASH_FLAG_DEAD. The function
1074 * can be used to declare the changes and to apply them.
1075 * If the object isn't striped, then just return success.
1077 * \param[in] env execution environment
1078 * \param[in] dt the striped object
1079 * \param[in] handle transaction handle
1080 * \param[in] declare whether to declare the change or apply
1082 * \retval 0 on success
1083 * \retval negative if failed
1085 static int lod_mark_dead_object(const struct lu_env *env,
1086 struct dt_object *dt,
1090 struct lod_object *lo = lod_dt_obj(dt);
1091 struct lmv_mds_md_v1 *lmv;
1092 __u32 dead_hash_type;
1098 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
1101 rc = lod_load_striping_locked(env, lo);
1105 if (lo->ldo_stripenr == 0)
1108 rc = lod_get_lmv_ea(env, lo);
1112 lmv = lod_env_info(env)->lti_ea_store;
1113 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1114 dead_hash_type = le32_to_cpu(lmv->lmv_hash_type) | LMV_HASH_FLAG_DEAD;
1115 lmv->lmv_hash_type = cpu_to_le32(dead_hash_type);
1116 for (i = 0; i < lo->ldo_stripenr; i++) {
1119 lmv->lmv_master_mdt_index = i;
1121 buf.lb_len = sizeof(*lmv);
1123 rc = lod_sub_object_declare_xattr_set(env,
1124 lo->ldo_stripe[i], &buf,
1126 LU_XATTR_REPLACE, th);
1128 rc = lod_sub_object_xattr_set(env, lo->ldo_stripe[i],
1129 &buf, XATTR_NAME_LMV,
1130 LU_XATTR_REPLACE, th);
1140 * Implementation of dt_object_operations::do_declare_attr_set.
1142 * If the object is striped, then apply the changes to all the stripes.
1144 * \see dt_object_operations::do_declare_attr_set() in the API description
1147 static int lod_declare_attr_set(const struct lu_env *env,
1148 struct dt_object *dt,
1149 const struct lu_attr *attr,
1152 struct dt_object *next = dt_object_child(dt);
1153 struct lod_object *lo = lod_dt_obj(dt);
1157 /* Set dead object on all other stripes */
1158 if (attr->la_valid & LA_FLAGS && !(attr->la_valid & ~LA_FLAGS) &&
1159 attr->la_flags & LUSTRE_SLAVE_DEAD_FL) {
1160 rc = lod_mark_dead_object(env, dt, th, true);
1165 * declare setattr on the local object
1167 rc = lod_sub_object_declare_attr_set(env, next, attr, th);
1171 /* osp_declare_attr_set() ignores all attributes other than
1172 * UID, GID, and size, and osp_attr_set() ignores all but UID
1173 * and GID. Declaration of size attr setting happens through
1174 * lod_declare_init_size(), and not through this function.
1175 * Therefore we need not load striping unless ownership is
1176 * changing. This should save memory and (we hope) speed up
1178 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1179 if (!(attr->la_valid & (LA_UID | LA_GID)))
1182 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1185 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE |
1186 LA_ATIME | LA_MTIME | LA_CTIME |
1191 * load striping information, notice we don't do this when object
1192 * is being initialized as we don't need this information till
1193 * few specific cases like destroy, chown
1195 rc = lod_load_striping(env, lo);
1199 if (lo->ldo_stripenr == 0)
1203 * if object is striped declare changes on the stripes
1205 LASSERT(lo->ldo_stripe);
1206 for (i = 0; i < lo->ldo_stripenr; i++) {
1207 if (lo->ldo_stripe[i] == NULL)
1209 rc = lod_sub_object_declare_attr_set(env,
1210 lo->ldo_stripe[i], attr,
1216 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE) &&
1217 dt_object_exists(next) != 0 &&
1218 dt_object_remote(next) == 0)
1219 lod_sub_object_declare_xattr_del(env, next,
1220 XATTR_NAME_LOV, th);
1222 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) &&
1223 dt_object_exists(next) &&
1224 dt_object_remote(next) == 0 && S_ISREG(attr->la_mode)) {
1225 struct lod_thread_info *info = lod_env_info(env);
1226 struct lu_buf *buf = &info->lti_buf;
1228 buf->lb_buf = info->lti_ea_store;
1229 buf->lb_len = info->lti_ea_store_size;
1230 lod_sub_object_declare_xattr_set(env, next, buf,
1232 LU_XATTR_REPLACE, th);
1239 * Implementation of dt_object_operations::do_attr_set.
1241 * If the object is striped, then apply the changes to all or subset of
1242 * the stripes depending on the object type and specific attributes.
1244 * \see dt_object_operations::do_attr_set() in the API description for details.
1246 static int lod_attr_set(const struct lu_env *env,
1247 struct dt_object *dt,
1248 const struct lu_attr *attr,
1251 struct dt_object *next = dt_object_child(dt);
1252 struct lod_object *lo = lod_dt_obj(dt);
1256 /* Set dead object on all other stripes */
1257 if (attr->la_valid & LA_FLAGS && !(attr->la_valid & ~LA_FLAGS) &&
1258 attr->la_flags & LUSTRE_SLAVE_DEAD_FL) {
1259 rc = lod_mark_dead_object(env, dt, th, false);
1264 * apply changes to the local object
1266 rc = lod_sub_object_attr_set(env, next, attr, th);
1270 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1271 if (!(attr->la_valid & (LA_UID | LA_GID)))
1274 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1277 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE |
1278 LA_ATIME | LA_MTIME | LA_CTIME |
1283 if (lo->ldo_stripenr == 0)
1287 * if object is striped, apply changes to all the stripes
1289 LASSERT(lo->ldo_stripe);
1290 for (i = 0; i < lo->ldo_stripenr; i++) {
1291 if (unlikely(lo->ldo_stripe[i] == NULL))
1294 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
1295 (dt_object_exists(lo->ldo_stripe[i]) == 0))
1298 rc = lod_sub_object_attr_set(env, lo->ldo_stripe[i], attr, th);
1303 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE) &&
1304 dt_object_exists(next) != 0 &&
1305 dt_object_remote(next) == 0)
1306 rc = lod_sub_object_xattr_del(env, next, XATTR_NAME_LOV, th);
1308 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) &&
1309 dt_object_exists(next) &&
1310 dt_object_remote(next) == 0 && S_ISREG(attr->la_mode)) {
1311 struct lod_thread_info *info = lod_env_info(env);
1312 struct lu_buf *buf = &info->lti_buf;
1313 struct ost_id *oi = &info->lti_ostid;
1314 struct lu_fid *fid = &info->lti_fid;
1315 struct lov_mds_md_v1 *lmm;
1316 struct lov_ost_data_v1 *objs;
1320 rc1 = lod_get_lov_ea(env, lo);
1324 buf->lb_buf = info->lti_ea_store;
1325 buf->lb_len = info->lti_ea_store_size;
1326 lmm = info->lti_ea_store;
1327 magic = le32_to_cpu(lmm->lmm_magic);
1328 if (magic == LOV_MAGIC_V1)
1329 objs = &(lmm->lmm_objects[0]);
1331 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1332 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1333 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1335 fid_to_ostid(fid, oi);
1336 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1338 rc = lod_sub_object_xattr_set(env, next, buf, XATTR_NAME_LOV,
1339 LU_XATTR_REPLACE, th);
1346 * Implementation of dt_object_operations::do_xattr_get.
1348 * If LOV EA is requested from the root object and it's not
1349 * found, then return default striping for the filesystem.
1351 * \see dt_object_operations::do_xattr_get() in the API description for details.
1353 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1354 struct lu_buf *buf, const char *name)
1356 struct lod_thread_info *info = lod_env_info(env);
1357 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1361 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1362 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1363 struct lmv_mds_md_v1 *lmv1;
1366 if (rc > (typeof(rc))sizeof(*lmv1))
1369 if (rc < (typeof(rc))sizeof(*lmv1))
1370 RETURN(rc = rc > 0 ? -EINVAL : rc);
1372 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1373 CLASSERT(sizeof(*lmv1) <= sizeof(info->lti_key));
1375 info->lti_buf.lb_buf = info->lti_key;
1376 info->lti_buf.lb_len = sizeof(*lmv1);
1377 rc = dt_xattr_get(env, dt_object_child(dt),
1378 &info->lti_buf, name);
1379 if (unlikely(rc != sizeof(*lmv1)))
1380 RETURN(rc = rc > 0 ? -EINVAL : rc);
1382 lmv1 = info->lti_buf.lb_buf;
1383 /* The on-disk LMV EA only contains header, but the
1384 * returned LMV EA size should contain the space for
1385 * the FIDs of all shards of the striped directory. */
1386 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1387 rc = lmv_mds_md_size(
1388 le32_to_cpu(lmv1->lmv_stripe_count),
1391 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1395 RETURN(rc = rc1 != 0 ? rc1 : rc);
1398 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1402 * lod returns default striping on the real root of the device
1403 * this is like the root stores default striping for the whole
1404 * filesystem. historically we've been using a different approach
1405 * and store it in the config.
1407 dt_root_get(env, dev->lod_child, &info->lti_fid);
1408 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1410 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1411 struct lov_user_md *lum = buf->lb_buf;
1412 struct lov_desc *desc = &dev->lod_desc;
1414 if (buf->lb_buf == NULL) {
1416 } else if (buf->lb_len >= sizeof(*lum)) {
1417 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1418 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1419 lmm_oi_set_id(&lum->lmm_oi, 0);
1420 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1421 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1422 lum->lmm_stripe_size = cpu_to_le32(
1423 desc->ld_default_stripe_size);
1424 lum->lmm_stripe_count = cpu_to_le16(
1425 desc->ld_default_stripe_count);
1426 lum->lmm_stripe_offset = cpu_to_le16(
1427 desc->ld_default_stripe_offset);
1440 * Checks that the magic of the stripe is sane.
1442 * \param[in] lod lod device
1443 * \param[in] lum a buffer storing LMV EA to verify
1445 * \retval 0 if the EA is sane
1446 * \retval negative otherwise
1448 static int lod_verify_md_striping(struct lod_device *lod,
1449 const struct lmv_user_md_v1 *lum)
1451 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1452 CERROR("%s: invalid lmv_user_md: magic = %x, "
1453 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1454 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1455 (int)le32_to_cpu(lum->lum_stripe_offset),
1456 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1464 * Initialize LMV EA for a slave.
1466 * Initialize slave's LMV EA from the master's LMV EA.
1468 * \param[in] master_lmv a buffer containing master's EA
1469 * \param[out] slave_lmv a buffer where slave's EA will be stored
1472 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1473 const struct lmv_mds_md_v1 *master_lmv)
1475 *slave_lmv = *master_lmv;
1476 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1482 * Generate LMV EA from the object passed as \a dt. The object must have
1483 * the stripes created and initialized.
1485 * \param[in] env execution environment
1486 * \param[in] dt object
1487 * \param[out] lmv_buf buffer storing generated LMV EA
1489 * \retval 0 on success
1490 * \retval negative if failed
1492 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1493 struct lu_buf *lmv_buf)
1495 struct lod_thread_info *info = lod_env_info(env);
1496 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1497 struct lod_object *lo = lod_dt_obj(dt);
1498 struct lmv_mds_md_v1 *lmm1;
1500 int type = LU_SEQ_RANGE_ANY;
1505 LASSERT(lo->ldo_dir_striped != 0);
1506 LASSERT(lo->ldo_stripenr > 0);
1507 stripe_count = lo->ldo_stripenr;
1508 /* Only store the LMV EA heahder on the disk. */
1509 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1510 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1514 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1517 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1518 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1519 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1520 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1521 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1526 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1527 lmv_buf->lb_buf = info->lti_ea_store;
1528 lmv_buf->lb_len = sizeof(*lmm1);
1534 * Create in-core represenation for a striped directory.
1536 * Parse the buffer containing LMV EA and instantiate LU objects
1537 * representing the stripe objects. The pointers to the objects are
1538 * stored in ldo_stripe field of \a lo. This function is used when
1539 * we need to access an already created object (i.e. load from a disk).
1541 * \param[in] env execution environment
1542 * \param[in] lo lod object
1543 * \param[in] buf buffer containing LMV EA
1545 * \retval 0 on success
1546 * \retval negative if failed
1548 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1549 const struct lu_buf *buf)
1551 struct lod_thread_info *info = lod_env_info(env);
1552 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1553 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1554 struct dt_object **stripe;
1555 union lmv_mds_md *lmm = buf->lb_buf;
1556 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1557 struct lu_fid *fid = &info->lti_fid;
1562 if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
1565 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1566 lo->ldo_dir_slave_stripe = 1;
1570 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1573 if (le32_to_cpu(lmv1->lmv_stripe_count) < 1)
1576 LASSERT(lo->ldo_stripe == NULL);
1577 OBD_ALLOC(stripe, sizeof(stripe[0]) *
1578 (le32_to_cpu(lmv1->lmv_stripe_count)));
1582 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1583 struct dt_device *tgt_dt;
1584 struct dt_object *dto;
1585 int type = LU_SEQ_RANGE_ANY;
1588 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1589 if (!fid_is_sane(fid))
1590 GOTO(out, rc = -ESTALE);
1592 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1596 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1597 tgt_dt = lod->lod_child;
1599 struct lod_tgt_desc *tgt;
1601 tgt = LTD_TGT(ltd, idx);
1603 GOTO(out, rc = -ESTALE);
1604 tgt_dt = tgt->ltd_tgt;
1607 dto = dt_locate_at(env, tgt_dt, fid,
1608 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1611 GOTO(out, rc = PTR_ERR(dto));
1616 lo->ldo_stripe = stripe;
1617 lo->ldo_stripenr = le32_to_cpu(lmv1->lmv_stripe_count);
1618 lo->ldo_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1620 lod_object_free_striping(env, lo);
1626 * Declare create a striped directory.
1628 * Declare creating a striped directory with a given stripe pattern on the
1629 * specified MDTs. A striped directory is represented as a regular directory
1630 * - an index listing all the stripes. The stripes point back to the master
1631 * object with ".." and LinkEA. The master object gets LMV EA which
1632 * identifies it as a striped directory. The function allocates FIDs
1635 * \param[in] env execution environment
1636 * \param[in] dt object
1637 * \param[in] attr attributes to initialize the objects with
1638 * \param[in] dof type of objects to be created
1639 * \param[in] th transaction handle
1641 * \retval 0 on success
1642 * \retval negative if failed
1644 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1645 struct dt_object *dt,
1646 struct lu_attr *attr,
1647 struct dt_object_format *dof,
1650 struct lod_thread_info *info = lod_env_info(env);
1651 struct lu_buf lmv_buf;
1652 struct lu_buf slave_lmv_buf;
1653 struct lmv_mds_md_v1 *lmm;
1654 struct lmv_mds_md_v1 *slave_lmm = NULL;
1655 struct dt_insert_rec *rec = &info->lti_dt_rec;
1656 struct lod_object *lo = lod_dt_obj(dt);
1661 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1664 lmm = lmv_buf.lb_buf;
1666 OBD_ALLOC_PTR(slave_lmm);
1667 if (slave_lmm == NULL)
1668 GOTO(out, rc = -ENOMEM);
1670 lod_prep_slave_lmv_md(slave_lmm, lmm);
1671 slave_lmv_buf.lb_buf = slave_lmm;
1672 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1674 if (!dt_try_as_dir(env, dt_object_child(dt)))
1675 GOTO(out, rc = -EINVAL);
1677 rec->rec_type = S_IFDIR;
1678 for (i = 0; i < lo->ldo_stripenr; i++) {
1679 struct dt_object *dto = lo->ldo_stripe[i];
1680 char *stripe_name = info->lti_key;
1681 struct lu_name *sname;
1682 struct linkea_data ldata = { NULL };
1683 struct lu_buf linkea_buf;
1685 rc = lod_sub_object_declare_create(env, dto, attr, NULL,
1690 if (!dt_try_as_dir(env, dto))
1691 GOTO(out, rc = -EINVAL);
1693 rc = lod_sub_object_declare_ref_add(env, dto, th);
1697 rec->rec_fid = lu_object_fid(&dto->do_lu);
1698 rc = lod_sub_object_declare_insert(env, dto,
1699 (const struct dt_rec *)rec,
1700 (const struct dt_key *)dot, th);
1704 /* master stripe FID will be put to .. */
1705 rec->rec_fid = lu_object_fid(&dt->do_lu);
1706 rc = lod_sub_object_declare_insert(env, dto,
1707 (const struct dt_rec *)rec,
1708 (const struct dt_key *)dotdot,
1713 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1714 cfs_fail_val != i) {
1715 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1717 slave_lmm->lmv_master_mdt_index =
1720 slave_lmm->lmv_master_mdt_index =
1722 rc = lod_sub_object_declare_xattr_set(env, dto,
1723 &slave_lmv_buf, XATTR_NAME_LMV, 0, th);
1728 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1730 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1731 PFID(lu_object_fid(&dto->do_lu)), i + 1);
1733 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1734 PFID(lu_object_fid(&dto->do_lu)), i);
1736 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
1737 rc = linkea_data_new(&ldata, &info->lti_linkea_buf);
1741 rc = linkea_add_buf(&ldata, sname, lu_object_fid(&dt->do_lu));
1745 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1746 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1747 rc = lod_sub_object_declare_xattr_set(env, dto, &linkea_buf,
1748 XATTR_NAME_LINK, 0, th);
1752 rec->rec_fid = lu_object_fid(&dto->do_lu);
1753 rc = lod_sub_object_declare_insert(env, dt_object_child(dt),
1754 (const struct dt_rec *)rec,
1755 (const struct dt_key *)stripe_name,
1760 rc = lod_sub_object_declare_ref_add(env, dt_object_child(dt),
1766 rc = lod_sub_object_declare_xattr_set(env, dt_object_child(dt),
1767 &lmv_buf, XATTR_NAME_LMV, 0, th);
1771 if (slave_lmm != NULL)
1772 OBD_FREE_PTR(slave_lmm);
1777 static int lod_prep_md_striped_create(const struct lu_env *env,
1778 struct dt_object *dt,
1779 struct lu_attr *attr,
1780 const struct lmv_user_md_v1 *lum,
1781 struct dt_object_format *dof,
1784 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1785 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1786 struct lod_object *lo = lod_dt_obj(dt);
1787 struct dt_object **stripe;
1796 /* The lum has been verifed in lod_verify_md_striping */
1797 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
1798 LASSERT(le32_to_cpu(lum->lum_stripe_count) > 0);
1800 stripe_count = le32_to_cpu(lum->lum_stripe_count);
1802 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_count);
1806 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
1807 if (idx_array == NULL)
1808 GOTO(out_free, rc = -ENOMEM);
1810 /* Start index will be the master MDT */
1811 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1812 idx_array[0] = master_index;
1813 for (i = 0; i < stripe_count; i++) {
1814 struct lod_tgt_desc *tgt = NULL;
1815 struct dt_object *dto;
1816 struct lu_fid fid = { 0 };
1818 struct lu_object_conf conf = { 0 };
1819 struct dt_device *tgt_dt = NULL;
1821 /* Try to find next avaible target */
1823 for (j = 0; j < lod->lod_remote_mdt_count;
1824 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
1825 bool already_allocated = false;
1828 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
1829 idx, lod->lod_remote_mdt_count + 1, i);
1830 if (idx == master_index) {
1831 /* Allocate the FID locally */
1832 rc = obd_fid_alloc(env, lod->lod_child_exp,
1836 tgt_dt = lod->lod_child;
1840 /* Find next available target */
1841 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx))
1844 if (likely(!OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
1845 /* check whether the idx already exists
1846 * in current allocated array */
1847 for (k = 0; k < i; k++) {
1848 if (idx_array[k] == idx) {
1849 already_allocated = true;
1854 if (already_allocated)
1858 /* check the status of the OSP */
1859 tgt = LTD_TGT(ltd, idx);
1863 tgt_dt = tgt->ltd_tgt;
1864 rc = dt_statfs(env, tgt_dt, NULL);
1866 /* this OSP doesn't feel well */
1871 rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
1880 /* Can not allocate more stripes */
1881 if (j == lod->lod_remote_mdt_count) {
1882 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
1883 lod2obd(lod)->obd_name, stripe_count, i - 1);
1887 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
1888 idx, i, PFID(&fid));
1890 /* Set the start index for next stripe allocation */
1891 if (i < stripe_count - 1)
1892 idx_array[i + 1] = (idx + 1) %
1893 (lod->lod_remote_mdt_count + 1);
1894 /* tgt_dt and fid must be ready after search avaible OSP
1895 * in the above loop */
1896 LASSERT(tgt_dt != NULL);
1897 LASSERT(fid_is_sane(&fid));
1898 conf.loc_flags = LOC_F_NEW;
1899 dto = dt_locate_at(env, tgt_dt, &fid,
1900 dt->do_lu.lo_dev->ld_site->ls_top_dev,
1903 GOTO(out_put, rc = PTR_ERR(dto));
1907 lo->ldo_dir_striped = 1;
1908 lo->ldo_stripe = stripe;
1909 lo->ldo_stripenr = i;
1910 lo->ldo_stripes_allocated = stripe_count;
1912 if (lo->ldo_stripenr == 0)
1913 GOTO(out_put, rc = -ENOSPC);
1915 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
1921 for (i = 0; i < stripe_count; i++)
1922 if (stripe[i] != NULL)
1923 lu_object_put(env, &stripe[i]->do_lu);
1924 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_count);
1925 lo->ldo_stripenr = 0;
1926 lo->ldo_stripes_allocated = 0;
1927 lo->ldo_stripe = NULL;
1931 if (idx_array != NULL)
1932 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
1938 * Declare create striped md object.
1940 * The function declares intention to create a striped directory. This is a
1941 * wrapper for lod_prep_md_striped_create(). The only additional functionality
1942 * is to verify pattern \a lum_buf is good. Check that function for the details.
1944 * \param[in] env execution environment
1945 * \param[in] dt object
1946 * \param[in] attr attributes to initialize the objects with
1947 * \param[in] lum_buf a pattern specifying the number of stripes and
1949 * \param[in] dof type of objects to be created
1950 * \param[in] th transaction handle
1952 * \retval 0 on success
1953 * \retval negative if failed
1956 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
1957 struct dt_object *dt,
1958 struct lu_attr *attr,
1959 const struct lu_buf *lum_buf,
1960 struct dt_object_format *dof,
1963 struct lod_object *lo = lod_dt_obj(dt);
1964 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1965 struct lmv_user_md_v1 *lum;
1969 lum = lum_buf->lb_buf;
1970 LASSERT(lum != NULL);
1972 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
1973 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
1974 (int)le32_to_cpu(lum->lum_stripe_offset));
1976 if (le32_to_cpu(lum->lum_stripe_count) == 0)
1979 rc = lod_verify_md_striping(lod, lum);
1983 /* prepare dir striped objects */
1984 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
1986 /* failed to create striping, let's reset
1987 * config so that others don't get confused */
1988 lod_object_free_striping(env, lo);
1996 * Implementation of dt_object_operations::do_declare_xattr_set.
1998 * Used with regular (non-striped) objects. Basically it
1999 * initializes the striping information and applies the
2000 * change to all the stripes.
2002 * \see dt_object_operations::do_declare_xattr_set() in the API description
2005 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2006 struct dt_object *dt,
2007 const struct lu_buf *buf,
2008 const char *name, int fl,
2011 struct dt_object *next = dt_object_child(dt);
2012 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2013 struct lod_object *lo = lod_dt_obj(dt);
2018 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2019 struct lmv_user_md_v1 *lum;
2021 LASSERT(buf != NULL && buf->lb_buf != NULL);
2023 rc = lod_verify_md_striping(d, lum);
2028 rc = lod_sub_object_declare_xattr_set(env, next, buf, name, fl, th);
2032 /* Note: Do not set LinkEA on sub-stripes, otherwise
2033 * it will confuse the fid2path process(see mdt_path_current()).
2034 * The linkEA between master and sub-stripes is set in
2035 * lod_xattr_set_lmv(). */
2036 if (strcmp(name, XATTR_NAME_LINK) == 0)
2039 /* set xattr to each stripes, if needed */
2040 rc = lod_load_striping(env, lo);
2044 if (lo->ldo_stripenr == 0)
2047 for (i = 0; i < lo->ldo_stripenr; i++) {
2048 LASSERT(lo->ldo_stripe[i]);
2050 rc = lod_sub_object_declare_xattr_set(env, lo->ldo_stripe[i],
2060 * Reset parent FID on OST object
2062 * Replace parent FID with @dt object FID, which is only called during migration
2063 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2064 * the FID is changed.
2066 * \param[in] env execution environment
2067 * \param[in] dt dt_object whose stripes's parent FID will be reset
2068 * \parem[in] th thandle
2069 * \param[in] declare if it is declare
2071 * \retval 0 if reset succeeds
2072 * \retval negative errno if reset fais
2074 static int lod_object_replace_parent_fid(const struct lu_env *env,
2075 struct dt_object *dt,
2076 struct thandle *th, bool declare)
2078 struct lod_object *lo = lod_dt_obj(dt);
2079 struct lod_thread_info *info = lod_env_info(env);
2080 struct lu_buf *buf = &info->lti_buf;
2081 struct filter_fid *ff;
2085 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2087 /* set xattr to each stripes, if needed */
2088 rc = lod_load_striping(env, lo);
2092 if (lo->ldo_stripenr == 0)
2095 if (info->lti_ea_store_size < sizeof(*ff)) {
2096 rc = lod_ea_store_resize(info, sizeof(*ff));
2101 buf->lb_buf = info->lti_ea_store;
2102 buf->lb_len = info->lti_ea_store_size;
2104 for (i = 0; i < lo->ldo_stripenr; i++) {
2105 if (lo->ldo_stripe[i] == NULL)
2108 rc = dt_xattr_get(env, lo->ldo_stripe[i], buf,
2116 fid_le_to_cpu(&ff->ff_parent, &ff->ff_parent);
2117 ff->ff_parent.f_seq = lu_object_fid(&dt->do_lu)->f_seq;
2118 ff->ff_parent.f_oid = lu_object_fid(&dt->do_lu)->f_oid;
2119 fid_cpu_to_le(&ff->ff_parent, &ff->ff_parent);
2122 rc = lod_sub_object_declare_xattr_set(env,
2123 lo->ldo_stripe[i], buf,
2125 LU_XATTR_REPLACE, th);
2127 rc = lod_sub_object_xattr_set(env, lo->ldo_stripe[i],
2128 buf, XATTR_NAME_FID,
2129 LU_XATTR_REPLACE, th);
2139 * Implementation of dt_object_operations::do_declare_xattr_set.
2141 * \see dt_object_operations::do_declare_xattr_set() in the API description
2144 * the extension to the API:
2145 * - declaring LOVEA requests striping creation
2146 * - LU_XATTR_REPLACE means layout swap
2148 static int lod_declare_xattr_set(const struct lu_env *env,
2149 struct dt_object *dt,
2150 const struct lu_buf *buf,
2151 const char *name, int fl,
2154 struct dt_object *next = dt_object_child(dt);
2155 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2161 * allow to declare predefined striping on a new (!mode) object
2162 * which is supposed to be replay of regular file creation
2163 * (when LOV setting is declared)
2164 * LU_XATTR_REPLACE is set to indicate a layout swap
2166 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
2167 if ((S_ISREG(mode) || mode == 0) && strcmp(name, XATTR_NAME_LOV) == 0 &&
2168 !(fl & LU_XATTR_REPLACE)) {
2170 * this is a request to manipulate object's striping
2172 if (dt_object_exists(dt)) {
2173 rc = dt_attr_get(env, next, attr);
2177 memset(attr, 0, sizeof(*attr));
2178 attr->la_valid = LA_TYPE | LA_MODE;
2179 attr->la_mode = S_IFREG;
2181 rc = lod_declare_striped_object(env, dt, attr, buf, th);
2182 } else if (S_ISDIR(mode)) {
2183 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
2184 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
2185 rc = lod_object_replace_parent_fid(env, dt, th, true);
2187 rc = lod_sub_object_declare_xattr_set(env, next, buf, name,
2195 * Resets cached default striping in the object.
2197 * \param[in] lo object
2199 static void lod_lov_stripe_cache_clear(struct lod_object *lo)
2201 lo->ldo_def_striping_set = 0;
2202 lo->ldo_def_striping_cached = 0;
2203 lod_object_set_pool(lo, NULL);
2204 lo->ldo_def_stripe_size = 0;
2205 lo->ldo_def_stripenr = 0;
2206 if (lo->ldo_dir_stripe != NULL)
2207 lo->ldo_dir_def_striping_cached = 0;
2211 * Apply xattr changes to the object.
2213 * Applies xattr changes to the object and the stripes if the latter exist.
2215 * \param[in] env execution environment
2216 * \param[in] dt object
2217 * \param[in] buf buffer pointing to the new value of xattr
2218 * \param[in] name name of xattr
2219 * \param[in] fl flags
2220 * \param[in] th transaction handle
2222 * \retval 0 on success
2223 * \retval negative if failed
2225 static int lod_xattr_set_internal(const struct lu_env *env,
2226 struct dt_object *dt,
2227 const struct lu_buf *buf,
2228 const char *name, int fl,
2231 struct dt_object *next = dt_object_child(dt);
2232 struct lod_object *lo = lod_dt_obj(dt);
2237 rc = lod_sub_object_xattr_set(env, next, buf, name, fl, th);
2238 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2241 /* Note: Do not set LinkEA on sub-stripes, otherwise
2242 * it will confuse the fid2path process(see mdt_path_current()).
2243 * The linkEA between master and sub-stripes is set in
2244 * lod_xattr_set_lmv(). */
2245 if (lo->ldo_stripenr == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
2248 for (i = 0; i < lo->ldo_stripenr; i++) {
2249 LASSERT(lo->ldo_stripe[i]);
2251 rc = lod_sub_object_xattr_set(env, lo->ldo_stripe[i], buf, name,
2261 * Delete an extended attribute.
2263 * Deletes specified xattr from the object and the stripes if the latter exist.
2265 * \param[in] env execution environment
2266 * \param[in] dt object
2267 * \param[in] name name of xattr
2268 * \param[in] th transaction handle
2270 * \retval 0 on success
2271 * \retval negative if failed
2273 static int lod_xattr_del_internal(const struct lu_env *env,
2274 struct dt_object *dt,
2275 const char *name, struct thandle *th)
2277 struct dt_object *next = dt_object_child(dt);
2278 struct lod_object *lo = lod_dt_obj(dt);
2283 rc = lod_sub_object_xattr_del(env, next, name, th);
2284 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2287 if (lo->ldo_stripenr == 0)
2290 for (i = 0; i < lo->ldo_stripenr; i++) {
2291 LASSERT(lo->ldo_stripe[i]);
2293 rc = lod_sub_object_xattr_del(env, lo->ldo_stripe[i], name,
2303 * Set default striping on a directory.
2305 * Sets specified striping on a directory object unless it matches the default
2306 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
2307 * EA. This striping will be used when regular file is being created in this
2310 * \param[in] env execution environment
2311 * \param[in] dt the striped object
2312 * \param[in] buf buffer with the striping
2313 * \param[in] name name of EA
2314 * \param[in] fl xattr flag (see OSD API description)
2315 * \param[in] th transaction handle
2317 * \retval 0 on success
2318 * \retval negative if failed
2320 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
2321 struct dt_object *dt,
2322 const struct lu_buf *buf,
2323 const char *name, int fl,
2326 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2327 struct lod_object *l = lod_dt_obj(dt);
2328 struct lov_user_md_v1 *lum;
2329 struct lov_user_md_v3 *v3 = NULL;
2330 const char *pool_name = NULL;
2334 /* If it is striped dir, we should clear the stripe cache for
2335 * slave stripe as well, but there are no effective way to
2336 * notify the LOD on the slave MDT, so we do not cache stripe
2337 * information for slave stripe for now. XXX*/
2338 lod_lov_stripe_cache_clear(l);
2339 LASSERT(buf != NULL && buf->lb_buf != NULL);
2342 rc = lod_verify_striping(d, buf, false);
2346 if (lum->lmm_magic == LOV_USER_MAGIC_V3) {
2348 if (v3->lmm_pool_name[0] != '\0')
2349 pool_name = v3->lmm_pool_name;
2352 /* if { size, offset, count } = { 0, -1, 0 } and no pool
2353 * (i.e. all default values specified) then delete default
2354 * striping from dir. */
2356 "set default striping: sz %u # %u offset %d %s %s\n",
2357 (unsigned)lum->lmm_stripe_size,
2358 (unsigned)lum->lmm_stripe_count,
2359 (int)lum->lmm_stripe_offset,
2360 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
2362 if (LOVEA_DELETE_VALUES(lum->lmm_stripe_size, lum->lmm_stripe_count,
2363 lum->lmm_stripe_offset, pool_name)) {
2364 rc = lod_xattr_del_internal(env, dt, name, th);
2368 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
2375 * Set default striping on a directory object.
2377 * Sets specified striping on a directory object unless it matches the default
2378 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
2379 * EA. This striping will be used when a new directory is being created in the
2382 * \param[in] env execution environment
2383 * \param[in] dt the striped object
2384 * \param[in] buf buffer with the striping
2385 * \param[in] name name of EA
2386 * \param[in] fl xattr flag (see OSD API description)
2387 * \param[in] th transaction handle
2389 * \retval 0 on success
2390 * \retval negative if failed
2392 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
2393 struct dt_object *dt,
2394 const struct lu_buf *buf,
2395 const char *name, int fl,
2398 struct lod_object *l = lod_dt_obj(dt);
2399 struct lmv_user_md_v1 *lum;
2403 LASSERT(buf != NULL && buf->lb_buf != NULL);
2406 CDEBUG(D_OTHER, "set default stripe_count # %u stripe_offset %d\n",
2407 le32_to_cpu(lum->lum_stripe_count),
2408 (int)le32_to_cpu(lum->lum_stripe_offset));
2410 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
2411 le32_to_cpu(lum->lum_stripe_offset)) &&
2412 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
2413 rc = lod_xattr_del_internal(env, dt, name, th);
2417 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
2422 /* Update default stripe cache */
2423 if (l->ldo_dir_stripe == NULL) {
2424 OBD_ALLOC_PTR(l->ldo_dir_stripe);
2425 if (l->ldo_dir_stripe == NULL)
2429 l->ldo_dir_def_striping_cached = 0;
2434 * Turn directory into a striped directory.
2436 * During replay the client sends the striping created before MDT
2437 * failure, then the layer above LOD sends this defined striping
2438 * using ->do_xattr_set(), so LOD uses this method to replay creation
2439 * of the stripes. Notice the original information for the striping
2440 * (#stripes, FIDs, etc) was transferred in declare path.
2442 * \param[in] env execution environment
2443 * \param[in] dt the striped object
2444 * \param[in] buf not used currently
2445 * \param[in] name not used currently
2446 * \param[in] fl xattr flag (see OSD API description)
2447 * \param[in] th transaction handle
2449 * \retval 0 on success
2450 * \retval negative if failed
2452 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
2453 const struct lu_buf *buf, const char *name,
2454 int fl, struct thandle *th)
2456 struct lod_object *lo = lod_dt_obj(dt);
2457 struct lod_thread_info *info = lod_env_info(env);
2458 struct lu_attr *attr = &info->lti_attr;
2459 struct dt_object_format *dof = &info->lti_format;
2460 struct lu_buf lmv_buf;
2461 struct lu_buf slave_lmv_buf;
2462 struct lmv_mds_md_v1 *lmm;
2463 struct lmv_mds_md_v1 *slave_lmm = NULL;
2464 struct dt_insert_rec *rec = &info->lti_dt_rec;
2469 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
2472 /* The stripes are supposed to be allocated in declare phase,
2473 * if there are no stripes being allocated, it will skip */
2474 if (lo->ldo_stripenr == 0)
2477 rc = dt_attr_get(env, dt_object_child(dt), attr);
2481 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME |
2482 LA_MODE | LA_UID | LA_GID | LA_TYPE;
2483 dof->dof_type = DFT_DIR;
2485 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
2488 lmm = lmv_buf.lb_buf;
2490 OBD_ALLOC_PTR(slave_lmm);
2491 if (slave_lmm == NULL)
2494 lod_prep_slave_lmv_md(slave_lmm, lmm);
2495 slave_lmv_buf.lb_buf = slave_lmm;
2496 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
2498 rec->rec_type = S_IFDIR;
2499 for (i = 0; i < lo->ldo_stripenr; i++) {
2500 struct dt_object *dto;
2501 char *stripe_name = info->lti_key;
2502 struct lu_name *sname;
2503 struct linkea_data ldata = { NULL };
2504 struct lu_buf linkea_buf;
2506 dto = lo->ldo_stripe[i];
2508 dt_write_lock(env, dto, MOR_TGT_CHILD);
2509 rc = lod_sub_object_create(env, dto, attr, NULL, dof,
2512 dt_write_unlock(env, dto);
2516 rc = lod_sub_object_ref_add(env, dto, th);
2517 dt_write_unlock(env, dto);
2521 rec->rec_fid = lu_object_fid(&dto->do_lu);
2522 rc = lod_sub_object_index_insert(env, dto,
2523 (const struct dt_rec *)rec,
2524 (const struct dt_key *)dot, th, 0);
2528 rec->rec_fid = lu_object_fid(&dt->do_lu);
2529 rc = lod_sub_object_index_insert(env, dto, (struct dt_rec *)rec,
2530 (const struct dt_key *)dotdot, th, 0);
2534 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
2535 cfs_fail_val != i) {
2536 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
2538 slave_lmm->lmv_master_mdt_index =
2541 slave_lmm->lmv_master_mdt_index =
2544 rc = lod_sub_object_xattr_set(env, dto, &slave_lmv_buf,
2545 XATTR_NAME_LMV, fl, th);
2550 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
2552 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2553 PFID(lu_object_fid(&dto->do_lu)), i + 1);
2555 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2556 PFID(lu_object_fid(&dto->do_lu)), i);
2558 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
2559 rc = linkea_data_new(&ldata, &info->lti_linkea_buf);
2563 rc = linkea_add_buf(&ldata, sname, lu_object_fid(&dt->do_lu));
2567 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
2568 linkea_buf.lb_len = ldata.ld_leh->leh_len;
2569 rc = lod_sub_object_xattr_set(env, dto, &linkea_buf,
2570 XATTR_NAME_LINK, 0, th);
2574 rec->rec_fid = lu_object_fid(&dto->do_lu);
2575 rc = lod_sub_object_index_insert(env, dt_object_child(dt),
2576 (const struct dt_rec *)rec,
2577 (const struct dt_key *)stripe_name, th, 0);
2581 rc = lod_sub_object_ref_add(env, dt_object_child(dt), th);
2586 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
2587 rc = lod_sub_object_xattr_set(env, dt_object_child(dt),
2588 &lmv_buf, XATTR_NAME_LMV, fl, th);
2590 if (slave_lmm != NULL)
2591 OBD_FREE_PTR(slave_lmm);
2597 * Helper function to declare/execute creation of a striped directory
2599 * Called in declare/create object path, prepare striping for a directory
2600 * and prepare defaults data striping for the objects to be created in
2601 * that directory. Notice the function calls "declaration" or "execution"
2602 * methods depending on \a declare param. This is a consequence of the
2603 * current approach while we don't have natural distributed transactions:
2604 * we basically execute non-local updates in the declare phase. So, the
2605 * arguments for the both phases are the same and this is the reason for
2606 * this function to exist.
2608 * \param[in] env execution environment
2609 * \param[in] dt object
2610 * \param[in] attr attributes the stripes will be created with
2611 * \param[in] dof format of stripes (see OSD API description)
2612 * \param[in] th transaction handle
2613 * \param[in] declare where to call "declare" or "execute" methods
2615 * \retval 0 on success
2616 * \retval negative if failed
2618 static int lod_dir_striping_create_internal(const struct lu_env *env,
2619 struct dt_object *dt,
2620 struct lu_attr *attr,
2621 struct dt_object_format *dof,
2625 struct lod_thread_info *info = lod_env_info(env);
2626 struct lod_object *lo = lod_dt_obj(dt);
2630 if (!LMVEA_DELETE_VALUES(lo->ldo_stripenr,
2631 lo->ldo_dir_stripe_offset)) {
2632 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
2633 int stripe_count = lo->ldo_stripenr;
2635 if (info->lti_ea_store_size < sizeof(*v1)) {
2636 rc = lod_ea_store_resize(info, sizeof(*v1));
2639 v1 = info->lti_ea_store;
2642 memset(v1, 0, sizeof(*v1));
2643 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
2644 v1->lum_stripe_count = cpu_to_le32(stripe_count);
2645 v1->lum_stripe_offset =
2646 cpu_to_le32(lo->ldo_dir_stripe_offset);
2648 info->lti_buf.lb_buf = v1;
2649 info->lti_buf.lb_len = sizeof(*v1);
2652 rc = lod_declare_xattr_set_lmv(env, dt, attr,
2653 &info->lti_buf, dof, th);
2655 rc = lod_xattr_set_lmv(env, dt, &info->lti_buf,
2656 XATTR_NAME_LMV, 0, th);
2661 /* Transfer default LMV striping from the parent */
2662 if (lo->ldo_dir_def_striping_set &&
2663 !LMVEA_DELETE_VALUES(lo->ldo_dir_def_stripenr,
2664 lo->ldo_dir_def_stripe_offset)) {
2665 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
2666 int def_stripe_count = lo->ldo_dir_def_stripenr;
2668 if (info->lti_ea_store_size < sizeof(*v1)) {
2669 rc = lod_ea_store_resize(info, sizeof(*v1));
2672 v1 = info->lti_ea_store;
2675 memset(v1, 0, sizeof(*v1));
2676 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
2677 v1->lum_stripe_count = cpu_to_le32(def_stripe_count);
2678 v1->lum_stripe_offset =
2679 cpu_to_le32(lo->ldo_dir_def_stripe_offset);
2681 cpu_to_le32(lo->ldo_dir_def_hash_type);
2683 info->lti_buf.lb_buf = v1;
2684 info->lti_buf.lb_len = sizeof(*v1);
2686 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
2687 XATTR_NAME_DEFAULT_LMV,
2690 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
2692 XATTR_NAME_DEFAULT_LMV, 0,
2698 /* Transfer default LOV striping from the parent */
2699 if (lo->ldo_def_striping_set &&
2700 !LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size,
2701 lo->ldo_def_stripenr,
2702 lo->ldo_def_stripe_offset,
2704 struct lov_user_md_v3 *v3 = info->lti_ea_store;
2706 if (info->lti_ea_store_size < sizeof(*v3)) {
2707 rc = lod_ea_store_resize(info, sizeof(*v3));
2710 v3 = info->lti_ea_store;
2713 memset(v3, 0, sizeof(*v3));
2714 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
2715 v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
2716 v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
2717 v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
2718 if (lo->ldo_pool != NULL)
2719 strlcpy(v3->lmm_pool_name, lo->ldo_pool,
2720 sizeof(v3->lmm_pool_name));
2722 info->lti_buf.lb_buf = v3;
2723 info->lti_buf.lb_len = sizeof(*v3);
2726 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
2727 XATTR_NAME_LOV, 0, th);
2729 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
2730 XATTR_NAME_LOV, 0, th);
2738 static int lod_declare_dir_striping_create(const struct lu_env *env,
2739 struct dt_object *dt,
2740 struct lu_attr *attr,
2741 struct dt_object_format *dof,
2744 return lod_dir_striping_create_internal(env, dt, attr, dof, th, true);
2747 static int lod_dir_striping_create(const struct lu_env *env,
2748 struct dt_object *dt,
2749 struct lu_attr *attr,
2750 struct dt_object_format *dof,
2753 struct lod_object *lo = lod_dt_obj(dt);
2756 rc = lod_dir_striping_create_internal(env, dt, attr, dof, th, false);
2758 lo->ldo_striping_cached = 1;
2764 * Implementation of dt_object_operations::do_xattr_set.
2766 * Sets specified extended attribute on the object. Three types of EAs are
2768 * LOV EA - stores striping for a regular file or default striping (when set
2770 * LMV EA - stores a marker for the striped directories
2771 * DMV EA - stores default directory striping
2773 * When striping is applied to a non-striped existing object (this is called
2774 * late striping), then LOD notices the caller wants to turn the object into a
2775 * striped one. The stripe objects are created and appropriate EA is set:
2776 * LOV EA storing all the stripes directly or LMV EA storing just a small header
2777 * with striping configuration.
2779 * \see dt_object_operations::do_xattr_set() in the API description for details.
2781 static int lod_xattr_set(const struct lu_env *env,
2782 struct dt_object *dt, const struct lu_buf *buf,
2783 const char *name, int fl, struct thandle *th)
2785 struct dt_object *next = dt_object_child(dt);
2789 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
2790 strcmp(name, XATTR_NAME_LMV) == 0) {
2791 struct lmv_mds_md_v1 *lmm = buf->lb_buf;
2793 if (lmm != NULL && le32_to_cpu(lmm->lmv_hash_type) &
2794 LMV_HASH_FLAG_MIGRATION)
2795 rc = lod_sub_object_xattr_set(env, next, buf, name, fl,
2798 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
2803 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
2804 strcmp(name, XATTR_NAME_LOV) == 0) {
2806 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl, th);
2808 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
2809 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2811 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
2814 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
2815 !strcmp(name, XATTR_NAME_LOV)) {
2816 /* in case of lov EA swap, just set it
2817 * if not, it is a replay so check striping match what we
2818 * already have during req replay, declare_xattr_set()
2819 * defines striping, then create() does the work */
2820 if (fl & LU_XATTR_REPLACE) {
2821 /* free stripes, then update disk */
2822 lod_object_free_striping(env, lod_dt_obj(dt));
2824 rc = lod_sub_object_xattr_set(env, next, buf, name,
2826 } else if (dt_object_remote(dt)) {
2827 /* This only happens during migration, see
2828 * mdd_migrate_create(), in which Master MDT will
2829 * create a remote target object, and only set
2830 * (migrating) stripe EA on the remote object,
2831 * and does not need creating each stripes. */
2832 rc = lod_sub_object_xattr_set(env, next, buf, name,
2835 rc = lod_striping_create(env, dt, NULL, NULL, th);
2838 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
2839 rc = lod_object_replace_parent_fid(env, dt, th, false);
2844 /* then all other xattr */
2845 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
2851 * Implementation of dt_object_operations::do_declare_xattr_del.
2853 * \see dt_object_operations::do_declare_xattr_del() in the API description
2856 static int lod_declare_xattr_del(const struct lu_env *env,
2857 struct dt_object *dt, const char *name,
2860 struct lod_object *lo = lod_dt_obj(dt);
2865 rc = lod_sub_object_declare_xattr_del(env, dt_object_child(dt),
2870 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
2873 /* set xattr to each stripes, if needed */
2874 rc = lod_load_striping(env, lo);
2878 if (lo->ldo_stripenr == 0)
2881 for (i = 0; i < lo->ldo_stripenr; i++) {
2882 LASSERT(lo->ldo_stripe[i]);
2883 rc = lod_sub_object_declare_xattr_del(env, lo->ldo_stripe[i],
2893 * Implementation of dt_object_operations::do_xattr_del.
2895 * If EA storing a regular striping is being deleted, then release
2896 * all the references to the stripe objects in core.
2898 * \see dt_object_operations::do_xattr_del() in the API description for details.
2900 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
2901 const char *name, struct thandle *th)
2903 struct dt_object *next = dt_object_child(dt);
2904 struct lod_object *lo = lod_dt_obj(dt);
2909 if (!strcmp(name, XATTR_NAME_LOV))
2910 lod_object_free_striping(env, lod_dt_obj(dt));
2912 rc = lod_sub_object_xattr_del(env, next, name, th);
2913 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2916 if (lo->ldo_stripenr == 0)
2919 for (i = 0; i < lo->ldo_stripenr; i++) {
2920 LASSERT(lo->ldo_stripe[i]);
2922 rc = lod_sub_object_xattr_del(env, lo->ldo_stripe[i], name, th);
2931 * Implementation of dt_object_operations::do_xattr_list.
2933 * \see dt_object_operations::do_xattr_list() in the API description
2936 static int lod_xattr_list(const struct lu_env *env,
2937 struct dt_object *dt, const struct lu_buf *buf)
2939 return dt_xattr_list(env, dt_object_child(dt), buf);
2943 * Initialize a pool the object belongs to.
2945 * When a striped object is being created, striping configuration
2946 * may demand the stripes are allocated on a limited set of the
2947 * targets. These limited sets are known as "pools". So we copy
2948 * a pool name into the object and later actual creation methods
2949 * (like lod_object_create()) will use this information to allocate
2950 * the stripes properly.
2952 * \param[in] o object
2953 * \param[in] pool pool name
2955 int lod_object_set_pool(struct lod_object *o, char *pool)
2960 len = strlen(o->ldo_pool);
2961 OBD_FREE(o->ldo_pool, len + 1);
2966 OBD_ALLOC(o->ldo_pool, len + 1);
2967 if (o->ldo_pool == NULL)
2969 strcpy(o->ldo_pool, pool);
2974 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
2976 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
2981 * Cache default regular striping in the object.
2983 * To improve performance of striped regular object creation we cache
2984 * default LOV striping (if it exists) in the parent directory object.
2986 * \param[in] env execution environment
2987 * \param[in] lp object
2989 * \retval 0 on success
2990 * \retval negative if failed
2992 static int lod_cache_parent_lov_striping(const struct lu_env *env,
2993 struct lod_object *lp)
2995 struct lod_thread_info *info = lod_env_info(env);
2996 struct lov_user_md_v1 *v1 = NULL;
2997 struct lov_user_md_v3 *v3 = NULL;
3001 /* called from MDD without parent being write locked,
3003 dt_write_lock(env, dt_object_child(&lp->ldo_obj), 0);
3004 rc = lod_get_lov_ea(env, lp);
3008 if (rc < (typeof(rc))sizeof(struct lov_user_md)) {
3009 /* don't lookup for non-existing or invalid striping */
3010 lp->ldo_def_striping_set = 0;
3011 lp->ldo_def_striping_cached = 1;
3012 lp->ldo_def_stripe_size = 0;
3013 lp->ldo_def_stripenr = 0;
3014 lp->ldo_def_stripe_offset = (typeof(v1->lmm_stripe_offset))(-1);
3015 GOTO(unlock, rc = 0);
3019 v1 = info->lti_ea_store;
3020 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
3021 lustre_swab_lov_user_md_v1(v1);
3022 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
3023 v3 = (struct lov_user_md_v3 *)v1;
3024 lustre_swab_lov_user_md_v3(v3);
3027 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1)
3028 GOTO(unlock, rc = 0);
3030 if (v1->lmm_pattern != LOV_PATTERN_RAID0 && v1->lmm_pattern != 0)
3031 GOTO(unlock, rc = 0);
3033 CDEBUG(D_INFO, DFID" stripe_count=%d stripe_size=%d stripe_offset=%d\n",
3034 PFID(lu_object_fid(&lp->ldo_obj.do_lu)),
3035 (int)v1->lmm_stripe_count,
3036 (int)v1->lmm_stripe_size, (int)v1->lmm_stripe_offset);
3038 lp->ldo_def_stripenr = v1->lmm_stripe_count;
3039 lp->ldo_def_stripe_size = v1->lmm_stripe_size;
3040 lp->ldo_def_stripe_offset = v1->lmm_stripe_offset;
3041 lp->ldo_def_striping_cached = 1;
3042 lp->ldo_def_striping_set = 1;
3043 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
3044 /* XXX: sanity check here */
3045 v3 = (struct lov_user_md_v3 *) v1;
3046 if (v3->lmm_pool_name[0])
3047 lod_object_set_pool(lp, v3->lmm_pool_name);
3051 dt_write_unlock(env, dt_object_child(&lp->ldo_obj));
3057 * Cache default directory striping in the object.
3059 * To improve performance of striped directory creation we cache default
3060 * directory striping (if it exists) in the parent directory object.
3062 * \param[in] env execution environment
3063 * \param[in] lp object
3065 * \retval 0 on success
3066 * \retval negative if failed
3068 static int lod_cache_parent_lmv_striping(const struct lu_env *env,
3069 struct lod_object *lp)
3071 struct lod_thread_info *info = lod_env_info(env);
3072 struct lmv_user_md_v1 *v1 = NULL;
3076 /* called from MDD without parent being write locked,
3078 dt_write_lock(env, dt_object_child(&lp->ldo_obj), 0);
3079 rc = lod_get_default_lmv_ea(env, lp);
3083 if (rc < (typeof(rc))sizeof(struct lmv_user_md)) {
3084 /* don't lookup for non-existing or invalid striping */
3085 lp->ldo_dir_def_striping_set = 0;
3086 lp->ldo_dir_def_striping_cached = 1;
3087 lp->ldo_dir_def_stripenr = 0;
3088 lp->ldo_dir_def_stripe_offset =
3089 (typeof(v1->lum_stripe_offset))(-1);
3090 lp->ldo_dir_def_hash_type = LMV_HASH_TYPE_FNV_1A_64;
3091 GOTO(unlock, rc = 0);
3095 v1 = info->lti_ea_store;
3097 lp->ldo_dir_def_stripenr = le32_to_cpu(v1->lum_stripe_count);
3098 lp->ldo_dir_def_stripe_offset = le32_to_cpu(v1->lum_stripe_offset);
3099 lp->ldo_dir_def_hash_type = le32_to_cpu(v1->lum_hash_type);
3100 lp->ldo_dir_def_striping_set = 1;
3101 lp->ldo_dir_def_striping_cached = 1;
3105 dt_write_unlock(env, dt_object_child(&lp->ldo_obj));
3110 * Cache default striping in the object.
3112 * To improve performance of striped object creation we cache default striping
3113 * (if it exists) in the parent directory object. We always cache default
3114 * striping for the regular files (stored in LOV EA) and we cache default
3115 * striping for the directories if requested by \a child_mode (when a new
3116 * directory is being created).
3118 * \param[in] env execution environment
3119 * \param[in] lp object
3120 * \param[in] child_mode new object's mode
3122 * \retval 0 on success
3123 * \retval negative if failed
3125 static int lod_cache_parent_striping(const struct lu_env *env,
3126 struct lod_object *lp,
3132 if (!lp->ldo_def_striping_cached) {
3133 /* we haven't tried to get default striping for
3134 * the directory yet, let's cache it in the object */
3135 rc = lod_cache_parent_lov_striping(env, lp);
3140 /* If the parent is on the remote MDT, we should always
3141 * try to refresh the default stripeEA cache, because we
3142 * do not cache default striping information for remote
3144 if (S_ISDIR(child_mode) && (!lp->ldo_dir_def_striping_cached ||
3145 dt_object_remote(&lp->ldo_obj)))
3146 rc = lod_cache_parent_lmv_striping(env, lp);
3152 * Implementation of dt_object_operations::do_ah_init.
3154 * This method is used to make a decision on the striping configuration for the
3155 * object being created. It can be taken from the \a parent object if it exists,
3156 * or filesystem's default. The resulting configuration (number of stripes,
3157 * stripe size/offset, pool name, etc) is stored in the object itself and will
3158 * be used by the methods like ->doo_declare_create().
3160 * \see dt_object_operations::do_ah_init() in the API description for details.
3162 static void lod_ah_init(const struct lu_env *env,
3163 struct dt_allocation_hint *ah,
3164 struct dt_object *parent,
3165 struct dt_object *child,
3168 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
3169 struct dt_object *nextp = NULL;
3170 struct dt_object *nextc;
3171 struct lod_object *lp = NULL;
3172 struct lod_object *lc;
3173 struct lov_desc *desc;
3179 if (likely(parent)) {
3180 nextp = dt_object_child(parent);
3181 lp = lod_dt_obj(parent);
3182 rc = lod_load_striping(env, lp);
3187 nextc = dt_object_child(child);
3188 lc = lod_dt_obj(child);
3190 LASSERT(lc->ldo_stripenr == 0);
3191 LASSERT(lc->ldo_stripe == NULL);
3193 if (!dt_object_exists(nextc))
3194 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
3196 if (S_ISDIR(child_mode)) {
3197 if (lc->ldo_dir_stripe == NULL) {
3198 OBD_ALLOC_PTR(lc->ldo_dir_stripe);
3199 if (lc->ldo_dir_stripe == NULL)
3203 LASSERT(lp != NULL);
3204 if (lp->ldo_dir_stripe == NULL) {
3205 OBD_ALLOC_PTR(lp->ldo_dir_stripe);
3206 if (lp->ldo_dir_stripe == NULL)
3210 rc = lod_cache_parent_striping(env, lp, child_mode);
3214 /* transfer defaults to new directory */
3215 if (lp->ldo_def_striping_set) {
3217 lod_object_set_pool(lc, lp->ldo_pool);
3218 lc->ldo_def_stripenr = lp->ldo_def_stripenr;
3219 lc->ldo_def_stripe_size = lp->ldo_def_stripe_size;
3220 lc->ldo_def_stripe_offset = lp->ldo_def_stripe_offset;
3221 lc->ldo_def_striping_set = 1;
3222 lc->ldo_def_striping_cached = 1;
3223 CDEBUG(D_OTHER, "inherite EA sz:%d off:%d nr:%d\n",
3224 (int)lc->ldo_def_stripe_size,
3225 (int)lc->ldo_def_stripe_offset,
3226 (int)lc->ldo_def_stripenr);
3229 /* transfer dir defaults to new directory */
3230 if (lp->ldo_dir_def_striping_set) {
3231 lc->ldo_dir_def_stripenr = lp->ldo_dir_def_stripenr;
3232 lc->ldo_dir_def_stripe_offset =
3233 lp->ldo_dir_def_stripe_offset;
3234 lc->ldo_dir_def_hash_type =
3235 lp->ldo_dir_def_hash_type;
3236 lc->ldo_dir_def_striping_set = 1;
3237 lc->ldo_dir_def_striping_cached = 1;
3238 CDEBUG(D_INFO, "inherit default EA nr:%d off:%d t%u\n",
3239 (int)lc->ldo_dir_def_stripenr,
3240 (int)lc->ldo_dir_def_stripe_offset,
3241 lc->ldo_dir_def_hash_type);
3244 /* It should always honour the specified stripes */
3245 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0) {
3246 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
3248 rc = lod_verify_md_striping(d, lum1);
3250 le32_to_cpu(lum1->lum_stripe_count) > 1) {
3252 le32_to_cpu(lum1->lum_stripe_count);
3253 lc->ldo_dir_stripe_offset =
3254 le32_to_cpu(lum1->lum_stripe_offset);
3255 lc->ldo_dir_hash_type =
3256 le32_to_cpu(lum1->lum_hash_type);
3257 CDEBUG(D_INFO, "set stripe EA nr:%hu off:%d\n",
3259 (int)lc->ldo_dir_stripe_offset);
3261 /* then check whether there is default stripes from parent */
3262 } else if (lp->ldo_dir_def_striping_set) {
3263 /* If there are default dir stripe from parent */
3264 lc->ldo_stripenr = lp->ldo_dir_def_stripenr;
3265 lc->ldo_dir_stripe_offset =
3266 lp->ldo_dir_def_stripe_offset;
3267 lc->ldo_dir_hash_type =
3268 lp->ldo_dir_def_hash_type;
3269 CDEBUG(D_INFO, "inherit EA nr:%hu off:%d\n",
3271 (int)lc->ldo_dir_stripe_offset);
3273 /* set default stripe for this directory */
3274 lc->ldo_stripenr = 0;
3275 lc->ldo_dir_stripe_offset = -1;
3278 /* shrink the stripe_count to the avaible MDT count */
3279 if (lc->ldo_stripenr > d->lod_remote_mdt_count + 1 &&
3280 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))
3281 lc->ldo_stripenr = d->lod_remote_mdt_count + 1;
3283 /* Directory will be striped only if stripe_count > 1, if
3284 * stripe_count == 1, let's reset stripenr = 0 to avoid
3285 * create single master stripe and also help to unify the
3286 * stripe handling of directories and files */
3287 if (lc->ldo_stripenr == 1)
3288 lc->ldo_stripenr = 0;
3290 CDEBUG(D_INFO, "final striping count:%hu, offset:%d\n",
3291 lc->ldo_stripenr, (int)lc->ldo_dir_stripe_offset);
3297 * if object is going to be striped over OSTs, transfer default
3298 * striping information to the child, so that we can use it
3299 * during declaration and creation
3301 if (!lod_object_will_be_striped(S_ISREG(child_mode),
3302 lu_object_fid(&child->do_lu)))
3305 * try from the parent
3307 if (likely(parent)) {
3308 lod_cache_parent_striping(env, lp, child_mode);
3310 lc->ldo_def_stripe_offset = LOV_OFFSET_DEFAULT;
3312 if (lp->ldo_def_striping_set) {
3314 lod_object_set_pool(lc, lp->ldo_pool);
3315 lc->ldo_stripenr = lp->ldo_def_stripenr;
3316 lc->ldo_stripe_size = lp->ldo_def_stripe_size;
3317 lc->ldo_def_stripe_offset = lp->ldo_def_stripe_offset;
3318 CDEBUG(D_OTHER, "striping from parent: #%d, sz %d %s\n",
3319 lc->ldo_stripenr, lc->ldo_stripe_size,
3320 lp->ldo_pool ? lp->ldo_pool : "");
3325 * if the parent doesn't provide with specific pattern, grab fs-wide one
3327 desc = &d->lod_desc;
3328 if (lc->ldo_stripenr == 0)
3329 lc->ldo_stripenr = desc->ld_default_stripe_count;
3330 if (lc->ldo_stripe_size == 0)
3331 lc->ldo_stripe_size = desc->ld_default_stripe_size;
3332 CDEBUG(D_OTHER, "final striping: # %d stripes, sz %d from %s\n",
3333 lc->ldo_stripenr, lc->ldo_stripe_size,
3334 lc->ldo_pool ? lc->ldo_pool : "");
3337 /* we do not cache stripe information for slave stripe, see
3338 * lod_xattr_set_lov_on_dir */
3339 if (lp != NULL && lp->ldo_dir_slave_stripe)
3340 lod_lov_stripe_cache_clear(lp);
3345 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
3347 * Size initialization on late striping.
3349 * Propagate the size of a truncated object to a deferred striping.
3350 * This function handles a special case when truncate was done on a
3351 * non-striped object and now while the striping is being created
3352 * we can't lose that size, so we have to propagate it to the stripes
3355 * \param[in] env execution environment
3356 * \param[in] dt object
3357 * \param[in] th transaction handle
3359 * \retval 0 on success
3360 * \retval negative if failed
3362 static int lod_declare_init_size(const struct lu_env *env,
3363 struct dt_object *dt, struct thandle *th)
3365 struct dt_object *next = dt_object_child(dt);
3366 struct lod_object *lo = lod_dt_obj(dt);
3367 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3368 uint64_t size, offs;
3372 /* XXX: we support the simplest (RAID0) striping so far */
3373 LASSERT(lo->ldo_stripe || lo->ldo_stripenr == 0);
3374 LASSERT(lo->ldo_stripe_size > 0);
3376 if (lo->ldo_stripenr == 0)
3379 rc = dt_attr_get(env, next, attr);
3380 LASSERT(attr->la_valid & LA_SIZE);
3384 size = attr->la_size;
3388 /* ll_do_div64(a, b) returns a % b, and a = a / b */
3389 ll_do_div64(size, (__u64) lo->ldo_stripe_size);
3390 stripe = ll_do_div64(size, (__u64) lo->ldo_stripenr);
3392 size = size * lo->ldo_stripe_size;
3393 offs = attr->la_size;
3394 size += ll_do_div64(offs, lo->ldo_stripe_size);
3396 attr->la_valid = LA_SIZE;
3397 attr->la_size = size;
3399 rc = lod_sub_object_declare_attr_set(env, lo->ldo_stripe[stripe], attr,
3406 * Declare creation of striped object.
3408 * The function declares creation stripes for a regular object. The function
3409 * also declares whether the stripes will be created with non-zero size if
3410 * previously size was set non-zero on the master object. If object \a dt is
3411 * not local, then only fully defined striping can be applied in \a lovea.
3412 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
3415 * \param[in] env execution environment
3416 * \param[in] dt object
3417 * \param[in] attr attributes the stripes will be created with
3418 * \param[in] lovea a buffer containing striping description
3419 * \param[in] th transaction handle
3421 * \retval 0 on success
3422 * \retval negative if failed
3424 int lod_declare_striped_object(const struct lu_env *env, struct dt_object *dt,
3425 struct lu_attr *attr,
3426 const struct lu_buf *lovea, struct thandle *th)
3428 struct lod_thread_info *info = lod_env_info(env);
3429 struct dt_object *next = dt_object_child(dt);
3430 struct lod_object *lo = lod_dt_obj(dt);
3434 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO)) {
3435 /* failed to create striping, let's reset
3436 * config so that others don't get confused */
3437 lod_object_free_striping(env, lo);
3438 GOTO(out, rc = -ENOMEM);
3441 if (!dt_object_remote(next)) {
3442 /* choose OST and generate appropriate objects */
3443 rc = lod_qos_prep_create(env, lo, attr, lovea, th);
3445 /* failed to create striping, let's reset
3446 * config so that others don't get confused */
3447 lod_object_free_striping(env, lo);
3452 * declare storage for striping data
3454 info->lti_buf.lb_len = lov_mds_md_size(lo->ldo_stripenr,
3455 lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1);
3457 /* LOD can not choose OST objects for remote objects, i.e.
3458 * stripes must be ready before that. Right now, it can only
3459 * happen during migrate, i.e. migrate process needs to create
3460 * remote regular file (mdd_migrate_create), then the migrate
3461 * process will provide stripeEA. */
3462 LASSERT(lovea != NULL);
3463 info->lti_buf = *lovea;
3466 rc = lod_sub_object_declare_xattr_set(env, next, &info->lti_buf,
3467 XATTR_NAME_LOV, 0, th);
3472 * if striping is created with local object's size > 0,
3473 * we have to propagate this size to specific object
3474 * the case is possible only when local object was created previously
3476 if (dt_object_exists(next))
3477 rc = lod_declare_init_size(env, dt, th);
3484 * Implementation of dt_object_operations::do_declare_create.
3486 * The method declares creation of a new object. If the object will be striped,
3487 * then helper functions are called to find FIDs for the stripes, declare
3488 * creation of the stripes and declare initialization of the striping
3489 * information to be stored in the master object.
3491 * \see dt_object_operations::do_declare_create() in the API description
3494 static int lod_declare_object_create(const struct lu_env *env,
3495 struct dt_object *dt,
3496 struct lu_attr *attr,
3497 struct dt_allocation_hint *hint,
3498 struct dt_object_format *dof,
3501 struct dt_object *next = dt_object_child(dt);
3502 struct lod_object *lo = lod_dt_obj(dt);
3511 * first of all, we declare creation of local object
3513 rc = lod_sub_object_declare_create(env, next, attr, hint, dof, th);
3517 if (dof->dof_type == DFT_SYM)
3518 dt->do_body_ops = &lod_body_lnk_ops;
3519 else if (dof->dof_type == DFT_REGULAR)
3520 dt->do_body_ops = &lod_body_ops;
3523 * it's lod_ah_init() that has decided the object will be striped
3525 if (dof->dof_type == DFT_REGULAR) {
3526 /* callers don't want stripes */
3527 /* XXX: all tricky interactions with ->ah_make_hint() decided
3528 * to use striping, then ->declare_create() behaving differently
3529 * should be cleaned */
3530 if (dof->u.dof_reg.striped == 0)
3531 lo->ldo_stripenr = 0;
3532 if (lo->ldo_stripenr > 0)
3533 rc = lod_declare_striped_object(env, dt, attr,
3535 } else if (dof->dof_type == DFT_DIR) {
3536 struct seq_server_site *ss;
3538 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
3540 /* If the parent has default stripeEA, and client
3541 * did not find it before sending create request,
3542 * then MDT will return -EREMOTE, and client will
3543 * retrieve the default stripeEA and re-create the
3546 * Note: if dah_eadata != NULL, it means creating the
3547 * striped directory with specified stripeEA, then it
3548 * should ignore the default stripeEA */
3549 if (hint != NULL && hint->dah_eadata == NULL) {
3550 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
3551 GOTO(out, rc = -EREMOTE);
3553 if (lo->ldo_dir_stripe_offset == -1) {
3554 /* child and parent should be in the same MDT */
3555 if (hint->dah_parent != NULL &&
3556 dt_object_remote(hint->dah_parent))
3557 GOTO(out, rc = -EREMOTE);
3558 } else if (lo->ldo_dir_stripe_offset !=
3560 struct lod_device *lod;
3561 struct lod_tgt_descs *ltd;
3562 struct lod_tgt_desc *tgt = NULL;
3563 bool found_mdt = false;
3566 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
3567 ltd = &lod->lod_mdt_descs;
3568 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
3569 tgt = LTD_TGT(ltd, i);
3570 if (tgt->ltd_index ==
3571 lo->ldo_dir_stripe_offset) {
3577 /* If the MDT indicated by stripe_offset can be
3578 * found, then tell client to resend the create
3579 * request to the correct MDT, otherwise return
3580 * error to client */
3582 GOTO(out, rc = -EREMOTE);
3584 GOTO(out, rc = -EINVAL);
3588 /* Orphan object (like migrating object) does not have
3589 * lod_dir_stripe, see lod_ah_init */
3590 if (lo->ldo_dir_stripe != NULL)
3591 rc = lod_declare_dir_striping_create(env, dt, attr,
3599 * Creation of a striped regular object.
3601 * The function is called to create the stripe objects for a regular
3602 * striped file. This can happen at the initial object creation or
3603 * when the caller asks LOD to do so using ->do_xattr_set() method
3604 * (so called late striping). Notice all the information are already
3605 * prepared in the form of the list of objects (ldo_stripe field).
3606 * This is done during declare phase.
3608 * \param[in] env execution environment
3609 * \param[in] dt object
3610 * \param[in] attr attributes the stripes will be created with
3611 * \param[in] dof format of stripes (see OSD API description)
3612 * \param[in] th transaction handle
3614 * \retval 0 on success
3615 * \retval negative if failed
3617 int lod_striping_create(const struct lu_env *env, struct dt_object *dt,
3618 struct lu_attr *attr, struct dt_object_format *dof,
3621 struct lod_object *lo = lod_dt_obj(dt);
3625 LASSERT(lo->ldo_striping_cached == 0);
3627 /* create all underlying objects */
3628 for (i = 0; i < lo->ldo_stripenr; i++) {
3629 LASSERT(lo->ldo_stripe[i]);
3630 rc = lod_sub_object_create(env, lo->ldo_stripe[i], attr, NULL,
3637 rc = lod_generate_and_set_lovea(env, lo, th);
3639 lo->ldo_striping_cached = 1;
3646 * Implementation of dt_object_operations::do_create.
3648 * If any of preceeding methods (like ->do_declare_create(),
3649 * ->do_ah_init(), etc) chose to create a striped object,
3650 * then this method will create the master and the stripes.
3652 * \see dt_object_operations::do_create() in the API description for details.
3654 static int lod_object_create(const struct lu_env *env, struct dt_object *dt,
3655 struct lu_attr *attr,
3656 struct dt_allocation_hint *hint,
3657 struct dt_object_format *dof, struct thandle *th)
3659 struct lod_object *lo = lod_dt_obj(dt);
3663 /* create local object */
3664 rc = lod_sub_object_create(env, dt_object_child(dt), attr, hint, dof,
3669 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
3670 lo->ldo_stripe && dof->u.dof_reg.striped != 0)
3671 rc = lod_striping_create(env, dt, attr, dof, th);
3677 * Implementation of dt_object_operations::do_declare_destroy.
3679 * If the object is a striped directory, then the function declares reference
3680 * removal from the master object (this is an index) to the stripes and declares
3681 * destroy of all the stripes. In all the cases, it declares an intention to
3682 * destroy the object itself.
3684 * \see dt_object_operations::do_declare_destroy() in the API description
3687 static int lod_declare_object_destroy(const struct lu_env *env,
3688 struct dt_object *dt,
3691 struct dt_object *next = dt_object_child(dt);
3692 struct lod_object *lo = lod_dt_obj(dt);
3693 struct lod_thread_info *info = lod_env_info(env);
3694 char *stripe_name = info->lti_key;
3699 * load striping information, notice we don't do this when object
3700 * is being initialized as we don't need this information till
3701 * few specific cases like destroy, chown
3703 rc = lod_load_striping(env, lo);
3707 /* declare destroy for all underlying objects */
3708 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
3709 rc = next->do_ops->do_index_try(env, next,
3710 &dt_directory_features);
3714 for (i = 0; i < lo->ldo_stripenr; i++) {
3715 rc = lod_sub_object_declare_ref_del(env, next, th);
3719 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3720 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
3722 rc = lod_sub_object_declare_delete(env, next,
3723 (const struct dt_key *)stripe_name, th);
3730 * we declare destroy for the local object
3732 rc = lod_sub_object_declare_destroy(env, next, th);
3736 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ))
3739 /* declare destroy all striped objects */
3740 for (i = 0; i < lo->ldo_stripenr; i++) {
3741 if (lo->ldo_stripe[i] == NULL)
3744 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
3745 rc = lod_sub_object_declare_ref_del(env,
3746 lo->ldo_stripe[i], th);
3748 rc = lod_sub_object_declare_destroy(env, lo->ldo_stripe[i],
3758 * Implementation of dt_object_operations::do_destroy.
3760 * If the object is a striped directory, then the function removes references
3761 * from the master object (this is an index) to the stripes and destroys all
3762 * the stripes. In all the cases, the function destroys the object itself.
3764 * \see dt_object_operations::do_destroy() in the API description for details.
3766 static int lod_object_destroy(const struct lu_env *env,
3767 struct dt_object *dt, struct thandle *th)
3769 struct dt_object *next = dt_object_child(dt);
3770 struct lod_object *lo = lod_dt_obj(dt);
3771 struct lod_thread_info *info = lod_env_info(env);
3772 char *stripe_name = info->lti_key;
3777 /* destroy sub-stripe of master object */
3778 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
3779 rc = next->do_ops->do_index_try(env, next,
3780 &dt_directory_features);
3784 for (i = 0; i < lo->ldo_stripenr; i++) {
3785 rc = lod_sub_object_ref_del(env, next, th);
3789 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3790 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
3793 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
3794 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
3795 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)));
3797 rc = lod_sub_object_delete(env, next,
3798 (const struct dt_key *)stripe_name, th);
3804 rc = lod_sub_object_destroy(env, next, th);
3808 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ))
3811 /* destroy all striped objects */
3812 for (i = 0; i < lo->ldo_stripenr; i++) {
3813 if (likely(lo->ldo_stripe[i] != NULL) &&
3814 (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
3815 i == cfs_fail_val)) {
3816 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
3817 dt_write_lock(env, lo->ldo_stripe[i],
3819 rc = lod_sub_object_ref_del(env,
3820 lo->ldo_stripe[i], th);
3821 dt_write_unlock(env, lo->ldo_stripe[i]);
3826 rc = lod_sub_object_destroy(env, lo->ldo_stripe[i], th);
3836 * Implementation of dt_object_operations::do_declare_ref_add.
3838 * \see dt_object_operations::do_declare_ref_add() in the API description
3841 static int lod_declare_ref_add(const struct lu_env *env,
3842 struct dt_object *dt, struct thandle *th)
3844 return lod_sub_object_declare_ref_add(env, dt_object_child(dt), th);
3848 * Implementation of dt_object_operations::do_ref_add.
3850 * \see dt_object_operations::do_ref_add() in the API description for details.
3852 static int lod_ref_add(const struct lu_env *env,
3853 struct dt_object *dt, struct thandle *th)
3855 return lod_sub_object_ref_add(env, dt_object_child(dt), th);
3859 * Implementation of dt_object_operations::do_declare_ref_del.
3861 * \see dt_object_operations::do_declare_ref_del() in the API description
3864 static int lod_declare_ref_del(const struct lu_env *env,
3865 struct dt_object *dt, struct thandle *th)
3867 return lod_sub_object_declare_ref_del(env, dt_object_child(dt), th);
3871 * Implementation of dt_object_operations::do_ref_del
3873 * \see dt_object_operations::do_ref_del() in the API description for details.
3875 static int lod_ref_del(const struct lu_env *env,
3876 struct dt_object *dt, struct thandle *th)
3878 return lod_sub_object_ref_del(env, dt_object_child(dt), th);
3882 * Implementation of dt_object_operations::do_object_sync.
3884 * \see dt_object_operations::do_object_sync() in the API description
3887 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
3888 __u64 start, __u64 end)
3890 return dt_object_sync(env, dt_object_child(dt), start, end);
3893 struct lod_slave_locks {
3895 struct lustre_handle lsl_handle[0];
3899 * Release LDLM locks on the stripes of a striped directory.
3901 * Iterates over all the locks taken on the stripe objects and
3902 * release them using ->do_object_unlock() method.
3904 * \param[in] env execution environment
3905 * \param[in] dt striped object
3906 * \param[in] einfo lock description
3907 * \param[in] policy data describing requested lock
3909 * \retval 0 on success
3910 * \retval negative if failed
3912 static int lod_object_unlock_internal(const struct lu_env *env,
3913 struct dt_object *dt,
3914 struct ldlm_enqueue_info *einfo,
3915 union ldlm_policy_data *policy)
3917 struct lod_slave_locks *slave_locks = einfo->ei_cbdata;
3922 if (slave_locks == NULL)
3925 for (i = 1; i < slave_locks->lsl_lock_count; i++) {
3926 if (lustre_handle_is_used(&slave_locks->lsl_handle[i]))
3927 ldlm_lock_decref(&slave_locks->lsl_handle[i],
3935 * Implementation of dt_object_operations::do_object_unlock.
3937 * Used to release LDLM lock(s).
3939 * \see dt_object_operations::do_object_unlock() in the API description
3942 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
3943 struct ldlm_enqueue_info *einfo,
3944 union ldlm_policy_data *policy)
3946 struct lod_object *lo = lod_dt_obj(dt);
3947 struct lod_slave_locks *slave_locks = einfo->ei_cbdata;
3948 int slave_locks_size;
3952 if (slave_locks == NULL)
3955 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3958 /* Note: for remote lock for single stripe dir, MDT will cancel
3959 * the lock by lockh directly */
3960 if (lo->ldo_stripenr <= 1 && dt_object_remote(dt_object_child(dt)))
3963 /* Only cancel slave lock for striped dir */
3964 rc = lod_object_unlock_internal(env, dt, einfo, policy);
3966 slave_locks_size = sizeof(*slave_locks) + slave_locks->lsl_lock_count *
3967 sizeof(slave_locks->lsl_handle[0]);
3968 OBD_FREE(slave_locks, slave_locks_size);
3969 einfo->ei_cbdata = NULL;
3975 * Implementation of dt_object_operations::do_object_lock.
3977 * Used to get LDLM lock on the non-striped and striped objects.
3979 * \see dt_object_operations::do_object_lock() in the API description
3982 static int lod_object_lock(const struct lu_env *env,
3983 struct dt_object *dt,
3984 struct lustre_handle *lh,
3985 struct ldlm_enqueue_info *einfo,
3986 union ldlm_policy_data *policy)
3988 struct lod_object *lo = lod_dt_obj(dt);
3991 int slave_locks_size;
3992 struct lod_slave_locks *slave_locks = NULL;
3995 /* remote object lock */
3996 if (!einfo->ei_enq_slave) {
3997 LASSERT(dt_object_remote(dt));
3998 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
4002 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4005 rc = lod_load_striping(env, lo);
4010 if (lo->ldo_stripenr <= 1)
4013 slave_locks_size = sizeof(*slave_locks) + lo->ldo_stripenr *
4014 sizeof(slave_locks->lsl_handle[0]);
4015 /* Freed in lod_object_unlock */
4016 OBD_ALLOC(slave_locks, slave_locks_size);
4017 if (slave_locks == NULL)
4019 slave_locks->lsl_lock_count = lo->ldo_stripenr;
4021 /* striped directory lock */
4022 for (i = 1; i < lo->ldo_stripenr; i++) {
4023 struct lustre_handle lockh;
4024 struct ldlm_res_id *res_id;
4026 res_id = &lod_env_info(env)->lti_res_id;
4027 fid_build_reg_res_name(lu_object_fid(&lo->ldo_stripe[i]->do_lu),
4029 einfo->ei_res_id = res_id;
4031 LASSERT(lo->ldo_stripe[i] != NULL);
4032 if (likely(dt_object_remote(lo->ldo_stripe[i]))) {
4033 rc = dt_object_lock(env, lo->ldo_stripe[i], &lockh,
4036 struct ldlm_namespace *ns = einfo->ei_namespace;
4037 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
4038 ldlm_completion_callback completion = einfo->ei_cb_cp;
4039 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
4041 /* This only happens if there are mulitple stripes
4042 * on the master MDT, i.e. except stripe0, there are
4043 * other stripes on the Master MDT as well, Only
4044 * happens in the test case right now. */
4045 LASSERT(ns != NULL);
4046 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS,
4047 policy, einfo->ei_mode,
4048 &dlmflags, blocking,
4050 NULL, 0, LVB_T_NONE,
4055 slave_locks->lsl_handle[i] = lockh;
4058 einfo->ei_cbdata = slave_locks;
4061 if (rc != 0 && slave_locks != NULL) {
4062 einfo->ei_cbdata = slave_locks;
4063 lod_object_unlock_internal(env, dt, einfo, policy);
4064 OBD_FREE(slave_locks, slave_locks_size);
4065 einfo->ei_cbdata = NULL;
4071 struct dt_object_operations lod_obj_ops = {
4072 .do_read_lock = lod_object_read_lock,
4073 .do_write_lock = lod_object_write_lock,
4074 .do_read_unlock = lod_object_read_unlock,
4075 .do_write_unlock = lod_object_write_unlock,
4076 .do_write_locked = lod_object_write_locked,
4077 .do_attr_get = lod_attr_get,
4078 .do_declare_attr_set = lod_declare_attr_set,
4079 .do_attr_set = lod_attr_set,
4080 .do_xattr_get = lod_xattr_get,
4081 .do_declare_xattr_set = lod_declare_xattr_set,
4082 .do_xattr_set = lod_xattr_set,
4083 .do_declare_xattr_del = lod_declare_xattr_del,
4084 .do_xattr_del = lod_xattr_del,
4085 .do_xattr_list = lod_xattr_list,
4086 .do_ah_init = lod_ah_init,
4087 .do_declare_create = lod_declare_object_create,
4088 .do_create = lod_object_create,
4089 .do_declare_destroy = lod_declare_object_destroy,
4090 .do_destroy = lod_object_destroy,
4091 .do_index_try = lod_index_try,
4092 .do_declare_ref_add = lod_declare_ref_add,
4093 .do_ref_add = lod_ref_add,
4094 .do_declare_ref_del = lod_declare_ref_del,
4095 .do_ref_del = lod_ref_del,
4096 .do_object_sync = lod_object_sync,
4097 .do_object_lock = lod_object_lock,
4098 .do_object_unlock = lod_object_unlock,
4102 * Implementation of dt_body_operations::dbo_read.
4104 * \see dt_body_operations::dbo_read() in the API description for details.
4106 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
4107 struct lu_buf *buf, loff_t *pos)
4109 struct dt_object *next = dt_object_child(dt);
4110 return next->do_body_ops->dbo_read(env, next, buf, pos);
4114 * Implementation of dt_body_operations::dbo_declare_write.
4116 * \see dt_body_operations::dbo_declare_write() in the API description
4119 static ssize_t lod_declare_write(const struct lu_env *env,
4120 struct dt_object *dt,
4121 const struct lu_buf *buf, loff_t pos,
4124 return lod_sub_object_declare_write(env, dt_object_child(dt), buf, pos,
4129 * Implementation of dt_body_operations::dbo_write.
4131 * \see dt_body_operations::dbo_write() in the API description for details.
4133 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
4134 const struct lu_buf *buf, loff_t *pos,
4135 struct thandle *th, int iq)
4137 return lod_sub_object_write(env, dt_object_child(dt), buf, pos, th, iq);
4140 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
4141 __u64 start, __u64 end, struct thandle *th)
4143 if (dt_object_remote(dt))
4146 return lod_sub_object_declare_punch(env, dt_object_child(dt), start,
4150 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
4151 __u64 start, __u64 end, struct thandle *th)
4153 if (dt_object_remote(dt))
4156 return lod_sub_object_punch(env, dt_object_child(dt), start, end, th);
4159 static const struct dt_body_operations lod_body_lnk_ops = {
4160 .dbo_read = lod_read,
4161 .dbo_declare_write = lod_declare_write,
4162 .dbo_write = lod_write
4165 static const struct dt_body_operations lod_body_ops = {
4166 .dbo_read = lod_read,
4167 .dbo_declare_write = lod_declare_write,
4168 .dbo_write = lod_write,
4169 .dbo_declare_punch = lod_declare_punch,
4170 .dbo_punch = lod_punch,
4174 * Implementation of lu_object_operations::loo_object_init.
4176 * The function determines the type and the index of the target device using
4177 * sequence of the object's FID. Then passes control down to the
4178 * corresponding device:
4179 * OSD for the local objects, OSP for remote
4181 * \see lu_object_operations::loo_object_init() in the API description
4184 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
4185 const struct lu_object_conf *conf)
4187 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
4188 struct lu_device *cdev = NULL;
4189 struct lu_object *cobj;
4190 struct lod_tgt_descs *ltd = NULL;
4191 struct lod_tgt_desc *tgt;
4193 int type = LU_SEQ_RANGE_ANY;
4197 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
4199 /* Note: Sometimes, it will Return EAGAIN here, see
4200 * ptrlpc_import_delay_req(), which might confuse
4201 * lu_object_find_at() and make it wait there incorrectly.
4202 * so we convert it to EIO here.*/
4209 if (type == LU_SEQ_RANGE_MDT &&
4210 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
4211 cdev = &lod->lod_child->dd_lu_dev;
4212 } else if (type == LU_SEQ_RANGE_MDT) {
4213 ltd = &lod->lod_mdt_descs;
4215 } else if (type == LU_SEQ_RANGE_OST) {
4216 ltd = &lod->lod_ost_descs;
4223 if (ltd->ltd_tgts_size > idx &&
4224 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
4225 tgt = LTD_TGT(ltd, idx);
4227 LASSERT(tgt != NULL);
4228 LASSERT(tgt->ltd_tgt != NULL);
4230 cdev = &(tgt->ltd_tgt->dd_lu_dev);
4232 lod_putref(lod, ltd);
4235 if (unlikely(cdev == NULL))
4238 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
4239 if (unlikely(cobj == NULL))
4242 lu_object_add(lo, cobj);
4249 * Release resources associated with striping.
4251 * If the object is striped (regular or directory), then release
4252 * the stripe objects references and free the ldo_stripe array.
4254 * \param[in] env execution environment
4255 * \param[in] lo object
4257 void lod_object_free_striping(const struct lu_env *env, struct lod_object *lo)
4261 if (lo->ldo_dir_stripe != NULL) {
4262 OBD_FREE_PTR(lo->ldo_dir_stripe);
4263 lo->ldo_dir_stripe = NULL;
4266 if (lo->ldo_stripe) {
4267 LASSERT(lo->ldo_stripes_allocated > 0);
4269 for (i = 0; i < lo->ldo_stripenr; i++) {
4270 if (lo->ldo_stripe[i])
4271 lu_object_put(env, &lo->ldo_stripe[i]->do_lu);
4274 i = sizeof(struct dt_object *) * lo->ldo_stripes_allocated;
4275 OBD_FREE(lo->ldo_stripe, i);
4276 lo->ldo_stripe = NULL;
4277 lo->ldo_stripes_allocated = 0;
4279 lo->ldo_striping_cached = 0;
4280 lo->ldo_stripenr = 0;
4281 lo->ldo_pattern = 0;
4285 * Implementation of lu_object_operations::loo_object_start.
4287 * \see lu_object_operations::loo_object_start() in the API description
4290 static int lod_object_start(const struct lu_env *env, struct lu_object *o)
4292 if (S_ISLNK(o->lo_header->loh_attr & S_IFMT)) {
4293 lu2lod_obj(o)->ldo_obj.do_body_ops = &lod_body_lnk_ops;
4294 } else if (S_ISREG(o->lo_header->loh_attr & S_IFMT) ||
4295 fid_is_local_file(lu_object_fid(o))) {
4296 /* Note: some local file (like last rcvd) is created
4297 * through bottom layer (OSD), so the object initialization
4298 * comes to lod, it does not set loh_attr yet, so
4299 * set do_body_ops for local file anyway */
4300 lu2lod_obj(o)->ldo_obj.do_body_ops = &lod_body_ops;
4306 * Implementation of lu_object_operations::loo_object_free.
4308 * \see lu_object_operations::loo_object_free() in the API description
4311 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
4313 struct lod_object *mo = lu2lod_obj(o);
4316 * release all underlying object pinned
4319 lod_object_free_striping(env, mo);
4321 lod_object_set_pool(mo, NULL);
4324 OBD_SLAB_FREE_PTR(mo, lod_object_kmem);
4328 * Implementation of lu_object_operations::loo_object_release.
4330 * \see lu_object_operations::loo_object_release() in the API description
4333 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
4335 /* XXX: shouldn't we release everything here in case if object
4336 * creation failed before? */
4340 * Implementation of lu_object_operations::loo_object_print.
4342 * \see lu_object_operations::loo_object_print() in the API description
4345 static int lod_object_print(const struct lu_env *env, void *cookie,
4346 lu_printer_t p, const struct lu_object *l)
4348 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
4350 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
4353 struct lu_object_operations lod_lu_obj_ops = {
4354 .loo_object_init = lod_object_init,
4355 .loo_object_start = lod_object_start,
4356 .loo_object_free = lod_object_free,
4357 .loo_object_release = lod_object_release,
4358 .loo_object_print = lod_object_print,