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, 2016, Intel Corporation.
29 * lustre/lod/lod_object.c
31 * This file contains implementations of methods for the OSD API
32 * for the Logical Object Device (LOD) layer, which provides a virtual
33 * local OSD object interface to the MDD layer, and abstracts the
34 * addressing of local (OSD) and remote (OSP) objects. The API is
35 * described in the file lustre/include/dt_object.h and in
36 * Documentation/osd-api.txt.
38 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
41 #define DEBUG_SUBSYSTEM S_MDS
44 #include <obd_class.h>
45 #include <obd_support.h>
47 #include <lustre_fid.h>
48 #include <lustre_linkea.h>
49 #include <lustre_lmv.h>
50 #include <lustre_param.h>
51 #include <lustre_swab.h>
52 #include <lustre_ver.h>
53 #include <lprocfs_status.h>
54 #include <md_object.h>
56 #include "lod_internal.h"
58 static const char dot[] = ".";
59 static const char dotdot[] = "..";
61 static const struct dt_body_operations lod_body_lnk_ops;
62 static const struct dt_body_operations lod_body_ops;
65 * Implementation of dt_index_operations::dio_lookup
67 * Used with regular (non-striped) objects.
69 * \see dt_index_operations::dio_lookup() in the API description for details.
71 static int lod_index_lookup(const struct lu_env *env, struct dt_object *dt,
72 struct dt_rec *rec, const struct dt_key *key)
74 struct dt_object *next = dt_object_child(dt);
75 return next->do_index_ops->dio_lookup(env, next, rec, key);
79 * Implementation of dt_index_operations::dio_declare_insert.
81 * Used with regular (non-striped) objects.
83 * \see dt_index_operations::dio_declare_insert() in the API description
86 static int lod_declare_index_insert(const struct lu_env *env,
88 const struct dt_rec *rec,
89 const struct dt_key *key,
92 return lod_sub_object_declare_insert(env, dt_object_child(dt),
97 * Implementation of dt_index_operations::dio_insert.
99 * Used with regular (non-striped) objects
101 * \see dt_index_operations::dio_insert() in the API description for details.
103 static int lod_index_insert(const struct lu_env *env,
104 struct dt_object *dt,
105 const struct dt_rec *rec,
106 const struct dt_key *key,
110 return lod_sub_object_index_insert(env, dt_object_child(dt), rec, key,
115 * Implementation of dt_index_operations::dio_declare_delete.
117 * Used with regular (non-striped) objects.
119 * \see dt_index_operations::dio_declare_delete() in the API description
122 static int lod_declare_index_delete(const struct lu_env *env,
123 struct dt_object *dt,
124 const struct dt_key *key,
127 return lod_sub_object_declare_delete(env, dt_object_child(dt), key,
132 * Implementation of dt_index_operations::dio_delete.
134 * Used with regular (non-striped) objects.
136 * \see dt_index_operations::dio_delete() in the API description for details.
138 static int lod_index_delete(const struct lu_env *env,
139 struct dt_object *dt,
140 const struct dt_key *key,
143 return lod_sub_object_delete(env, dt_object_child(dt), key, th);
147 * Implementation of dt_it_ops::init.
149 * Used with regular (non-striped) objects.
151 * \see dt_it_ops::init() in the API description for details.
153 static struct dt_it *lod_it_init(const struct lu_env *env,
154 struct dt_object *dt, __u32 attr)
156 struct dt_object *next = dt_object_child(dt);
157 struct lod_it *it = &lod_env_info(env)->lti_it;
158 struct dt_it *it_next;
160 it_next = next->do_index_ops->dio_it.init(env, next, attr);
164 /* currently we do not use more than one iterator per thread
165 * so we store it in thread info. if at some point we need
166 * more active iterators in a single thread, we can allocate
168 LASSERT(it->lit_obj == NULL);
170 it->lit_it = it_next;
173 return (struct dt_it *)it;
176 #define LOD_CHECK_IT(env, it) \
178 LASSERT((it)->lit_obj != NULL); \
179 LASSERT((it)->lit_it != NULL); \
183 * Implementation of dt_index_operations::dio_it.fini.
185 * Used with regular (non-striped) objects.
187 * \see dt_index_operations::dio_it.fini() in the API description for details.
189 static void lod_it_fini(const struct lu_env *env, struct dt_it *di)
191 struct lod_it *it = (struct lod_it *)di;
193 LOD_CHECK_IT(env, it);
194 it->lit_obj->do_index_ops->dio_it.fini(env, it->lit_it);
196 /* the iterator not in use any more */
202 * Implementation of dt_it_ops::get.
204 * Used with regular (non-striped) objects.
206 * \see dt_it_ops::get() in the API description for details.
208 static int lod_it_get(const struct lu_env *env, struct dt_it *di,
209 const struct dt_key *key)
211 const struct lod_it *it = (const struct lod_it *)di;
213 LOD_CHECK_IT(env, it);
214 return it->lit_obj->do_index_ops->dio_it.get(env, it->lit_it, key);
218 * Implementation of dt_it_ops::put.
220 * Used with regular (non-striped) objects.
222 * \see dt_it_ops::put() in the API description for details.
224 static void lod_it_put(const struct lu_env *env, struct dt_it *di)
226 struct lod_it *it = (struct lod_it *)di;
228 LOD_CHECK_IT(env, it);
229 return it->lit_obj->do_index_ops->dio_it.put(env, it->lit_it);
233 * Implementation of dt_it_ops::next.
235 * Used with regular (non-striped) objects
237 * \see dt_it_ops::next() in the API description for details.
239 static int lod_it_next(const struct lu_env *env, struct dt_it *di)
241 struct lod_it *it = (struct lod_it *)di;
243 LOD_CHECK_IT(env, it);
244 return it->lit_obj->do_index_ops->dio_it.next(env, it->lit_it);
248 * Implementation of dt_it_ops::key.
250 * Used with regular (non-striped) objects.
252 * \see dt_it_ops::key() in the API description for details.
254 static struct dt_key *lod_it_key(const struct lu_env *env,
255 const struct dt_it *di)
257 const struct lod_it *it = (const struct lod_it *)di;
259 LOD_CHECK_IT(env, it);
260 return it->lit_obj->do_index_ops->dio_it.key(env, it->lit_it);
264 * Implementation of dt_it_ops::key_size.
266 * Used with regular (non-striped) objects.
268 * \see dt_it_ops::key_size() in the API description for details.
270 static int lod_it_key_size(const struct lu_env *env, const struct dt_it *di)
272 struct lod_it *it = (struct lod_it *)di;
274 LOD_CHECK_IT(env, it);
275 return it->lit_obj->do_index_ops->dio_it.key_size(env, it->lit_it);
279 * Implementation of dt_it_ops::rec.
281 * Used with regular (non-striped) objects.
283 * \see dt_it_ops::rec() in the API description for details.
285 static int lod_it_rec(const struct lu_env *env, const struct dt_it *di,
286 struct dt_rec *rec, __u32 attr)
288 const struct lod_it *it = (const struct lod_it *)di;
290 LOD_CHECK_IT(env, it);
291 return it->lit_obj->do_index_ops->dio_it.rec(env, it->lit_it, rec,
296 * Implementation of dt_it_ops::rec_size.
298 * Used with regular (non-striped) objects.
300 * \see dt_it_ops::rec_size() in the API description for details.
302 static int lod_it_rec_size(const struct lu_env *env, const struct dt_it *di,
305 const struct lod_it *it = (const struct lod_it *)di;
307 LOD_CHECK_IT(env, it);
308 return it->lit_obj->do_index_ops->dio_it.rec_size(env, it->lit_it,
313 * Implementation of dt_it_ops::store.
315 * Used with regular (non-striped) objects.
317 * \see dt_it_ops::store() in the API description for details.
319 static __u64 lod_it_store(const struct lu_env *env, const struct dt_it *di)
321 const struct lod_it *it = (const struct lod_it *)di;
323 LOD_CHECK_IT(env, it);
324 return it->lit_obj->do_index_ops->dio_it.store(env, it->lit_it);
328 * Implementation of dt_it_ops::load.
330 * Used with regular (non-striped) objects.
332 * \see dt_it_ops::load() in the API description for details.
334 static int lod_it_load(const struct lu_env *env, const struct dt_it *di,
337 const struct lod_it *it = (const struct lod_it *)di;
339 LOD_CHECK_IT(env, it);
340 return it->lit_obj->do_index_ops->dio_it.load(env, it->lit_it, hash);
344 * Implementation of dt_it_ops::key_rec.
346 * Used with regular (non-striped) objects.
348 * \see dt_it_ops::rec() in the API description for details.
350 static int lod_it_key_rec(const struct lu_env *env, const struct dt_it *di,
353 const struct lod_it *it = (const struct lod_it *)di;
355 LOD_CHECK_IT(env, it);
356 return it->lit_obj->do_index_ops->dio_it.key_rec(env, it->lit_it,
360 static struct dt_index_operations lod_index_ops = {
361 .dio_lookup = lod_index_lookup,
362 .dio_declare_insert = lod_declare_index_insert,
363 .dio_insert = lod_index_insert,
364 .dio_declare_delete = lod_declare_index_delete,
365 .dio_delete = lod_index_delete,
373 .key_size = lod_it_key_size,
375 .rec_size = lod_it_rec_size,
376 .store = lod_it_store,
378 .key_rec = lod_it_key_rec,
383 * Implementation of dt_it_ops::init.
385 * Used with striped objects. Internally just initializes the iterator
386 * on the first stripe.
388 * \see dt_it_ops::init() in the API description for details.
390 static struct dt_it *lod_striped_it_init(const struct lu_env *env,
391 struct dt_object *dt, __u32 attr)
393 struct lod_object *lo = lod_dt_obj(dt);
394 struct dt_object *next;
395 struct lod_it *it = &lod_env_info(env)->lti_it;
396 struct dt_it *it_next;
399 LASSERT(lo->ldo_dir_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_dir_stripenr > 0); \
427 LASSERT((it)->lit_stripe_index < (lo)->ldo_dir_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_dir_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_dir_stripenr > 0) {
976 for (i = 0; i < lo->ldo_dir_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);
1067 int lod_obj_for_each_stripe(const struct lu_env *env, struct lod_object *lo,
1068 struct thandle *th, lod_obj_stripe_cb_t cb,
1069 struct lod_obj_stripe_cb_data *data)
1071 struct lod_layout_component *lod_comp;
1075 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1076 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1077 lod_comp = &lo->ldo_comp_entries[i];
1079 if (lod_comp->llc_stripe == NULL)
1082 LASSERT(lod_comp->llc_stripenr > 0);
1083 for (j = 0; j < lod_comp->llc_stripenr; j++) {
1084 struct dt_object *dt = lod_comp->llc_stripe[j];
1088 rc = cb(env, lo, dt, th, j, data);
1097 lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
1098 struct dt_object *dt, struct thandle *th,
1099 int stripe_idx, struct lod_obj_stripe_cb_data *data)
1101 if (data->locd_declare)
1102 return lod_sub_object_declare_attr_set(env, dt,
1103 data->locd_attr, th);
1105 return lod_sub_object_attr_set(env, dt, data->locd_attr, th);
1109 * Implementation of dt_object_operations::do_declare_attr_set.
1111 * If the object is striped, then apply the changes to all the stripes.
1113 * \see dt_object_operations::do_declare_attr_set() in the API description
1116 static int lod_declare_attr_set(const struct lu_env *env,
1117 struct dt_object *dt,
1118 const struct lu_attr *attr,
1121 struct dt_object *next = dt_object_child(dt);
1122 struct lod_object *lo = lod_dt_obj(dt);
1127 * declare setattr on the local object
1129 rc = lod_sub_object_declare_attr_set(env, next, attr, th);
1133 /* osp_declare_attr_set() ignores all attributes other than
1134 * UID, GID, PID, and size, and osp_attr_set() ignores all but
1135 * UID, GID and PID. Declaration of size attr setting happens
1136 * through lod_declare_init_size(), and not through this function.
1137 * Therefore we need not load striping unless ownership is
1138 * changing. This should save memory and (we hope) speed up
1140 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1141 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID)))
1144 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1147 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID | LA_MODE |
1148 LA_ATIME | LA_MTIME | LA_CTIME |
1153 * load striping information, notice we don't do this when object
1154 * is being initialized as we don't need this information till
1155 * few specific cases like destroy, chown
1157 rc = lod_load_striping(env, lo);
1161 if (!lod_obj_is_striped(dt))
1165 * if object is striped declare changes on the stripes
1167 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1168 LASSERT(lo->ldo_stripe);
1169 for (i = 0; i < lo->ldo_dir_stripenr; i++) {
1170 if (lo->ldo_stripe[i] == NULL)
1172 rc = lod_sub_object_declare_attr_set(env,
1173 lo->ldo_stripe[i], attr,
1179 struct lod_obj_stripe_cb_data data;
1181 data.locd_attr = attr;
1182 data.locd_declare = true;
1183 rc = lod_obj_for_each_stripe(env, lo, th,
1184 lod_obj_stripe_attr_set_cb, &data);
1190 if (!dt_object_exists(next) || dt_object_remote(next) ||
1191 !S_ISREG(attr->la_mode))
1194 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1195 rc = lod_sub_object_declare_xattr_del(env, next,
1196 XATTR_NAME_LOV, th);
1200 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) ||
1201 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1202 struct lod_thread_info *info = lod_env_info(env);
1203 struct lu_buf *buf = &info->lti_buf;
1205 buf->lb_buf = info->lti_ea_store;
1206 buf->lb_len = info->lti_ea_store_size;
1207 rc = lod_sub_object_declare_xattr_set(env, next, buf,
1209 LU_XATTR_REPLACE, th);
1216 * Implementation of dt_object_operations::do_attr_set.
1218 * If the object is striped, then apply the changes to all or subset of
1219 * the stripes depending on the object type and specific attributes.
1221 * \see dt_object_operations::do_attr_set() in the API description for details.
1223 static int lod_attr_set(const struct lu_env *env,
1224 struct dt_object *dt,
1225 const struct lu_attr *attr,
1228 struct dt_object *next = dt_object_child(dt);
1229 struct lod_object *lo = lod_dt_obj(dt);
1234 * apply changes to the local object
1236 rc = lod_sub_object_attr_set(env, next, attr, th);
1240 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1241 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID)))
1244 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1247 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE | LA_PROJID |
1248 LA_ATIME | LA_MTIME | LA_CTIME |
1253 if (!lod_obj_is_striped(dt))
1257 * if object is striped, apply changes to all the stripes
1259 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1260 LASSERT(lo->ldo_stripe);
1261 for (i = 0; i < lo->ldo_dir_stripenr; i++) {
1262 if (unlikely(lo->ldo_stripe[i] == NULL))
1265 if ((dt_object_exists(lo->ldo_stripe[i]) == 0))
1268 rc = lod_sub_object_attr_set(env, lo->ldo_stripe[i],
1274 struct lod_obj_stripe_cb_data data;
1276 data.locd_attr = attr;
1277 data.locd_declare = false;
1278 rc = lod_obj_for_each_stripe(env, lo, th,
1279 lod_obj_stripe_attr_set_cb, &data);
1285 if (!dt_object_exists(next) || dt_object_remote(next) ||
1286 !S_ISREG(attr->la_mode))
1289 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1290 rc = lod_sub_object_xattr_del(env, next, XATTR_NAME_LOV, th);
1294 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE)) {
1295 struct lod_thread_info *info = lod_env_info(env);
1296 struct lu_buf *buf = &info->lti_buf;
1297 struct ost_id *oi = &info->lti_ostid;
1298 struct lu_fid *fid = &info->lti_fid;
1299 struct lov_mds_md_v1 *lmm;
1300 struct lov_ost_data_v1 *objs;
1303 rc = lod_get_lov_ea(env, lo);
1307 buf->lb_buf = info->lti_ea_store;
1308 buf->lb_len = info->lti_ea_store_size;
1309 lmm = info->lti_ea_store;
1310 magic = le32_to_cpu(lmm->lmm_magic);
1311 if (magic == LOV_MAGIC_COMP_V1) {
1312 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1313 struct lov_comp_md_entry_v1 *lcme =
1314 &lcm->lcm_entries[0];
1316 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
1317 magic = le32_to_cpu(lmm->lmm_magic);
1320 if (magic == LOV_MAGIC_V1)
1321 objs = &(lmm->lmm_objects[0]);
1323 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1324 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1325 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1327 fid_to_ostid(fid, oi);
1328 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1330 rc = lod_sub_object_xattr_set(env, next, buf, XATTR_NAME_LOV,
1331 LU_XATTR_REPLACE, th);
1332 } else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1333 struct lod_thread_info *info = lod_env_info(env);
1334 struct lu_buf *buf = &info->lti_buf;
1335 struct lov_comp_md_v1 *lcm;
1336 struct lov_comp_md_entry_v1 *lcme;
1338 rc = lod_get_lov_ea(env, lo);
1342 buf->lb_buf = info->lti_ea_store;
1343 buf->lb_len = info->lti_ea_store_size;
1345 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
1348 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1349 lcme = &lcm->lcm_entries[0];
1350 le64_add_cpu(&lcme->lcme_extent.e_start, 1);
1351 le64_add_cpu(&lcme->lcme_extent.e_end, -1);
1353 rc = lod_sub_object_xattr_set(env, next, buf, XATTR_NAME_LOV,
1354 LU_XATTR_REPLACE, th);
1361 * Implementation of dt_object_operations::do_xattr_get.
1363 * If LOV EA is requested from the root object and it's not
1364 * found, then return default striping for the filesystem.
1366 * \see dt_object_operations::do_xattr_get() in the API description for details.
1368 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1369 struct lu_buf *buf, const char *name)
1371 struct lod_thread_info *info = lod_env_info(env);
1372 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1377 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1378 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1379 struct lmv_mds_md_v1 *lmv1;
1382 if (rc > (typeof(rc))sizeof(*lmv1))
1385 if (rc < (typeof(rc))sizeof(*lmv1))
1386 RETURN(rc = rc > 0 ? -EINVAL : rc);
1388 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1389 CLASSERT(sizeof(*lmv1) <= sizeof(info->lti_key));
1391 info->lti_buf.lb_buf = info->lti_key;
1392 info->lti_buf.lb_len = sizeof(*lmv1);
1393 rc = dt_xattr_get(env, dt_object_child(dt),
1394 &info->lti_buf, name);
1395 if (unlikely(rc != sizeof(*lmv1)))
1396 RETURN(rc = rc > 0 ? -EINVAL : rc);
1398 lmv1 = info->lti_buf.lb_buf;
1399 /* The on-disk LMV EA only contains header, but the
1400 * returned LMV EA size should contain the space for
1401 * the FIDs of all shards of the striped directory. */
1402 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1403 rc = lmv_mds_md_size(
1404 le32_to_cpu(lmv1->lmv_stripe_count),
1407 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1411 RETURN(rc = rc1 != 0 ? rc1 : rc);
1414 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1418 * XXX: Only used by lfsck
1420 * lod returns default striping on the real root of the device
1421 * this is like the root stores default striping for the whole
1422 * filesystem. historically we've been using a different approach
1423 * and store it in the config.
1425 dt_root_get(env, dev->lod_child, &info->lti_fid);
1426 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1428 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1429 struct lov_user_md *lum = buf->lb_buf;
1430 struct lov_desc *desc = &dev->lod_desc;
1432 if (buf->lb_buf == NULL) {
1434 } else if (buf->lb_len >= sizeof(*lum)) {
1435 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1436 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1437 lmm_oi_set_id(&lum->lmm_oi, 0);
1438 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1439 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1440 lum->lmm_stripe_size = cpu_to_le32(
1441 desc->ld_default_stripe_size);
1442 lum->lmm_stripe_count = cpu_to_le16(
1443 desc->ld_default_stripe_count);
1444 lum->lmm_stripe_offset = cpu_to_le16(
1445 desc->ld_default_stripe_offset);
1458 * Checks that the magic of the stripe is sane.
1460 * \param[in] lod lod device
1461 * \param[in] lum a buffer storing LMV EA to verify
1463 * \retval 0 if the EA is sane
1464 * \retval negative otherwise
1466 static int lod_verify_md_striping(struct lod_device *lod,
1467 const struct lmv_user_md_v1 *lum)
1469 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1470 CERROR("%s: invalid lmv_user_md: magic = %x, "
1471 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1472 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1473 (int)le32_to_cpu(lum->lum_stripe_offset),
1474 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1482 * Initialize LMV EA for a slave.
1484 * Initialize slave's LMV EA from the master's LMV EA.
1486 * \param[in] master_lmv a buffer containing master's EA
1487 * \param[out] slave_lmv a buffer where slave's EA will be stored
1490 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1491 const struct lmv_mds_md_v1 *master_lmv)
1493 *slave_lmv = *master_lmv;
1494 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1500 * Generate LMV EA from the object passed as \a dt. The object must have
1501 * the stripes created and initialized.
1503 * \param[in] env execution environment
1504 * \param[in] dt object
1505 * \param[out] lmv_buf buffer storing generated LMV EA
1507 * \retval 0 on success
1508 * \retval negative if failed
1510 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1511 struct lu_buf *lmv_buf)
1513 struct lod_thread_info *info = lod_env_info(env);
1514 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1515 struct lod_object *lo = lod_dt_obj(dt);
1516 struct lmv_mds_md_v1 *lmm1;
1518 int type = LU_SEQ_RANGE_ANY;
1523 LASSERT(lo->ldo_dir_striped != 0);
1524 LASSERT(lo->ldo_dir_stripenr > 0);
1525 stripe_count = lo->ldo_dir_stripenr;
1526 /* Only store the LMV EA heahder on the disk. */
1527 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1528 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1532 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1535 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1536 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1537 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1538 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1539 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1544 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1545 lmv_buf->lb_buf = info->lti_ea_store;
1546 lmv_buf->lb_len = sizeof(*lmm1);
1552 * Create in-core represenation for a striped directory.
1554 * Parse the buffer containing LMV EA and instantiate LU objects
1555 * representing the stripe objects. The pointers to the objects are
1556 * stored in ldo_stripe field of \a lo. This function is used when
1557 * we need to access an already created object (i.e. load from a disk).
1559 * \param[in] env execution environment
1560 * \param[in] lo lod object
1561 * \param[in] buf buffer containing LMV EA
1563 * \retval 0 on success
1564 * \retval negative if failed
1566 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1567 const struct lu_buf *buf)
1569 struct lod_thread_info *info = lod_env_info(env);
1570 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1571 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1572 struct dt_object **stripe;
1573 union lmv_mds_md *lmm = buf->lb_buf;
1574 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1575 struct lu_fid *fid = &info->lti_fid;
1580 if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
1583 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1584 lo->ldo_dir_slave_stripe = 1;
1588 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1591 if (le32_to_cpu(lmv1->lmv_stripe_count) < 1)
1594 LASSERT(lo->ldo_stripe == NULL);
1595 OBD_ALLOC(stripe, sizeof(stripe[0]) *
1596 (le32_to_cpu(lmv1->lmv_stripe_count)));
1600 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1601 struct dt_device *tgt_dt;
1602 struct dt_object *dto;
1603 int type = LU_SEQ_RANGE_ANY;
1606 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1607 if (!fid_is_sane(fid))
1608 GOTO(out, rc = -ESTALE);
1610 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1614 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1615 tgt_dt = lod->lod_child;
1617 struct lod_tgt_desc *tgt;
1619 tgt = LTD_TGT(ltd, idx);
1621 GOTO(out, rc = -ESTALE);
1622 tgt_dt = tgt->ltd_tgt;
1625 dto = dt_locate_at(env, tgt_dt, fid,
1626 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1629 GOTO(out, rc = PTR_ERR(dto));
1634 lo->ldo_stripe = stripe;
1635 lo->ldo_dir_stripenr = le32_to_cpu(lmv1->lmv_stripe_count);
1636 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1638 lod_object_free_striping(env, lo);
1644 * Declare create a striped directory.
1646 * Declare creating a striped directory with a given stripe pattern on the
1647 * specified MDTs. A striped directory is represented as a regular directory
1648 * - an index listing all the stripes. The stripes point back to the master
1649 * object with ".." and LinkEA. The master object gets LMV EA which
1650 * identifies it as a striped directory. The function allocates FIDs
1653 * \param[in] env execution environment
1654 * \param[in] dt object
1655 * \param[in] attr attributes to initialize the objects with
1656 * \param[in] dof type of objects to be created
1657 * \param[in] th transaction handle
1659 * \retval 0 on success
1660 * \retval negative if failed
1662 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1663 struct dt_object *dt,
1664 struct lu_attr *attr,
1665 struct dt_object_format *dof,
1668 struct lod_thread_info *info = lod_env_info(env);
1669 struct lu_buf lmv_buf;
1670 struct lu_buf slave_lmv_buf;
1671 struct lmv_mds_md_v1 *lmm;
1672 struct lmv_mds_md_v1 *slave_lmm = NULL;
1673 struct dt_insert_rec *rec = &info->lti_dt_rec;
1674 struct lod_object *lo = lod_dt_obj(dt);
1679 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1682 lmm = lmv_buf.lb_buf;
1684 OBD_ALLOC_PTR(slave_lmm);
1685 if (slave_lmm == NULL)
1686 GOTO(out, rc = -ENOMEM);
1688 lod_prep_slave_lmv_md(slave_lmm, lmm);
1689 slave_lmv_buf.lb_buf = slave_lmm;
1690 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1692 if (!dt_try_as_dir(env, dt_object_child(dt)))
1693 GOTO(out, rc = -EINVAL);
1695 rec->rec_type = S_IFDIR;
1696 for (i = 0; i < lo->ldo_dir_stripenr; i++) {
1697 struct dt_object *dto = lo->ldo_stripe[i];
1698 char *stripe_name = info->lti_key;
1699 struct lu_name *sname;
1700 struct linkea_data ldata = { NULL };
1701 struct lu_buf linkea_buf;
1703 rc = lod_sub_object_declare_create(env, dto, attr, NULL,
1708 if (!dt_try_as_dir(env, dto))
1709 GOTO(out, rc = -EINVAL);
1711 rc = lod_sub_object_declare_ref_add(env, dto, th);
1715 rec->rec_fid = lu_object_fid(&dto->do_lu);
1716 rc = lod_sub_object_declare_insert(env, dto,
1717 (const struct dt_rec *)rec,
1718 (const struct dt_key *)dot, th);
1722 /* master stripe FID will be put to .. */
1723 rec->rec_fid = lu_object_fid(&dt->do_lu);
1724 rc = lod_sub_object_declare_insert(env, dto,
1725 (const struct dt_rec *)rec,
1726 (const struct dt_key *)dotdot,
1731 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1732 cfs_fail_val != i) {
1733 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1735 slave_lmm->lmv_master_mdt_index =
1738 slave_lmm->lmv_master_mdt_index =
1740 rc = lod_sub_object_declare_xattr_set(env, dto,
1741 &slave_lmv_buf, XATTR_NAME_LMV, 0, th);
1746 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1748 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1749 PFID(lu_object_fid(&dto->do_lu)), i + 1);
1751 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1752 PFID(lu_object_fid(&dto->do_lu)), i);
1754 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
1755 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1756 sname, lu_object_fid(&dt->do_lu));
1760 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1761 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1762 rc = lod_sub_object_declare_xattr_set(env, dto, &linkea_buf,
1763 XATTR_NAME_LINK, 0, th);
1767 rec->rec_fid = lu_object_fid(&dto->do_lu);
1768 rc = lod_sub_object_declare_insert(env, dt_object_child(dt),
1769 (const struct dt_rec *)rec,
1770 (const struct dt_key *)stripe_name,
1775 rc = lod_sub_object_declare_ref_add(env, dt_object_child(dt),
1781 rc = lod_sub_object_declare_xattr_set(env, dt_object_child(dt),
1782 &lmv_buf, XATTR_NAME_LMV, 0, th);
1786 if (slave_lmm != NULL)
1787 OBD_FREE_PTR(slave_lmm);
1792 static int lod_prep_md_striped_create(const struct lu_env *env,
1793 struct dt_object *dt,
1794 struct lu_attr *attr,
1795 const struct lmv_user_md_v1 *lum,
1796 struct dt_object_format *dof,
1799 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1800 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1801 struct lod_object *lo = lod_dt_obj(dt);
1802 struct dt_object **stripe;
1811 /* The lum has been verifed in lod_verify_md_striping */
1812 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
1813 LASSERT(le32_to_cpu(lum->lum_stripe_count) > 0);
1815 stripe_count = le32_to_cpu(lum->lum_stripe_count);
1817 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
1818 if (idx_array == NULL)
1821 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_count);
1823 GOTO(out_free, rc = -ENOMEM);
1825 /* Start index must be the master MDT */
1826 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1827 idx_array[0] = master_index;
1828 for (i = 0; i < stripe_count; i++) {
1829 struct lod_tgt_desc *tgt = NULL;
1830 struct dt_object *dto;
1831 struct lu_fid fid = { 0 };
1833 struct lu_object_conf conf = { 0 };
1834 struct dt_device *tgt_dt = NULL;
1836 /* Try to find next avaible target */
1838 for (j = 0; j < lod->lod_remote_mdt_count;
1839 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
1840 bool already_allocated = false;
1843 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
1844 idx, lod->lod_remote_mdt_count + 1, i);
1846 if (likely(!OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
1847 /* check whether the idx already exists
1848 * in current allocated array */
1849 for (k = 0; k < i; k++) {
1850 if (idx_array[k] == idx) {
1851 already_allocated = true;
1856 if (already_allocated)
1860 /* Sigh, this index is not in the bitmap, let's check
1861 * next available target */
1862 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) &&
1863 idx != master_index)
1866 if (idx == master_index) {
1867 /* Allocate the FID locally */
1868 rc = obd_fid_alloc(env, lod->lod_child_exp,
1872 tgt_dt = lod->lod_child;
1876 /* check the status of the OSP */
1877 tgt = LTD_TGT(ltd, idx);
1881 tgt_dt = tgt->ltd_tgt;
1882 rc = dt_statfs(env, tgt_dt, NULL);
1884 /* this OSP doesn't feel well */
1889 rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
1898 /* Can not allocate more stripes */
1899 if (j == lod->lod_remote_mdt_count) {
1900 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
1901 lod2obd(lod)->obd_name, stripe_count, i);
1905 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
1906 idx, i, PFID(&fid));
1908 /* Set the start index for next stripe allocation */
1909 if (i < stripe_count - 1)
1910 idx_array[i + 1] = (idx + 1) %
1911 (lod->lod_remote_mdt_count + 1);
1912 /* tgt_dt and fid must be ready after search avaible OSP
1913 * in the above loop */
1914 LASSERT(tgt_dt != NULL);
1915 LASSERT(fid_is_sane(&fid));
1916 conf.loc_flags = LOC_F_NEW;
1917 dto = dt_locate_at(env, tgt_dt, &fid,
1918 dt->do_lu.lo_dev->ld_site->ls_top_dev,
1921 GOTO(out_put, rc = PTR_ERR(dto));
1925 lo->ldo_dir_striped = 1;
1926 lo->ldo_stripe = stripe;
1927 lo->ldo_dir_stripenr = i;
1928 lo->ldo_dir_stripes_allocated = stripe_count;
1930 if (lo->ldo_dir_stripenr == 0)
1931 GOTO(out_put, rc = -ENOSPC);
1933 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
1939 for (i = 0; i < stripe_count; i++)
1940 if (stripe[i] != NULL)
1941 dt_object_put(env, stripe[i]);
1942 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_count);
1943 lo->ldo_dir_stripenr = 0;
1944 lo->ldo_dir_stripes_allocated = 0;
1945 lo->ldo_stripe = NULL;
1949 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
1955 * Declare create striped md object.
1957 * The function declares intention to create a striped directory. This is a
1958 * wrapper for lod_prep_md_striped_create(). The only additional functionality
1959 * is to verify pattern \a lum_buf is good. Check that function for the details.
1961 * \param[in] env execution environment
1962 * \param[in] dt object
1963 * \param[in] attr attributes to initialize the objects with
1964 * \param[in] lum_buf a pattern specifying the number of stripes and
1966 * \param[in] dof type of objects to be created
1967 * \param[in] th transaction handle
1969 * \retval 0 on success
1970 * \retval negative if failed
1973 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
1974 struct dt_object *dt,
1975 struct lu_attr *attr,
1976 const struct lu_buf *lum_buf,
1977 struct dt_object_format *dof,
1980 struct lod_object *lo = lod_dt_obj(dt);
1981 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1982 struct lmv_user_md_v1 *lum;
1986 lum = lum_buf->lb_buf;
1987 LASSERT(lum != NULL);
1989 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
1990 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
1991 (int)le32_to_cpu(lum->lum_stripe_offset));
1993 if (le32_to_cpu(lum->lum_stripe_count) == 0)
1996 rc = lod_verify_md_striping(lod, lum);
2000 /* prepare dir striped objects */
2001 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2003 /* failed to create striping, let's reset
2004 * config so that others don't get confused */
2005 lod_object_free_striping(env, lo);
2013 * Implementation of dt_object_operations::do_declare_xattr_set.
2015 * Used with regular (non-striped) objects. Basically it
2016 * initializes the striping information and applies the
2017 * change to all the stripes.
2019 * \see dt_object_operations::do_declare_xattr_set() in the API description
2022 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2023 struct dt_object *dt,
2024 const struct lu_buf *buf,
2025 const char *name, int fl,
2028 struct dt_object *next = dt_object_child(dt);
2029 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2030 struct lod_object *lo = lod_dt_obj(dt);
2035 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2036 struct lmv_user_md_v1 *lum;
2038 LASSERT(buf != NULL && buf->lb_buf != NULL);
2040 rc = lod_verify_md_striping(d, lum);
2043 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2044 rc = lod_verify_striping(d, buf, false, 0);
2049 rc = lod_sub_object_declare_xattr_set(env, next, buf, name, fl, th);
2053 /* Note: Do not set LinkEA on sub-stripes, otherwise
2054 * it will confuse the fid2path process(see mdt_path_current()).
2055 * The linkEA between master and sub-stripes is set in
2056 * lod_xattr_set_lmv(). */
2057 if (strcmp(name, XATTR_NAME_LINK) == 0)
2060 /* set xattr to each stripes, if needed */
2061 rc = lod_load_striping(env, lo);
2065 if (lo->ldo_dir_stripenr == 0)
2068 for (i = 0; i < lo->ldo_dir_stripenr; i++) {
2069 LASSERT(lo->ldo_stripe[i]);
2071 rc = lod_sub_object_declare_xattr_set(env, lo->ldo_stripe[i],
2081 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2082 struct lod_object *lo,
2083 struct dt_object *dt, struct thandle *th,
2085 struct lod_obj_stripe_cb_data *data)
2087 struct lod_thread_info *info = lod_env_info(env);
2088 struct filter_fid *ff = &info->lti_ff;
2089 struct lu_buf *buf = &info->lti_buf;
2093 buf->lb_len = sizeof(*ff);
2094 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2101 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2102 ff->ff_parent.f_ver = stripe_idx;
2103 fid_cpu_to_le(&ff->ff_parent, &ff->ff_parent);
2104 if (data->locd_declare)
2105 rc = lod_sub_object_declare_xattr_set(env, dt, buf,
2107 LU_XATTR_REPLACE, th);
2109 rc = lod_sub_object_xattr_set(env, dt, buf, XATTR_NAME_FID,
2110 LU_XATTR_REPLACE, th);
2116 * Reset parent FID on OST object
2118 * Replace parent FID with @dt object FID, which is only called during migration
2119 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2120 * the FID is changed.
2122 * \param[in] env execution environment
2123 * \param[in] dt dt_object whose stripes's parent FID will be reset
2124 * \parem[in] th thandle
2125 * \param[in] declare if it is declare
2127 * \retval 0 if reset succeeds
2128 * \retval negative errno if reset fails
2130 static int lod_object_replace_parent_fid(const struct lu_env *env,
2131 struct dt_object *dt,
2132 struct thandle *th, bool declare)
2134 struct lod_object *lo = lod_dt_obj(dt);
2135 struct lod_thread_info *info = lod_env_info(env);
2136 struct lu_buf *buf = &info->lti_buf;
2137 struct filter_fid *ff;
2138 struct lod_obj_stripe_cb_data data;
2142 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2144 /* set xattr to each stripes, if needed */
2145 rc = lod_load_striping(env, lo);
2149 if (!lod_obj_is_striped(dt))
2152 if (info->lti_ea_store_size < sizeof(*ff)) {
2153 rc = lod_ea_store_resize(info, sizeof(*ff));
2158 buf->lb_buf = info->lti_ea_store;
2159 buf->lb_len = info->lti_ea_store_size;
2161 data.locd_declare = declare;
2162 rc = lod_obj_for_each_stripe(env, lo, th,
2163 lod_obj_stripe_replace_parent_fid_cb, &data);
2168 inline __u16 lod_comp_entry_stripecnt(struct lod_object *lo,
2169 struct lod_layout_component *entry,
2172 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2176 else if (lod_comp_inited(entry))
2177 return entry->llc_stripenr;
2178 else if ((__u16)-1 == entry->llc_stripenr)
2179 return lod->lod_desc.ld_tgt_count;
2181 return lod_get_stripecnt(lod, lo, entry->llc_stripenr);
2184 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2186 int magic, size = 0, i;
2187 struct lod_layout_component *comp_entries;
2192 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2193 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2195 lo->ldo_def_striping->lds_def_striping_is_composite;
2197 comp_cnt = lo->ldo_comp_cnt;
2198 comp_entries = lo->ldo_comp_entries;
2199 is_composite = lo->ldo_is_composite;
2203 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2205 size = sizeof(struct lov_comp_md_v1) +
2206 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2207 LASSERT(size % sizeof(__u64) == 0);
2210 for (i = 0; i < comp_cnt; i++) {
2213 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2214 stripenr = lod_comp_entry_stripecnt(lo, &comp_entries[i],
2216 size += lov_user_md_size(stripenr, magic);
2217 LASSERT(size % sizeof(__u64) == 0);
2223 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2224 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2227 * \param[in] env execution environment
2228 * \param[in] dt dt_object to add components on
2229 * \param[in] buf buffer contains components to be added
2230 * \parem[in] th thandle
2232 * \retval 0 on success
2233 * \retval negative errno on failure
2235 static int lod_declare_layout_add(const struct lu_env *env,
2236 struct dt_object *dt,
2237 const struct lu_buf *buf,
2240 struct lod_thread_info *info = lod_env_info(env);
2241 struct lod_layout_component *comp_array, *lod_comp;
2242 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2243 struct dt_object *next = dt_object_child(dt);
2244 struct lov_desc *desc = &d->lod_desc;
2245 struct lod_object *lo = lod_dt_obj(dt);
2246 struct lov_user_md_v3 *v3;
2247 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2250 int i, rc, array_cnt;
2253 LASSERT(lo->ldo_is_composite);
2255 prev_end = lo->ldo_comp_entries[lo->ldo_comp_cnt - 1].llc_extent.e_end;
2256 rc = lod_verify_striping(d, buf, false, prev_end);
2260 magic = comp_v1->lcm_magic;
2261 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2262 lustre_swab_lov_comp_md_v1(comp_v1);
2263 magic = comp_v1->lcm_magic;
2266 if (magic != LOV_USER_MAGIC_COMP_V1)
2269 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2270 OBD_ALLOC(comp_array, sizeof(*comp_array) * array_cnt);
2271 if (comp_array == NULL)
2274 memcpy(comp_array, lo->ldo_comp_entries,
2275 sizeof(*comp_array) * lo->ldo_comp_cnt);
2277 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2278 struct lov_user_md_v1 *v1;
2279 struct lu_extent *ext;
2281 v1 = (struct lov_user_md *)((char *)comp_v1 +
2282 comp_v1->lcm_entries[i].lcme_offset);
2283 ext = &comp_v1->lcm_entries[i].lcme_extent;
2285 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2286 lod_comp->llc_extent.e_start = ext->e_start;
2287 lod_comp->llc_extent.e_end = ext->e_end;
2288 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2290 lod_comp->llc_stripenr = v1->lmm_stripe_count;
2291 if (!lod_comp->llc_stripenr ||
2292 lod_comp->llc_stripenr == (__u16)-1)
2293 lod_comp->llc_stripenr = desc->ld_default_stripe_count;
2294 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2295 if (!lod_comp->llc_stripe_size)
2296 lod_comp->llc_stripe_size =
2297 desc->ld_default_stripe_size;
2299 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2300 v3 = (struct lov_user_md_v3 *) v1;
2301 if (v3->lmm_pool_name[0] != '\0') {
2302 rc = lod_set_pool(&lod_comp->llc_pool,
2310 OBD_FREE(lo->ldo_comp_entries, sizeof(*lod_comp) * lo->ldo_comp_cnt);
2311 lo->ldo_comp_entries = comp_array;
2312 lo->ldo_comp_cnt = array_cnt;
2313 /* No need to increase layout generation here, it will be increased
2314 * later when generating component ID for the new components */
2316 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2317 rc = lod_sub_object_declare_xattr_set(env, next, &info->lti_buf,
2318 XATTR_NAME_LOV, 0, th);
2325 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2326 lod_comp = &comp_array[i];
2327 if (lod_comp->llc_pool != NULL) {
2328 OBD_FREE(lod_comp->llc_pool,
2329 strlen(lod_comp->llc_pool) + 1);
2330 lod_comp->llc_pool = NULL;
2333 OBD_FREE(comp_array, sizeof(*comp_array) * array_cnt);
2338 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2339 * the '$field' can only be 'flags' now. The xattr value is binary
2340 * lov_comp_md_v1 which contains the component ID(s) and the value of
2341 * the field to be modified.
2343 * \param[in] env execution environment
2344 * \param[in] dt dt_object to be modified
2345 * \param[in] op operation string, like "set.flags"
2346 * \param[in] buf buffer contains components to be set
2347 * \parem[in] th thandle
2349 * \retval 0 on success
2350 * \retval negative errno on failure
2352 static int lod_declare_layout_set(const struct lu_env *env,
2353 struct dt_object *dt,
2354 char *op, const struct lu_buf *buf,
2357 struct lod_layout_component *lod_comp;
2358 struct lod_thread_info *info = lod_env_info(env);
2359 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2360 struct lod_object *lo = lod_dt_obj(dt);
2361 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2364 bool changed = false;
2367 if (strcmp(op, "set.flags") != 0) {
2368 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2369 lod2obd(d)->obd_name, op);
2373 magic = comp_v1->lcm_magic;
2374 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2375 lustre_swab_lov_comp_md_v1(comp_v1);
2376 magic = comp_v1->lcm_magic;
2379 if (magic != LOV_USER_MAGIC_COMP_V1)
2382 if (comp_v1->lcm_entry_count == 0) {
2383 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2384 lod2obd(d)->obd_name);
2388 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2389 id = comp_v1->lcm_entries[i].lcme_id;
2391 for (j = 0; j < lo->ldo_comp_cnt; j++) {
2392 lod_comp = &lo->ldo_comp_entries[j];
2393 if (id == lod_comp->llc_id || id == LCME_ID_ALL) {
2394 lod_comp->llc_flags =
2395 comp_v1->lcm_entries[i].lcme_flags;
2402 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
2403 lod2obd(d)->obd_name);
2407 lod_obj_inc_layout_gen(lo);
2409 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2410 rc = lod_sub_object_declare_xattr_set(env, dt, &info->lti_buf,
2411 XATTR_NAME_LOV, 0, th);
2416 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
2417 * and the xattr value is a unique component ID or a special lcme_id.
2419 * \param[in] env execution environment
2420 * \param[in] dt dt_object to be operated on
2421 * \param[in] buf buffer contains component ID or lcme_id
2422 * \parem[in] th thandle
2424 * \retval 0 on success
2425 * \retval negative errno on failure
2427 static int lod_declare_layout_del(const struct lu_env *env,
2428 struct dt_object *dt,
2429 const struct lu_buf *buf,
2432 struct lod_thread_info *info = lod_env_info(env);
2433 struct dt_object *next = dt_object_child(dt);
2434 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2435 struct lod_object *lo = lod_dt_obj(dt);
2436 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2441 LASSERT(lo->ldo_is_composite);
2443 id = *(__u32 *)buf->lb_buf;
2444 if (id == 0 || id == LCME_ID_NONE) {
2445 CDEBUG(D_LAYOUT, "%s: invalid component id %#x\n",
2446 lod2obd(d)->obd_name, id);
2450 left = lo->ldo_comp_cnt;
2454 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
2455 struct lod_layout_component *lod_comp;
2457 lod_comp = &lo->ldo_comp_entries[i];
2459 if (id <= LCME_ID_MAX && id != lod_comp->llc_id)
2461 else if (id > LCME_ID_MAX && id < LCME_ID_ALL &&
2462 !(id & lod_comp->llc_flags))
2465 if (left != (i + 1)) {
2466 CDEBUG(D_LAYOUT, "%s: this deletion will create "
2467 "a hole.\n", lod2obd(d)->obd_name);
2472 /* Mark the component as deleted */
2473 lod_comp->llc_id = LCME_ID_INVAL;
2475 /* Not instantiated component */
2476 if (lod_comp->llc_stripe == NULL)
2479 LASSERT(lod_comp->llc_stripenr > 0);
2480 for (j = 0; j < lod_comp->llc_stripenr; j++) {
2481 struct dt_object *obj = lod_comp->llc_stripe[j];
2485 rc = lod_sub_object_declare_destroy(env, obj, th);
2491 LASSERTF(left >= 0, "left = %d\n", left);
2492 if (left == lo->ldo_comp_cnt) {
2493 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
2494 lod2obd(d)->obd_name, id);
2498 memset(attr, 0, sizeof(*attr));
2499 attr->la_valid = LA_SIZE;
2500 rc = lod_sub_object_declare_attr_set(env, next, attr, th);
2505 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2506 rc = lod_sub_object_declare_xattr_set(env, next, &info->lti_buf,
2507 XATTR_NAME_LOV, 0, th);
2509 rc = lod_sub_object_declare_xattr_del(env, next, XATTR_NAME_LOV,
2517 * Declare layout add/set/del operations issued by special xattr names:
2519 * XATTR_LUSTRE_LOV.add add component(s) to existing file
2520 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
2521 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
2523 * \param[in] env execution environment
2524 * \param[in] dt object
2525 * \param[in] name name of xattr
2526 * \param[in] buf lu_buf contains xattr value
2527 * \param[in] th transaction handle
2529 * \retval 0 on success
2530 * \retval negative if failed
2532 static int lod_declare_modify_layout(const struct lu_env *env,
2533 struct dt_object *dt,
2535 const struct lu_buf *buf,
2538 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2539 struct lod_object *lo = lod_dt_obj(dt);
2540 struct dt_object *next = dt_object_child(&lo->ldo_obj);
2542 int rc, len = strlen(XATTR_LUSTRE_LOV);
2545 LASSERT(dt_object_exists(dt));
2547 if (strlen(name) <= len || name[len] != '.') {
2548 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
2549 lod2obd(d)->obd_name, name);
2554 dt_write_lock(env, next, 0);
2555 rc = lod_load_striping_locked(env, lo);
2559 /* the layout to be modified must be a composite layout */
2560 if (!lo->ldo_is_composite) {
2561 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
2562 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
2563 GOTO(unlock, rc = -EINVAL);
2566 op = (char *)name + len;
2567 if (strcmp(op, "add") == 0) {
2568 rc = lod_declare_layout_add(env, dt, buf, th);
2569 } else if (strcmp(op, "del") == 0) {
2570 rc = lod_declare_layout_del(env, dt, buf, th);
2571 } else if (strncmp(op, "set", strlen("set")) == 0) {
2572 rc = lod_declare_layout_set(env, dt, op, buf, th);
2574 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
2575 lod2obd(d)->obd_name, name);
2576 GOTO(unlock, rc = -ENOTSUPP);
2580 lod_object_free_striping(env, lo);
2581 dt_write_unlock(env, next);
2587 * Implementation of dt_object_operations::do_declare_xattr_set.
2589 * \see dt_object_operations::do_declare_xattr_set() in the API description
2592 * the extension to the API:
2593 * - declaring LOVEA requests striping creation
2594 * - LU_XATTR_REPLACE means layout swap
2596 static int lod_declare_xattr_set(const struct lu_env *env,
2597 struct dt_object *dt,
2598 const struct lu_buf *buf,
2599 const char *name, int fl,
2602 struct dt_object *next = dt_object_child(dt);
2603 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2608 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
2609 if ((S_ISREG(mode) || mode == 0) && !(fl & LU_XATTR_REPLACE) &&
2610 (strcmp(name, XATTR_NAME_LOV) == 0 ||
2611 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
2613 * this is a request to create object's striping.
2615 * allow to declare predefined striping on a new (!mode) object
2616 * which is supposed to be replay of regular file creation
2617 * (when LOV setting is declared)
2619 * LU_XATTR_REPLACE is set to indicate a layout swap
2621 if (dt_object_exists(dt)) {
2622 rc = dt_attr_get(env, next, attr);
2626 memset(attr, 0, sizeof(*attr));
2627 attr->la_valid = LA_TYPE | LA_MODE;
2628 attr->la_mode = S_IFREG;
2630 rc = lod_declare_striped_object(env, dt, attr, buf, th);
2631 } else if (S_ISREG(mode) &&
2632 strlen(name) > strlen(XATTR_LUSTRE_LOV) + 1 &&
2633 strncmp(name, XATTR_LUSTRE_LOV,
2634 strlen(XATTR_LUSTRE_LOV)) == 0) {
2636 * this is a request to modify object's striping.
2637 * add/set/del component(s).
2639 if (!dt_object_exists(dt))
2642 rc = lod_declare_modify_layout(env, dt, name, buf, th);
2643 } else if (S_ISDIR(mode)) {
2644 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
2645 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
2646 rc = lod_object_replace_parent_fid(env, dt, th, true);
2648 rc = lod_sub_object_declare_xattr_set(env, next, buf, name,
2656 * Apply xattr changes to the object.
2658 * Applies xattr changes to the object and the stripes if the latter exist.
2660 * \param[in] env execution environment
2661 * \param[in] dt object
2662 * \param[in] buf buffer pointing to the new value of xattr
2663 * \param[in] name name of xattr
2664 * \param[in] fl flags
2665 * \param[in] th transaction handle
2667 * \retval 0 on success
2668 * \retval negative if failed
2670 static int lod_xattr_set_internal(const struct lu_env *env,
2671 struct dt_object *dt,
2672 const struct lu_buf *buf,
2673 const char *name, int fl,
2676 struct dt_object *next = dt_object_child(dt);
2677 struct lod_object *lo = lod_dt_obj(dt);
2682 rc = lod_sub_object_xattr_set(env, next, buf, name, fl, th);
2683 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2686 /* Note: Do not set LinkEA on sub-stripes, otherwise
2687 * it will confuse the fid2path process(see mdt_path_current()).
2688 * The linkEA between master and sub-stripes is set in
2689 * lod_xattr_set_lmv(). */
2690 if (lo->ldo_dir_stripenr == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
2693 for (i = 0; i < lo->ldo_dir_stripenr; i++) {
2694 LASSERT(lo->ldo_stripe[i]);
2696 rc = lod_sub_object_xattr_set(env, lo->ldo_stripe[i], buf, name,
2706 * Delete an extended attribute.
2708 * Deletes specified xattr from the object and the stripes if the latter exist.
2710 * \param[in] env execution environment
2711 * \param[in] dt object
2712 * \param[in] name name of xattr
2713 * \param[in] th transaction handle
2715 * \retval 0 on success
2716 * \retval negative if failed
2718 static int lod_xattr_del_internal(const struct lu_env *env,
2719 struct dt_object *dt,
2720 const char *name, struct thandle *th)
2722 struct dt_object *next = dt_object_child(dt);
2723 struct lod_object *lo = lod_dt_obj(dt);
2728 rc = lod_sub_object_xattr_del(env, next, name, th);
2729 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2732 if (lo->ldo_dir_stripenr == 0)
2735 for (i = 0; i < lo->ldo_dir_stripenr; i++) {
2736 LASSERT(lo->ldo_stripe[i]);
2738 rc = lod_sub_object_xattr_del(env, lo->ldo_stripe[i], name,
2748 * Set default striping on a directory.
2750 * Sets specified striping on a directory object unless it matches the default
2751 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
2752 * EA. This striping will be used when regular file is being created in this
2755 * \param[in] env execution environment
2756 * \param[in] dt the striped object
2757 * \param[in] buf buffer with the striping
2758 * \param[in] name name of EA
2759 * \param[in] fl xattr flag (see OSD API description)
2760 * \param[in] th transaction handle
2762 * \retval 0 on success
2763 * \retval negative if failed
2765 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
2766 struct dt_object *dt,
2767 const struct lu_buf *buf,
2768 const char *name, int fl,
2771 struct lov_user_md_v1 *lum;
2772 struct lov_user_md_v3 *v3 = NULL;
2773 const char *pool_name = NULL;
2778 LASSERT(buf != NULL && buf->lb_buf != NULL);
2781 switch (lum->lmm_magic) {
2782 case LOV_USER_MAGIC_V3:
2784 if (v3->lmm_pool_name[0] != '\0')
2785 pool_name = v3->lmm_pool_name;
2787 case LOV_USER_MAGIC_V1:
2788 /* if { size, offset, count } = { 0, -1, 0 } and no pool
2789 * (i.e. all default values specified) then delete default
2790 * striping from dir. */
2792 "set default striping: sz %u # %u offset %d %s %s\n",
2793 (unsigned)lum->lmm_stripe_size,
2794 (unsigned)lum->lmm_stripe_count,
2795 (int)lum->lmm_stripe_offset,
2796 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
2798 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
2799 lum->lmm_stripe_count,
2800 lum->lmm_stripe_offset,
2803 case LOV_USER_MAGIC_COMP_V1:
2807 CERROR("Invalid magic %x\n", lum->lmm_magic);
2812 rc = lod_xattr_del_internal(env, dt, name, th);
2816 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
2823 * Set default striping on a directory object.
2825 * Sets specified striping on a directory object unless it matches the default
2826 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
2827 * EA. This striping will be used when a new directory is being created in the
2830 * \param[in] env execution environment
2831 * \param[in] dt the striped object
2832 * \param[in] buf buffer with the striping
2833 * \param[in] name name of EA
2834 * \param[in] fl xattr flag (see OSD API description)
2835 * \param[in] th transaction handle
2837 * \retval 0 on success
2838 * \retval negative if failed
2840 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
2841 struct dt_object *dt,
2842 const struct lu_buf *buf,
2843 const char *name, int fl,
2846 struct lmv_user_md_v1 *lum;
2850 LASSERT(buf != NULL && buf->lb_buf != NULL);
2853 CDEBUG(D_OTHER, "set default stripe_count # %u stripe_offset %d\n",
2854 le32_to_cpu(lum->lum_stripe_count),
2855 (int)le32_to_cpu(lum->lum_stripe_offset));
2857 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
2858 le32_to_cpu(lum->lum_stripe_offset)) &&
2859 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
2860 rc = lod_xattr_del_internal(env, dt, name, th);
2864 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
2873 * Turn directory into a striped directory.
2875 * During replay the client sends the striping created before MDT
2876 * failure, then the layer above LOD sends this defined striping
2877 * using ->do_xattr_set(), so LOD uses this method to replay creation
2878 * of the stripes. Notice the original information for the striping
2879 * (#stripes, FIDs, etc) was transferred in declare path.
2881 * \param[in] env execution environment
2882 * \param[in] dt the striped object
2883 * \param[in] buf not used currently
2884 * \param[in] name not used currently
2885 * \param[in] fl xattr flag (see OSD API description)
2886 * \param[in] th transaction handle
2888 * \retval 0 on success
2889 * \retval negative if failed
2891 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
2892 const struct lu_buf *buf, const char *name,
2893 int fl, struct thandle *th)
2895 struct lod_object *lo = lod_dt_obj(dt);
2896 struct lod_thread_info *info = lod_env_info(env);
2897 struct lu_attr *attr = &info->lti_attr;
2898 struct dt_object_format *dof = &info->lti_format;
2899 struct lu_buf lmv_buf;
2900 struct lu_buf slave_lmv_buf;
2901 struct lmv_mds_md_v1 *lmm;
2902 struct lmv_mds_md_v1 *slave_lmm = NULL;
2903 struct dt_insert_rec *rec = &info->lti_dt_rec;
2908 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
2911 /* The stripes are supposed to be allocated in declare phase,
2912 * if there are no stripes being allocated, it will skip */
2913 if (lo->ldo_dir_stripenr == 0)
2916 rc = dt_attr_get(env, dt_object_child(dt), attr);
2920 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME |
2921 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
2922 dof->dof_type = DFT_DIR;
2924 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
2927 lmm = lmv_buf.lb_buf;
2929 OBD_ALLOC_PTR(slave_lmm);
2930 if (slave_lmm == NULL)
2933 lod_prep_slave_lmv_md(slave_lmm, lmm);
2934 slave_lmv_buf.lb_buf = slave_lmm;
2935 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
2937 rec->rec_type = S_IFDIR;
2938 for (i = 0; i < lo->ldo_dir_stripenr; i++) {
2939 struct dt_object *dto;
2940 char *stripe_name = info->lti_key;
2941 struct lu_name *sname;
2942 struct linkea_data ldata = { NULL };
2943 struct lu_buf linkea_buf;
2945 dto = lo->ldo_stripe[i];
2947 dt_write_lock(env, dto, MOR_TGT_CHILD);
2948 rc = lod_sub_object_create(env, dto, attr, NULL, dof,
2951 dt_write_unlock(env, dto);
2955 rc = lod_sub_object_ref_add(env, dto, th);
2956 dt_write_unlock(env, dto);
2960 rec->rec_fid = lu_object_fid(&dto->do_lu);
2961 rc = lod_sub_object_index_insert(env, dto,
2962 (const struct dt_rec *)rec,
2963 (const struct dt_key *)dot, th, 0);
2967 rec->rec_fid = lu_object_fid(&dt->do_lu);
2968 rc = lod_sub_object_index_insert(env, dto, (struct dt_rec *)rec,
2969 (const struct dt_key *)dotdot, th, 0);
2973 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
2974 cfs_fail_val != i) {
2975 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
2977 slave_lmm->lmv_master_mdt_index =
2980 slave_lmm->lmv_master_mdt_index =
2983 rc = lod_sub_object_xattr_set(env, dto, &slave_lmv_buf,
2984 XATTR_NAME_LMV, fl, th);
2989 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
2991 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2992 PFID(lu_object_fid(&dto->do_lu)), i + 1);
2994 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2995 PFID(lu_object_fid(&dto->do_lu)), i);
2997 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
2998 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
2999 sname, lu_object_fid(&dt->do_lu));
3003 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
3004 linkea_buf.lb_len = ldata.ld_leh->leh_len;
3005 rc = lod_sub_object_xattr_set(env, dto, &linkea_buf,
3006 XATTR_NAME_LINK, 0, th);
3010 rec->rec_fid = lu_object_fid(&dto->do_lu);
3011 rc = lod_sub_object_index_insert(env, dt_object_child(dt),
3012 (const struct dt_rec *)rec,
3013 (const struct dt_key *)stripe_name, th, 0);
3017 rc = lod_sub_object_ref_add(env, dt_object_child(dt), th);
3022 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
3023 rc = lod_sub_object_xattr_set(env, dt_object_child(dt),
3024 &lmv_buf, XATTR_NAME_LMV, fl, th);
3026 if (slave_lmm != NULL)
3027 OBD_FREE_PTR(slave_lmm);
3033 * Helper function to declare/execute creation of a striped directory
3035 * Called in declare/create object path, prepare striping for a directory
3036 * and prepare defaults data striping for the objects to be created in
3037 * that directory. Notice the function calls "declaration" or "execution"
3038 * methods depending on \a declare param. This is a consequence of the
3039 * current approach while we don't have natural distributed transactions:
3040 * we basically execute non-local updates in the declare phase. So, the
3041 * arguments for the both phases are the same and this is the reason for
3042 * this function to exist.
3044 * \param[in] env execution environment
3045 * \param[in] dt object
3046 * \param[in] attr attributes the stripes will be created with
3047 * \param[in] dof format of stripes (see OSD API description)
3048 * \param[in] th transaction handle
3049 * \param[in] declare where to call "declare" or "execute" methods
3051 * \retval 0 on success
3052 * \retval negative if failed
3054 static int lod_dir_striping_create_internal(const struct lu_env *env,
3055 struct dt_object *dt,
3056 struct lu_attr *attr,
3057 struct dt_object_format *dof,
3061 struct lod_thread_info *info = lod_env_info(env);
3062 struct lod_object *lo = lod_dt_obj(dt);
3063 const struct lod_default_striping *lds = lo->ldo_def_striping;
3067 LASSERT(ergo(lds != NULL,
3068 lds->lds_def_striping_set ||
3069 lds->lds_dir_def_striping_set));
3071 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripenr,
3072 lo->ldo_dir_stripe_offset)) {
3073 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3074 int stripe_count = lo->ldo_dir_stripenr;
3076 if (info->lti_ea_store_size < sizeof(*v1)) {
3077 rc = lod_ea_store_resize(info, sizeof(*v1));
3080 v1 = info->lti_ea_store;
3083 memset(v1, 0, sizeof(*v1));
3084 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3085 v1->lum_stripe_count = cpu_to_le32(stripe_count);
3086 v1->lum_stripe_offset =
3087 cpu_to_le32(lo->ldo_dir_stripe_offset);
3089 info->lti_buf.lb_buf = v1;
3090 info->lti_buf.lb_len = sizeof(*v1);
3093 rc = lod_declare_xattr_set_lmv(env, dt, attr,
3094 &info->lti_buf, dof, th);
3096 rc = lod_xattr_set_lmv(env, dt, &info->lti_buf,
3097 XATTR_NAME_LMV, 0, th);
3102 /* Transfer default LMV striping from the parent */
3103 if (lds != NULL && lds->lds_dir_def_striping_set &&
3104 !LMVEA_DELETE_VALUES(lds->lds_dir_def_stripenr,
3105 lds->lds_dir_def_stripe_offset)) {
3106 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3108 if (info->lti_ea_store_size < sizeof(*v1)) {
3109 rc = lod_ea_store_resize(info, sizeof(*v1));
3112 v1 = info->lti_ea_store;
3115 memset(v1, 0, sizeof(*v1));
3116 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3117 v1->lum_stripe_count = cpu_to_le32(lds->lds_dir_def_stripenr);
3118 v1->lum_stripe_offset =
3119 cpu_to_le32(lds->lds_dir_def_stripe_offset);
3121 cpu_to_le32(lds->lds_dir_def_hash_type);
3123 info->lti_buf.lb_buf = v1;
3124 info->lti_buf.lb_len = sizeof(*v1);
3126 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3127 XATTR_NAME_DEFAULT_LMV,
3130 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
3132 XATTR_NAME_DEFAULT_LMV, 0,
3138 /* Transfer default LOV striping from the parent */
3139 if (lds != NULL && lds->lds_def_striping_set &&
3140 lds->lds_def_comp_cnt != 0) {
3141 struct lov_mds_md *lmm;
3142 int lmm_size = lod_comp_md_size(lo, true);
3144 if (info->lti_ea_store_size < lmm_size) {
3145 rc = lod_ea_store_resize(info, lmm_size);
3149 lmm = info->lti_ea_store;
3151 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
3155 info->lti_buf.lb_buf = lmm;
3156 info->lti_buf.lb_len = lmm_size;
3159 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3160 XATTR_NAME_LOV, 0, th);
3162 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
3163 XATTR_NAME_LOV, 0, th);
3171 static int lod_declare_dir_striping_create(const struct lu_env *env,
3172 struct dt_object *dt,
3173 struct lu_attr *attr,
3174 struct dt_object_format *dof,
3177 return lod_dir_striping_create_internal(env, dt, attr, dof, th, true);
3180 static int lod_dir_striping_create(const struct lu_env *env,
3181 struct dt_object *dt,
3182 struct lu_attr *attr,
3183 struct dt_object_format *dof,
3186 return lod_dir_striping_create_internal(env, dt, attr, dof, th, false);
3190 * Make LOV EA for striped object.
3192 * Generate striping information and store it in the LOV EA of the given
3193 * object. The caller must ensure nobody else is calling the function
3194 * against the object concurrently. The transaction must be started.
3195 * FLDB service must be running as well; it's used to map FID to the target,
3196 * which is stored in LOV EA.
3198 * \param[in] env execution environment for this thread
3199 * \param[in] lo LOD object
3200 * \param[in] th transaction handle
3202 * \retval 0 if LOV EA is stored successfully
3203 * \retval negative error number on failure
3205 static int lod_generate_and_set_lovea(const struct lu_env *env,
3206 struct lod_object *lo,
3209 struct lod_thread_info *info = lod_env_info(env);
3210 struct dt_object *next = dt_object_child(&lo->ldo_obj);
3211 struct lov_mds_md_v1 *lmm;
3217 if (lo->ldo_comp_cnt == 0) {
3218 lod_object_free_striping(env, lo);
3219 rc = lod_sub_object_xattr_del(env, next, XATTR_NAME_LOV, th);
3223 lmm_size = lod_comp_md_size(lo, false);
3224 if (info->lti_ea_store_size < lmm_size) {
3225 rc = lod_ea_store_resize(info, lmm_size);
3229 lmm = info->lti_ea_store;
3231 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
3235 info->lti_buf.lb_buf = lmm;
3236 info->lti_buf.lb_len = lmm_size;
3237 rc = lod_sub_object_xattr_set(env, next, &info->lti_buf,
3238 XATTR_NAME_LOV, 0, th);
3243 * Delete layout component(s)
3245 * \param[in] env execution environment for this thread
3246 * \param[in] dt object
3247 * \param[in] th transaction handle
3249 * \retval 0 on success
3250 * \retval negative error number on failure
3252 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
3255 struct lod_layout_component *lod_comp;
3256 struct lod_object *lo = lod_dt_obj(dt);
3257 struct dt_object *next = dt_object_child(dt);
3258 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3261 LASSERT(lo->ldo_is_composite);
3262 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
3264 left = lo->ldo_comp_cnt;
3265 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3266 lod_comp = &lo->ldo_comp_entries[i];
3268 if (lod_comp->llc_id != LCME_ID_INVAL)
3272 /* Not instantiated component */
3273 if (lod_comp->llc_stripe == NULL)
3276 LASSERT(lod_comp->llc_stripenr > 0);
3277 for (j = 0; j < lod_comp->llc_stripenr; j++) {
3278 struct dt_object *obj = lod_comp->llc_stripe[j];
3282 rc = lod_sub_object_destroy(env, obj, th);
3286 lu_object_put(env, &obj->do_lu);
3287 lod_comp->llc_stripe[j] = NULL;
3289 OBD_FREE(lod_comp->llc_stripe, sizeof(struct dt_object *) *
3290 lod_comp->llc_stripes_allocated);
3291 lod_comp->llc_stripe = NULL;
3292 lod_comp->llc_stripes_allocated = 0;
3293 lod_obj_set_pool(lo, i, NULL);
3294 if (lod_comp->llc_ostlist.op_array) {
3295 OBD_FREE(lod_comp->llc_ostlist.op_array,
3296 lod_comp->llc_ostlist.op_size);
3297 lod_comp->llc_ostlist.op_array = NULL;
3298 lod_comp->llc_ostlist.op_size = 0;
3302 LASSERTF(left >= 0 && left < lo->ldo_comp_cnt, "left = %d\n", left);
3304 struct lod_layout_component *comp_array;
3306 OBD_ALLOC(comp_array, sizeof(*comp_array) * left);
3307 if (comp_array == NULL)
3308 GOTO(out, rc = -ENOMEM);
3310 memcpy(&comp_array[0], &lo->ldo_comp_entries[0],
3311 sizeof(*comp_array) * left);
3313 OBD_FREE(lo->ldo_comp_entries,
3314 sizeof(*comp_array) * lo->ldo_comp_cnt);
3315 lo->ldo_comp_entries = comp_array;
3316 lo->ldo_comp_cnt = left;
3317 lod_obj_inc_layout_gen(lo);
3319 lod_free_comp_entries(lo);
3322 LASSERT(dt_object_exists(dt));
3323 rc = dt_attr_get(env, next, attr);
3327 if (attr->la_size > 0) {
3329 attr->la_valid = LA_SIZE;
3330 rc = lod_sub_object_attr_set(env, next, attr, th);
3335 rc = lod_generate_and_set_lovea(env, lo, th);
3339 lod_object_free_striping(env, lo);
3344 * Implementation of dt_object_operations::do_xattr_set.
3346 * Sets specified extended attribute on the object. Three types of EAs are
3348 * LOV EA - stores striping for a regular file or default striping (when set
3350 * LMV EA - stores a marker for the striped directories
3351 * DMV EA - stores default directory striping
3353 * When striping is applied to a non-striped existing object (this is called
3354 * late striping), then LOD notices the caller wants to turn the object into a
3355 * striped one. The stripe objects are created and appropriate EA is set:
3356 * LOV EA storing all the stripes directly or LMV EA storing just a small header
3357 * with striping configuration.
3359 * \see dt_object_operations::do_xattr_set() in the API description for details.
3361 static int lod_xattr_set(const struct lu_env *env,
3362 struct dt_object *dt, const struct lu_buf *buf,
3363 const char *name, int fl, struct thandle *th)
3365 struct dt_object *next = dt_object_child(dt);
3369 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
3370 strcmp(name, XATTR_NAME_LMV) == 0) {
3371 struct lmv_mds_md_v1 *lmm = buf->lb_buf;
3373 if (lmm != NULL && le32_to_cpu(lmm->lmv_hash_type) &
3374 LMV_HASH_FLAG_MIGRATION)
3375 rc = lod_sub_object_xattr_set(env, next, buf, name, fl,
3378 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
3383 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
3384 strcmp(name, XATTR_NAME_LOV) == 0) {
3386 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl, th);
3388 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
3389 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
3391 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
3394 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
3395 (!strcmp(name, XATTR_NAME_LOV) ||
3396 !strncmp(name, XATTR_LUSTRE_LOV,
3397 strlen(XATTR_LUSTRE_LOV)))) {
3398 /* in case of lov EA swap, just set it
3399 * if not, it is a replay so check striping match what we
3400 * already have during req replay, declare_xattr_set()
3401 * defines striping, then create() does the work */
3402 if (fl & LU_XATTR_REPLACE) {
3403 /* free stripes, then update disk */
3404 lod_object_free_striping(env, lod_dt_obj(dt));
3406 rc = lod_sub_object_xattr_set(env, next, buf, name,
3408 } else if (dt_object_remote(dt)) {
3409 /* This only happens during migration, see
3410 * mdd_migrate_create(), in which Master MDT will
3411 * create a remote target object, and only set
3412 * (migrating) stripe EA on the remote object,
3413 * and does not need creating each stripes. */
3414 rc = lod_sub_object_xattr_set(env, next, buf, name,
3416 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
3417 /* delete component(s) */
3418 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
3419 rc = lod_layout_del(env, dt, th);
3422 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
3423 * it's going to create create file with specified
3424 * component(s), the striping must have not being
3425 * cached in this case;
3427 * Otherwise, it's going to add/change component(s) to
3428 * an existing file, the striping must have been cached
3431 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
3432 !strcmp(name, XATTR_NAME_LOV),
3433 !lod_dt_obj(dt)->ldo_comp_cached));
3435 rc = lod_striping_create(env, dt, NULL, NULL, th);
3438 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3439 rc = lod_object_replace_parent_fid(env, dt, th, false);
3444 /* then all other xattr */
3445 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3451 * Implementation of dt_object_operations::do_declare_xattr_del.
3453 * \see dt_object_operations::do_declare_xattr_del() in the API description
3456 static int lod_declare_xattr_del(const struct lu_env *env,
3457 struct dt_object *dt, const char *name,
3460 struct lod_object *lo = lod_dt_obj(dt);
3465 rc = lod_sub_object_declare_xattr_del(env, dt_object_child(dt),
3470 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3473 /* set xattr to each stripes, if needed */
3474 rc = lod_load_striping(env, lo);
3478 if (lo->ldo_dir_stripenr == 0)
3481 for (i = 0; i < lo->ldo_dir_stripenr; i++) {
3482 LASSERT(lo->ldo_stripe[i]);
3483 rc = lod_sub_object_declare_xattr_del(env, lo->ldo_stripe[i],
3493 * Implementation of dt_object_operations::do_xattr_del.
3495 * If EA storing a regular striping is being deleted, then release
3496 * all the references to the stripe objects in core.
3498 * \see dt_object_operations::do_xattr_del() in the API description for details.
3500 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
3501 const char *name, struct thandle *th)
3503 struct dt_object *next = dt_object_child(dt);
3504 struct lod_object *lo = lod_dt_obj(dt);
3509 if (!strcmp(name, XATTR_NAME_LOV))
3510 lod_object_free_striping(env, lod_dt_obj(dt));
3512 rc = lod_sub_object_xattr_del(env, next, name, th);
3513 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3516 if (lo->ldo_dir_stripenr == 0)
3519 for (i = 0; i < lo->ldo_dir_stripenr; i++) {
3520 LASSERT(lo->ldo_stripe[i]);
3522 rc = lod_sub_object_xattr_del(env, lo->ldo_stripe[i], name, th);
3531 * Implementation of dt_object_operations::do_xattr_list.
3533 * \see dt_object_operations::do_xattr_list() in the API description
3536 static int lod_xattr_list(const struct lu_env *env,
3537 struct dt_object *dt, const struct lu_buf *buf)
3539 return dt_xattr_list(env, dt_object_child(dt), buf);
3542 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
3544 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
3549 * Get default striping.
3551 * \param[in] env execution environment
3552 * \param[in] lo object
3553 * \param[out] lds default striping
3555 * \retval 0 on success
3556 * \retval negative if failed
3558 static int lod_get_default_lov_striping(const struct lu_env *env,
3559 struct lod_object *lo,
3560 struct lod_default_striping *lds)
3562 struct lod_thread_info *info = lod_env_info(env);
3563 struct lov_user_md_v1 *v1 = NULL;
3564 struct lov_user_md_v3 *v3 = NULL;
3565 struct lov_comp_md_v1 *comp_v1 = NULL;
3571 lds->lds_def_striping_set = 0;
3573 rc = lod_get_lov_ea(env, lo);
3577 if (rc < (typeof(rc))sizeof(struct lov_user_md))
3580 v1 = info->lti_ea_store;
3581 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
3582 lustre_swab_lov_user_md_v1(v1);
3583 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
3584 v3 = (struct lov_user_md_v3 *)v1;
3585 lustre_swab_lov_user_md_v3(v3);
3586 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
3587 comp_v1 = (struct lov_comp_md_v1 *)v1;
3588 lustre_swab_lov_comp_md_v1(comp_v1);
3591 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1 &&
3592 v1->lmm_magic != LOV_MAGIC_COMP_V1)
3595 if (v1->lmm_magic == LOV_MAGIC_COMP_V1) {
3596 comp_v1 = (struct lov_comp_md_v1 *)v1;
3597 comp_cnt = comp_v1->lcm_entry_count;
3606 /* realloc default comp entries if necessary */
3607 rc = lod_def_striping_comp_resize(lds, comp_cnt);
3611 lds->lds_def_comp_cnt = comp_cnt;
3612 lds->lds_def_striping_is_composite = composite ? 1 : 0;
3614 for (i = 0; i < comp_cnt; i++) {
3615 struct lod_layout_component *lod_comp;
3616 struct lu_extent *ext;
3619 lod_comp = &lds->lds_def_comp_entries[i];
3621 * reset lod_comp values, llc_stripes is always NULL in
3622 * the default striping template, llc_pool will be reset
3625 memset(lod_comp, 0, offsetof(typeof(*lod_comp), llc_pool));
3628 v1 = (struct lov_user_md *)((char *)comp_v1 +
3629 comp_v1->lcm_entries[i].lcme_offset);
3630 ext = &comp_v1->lcm_entries[i].lcme_extent;
3631 lod_comp->llc_extent = *ext;
3634 if (v1->lmm_pattern != LOV_PATTERN_RAID0 &&
3635 v1->lmm_pattern != 0) {
3636 lod_free_def_comp_entries(lds);
3640 CDEBUG(D_LAYOUT, DFID" stripe_count=%d stripe_size=%d "
3641 "stripe_offset=%d\n",
3642 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
3643 (int)v1->lmm_stripe_count, (int)v1->lmm_stripe_size,
3644 (int)v1->lmm_stripe_offset);
3646 lod_comp->llc_stripenr = v1->lmm_stripe_count;
3647 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
3648 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
3651 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
3652 /* XXX: sanity check here */
3653 v3 = (struct lov_user_md_v3 *) v1;
3654 if (v3->lmm_pool_name[0] != '\0')
3655 pool = v3->lmm_pool_name;
3657 lod_set_def_pool(lds, i, pool);
3660 lds->lds_def_striping_set = 1;
3665 * Get default directory striping.
3667 * \param[in] env execution environment
3668 * \param[in] lo object
3669 * \param[out] lds default striping
3671 * \retval 0 on success
3672 * \retval negative if failed
3674 static int lod_get_default_lmv_striping(const struct lu_env *env,
3675 struct lod_object *lo,
3676 struct lod_default_striping *lds)
3678 struct lod_thread_info *info = lod_env_info(env);
3679 struct lmv_user_md_v1 *v1 = NULL;
3683 lds->lds_dir_def_striping_set = 0;
3684 rc = lod_get_default_lmv_ea(env, lo);
3688 if (rc < (typeof(rc))sizeof(struct lmv_user_md))
3691 v1 = info->lti_ea_store;
3693 lds->lds_dir_def_stripenr = le32_to_cpu(v1->lum_stripe_count);
3694 lds->lds_dir_def_stripe_offset = le32_to_cpu(v1->lum_stripe_offset);
3695 lds->lds_dir_def_hash_type = le32_to_cpu(v1->lum_hash_type);
3696 lds->lds_dir_def_striping_set = 1;
3702 * Get default striping in the object.
3704 * Get object default striping and default directory striping.
3706 * \param[in] env execution environment
3707 * \param[in] lo object
3708 * \param[out] lds default striping
3710 * \retval 0 on success
3711 * \retval negative if failed
3713 static int lod_get_default_striping(const struct lu_env *env,
3714 struct lod_object *lo,
3715 struct lod_default_striping *lds)
3719 rc = lod_get_default_lov_striping(env, lo, lds);
3720 rc1 = lod_get_default_lmv_striping(env, lo, lds);
3721 if (rc == 0 && rc1 < 0)
3728 * Apply default striping on object.
3730 * If object striping pattern is not set, set to the one in default striping.
3731 * The default striping is from parent or fs.
3733 * \param[in] lo new object
3734 * \param[in] lds default striping
3735 * \param[in] mode new object's mode
3737 static void lod_striping_from_default(struct lod_object *lo,
3738 const struct lod_default_striping *lds,
3741 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
3742 struct lov_desc *desc = &d->lod_desc;
3745 if (lds->lds_def_striping_set && S_ISREG(mode)) {
3746 rc = lod_alloc_comp_entries(lo, lds->lds_def_comp_cnt);
3750 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
3752 for (i = 0; i < lo->ldo_comp_cnt; i++) {
3753 struct lod_layout_component *obj_comp =
3754 &lo->ldo_comp_entries[i];
3755 struct lod_layout_component *def_comp =
3756 &lds->lds_def_comp_entries[i];
3758 CDEBUG(D_LAYOUT, "Inherite from default: size:%hu "
3759 "nr:%u offset:%u %s\n",
3760 def_comp->llc_stripe_size,
3761 def_comp->llc_stripenr,
3762 def_comp->llc_stripe_offset,
3763 def_comp->llc_pool ?: "");
3765 *obj_comp = *def_comp;
3766 if (def_comp->llc_pool != NULL) {
3767 /* pointer was copied from def_comp */
3768 obj_comp->llc_pool = NULL;
3769 lod_obj_set_pool(lo, i, def_comp->llc_pool);
3773 * Don't initialize these fields for plain layout
3774 * (v1/v3) here, they are inherited in the order of
3775 * 'parent' -> 'fs default (root)' -> 'global default
3776 * values for stripe_count & stripe_size'.
3778 * see lod_ah_init().
3780 if (!lo->ldo_is_composite)
3783 if (obj_comp->llc_stripenr <= 0)
3784 obj_comp->llc_stripenr =
3785 desc->ld_default_stripe_count;
3786 if (obj_comp->llc_stripe_size <= 0)
3787 obj_comp->llc_stripe_size =
3788 desc->ld_default_stripe_size;
3790 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
3791 if (lo->ldo_dir_stripenr == 0)
3792 lo->ldo_dir_stripenr = lds->lds_dir_def_stripenr;
3793 if (lo->ldo_dir_stripe_offset == -1)
3794 lo->ldo_dir_stripe_offset =
3795 lds->lds_dir_def_stripe_offset;
3796 if (lo->ldo_dir_hash_type == 0)
3797 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
3799 CDEBUG(D_LAYOUT, "striping from default dir: nr:%hu, "
3800 "offset:%u, hash_type:%u\n",
3801 lo->ldo_dir_stripenr, lo->ldo_dir_stripe_offset,
3802 lo->ldo_dir_hash_type);
3806 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root)
3808 struct lod_layout_component *lod_comp;
3810 if (lo->ldo_comp_cnt == 0)
3813 if (lo->ldo_is_composite)
3816 lod_comp = &lo->ldo_comp_entries[0];
3818 if (lod_comp->llc_stripenr <= 0 ||
3819 lod_comp->llc_stripe_size <= 0)
3822 if (from_root && (lod_comp->llc_pool == NULL ||
3823 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
3830 * Implementation of dt_object_operations::do_ah_init.
3832 * This method is used to make a decision on the striping configuration for the
3833 * object being created. It can be taken from the \a parent object if it exists,
3834 * or filesystem's default. The resulting configuration (number of stripes,
3835 * stripe size/offset, pool name, etc) is stored in the object itself and will
3836 * be used by the methods like ->doo_declare_create().
3838 * \see dt_object_operations::do_ah_init() in the API description for details.
3840 static void lod_ah_init(const struct lu_env *env,
3841 struct dt_allocation_hint *ah,
3842 struct dt_object *parent,
3843 struct dt_object *child,
3846 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
3847 struct lod_thread_info *info = lod_env_info(env);
3848 struct lod_default_striping *lds = &info->lti_def_striping;
3849 struct dt_object *nextp = NULL;
3850 struct dt_object *nextc;
3851 struct lod_object *lp = NULL;
3852 struct lod_object *lc;
3853 struct lov_desc *desc;
3854 struct lod_layout_component *lod_comp;
3860 if (likely(parent)) {
3861 nextp = dt_object_child(parent);
3862 lp = lod_dt_obj(parent);
3865 nextc = dt_object_child(child);
3866 lc = lod_dt_obj(child);
3868 LASSERT(!lod_obj_is_striped(child));
3869 /* default layout template may have been set on the regular file
3870 * when this is called from mdd_create_data() */
3871 if (S_ISREG(child_mode))
3872 lod_free_comp_entries(lc);
3874 if (!dt_object_exists(nextc))
3875 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
3877 if (S_ISDIR(child_mode)) {
3878 /* other default values are 0 */
3879 lc->ldo_dir_stripe_offset = -1;
3881 /* get default striping from parent object */
3882 if (likely(lp != NULL))
3883 lod_get_default_striping(env, lp, lds);
3885 /* set child default striping info, default value is NULL */
3886 if (lds->lds_def_striping_set || lds->lds_dir_def_striping_set)
3887 lc->ldo_def_striping = lds;
3889 /* It should always honour the specified stripes */
3890 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
3891 lod_verify_md_striping(d, ah->dah_eadata) == 0) {
3892 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
3894 lc->ldo_dir_stripenr =
3895 le32_to_cpu(lum1->lum_stripe_count);
3896 lc->ldo_dir_stripe_offset =
3897 le32_to_cpu(lum1->lum_stripe_offset);
3898 lc->ldo_dir_hash_type =
3899 le32_to_cpu(lum1->lum_hash_type);
3900 CDEBUG(D_INFO, "set dir stripe: count %hu, offset %d, "
3902 lc->ldo_dir_stripenr,
3903 (int)lc->ldo_dir_stripe_offset,
3904 lc->ldo_dir_hash_type);
3906 /* transfer defaults LMV to new directory */
3907 lod_striping_from_default(lc, lds, child_mode);
3910 /* shrink the stripe_count to the avaible MDT count */
3911 if (lc->ldo_dir_stripenr > d->lod_remote_mdt_count + 1 &&
3912 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))
3913 lc->ldo_dir_stripenr = d->lod_remote_mdt_count + 1;
3915 /* Directory will be striped only if stripe_count > 1, if
3916 * stripe_count == 1, let's reset stripenr = 0 to avoid
3917 * create single master stripe and also help to unify the
3918 * stripe handling of directories and files */
3919 if (lc->ldo_dir_stripenr == 1)
3920 lc->ldo_dir_stripenr = 0;
3922 CDEBUG(D_INFO, "final dir stripe [%hu %d %u]\n",
3923 lc->ldo_dir_stripenr, (int)lc->ldo_dir_stripe_offset,
3924 lc->ldo_dir_hash_type);
3929 /* child object regular file*/
3931 if (!lod_object_will_be_striped(S_ISREG(child_mode),
3932 lu_object_fid(&child->do_lu)))
3935 /* If object is going to be striped over OSTs, transfer default
3936 * striping information to the child, so that we can use it
3937 * during declaration and creation.
3939 * Try from the parent first.
3941 if (likely(lp != NULL)) {
3942 rc = lod_get_default_lov_striping(env, lp, lds);
3944 lod_striping_from_default(lc, lds, child_mode);
3947 /* Initialize lod_device::lod_md_root object reference */
3948 if (d->lod_md_root == NULL) {
3949 struct dt_object *root;
3950 struct lod_object *lroot;
3952 lu_root_fid(&info->lti_fid);
3953 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
3954 if (!IS_ERR(root)) {
3955 lroot = lod_dt_obj(root);
3957 spin_lock(&d->lod_lock);
3958 if (d->lod_md_root != NULL)
3959 dt_object_put(env, &d->lod_md_root->ldo_obj);
3960 d->lod_md_root = lroot;
3961 spin_unlock(&d->lod_lock);
3965 /* try inherit layout from the root object (fs default) when:
3966 * - parent does not have default layout; or
3967 * - parent has plain(v1/v3) default layout, and some attributes
3968 * are not specified in the default layout;
3970 if (d->lod_md_root != NULL && lod_need_inherit_more(lc, true)) {
3971 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds);
3974 if (lc->ldo_comp_cnt == 0) {
3975 lod_striping_from_default(lc, lds, child_mode);
3976 } else if (!lds->lds_def_striping_is_composite) {
3977 struct lod_layout_component *def_comp;
3979 LASSERT(!lc->ldo_is_composite);
3980 lod_comp = &lc->ldo_comp_entries[0];
3981 def_comp = &lds->lds_def_comp_entries[0];
3983 if (lod_comp->llc_stripenr <= 0)
3984 lod_comp->llc_stripenr = def_comp->llc_stripenr;
3985 if (lod_comp->llc_stripe_size <= 0)
3986 lod_comp->llc_stripe_size =
3987 def_comp->llc_stripe_size;
3988 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT)
3989 lod_comp->llc_stripe_offset =
3990 def_comp->llc_stripe_offset;
3991 if (lod_comp->llc_pool == NULL)
3992 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
3997 * fs default striping may not be explicitly set, or historically set
3998 * in config log, use them.
4000 if (lod_need_inherit_more(lc, false)) {
4002 if (lc->ldo_comp_cnt == 0) {
4003 rc = lod_alloc_comp_entries(lc, 1);
4005 /* fail to allocate memory, will create a
4006 * non-striped file. */
4008 lc->ldo_is_composite = 0;
4009 lod_comp = &lc->ldo_comp_entries[0];
4010 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4012 LASSERT(!lc->ldo_is_composite);
4013 lod_comp = &lc->ldo_comp_entries[0];
4014 desc = &d->lod_desc;
4015 if (lod_comp->llc_stripenr <= 0)
4016 lod_comp->llc_stripenr = desc->ld_default_stripe_count;
4017 if (lod_comp->llc_stripe_size <= 0)
4018 lod_comp->llc_stripe_size =
4019 desc->ld_default_stripe_size;
4025 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
4027 * Size initialization on late striping.
4029 * Propagate the size of a truncated object to a deferred striping.
4030 * This function handles a special case when truncate was done on a
4031 * non-striped object and now while the striping is being created
4032 * we can't lose that size, so we have to propagate it to the stripes
4035 * \param[in] env execution environment
4036 * \param[in] dt object
4037 * \param[in] th transaction handle
4039 * \retval 0 on success
4040 * \retval negative if failed
4042 static int lod_declare_init_size(const struct lu_env *env,
4043 struct dt_object *dt, struct thandle *th)
4045 struct dt_object *next = dt_object_child(dt);
4046 struct lod_object *lo = lod_dt_obj(dt);
4047 struct dt_object **objects = NULL;
4048 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4049 uint64_t size, offs;
4050 int i, rc, stripe, stripenr = 0, stripe_size = 0;
4053 if (!lod_obj_is_striped(dt))
4056 rc = dt_attr_get(env, next, attr);
4057 LASSERT(attr->la_valid & LA_SIZE);
4061 size = attr->la_size;
4065 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4066 struct lod_layout_component *lod_comp;
4067 struct lu_extent *extent;
4069 lod_comp = &lo->ldo_comp_entries[i];
4071 if (lod_comp->llc_stripe == NULL)
4074 extent = &lod_comp->llc_extent;
4075 CDEBUG(D_INFO, "%lld [%lld, %lld)\n",
4076 size, extent->e_start, extent->e_end);
4077 if (!lo->ldo_is_composite ||
4078 (size >= extent->e_start && size < extent->e_end)) {
4079 objects = lod_comp->llc_stripe;
4080 stripenr = lod_comp->llc_stripenr;
4081 stripe_size = lod_comp->llc_stripe_size;
4089 LASSERT(objects != NULL && stripe_size != 0);
4091 /* ll_do_div64(a, b) returns a % b, and a = a / b */
4092 ll_do_div64(size, (__u64)stripe_size);
4093 stripe = ll_do_div64(size, (__u64)stripenr);
4094 LASSERT(objects[stripe] != NULL);
4096 size = size * stripe_size;
4097 offs = attr->la_size;
4098 size += ll_do_div64(offs, stripe_size);
4100 attr->la_valid = LA_SIZE;
4101 attr->la_size = size;
4103 rc = lod_sub_object_declare_attr_set(env, objects[stripe], attr, th);
4109 * Declare creation of striped object.
4111 * The function declares creation stripes for a regular object. The function
4112 * also declares whether the stripes will be created with non-zero size if
4113 * previously size was set non-zero on the master object. If object \a dt is
4114 * not local, then only fully defined striping can be applied in \a lovea.
4115 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
4118 * \param[in] env execution environment
4119 * \param[in] dt object
4120 * \param[in] attr attributes the stripes will be created with
4121 * \param[in] lovea a buffer containing striping description
4122 * \param[in] th transaction handle
4124 * \retval 0 on success
4125 * \retval negative if failed
4127 int lod_declare_striped_object(const struct lu_env *env, struct dt_object *dt,
4128 struct lu_attr *attr,
4129 const struct lu_buf *lovea, struct thandle *th)
4131 struct lod_thread_info *info = lod_env_info(env);
4132 struct dt_object *next = dt_object_child(dt);
4133 struct lod_object *lo = lod_dt_obj(dt);
4137 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
4138 GOTO(out, rc = -ENOMEM);
4140 if (!dt_object_remote(next)) {
4141 /* choose OST and generate appropriate objects */
4142 rc = lod_prepare_create(env, lo, attr, lovea, th);
4147 * declare storage for striping data
4149 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
4151 /* LOD can not choose OST objects for remote objects, i.e.
4152 * stripes must be ready before that. Right now, it can only
4153 * happen during migrate, i.e. migrate process needs to create
4154 * remote regular file (mdd_migrate_create), then the migrate
4155 * process will provide stripeEA. */
4156 LASSERT(lovea != NULL);
4157 info->lti_buf = *lovea;
4160 rc = lod_sub_object_declare_xattr_set(env, next, &info->lti_buf,
4161 XATTR_NAME_LOV, 0, th);
4166 * if striping is created with local object's size > 0,
4167 * we have to propagate this size to specific object
4168 * the case is possible only when local object was created previously
4170 if (dt_object_exists(next))
4171 rc = lod_declare_init_size(env, dt, th);
4174 /* failed to create striping or to set initial size, let's reset
4175 * config so that others don't get confused */
4177 lod_object_free_striping(env, lo);
4183 * Implementation of dt_object_operations::do_declare_create.
4185 * The method declares creation of a new object. If the object will be striped,
4186 * then helper functions are called to find FIDs for the stripes, declare
4187 * creation of the stripes and declare initialization of the striping
4188 * information to be stored in the master object.
4190 * \see dt_object_operations::do_declare_create() in the API description
4193 static int lod_declare_object_create(const struct lu_env *env,
4194 struct dt_object *dt,
4195 struct lu_attr *attr,
4196 struct dt_allocation_hint *hint,
4197 struct dt_object_format *dof,
4200 struct dt_object *next = dt_object_child(dt);
4201 struct lod_object *lo = lod_dt_obj(dt);
4210 * first of all, we declare creation of local object
4212 rc = lod_sub_object_declare_create(env, next, attr, hint, dof, th);
4216 if (dof->dof_type == DFT_SYM)
4217 dt->do_body_ops = &lod_body_lnk_ops;
4218 else if (dof->dof_type == DFT_REGULAR)
4219 dt->do_body_ops = &lod_body_ops;
4222 * it's lod_ah_init() that has decided the object will be striped
4224 if (dof->dof_type == DFT_REGULAR) {
4225 /* callers don't want stripes */
4226 /* XXX: all tricky interactions with ->ah_make_hint() decided
4227 * to use striping, then ->declare_create() behaving differently
4228 * should be cleaned */
4229 if (dof->u.dof_reg.striped != 0)
4230 rc = lod_declare_striped_object(env, dt, attr,
4232 } else if (dof->dof_type == DFT_DIR) {
4233 struct seq_server_site *ss;
4235 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
4237 /* If the parent has default stripeEA, and client
4238 * did not find it before sending create request,
4239 * then MDT will return -EREMOTE, and client will
4240 * retrieve the default stripeEA and re-create the
4243 * Note: if dah_eadata != NULL, it means creating the
4244 * striped directory with specified stripeEA, then it
4245 * should ignore the default stripeEA */
4246 if (hint != NULL && hint->dah_eadata == NULL) {
4247 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
4248 GOTO(out, rc = -EREMOTE);
4250 if (lo->ldo_dir_stripe_offset == -1) {
4251 /* child and parent should be in the same MDT */
4252 if (hint->dah_parent != NULL &&
4253 dt_object_remote(hint->dah_parent))
4254 GOTO(out, rc = -EREMOTE);
4255 } else if (lo->ldo_dir_stripe_offset !=
4257 struct lod_device *lod;
4258 struct lod_tgt_descs *ltd;
4259 struct lod_tgt_desc *tgt = NULL;
4260 bool found_mdt = false;
4263 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
4264 ltd = &lod->lod_mdt_descs;
4265 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
4266 tgt = LTD_TGT(ltd, i);
4267 if (tgt->ltd_index ==
4268 lo->ldo_dir_stripe_offset) {
4274 /* If the MDT indicated by stripe_offset can be
4275 * found, then tell client to resend the create
4276 * request to the correct MDT, otherwise return
4277 * error to client */
4279 GOTO(out, rc = -EREMOTE);
4281 GOTO(out, rc = -EINVAL);
4285 rc = lod_declare_dir_striping_create(env, dt, attr, dof, th);
4288 /* failed to create striping or to set initial size, let's reset
4289 * config so that others don't get confused */
4291 lod_object_free_striping(env, lo);
4296 * Creation of a striped regular object.
4298 * The function is called to create the stripe objects for a regular
4299 * striped file. This can happen at the initial object creation or
4300 * when the caller asks LOD to do so using ->do_xattr_set() method
4301 * (so called late striping). Notice all the information are already
4302 * prepared in the form of the list of objects (ldo_stripe field).
4303 * This is done during declare phase.
4305 * \param[in] env execution environment
4306 * \param[in] dt object
4307 * \param[in] attr attributes the stripes will be created with
4308 * \param[in] dof format of stripes (see OSD API description)
4309 * \param[in] th transaction handle
4311 * \retval 0 on success
4312 * \retval negative if failed
4314 int lod_striping_create(const struct lu_env *env, struct dt_object *dt,
4315 struct lu_attr *attr, struct dt_object_format *dof,
4318 struct lod_layout_component *lod_comp;
4319 struct lod_object *lo = lod_dt_obj(dt);
4323 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
4325 /* create all underlying objects */
4326 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4327 lod_comp = &lo->ldo_comp_entries[i];
4329 if (lod_comp_inited(lod_comp))
4332 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
4333 lod_comp_set_init(lod_comp);
4335 if (lod_comp->llc_stripe == NULL)
4338 LASSERT(lod_comp->llc_stripenr);
4339 for (j = 0; j < lod_comp->llc_stripenr; j++) {
4340 struct dt_object *object = lod_comp->llc_stripe[j];
4341 LASSERT(object != NULL);
4342 rc = lod_sub_object_create(env, object, attr, NULL,
4347 lod_comp_set_init(lod_comp);
4351 rc = lod_generate_and_set_lovea(env, lo, th);
4354 lo->ldo_comp_cached = 1;
4356 lod_object_free_striping(env, lo);
4362 * Implementation of dt_object_operations::do_create.
4364 * If any of preceeding methods (like ->do_declare_create(),
4365 * ->do_ah_init(), etc) chose to create a striped object,
4366 * then this method will create the master and the stripes.
4368 * \see dt_object_operations::do_create() in the API description for details.
4370 static int lod_object_create(const struct lu_env *env, struct dt_object *dt,
4371 struct lu_attr *attr,
4372 struct dt_allocation_hint *hint,
4373 struct dt_object_format *dof, struct thandle *th)
4378 /* create local object */
4379 rc = lod_sub_object_create(env, dt_object_child(dt), attr, hint, dof,
4384 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
4385 lod_obj_is_striped(dt) && dof->u.dof_reg.striped != 0) {
4386 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
4387 rc = lod_striping_create(env, dt, attr, dof, th);
4394 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
4395 struct dt_object *dt, struct thandle *th,
4396 int stripe_idx, struct lod_obj_stripe_cb_data *data)
4398 if (data->locd_declare)
4399 return lod_sub_object_declare_destroy(env, dt, th);
4400 else if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
4401 stripe_idx == cfs_fail_val)
4402 return lod_sub_object_destroy(env, dt, th);
4408 * Implementation of dt_object_operations::do_declare_destroy.
4410 * If the object is a striped directory, then the function declares reference
4411 * removal from the master object (this is an index) to the stripes and declares
4412 * destroy of all the stripes. In all the cases, it declares an intention to
4413 * destroy the object itself.
4415 * \see dt_object_operations::do_declare_destroy() in the API description
4418 static int lod_declare_object_destroy(const struct lu_env *env,
4419 struct dt_object *dt,
4422 struct dt_object *next = dt_object_child(dt);
4423 struct lod_object *lo = lod_dt_obj(dt);
4424 struct lod_thread_info *info = lod_env_info(env);
4425 char *stripe_name = info->lti_key;
4430 * load striping information, notice we don't do this when object
4431 * is being initialized as we don't need this information till
4432 * few specific cases like destroy, chown
4434 rc = lod_load_striping(env, lo);
4438 /* declare destroy for all underlying objects */
4439 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4440 rc = next->do_ops->do_index_try(env, next,
4441 &dt_directory_features);
4445 for (i = 0; i < lo->ldo_dir_stripenr; i++) {
4446 rc = lod_sub_object_declare_ref_del(env, next, th);
4450 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4451 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
4453 rc = lod_sub_object_declare_delete(env, next,
4454 (const struct dt_key *)stripe_name, th);
4461 * we declare destroy for the local object
4463 rc = lod_sub_object_declare_destroy(env, next, th);
4467 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
4468 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
4471 if (!lod_obj_is_striped(dt))
4474 /* declare destroy all striped objects */
4475 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4476 for (i = 0; i < lo->ldo_dir_stripenr; i++) {
4477 if (lo->ldo_stripe[i] == NULL)
4480 rc = lod_sub_object_declare_ref_del(env,
4481 lo->ldo_stripe[i], th);
4483 rc = lod_sub_object_declare_destroy(env,
4484 lo->ldo_stripe[i], th);
4489 struct lod_obj_stripe_cb_data data;
4491 data.locd_declare = true;
4492 rc = lod_obj_for_each_stripe(env, lo, th,
4493 lod_obj_stripe_destroy_cb, &data);
4500 * Implementation of dt_object_operations::do_destroy.
4502 * If the object is a striped directory, then the function removes references
4503 * from the master object (this is an index) to the stripes and destroys all
4504 * the stripes. In all the cases, the function destroys the object itself.
4506 * \see dt_object_operations::do_destroy() in the API description for details.
4508 static int lod_object_destroy(const struct lu_env *env,
4509 struct dt_object *dt, struct thandle *th)
4511 struct dt_object *next = dt_object_child(dt);
4512 struct lod_object *lo = lod_dt_obj(dt);
4513 struct lod_thread_info *info = lod_env_info(env);
4514 char *stripe_name = info->lti_key;
4519 /* destroy sub-stripe of master object */
4520 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4521 rc = next->do_ops->do_index_try(env, next,
4522 &dt_directory_features);
4526 for (i = 0; i < lo->ldo_dir_stripenr; i++) {
4527 rc = lod_sub_object_ref_del(env, next, th);
4531 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4532 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
4535 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
4536 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
4537 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)));
4539 rc = lod_sub_object_delete(env, next,
4540 (const struct dt_key *)stripe_name, th);
4546 rc = lod_sub_object_destroy(env, next, th);
4550 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
4551 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
4554 if (!lod_obj_is_striped(dt))
4557 /* destroy all striped objects */
4558 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4559 for (i = 0; i < lo->ldo_dir_stripenr; i++) {
4560 if (lo->ldo_stripe[i] == NULL)
4562 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
4563 i == cfs_fail_val) {
4564 dt_write_lock(env, lo->ldo_stripe[i],
4566 rc = lod_sub_object_ref_del(env,
4567 lo->ldo_stripe[i], th);
4568 dt_write_unlock(env, lo->ldo_stripe[i]);
4572 rc = lod_sub_object_destroy(env,
4573 lo->ldo_stripe[i], th);
4579 struct lod_obj_stripe_cb_data data;
4581 data.locd_declare = false;
4582 rc = lod_obj_for_each_stripe(env, lo, th,
4583 lod_obj_stripe_destroy_cb, &data);
4590 * Implementation of dt_object_operations::do_declare_ref_add.
4592 * \see dt_object_operations::do_declare_ref_add() in the API description
4595 static int lod_declare_ref_add(const struct lu_env *env,
4596 struct dt_object *dt, struct thandle *th)
4598 return lod_sub_object_declare_ref_add(env, dt_object_child(dt), th);
4602 * Implementation of dt_object_operations::do_ref_add.
4604 * \see dt_object_operations::do_ref_add() in the API description for details.
4606 static int lod_ref_add(const struct lu_env *env,
4607 struct dt_object *dt, struct thandle *th)
4609 return lod_sub_object_ref_add(env, dt_object_child(dt), th);
4613 * Implementation of dt_object_operations::do_declare_ref_del.
4615 * \see dt_object_operations::do_declare_ref_del() in the API description
4618 static int lod_declare_ref_del(const struct lu_env *env,
4619 struct dt_object *dt, struct thandle *th)
4621 return lod_sub_object_declare_ref_del(env, dt_object_child(dt), th);
4625 * Implementation of dt_object_operations::do_ref_del
4627 * \see dt_object_operations::do_ref_del() in the API description for details.
4629 static int lod_ref_del(const struct lu_env *env,
4630 struct dt_object *dt, struct thandle *th)
4632 return lod_sub_object_ref_del(env, dt_object_child(dt), th);
4636 * Implementation of dt_object_operations::do_object_sync.
4638 * \see dt_object_operations::do_object_sync() in the API description
4641 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
4642 __u64 start, __u64 end)
4644 return dt_object_sync(env, dt_object_child(dt), start, end);
4648 * Release LDLM locks on the stripes of a striped directory.
4650 * Iterates over all the locks taken on the stripe objects and
4653 * \param[in] env execution environment
4654 * \param[in] dt striped object
4655 * \param[in] einfo lock description
4656 * \param[in] policy data describing requested lock
4658 * \retval 0 on success
4659 * \retval negative if failed
4661 static int lod_object_unlock_internal(const struct lu_env *env,
4662 struct dt_object *dt,
4663 struct ldlm_enqueue_info *einfo,
4664 union ldlm_policy_data *policy)
4666 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
4671 if (slave_locks == NULL)
4674 for (i = 1; i < slave_locks->count; i++) {
4675 if (lustre_handle_is_used(&slave_locks->handles[i]))
4676 ldlm_lock_decref_and_cancel(&slave_locks->handles[i],
4684 * Implementation of dt_object_operations::do_object_unlock.
4686 * Used to release LDLM lock(s).
4688 * \see dt_object_operations::do_object_unlock() in the API description
4691 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
4692 struct ldlm_enqueue_info *einfo,
4693 union ldlm_policy_data *policy)
4695 struct lod_object *lo = lod_dt_obj(dt);
4696 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
4697 int slave_locks_size;
4701 if (slave_locks == NULL)
4704 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
4705 LASSERT(lo->ldo_dir_stripenr > 1);
4706 /* Note: for remote lock for single stripe dir, MDT will cancel
4707 * the lock by lockh directly */
4708 LASSERT(!dt_object_remote(dt_object_child(dt)));
4710 /* locks were unlocked in MDT layer */
4711 for (i = 1; i < slave_locks->count; i++) {
4712 LASSERT(!lustre_handle_is_used(&slave_locks->handles[i]));
4713 dt_invalidate(env, lo->ldo_stripe[i]);
4716 slave_locks_size = sizeof(*slave_locks) + slave_locks->count *
4717 sizeof(slave_locks->handles[0]);
4718 OBD_FREE(slave_locks, slave_locks_size);
4719 einfo->ei_cbdata = NULL;
4725 * Implementation of dt_object_operations::do_object_lock.
4727 * Used to get LDLM lock on the non-striped and striped objects.
4729 * \see dt_object_operations::do_object_lock() in the API description
4732 static int lod_object_lock(const struct lu_env *env,
4733 struct dt_object *dt,
4734 struct lustre_handle *lh,
4735 struct ldlm_enqueue_info *einfo,
4736 union ldlm_policy_data *policy)
4738 struct lod_object *lo = lod_dt_obj(dt);
4741 int slave_locks_size;
4742 struct lustre_handle_array *slave_locks = NULL;
4745 /* remote object lock */
4746 if (!einfo->ei_enq_slave) {
4747 LASSERT(dt_object_remote(dt));
4748 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
4752 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4753 GOTO(out, rc = -ENOTDIR);
4755 rc = lod_load_striping(env, lo);
4760 if (lo->ldo_dir_stripenr <= 1) {
4762 * NB, ei_cbdata stores pointer to slave locks, if no locks
4763 * taken, make sure it's set to NULL, otherwise MDT will try to
4766 einfo->ei_cbdata = NULL;
4770 slave_locks_size = sizeof(*slave_locks) + lo->ldo_dir_stripenr *
4771 sizeof(slave_locks->handles[0]);
4772 /* Freed in lod_object_unlock */
4773 OBD_ALLOC(slave_locks, slave_locks_size);
4774 if (slave_locks == NULL)
4775 GOTO(out, rc = -ENOMEM);
4776 slave_locks->count = lo->ldo_dir_stripenr;
4778 /* striped directory lock */
4779 for (i = 1; i < lo->ldo_dir_stripenr; i++) {
4780 struct lustre_handle lockh;
4781 struct ldlm_res_id *res_id;
4783 res_id = &lod_env_info(env)->lti_res_id;
4784 fid_build_reg_res_name(lu_object_fid(&lo->ldo_stripe[i]->do_lu),
4786 einfo->ei_res_id = res_id;
4788 LASSERT(lo->ldo_stripe[i] != NULL);
4789 if (likely(dt_object_remote(lo->ldo_stripe[i]))) {
4790 rc = dt_object_lock(env, lo->ldo_stripe[i], &lockh,
4793 struct ldlm_namespace *ns = einfo->ei_namespace;
4794 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
4795 ldlm_completion_callback completion = einfo->ei_cb_cp;
4796 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
4798 if (einfo->ei_mode == LCK_PW ||
4799 einfo->ei_mode == LCK_EX)
4800 dlmflags |= LDLM_FL_COS_INCOMPAT;
4802 /* This only happens if there are mulitple stripes
4803 * on the master MDT, i.e. except stripe0, there are
4804 * other stripes on the Master MDT as well, Only
4805 * happens in the test case right now. */
4806 LASSERT(ns != NULL);
4807 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS,
4808 policy, einfo->ei_mode,
4809 &dlmflags, blocking,
4811 NULL, 0, LVB_T_NONE,
4816 slave_locks->handles[i] = lockh;
4818 einfo->ei_cbdata = slave_locks;
4820 if (rc != 0 && slave_locks != NULL) {
4821 lod_object_unlock_internal(env, dt, einfo, policy);
4822 OBD_FREE(slave_locks, slave_locks_size);
4827 einfo->ei_cbdata = NULL;
4832 * Implementation of dt_object_operations::do_invalidate.
4834 * \see dt_object_operations::do_invalidate() in the API description for details
4836 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
4838 return dt_invalidate(env, dt_object_child(dt));
4841 static int lod_declare_layout_change(const struct lu_env *env,
4842 struct dt_object *dt,
4843 struct layout_intent *layout,
4844 const struct lu_buf *buf,
4847 struct lod_thread_info *info = lod_env_info(env);
4848 struct lod_object *lo = lod_dt_obj(dt);
4849 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
4850 struct dt_object *next = dt_object_child(dt);
4851 struct ost_pool *inuse = &info->lti_inuse_osts;
4852 struct lod_layout_component *lod_comp;
4853 struct lov_comp_md_v1 *comp_v1 = NULL;
4854 bool replay = false;
4855 bool need_create = false;
4859 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
4860 dt_object_remote(next))
4863 dt_write_lock(env, next, 0);
4865 * In case the client is passing lovea, which only happens during
4866 * the replay of layout intent write RPC for now, we may need to
4867 * parse the lovea and apply new layout configuration.
4869 if (buf && buf->lb_len) {
4870 struct lov_user_md_v1 *v1 = buf->lb_buf;
4872 if (v1->lmm_magic != (LOV_MAGIC_DEF | LOV_MAGIC_COMP_V1) &&
4874 __swab32(LOV_MAGIC_DEF | LOV_MAGIC_COMP_V1)) {
4875 CERROR("%s: the replay buffer of layout extend "
4876 "(magic %#x) does not contain expected "
4877 "composite layout.\n",
4878 lod2obd(d)->obd_name, v1->lmm_magic);
4879 GOTO(out, rc = -EINVAL);
4882 lod_object_free_striping(env, lo);
4883 rc = lod_use_defined_striping(env, lo, buf);
4887 rc = lod_get_lov_ea(env, lo);
4890 /* old on-disk EA is stored in info->lti_buf */
4891 comp_v1 = (struct lov_comp_md_v1 *)&info->lti_buf.lb_buf;
4894 /* non replay path */
4895 rc = lod_load_striping_locked(env, lo);
4899 /* Prepare inuse array for composite file */
4900 rc = lod_prepare_inuse(env, lo);
4905 /* Make sure defined layout covers the requested write range. */
4906 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
4907 if (lo->ldo_comp_cnt > 1 &&
4908 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
4909 lod_comp->llc_extent.e_end < layout->li_end) {
4910 CDEBUG(replay ? D_ERROR : D_LAYOUT,
4911 "%s: the defined layout [0, %#llx) does not covers "
4912 "the write range [%#llx, %#llx).\n",
4913 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
4914 layout->li_start, layout->li_end);
4915 GOTO(out, rc = -EINVAL);
4919 * Iterate ld->ldo_comp_entries, find the component whose extent under
4920 * the write range and not instantianted.
4922 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4923 lod_comp = &lo->ldo_comp_entries[i];
4925 if (lod_comp->llc_extent.e_start >= layout->li_end)
4929 if (lod_comp_inited(lod_comp))
4933 * In replay path, lod_comp is the EA passed by
4934 * client replay buffer, comp_v1 is the pre-recovery
4935 * on-disk EA, we'd sift out those components which
4936 * were init-ed in the on-disk EA.
4938 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
4943 * this component hasn't instantiated in normal path, or during
4944 * replay it needs replay the instantiation.
4947 /* A released component is being extended */
4948 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
4949 GOTO(out, rc = -EINVAL);
4953 rc = lod_qos_prep_create(env, lo, NULL, th, i, inuse);
4959 lod_obj_inc_layout_gen(lo);
4961 GOTO(unlock, rc = -EALREADY);
4964 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
4965 rc = lod_sub_object_declare_xattr_set(env, next, &info->lti_buf,
4966 XATTR_NAME_LOV, 0, th);
4970 lod_object_free_striping(env, lo);
4973 dt_write_unlock(env, next);
4979 * Instantiate layout component objects which covers the intent write offset.
4981 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
4982 struct layout_intent *layout,
4983 const struct lu_buf *buf, struct thandle *th)
4985 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4987 RETURN(lod_striping_create(env, dt, attr, NULL, th));
4990 struct dt_object_operations lod_obj_ops = {
4991 .do_read_lock = lod_object_read_lock,
4992 .do_write_lock = lod_object_write_lock,
4993 .do_read_unlock = lod_object_read_unlock,
4994 .do_write_unlock = lod_object_write_unlock,
4995 .do_write_locked = lod_object_write_locked,
4996 .do_attr_get = lod_attr_get,
4997 .do_declare_attr_set = lod_declare_attr_set,
4998 .do_attr_set = lod_attr_set,
4999 .do_xattr_get = lod_xattr_get,
5000 .do_declare_xattr_set = lod_declare_xattr_set,
5001 .do_xattr_set = lod_xattr_set,
5002 .do_declare_xattr_del = lod_declare_xattr_del,
5003 .do_xattr_del = lod_xattr_del,
5004 .do_xattr_list = lod_xattr_list,
5005 .do_ah_init = lod_ah_init,
5006 .do_declare_create = lod_declare_object_create,
5007 .do_create = lod_object_create,
5008 .do_declare_destroy = lod_declare_object_destroy,
5009 .do_destroy = lod_object_destroy,
5010 .do_index_try = lod_index_try,
5011 .do_declare_ref_add = lod_declare_ref_add,
5012 .do_ref_add = lod_ref_add,
5013 .do_declare_ref_del = lod_declare_ref_del,
5014 .do_ref_del = lod_ref_del,
5015 .do_object_sync = lod_object_sync,
5016 .do_object_lock = lod_object_lock,
5017 .do_object_unlock = lod_object_unlock,
5018 .do_invalidate = lod_invalidate,
5019 .do_declare_layout_change = lod_declare_layout_change,
5020 .do_layout_change = lod_layout_change,
5024 * Implementation of dt_body_operations::dbo_read.
5026 * \see dt_body_operations::dbo_read() in the API description for details.
5028 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
5029 struct lu_buf *buf, loff_t *pos)
5031 struct dt_object *next = dt_object_child(dt);
5032 return next->do_body_ops->dbo_read(env, next, buf, pos);
5036 * Implementation of dt_body_operations::dbo_declare_write.
5038 * \see dt_body_operations::dbo_declare_write() in the API description
5041 static ssize_t lod_declare_write(const struct lu_env *env,
5042 struct dt_object *dt,
5043 const struct lu_buf *buf, loff_t pos,
5046 return lod_sub_object_declare_write(env, dt_object_child(dt), buf, pos,
5051 * Implementation of dt_body_operations::dbo_write.
5053 * \see dt_body_operations::dbo_write() in the API description for details.
5055 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
5056 const struct lu_buf *buf, loff_t *pos,
5057 struct thandle *th, int iq)
5059 return lod_sub_object_write(env, dt_object_child(dt), buf, pos, th, iq);
5062 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
5063 __u64 start, __u64 end, struct thandle *th)
5065 if (dt_object_remote(dt))
5068 return lod_sub_object_declare_punch(env, dt_object_child(dt), start,
5072 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
5073 __u64 start, __u64 end, struct thandle *th)
5075 if (dt_object_remote(dt))
5078 return lod_sub_object_punch(env, dt_object_child(dt), start, end, th);
5081 static const struct dt_body_operations lod_body_lnk_ops = {
5082 .dbo_read = lod_read,
5083 .dbo_declare_write = lod_declare_write,
5084 .dbo_write = lod_write
5087 static const struct dt_body_operations lod_body_ops = {
5088 .dbo_read = lod_read,
5089 .dbo_declare_write = lod_declare_write,
5090 .dbo_write = lod_write,
5091 .dbo_declare_punch = lod_declare_punch,
5092 .dbo_punch = lod_punch,
5096 * Implementation of lu_object_operations::loo_object_init.
5098 * The function determines the type and the index of the target device using
5099 * sequence of the object's FID. Then passes control down to the
5100 * corresponding device:
5101 * OSD for the local objects, OSP for remote
5103 * \see lu_object_operations::loo_object_init() in the API description
5106 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
5107 const struct lu_object_conf *conf)
5109 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
5110 struct lu_device *cdev = NULL;
5111 struct lu_object *cobj;
5112 struct lod_tgt_descs *ltd = NULL;
5113 struct lod_tgt_desc *tgt;
5115 int type = LU_SEQ_RANGE_ANY;
5119 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
5121 /* Note: Sometimes, it will Return EAGAIN here, see
5122 * ptrlpc_import_delay_req(), which might confuse
5123 * lu_object_find_at() and make it wait there incorrectly.
5124 * so we convert it to EIO here.*/
5131 if (type == LU_SEQ_RANGE_MDT &&
5132 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
5133 cdev = &lod->lod_child->dd_lu_dev;
5134 } else if (type == LU_SEQ_RANGE_MDT) {
5135 ltd = &lod->lod_mdt_descs;
5137 } else if (type == LU_SEQ_RANGE_OST) {
5138 ltd = &lod->lod_ost_descs;
5145 if (ltd->ltd_tgts_size > idx &&
5146 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
5147 tgt = LTD_TGT(ltd, idx);
5149 LASSERT(tgt != NULL);
5150 LASSERT(tgt->ltd_tgt != NULL);
5152 cdev = &(tgt->ltd_tgt->dd_lu_dev);
5154 lod_putref(lod, ltd);
5157 if (unlikely(cdev == NULL))
5160 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
5161 if (unlikely(cobj == NULL))
5164 lu_object_add(lo, cobj);
5171 * Release resources associated with striping.
5173 * If the object is striped (regular or directory), then release
5174 * the stripe objects references and free the ldo_stripe array.
5176 * \param[in] env execution environment
5177 * \param[in] lo object
5179 void lod_object_free_striping(const struct lu_env *env, struct lod_object *lo)
5181 struct lod_layout_component *lod_comp;
5184 if (lo->ldo_stripe != NULL) {
5185 LASSERT(lo->ldo_comp_entries == NULL);
5186 LASSERT(lo->ldo_dir_stripes_allocated > 0);
5188 for (i = 0; i < lo->ldo_dir_stripenr; i++) {
5189 if (lo->ldo_stripe[i])
5190 dt_object_put(env, lo->ldo_stripe[i]);
5193 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
5194 OBD_FREE(lo->ldo_stripe, j);
5195 lo->ldo_stripe = NULL;
5196 lo->ldo_dir_stripes_allocated = 0;
5197 lo->ldo_dir_stripenr = 0;
5198 } else if (lo->ldo_comp_entries != NULL) {
5199 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5200 /* free lod_layout_component::llc_stripe array */
5201 lod_comp = &lo->ldo_comp_entries[i];
5203 if (lod_comp->llc_stripe == NULL)
5205 LASSERT(lod_comp->llc_stripes_allocated != 0);
5206 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
5207 if (lod_comp->llc_stripe[j] != NULL)
5209 &lod_comp->llc_stripe[j]->do_lu);
5211 OBD_FREE(lod_comp->llc_stripe,
5212 sizeof(struct dt_object *) *
5213 lod_comp->llc_stripes_allocated);
5214 lod_comp->llc_stripe = NULL;
5215 lod_comp->llc_stripes_allocated = 0;
5217 lod_free_comp_entries(lo);
5218 lo->ldo_comp_cached = 0;
5223 * Implementation of lu_object_operations::loo_object_start.
5225 * \see lu_object_operations::loo_object_start() in the API description
5228 static int lod_object_start(const struct lu_env *env, struct lu_object *o)
5230 if (S_ISLNK(o->lo_header->loh_attr & S_IFMT)) {
5231 lu2lod_obj(o)->ldo_obj.do_body_ops = &lod_body_lnk_ops;
5232 } else if (S_ISREG(o->lo_header->loh_attr & S_IFMT) ||
5233 fid_is_local_file(lu_object_fid(o))) {
5234 /* Note: some local file (like last rcvd) is created
5235 * through bottom layer (OSD), so the object initialization
5236 * comes to lod, it does not set loh_attr yet, so
5237 * set do_body_ops for local file anyway */
5238 lu2lod_obj(o)->ldo_obj.do_body_ops = &lod_body_ops;
5244 * Implementation of lu_object_operations::loo_object_free.
5246 * \see lu_object_operations::loo_object_free() in the API description
5249 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
5251 struct lod_object *lo = lu2lod_obj(o);
5253 /* release all underlying object pinned */
5254 lod_object_free_striping(env, lo);
5256 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
5260 * Implementation of lu_object_operations::loo_object_release.
5262 * \see lu_object_operations::loo_object_release() in the API description
5265 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
5267 /* XXX: shouldn't we release everything here in case if object
5268 * creation failed before? */
5272 * Implementation of lu_object_operations::loo_object_print.
5274 * \see lu_object_operations::loo_object_print() in the API description
5277 static int lod_object_print(const struct lu_env *env, void *cookie,
5278 lu_printer_t p, const struct lu_object *l)
5280 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
5282 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
5285 struct lu_object_operations lod_lu_obj_ops = {
5286 .loo_object_init = lod_object_init,
5287 .loo_object_start = lod_object_start,
5288 .loo_object_free = lod_object_free,
5289 .loo_object_release = lod_object_release,
5290 .loo_object_print = lod_object_print,