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 <uapi/linux/lustre/lustre_param.h>
51 #include <lustre_swab.h>
52 #include <uapi/linux/lustre/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[] = "..";
62 * Implementation of dt_index_operations::dio_lookup
64 * Used with regular (non-striped) objects.
66 * \see dt_index_operations::dio_lookup() in the API description for details.
68 static int lod_lookup(const struct lu_env *env, struct dt_object *dt,
69 struct dt_rec *rec, const struct dt_key *key)
71 struct dt_object *next = dt_object_child(dt);
72 return next->do_index_ops->dio_lookup(env, next, rec, key);
76 * Implementation of dt_index_operations::dio_declare_insert.
78 * Used with regular (non-striped) objects.
80 * \see dt_index_operations::dio_declare_insert() in the API description
83 static int lod_declare_insert(const struct lu_env *env, struct dt_object *dt,
84 const struct dt_rec *rec,
85 const struct dt_key *key, struct thandle *th)
87 return lod_sub_declare_insert(env, dt_object_child(dt), rec, key, th);
91 * Implementation of dt_index_operations::dio_insert.
93 * Used with regular (non-striped) objects
95 * \see dt_index_operations::dio_insert() in the API description for details.
97 static int lod_insert(const struct lu_env *env, struct dt_object *dt,
98 const struct dt_rec *rec, const struct dt_key *key,
99 struct thandle *th, int ign)
101 return lod_sub_insert(env, dt_object_child(dt), rec, key, th, ign);
105 * Implementation of dt_index_operations::dio_declare_delete.
107 * Used with regular (non-striped) objects.
109 * \see dt_index_operations::dio_declare_delete() in the API description
112 static int lod_declare_delete(const struct lu_env *env, struct dt_object *dt,
113 const struct dt_key *key, struct thandle *th)
115 return lod_sub_declare_delete(env, dt_object_child(dt), key, th);
119 * Implementation of dt_index_operations::dio_delete.
121 * Used with regular (non-striped) objects.
123 * \see dt_index_operations::dio_delete() in the API description for details.
125 static int lod_delete(const struct lu_env *env, struct dt_object *dt,
126 const struct dt_key *key, struct thandle *th)
128 return lod_sub_delete(env, dt_object_child(dt), key, th);
132 * Implementation of dt_it_ops::init.
134 * Used with regular (non-striped) objects.
136 * \see dt_it_ops::init() in the API description for details.
138 static struct dt_it *lod_it_init(const struct lu_env *env,
139 struct dt_object *dt, __u32 attr)
141 struct dt_object *next = dt_object_child(dt);
142 struct lod_it *it = &lod_env_info(env)->lti_it;
143 struct dt_it *it_next;
145 it_next = next->do_index_ops->dio_it.init(env, next, attr);
149 /* currently we do not use more than one iterator per thread
150 * so we store it in thread info. if at some point we need
151 * more active iterators in a single thread, we can allocate
153 LASSERT(it->lit_obj == NULL);
155 it->lit_it = it_next;
158 return (struct dt_it *)it;
161 #define LOD_CHECK_IT(env, it) \
163 LASSERT((it)->lit_obj != NULL); \
164 LASSERT((it)->lit_it != NULL); \
168 * Implementation of dt_index_operations::dio_it.fini.
170 * Used with regular (non-striped) objects.
172 * \see dt_index_operations::dio_it.fini() in the API description for details.
174 static void lod_it_fini(const struct lu_env *env, struct dt_it *di)
176 struct lod_it *it = (struct lod_it *)di;
178 LOD_CHECK_IT(env, it);
179 it->lit_obj->do_index_ops->dio_it.fini(env, it->lit_it);
181 /* the iterator not in use any more */
187 * Implementation of dt_it_ops::get.
189 * Used with regular (non-striped) objects.
191 * \see dt_it_ops::get() in the API description for details.
193 static int lod_it_get(const struct lu_env *env, struct dt_it *di,
194 const struct dt_key *key)
196 const struct lod_it *it = (const struct lod_it *)di;
198 LOD_CHECK_IT(env, it);
199 return it->lit_obj->do_index_ops->dio_it.get(env, it->lit_it, key);
203 * Implementation of dt_it_ops::put.
205 * Used with regular (non-striped) objects.
207 * \see dt_it_ops::put() in the API description for details.
209 static void lod_it_put(const struct lu_env *env, struct dt_it *di)
211 struct lod_it *it = (struct lod_it *)di;
213 LOD_CHECK_IT(env, it);
214 return it->lit_obj->do_index_ops->dio_it.put(env, it->lit_it);
218 * Implementation of dt_it_ops::next.
220 * Used with regular (non-striped) objects
222 * \see dt_it_ops::next() in the API description for details.
224 static int lod_it_next(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.next(env, it->lit_it);
233 * Implementation of dt_it_ops::key.
235 * Used with regular (non-striped) objects.
237 * \see dt_it_ops::key() in the API description for details.
239 static struct dt_key *lod_it_key(const struct lu_env *env,
240 const struct dt_it *di)
242 const struct lod_it *it = (const struct lod_it *)di;
244 LOD_CHECK_IT(env, it);
245 return it->lit_obj->do_index_ops->dio_it.key(env, it->lit_it);
249 * Implementation of dt_it_ops::key_size.
251 * Used with regular (non-striped) objects.
253 * \see dt_it_ops::key_size() in the API description for details.
255 static int lod_it_key_size(const struct lu_env *env, const struct dt_it *di)
257 struct lod_it *it = (struct lod_it *)di;
259 LOD_CHECK_IT(env, it);
260 return it->lit_obj->do_index_ops->dio_it.key_size(env, it->lit_it);
264 * Implementation of dt_it_ops::rec.
266 * Used with regular (non-striped) objects.
268 * \see dt_it_ops::rec() in the API description for details.
270 static int lod_it_rec(const struct lu_env *env, const struct dt_it *di,
271 struct dt_rec *rec, __u32 attr)
273 const struct lod_it *it = (const struct lod_it *)di;
275 LOD_CHECK_IT(env, it);
276 return it->lit_obj->do_index_ops->dio_it.rec(env, it->lit_it, rec,
281 * Implementation of dt_it_ops::rec_size.
283 * Used with regular (non-striped) objects.
285 * \see dt_it_ops::rec_size() in the API description for details.
287 static int lod_it_rec_size(const struct lu_env *env, const struct dt_it *di,
290 const struct lod_it *it = (const struct lod_it *)di;
292 LOD_CHECK_IT(env, it);
293 return it->lit_obj->do_index_ops->dio_it.rec_size(env, it->lit_it,
298 * Implementation of dt_it_ops::store.
300 * Used with regular (non-striped) objects.
302 * \see dt_it_ops::store() in the API description for details.
304 static __u64 lod_it_store(const struct lu_env *env, const struct dt_it *di)
306 const struct lod_it *it = (const struct lod_it *)di;
308 LOD_CHECK_IT(env, it);
309 return it->lit_obj->do_index_ops->dio_it.store(env, it->lit_it);
313 * Implementation of dt_it_ops::load.
315 * Used with regular (non-striped) objects.
317 * \see dt_it_ops::load() in the API description for details.
319 static int lod_it_load(const struct lu_env *env, const struct dt_it *di,
322 const struct lod_it *it = (const struct lod_it *)di;
324 LOD_CHECK_IT(env, it);
325 return it->lit_obj->do_index_ops->dio_it.load(env, it->lit_it, hash);
329 * Implementation of dt_it_ops::key_rec.
331 * Used with regular (non-striped) objects.
333 * \see dt_it_ops::rec() in the API description for details.
335 static int lod_it_key_rec(const struct lu_env *env, const struct dt_it *di,
338 const struct lod_it *it = (const struct lod_it *)di;
340 LOD_CHECK_IT(env, it);
341 return it->lit_obj->do_index_ops->dio_it.key_rec(env, it->lit_it,
345 static struct dt_index_operations lod_index_ops = {
346 .dio_lookup = lod_lookup,
347 .dio_declare_insert = lod_declare_insert,
348 .dio_insert = lod_insert,
349 .dio_declare_delete = lod_declare_delete,
350 .dio_delete = lod_delete,
358 .key_size = lod_it_key_size,
360 .rec_size = lod_it_rec_size,
361 .store = lod_it_store,
363 .key_rec = lod_it_key_rec,
368 * Implementation of dt_it_ops::init.
370 * Used with striped objects. Internally just initializes the iterator
371 * on the first stripe.
373 * \see dt_it_ops::init() in the API description for details.
375 static struct dt_it *lod_striped_it_init(const struct lu_env *env,
376 struct dt_object *dt, __u32 attr)
378 struct lod_object *lo = lod_dt_obj(dt);
379 struct dt_object *next;
380 struct lod_it *it = &lod_env_info(env)->lti_it;
381 struct dt_it *it_next;
384 LASSERT(lo->ldo_dir_stripe_count > 0);
385 next = lo->ldo_stripe[0];
386 LASSERT(next != NULL);
387 LASSERT(next->do_index_ops != NULL);
389 it_next = next->do_index_ops->dio_it.init(env, next, attr);
393 /* currently we do not use more than one iterator per thread
394 * so we store it in thread info. if at some point we need
395 * more active iterators in a single thread, we can allocate
397 LASSERT(it->lit_obj == NULL);
399 it->lit_stripe_index = 0;
401 it->lit_it = it_next;
404 return (struct dt_it *)it;
407 #define LOD_CHECK_STRIPED_IT(env, it, lo) \
409 LASSERT((it)->lit_obj != NULL); \
410 LASSERT((it)->lit_it != NULL); \
411 LASSERT((lo)->ldo_dir_stripe_count > 0); \
412 LASSERT((it)->lit_stripe_index < (lo)->ldo_dir_stripe_count); \
416 * Implementation of dt_it_ops::fini.
418 * Used with striped objects.
420 * \see dt_it_ops::fini() in the API description for details.
422 static void lod_striped_it_fini(const struct lu_env *env, struct dt_it *di)
424 struct lod_it *it = (struct lod_it *)di;
425 struct lod_object *lo = lod_dt_obj(it->lit_obj);
426 struct dt_object *next;
428 /* If lit_it == NULL, then it means the sub_it has been finished,
429 * which only happens in failure cases, see lod_striped_it_next() */
430 if (it->lit_it != NULL) {
431 LOD_CHECK_STRIPED_IT(env, it, lo);
433 next = lo->ldo_stripe[it->lit_stripe_index];
434 LASSERT(next != NULL);
435 LASSERT(next->do_index_ops != NULL);
437 next->do_index_ops->dio_it.fini(env, it->lit_it);
440 /* the iterator not in use any more */
443 it->lit_stripe_index = 0;
447 * Implementation of dt_it_ops::get.
449 * Right now it's not used widely, only to reset the iterator to the
450 * initial position. It should be possible to implement a full version
451 * which chooses a correct stripe to be able to position with any key.
453 * \see dt_it_ops::get() in the API description for details.
455 static int lod_striped_it_get(const struct lu_env *env, struct dt_it *di,
456 const struct dt_key *key)
458 const struct lod_it *it = (const struct lod_it *)di;
459 struct lod_object *lo = lod_dt_obj(it->lit_obj);
460 struct dt_object *next;
463 LOD_CHECK_STRIPED_IT(env, it, lo);
465 next = lo->ldo_stripe[it->lit_stripe_index];
466 LASSERT(next != NULL);
467 LASSERT(next->do_index_ops != NULL);
469 return next->do_index_ops->dio_it.get(env, it->lit_it, key);
473 * Implementation of dt_it_ops::put.
475 * Used with striped objects.
477 * \see dt_it_ops::put() in the API description for details.
479 static void lod_striped_it_put(const struct lu_env *env, struct dt_it *di)
481 struct lod_it *it = (struct lod_it *)di;
482 struct lod_object *lo = lod_dt_obj(it->lit_obj);
483 struct dt_object *next;
485 LOD_CHECK_STRIPED_IT(env, it, lo);
487 next = lo->ldo_stripe[it->lit_stripe_index];
488 LASSERT(next != NULL);
489 LASSERT(next->do_index_ops != NULL);
491 return next->do_index_ops->dio_it.put(env, it->lit_it);
495 * Implementation of dt_it_ops::next.
497 * Used with striped objects. When the end of the current stripe is
498 * reached, the method takes the next stripe's iterator.
500 * \see dt_it_ops::next() in the API description for details.
502 static int lod_striped_it_next(const struct lu_env *env, struct dt_it *di)
504 struct lod_it *it = (struct lod_it *)di;
505 struct lod_object *lo = lod_dt_obj(it->lit_obj);
506 struct dt_object *next;
507 struct dt_it *it_next;
511 LOD_CHECK_STRIPED_IT(env, it, lo);
513 next = lo->ldo_stripe[it->lit_stripe_index];
514 LASSERT(next != NULL);
515 LASSERT(next->do_index_ops != NULL);
517 rc = next->do_index_ops->dio_it.next(env, it->lit_it);
521 if (rc == 0 && it->lit_stripe_index == 0)
524 if (rc == 0 && it->lit_stripe_index > 0) {
525 struct lu_dirent *ent;
527 ent = (struct lu_dirent *)lod_env_info(env)->lti_key;
529 rc = next->do_index_ops->dio_it.rec(env, it->lit_it,
530 (struct dt_rec *)ent,
535 /* skip . and .. for slave stripe */
536 if ((strncmp(ent->lde_name, ".",
537 le16_to_cpu(ent->lde_namelen)) == 0 &&
538 le16_to_cpu(ent->lde_namelen) == 1) ||
539 (strncmp(ent->lde_name, "..",
540 le16_to_cpu(ent->lde_namelen)) == 0 &&
541 le16_to_cpu(ent->lde_namelen) == 2))
547 /* go to next stripe */
548 if (it->lit_stripe_index + 1 >= lo->ldo_dir_stripe_count)
551 it->lit_stripe_index++;
553 next->do_index_ops->dio_it.put(env, it->lit_it);
554 next->do_index_ops->dio_it.fini(env, it->lit_it);
557 next = lo->ldo_stripe[it->lit_stripe_index];
558 LASSERT(next != NULL);
559 rc = next->do_ops->do_index_try(env, next, &dt_directory_features);
563 LASSERT(next->do_index_ops != NULL);
565 it_next = next->do_index_ops->dio_it.init(env, next, it->lit_attr);
566 if (!IS_ERR(it_next)) {
567 it->lit_it = it_next;
570 rc = PTR_ERR(it_next);
577 * Implementation of dt_it_ops::key.
579 * Used with striped objects.
581 * \see dt_it_ops::key() in the API description for details.
583 static struct dt_key *lod_striped_it_key(const struct lu_env *env,
584 const struct dt_it *di)
586 const struct lod_it *it = (const struct lod_it *)di;
587 struct lod_object *lo = lod_dt_obj(it->lit_obj);
588 struct dt_object *next;
590 LOD_CHECK_STRIPED_IT(env, it, lo);
592 next = lo->ldo_stripe[it->lit_stripe_index];
593 LASSERT(next != NULL);
594 LASSERT(next->do_index_ops != NULL);
596 return next->do_index_ops->dio_it.key(env, it->lit_it);
600 * Implementation of dt_it_ops::key_size.
602 * Used with striped objects.
604 * \see dt_it_ops::size() in the API description for details.
606 static int lod_striped_it_key_size(const struct lu_env *env,
607 const struct dt_it *di)
609 struct lod_it *it = (struct lod_it *)di;
610 struct lod_object *lo = lod_dt_obj(it->lit_obj);
611 struct dt_object *next;
613 LOD_CHECK_STRIPED_IT(env, it, lo);
615 next = lo->ldo_stripe[it->lit_stripe_index];
616 LASSERT(next != NULL);
617 LASSERT(next->do_index_ops != NULL);
619 return next->do_index_ops->dio_it.key_size(env, it->lit_it);
623 * Implementation of dt_it_ops::rec.
625 * Used with striped objects.
627 * \see dt_it_ops::rec() in the API description for details.
629 static int lod_striped_it_rec(const struct lu_env *env, const struct dt_it *di,
630 struct dt_rec *rec, __u32 attr)
632 const struct lod_it *it = (const struct lod_it *)di;
633 struct lod_object *lo = lod_dt_obj(it->lit_obj);
634 struct dt_object *next;
636 LOD_CHECK_STRIPED_IT(env, it, lo);
638 next = lo->ldo_stripe[it->lit_stripe_index];
639 LASSERT(next != NULL);
640 LASSERT(next->do_index_ops != NULL);
642 return next->do_index_ops->dio_it.rec(env, it->lit_it, rec, attr);
646 * Implementation of dt_it_ops::rec_size.
648 * Used with striped objects.
650 * \see dt_it_ops::rec_size() in the API description for details.
652 static int lod_striped_it_rec_size(const struct lu_env *env,
653 const struct dt_it *di, __u32 attr)
655 struct lod_it *it = (struct lod_it *)di;
656 struct lod_object *lo = lod_dt_obj(it->lit_obj);
657 struct dt_object *next;
659 LOD_CHECK_STRIPED_IT(env, it, lo);
661 next = lo->ldo_stripe[it->lit_stripe_index];
662 LASSERT(next != NULL);
663 LASSERT(next->do_index_ops != NULL);
665 return next->do_index_ops->dio_it.rec_size(env, it->lit_it, attr);
669 * Implementation of dt_it_ops::store.
671 * Used with striped objects.
673 * \see dt_it_ops::store() in the API description for details.
675 static __u64 lod_striped_it_store(const struct lu_env *env,
676 const struct dt_it *di)
678 const struct lod_it *it = (const struct lod_it *)di;
679 struct lod_object *lo = lod_dt_obj(it->lit_obj);
680 struct dt_object *next;
682 LOD_CHECK_STRIPED_IT(env, it, lo);
684 next = lo->ldo_stripe[it->lit_stripe_index];
685 LASSERT(next != NULL);
686 LASSERT(next->do_index_ops != NULL);
688 return next->do_index_ops->dio_it.store(env, it->lit_it);
692 * Implementation of dt_it_ops::load.
694 * Used with striped objects.
696 * \see dt_it_ops::load() in the API description for details.
698 static int lod_striped_it_load(const struct lu_env *env,
699 const struct dt_it *di, __u64 hash)
701 const struct lod_it *it = (const struct lod_it *)di;
702 struct lod_object *lo = lod_dt_obj(it->lit_obj);
703 struct dt_object *next;
705 LOD_CHECK_STRIPED_IT(env, it, lo);
707 next = lo->ldo_stripe[it->lit_stripe_index];
708 LASSERT(next != NULL);
709 LASSERT(next->do_index_ops != NULL);
711 return next->do_index_ops->dio_it.load(env, it->lit_it, hash);
714 static struct dt_index_operations lod_striped_index_ops = {
715 .dio_lookup = lod_lookup,
716 .dio_declare_insert = lod_declare_insert,
717 .dio_insert = lod_insert,
718 .dio_declare_delete = lod_declare_delete,
719 .dio_delete = lod_delete,
721 .init = lod_striped_it_init,
722 .fini = lod_striped_it_fini,
723 .get = lod_striped_it_get,
724 .put = lod_striped_it_put,
725 .next = lod_striped_it_next,
726 .key = lod_striped_it_key,
727 .key_size = lod_striped_it_key_size,
728 .rec = lod_striped_it_rec,
729 .rec_size = lod_striped_it_rec_size,
730 .store = lod_striped_it_store,
731 .load = lod_striped_it_load,
736 * Append the FID for each shard of the striped directory after the
737 * given LMV EA header.
739 * To simplify striped directory and the consistency verification,
740 * we only store the LMV EA header on disk, for both master object
741 * and slave objects. When someone wants to know the whole LMV EA,
742 * such as client readdir(), we can build the entrie LMV EA on the
743 * MDT side (in RAM) via iterating the sub-directory entries that
744 * are contained in the master object of the stripe directory.
746 * For the master object of the striped directroy, the valid name
747 * for each shard is composed of the ${shard_FID}:${shard_idx}.
749 * There may be holes in the LMV EA if some shards' name entries
750 * are corrupted or lost.
752 * \param[in] env pointer to the thread context
753 * \param[in] lo pointer to the master object of the striped directory
754 * \param[in] buf pointer to the lu_buf which will hold the LMV EA
755 * \param[in] resize whether re-allocate the buffer if it is not big enough
757 * \retval positive size of the LMV EA
758 * \retval 0 for nothing to be loaded
759 * \retval negative error number on failure
761 int lod_load_lmv_shards(const struct lu_env *env, struct lod_object *lo,
762 struct lu_buf *buf, bool resize)
764 struct lu_dirent *ent =
765 (struct lu_dirent *)lod_env_info(env)->lti_key;
766 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
767 struct dt_object *obj = dt_object_child(&lo->ldo_obj);
768 struct lmv_mds_md_v1 *lmv1 = buf->lb_buf;
770 const struct dt_it_ops *iops;
772 __u32 magic = le32_to_cpu(lmv1->lmv_magic);
777 /* If it is not a striped directory, then load nothing. */
778 if (magic != LMV_MAGIC_V1)
781 /* If it is in migration (or failure), then load nothing. */
782 if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
785 stripes = le32_to_cpu(lmv1->lmv_stripe_count);
789 rc = lmv_mds_md_size(stripes, magic);
793 if (buf->lb_len < lmv1_size) {
802 lu_buf_alloc(buf, lmv1_size);
807 memcpy(buf->lb_buf, tbuf.lb_buf, tbuf.lb_len);
810 if (unlikely(!dt_try_as_dir(env, obj)))
813 memset(&lmv1->lmv_stripe_fids[0], 0, stripes * sizeof(struct lu_fid));
814 iops = &obj->do_index_ops->dio_it;
815 it = iops->init(env, obj, LUDA_64BITHASH);
819 rc = iops->load(env, it, 0);
821 rc = iops->next(env, it);
826 char name[FID_LEN + 2] = "";
831 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
837 fid_le_to_cpu(&fid, &ent->lde_fid);
838 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
839 if (ent->lde_name[0] == '.') {
840 if (ent->lde_namelen == 1)
843 if (ent->lde_namelen == 2 && ent->lde_name[1] == '.')
847 len = snprintf(name, sizeof(name),
848 DFID":", PFID(&ent->lde_fid));
849 /* The ent->lde_name is composed of ${FID}:${index} */
850 if (ent->lde_namelen < len + 1 ||
851 memcmp(ent->lde_name, name, len) != 0) {
852 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
853 "%s: invalid shard name %.*s with the FID "DFID
854 " for the striped directory "DFID", %s\n",
855 lod2obd(lod)->obd_name, ent->lde_namelen,
856 ent->lde_name, PFID(&fid),
857 PFID(lu_object_fid(&obj->do_lu)),
858 lod->lod_lmv_failout ? "failout" : "skip");
860 if (lod->lod_lmv_failout)
868 if (ent->lde_name[len] < '0' ||
869 ent->lde_name[len] > '9') {
870 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
871 "%s: invalid shard name %.*s with the "
872 "FID "DFID" for the striped directory "
874 lod2obd(lod)->obd_name, ent->lde_namelen,
875 ent->lde_name, PFID(&fid),
876 PFID(lu_object_fid(&obj->do_lu)),
877 lod->lod_lmv_failout ?
880 if (lod->lod_lmv_failout)
886 index = index * 10 + ent->lde_name[len++] - '0';
887 } while (len < ent->lde_namelen);
889 if (len == ent->lde_namelen) {
890 /* Out of LMV EA range. */
891 if (index >= stripes) {
892 CERROR("%s: the shard %.*s for the striped "
893 "directory "DFID" is out of the known "
894 "LMV EA range [0 - %u], failout\n",
895 lod2obd(lod)->obd_name, ent->lde_namelen,
897 PFID(lu_object_fid(&obj->do_lu)),
903 /* The slot has been occupied. */
904 if (!fid_is_zero(&lmv1->lmv_stripe_fids[index])) {
908 &lmv1->lmv_stripe_fids[index]);
909 CERROR("%s: both the shard "DFID" and "DFID
910 " for the striped directory "DFID
911 " claim the same LMV EA slot at the "
912 "index %d, failout\n",
913 lod2obd(lod)->obd_name,
914 PFID(&fid0), PFID(&fid),
915 PFID(lu_object_fid(&obj->do_lu)), index);
920 /* stored as LE mode */
921 lmv1->lmv_stripe_fids[index] = ent->lde_fid;
924 rc = iops->next(env, it);
931 RETURN(rc > 0 ? lmv_mds_md_size(stripes, magic) : rc);
935 * Implementation of dt_object_operations::do_index_try.
937 * \see dt_object_operations::do_index_try() in the API description for details.
939 static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
940 const struct dt_index_features *feat)
942 struct lod_object *lo = lod_dt_obj(dt);
943 struct dt_object *next = dt_object_child(dt);
947 LASSERT(next->do_ops);
948 LASSERT(next->do_ops->do_index_try);
950 rc = lod_load_striping_locked(env, lo);
954 rc = next->do_ops->do_index_try(env, next, feat);
958 if (lo->ldo_dir_stripe_count > 0) {
961 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
962 if (dt_object_exists(lo->ldo_stripe[i]) == 0)
964 rc = lo->ldo_stripe[i]->do_ops->do_index_try(env,
965 lo->ldo_stripe[i], feat);
969 dt->do_index_ops = &lod_striped_index_ops;
971 dt->do_index_ops = &lod_index_ops;
978 * Implementation of dt_object_operations::do_read_lock.
980 * \see dt_object_operations::do_read_lock() in the API description for details.
982 static void lod_read_lock(const struct lu_env *env, struct dt_object *dt,
985 dt_read_lock(env, dt_object_child(dt), role);
989 * Implementation of dt_object_operations::do_write_lock.
991 * \see dt_object_operations::do_write_lock() in the API description for
994 static void lod_write_lock(const struct lu_env *env, struct dt_object *dt,
997 dt_write_lock(env, dt_object_child(dt), role);
1001 * Implementation of dt_object_operations::do_read_unlock.
1003 * \see dt_object_operations::do_read_unlock() in the API description for
1006 static void lod_read_unlock(const struct lu_env *env, struct dt_object *dt)
1008 dt_read_unlock(env, dt_object_child(dt));
1012 * Implementation of dt_object_operations::do_write_unlock.
1014 * \see dt_object_operations::do_write_unlock() in the API description for
1017 static void lod_write_unlock(const struct lu_env *env, struct dt_object *dt)
1019 dt_write_unlock(env, dt_object_child(dt));
1023 * Implementation of dt_object_operations::do_write_locked.
1025 * \see dt_object_operations::do_write_locked() in the API description for
1028 static int lod_write_locked(const struct lu_env *env, struct dt_object *dt)
1030 return dt_write_locked(env, dt_object_child(dt));
1034 * Implementation of dt_object_operations::do_attr_get.
1036 * \see dt_object_operations::do_attr_get() in the API description for details.
1038 static int lod_attr_get(const struct lu_env *env,
1039 struct dt_object *dt,
1040 struct lu_attr *attr)
1042 /* Note: for striped directory, client will merge attributes
1043 * from all of the sub-stripes see lmv_merge_attr(), and there
1044 * no MDD logic depend on directory nlink/size/time, so we can
1045 * always use master inode nlink and size for now. */
1046 return dt_attr_get(env, dt_object_child(dt), attr);
1049 int lod_obj_for_each_stripe(const struct lu_env *env, struct lod_object *lo,
1051 struct lod_obj_stripe_cb_data *data)
1053 struct lod_layout_component *lod_comp;
1057 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1058 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1059 lod_comp = &lo->ldo_comp_entries[i];
1061 if (lod_comp->llc_stripe == NULL)
1064 /* has stripe but not inited yet, this component has been
1065 * declared to be created, but hasn't created yet.
1067 if (!lod_comp_inited(lod_comp))
1070 if (data->locd_comp_skip_cb &&
1071 data->locd_comp_skip_cb(env, lo, i, data))
1074 LASSERT(lod_comp->llc_stripe_count > 0);
1075 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
1076 struct dt_object *dt = lod_comp->llc_stripe[j];
1080 rc = data->locd_stripe_cb(env, lo, dt, th, i, j, data);
1088 static bool lod_obj_attr_set_comp_skip_cb(const struct lu_env *env,
1089 struct lod_object *lo, int comp_idx,
1090 struct lod_obj_stripe_cb_data *data)
1092 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[comp_idx];
1093 bool skipped = false;
1095 if (!(data->locd_attr->la_valid & LA_LAYOUT_VERSION))
1098 switch (lo->ldo_flr_state) {
1099 case LCM_FL_WRITE_PENDING: {
1102 /* skip stale components */
1103 if (lod_comp->llc_flags & LCME_FL_STALE) {
1108 /* skip valid and overlapping components, therefore any
1109 * attempts to write overlapped components will never succeed
1110 * because client will get EINPROGRESS. */
1111 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1115 if (lo->ldo_comp_entries[i].llc_flags & LCME_FL_STALE)
1118 if (lu_extent_is_overlapped(&lod_comp->llc_extent,
1119 &lo->ldo_comp_entries[i].llc_extent)) {
1127 LASSERTF(0, "impossible: %d\n", lo->ldo_flr_state);
1128 case LCM_FL_SYNC_PENDING:
1132 CDEBUG(D_LAYOUT, DFID": %s to set component %x to version: %u\n",
1133 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
1134 skipped ? "skipped" : "chose", lod_comp->llc_id,
1135 data->locd_attr->la_layout_version);
1141 lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
1142 struct dt_object *dt, struct thandle *th,
1143 int comp_idx, int stripe_idx,
1144 struct lod_obj_stripe_cb_data *data)
1146 if (data->locd_declare)
1147 return lod_sub_declare_attr_set(env, dt, data->locd_attr, th);
1149 if (data->locd_attr->la_valid & LA_LAYOUT_VERSION) {
1150 CDEBUG(D_LAYOUT, DFID": set layout version: %u, comp_idx: %d\n",
1151 PFID(lu_object_fid(&dt->do_lu)),
1152 data->locd_attr->la_layout_version, comp_idx);
1155 return lod_sub_attr_set(env, dt, data->locd_attr, th);
1159 * Implementation of dt_object_operations::do_declare_attr_set.
1161 * If the object is striped, then apply the changes to all the stripes.
1163 * \see dt_object_operations::do_declare_attr_set() in the API description
1166 static int lod_declare_attr_set(const struct lu_env *env,
1167 struct dt_object *dt,
1168 const struct lu_attr *attr,
1171 struct dt_object *next = dt_object_child(dt);
1172 struct lod_object *lo = lod_dt_obj(dt);
1177 * declare setattr on the local object
1179 rc = lod_sub_declare_attr_set(env, next, attr, th);
1183 /* osp_declare_attr_set() ignores all attributes other than
1184 * UID, GID, PROJID, and size, and osp_attr_set() ignores all
1185 * but UID, GID and PROJID. Declaration of size attr setting
1186 * happens through lod_declare_init_size(), and not through
1187 * this function. Therefore we need not load striping unless
1188 * ownership is changing. This should save memory and (we hope)
1189 * speed up rename().
1191 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1192 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1195 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1198 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID | LA_MODE |
1199 LA_ATIME | LA_MTIME | LA_CTIME |
1204 * load striping information, notice we don't do this when object
1205 * is being initialized as we don't need this information till
1206 * few specific cases like destroy, chown
1208 rc = lod_load_striping(env, lo);
1212 if (!lod_obj_is_striped(dt))
1216 * if object is striped declare changes on the stripes
1218 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1219 LASSERT(lo->ldo_stripe);
1220 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1221 if (lo->ldo_stripe[i] == NULL)
1223 rc = lod_sub_declare_attr_set(env, lo->ldo_stripe[i],
1229 struct lod_obj_stripe_cb_data data = { { 0 } };
1231 data.locd_attr = attr;
1232 data.locd_declare = true;
1233 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1234 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1240 if (!dt_object_exists(next) || dt_object_remote(next) ||
1241 !S_ISREG(attr->la_mode))
1244 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1245 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
1249 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) ||
1250 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1251 struct lod_thread_info *info = lod_env_info(env);
1252 struct lu_buf *buf = &info->lti_buf;
1254 buf->lb_buf = info->lti_ea_store;
1255 buf->lb_len = info->lti_ea_store_size;
1256 rc = lod_sub_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1257 LU_XATTR_REPLACE, th);
1264 * Implementation of dt_object_operations::do_attr_set.
1266 * If the object is striped, then apply the changes to all or subset of
1267 * the stripes depending on the object type and specific attributes.
1269 * \see dt_object_operations::do_attr_set() in the API description for details.
1271 static int lod_attr_set(const struct lu_env *env,
1272 struct dt_object *dt,
1273 const struct lu_attr *attr,
1276 struct dt_object *next = dt_object_child(dt);
1277 struct lod_object *lo = lod_dt_obj(dt);
1282 * apply changes to the local object
1284 rc = lod_sub_attr_set(env, next, attr, th);
1288 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1289 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1292 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1295 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE | LA_PROJID |
1296 LA_ATIME | LA_MTIME | LA_CTIME |
1301 /* FIXME: a tricky case in the code path of mdd_layout_change():
1302 * the in-memory striping information has been freed in lod_xattr_set()
1303 * due to layout change. It has to load stripe here again. It only
1304 * changes flags of layout so declare_attr_set() is still accurate */
1305 rc = lod_load_striping_locked(env, lo);
1309 if (!lod_obj_is_striped(dt))
1313 * if object is striped, apply changes to all the stripes
1315 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1316 LASSERT(lo->ldo_stripe);
1317 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1318 if (unlikely(lo->ldo_stripe[i] == NULL))
1321 if ((dt_object_exists(lo->ldo_stripe[i]) == 0))
1324 rc = lod_sub_attr_set(env, lo->ldo_stripe[i], attr, th);
1329 struct lod_obj_stripe_cb_data data = { { 0 } };
1331 data.locd_attr = attr;
1332 data.locd_declare = false;
1333 data.locd_comp_skip_cb = lod_obj_attr_set_comp_skip_cb;
1334 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1335 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1341 if (!dt_object_exists(next) || dt_object_remote(next) ||
1342 !S_ISREG(attr->la_mode))
1345 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1346 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
1350 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE)) {
1351 struct lod_thread_info *info = lod_env_info(env);
1352 struct lu_buf *buf = &info->lti_buf;
1353 struct ost_id *oi = &info->lti_ostid;
1354 struct lu_fid *fid = &info->lti_fid;
1355 struct lov_mds_md_v1 *lmm;
1356 struct lov_ost_data_v1 *objs;
1359 rc = lod_get_lov_ea(env, lo);
1363 buf->lb_buf = info->lti_ea_store;
1364 buf->lb_len = info->lti_ea_store_size;
1365 lmm = info->lti_ea_store;
1366 magic = le32_to_cpu(lmm->lmm_magic);
1367 if (magic == LOV_MAGIC_COMP_V1) {
1368 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1369 struct lov_comp_md_entry_v1 *lcme =
1370 &lcm->lcm_entries[0];
1372 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
1373 magic = le32_to_cpu(lmm->lmm_magic);
1376 if (magic == LOV_MAGIC_V1)
1377 objs = &(lmm->lmm_objects[0]);
1379 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1380 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1381 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1383 fid_to_ostid(fid, oi);
1384 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1386 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1387 LU_XATTR_REPLACE, th);
1388 } else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1389 struct lod_thread_info *info = lod_env_info(env);
1390 struct lu_buf *buf = &info->lti_buf;
1391 struct lov_comp_md_v1 *lcm;
1392 struct lov_comp_md_entry_v1 *lcme;
1394 rc = lod_get_lov_ea(env, lo);
1398 buf->lb_buf = info->lti_ea_store;
1399 buf->lb_len = info->lti_ea_store_size;
1401 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
1404 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1405 lcme = &lcm->lcm_entries[0];
1406 le64_add_cpu(&lcme->lcme_extent.e_start, 1);
1407 le64_add_cpu(&lcme->lcme_extent.e_end, -1);
1409 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1410 LU_XATTR_REPLACE, th);
1417 * Implementation of dt_object_operations::do_xattr_get.
1419 * If LOV EA is requested from the root object and it's not
1420 * found, then return default striping for the filesystem.
1422 * \see dt_object_operations::do_xattr_get() in the API description for details.
1424 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1425 struct lu_buf *buf, const char *name)
1427 struct lod_thread_info *info = lod_env_info(env);
1428 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1433 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1434 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1435 struct lmv_mds_md_v1 *lmv1;
1438 if (rc > (typeof(rc))sizeof(*lmv1))
1441 if (rc < (typeof(rc))sizeof(*lmv1))
1442 RETURN(rc = rc > 0 ? -EINVAL : rc);
1444 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1445 CLASSERT(sizeof(*lmv1) <= sizeof(info->lti_key));
1447 info->lti_buf.lb_buf = info->lti_key;
1448 info->lti_buf.lb_len = sizeof(*lmv1);
1449 rc = dt_xattr_get(env, dt_object_child(dt),
1450 &info->lti_buf, name);
1451 if (unlikely(rc != sizeof(*lmv1)))
1452 RETURN(rc = rc > 0 ? -EINVAL : rc);
1454 lmv1 = info->lti_buf.lb_buf;
1455 /* The on-disk LMV EA only contains header, but the
1456 * returned LMV EA size should contain the space for
1457 * the FIDs of all shards of the striped directory. */
1458 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1459 rc = lmv_mds_md_size(
1460 le32_to_cpu(lmv1->lmv_stripe_count),
1463 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1467 RETURN(rc = rc1 != 0 ? rc1 : rc);
1470 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1474 * XXX: Only used by lfsck
1476 * lod returns default striping on the real root of the device
1477 * this is like the root stores default striping for the whole
1478 * filesystem. historically we've been using a different approach
1479 * and store it in the config.
1481 dt_root_get(env, dev->lod_child, &info->lti_fid);
1482 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1484 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1485 struct lov_user_md *lum = buf->lb_buf;
1486 struct lov_desc *desc = &dev->lod_desc;
1488 if (buf->lb_buf == NULL) {
1490 } else if (buf->lb_len >= sizeof(*lum)) {
1491 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1492 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1493 lmm_oi_set_id(&lum->lmm_oi, 0);
1494 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1495 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1496 lum->lmm_stripe_size = cpu_to_le32(
1497 desc->ld_default_stripe_size);
1498 lum->lmm_stripe_count = cpu_to_le16(
1499 desc->ld_default_stripe_count);
1500 lum->lmm_stripe_offset = cpu_to_le16(
1501 desc->ld_default_stripe_offset);
1514 * Checks that the magic of the stripe is sane.
1516 * \param[in] lod lod device
1517 * \param[in] lum a buffer storing LMV EA to verify
1519 * \retval 0 if the EA is sane
1520 * \retval negative otherwise
1522 static int lod_verify_md_striping(struct lod_device *lod,
1523 const struct lmv_user_md_v1 *lum)
1525 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1526 CERROR("%s: invalid lmv_user_md: magic = %x, "
1527 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1528 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1529 (int)le32_to_cpu(lum->lum_stripe_offset),
1530 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1538 * Initialize LMV EA for a slave.
1540 * Initialize slave's LMV EA from the master's LMV EA.
1542 * \param[in] master_lmv a buffer containing master's EA
1543 * \param[out] slave_lmv a buffer where slave's EA will be stored
1546 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1547 const struct lmv_mds_md_v1 *master_lmv)
1549 *slave_lmv = *master_lmv;
1550 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1556 * Generate LMV EA from the object passed as \a dt. The object must have
1557 * the stripes created and initialized.
1559 * \param[in] env execution environment
1560 * \param[in] dt object
1561 * \param[out] lmv_buf buffer storing generated LMV EA
1563 * \retval 0 on success
1564 * \retval negative if failed
1566 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1567 struct lu_buf *lmv_buf)
1569 struct lod_thread_info *info = lod_env_info(env);
1570 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1571 struct lod_object *lo = lod_dt_obj(dt);
1572 struct lmv_mds_md_v1 *lmm1;
1574 int type = LU_SEQ_RANGE_ANY;
1579 LASSERT(lo->ldo_dir_striped != 0);
1580 LASSERT(lo->ldo_dir_stripe_count > 0);
1581 stripe_count = lo->ldo_dir_stripe_count;
1582 /* Only store the LMV EA heahder on the disk. */
1583 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1584 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1588 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1591 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1592 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1593 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1594 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1595 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1600 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1601 lmv_buf->lb_buf = info->lti_ea_store;
1602 lmv_buf->lb_len = sizeof(*lmm1);
1608 * Create in-core represenation for a striped directory.
1610 * Parse the buffer containing LMV EA and instantiate LU objects
1611 * representing the stripe objects. The pointers to the objects are
1612 * stored in ldo_stripe field of \a lo. This function is used when
1613 * we need to access an already created object (i.e. load from a disk).
1615 * \param[in] env execution environment
1616 * \param[in] lo lod object
1617 * \param[in] buf buffer containing LMV EA
1619 * \retval 0 on success
1620 * \retval negative if failed
1622 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1623 const struct lu_buf *buf)
1625 struct lod_thread_info *info = lod_env_info(env);
1626 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1627 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1628 struct dt_object **stripe;
1629 union lmv_mds_md *lmm = buf->lb_buf;
1630 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1631 struct lu_fid *fid = &info->lti_fid;
1636 if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
1639 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1640 lo->ldo_dir_slave_stripe = 1;
1644 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1647 if (le32_to_cpu(lmv1->lmv_stripe_count) < 1)
1650 LASSERT(lo->ldo_stripe == NULL);
1651 OBD_ALLOC(stripe, sizeof(stripe[0]) *
1652 (le32_to_cpu(lmv1->lmv_stripe_count)));
1656 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1657 struct dt_device *tgt_dt;
1658 struct dt_object *dto;
1659 int type = LU_SEQ_RANGE_ANY;
1662 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1663 if (!fid_is_sane(fid))
1664 GOTO(out, rc = -ESTALE);
1666 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1670 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1671 tgt_dt = lod->lod_child;
1673 struct lod_tgt_desc *tgt;
1675 tgt = LTD_TGT(ltd, idx);
1677 GOTO(out, rc = -ESTALE);
1678 tgt_dt = tgt->ltd_tgt;
1681 dto = dt_locate_at(env, tgt_dt, fid,
1682 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1685 GOTO(out, rc = PTR_ERR(dto));
1690 lo->ldo_stripe = stripe;
1691 lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1692 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1694 lod_object_free_striping(env, lo);
1700 * Declare create a striped directory.
1702 * Declare creating a striped directory with a given stripe pattern on the
1703 * specified MDTs. A striped directory is represented as a regular directory
1704 * - an index listing all the stripes. The stripes point back to the master
1705 * object with ".." and LinkEA. The master object gets LMV EA which
1706 * identifies it as a striped directory. The function allocates FIDs
1709 * \param[in] env execution environment
1710 * \param[in] dt object
1711 * \param[in] attr attributes to initialize the objects with
1712 * \param[in] dof type of objects to be created
1713 * \param[in] th transaction handle
1715 * \retval 0 on success
1716 * \retval negative if failed
1718 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1719 struct dt_object *dt,
1720 struct lu_attr *attr,
1721 struct dt_object_format *dof,
1724 struct lod_thread_info *info = lod_env_info(env);
1725 struct lu_buf lmv_buf;
1726 struct lu_buf slave_lmv_buf;
1727 struct lmv_mds_md_v1 *lmm;
1728 struct lmv_mds_md_v1 *slave_lmm = NULL;
1729 struct dt_insert_rec *rec = &info->lti_dt_rec;
1730 struct lod_object *lo = lod_dt_obj(dt);
1735 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1738 lmm = lmv_buf.lb_buf;
1740 OBD_ALLOC_PTR(slave_lmm);
1741 if (slave_lmm == NULL)
1742 GOTO(out, rc = -ENOMEM);
1744 lod_prep_slave_lmv_md(slave_lmm, lmm);
1745 slave_lmv_buf.lb_buf = slave_lmm;
1746 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1748 if (!dt_try_as_dir(env, dt_object_child(dt)))
1749 GOTO(out, rc = -EINVAL);
1751 rec->rec_type = S_IFDIR;
1752 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1753 struct dt_object *dto = lo->ldo_stripe[i];
1754 char *stripe_name = info->lti_key;
1755 struct lu_name *sname;
1756 struct linkea_data ldata = { NULL };
1757 struct lu_buf linkea_buf;
1759 rc = lod_sub_declare_create(env, dto, attr, NULL, dof, th);
1763 if (!dt_try_as_dir(env, dto))
1764 GOTO(out, rc = -EINVAL);
1766 rc = lod_sub_declare_ref_add(env, dto, th);
1770 rec->rec_fid = lu_object_fid(&dto->do_lu);
1771 rc = lod_sub_declare_insert(env, dto,
1772 (const struct dt_rec *)rec,
1773 (const struct dt_key *)dot, th);
1777 /* master stripe FID will be put to .. */
1778 rec->rec_fid = lu_object_fid(&dt->do_lu);
1779 rc = lod_sub_declare_insert(env, dto,
1780 (const struct dt_rec *)rec,
1781 (const struct dt_key *)dotdot, th);
1785 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1786 cfs_fail_val != i) {
1787 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1789 slave_lmm->lmv_master_mdt_index =
1792 slave_lmm->lmv_master_mdt_index =
1794 rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
1795 XATTR_NAME_LMV, 0, th);
1800 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1802 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1803 PFID(lu_object_fid(&dto->do_lu)), i + 1);
1805 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1806 PFID(lu_object_fid(&dto->do_lu)), i);
1808 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
1809 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1810 sname, lu_object_fid(&dt->do_lu));
1814 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1815 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1816 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
1817 XATTR_NAME_LINK, 0, th);
1821 rec->rec_fid = lu_object_fid(&dto->do_lu);
1822 rc = lod_sub_declare_insert(env, dt_object_child(dt),
1823 (const struct dt_rec *)rec,
1824 (const struct dt_key *)stripe_name,
1829 rc = lod_sub_declare_ref_add(env, dt_object_child(dt), th);
1834 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
1835 &lmv_buf, XATTR_NAME_LMV, 0, th);
1839 if (slave_lmm != NULL)
1840 OBD_FREE_PTR(slave_lmm);
1845 static int lod_prep_md_striped_create(const struct lu_env *env,
1846 struct dt_object *dt,
1847 struct lu_attr *attr,
1848 const struct lmv_user_md_v1 *lum,
1849 struct dt_object_format *dof,
1852 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1853 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1854 struct lod_object *lo = lod_dt_obj(dt);
1855 struct dt_object **stripe;
1864 /* The lum has been verifed in lod_verify_md_striping */
1865 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
1866 LASSERT(le32_to_cpu(lum->lum_stripe_count) > 0);
1868 stripe_count = le32_to_cpu(lum->lum_stripe_count);
1870 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
1871 if (idx_array == NULL)
1874 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_count);
1876 GOTO(out_free, rc = -ENOMEM);
1878 /* Start index must be the master MDT */
1879 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1880 idx_array[0] = master_index;
1881 for (i = 0; i < stripe_count; i++) {
1882 struct lod_tgt_desc *tgt = NULL;
1883 struct dt_object *dto;
1884 struct lu_fid fid = { 0 };
1886 struct lu_object_conf conf = { 0 };
1887 struct dt_device *tgt_dt = NULL;
1889 /* Try to find next avaible target */
1891 for (j = 0; j < lod->lod_remote_mdt_count;
1892 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
1893 bool already_allocated = false;
1896 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
1897 idx, lod->lod_remote_mdt_count + 1, i);
1899 if (likely(!OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
1900 /* check whether the idx already exists
1901 * in current allocated array */
1902 for (k = 0; k < i; k++) {
1903 if (idx_array[k] == idx) {
1904 already_allocated = true;
1909 if (already_allocated)
1913 /* Sigh, this index is not in the bitmap, let's check
1914 * next available target */
1915 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) &&
1916 idx != master_index)
1919 if (idx == master_index) {
1920 /* Allocate the FID locally */
1921 rc = obd_fid_alloc(env, lod->lod_child_exp,
1925 tgt_dt = lod->lod_child;
1929 /* check the status of the OSP */
1930 tgt = LTD_TGT(ltd, idx);
1934 tgt_dt = tgt->ltd_tgt;
1935 rc = dt_statfs(env, tgt_dt, NULL);
1937 /* this OSP doesn't feel well */
1942 rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
1951 /* Can not allocate more stripes */
1952 if (j == lod->lod_remote_mdt_count) {
1953 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
1954 lod2obd(lod)->obd_name, stripe_count, i);
1958 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
1959 idx, i, PFID(&fid));
1961 /* Set the start index for next stripe allocation */
1962 if (i < stripe_count - 1)
1963 idx_array[i + 1] = (idx + 1) %
1964 (lod->lod_remote_mdt_count + 1);
1965 /* tgt_dt and fid must be ready after search avaible OSP
1966 * in the above loop */
1967 LASSERT(tgt_dt != NULL);
1968 LASSERT(fid_is_sane(&fid));
1969 conf.loc_flags = LOC_F_NEW;
1970 dto = dt_locate_at(env, tgt_dt, &fid,
1971 dt->do_lu.lo_dev->ld_site->ls_top_dev,
1974 GOTO(out_put, rc = PTR_ERR(dto));
1978 lo->ldo_dir_stripe_loaded = 1;
1979 lo->ldo_dir_striped = 1;
1980 lo->ldo_stripe = stripe;
1981 lo->ldo_dir_stripe_count = i;
1982 lo->ldo_dir_stripes_allocated = stripe_count;
1984 if (lo->ldo_dir_stripe_count == 0)
1985 GOTO(out_put, rc = -ENOSPC);
1987 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
1993 for (i = 0; i < stripe_count; i++)
1994 if (stripe[i] != NULL)
1995 dt_object_put(env, stripe[i]);
1996 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_count);
1997 lo->ldo_dir_stripe_count = 0;
1998 lo->ldo_dir_stripes_allocated = 0;
1999 lo->ldo_stripe = NULL;
2003 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
2009 * Declare create striped md object.
2011 * The function declares intention to create a striped directory. This is a
2012 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2013 * is to verify pattern \a lum_buf is good. Check that function for the details.
2015 * \param[in] env execution environment
2016 * \param[in] dt object
2017 * \param[in] attr attributes to initialize the objects with
2018 * \param[in] lum_buf a pattern specifying the number of stripes and
2020 * \param[in] dof type of objects to be created
2021 * \param[in] th transaction handle
2023 * \retval 0 on success
2024 * \retval negative if failed
2027 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2028 struct dt_object *dt,
2029 struct lu_attr *attr,
2030 const struct lu_buf *lum_buf,
2031 struct dt_object_format *dof,
2034 struct lod_object *lo = lod_dt_obj(dt);
2035 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
2036 struct lmv_user_md_v1 *lum;
2040 lum = lum_buf->lb_buf;
2041 LASSERT(lum != NULL);
2043 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
2044 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
2045 (int)le32_to_cpu(lum->lum_stripe_offset));
2047 if (le32_to_cpu(lum->lum_stripe_count) == 0)
2050 rc = lod_verify_md_striping(lod, lum);
2054 /* prepare dir striped objects */
2055 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2057 /* failed to create striping, let's reset
2058 * config so that others don't get confused */
2059 lod_object_free_striping(env, lo);
2067 * Implementation of dt_object_operations::do_declare_xattr_set.
2069 * Used with regular (non-striped) objects. Basically it
2070 * initializes the striping information and applies the
2071 * change to all the stripes.
2073 * \see dt_object_operations::do_declare_xattr_set() in the API description
2076 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2077 struct dt_object *dt,
2078 const struct lu_buf *buf,
2079 const char *name, int fl,
2082 struct dt_object *next = dt_object_child(dt);
2083 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2084 struct lod_object *lo = lod_dt_obj(dt);
2089 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2090 struct lmv_user_md_v1 *lum;
2092 LASSERT(buf != NULL && buf->lb_buf != NULL);
2094 rc = lod_verify_md_striping(d, lum);
2097 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2098 rc = lod_verify_striping(d, lo, buf, false);
2103 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2107 /* Note: Do not set LinkEA on sub-stripes, otherwise
2108 * it will confuse the fid2path process(see mdt_path_current()).
2109 * The linkEA between master and sub-stripes is set in
2110 * lod_xattr_set_lmv(). */
2111 if (strcmp(name, XATTR_NAME_LINK) == 0)
2114 /* set xattr to each stripes, if needed */
2115 rc = lod_load_striping(env, lo);
2119 if (lo->ldo_dir_stripe_count == 0)
2122 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2123 LASSERT(lo->ldo_stripe[i]);
2125 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2135 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2136 struct lod_object *lo,
2137 struct dt_object *dt, struct thandle *th,
2138 int comp_idx, int stripe_idx,
2139 struct lod_obj_stripe_cb_data *data)
2141 struct lod_thread_info *info = lod_env_info(env);
2142 struct filter_fid *ff = &info->lti_ff;
2143 struct lu_buf *buf = &info->lti_buf;
2147 buf->lb_len = sizeof(*ff);
2148 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2155 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2156 ff->ff_parent.f_ver = stripe_idx;
2157 fid_cpu_to_le(&ff->ff_parent, &ff->ff_parent);
2158 if (data->locd_declare)
2159 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2160 LU_XATTR_REPLACE, th);
2162 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2163 LU_XATTR_REPLACE, th);
2169 * Reset parent FID on OST object
2171 * Replace parent FID with @dt object FID, which is only called during migration
2172 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2173 * the FID is changed.
2175 * \param[in] env execution environment
2176 * \param[in] dt dt_object whose stripes's parent FID will be reset
2177 * \parem[in] th thandle
2178 * \param[in] declare if it is declare
2180 * \retval 0 if reset succeeds
2181 * \retval negative errno if reset fails
2183 static int lod_replace_parent_fid(const struct lu_env *env,
2184 struct dt_object *dt,
2185 struct thandle *th, bool declare)
2187 struct lod_object *lo = lod_dt_obj(dt);
2188 struct lod_thread_info *info = lod_env_info(env);
2189 struct lu_buf *buf = &info->lti_buf;
2190 struct filter_fid *ff;
2191 struct lod_obj_stripe_cb_data data = { { 0 } };
2195 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2197 /* set xattr to each stripes, if needed */
2198 rc = lod_load_striping(env, lo);
2202 if (!lod_obj_is_striped(dt))
2205 if (info->lti_ea_store_size < sizeof(*ff)) {
2206 rc = lod_ea_store_resize(info, sizeof(*ff));
2211 buf->lb_buf = info->lti_ea_store;
2212 buf->lb_len = info->lti_ea_store_size;
2214 data.locd_declare = declare;
2215 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2216 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2221 inline __u16 lod_comp_entry_stripe_count(struct lod_object *lo,
2222 struct lod_layout_component *entry,
2225 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2229 else if (lod_comp_inited(entry))
2230 return entry->llc_stripe_count;
2231 else if ((__u16)-1 == entry->llc_stripe_count)
2232 return lod->lod_desc.ld_tgt_count;
2234 return lod_get_stripe_count(lod, lo, entry->llc_stripe_count);
2237 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2239 int magic, size = 0, i;
2240 struct lod_layout_component *comp_entries;
2245 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2246 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2248 lo->ldo_def_striping->lds_def_striping_is_composite;
2250 comp_cnt = lo->ldo_comp_cnt;
2251 comp_entries = lo->ldo_comp_entries;
2252 is_composite = lo->ldo_is_composite;
2256 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2258 size = sizeof(struct lov_comp_md_v1) +
2259 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2260 LASSERT(size % sizeof(__u64) == 0);
2263 for (i = 0; i < comp_cnt; i++) {
2266 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2267 stripe_count = lod_comp_entry_stripe_count(lo, &comp_entries[i],
2269 if (!is_dir && is_composite)
2270 lod_comp_shrink_stripe_count(&comp_entries[i],
2273 size += lov_user_md_size(stripe_count, magic);
2274 LASSERT(size % sizeof(__u64) == 0);
2280 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2281 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2284 * \param[in] env execution environment
2285 * \param[in] dt dt_object to add components on
2286 * \param[in] buf buffer contains components to be added
2287 * \parem[in] th thandle
2289 * \retval 0 on success
2290 * \retval negative errno on failure
2292 static int lod_declare_layout_add(const struct lu_env *env,
2293 struct dt_object *dt,
2294 const struct lu_buf *buf,
2297 struct lod_thread_info *info = lod_env_info(env);
2298 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2299 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2300 struct dt_object *next = dt_object_child(dt);
2301 struct lov_desc *desc = &d->lod_desc;
2302 struct lod_object *lo = lod_dt_obj(dt);
2303 struct lov_user_md_v3 *v3;
2304 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2306 int i, rc, array_cnt, old_array_cnt;
2309 LASSERT(lo->ldo_is_composite);
2311 if (lo->ldo_flr_state != LCM_FL_NOT_FLR)
2314 rc = lod_verify_striping(d, lo, buf, false);
2318 magic = comp_v1->lcm_magic;
2319 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2320 lustre_swab_lov_comp_md_v1(comp_v1);
2321 magic = comp_v1->lcm_magic;
2324 if (magic != LOV_USER_MAGIC_COMP_V1)
2327 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2328 OBD_ALLOC(comp_array, sizeof(*comp_array) * array_cnt);
2329 if (comp_array == NULL)
2332 memcpy(comp_array, lo->ldo_comp_entries,
2333 sizeof(*comp_array) * lo->ldo_comp_cnt);
2335 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2336 struct lov_user_md_v1 *v1;
2337 struct lu_extent *ext;
2339 v1 = (struct lov_user_md *)((char *)comp_v1 +
2340 comp_v1->lcm_entries[i].lcme_offset);
2341 ext = &comp_v1->lcm_entries[i].lcme_extent;
2343 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2344 lod_comp->llc_extent.e_start = ext->e_start;
2345 lod_comp->llc_extent.e_end = ext->e_end;
2346 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2347 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2349 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2350 if (!lod_comp->llc_stripe_count ||
2351 lod_comp->llc_stripe_count == (__u16)-1)
2352 lod_comp->llc_stripe_count =
2353 desc->ld_default_stripe_count;
2354 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2355 if (!lod_comp->llc_stripe_size)
2356 lod_comp->llc_stripe_size =
2357 desc->ld_default_stripe_size;
2359 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2360 v3 = (struct lov_user_md_v3 *) v1;
2361 if (v3->lmm_pool_name[0] != '\0') {
2362 rc = lod_set_pool(&lod_comp->llc_pool,
2370 old_array = lo->ldo_comp_entries;
2371 old_array_cnt = lo->ldo_comp_cnt;
2373 lo->ldo_comp_entries = comp_array;
2374 lo->ldo_comp_cnt = array_cnt;
2376 /* No need to increase layout generation here, it will be increased
2377 * later when generating component ID for the new components */
2379 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2380 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2381 XATTR_NAME_LOV, 0, th);
2383 lo->ldo_comp_entries = old_array;
2384 lo->ldo_comp_cnt = old_array_cnt;
2388 OBD_FREE(old_array, sizeof(*lod_comp) * old_array_cnt);
2390 LASSERT(lo->ldo_mirror_count == 1);
2391 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2396 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2397 lod_comp = &comp_array[i];
2398 if (lod_comp->llc_pool != NULL) {
2399 OBD_FREE(lod_comp->llc_pool,
2400 strlen(lod_comp->llc_pool) + 1);
2401 lod_comp->llc_pool = NULL;
2404 OBD_FREE(comp_array, sizeof(*comp_array) * array_cnt);
2409 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2410 * the '$field' can only be 'flags' now. The xattr value is binary
2411 * lov_comp_md_v1 which contains the component ID(s) and the value of
2412 * the field to be modified.
2414 * \param[in] env execution environment
2415 * \param[in] dt dt_object to be modified
2416 * \param[in] op operation string, like "set.flags"
2417 * \param[in] buf buffer contains components to be set
2418 * \parem[in] th thandle
2420 * \retval 0 on success
2421 * \retval negative errno on failure
2423 static int lod_declare_layout_set(const struct lu_env *env,
2424 struct dt_object *dt,
2425 char *op, const struct lu_buf *buf,
2428 struct lod_layout_component *lod_comp;
2429 struct lod_thread_info *info = lod_env_info(env);
2430 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2431 struct lod_object *lo = lod_dt_obj(dt);
2432 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2435 bool changed = false;
2438 if (strcmp(op, "set.flags") != 0) {
2439 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2440 lod2obd(d)->obd_name, op);
2444 magic = comp_v1->lcm_magic;
2445 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2446 lustre_swab_lov_comp_md_v1(comp_v1);
2447 magic = comp_v1->lcm_magic;
2450 if (magic != LOV_USER_MAGIC_COMP_V1)
2453 if (comp_v1->lcm_entry_count == 0) {
2454 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2455 lod2obd(d)->obd_name);
2459 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2460 id = comp_v1->lcm_entries[i].lcme_id;
2462 for (j = 0; j < lo->ldo_comp_cnt; j++) {
2463 lod_comp = &lo->ldo_comp_entries[j];
2464 if (id == lod_comp->llc_id || id == LCME_ID_ALL) {
2465 lod_comp->llc_flags =
2466 comp_v1->lcm_entries[i].lcme_flags;
2473 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
2474 lod2obd(d)->obd_name);
2478 lod_obj_inc_layout_gen(lo);
2480 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2481 rc = lod_sub_declare_xattr_set(env, dt, &info->lti_buf,
2482 XATTR_NAME_LOV, 0, th);
2487 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
2488 * and the xattr value is a unique component ID or a special lcme_id.
2490 * \param[in] env execution environment
2491 * \param[in] dt dt_object to be operated on
2492 * \param[in] buf buffer contains component ID or lcme_id
2493 * \parem[in] th thandle
2495 * \retval 0 on success
2496 * \retval negative errno on failure
2498 static int lod_declare_layout_del(const struct lu_env *env,
2499 struct dt_object *dt,
2500 const struct lu_buf *buf,
2503 struct lod_thread_info *info = lod_env_info(env);
2504 struct dt_object *next = dt_object_child(dt);
2505 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2506 struct lod_object *lo = lod_dt_obj(dt);
2507 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2508 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2509 __u32 magic, id, flags, neg_flags = 0;
2513 LASSERT(lo->ldo_is_composite);
2515 if (lo->ldo_flr_state != LCM_FL_NOT_FLR)
2518 magic = comp_v1->lcm_magic;
2519 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2520 lustre_swab_lov_comp_md_v1(comp_v1);
2521 magic = comp_v1->lcm_magic;
2524 if (magic != LOV_USER_MAGIC_COMP_V1)
2527 id = comp_v1->lcm_entries[0].lcme_id;
2528 flags = comp_v1->lcm_entries[0].lcme_flags;
2530 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
2531 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
2532 lod2obd(d)->obd_name, id, flags);
2536 if (id != LCME_ID_INVAL && flags != 0) {
2537 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
2538 lod2obd(d)->obd_name);
2542 if (flags & LCME_FL_NEG) {
2543 neg_flags = flags & ~LCME_FL_NEG;
2547 left = lo->ldo_comp_cnt;
2551 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
2552 struct lod_layout_component *lod_comp;
2554 lod_comp = &lo->ldo_comp_entries[i];
2556 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
2558 else if (flags && !(flags & lod_comp->llc_flags))
2560 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
2563 if (left != (i + 1)) {
2564 CDEBUG(D_LAYOUT, "%s: this deletion will create "
2565 "a hole.\n", lod2obd(d)->obd_name);
2570 /* Mark the component as deleted */
2571 lod_comp->llc_id = LCME_ID_INVAL;
2573 /* Not instantiated component */
2574 if (lod_comp->llc_stripe == NULL)
2577 LASSERT(lod_comp->llc_stripe_count > 0);
2578 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
2579 struct dt_object *obj = lod_comp->llc_stripe[j];
2583 rc = lod_sub_declare_destroy(env, obj, th);
2589 LASSERTF(left >= 0, "left = %d\n", left);
2590 if (left == lo->ldo_comp_cnt) {
2591 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
2592 lod2obd(d)->obd_name, id);
2596 memset(attr, 0, sizeof(*attr));
2597 attr->la_valid = LA_SIZE;
2598 rc = lod_sub_declare_attr_set(env, next, attr, th);
2603 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2604 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2605 XATTR_NAME_LOV, 0, th);
2607 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
2614 * Declare layout add/set/del operations issued by special xattr names:
2616 * XATTR_LUSTRE_LOV.add add component(s) to existing file
2617 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
2618 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
2620 * \param[in] env execution environment
2621 * \param[in] dt object
2622 * \param[in] name name of xattr
2623 * \param[in] buf lu_buf contains xattr value
2624 * \param[in] th transaction handle
2626 * \retval 0 on success
2627 * \retval negative if failed
2629 static int lod_declare_modify_layout(const struct lu_env *env,
2630 struct dt_object *dt,
2632 const struct lu_buf *buf,
2635 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2636 struct lod_object *lo = lod_dt_obj(dt);
2637 struct dt_object *next = dt_object_child(&lo->ldo_obj);
2639 int rc, len = strlen(XATTR_LUSTRE_LOV);
2642 LASSERT(dt_object_exists(dt));
2644 if (strlen(name) <= len || name[len] != '.') {
2645 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
2646 lod2obd(d)->obd_name, name);
2651 dt_write_lock(env, next, 0);
2652 rc = lod_load_striping_locked(env, lo);
2656 /* the layout to be modified must be a composite layout */
2657 if (!lo->ldo_is_composite) {
2658 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
2659 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
2660 GOTO(unlock, rc = -EINVAL);
2663 op = (char *)name + len;
2664 if (strcmp(op, "add") == 0) {
2665 rc = lod_declare_layout_add(env, dt, buf, th);
2666 } else if (strcmp(op, "del") == 0) {
2667 rc = lod_declare_layout_del(env, dt, buf, th);
2668 } else if (strncmp(op, "set", strlen("set")) == 0) {
2669 rc = lod_declare_layout_set(env, dt, op, buf, th);
2671 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
2672 lod2obd(d)->obd_name, name);
2673 GOTO(unlock, rc = -ENOTSUPP);
2677 lod_object_free_striping(env, lo);
2678 dt_write_unlock(env, next);
2684 * Merge layouts to form a mirrored file.
2686 static int lod_declare_layout_merge(const struct lu_env *env,
2687 struct dt_object *dt, const struct lu_buf *mbuf,
2690 struct lod_thread_info *info = lod_env_info(env);
2691 struct lu_buf *buf = &info->lti_buf;
2692 struct lod_object *lo = lod_dt_obj(dt);
2693 struct lov_comp_md_v1 *lcm;
2694 struct lov_comp_md_v1 *cur_lcm;
2695 struct lov_comp_md_v1 *merge_lcm;
2696 struct lov_comp_md_entry_v1 *lcme;
2699 __u16 cur_entry_count;
2700 __u16 merge_entry_count;
2702 __u16 mirror_id = 0;
2707 merge_lcm = mbuf->lb_buf;
2708 if (mbuf->lb_len < sizeof(*merge_lcm))
2711 /* must be an existing layout from disk */
2712 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
2715 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
2717 /* do not allow to merge two mirrored files */
2718 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
2721 /* verify the target buffer */
2722 rc = lod_get_lov_ea(env, lo);
2724 RETURN(rc ? : -ENODATA);
2726 cur_lcm = info->lti_ea_store;
2727 if (le32_to_cpu(cur_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
2730 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
2732 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
2733 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
2734 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
2737 /* size of new layout */
2738 size = le32_to_cpu(cur_lcm->lcm_size) +
2739 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
2741 memset(buf, 0, sizeof(*buf));
2742 lu_buf_alloc(buf, size);
2743 if (buf->lb_buf == NULL)
2747 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
2749 offset = sizeof(*lcm) +
2750 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
2751 for (i = 0; i < cur_entry_count; i++) {
2752 struct lov_comp_md_entry_v1 *cur_lcme;
2754 lcme = &lcm->lcm_entries[i];
2755 cur_lcme = &cur_lcm->lcm_entries[i];
2757 lcme->lcme_offset = cpu_to_le32(offset);
2758 memcpy((char *)lcm + offset,
2759 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
2760 le32_to_cpu(lcme->lcme_size));
2762 offset += le32_to_cpu(lcme->lcme_size);
2764 if (mirror_count == 1) {
2765 /* new mirrored file, create new mirror ID */
2766 id = pflr_id(1, i + 1);
2767 lcme->lcme_id = cpu_to_le32(id);
2770 id = MAX(le32_to_cpu(lcme->lcme_id), id);
2773 mirror_id = mirror_id_of(id) + 1;
2774 for (i = 0; i < merge_entry_count; i++) {
2775 struct lov_comp_md_entry_v1 *merge_lcme;
2777 merge_lcme = &merge_lcm->lcm_entries[i];
2778 lcme = &lcm->lcm_entries[cur_entry_count + i];
2780 *lcme = *merge_lcme;
2781 lcme->lcme_offset = cpu_to_le32(offset);
2783 id = pflr_id(mirror_id, i + 1);
2784 lcme->lcme_id = cpu_to_le32(id);
2786 memcpy((char *)lcm + offset,
2787 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
2788 le32_to_cpu(lcme->lcme_size));
2790 offset += le32_to_cpu(lcme->lcme_size);
2793 /* fixup layout information */
2794 lod_obj_inc_layout_gen(lo);
2795 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
2796 lcm->lcm_size = cpu_to_le32(size);
2797 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
2798 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
2799 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NOT_FLR)
2800 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
2802 LASSERT(dt_write_locked(env, dt_object_child(dt)));
2803 lod_object_free_striping(env, lo);
2804 rc = lod_parse_striping(env, lo, buf);
2808 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
2809 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
2817 * Implementation of dt_object_operations::do_declare_xattr_set.
2819 * \see dt_object_operations::do_declare_xattr_set() in the API description
2822 * the extension to the API:
2823 * - declaring LOVEA requests striping creation
2824 * - LU_XATTR_REPLACE means layout swap
2826 static int lod_declare_xattr_set(const struct lu_env *env,
2827 struct dt_object *dt,
2828 const struct lu_buf *buf,
2829 const char *name, int fl,
2832 struct dt_object *next = dt_object_child(dt);
2833 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2838 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
2839 if ((S_ISREG(mode) || mode == 0) &&
2840 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE)) &&
2841 (strcmp(name, XATTR_NAME_LOV) == 0 ||
2842 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
2844 * this is a request to create object's striping.
2846 * allow to declare predefined striping on a new (!mode) object
2847 * which is supposed to be replay of regular file creation
2848 * (when LOV setting is declared)
2850 * LU_XATTR_REPLACE is set to indicate a layout swap
2852 if (dt_object_exists(dt)) {
2853 rc = dt_attr_get(env, next, attr);
2857 memset(attr, 0, sizeof(*attr));
2858 attr->la_valid = LA_TYPE | LA_MODE;
2859 attr->la_mode = S_IFREG;
2861 rc = lod_declare_striped_create(env, dt, attr, buf, th);
2862 } else if (fl & LU_XATTR_MERGE) {
2863 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
2864 strcmp(name, XATTR_LUSTRE_LOV) == 0);
2865 rc = lod_declare_layout_merge(env, dt, buf, th);
2866 } else if (S_ISREG(mode) &&
2867 strlen(name) > strlen(XATTR_LUSTRE_LOV) + 1 &&
2868 strncmp(name, XATTR_LUSTRE_LOV,
2869 strlen(XATTR_LUSTRE_LOV)) == 0) {
2871 * this is a request to modify object's striping.
2872 * add/set/del component(s).
2874 if (!dt_object_exists(dt))
2877 rc = lod_declare_modify_layout(env, dt, name, buf, th);
2878 } else if (S_ISDIR(mode)) {
2879 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
2880 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
2881 rc = lod_replace_parent_fid(env, dt, th, true);
2883 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2890 * Apply xattr changes to the object.
2892 * Applies xattr changes to the object and the stripes if the latter exist.
2894 * \param[in] env execution environment
2895 * \param[in] dt object
2896 * \param[in] buf buffer pointing to the new value of xattr
2897 * \param[in] name name of xattr
2898 * \param[in] fl flags
2899 * \param[in] th transaction handle
2901 * \retval 0 on success
2902 * \retval negative if failed
2904 static int lod_xattr_set_internal(const struct lu_env *env,
2905 struct dt_object *dt,
2906 const struct lu_buf *buf,
2907 const char *name, int fl,
2910 struct dt_object *next = dt_object_child(dt);
2911 struct lod_object *lo = lod_dt_obj(dt);
2916 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
2917 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2920 /* Note: Do not set LinkEA on sub-stripes, otherwise
2921 * it will confuse the fid2path process(see mdt_path_current()).
2922 * The linkEA between master and sub-stripes is set in
2923 * lod_xattr_set_lmv(). */
2924 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
2927 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2928 LASSERT(lo->ldo_stripe[i]);
2930 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
2940 * Delete an extended attribute.
2942 * Deletes specified xattr from the object and the stripes if the latter exist.
2944 * \param[in] env execution environment
2945 * \param[in] dt object
2946 * \param[in] name name of xattr
2947 * \param[in] th transaction handle
2949 * \retval 0 on success
2950 * \retval negative if failed
2952 static int lod_xattr_del_internal(const struct lu_env *env,
2953 struct dt_object *dt,
2954 const char *name, struct thandle *th)
2956 struct dt_object *next = dt_object_child(dt);
2957 struct lod_object *lo = lod_dt_obj(dt);
2962 rc = lod_sub_xattr_del(env, next, name, th);
2963 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2966 if (lo->ldo_dir_stripe_count == 0)
2969 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2970 LASSERT(lo->ldo_stripe[i]);
2972 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
2981 * Set default striping on a directory.
2983 * Sets specified striping on a directory object unless it matches the default
2984 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
2985 * EA. This striping will be used when regular file is being created in this
2988 * \param[in] env execution environment
2989 * \param[in] dt the striped object
2990 * \param[in] buf buffer with the striping
2991 * \param[in] name name of EA
2992 * \param[in] fl xattr flag (see OSD API description)
2993 * \param[in] th transaction handle
2995 * \retval 0 on success
2996 * \retval negative if failed
2998 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
2999 struct dt_object *dt,
3000 const struct lu_buf *buf,
3001 const char *name, int fl,
3004 struct lov_user_md_v1 *lum;
3005 struct lov_user_md_v3 *v3 = NULL;
3006 const char *pool_name = NULL;
3011 LASSERT(buf != NULL && buf->lb_buf != NULL);
3014 switch (lum->lmm_magic) {
3015 case LOV_USER_MAGIC_V3:
3017 if (v3->lmm_pool_name[0] != '\0')
3018 pool_name = v3->lmm_pool_name;
3020 case LOV_USER_MAGIC_V1:
3021 /* if { size, offset, count } = { 0, -1, 0 } and no pool
3022 * (i.e. all default values specified) then delete default
3023 * striping from dir. */
3025 "set default striping: sz %u # %u offset %d %s %s\n",
3026 (unsigned)lum->lmm_stripe_size,
3027 (unsigned)lum->lmm_stripe_count,
3028 (int)lum->lmm_stripe_offset,
3029 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
3031 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
3032 lum->lmm_stripe_count,
3033 lum->lmm_stripe_offset,
3036 case LOV_USER_MAGIC_COMP_V1:
3040 CERROR("Invalid magic %x\n", lum->lmm_magic);
3045 rc = lod_xattr_del_internal(env, dt, name, th);
3049 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3056 * Set default striping on a directory object.
3058 * Sets specified striping on a directory object unless it matches the default
3059 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3060 * EA. This striping will be used when a new directory is being created in the
3063 * \param[in] env execution environment
3064 * \param[in] dt the striped object
3065 * \param[in] buf buffer with the striping
3066 * \param[in] name name of EA
3067 * \param[in] fl xattr flag (see OSD API description)
3068 * \param[in] th transaction handle
3070 * \retval 0 on success
3071 * \retval negative if failed
3073 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
3074 struct dt_object *dt,
3075 const struct lu_buf *buf,
3076 const char *name, int fl,
3079 struct lmv_user_md_v1 *lum;
3083 LASSERT(buf != NULL && buf->lb_buf != NULL);
3086 CDEBUG(D_OTHER, "set default stripe_count # %u stripe_offset %d\n",
3087 le32_to_cpu(lum->lum_stripe_count),
3088 (int)le32_to_cpu(lum->lum_stripe_offset));
3090 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
3091 le32_to_cpu(lum->lum_stripe_offset)) &&
3092 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
3093 rc = lod_xattr_del_internal(env, dt, name, th);
3097 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3106 * Turn directory into a striped directory.
3108 * During replay the client sends the striping created before MDT
3109 * failure, then the layer above LOD sends this defined striping
3110 * using ->do_xattr_set(), so LOD uses this method to replay creation
3111 * of the stripes. Notice the original information for the striping
3112 * (#stripes, FIDs, etc) was transferred in declare path.
3114 * \param[in] env execution environment
3115 * \param[in] dt the striped object
3116 * \param[in] buf not used currently
3117 * \param[in] name not used currently
3118 * \param[in] fl xattr flag (see OSD API description)
3119 * \param[in] th transaction handle
3121 * \retval 0 on success
3122 * \retval negative if failed
3124 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
3125 const struct lu_buf *buf, const char *name,
3126 int fl, struct thandle *th)
3128 struct lod_object *lo = lod_dt_obj(dt);
3129 struct lod_thread_info *info = lod_env_info(env);
3130 struct lu_attr *attr = &info->lti_attr;
3131 struct dt_object_format *dof = &info->lti_format;
3132 struct lu_buf lmv_buf;
3133 struct lu_buf slave_lmv_buf;
3134 struct lmv_mds_md_v1 *lmm;
3135 struct lmv_mds_md_v1 *slave_lmm = NULL;
3136 struct dt_insert_rec *rec = &info->lti_dt_rec;
3141 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3144 /* The stripes are supposed to be allocated in declare phase,
3145 * if there are no stripes being allocated, it will skip */
3146 if (lo->ldo_dir_stripe_count == 0)
3149 rc = dt_attr_get(env, dt_object_child(dt), attr);
3153 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME |
3154 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
3155 dof->dof_type = DFT_DIR;
3157 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
3160 lmm = lmv_buf.lb_buf;
3162 OBD_ALLOC_PTR(slave_lmm);
3163 if (slave_lmm == NULL)
3166 lod_prep_slave_lmv_md(slave_lmm, lmm);
3167 slave_lmv_buf.lb_buf = slave_lmm;
3168 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
3170 rec->rec_type = S_IFDIR;
3171 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3172 struct dt_object *dto;
3173 char *stripe_name = info->lti_key;
3174 struct lu_name *sname;
3175 struct linkea_data ldata = { NULL };
3176 struct lu_buf linkea_buf;
3178 dto = lo->ldo_stripe[i];
3180 dt_write_lock(env, dto, MOR_TGT_CHILD);
3181 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
3183 dt_write_unlock(env, dto);
3187 rc = lod_sub_ref_add(env, dto, th);
3188 dt_write_unlock(env, dto);
3192 rec->rec_fid = lu_object_fid(&dto->do_lu);
3193 rc = lod_sub_insert(env, dto, (const struct dt_rec *)rec,
3194 (const struct dt_key *)dot, th, 0);
3198 rec->rec_fid = lu_object_fid(&dt->do_lu);
3199 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
3200 (const struct dt_key *)dotdot, th, 0);
3204 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
3205 cfs_fail_val != i) {
3206 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
3208 slave_lmm->lmv_master_mdt_index =
3211 slave_lmm->lmv_master_mdt_index =
3214 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
3215 XATTR_NAME_LMV, fl, th);
3220 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
3222 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3223 PFID(lu_object_fid(&dto->do_lu)), i + 1);
3225 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3226 PFID(lu_object_fid(&dto->do_lu)), i);
3228 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
3229 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
3230 sname, lu_object_fid(&dt->do_lu));
3234 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
3235 linkea_buf.lb_len = ldata.ld_leh->leh_len;
3236 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
3237 XATTR_NAME_LINK, 0, th);
3241 rec->rec_fid = lu_object_fid(&dto->do_lu);
3242 rc = lod_sub_insert(env, dt_object_child(dt),
3243 (const struct dt_rec *)rec,
3244 (const struct dt_key *)stripe_name, th, 0);
3248 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
3253 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
3254 rc = lod_sub_xattr_set(env, dt_object_child(dt),
3255 &lmv_buf, XATTR_NAME_LMV, fl, th);
3257 if (slave_lmm != NULL)
3258 OBD_FREE_PTR(slave_lmm);
3264 * Helper function to declare/execute creation of a striped directory
3266 * Called in declare/create object path, prepare striping for a directory
3267 * and prepare defaults data striping for the objects to be created in
3268 * that directory. Notice the function calls "declaration" or "execution"
3269 * methods depending on \a declare param. This is a consequence of the
3270 * current approach while we don't have natural distributed transactions:
3271 * we basically execute non-local updates in the declare phase. So, the
3272 * arguments for the both phases are the same and this is the reason for
3273 * this function to exist.
3275 * \param[in] env execution environment
3276 * \param[in] dt object
3277 * \param[in] attr attributes the stripes will be created with
3278 * \param[in] dof format of stripes (see OSD API description)
3279 * \param[in] th transaction handle
3280 * \param[in] declare where to call "declare" or "execute" methods
3282 * \retval 0 on success
3283 * \retval negative if failed
3285 static int lod_dir_striping_create_internal(const struct lu_env *env,
3286 struct dt_object *dt,
3287 struct lu_attr *attr,
3288 struct dt_object_format *dof,
3292 struct lod_thread_info *info = lod_env_info(env);
3293 struct lod_object *lo = lod_dt_obj(dt);
3294 const struct lod_default_striping *lds = lo->ldo_def_striping;
3298 LASSERT(ergo(lds != NULL,
3299 lds->lds_def_striping_set ||
3300 lds->lds_dir_def_striping_set));
3302 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
3303 lo->ldo_dir_stripe_offset)) {
3304 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3305 int stripe_count = lo->ldo_dir_stripe_count;
3307 if (info->lti_ea_store_size < sizeof(*v1)) {
3308 rc = lod_ea_store_resize(info, sizeof(*v1));
3311 v1 = info->lti_ea_store;
3314 memset(v1, 0, sizeof(*v1));
3315 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3316 v1->lum_stripe_count = cpu_to_le32(stripe_count);
3317 v1->lum_stripe_offset =
3318 cpu_to_le32(lo->ldo_dir_stripe_offset);
3320 info->lti_buf.lb_buf = v1;
3321 info->lti_buf.lb_len = sizeof(*v1);
3324 rc = lod_declare_xattr_set_lmv(env, dt, attr,
3325 &info->lti_buf, dof, th);
3327 rc = lod_xattr_set_lmv(env, dt, &info->lti_buf,
3328 XATTR_NAME_LMV, 0, th);
3333 /* Transfer default LMV striping from the parent */
3334 if (lds != NULL && lds->lds_dir_def_striping_set &&
3335 !LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
3336 lds->lds_dir_def_stripe_offset)) {
3337 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3339 if (info->lti_ea_store_size < sizeof(*v1)) {
3340 rc = lod_ea_store_resize(info, sizeof(*v1));
3343 v1 = info->lti_ea_store;
3346 memset(v1, 0, sizeof(*v1));
3347 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3348 v1->lum_stripe_count =
3349 cpu_to_le32(lds->lds_dir_def_stripe_count);
3350 v1->lum_stripe_offset =
3351 cpu_to_le32(lds->lds_dir_def_stripe_offset);
3353 cpu_to_le32(lds->lds_dir_def_hash_type);
3355 info->lti_buf.lb_buf = v1;
3356 info->lti_buf.lb_len = sizeof(*v1);
3358 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3359 XATTR_NAME_DEFAULT_LMV,
3362 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
3364 XATTR_NAME_DEFAULT_LMV, 0,
3370 /* Transfer default LOV striping from the parent */
3371 if (lds != NULL && lds->lds_def_striping_set &&
3372 lds->lds_def_comp_cnt != 0) {
3373 struct lov_mds_md *lmm;
3374 int lmm_size = lod_comp_md_size(lo, true);
3376 if (info->lti_ea_store_size < lmm_size) {
3377 rc = lod_ea_store_resize(info, lmm_size);
3381 lmm = info->lti_ea_store;
3383 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
3387 info->lti_buf.lb_buf = lmm;
3388 info->lti_buf.lb_len = lmm_size;
3391 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3392 XATTR_NAME_LOV, 0, th);
3394 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
3395 XATTR_NAME_LOV, 0, th);
3403 static int lod_declare_dir_striping_create(const struct lu_env *env,
3404 struct dt_object *dt,
3405 struct lu_attr *attr,
3406 struct dt_object_format *dof,
3409 return lod_dir_striping_create_internal(env, dt, attr, dof, th, true);
3412 static int lod_dir_striping_create(const struct lu_env *env,
3413 struct dt_object *dt,
3414 struct lu_attr *attr,
3415 struct dt_object_format *dof,
3418 return lod_dir_striping_create_internal(env, dt, attr, dof, th, false);
3422 * Make LOV EA for striped object.
3424 * Generate striping information and store it in the LOV EA of the given
3425 * object. The caller must ensure nobody else is calling the function
3426 * against the object concurrently. The transaction must be started.
3427 * FLDB service must be running as well; it's used to map FID to the target,
3428 * which is stored in LOV EA.
3430 * \param[in] env execution environment for this thread
3431 * \param[in] lo LOD object
3432 * \param[in] th transaction handle
3434 * \retval 0 if LOV EA is stored successfully
3435 * \retval negative error number on failure
3437 static int lod_generate_and_set_lovea(const struct lu_env *env,
3438 struct lod_object *lo,
3441 struct lod_thread_info *info = lod_env_info(env);
3442 struct dt_object *next = dt_object_child(&lo->ldo_obj);
3443 struct lov_mds_md_v1 *lmm;
3449 if (lo->ldo_comp_cnt == 0) {
3450 lod_object_free_striping(env, lo);
3451 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
3455 lmm_size = lod_comp_md_size(lo, false);
3456 if (info->lti_ea_store_size < lmm_size) {
3457 rc = lod_ea_store_resize(info, lmm_size);
3461 lmm = info->lti_ea_store;
3463 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
3467 info->lti_buf.lb_buf = lmm;
3468 info->lti_buf.lb_len = lmm_size;
3469 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
3470 XATTR_NAME_LOV, 0, th);
3475 * Delete layout component(s)
3477 * \param[in] env execution environment for this thread
3478 * \param[in] dt object
3479 * \param[in] th transaction handle
3481 * \retval 0 on success
3482 * \retval negative error number on failure
3484 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
3487 struct lod_layout_component *lod_comp;
3488 struct lod_object *lo = lod_dt_obj(dt);
3489 struct dt_object *next = dt_object_child(dt);
3490 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3493 LASSERT(lo->ldo_is_composite);
3494 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
3496 left = lo->ldo_comp_cnt;
3497 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3498 lod_comp = &lo->ldo_comp_entries[i];
3500 if (lod_comp->llc_id != LCME_ID_INVAL)
3504 /* Not instantiated component */
3505 if (lod_comp->llc_stripe == NULL)
3508 LASSERT(lod_comp->llc_stripe_count > 0);
3509 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
3510 struct dt_object *obj = lod_comp->llc_stripe[j];
3514 rc = lod_sub_destroy(env, obj, th);
3518 lu_object_put(env, &obj->do_lu);
3519 lod_comp->llc_stripe[j] = NULL;
3521 OBD_FREE(lod_comp->llc_stripe, sizeof(struct dt_object *) *
3522 lod_comp->llc_stripes_allocated);
3523 lod_comp->llc_stripe = NULL;
3524 lod_comp->llc_stripes_allocated = 0;
3525 lod_obj_set_pool(lo, i, NULL);
3526 if (lod_comp->llc_ostlist.op_array) {
3527 OBD_FREE(lod_comp->llc_ostlist.op_array,
3528 lod_comp->llc_ostlist.op_size);
3529 lod_comp->llc_ostlist.op_array = NULL;
3530 lod_comp->llc_ostlist.op_size = 0;
3534 LASSERTF(left >= 0 && left < lo->ldo_comp_cnt, "left = %d\n", left);
3536 struct lod_layout_component *comp_array;
3538 OBD_ALLOC(comp_array, sizeof(*comp_array) * left);
3539 if (comp_array == NULL)
3540 GOTO(out, rc = -ENOMEM);
3542 memcpy(&comp_array[0], &lo->ldo_comp_entries[0],
3543 sizeof(*comp_array) * left);
3545 OBD_FREE(lo->ldo_comp_entries,
3546 sizeof(*comp_array) * lo->ldo_comp_cnt);
3547 lo->ldo_comp_entries = comp_array;
3548 lo->ldo_comp_cnt = left;
3550 LASSERT(lo->ldo_mirror_count == 1);
3551 lo->ldo_mirrors[0].lme_end = left - 1;
3552 lod_obj_inc_layout_gen(lo);
3554 lod_free_comp_entries(lo);
3557 LASSERT(dt_object_exists(dt));
3558 rc = dt_attr_get(env, next, attr);
3562 if (attr->la_size > 0) {
3564 attr->la_valid = LA_SIZE;
3565 rc = lod_sub_attr_set(env, next, attr, th);
3570 rc = lod_generate_and_set_lovea(env, lo, th);
3574 lod_object_free_striping(env, lo);
3579 * Implementation of dt_object_operations::do_xattr_set.
3581 * Sets specified extended attribute on the object. Three types of EAs are
3583 * LOV EA - stores striping for a regular file or default striping (when set
3585 * LMV EA - stores a marker for the striped directories
3586 * DMV EA - stores default directory striping
3588 * When striping is applied to a non-striped existing object (this is called
3589 * late striping), then LOD notices the caller wants to turn the object into a
3590 * striped one. The stripe objects are created and appropriate EA is set:
3591 * LOV EA storing all the stripes directly or LMV EA storing just a small header
3592 * with striping configuration.
3594 * \see dt_object_operations::do_xattr_set() in the API description for details.
3596 static int lod_xattr_set(const struct lu_env *env,
3597 struct dt_object *dt, const struct lu_buf *buf,
3598 const char *name, int fl, struct thandle *th)
3600 struct dt_object *next = dt_object_child(dt);
3604 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
3605 strcmp(name, XATTR_NAME_LMV) == 0) {
3606 struct lmv_mds_md_v1 *lmm = buf->lb_buf;
3608 if (lmm != NULL && le32_to_cpu(lmm->lmv_hash_type) &
3609 LMV_HASH_FLAG_MIGRATION)
3610 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3612 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
3617 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
3618 strcmp(name, XATTR_NAME_LOV) == 0) {
3620 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl, th);
3622 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
3623 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
3625 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
3628 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
3629 (!strcmp(name, XATTR_NAME_LOV) ||
3630 !strncmp(name, XATTR_LUSTRE_LOV,
3631 strlen(XATTR_LUSTRE_LOV)))) {
3632 /* in case of lov EA swap, just set it
3633 * if not, it is a replay so check striping match what we
3634 * already have during req replay, declare_xattr_set()
3635 * defines striping, then create() does the work */
3636 if (fl & LU_XATTR_REPLACE) {
3637 /* free stripes, then update disk */
3638 lod_object_free_striping(env, lod_dt_obj(dt));
3640 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3641 } else if (dt_object_remote(dt)) {
3642 /* This only happens during migration, see
3643 * mdd_migrate_create(), in which Master MDT will
3644 * create a remote target object, and only set
3645 * (migrating) stripe EA on the remote object,
3646 * and does not need creating each stripes. */
3647 rc = lod_sub_xattr_set(env, next, buf, name,
3649 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
3650 /* delete component(s) */
3651 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
3652 rc = lod_layout_del(env, dt, th);
3655 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
3656 * it's going to create create file with specified
3657 * component(s), the striping must have not being
3658 * cached in this case;
3660 * Otherwise, it's going to add/change component(s) to
3661 * an existing file, the striping must have been cached
3664 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
3665 !strcmp(name, XATTR_NAME_LOV),
3666 !lod_dt_obj(dt)->ldo_comp_cached));
3668 rc = lod_striped_create(env, dt, NULL, NULL, th);
3671 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3672 rc = lod_replace_parent_fid(env, dt, th, false);
3677 /* then all other xattr */
3678 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3684 * Implementation of dt_object_operations::do_declare_xattr_del.
3686 * \see dt_object_operations::do_declare_xattr_del() in the API description
3689 static int lod_declare_xattr_del(const struct lu_env *env,
3690 struct dt_object *dt, const char *name,
3693 struct lod_object *lo = lod_dt_obj(dt);
3698 rc = lod_sub_declare_xattr_del(env, dt_object_child(dt), name, th);
3702 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3705 /* set xattr to each stripes, if needed */
3706 rc = lod_load_striping(env, lo);
3710 if (lo->ldo_dir_stripe_count == 0)
3713 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3714 LASSERT(lo->ldo_stripe[i]);
3715 rc = lod_sub_declare_xattr_del(env, lo->ldo_stripe[i],
3725 * Implementation of dt_object_operations::do_xattr_del.
3727 * If EA storing a regular striping is being deleted, then release
3728 * all the references to the stripe objects in core.
3730 * \see dt_object_operations::do_xattr_del() in the API description for details.
3732 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
3733 const char *name, struct thandle *th)
3735 struct dt_object *next = dt_object_child(dt);
3736 struct lod_object *lo = lod_dt_obj(dt);
3741 if (!strcmp(name, XATTR_NAME_LOV))
3742 lod_object_free_striping(env, lod_dt_obj(dt));
3744 rc = lod_sub_xattr_del(env, next, name, th);
3745 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3748 if (lo->ldo_dir_stripe_count == 0)
3751 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3752 LASSERT(lo->ldo_stripe[i]);
3754 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3763 * Implementation of dt_object_operations::do_xattr_list.
3765 * \see dt_object_operations::do_xattr_list() in the API description
3768 static int lod_xattr_list(const struct lu_env *env,
3769 struct dt_object *dt, const struct lu_buf *buf)
3771 return dt_xattr_list(env, dt_object_child(dt), buf);
3774 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
3776 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
3781 * Get default striping.
3783 * \param[in] env execution environment
3784 * \param[in] lo object
3785 * \param[out] lds default striping
3787 * \retval 0 on success
3788 * \retval negative if failed
3790 static int lod_get_default_lov_striping(const struct lu_env *env,
3791 struct lod_object *lo,
3792 struct lod_default_striping *lds)
3794 struct lod_thread_info *info = lod_env_info(env);
3795 struct lov_user_md_v1 *v1 = NULL;
3796 struct lov_user_md_v3 *v3 = NULL;
3797 struct lov_comp_md_v1 *comp_v1 = NULL;
3804 lds->lds_def_striping_set = 0;
3806 rc = lod_get_lov_ea(env, lo);
3810 if (rc < (typeof(rc))sizeof(struct lov_user_md))
3813 v1 = info->lti_ea_store;
3814 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
3815 lustre_swab_lov_user_md_v1(v1);
3816 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
3817 v3 = (struct lov_user_md_v3 *)v1;
3818 lustre_swab_lov_user_md_v3(v3);
3819 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
3820 comp_v1 = (struct lov_comp_md_v1 *)v1;
3821 lustre_swab_lov_comp_md_v1(comp_v1);
3824 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1 &&
3825 v1->lmm_magic != LOV_MAGIC_COMP_V1)
3828 if (v1->lmm_magic == LOV_MAGIC_COMP_V1) {
3829 comp_v1 = (struct lov_comp_md_v1 *)v1;
3830 comp_cnt = comp_v1->lcm_entry_count;
3833 mirror_cnt = comp_v1->lcm_mirror_count + 1;
3841 /* realloc default comp entries if necessary */
3842 rc = lod_def_striping_comp_resize(lds, comp_cnt);
3846 lds->lds_def_comp_cnt = comp_cnt;
3847 lds->lds_def_striping_is_composite = composite;
3848 lds->lds_def_mirror_cnt = mirror_cnt;
3850 for (i = 0; i < comp_cnt; i++) {
3851 struct lod_layout_component *lod_comp;
3852 struct lu_extent *ext;
3855 lod_comp = &lds->lds_def_comp_entries[i];
3857 * reset lod_comp values, llc_stripes is always NULL in
3858 * the default striping template, llc_pool will be reset
3861 memset(lod_comp, 0, offsetof(typeof(*lod_comp), llc_pool));
3864 v1 = (struct lov_user_md *)((char *)comp_v1 +
3865 comp_v1->lcm_entries[i].lcme_offset);
3866 ext = &comp_v1->lcm_entries[i].lcme_extent;
3867 lod_comp->llc_extent = *ext;
3870 if (v1->lmm_pattern != LOV_PATTERN_RAID0 &&
3871 v1->lmm_pattern != LOV_PATTERN_MDT &&
3872 v1->lmm_pattern != 0) {
3873 lod_free_def_comp_entries(lds);
3877 CDEBUG(D_LAYOUT, DFID" stripe_count=%d stripe_size=%d "
3878 "stripe_offset=%d\n",
3879 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
3880 (int)v1->lmm_stripe_count, (int)v1->lmm_stripe_size,
3881 (int)v1->lmm_stripe_offset);
3883 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
3884 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
3885 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
3886 lod_comp->llc_pattern = v1->lmm_pattern;
3889 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
3890 /* XXX: sanity check here */
3891 v3 = (struct lov_user_md_v3 *) v1;
3892 if (v3->lmm_pool_name[0] != '\0')
3893 pool = v3->lmm_pool_name;
3895 lod_set_def_pool(lds, i, pool);
3898 lds->lds_def_striping_set = 1;
3903 * Get default directory striping.
3905 * \param[in] env execution environment
3906 * \param[in] lo object
3907 * \param[out] lds default striping
3909 * \retval 0 on success
3910 * \retval negative if failed
3912 static int lod_get_default_lmv_striping(const struct lu_env *env,
3913 struct lod_object *lo,
3914 struct lod_default_striping *lds)
3916 struct lod_thread_info *info = lod_env_info(env);
3917 struct lmv_user_md_v1 *v1 = NULL;
3921 lds->lds_dir_def_striping_set = 0;
3922 rc = lod_get_default_lmv_ea(env, lo);
3926 if (rc < (typeof(rc))sizeof(struct lmv_user_md))
3929 v1 = info->lti_ea_store;
3931 lds->lds_dir_def_stripe_count = le32_to_cpu(v1->lum_stripe_count);
3932 lds->lds_dir_def_stripe_offset = le32_to_cpu(v1->lum_stripe_offset);
3933 lds->lds_dir_def_hash_type = le32_to_cpu(v1->lum_hash_type);
3934 lds->lds_dir_def_striping_set = 1;
3940 * Get default striping in the object.
3942 * Get object default striping and default directory striping.
3944 * \param[in] env execution environment
3945 * \param[in] lo object
3946 * \param[out] lds default striping
3948 * \retval 0 on success
3949 * \retval negative if failed
3951 static int lod_get_default_striping(const struct lu_env *env,
3952 struct lod_object *lo,
3953 struct lod_default_striping *lds)
3957 rc = lod_get_default_lov_striping(env, lo, lds);
3958 rc1 = lod_get_default_lmv_striping(env, lo, lds);
3959 if (rc == 0 && rc1 < 0)
3966 * Apply default striping on object.
3968 * If object striping pattern is not set, set to the one in default striping.
3969 * The default striping is from parent or fs.
3971 * \param[in] lo new object
3972 * \param[in] lds default striping
3973 * \param[in] mode new object's mode
3975 static void lod_striping_from_default(struct lod_object *lo,
3976 const struct lod_default_striping *lds,
3979 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
3980 struct lov_desc *desc = &d->lod_desc;
3983 if (lds->lds_def_striping_set && S_ISREG(mode)) {
3984 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
3985 lds->lds_def_comp_cnt);
3989 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
3991 for (i = 0; i < lo->ldo_comp_cnt; i++) {
3992 struct lod_layout_component *obj_comp =
3993 &lo->ldo_comp_entries[i];
3994 struct lod_layout_component *def_comp =
3995 &lds->lds_def_comp_entries[i];
3997 CDEBUG(D_LAYOUT, "Inherite from default: size:%hu "
3998 "nr:%u offset:%u pattern %#x %s\n",
3999 def_comp->llc_stripe_size,
4000 def_comp->llc_stripe_count,
4001 def_comp->llc_stripe_offset,
4002 def_comp->llc_pattern,
4003 def_comp->llc_pool ?: "");
4005 *obj_comp = *def_comp;
4006 if (def_comp->llc_pool != NULL) {
4007 /* pointer was copied from def_comp */
4008 obj_comp->llc_pool = NULL;
4009 lod_obj_set_pool(lo, i, def_comp->llc_pool);
4013 * Don't initialize these fields for plain layout
4014 * (v1/v3) here, they are inherited in the order of
4015 * 'parent' -> 'fs default (root)' -> 'global default
4016 * values for stripe_count & stripe_size'.
4018 * see lod_ah_init().
4020 if (!lo->ldo_is_composite)
4023 if (obj_comp->llc_stripe_count <= 0 &&
4024 obj_comp->llc_pattern != LOV_PATTERN_MDT)
4025 obj_comp->llc_stripe_count =
4026 desc->ld_default_stripe_count;
4027 if (obj_comp->llc_stripe_size <= 0)
4028 obj_comp->llc_stripe_size =
4029 desc->ld_default_stripe_size;
4031 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
4032 if (lo->ldo_dir_stripe_count == 0)
4033 lo->ldo_dir_stripe_count =
4034 lds->lds_dir_def_stripe_count;
4035 if (lo->ldo_dir_stripe_offset == -1)
4036 lo->ldo_dir_stripe_offset =
4037 lds->lds_dir_def_stripe_offset;
4038 if (lo->ldo_dir_hash_type == 0)
4039 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
4041 CDEBUG(D_LAYOUT, "striping from default dir: count:%hu, "
4042 "offset:%u, hash_type:%u\n",
4043 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
4044 lo->ldo_dir_hash_type);
4048 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root)
4050 struct lod_layout_component *lod_comp;
4052 if (lo->ldo_comp_cnt == 0)
4055 if (lo->ldo_is_composite)
4058 lod_comp = &lo->ldo_comp_entries[0];
4060 if (lod_comp->llc_stripe_count <= 0 ||
4061 lod_comp->llc_stripe_size <= 0)
4064 if (from_root && (lod_comp->llc_pool == NULL ||
4065 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
4072 * Implementation of dt_object_operations::do_ah_init.
4074 * This method is used to make a decision on the striping configuration for the
4075 * object being created. It can be taken from the \a parent object if it exists,
4076 * or filesystem's default. The resulting configuration (number of stripes,
4077 * stripe size/offset, pool name, etc) is stored in the object itself and will
4078 * be used by the methods like ->doo_declare_create().
4080 * \see dt_object_operations::do_ah_init() in the API description for details.
4082 static void lod_ah_init(const struct lu_env *env,
4083 struct dt_allocation_hint *ah,
4084 struct dt_object *parent,
4085 struct dt_object *child,
4088 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
4089 struct lod_thread_info *info = lod_env_info(env);
4090 struct lod_default_striping *lds = &info->lti_def_striping;
4091 struct dt_object *nextp = NULL;
4092 struct dt_object *nextc;
4093 struct lod_object *lp = NULL;
4094 struct lod_object *lc;
4095 struct lov_desc *desc;
4096 struct lod_layout_component *lod_comp;
4102 if (likely(parent)) {
4103 nextp = dt_object_child(parent);
4104 lp = lod_dt_obj(parent);
4107 nextc = dt_object_child(child);
4108 lc = lod_dt_obj(child);
4110 LASSERT(!lod_obj_is_striped(child));
4111 /* default layout template may have been set on the regular file
4112 * when this is called from mdd_create_data() */
4113 if (S_ISREG(child_mode))
4114 lod_free_comp_entries(lc);
4116 if (!dt_object_exists(nextc))
4117 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
4119 if (S_ISDIR(child_mode)) {
4120 /* other default values are 0 */
4121 lc->ldo_dir_stripe_offset = -1;
4123 /* get default striping from parent object */
4124 if (likely(lp != NULL))
4125 lod_get_default_striping(env, lp, lds);
4127 /* set child default striping info, default value is NULL */
4128 if (lds->lds_def_striping_set || lds->lds_dir_def_striping_set)
4129 lc->ldo_def_striping = lds;
4131 /* It should always honour the specified stripes */
4132 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
4133 lod_verify_md_striping(d, ah->dah_eadata) == 0) {
4134 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
4136 lc->ldo_dir_stripe_count =
4137 le32_to_cpu(lum1->lum_stripe_count);
4138 lc->ldo_dir_stripe_offset =
4139 le32_to_cpu(lum1->lum_stripe_offset);
4140 lc->ldo_dir_hash_type =
4141 le32_to_cpu(lum1->lum_hash_type);
4143 "set dirstripe: count %hu, offset %d, hash %u\n",
4144 lc->ldo_dir_stripe_count,
4145 (int)lc->ldo_dir_stripe_offset,
4146 lc->ldo_dir_hash_type);
4148 /* transfer defaults LMV to new directory */
4149 lod_striping_from_default(lc, lds, child_mode);
4152 /* shrink the stripe_count to the avaible MDT count */
4153 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
4154 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))
4155 lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
4157 /* Directory will be striped only if stripe_count > 1, if
4158 * stripe_count == 1, let's reset stripe_count = 0 to avoid
4159 * create single master stripe and also help to unify the
4160 * stripe handling of directories and files */
4161 if (lc->ldo_dir_stripe_count == 1)
4162 lc->ldo_dir_stripe_count = 0;
4164 CDEBUG(D_INFO, "final dir stripe [%hu %d %u]\n",
4165 lc->ldo_dir_stripe_count,
4166 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
4171 /* child object regular file*/
4173 if (!lod_object_will_be_striped(S_ISREG(child_mode),
4174 lu_object_fid(&child->do_lu)))
4177 /* If object is going to be striped over OSTs, transfer default
4178 * striping information to the child, so that we can use it
4179 * during declaration and creation.
4181 * Try from the parent first.
4183 if (likely(lp != NULL)) {
4184 rc = lod_get_default_lov_striping(env, lp, lds);
4186 lod_striping_from_default(lc, lds, child_mode);
4189 /* Initialize lod_device::lod_md_root object reference */
4190 if (d->lod_md_root == NULL) {
4191 struct dt_object *root;
4192 struct lod_object *lroot;
4194 lu_root_fid(&info->lti_fid);
4195 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
4196 if (!IS_ERR(root)) {
4197 lroot = lod_dt_obj(root);
4199 spin_lock(&d->lod_lock);
4200 if (d->lod_md_root != NULL)
4201 dt_object_put(env, &d->lod_md_root->ldo_obj);
4202 d->lod_md_root = lroot;
4203 spin_unlock(&d->lod_lock);
4207 /* try inherit layout from the root object (fs default) when:
4208 * - parent does not have default layout; or
4209 * - parent has plain(v1/v3) default layout, and some attributes
4210 * are not specified in the default layout;
4212 if (d->lod_md_root != NULL && lod_need_inherit_more(lc, true)) {
4213 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds);
4216 if (lc->ldo_comp_cnt == 0) {
4217 lod_striping_from_default(lc, lds, child_mode);
4218 } else if (!lds->lds_def_striping_is_composite) {
4219 struct lod_layout_component *def_comp;
4221 LASSERT(!lc->ldo_is_composite);
4222 lod_comp = &lc->ldo_comp_entries[0];
4223 def_comp = &lds->lds_def_comp_entries[0];
4225 if (lod_comp->llc_stripe_count <= 0)
4226 lod_comp->llc_stripe_count =
4227 def_comp->llc_stripe_count;
4228 if (lod_comp->llc_stripe_size <= 0)
4229 lod_comp->llc_stripe_size =
4230 def_comp->llc_stripe_size;
4231 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT)
4232 lod_comp->llc_stripe_offset =
4233 def_comp->llc_stripe_offset;
4234 if (lod_comp->llc_pool == NULL)
4235 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
4240 * fs default striping may not be explicitly set, or historically set
4241 * in config log, use them.
4243 if (lod_need_inherit_more(lc, false)) {
4244 if (lc->ldo_comp_cnt == 0) {
4245 rc = lod_alloc_comp_entries(lc, 0, 1);
4247 /* fail to allocate memory, will create a
4248 * non-striped file. */
4250 lc->ldo_is_composite = 0;
4251 lod_comp = &lc->ldo_comp_entries[0];
4252 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4254 LASSERT(!lc->ldo_is_composite);
4255 lod_comp = &lc->ldo_comp_entries[0];
4256 desc = &d->lod_desc;
4257 if (lod_comp->llc_stripe_count <= 0)
4258 lod_comp->llc_stripe_count =
4259 desc->ld_default_stripe_count;
4260 if (lod_comp->llc_stripe_size <= 0)
4261 lod_comp->llc_stripe_size =
4262 desc->ld_default_stripe_size;
4268 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
4270 * Size initialization on late striping.
4272 * Propagate the size of a truncated object to a deferred striping.
4273 * This function handles a special case when truncate was done on a
4274 * non-striped object and now while the striping is being created
4275 * we can't lose that size, so we have to propagate it to the stripes
4278 * \param[in] env execution environment
4279 * \param[in] dt object
4280 * \param[in] th transaction handle
4282 * \retval 0 on success
4283 * \retval negative if failed
4285 static int lod_declare_init_size(const struct lu_env *env,
4286 struct dt_object *dt, struct thandle *th)
4288 struct dt_object *next = dt_object_child(dt);
4289 struct lod_object *lo = lod_dt_obj(dt);
4290 struct dt_object **objects = NULL;
4291 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4292 uint64_t size, offs;
4293 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
4294 struct lu_extent size_ext;
4297 if (!lod_obj_is_striped(dt))
4300 rc = dt_attr_get(env, next, attr);
4301 LASSERT(attr->la_valid & LA_SIZE);
4305 size = attr->la_size;
4309 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
4310 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4311 struct lod_layout_component *lod_comp;
4312 struct lu_extent *extent;
4314 lod_comp = &lo->ldo_comp_entries[i];
4316 if (lod_comp->llc_stripe == NULL)
4319 extent = &lod_comp->llc_extent;
4320 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
4321 if (!lo->ldo_is_composite ||
4322 lu_extent_is_overlapped(extent, &size_ext)) {
4323 objects = lod_comp->llc_stripe;
4324 stripe_count = lod_comp->llc_stripe_count;
4325 stripe_size = lod_comp->llc_stripe_size;
4328 if (stripe_count == 0)
4331 LASSERT(objects != NULL && stripe_size != 0);
4332 /* ll_do_div64(a, b) returns a % b, and a = a / b */
4333 ll_do_div64(size, (__u64)stripe_size);
4334 stripe = ll_do_div64(size, (__u64)stripe_count);
4335 LASSERT(objects[stripe] != NULL);
4337 size = size * stripe_size;
4338 offs = attr->la_size;
4339 size += ll_do_div64(offs, stripe_size);
4341 attr->la_valid = LA_SIZE;
4342 attr->la_size = size;
4344 rc = lod_sub_declare_attr_set(env, objects[stripe],
4353 * Declare creation of striped object.
4355 * The function declares creation stripes for a regular object. The function
4356 * also declares whether the stripes will be created with non-zero size if
4357 * previously size was set non-zero on the master object. If object \a dt is
4358 * not local, then only fully defined striping can be applied in \a lovea.
4359 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
4362 * \param[in] env execution environment
4363 * \param[in] dt object
4364 * \param[in] attr attributes the stripes will be created with
4365 * \param[in] lovea a buffer containing striping description
4366 * \param[in] th transaction handle
4368 * \retval 0 on success
4369 * \retval negative if failed
4371 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
4372 struct lu_attr *attr,
4373 const struct lu_buf *lovea, struct thandle *th)
4375 struct lod_thread_info *info = lod_env_info(env);
4376 struct dt_object *next = dt_object_child(dt);
4377 struct lod_object *lo = lod_dt_obj(dt);
4381 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
4382 GOTO(out, rc = -ENOMEM);
4384 if (!dt_object_remote(next)) {
4385 /* choose OST and generate appropriate objects */
4386 rc = lod_prepare_create(env, lo, attr, lovea, th);
4391 * declare storage for striping data
4393 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
4395 /* LOD can not choose OST objects for remote objects, i.e.
4396 * stripes must be ready before that. Right now, it can only
4397 * happen during migrate, i.e. migrate process needs to create
4398 * remote regular file (mdd_migrate_create), then the migrate
4399 * process will provide stripeEA. */
4400 LASSERT(lovea != NULL);
4401 info->lti_buf = *lovea;
4404 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
4405 XATTR_NAME_LOV, 0, th);
4410 * if striping is created with local object's size > 0,
4411 * we have to propagate this size to specific object
4412 * the case is possible only when local object was created previously
4414 if (dt_object_exists(next))
4415 rc = lod_declare_init_size(env, dt, th);
4418 /* failed to create striping or to set initial size, let's reset
4419 * config so that others don't get confused */
4421 lod_object_free_striping(env, lo);
4427 * Implementation of dt_object_operations::do_declare_create.
4429 * The method declares creation of a new object. If the object will be striped,
4430 * then helper functions are called to find FIDs for the stripes, declare
4431 * creation of the stripes and declare initialization of the striping
4432 * information to be stored in the master object.
4434 * \see dt_object_operations::do_declare_create() in the API description
4437 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
4438 struct lu_attr *attr,
4439 struct dt_allocation_hint *hint,
4440 struct dt_object_format *dof, struct thandle *th)
4442 struct dt_object *next = dt_object_child(dt);
4443 struct lod_object *lo = lod_dt_obj(dt);
4452 * first of all, we declare creation of local object
4454 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
4459 * it's lod_ah_init() that has decided the object will be striped
4461 if (dof->dof_type == DFT_REGULAR) {
4462 /* callers don't want stripes */
4463 /* XXX: all tricky interactions with ->ah_make_hint() decided
4464 * to use striping, then ->declare_create() behaving differently
4465 * should be cleaned */
4466 if (dof->u.dof_reg.striped != 0)
4467 rc = lod_declare_striped_create(env, dt, attr,
4469 } else if (dof->dof_type == DFT_DIR) {
4470 struct seq_server_site *ss;
4472 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
4474 /* If the parent has default stripeEA, and client
4475 * did not find it before sending create request,
4476 * then MDT will return -EREMOTE, and client will
4477 * retrieve the default stripeEA and re-create the
4480 * Note: if dah_eadata != NULL, it means creating the
4481 * striped directory with specified stripeEA, then it
4482 * should ignore the default stripeEA */
4483 if (hint != NULL && hint->dah_eadata == NULL) {
4484 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
4485 GOTO(out, rc = -EREMOTE);
4487 if (lo->ldo_dir_stripe_offset == -1) {
4488 /* child and parent should be in the same MDT */
4489 if (hint->dah_parent != NULL &&
4490 dt_object_remote(hint->dah_parent))
4491 GOTO(out, rc = -EREMOTE);
4492 } else if (lo->ldo_dir_stripe_offset !=
4494 struct lod_device *lod;
4495 struct lod_tgt_descs *ltd;
4496 struct lod_tgt_desc *tgt = NULL;
4497 bool found_mdt = false;
4500 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
4501 ltd = &lod->lod_mdt_descs;
4502 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
4503 tgt = LTD_TGT(ltd, i);
4504 if (tgt->ltd_index ==
4505 lo->ldo_dir_stripe_offset) {
4511 /* If the MDT indicated by stripe_offset can be
4512 * found, then tell client to resend the create
4513 * request to the correct MDT, otherwise return
4514 * error to client */
4516 GOTO(out, rc = -EREMOTE);
4518 GOTO(out, rc = -EINVAL);
4522 rc = lod_declare_dir_striping_create(env, dt, attr, dof, th);
4525 /* failed to create striping or to set initial size, let's reset
4526 * config so that others don't get confused */
4528 lod_object_free_striping(env, lo);
4533 * Generate component ID for new created component.
4535 * \param[in] lo LOD object
4536 * \param[in] comp_idx index of ldo_comp_entries
4538 * \retval component ID on success
4539 * \retval LCME_ID_INVAL on failure
4541 static __u32 lod_gen_component_id(struct lod_object *lo,
4542 int mirror_id, int comp_idx)
4544 struct lod_layout_component *lod_comp;
4545 __u32 id, start, end;
4548 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
4550 lod_obj_inc_layout_gen(lo);
4551 id = lo->ldo_layout_gen;
4552 if (likely(id <= SEQ_ID_MAX))
4553 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
4555 /* Layout generation wraps, need to check collisions. */
4556 start = id & SEQ_ID_MASK;
4559 for (id = start; id <= end; id++) {
4560 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4561 lod_comp = &lo->ldo_comp_entries[i];
4562 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
4565 /* Found the ununsed ID */
4566 if (i == lo->ldo_comp_cnt)
4567 RETURN(pflr_id(mirror_id, id));
4569 if (end == LCME_ID_MAX) {
4571 end = min(lo->ldo_layout_gen & LCME_ID_MASK,
4572 (__u32)(LCME_ID_MAX - 1));
4576 RETURN(LCME_ID_INVAL);
4580 * Creation of a striped regular object.
4582 * The function is called to create the stripe objects for a regular
4583 * striped file. This can happen at the initial object creation or
4584 * when the caller asks LOD to do so using ->do_xattr_set() method
4585 * (so called late striping). Notice all the information are already
4586 * prepared in the form of the list of objects (ldo_stripe field).
4587 * This is done during declare phase.
4589 * \param[in] env execution environment
4590 * \param[in] dt object
4591 * \param[in] attr attributes the stripes will be created with
4592 * \param[in] dof format of stripes (see OSD API description)
4593 * \param[in] th transaction handle
4595 * \retval 0 on success
4596 * \retval negative if failed
4598 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
4599 struct lu_attr *attr, struct dt_object_format *dof,
4602 struct lod_layout_component *lod_comp;
4603 struct lod_object *lo = lod_dt_obj(dt);
4608 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
4610 mirror_id = lo->ldo_mirror_count > 1 ? 1 : 0;
4612 /* create all underlying objects */
4613 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4614 lod_comp = &lo->ldo_comp_entries[i];
4616 if (lod_comp->llc_extent.e_start == 0 && i > 0) /* new mirror */
4619 if (lod_comp->llc_id == LCME_ID_INVAL) {
4620 lod_comp->llc_id = lod_gen_component_id(lo,
4622 if (lod_comp->llc_id == LCME_ID_INVAL)
4623 GOTO(out, rc = -ERANGE);
4626 if (lod_comp_inited(lod_comp))
4629 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
4630 lod_comp_set_init(lod_comp);
4632 if (lov_pattern(lod_comp->llc_pattern) == LOV_PATTERN_MDT)
4633 lod_comp_set_init(lod_comp);
4635 if (lod_comp->llc_stripe == NULL)
4638 LASSERT(lod_comp->llc_stripe_count);
4639 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4640 struct dt_object *object = lod_comp->llc_stripe[j];
4641 LASSERT(object != NULL);
4642 rc = lod_sub_create(env, object, attr, NULL, dof, th);
4646 lod_comp_set_init(lod_comp);
4649 rc = lod_fill_mirrors(lo);
4653 rc = lod_generate_and_set_lovea(env, lo, th);
4657 lo->ldo_comp_cached = 1;
4661 lod_object_free_striping(env, lo);
4666 * Implementation of dt_object_operations::do_create.
4668 * If any of preceeding methods (like ->do_declare_create(),
4669 * ->do_ah_init(), etc) chose to create a striped object,
4670 * then this method will create the master and the stripes.
4672 * \see dt_object_operations::do_create() in the API description for details.
4674 static int lod_create(const struct lu_env *env, struct dt_object *dt,
4675 struct lu_attr *attr, struct dt_allocation_hint *hint,
4676 struct dt_object_format *dof, struct thandle *th)
4681 /* create local object */
4682 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
4686 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
4687 lod_obj_is_striped(dt) && dof->u.dof_reg.striped != 0) {
4688 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
4689 rc = lod_striped_create(env, dt, attr, dof, th);
4696 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
4697 struct dt_object *dt, struct thandle *th,
4698 int comp_idx, int stripe_idx,
4699 struct lod_obj_stripe_cb_data *data)
4701 if (data->locd_declare)
4702 return lod_sub_declare_destroy(env, dt, th);
4703 else if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
4704 stripe_idx == cfs_fail_val)
4705 return lod_sub_destroy(env, dt, th);
4711 * Implementation of dt_object_operations::do_declare_destroy.
4713 * If the object is a striped directory, then the function declares reference
4714 * removal from the master object (this is an index) to the stripes and declares
4715 * destroy of all the stripes. In all the cases, it declares an intention to
4716 * destroy the object itself.
4718 * \see dt_object_operations::do_declare_destroy() in the API description
4721 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
4724 struct dt_object *next = dt_object_child(dt);
4725 struct lod_object *lo = lod_dt_obj(dt);
4726 struct lod_thread_info *info = lod_env_info(env);
4727 char *stripe_name = info->lti_key;
4732 * load striping information, notice we don't do this when object
4733 * is being initialized as we don't need this information till
4734 * few specific cases like destroy, chown
4736 rc = lod_load_striping(env, lo);
4740 /* declare destroy for all underlying objects */
4741 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4742 rc = next->do_ops->do_index_try(env, next,
4743 &dt_directory_features);
4747 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4748 rc = lod_sub_declare_ref_del(env, next, th);
4752 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4753 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
4755 rc = lod_sub_declare_delete(env, next,
4756 (const struct dt_key *)stripe_name, th);
4763 * we declare destroy for the local object
4765 rc = lod_sub_declare_destroy(env, next, th);
4769 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
4770 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
4773 if (!lod_obj_is_striped(dt))
4776 /* declare destroy all striped objects */
4777 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4778 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4779 if (lo->ldo_stripe[i] == NULL)
4782 rc = lod_sub_declare_ref_del(env, lo->ldo_stripe[i],
4785 rc = lod_sub_declare_destroy(env, lo->ldo_stripe[i],
4791 struct lod_obj_stripe_cb_data data = { { 0 } };
4793 data.locd_declare = true;
4794 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
4795 rc = lod_obj_for_each_stripe(env, lo, th, &data);
4802 * Implementation of dt_object_operations::do_destroy.
4804 * If the object is a striped directory, then the function removes references
4805 * from the master object (this is an index) to the stripes and destroys all
4806 * the stripes. In all the cases, the function destroys the object itself.
4808 * \see dt_object_operations::do_destroy() in the API description for details.
4810 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
4813 struct dt_object *next = dt_object_child(dt);
4814 struct lod_object *lo = lod_dt_obj(dt);
4815 struct lod_thread_info *info = lod_env_info(env);
4816 char *stripe_name = info->lti_key;
4821 /* destroy sub-stripe of master object */
4822 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4823 rc = next->do_ops->do_index_try(env, next,
4824 &dt_directory_features);
4828 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4829 rc = lod_sub_ref_del(env, next, th);
4833 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4834 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
4837 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
4838 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
4839 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)));
4841 rc = lod_sub_delete(env, next,
4842 (const struct dt_key *)stripe_name, th);
4848 rc = lod_sub_destroy(env, next, th);
4852 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
4853 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
4856 if (!lod_obj_is_striped(dt))
4859 /* destroy all striped objects */
4860 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4861 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4862 if (lo->ldo_stripe[i] == NULL)
4864 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
4865 i == cfs_fail_val) {
4866 dt_write_lock(env, lo->ldo_stripe[i],
4868 rc = lod_sub_ref_del(env, lo->ldo_stripe[i],
4870 dt_write_unlock(env, lo->ldo_stripe[i]);
4874 rc = lod_sub_destroy(env, lo->ldo_stripe[i],
4881 struct lod_obj_stripe_cb_data data = { { 0 } };
4883 data.locd_declare = false;
4884 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
4885 rc = lod_obj_for_each_stripe(env, lo, th, &data);
4892 * Implementation of dt_object_operations::do_declare_ref_add.
4894 * \see dt_object_operations::do_declare_ref_add() in the API description
4897 static int lod_declare_ref_add(const struct lu_env *env,
4898 struct dt_object *dt, struct thandle *th)
4900 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
4904 * Implementation of dt_object_operations::do_ref_add.
4906 * \see dt_object_operations::do_ref_add() in the API description for details.
4908 static int lod_ref_add(const struct lu_env *env,
4909 struct dt_object *dt, struct thandle *th)
4911 return lod_sub_ref_add(env, dt_object_child(dt), th);
4915 * Implementation of dt_object_operations::do_declare_ref_del.
4917 * \see dt_object_operations::do_declare_ref_del() in the API description
4920 static int lod_declare_ref_del(const struct lu_env *env,
4921 struct dt_object *dt, struct thandle *th)
4923 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
4927 * Implementation of dt_object_operations::do_ref_del
4929 * \see dt_object_operations::do_ref_del() in the API description for details.
4931 static int lod_ref_del(const struct lu_env *env,
4932 struct dt_object *dt, struct thandle *th)
4934 return lod_sub_ref_del(env, dt_object_child(dt), th);
4938 * Implementation of dt_object_operations::do_object_sync.
4940 * \see dt_object_operations::do_object_sync() in the API description
4943 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
4944 __u64 start, __u64 end)
4946 return dt_object_sync(env, dt_object_child(dt), start, end);
4950 * Release LDLM locks on the stripes of a striped directory.
4952 * Iterates over all the locks taken on the stripe objects and
4955 * \param[in] env execution environment
4956 * \param[in] dt striped object
4957 * \param[in] einfo lock description
4958 * \param[in] policy data describing requested lock
4960 * \retval 0 on success
4961 * \retval negative if failed
4963 static int lod_object_unlock_internal(const struct lu_env *env,
4964 struct dt_object *dt,
4965 struct ldlm_enqueue_info *einfo,
4966 union ldlm_policy_data *policy)
4968 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
4973 if (slave_locks == NULL)
4976 for (i = 1; i < slave_locks->count; i++) {
4977 if (lustre_handle_is_used(&slave_locks->handles[i]))
4978 ldlm_lock_decref_and_cancel(&slave_locks->handles[i],
4986 * Implementation of dt_object_operations::do_object_unlock.
4988 * Used to release LDLM lock(s).
4990 * \see dt_object_operations::do_object_unlock() in the API description
4993 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
4994 struct ldlm_enqueue_info *einfo,
4995 union ldlm_policy_data *policy)
4997 struct lod_object *lo = lod_dt_obj(dt);
4998 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
4999 int slave_locks_size;
5003 if (slave_locks == NULL)
5006 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
5007 LASSERT(lo->ldo_dir_stripe_count > 1);
5008 /* Note: for remote lock for single stripe dir, MDT will cancel
5009 * the lock by lockh directly */
5010 LASSERT(!dt_object_remote(dt_object_child(dt)));
5012 /* locks were unlocked in MDT layer */
5013 for (i = 1; i < slave_locks->count; i++) {
5014 LASSERT(!lustre_handle_is_used(&slave_locks->handles[i]));
5015 dt_invalidate(env, lo->ldo_stripe[i]);
5018 slave_locks_size = sizeof(*slave_locks) + slave_locks->count *
5019 sizeof(slave_locks->handles[0]);
5020 OBD_FREE(slave_locks, slave_locks_size);
5021 einfo->ei_cbdata = NULL;
5027 * Implementation of dt_object_operations::do_object_lock.
5029 * Used to get LDLM lock on the non-striped and striped objects.
5031 * \see dt_object_operations::do_object_lock() in the API description
5034 static int lod_object_lock(const struct lu_env *env,
5035 struct dt_object *dt,
5036 struct lustre_handle *lh,
5037 struct ldlm_enqueue_info *einfo,
5038 union ldlm_policy_data *policy)
5040 struct lod_object *lo = lod_dt_obj(dt);
5043 int slave_locks_size;
5044 struct lustre_handle_array *slave_locks = NULL;
5047 /* remote object lock */
5048 if (!einfo->ei_enq_slave) {
5049 LASSERT(dt_object_remote(dt));
5050 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
5054 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
5055 GOTO(out, rc = -ENOTDIR);
5057 rc = lod_load_striping(env, lo);
5062 if (lo->ldo_dir_stripe_count <= 1) {
5064 * NB, ei_cbdata stores pointer to slave locks, if no locks
5065 * taken, make sure it's set to NULL, otherwise MDT will try to
5068 einfo->ei_cbdata = NULL;
5072 slave_locks_size = sizeof(*slave_locks) + lo->ldo_dir_stripe_count *
5073 sizeof(slave_locks->handles[0]);
5074 /* Freed in lod_object_unlock */
5075 OBD_ALLOC(slave_locks, slave_locks_size);
5076 if (slave_locks == NULL)
5077 GOTO(out, rc = -ENOMEM);
5078 slave_locks->count = lo->ldo_dir_stripe_count;
5080 /* striped directory lock */
5081 for (i = 1; i < lo->ldo_dir_stripe_count; i++) {
5082 struct lustre_handle lockh;
5083 struct ldlm_res_id *res_id;
5085 res_id = &lod_env_info(env)->lti_res_id;
5086 fid_build_reg_res_name(lu_object_fid(&lo->ldo_stripe[i]->do_lu),
5088 einfo->ei_res_id = res_id;
5090 LASSERT(lo->ldo_stripe[i] != NULL);
5091 if (likely(dt_object_remote(lo->ldo_stripe[i]))) {
5092 rc = dt_object_lock(env, lo->ldo_stripe[i], &lockh,
5095 struct ldlm_namespace *ns = einfo->ei_namespace;
5096 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
5097 ldlm_completion_callback completion = einfo->ei_cb_cp;
5098 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
5100 if (einfo->ei_mode == LCK_PW ||
5101 einfo->ei_mode == LCK_EX)
5102 dlmflags |= LDLM_FL_COS_INCOMPAT;
5104 /* This only happens if there are mulitple stripes
5105 * on the master MDT, i.e. except stripe0, there are
5106 * other stripes on the Master MDT as well, Only
5107 * happens in the test case right now. */
5108 LASSERT(ns != NULL);
5109 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS,
5110 policy, einfo->ei_mode,
5111 &dlmflags, blocking,
5113 NULL, 0, LVB_T_NONE,
5118 slave_locks->handles[i] = lockh;
5120 einfo->ei_cbdata = slave_locks;
5122 if (rc != 0 && slave_locks != NULL) {
5123 lod_object_unlock_internal(env, dt, einfo, policy);
5124 OBD_FREE(slave_locks, slave_locks_size);
5129 einfo->ei_cbdata = NULL;
5134 * Implementation of dt_object_operations::do_invalidate.
5136 * \see dt_object_operations::do_invalidate() in the API description for details
5138 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
5140 return dt_invalidate(env, dt_object_child(dt));
5143 static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
5147 /* clear memory region that will be used for layout change */
5148 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
5149 info->lti_count = 0;
5151 if (info->lti_comp_size >= comp_cnt)
5154 if (info->lti_comp_size > 0) {
5155 OBD_FREE(info->lti_comp_idx,
5156 info->lti_comp_size * sizeof(__u32));
5157 info->lti_comp_size = 0;
5160 OBD_ALLOC(info->lti_comp_idx, comp_cnt * sizeof(__u32));
5161 if (!info->lti_comp_idx)
5164 info->lti_comp_size = comp_cnt;
5168 static int lod_declare_instantiate_components(const struct lu_env *env,
5169 struct lod_object *lo, struct thandle *th)
5171 struct lod_thread_info *info = lod_env_info(env);
5172 struct ost_pool *inuse = &info->lti_inuse_osts;
5177 LASSERT(info->lti_count < lo->ldo_comp_cnt);
5178 if (info->lti_count > 0) {
5179 /* Prepare inuse array for composite file */
5180 rc = lod_prepare_inuse(env, lo);
5185 for (i = 0; i < info->lti_count; i++) {
5186 rc = lod_qos_prep_create(env, lo, NULL, th,
5187 info->lti_comp_idx[i], inuse);
5193 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
5194 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
5195 &info->lti_buf, XATTR_NAME_LOV, 0, th);
5201 static int lod_declare_update_plain(const struct lu_env *env,
5202 struct lod_object *lo, struct layout_intent *layout,
5203 const struct lu_buf *buf, struct thandle *th)
5205 struct lod_thread_info *info = lod_env_info(env);
5206 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5207 struct lod_layout_component *lod_comp;
5208 struct lov_comp_md_v1 *comp_v1 = NULL;
5209 bool replay = false;
5213 LASSERT(lo->ldo_flr_state == LCM_FL_NOT_FLR);
5216 * In case the client is passing lovea, which only happens during
5217 * the replay of layout intent write RPC for now, we may need to
5218 * parse the lovea and apply new layout configuration.
5220 if (buf && buf->lb_len) {
5221 struct lov_user_md_v1 *v1 = buf->lb_buf;
5223 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
5224 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
5225 LOV_MAGIC_COMP_V1)) {
5226 CERROR("%s: the replay buffer of layout extend "
5227 "(magic %#x) does not contain expected "
5228 "composite layout.\n",
5229 lod2obd(d)->obd_name, v1->lmm_magic);
5230 GOTO(out, rc = -EINVAL);
5233 lod_object_free_striping(env, lo);
5234 rc = lod_use_defined_striping(env, lo, buf);
5238 rc = lod_get_lov_ea(env, lo);
5241 /* old on-disk EA is stored in info->lti_buf */
5242 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
5245 /* non replay path */
5246 rc = lod_load_striping_locked(env, lo);
5251 /* Make sure defined layout covers the requested write range. */
5252 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
5253 if (lo->ldo_comp_cnt > 1 &&
5254 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
5255 lod_comp->llc_extent.e_end < layout->li_extent.e_end) {
5256 CDEBUG(replay ? D_ERROR : D_LAYOUT,
5257 "%s: the defined layout [0, %#llx) does not covers "
5258 "the write range "DEXT"\n",
5259 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
5260 PEXT(&layout->li_extent));
5261 GOTO(out, rc = -EINVAL);
5264 CDEBUG(D_LAYOUT, "%s: "DFID": instantiate components "DEXT"\n",
5265 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
5266 PEXT(&layout->li_extent));
5269 * Iterate ld->ldo_comp_entries, find the component whose extent under
5270 * the write range and not instantianted.
5272 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5273 lod_comp = &lo->ldo_comp_entries[i];
5275 if (lod_comp->llc_extent.e_start >= layout->li_extent.e_end)
5279 if (lod_comp_inited(lod_comp))
5283 * In replay path, lod_comp is the EA passed by
5284 * client replay buffer, comp_v1 is the pre-recovery
5285 * on-disk EA, we'd sift out those components which
5286 * were init-ed in the on-disk EA.
5288 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
5293 * this component hasn't instantiated in normal path, or during
5294 * replay it needs replay the instantiation.
5297 /* A released component is being extended */
5298 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
5299 GOTO(out, rc = -EINVAL);
5301 LASSERT(info->lti_comp_idx != NULL);
5302 info->lti_comp_idx[info->lti_count++] = i;
5305 if (info->lti_count == 0)
5308 lod_obj_inc_layout_gen(lo);
5309 rc = lod_declare_instantiate_components(env, lo, th);
5312 lod_object_free_striping(env, lo);
5316 #define lod_foreach_mirror_comp(comp, lo, mirror_idx) \
5317 for (comp = &lo->ldo_comp_entries[lo->ldo_mirrors[mirror_idx].lme_start]; \
5318 comp <= &lo->ldo_comp_entries[lo->ldo_mirrors[mirror_idx].lme_end]; \
5321 static inline int lod_comp_index(struct lod_object *lo,
5322 struct lod_layout_component *lod_comp)
5324 LASSERT(lod_comp >= lo->ldo_comp_entries &&
5325 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
5327 return lod_comp - lo->ldo_comp_entries;
5331 * Stale other mirrors by writing extent.
5333 static void lod_stale_components(struct lod_object *lo, int primary,
5334 struct lu_extent *extent)
5336 struct lod_layout_component *pri_comp, *lod_comp;
5339 /* The writing extent decides which components in the primary
5340 * are affected... */
5341 lod_foreach_mirror_comp(pri_comp, lo, primary) {
5342 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
5345 for (i = 0; i < lo->ldo_mirror_count; i++) {
5349 /* ... and then stale other components that are
5350 * overlapping with primary components */
5351 lod_foreach_mirror_comp(lod_comp, lo, i) {
5352 if (!lu_extent_is_overlapped(
5353 &pri_comp->llc_extent,
5354 &lod_comp->llc_extent))
5357 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
5358 i, lod_comp_index(lo, lod_comp));
5360 lod_comp->llc_flags |= LCME_FL_STALE;
5361 lo->ldo_mirrors[i].lme_stale = 1;
5367 static int lod_declare_update_rdonly(const struct lu_env *env,
5368 struct lod_object *lo, struct md_layout_change *mlc,
5371 struct lod_thread_info *info = lod_env_info(env);
5372 struct lu_attr *layout_attr = &info->lti_layout_attr;
5373 struct lod_layout_component *lod_comp;
5374 struct layout_intent *layout = mlc->mlc_intent;
5375 struct lu_extent extent = layout->li_extent;
5376 unsigned int seq = 0;
5382 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE);
5383 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
5384 LASSERT(lo->ldo_mirror_count > 0);
5386 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
5387 PFID(lod_object_fid(lo)), PEXT(&extent));
5389 if (OBD_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
5390 get_random_bytes(&seq, sizeof(seq));
5391 seq %= lo->ldo_mirror_count;
5395 * Pick a mirror as the primary.
5396 * Now it only picks the first mirror, this algo can be
5397 * revised later after knowing the topology of cluster or
5398 * the availability of OSTs.
5400 for (picked = -1, i = 0; i < lo->ldo_mirror_count; i++) {
5401 int index = (i + seq) % lo->ldo_mirror_count;
5403 if (!lo->ldo_mirrors[index].lme_stale) {
5408 if (picked < 0) /* failed to pick a primary */
5411 CDEBUG(D_LAYOUT, DFID": picked mirror %u as primary\n",
5412 PFID(lod_object_fid(lo)), lo->ldo_mirrors[picked].lme_id);
5414 /* stale overlapping components from other mirrors */
5415 lod_stale_components(lo, picked, &extent);
5417 /* instantiate components for the picked mirror, start from 0 */
5418 extent = (struct lu_extent) { 0, layout->li_extent.e_end };
5419 lod_foreach_mirror_comp(lod_comp, lo, picked) {
5420 if (!lu_extent_is_overlapped(&extent,
5421 &lod_comp->llc_extent))
5424 if (lod_comp_inited(lod_comp))
5427 CDEBUG(D_LAYOUT, "instantiate: %u / %u\n",
5428 i, lod_comp_index(lo, lod_comp));
5430 info->lti_comp_idx[info->lti_count++] =
5431 lod_comp_index(lo, lod_comp);
5434 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
5436 /* Reset the layout version once it's becoming too large.
5437 * This way it can make sure that the layout version is
5438 * monotonously increased in this writing era. */
5439 lod_obj_inc_layout_gen(lo);
5440 if (lo->ldo_layout_gen > (LCME_ID_MAX >> 1)) {
5441 __u32 layout_version;
5443 cfs_get_random_bytes(&layout_version, sizeof(layout_version));
5444 lo->ldo_layout_gen = layout_version & 0xffff;
5447 rc = lod_declare_instantiate_components(env, lo, th);
5451 layout_attr->la_valid = LA_LAYOUT_VERSION;
5452 layout_attr->la_layout_version = 0; /* set current version */
5453 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
5459 lod_object_free_striping(env, lo);
5463 static int lod_declare_update_write_pending(const struct lu_env *env,
5464 struct lod_object *lo, struct md_layout_change *mlc,
5467 struct lod_thread_info *info = lod_env_info(env);
5468 struct lu_attr *layout_attr = &info->lti_layout_attr;
5469 struct lod_layout_component *lod_comp;
5470 struct lu_extent extent = { 0 };
5476 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
5477 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
5478 mlc->mlc_opc == MD_LAYOUT_RESYNC);
5480 /* look for the primary mirror */
5481 for (i = 0; i < lo->ldo_mirror_count; i++) {
5482 if (lo->ldo_mirrors[i].lme_stale)
5485 LASSERTF(primary < 0, DFID " has multiple primary: %u / %u",
5486 PFID(lod_object_fid(lo)),
5487 lo->ldo_mirrors[i].lme_id,
5488 lo->ldo_mirrors[primary].lme_id);
5493 CERROR(DFID ": doesn't have a primary mirror\n",
5494 PFID(lod_object_fid(lo)));
5495 GOTO(out, rc = -ENODATA);
5498 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
5499 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
5501 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
5503 /* for LAYOUT_WRITE opc, it has to do the following operations:
5504 * 1. stale overlapping componets from stale mirrors;
5505 * 2. instantiate components of the primary mirror;
5506 * 3. transfter layout version to all objects of the primary;
5508 * for LAYOUT_RESYNC opc, it will do:
5509 * 1. instantiate components of all stale mirrors;
5510 * 2. transfer layout version to all objects to close write era. */
5512 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
5513 LASSERT(mlc->mlc_intent != NULL);
5515 extent = mlc->mlc_intent->li_extent;
5517 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
5518 PFID(lod_object_fid(lo)), PEXT(&extent));
5520 /* 1. stale overlapping components */
5521 lod_stale_components(lo, primary, &extent);
5523 /* 2. find out the components need instantiating.
5524 * instantiate [0, mlc->mlc_intent->e_end) */
5526 lod_foreach_mirror_comp(lod_comp, lo, primary) {
5527 if (!lu_extent_is_overlapped(&extent,
5528 &lod_comp->llc_extent))
5531 if (lod_comp_inited(lod_comp))
5534 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
5535 primary, lod_comp_index(lo, lod_comp));
5536 info->lti_comp_idx[info->lti_count++] =
5537 lod_comp_index(lo, lod_comp);
5539 } else { /* MD_LAYOUT_RESYNC */
5540 /* figure out the components that have been instantiated in
5541 * in primary to decide what components should be instantiated
5542 * in stale mirrors */
5543 lod_foreach_mirror_comp(lod_comp, lo, primary) {
5544 if (!lod_comp_inited(lod_comp))
5547 extent.e_end = lod_comp->llc_extent.e_end;
5551 DFID": instantiate all stale components in "DEXT"\n",
5552 PFID(lod_object_fid(lo)), PEXT(&extent));
5554 /* 1. instantiate all components within this extent, even
5555 * non-stale components so that it won't need to instantiate
5556 * those components for mirror truncate later. */
5557 for (i = 0; i < lo->ldo_mirror_count; i++) {
5561 LASSERTF(lo->ldo_mirrors[i].lme_stale,
5562 "both %d and %d are primary\n", i, primary);
5564 lod_foreach_mirror_comp(lod_comp, lo, i) {
5565 if (!lu_extent_is_overlapped(&extent,
5566 &lod_comp->llc_extent))
5569 if (lod_comp_inited(lod_comp))
5572 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
5573 i, lod_comp_index(lo, lod_comp));
5575 info->lti_comp_idx[info->lti_count++] =
5576 lod_comp_index(lo, lod_comp);
5580 /* change the file state to SYNC_PENDING */
5581 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
5584 rc = lod_declare_instantiate_components(env, lo, th);
5588 /* 3. transfer layout version to OST objects.
5589 * transfer new layout version to OST objects so that stale writes
5590 * can be denied. It also ends an era of writing by setting
5591 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
5592 * send write RPC; only resync RPCs could do it. */
5593 layout_attr->la_valid = LA_LAYOUT_VERSION;
5594 layout_attr->la_layout_version = 0; /* set current version */
5595 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
5596 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
5597 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
5601 lod_obj_inc_layout_gen(lo);
5604 lod_object_free_striping(env, lo);
5608 static int lod_declare_update_sync_pending(const struct lu_env *env,
5609 struct lod_object *lo, struct md_layout_change *mlc,
5612 struct lod_thread_info *info = lod_env_info(env);
5613 unsigned sync_components = 0;
5614 unsigned resync_components = 0;
5619 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
5620 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
5621 mlc->mlc_opc == MD_LAYOUT_WRITE);
5623 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
5624 PFID(lod_object_fid(lo)), mlc->mlc_opc);
5626 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
5627 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
5628 PFID(lod_object_fid(lo)));
5630 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
5631 return lod_declare_update_write_pending(env, lo, mlc, th);
5634 /* MD_LAYOUT_RESYNC_DONE */
5636 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5637 struct lod_layout_component *lod_comp;
5640 lod_comp = &lo->ldo_comp_entries[i];
5642 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
5647 for (j = 0; j < mlc->mlc_resync_count; j++) {
5648 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
5651 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
5652 lod_comp->llc_flags &= ~LCME_FL_STALE;
5653 resync_components++;
5659 for (i = 0; i < mlc->mlc_resync_count; i++) {
5660 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
5663 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
5664 "or already synced\n", PFID(lod_object_fid(lo)),
5665 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
5666 GOTO(out, rc = -EINVAL);
5669 if (!sync_components || !resync_components) {
5670 CDEBUG(D_LAYOUT, DFID": no mirror in sync or resync\n",
5671 PFID(lod_object_fid(lo)));
5673 /* tend to return an error code here to prevent
5674 * the MDT from setting SoM attribute */
5675 GOTO(out, rc = -EINVAL);
5678 CDEBUG(D_LAYOUT, DFID": resynced %u/%zu components\n",
5679 PFID(lod_object_fid(lo)),
5680 resync_components, mlc->mlc_resync_count);
5682 lo->ldo_flr_state = LCM_FL_RDONLY;
5683 lod_obj_inc_layout_gen(lo);
5685 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
5686 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
5687 &info->lti_buf, XATTR_NAME_LOV, 0, th);
5692 lod_object_free_striping(env, lo);
5696 static int lod_declare_layout_change(const struct lu_env *env,
5697 struct dt_object *dt, struct md_layout_change *mlc,
5700 struct lod_thread_info *info = lod_env_info(env);
5701 struct lod_object *lo = lod_dt_obj(dt);
5705 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
5706 dt_object_remote(dt_object_child(dt)))
5709 lod_write_lock(env, dt, 0);
5710 rc = lod_load_striping_locked(env, lo);
5714 LASSERT(lo->ldo_comp_cnt > 0);
5716 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
5720 switch (lo->ldo_flr_state) {
5721 case LCM_FL_NOT_FLR:
5722 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
5726 rc = lod_declare_update_rdonly(env, lo, mlc, th);
5728 case LCM_FL_WRITE_PENDING:
5729 rc = lod_declare_update_write_pending(env, lo, mlc, th);
5731 case LCM_FL_SYNC_PENDING:
5732 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
5739 dt_write_unlock(env, dt);
5744 * Instantiate layout component objects which covers the intent write offset.
5746 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
5747 struct md_layout_change *mlc, struct thandle *th)
5749 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
5750 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
5751 struct lod_object *lo = lod_dt_obj(dt);
5754 rc = lod_striped_create(env, dt, attr, NULL, th);
5755 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
5756 layout_attr->la_layout_version |= lo->ldo_layout_gen;
5757 rc = lod_attr_set(env, dt, layout_attr, th);
5763 struct dt_object_operations lod_obj_ops = {
5764 .do_read_lock = lod_read_lock,
5765 .do_write_lock = lod_write_lock,
5766 .do_read_unlock = lod_read_unlock,
5767 .do_write_unlock = lod_write_unlock,
5768 .do_write_locked = lod_write_locked,
5769 .do_attr_get = lod_attr_get,
5770 .do_declare_attr_set = lod_declare_attr_set,
5771 .do_attr_set = lod_attr_set,
5772 .do_xattr_get = lod_xattr_get,
5773 .do_declare_xattr_set = lod_declare_xattr_set,
5774 .do_xattr_set = lod_xattr_set,
5775 .do_declare_xattr_del = lod_declare_xattr_del,
5776 .do_xattr_del = lod_xattr_del,
5777 .do_xattr_list = lod_xattr_list,
5778 .do_ah_init = lod_ah_init,
5779 .do_declare_create = lod_declare_create,
5780 .do_create = lod_create,
5781 .do_declare_destroy = lod_declare_destroy,
5782 .do_destroy = lod_destroy,
5783 .do_index_try = lod_index_try,
5784 .do_declare_ref_add = lod_declare_ref_add,
5785 .do_ref_add = lod_ref_add,
5786 .do_declare_ref_del = lod_declare_ref_del,
5787 .do_ref_del = lod_ref_del,
5788 .do_object_sync = lod_object_sync,
5789 .do_object_lock = lod_object_lock,
5790 .do_object_unlock = lod_object_unlock,
5791 .do_invalidate = lod_invalidate,
5792 .do_declare_layout_change = lod_declare_layout_change,
5793 .do_layout_change = lod_layout_change,
5797 * Implementation of dt_body_operations::dbo_read.
5799 * \see dt_body_operations::dbo_read() in the API description for details.
5801 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
5802 struct lu_buf *buf, loff_t *pos)
5804 struct dt_object *next = dt_object_child(dt);
5806 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
5807 S_ISLNK(dt->do_lu.lo_header->loh_attr));
5808 return next->do_body_ops->dbo_read(env, next, buf, pos);
5812 * Implementation of dt_body_operations::dbo_declare_write.
5814 * \see dt_body_operations::dbo_declare_write() in the API description
5817 static ssize_t lod_declare_write(const struct lu_env *env,
5818 struct dt_object *dt,
5819 const struct lu_buf *buf, loff_t pos,
5822 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
5826 * Implementation of dt_body_operations::dbo_write.
5828 * \see dt_body_operations::dbo_write() in the API description for details.
5830 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
5831 const struct lu_buf *buf, loff_t *pos,
5832 struct thandle *th, int iq)
5834 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
5835 S_ISLNK(dt->do_lu.lo_header->loh_attr));
5836 return lod_sub_write(env, dt_object_child(dt), buf, pos, th, iq);
5839 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
5840 __u64 start, __u64 end, struct thandle *th)
5842 if (dt_object_remote(dt))
5845 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
5848 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
5849 __u64 start, __u64 end, struct thandle *th)
5851 if (dt_object_remote(dt))
5854 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
5855 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
5859 * different type of files use the same body_ops because object may be created
5860 * in OUT, where there is no chance to set correct body_ops for each type, so
5861 * body_ops themselves will check file type inside, see lod_read/write/punch for
5864 const struct dt_body_operations lod_body_ops = {
5865 .dbo_read = lod_read,
5866 .dbo_declare_write = lod_declare_write,
5867 .dbo_write = lod_write,
5868 .dbo_declare_punch = lod_declare_punch,
5869 .dbo_punch = lod_punch,
5873 * Implementation of lu_object_operations::loo_object_init.
5875 * The function determines the type and the index of the target device using
5876 * sequence of the object's FID. Then passes control down to the
5877 * corresponding device:
5878 * OSD for the local objects, OSP for remote
5880 * \see lu_object_operations::loo_object_init() in the API description
5883 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
5884 const struct lu_object_conf *conf)
5886 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
5887 struct lu_device *cdev = NULL;
5888 struct lu_object *cobj;
5889 struct lod_tgt_descs *ltd = NULL;
5890 struct lod_tgt_desc *tgt;
5892 int type = LU_SEQ_RANGE_ANY;
5896 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
5898 /* Note: Sometimes, it will Return EAGAIN here, see
5899 * ptrlpc_import_delay_req(), which might confuse
5900 * lu_object_find_at() and make it wait there incorrectly.
5901 * so we convert it to EIO here.*/
5908 if (type == LU_SEQ_RANGE_MDT &&
5909 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
5910 cdev = &lod->lod_child->dd_lu_dev;
5911 } else if (type == LU_SEQ_RANGE_MDT) {
5912 ltd = &lod->lod_mdt_descs;
5914 } else if (type == LU_SEQ_RANGE_OST) {
5915 ltd = &lod->lod_ost_descs;
5922 if (ltd->ltd_tgts_size > idx &&
5923 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
5924 tgt = LTD_TGT(ltd, idx);
5926 LASSERT(tgt != NULL);
5927 LASSERT(tgt->ltd_tgt != NULL);
5929 cdev = &(tgt->ltd_tgt->dd_lu_dev);
5931 lod_putref(lod, ltd);
5934 if (unlikely(cdev == NULL))
5937 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
5938 if (unlikely(cobj == NULL))
5941 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
5943 lu_object_add(lo, cobj);
5950 * Release resources associated with striping.
5952 * If the object is striped (regular or directory), then release
5953 * the stripe objects references and free the ldo_stripe array.
5955 * \param[in] env execution environment
5956 * \param[in] lo object
5958 void lod_object_free_striping(const struct lu_env *env, struct lod_object *lo)
5960 struct lod_layout_component *lod_comp;
5963 if (lo->ldo_stripe != NULL) {
5964 LASSERT(lo->ldo_comp_entries == NULL);
5965 LASSERT(lo->ldo_dir_stripes_allocated > 0);
5967 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5968 if (lo->ldo_stripe[i])
5969 dt_object_put(env, lo->ldo_stripe[i]);
5972 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
5973 OBD_FREE(lo->ldo_stripe, j);
5974 lo->ldo_stripe = NULL;
5975 lo->ldo_dir_stripes_allocated = 0;
5976 lo->ldo_dir_stripe_loaded = 0;
5977 lo->ldo_dir_stripe_count = 0;
5978 } else if (lo->ldo_comp_entries != NULL) {
5979 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5980 /* free lod_layout_component::llc_stripe array */
5981 lod_comp = &lo->ldo_comp_entries[i];
5983 if (lod_comp->llc_stripe == NULL)
5985 LASSERT(lod_comp->llc_stripes_allocated != 0);
5986 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
5987 if (lod_comp->llc_stripe[j] != NULL)
5989 &lod_comp->llc_stripe[j]->do_lu);
5991 OBD_FREE(lod_comp->llc_stripe,
5992 sizeof(struct dt_object *) *
5993 lod_comp->llc_stripes_allocated);
5994 lod_comp->llc_stripe = NULL;
5995 lod_comp->llc_stripes_allocated = 0;
5997 lod_free_comp_entries(lo);
5998 lo->ldo_comp_cached = 0;
6003 * Implementation of lu_object_operations::loo_object_free.
6005 * \see lu_object_operations::loo_object_free() in the API description
6008 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
6010 struct lod_object *lo = lu2lod_obj(o);
6012 /* release all underlying object pinned */
6013 lod_object_free_striping(env, lo);
6015 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
6019 * Implementation of lu_object_operations::loo_object_release.
6021 * \see lu_object_operations::loo_object_release() in the API description
6024 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
6026 /* XXX: shouldn't we release everything here in case if object
6027 * creation failed before? */
6031 * Implementation of lu_object_operations::loo_object_print.
6033 * \see lu_object_operations::loo_object_print() in the API description
6036 static int lod_object_print(const struct lu_env *env, void *cookie,
6037 lu_printer_t p, const struct lu_object *l)
6039 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
6041 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
6044 struct lu_object_operations lod_lu_obj_ops = {
6045 .loo_object_init = lod_object_init,
6046 .loo_object_free = lod_object_free,
6047 .loo_object_release = lod_object_release,
6048 .loo_object_print = lod_object_print,