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,
1050 struct thandle *th, lod_obj_stripe_cb_t cb,
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 LASSERT(lod_comp->llc_stripe_count > 0);
1065 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
1066 struct dt_object *dt = lod_comp->llc_stripe[j];
1070 rc = cb(env, lo, dt, th, j, data);
1079 lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
1080 struct dt_object *dt, struct thandle *th,
1081 int stripe_idx, struct lod_obj_stripe_cb_data *data)
1083 if (data->locd_declare)
1084 return lod_sub_declare_attr_set(env, dt, data->locd_attr, th);
1086 return lod_sub_attr_set(env, dt, data->locd_attr, th);
1090 * Implementation of dt_object_operations::do_declare_attr_set.
1092 * If the object is striped, then apply the changes to all the stripes.
1094 * \see dt_object_operations::do_declare_attr_set() in the API description
1097 static int lod_declare_attr_set(const struct lu_env *env,
1098 struct dt_object *dt,
1099 const struct lu_attr *attr,
1102 struct dt_object *next = dt_object_child(dt);
1103 struct lod_object *lo = lod_dt_obj(dt);
1108 * declare setattr on the local object
1110 rc = lod_sub_declare_attr_set(env, next, attr, th);
1114 /* osp_declare_attr_set() ignores all attributes other than
1115 * UID, GID, PROJID, and size, and osp_attr_set() ignores all
1116 * but UID, GID and PROJID. Declaration of size attr setting
1117 * happens through lod_declare_init_size(), and not through
1118 * this function. Therefore we need not load striping unless
1119 * ownership is changing. This should save memory and (we hope)
1120 * speed up rename().
1122 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1123 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID)))
1126 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1129 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID | LA_MODE |
1130 LA_ATIME | LA_MTIME | LA_CTIME |
1135 * load striping information, notice we don't do this when object
1136 * is being initialized as we don't need this information till
1137 * few specific cases like destroy, chown
1139 rc = lod_load_striping(env, lo);
1143 if (!lod_obj_is_striped(dt))
1147 * if object is striped declare changes on the stripes
1149 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1150 LASSERT(lo->ldo_stripe);
1151 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1152 if (lo->ldo_stripe[i] == NULL)
1154 rc = lod_sub_declare_attr_set(env, lo->ldo_stripe[i],
1160 struct lod_obj_stripe_cb_data data;
1162 data.locd_attr = attr;
1163 data.locd_declare = true;
1164 rc = lod_obj_for_each_stripe(env, lo, th,
1165 lod_obj_stripe_attr_set_cb, &data);
1171 if (!dt_object_exists(next) || dt_object_remote(next) ||
1172 !S_ISREG(attr->la_mode))
1175 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1176 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
1180 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) ||
1181 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1182 struct lod_thread_info *info = lod_env_info(env);
1183 struct lu_buf *buf = &info->lti_buf;
1185 buf->lb_buf = info->lti_ea_store;
1186 buf->lb_len = info->lti_ea_store_size;
1187 rc = lod_sub_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1188 LU_XATTR_REPLACE, th);
1195 * Implementation of dt_object_operations::do_attr_set.
1197 * If the object is striped, then apply the changes to all or subset of
1198 * the stripes depending on the object type and specific attributes.
1200 * \see dt_object_operations::do_attr_set() in the API description for details.
1202 static int lod_attr_set(const struct lu_env *env,
1203 struct dt_object *dt,
1204 const struct lu_attr *attr,
1207 struct dt_object *next = dt_object_child(dt);
1208 struct lod_object *lo = lod_dt_obj(dt);
1213 * apply changes to the local object
1215 rc = lod_sub_attr_set(env, next, attr, th);
1219 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1220 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID)))
1223 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1226 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE | LA_PROJID |
1227 LA_ATIME | LA_MTIME | LA_CTIME |
1232 if (!lod_obj_is_striped(dt))
1236 * if object is striped, apply changes to all the stripes
1238 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1239 LASSERT(lo->ldo_stripe);
1240 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1241 if (unlikely(lo->ldo_stripe[i] == NULL))
1244 if ((dt_object_exists(lo->ldo_stripe[i]) == 0))
1247 rc = lod_sub_attr_set(env, lo->ldo_stripe[i], attr, th);
1252 struct lod_obj_stripe_cb_data data;
1254 data.locd_attr = attr;
1255 data.locd_declare = false;
1256 rc = lod_obj_for_each_stripe(env, lo, th,
1257 lod_obj_stripe_attr_set_cb, &data);
1263 if (!dt_object_exists(next) || dt_object_remote(next) ||
1264 !S_ISREG(attr->la_mode))
1267 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1268 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
1272 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE)) {
1273 struct lod_thread_info *info = lod_env_info(env);
1274 struct lu_buf *buf = &info->lti_buf;
1275 struct ost_id *oi = &info->lti_ostid;
1276 struct lu_fid *fid = &info->lti_fid;
1277 struct lov_mds_md_v1 *lmm;
1278 struct lov_ost_data_v1 *objs;
1281 rc = lod_get_lov_ea(env, lo);
1285 buf->lb_buf = info->lti_ea_store;
1286 buf->lb_len = info->lti_ea_store_size;
1287 lmm = info->lti_ea_store;
1288 magic = le32_to_cpu(lmm->lmm_magic);
1289 if (magic == LOV_MAGIC_COMP_V1) {
1290 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1291 struct lov_comp_md_entry_v1 *lcme =
1292 &lcm->lcm_entries[0];
1294 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
1295 magic = le32_to_cpu(lmm->lmm_magic);
1298 if (magic == LOV_MAGIC_V1)
1299 objs = &(lmm->lmm_objects[0]);
1301 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1302 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1303 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1305 fid_to_ostid(fid, oi);
1306 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1308 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1309 LU_XATTR_REPLACE, th);
1310 } else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1311 struct lod_thread_info *info = lod_env_info(env);
1312 struct lu_buf *buf = &info->lti_buf;
1313 struct lov_comp_md_v1 *lcm;
1314 struct lov_comp_md_entry_v1 *lcme;
1316 rc = lod_get_lov_ea(env, lo);
1320 buf->lb_buf = info->lti_ea_store;
1321 buf->lb_len = info->lti_ea_store_size;
1323 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
1326 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1327 lcme = &lcm->lcm_entries[0];
1328 le64_add_cpu(&lcme->lcme_extent.e_start, 1);
1329 le64_add_cpu(&lcme->lcme_extent.e_end, -1);
1331 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1332 LU_XATTR_REPLACE, th);
1339 * Implementation of dt_object_operations::do_xattr_get.
1341 * If LOV EA is requested from the root object and it's not
1342 * found, then return default striping for the filesystem.
1344 * \see dt_object_operations::do_xattr_get() in the API description for details.
1346 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1347 struct lu_buf *buf, const char *name)
1349 struct lod_thread_info *info = lod_env_info(env);
1350 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1355 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1356 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1357 struct lmv_mds_md_v1 *lmv1;
1360 if (rc > (typeof(rc))sizeof(*lmv1))
1363 if (rc < (typeof(rc))sizeof(*lmv1))
1364 RETURN(rc = rc > 0 ? -EINVAL : rc);
1366 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1367 CLASSERT(sizeof(*lmv1) <= sizeof(info->lti_key));
1369 info->lti_buf.lb_buf = info->lti_key;
1370 info->lti_buf.lb_len = sizeof(*lmv1);
1371 rc = dt_xattr_get(env, dt_object_child(dt),
1372 &info->lti_buf, name);
1373 if (unlikely(rc != sizeof(*lmv1)))
1374 RETURN(rc = rc > 0 ? -EINVAL : rc);
1376 lmv1 = info->lti_buf.lb_buf;
1377 /* The on-disk LMV EA only contains header, but the
1378 * returned LMV EA size should contain the space for
1379 * the FIDs of all shards of the striped directory. */
1380 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1381 rc = lmv_mds_md_size(
1382 le32_to_cpu(lmv1->lmv_stripe_count),
1385 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1389 RETURN(rc = rc1 != 0 ? rc1 : rc);
1392 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1396 * XXX: Only used by lfsck
1398 * lod returns default striping on the real root of the device
1399 * this is like the root stores default striping for the whole
1400 * filesystem. historically we've been using a different approach
1401 * and store it in the config.
1403 dt_root_get(env, dev->lod_child, &info->lti_fid);
1404 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1406 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1407 struct lov_user_md *lum = buf->lb_buf;
1408 struct lov_desc *desc = &dev->lod_desc;
1410 if (buf->lb_buf == NULL) {
1412 } else if (buf->lb_len >= sizeof(*lum)) {
1413 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1414 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1415 lmm_oi_set_id(&lum->lmm_oi, 0);
1416 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1417 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1418 lum->lmm_stripe_size = cpu_to_le32(
1419 desc->ld_default_stripe_size);
1420 lum->lmm_stripe_count = cpu_to_le16(
1421 desc->ld_default_stripe_count);
1422 lum->lmm_stripe_offset = cpu_to_le16(
1423 desc->ld_default_stripe_offset);
1436 * Checks that the magic of the stripe is sane.
1438 * \param[in] lod lod device
1439 * \param[in] lum a buffer storing LMV EA to verify
1441 * \retval 0 if the EA is sane
1442 * \retval negative otherwise
1444 static int lod_verify_md_striping(struct lod_device *lod,
1445 const struct lmv_user_md_v1 *lum)
1447 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1448 CERROR("%s: invalid lmv_user_md: magic = %x, "
1449 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1450 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1451 (int)le32_to_cpu(lum->lum_stripe_offset),
1452 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1460 * Initialize LMV EA for a slave.
1462 * Initialize slave's LMV EA from the master's LMV EA.
1464 * \param[in] master_lmv a buffer containing master's EA
1465 * \param[out] slave_lmv a buffer where slave's EA will be stored
1468 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1469 const struct lmv_mds_md_v1 *master_lmv)
1471 *slave_lmv = *master_lmv;
1472 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1478 * Generate LMV EA from the object passed as \a dt. The object must have
1479 * the stripes created and initialized.
1481 * \param[in] env execution environment
1482 * \param[in] dt object
1483 * \param[out] lmv_buf buffer storing generated LMV EA
1485 * \retval 0 on success
1486 * \retval negative if failed
1488 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1489 struct lu_buf *lmv_buf)
1491 struct lod_thread_info *info = lod_env_info(env);
1492 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1493 struct lod_object *lo = lod_dt_obj(dt);
1494 struct lmv_mds_md_v1 *lmm1;
1496 int type = LU_SEQ_RANGE_ANY;
1501 LASSERT(lo->ldo_dir_striped != 0);
1502 LASSERT(lo->ldo_dir_stripe_count > 0);
1503 stripe_count = lo->ldo_dir_stripe_count;
1504 /* Only store the LMV EA heahder on the disk. */
1505 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1506 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1510 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1513 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1514 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1515 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1516 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1517 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1522 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1523 lmv_buf->lb_buf = info->lti_ea_store;
1524 lmv_buf->lb_len = sizeof(*lmm1);
1530 * Create in-core represenation for a striped directory.
1532 * Parse the buffer containing LMV EA and instantiate LU objects
1533 * representing the stripe objects. The pointers to the objects are
1534 * stored in ldo_stripe field of \a lo. This function is used when
1535 * we need to access an already created object (i.e. load from a disk).
1537 * \param[in] env execution environment
1538 * \param[in] lo lod object
1539 * \param[in] buf buffer containing LMV EA
1541 * \retval 0 on success
1542 * \retval negative if failed
1544 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1545 const struct lu_buf *buf)
1547 struct lod_thread_info *info = lod_env_info(env);
1548 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1549 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1550 struct dt_object **stripe;
1551 union lmv_mds_md *lmm = buf->lb_buf;
1552 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1553 struct lu_fid *fid = &info->lti_fid;
1558 if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
1561 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1562 lo->ldo_dir_slave_stripe = 1;
1566 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1569 if (le32_to_cpu(lmv1->lmv_stripe_count) < 1)
1572 LASSERT(lo->ldo_stripe == NULL);
1573 OBD_ALLOC(stripe, sizeof(stripe[0]) *
1574 (le32_to_cpu(lmv1->lmv_stripe_count)));
1578 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1579 struct dt_device *tgt_dt;
1580 struct dt_object *dto;
1581 int type = LU_SEQ_RANGE_ANY;
1584 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1585 if (!fid_is_sane(fid))
1586 GOTO(out, rc = -ESTALE);
1588 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1592 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1593 tgt_dt = lod->lod_child;
1595 struct lod_tgt_desc *tgt;
1597 tgt = LTD_TGT(ltd, idx);
1599 GOTO(out, rc = -ESTALE);
1600 tgt_dt = tgt->ltd_tgt;
1603 dto = dt_locate_at(env, tgt_dt, fid,
1604 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1607 GOTO(out, rc = PTR_ERR(dto));
1612 lo->ldo_stripe = stripe;
1613 lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1614 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1616 lod_object_free_striping(env, lo);
1622 * Declare create a striped directory.
1624 * Declare creating a striped directory with a given stripe pattern on the
1625 * specified MDTs. A striped directory is represented as a regular directory
1626 * - an index listing all the stripes. The stripes point back to the master
1627 * object with ".." and LinkEA. The master object gets LMV EA which
1628 * identifies it as a striped directory. The function allocates FIDs
1631 * \param[in] env execution environment
1632 * \param[in] dt object
1633 * \param[in] attr attributes to initialize the objects with
1634 * \param[in] dof type of objects to be created
1635 * \param[in] th transaction handle
1637 * \retval 0 on success
1638 * \retval negative if failed
1640 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1641 struct dt_object *dt,
1642 struct lu_attr *attr,
1643 struct dt_object_format *dof,
1646 struct lod_thread_info *info = lod_env_info(env);
1647 struct lu_buf lmv_buf;
1648 struct lu_buf slave_lmv_buf;
1649 struct lmv_mds_md_v1 *lmm;
1650 struct lmv_mds_md_v1 *slave_lmm = NULL;
1651 struct dt_insert_rec *rec = &info->lti_dt_rec;
1652 struct lod_object *lo = lod_dt_obj(dt);
1657 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1660 lmm = lmv_buf.lb_buf;
1662 OBD_ALLOC_PTR(slave_lmm);
1663 if (slave_lmm == NULL)
1664 GOTO(out, rc = -ENOMEM);
1666 lod_prep_slave_lmv_md(slave_lmm, lmm);
1667 slave_lmv_buf.lb_buf = slave_lmm;
1668 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1670 if (!dt_try_as_dir(env, dt_object_child(dt)))
1671 GOTO(out, rc = -EINVAL);
1673 rec->rec_type = S_IFDIR;
1674 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1675 struct dt_object *dto = lo->ldo_stripe[i];
1676 char *stripe_name = info->lti_key;
1677 struct lu_name *sname;
1678 struct linkea_data ldata = { NULL };
1679 struct lu_buf linkea_buf;
1681 rc = lod_sub_declare_create(env, dto, attr, NULL, dof, th);
1685 if (!dt_try_as_dir(env, dto))
1686 GOTO(out, rc = -EINVAL);
1688 rc = lod_sub_declare_ref_add(env, dto, th);
1692 rec->rec_fid = lu_object_fid(&dto->do_lu);
1693 rc = lod_sub_declare_insert(env, dto,
1694 (const struct dt_rec *)rec,
1695 (const struct dt_key *)dot, th);
1699 /* master stripe FID will be put to .. */
1700 rec->rec_fid = lu_object_fid(&dt->do_lu);
1701 rc = lod_sub_declare_insert(env, dto,
1702 (const struct dt_rec *)rec,
1703 (const struct dt_key *)dotdot, th);
1707 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1708 cfs_fail_val != i) {
1709 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1711 slave_lmm->lmv_master_mdt_index =
1714 slave_lmm->lmv_master_mdt_index =
1716 rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
1717 XATTR_NAME_LMV, 0, th);
1722 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1724 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1725 PFID(lu_object_fid(&dto->do_lu)), i + 1);
1727 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1728 PFID(lu_object_fid(&dto->do_lu)), i);
1730 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
1731 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1732 sname, lu_object_fid(&dt->do_lu));
1736 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1737 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1738 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
1739 XATTR_NAME_LINK, 0, th);
1743 rec->rec_fid = lu_object_fid(&dto->do_lu);
1744 rc = lod_sub_declare_insert(env, dt_object_child(dt),
1745 (const struct dt_rec *)rec,
1746 (const struct dt_key *)stripe_name,
1751 rc = lod_sub_declare_ref_add(env, dt_object_child(dt), th);
1756 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
1757 &lmv_buf, XATTR_NAME_LMV, 0, th);
1761 if (slave_lmm != NULL)
1762 OBD_FREE_PTR(slave_lmm);
1767 static int lod_prep_md_striped_create(const struct lu_env *env,
1768 struct dt_object *dt,
1769 struct lu_attr *attr,
1770 const struct lmv_user_md_v1 *lum,
1771 struct dt_object_format *dof,
1774 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1775 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1776 struct lod_object *lo = lod_dt_obj(dt);
1777 struct dt_object **stripe;
1786 /* The lum has been verifed in lod_verify_md_striping */
1787 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
1788 LASSERT(le32_to_cpu(lum->lum_stripe_count) > 0);
1790 stripe_count = le32_to_cpu(lum->lum_stripe_count);
1792 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
1793 if (idx_array == NULL)
1796 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_count);
1798 GOTO(out_free, rc = -ENOMEM);
1800 /* Start index must be the master MDT */
1801 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1802 idx_array[0] = master_index;
1803 for (i = 0; i < stripe_count; i++) {
1804 struct lod_tgt_desc *tgt = NULL;
1805 struct dt_object *dto;
1806 struct lu_fid fid = { 0 };
1808 struct lu_object_conf conf = { 0 };
1809 struct dt_device *tgt_dt = NULL;
1811 /* Try to find next avaible target */
1813 for (j = 0; j < lod->lod_remote_mdt_count;
1814 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
1815 bool already_allocated = false;
1818 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
1819 idx, lod->lod_remote_mdt_count + 1, i);
1821 if (likely(!OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
1822 /* check whether the idx already exists
1823 * in current allocated array */
1824 for (k = 0; k < i; k++) {
1825 if (idx_array[k] == idx) {
1826 already_allocated = true;
1831 if (already_allocated)
1835 /* Sigh, this index is not in the bitmap, let's check
1836 * next available target */
1837 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) &&
1838 idx != master_index)
1841 if (idx == master_index) {
1842 /* Allocate the FID locally */
1843 rc = obd_fid_alloc(env, lod->lod_child_exp,
1847 tgt_dt = lod->lod_child;
1851 /* check the status of the OSP */
1852 tgt = LTD_TGT(ltd, idx);
1856 tgt_dt = tgt->ltd_tgt;
1857 rc = dt_statfs(env, tgt_dt, NULL);
1859 /* this OSP doesn't feel well */
1864 rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
1873 /* Can not allocate more stripes */
1874 if (j == lod->lod_remote_mdt_count) {
1875 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
1876 lod2obd(lod)->obd_name, stripe_count, i);
1880 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
1881 idx, i, PFID(&fid));
1883 /* Set the start index for next stripe allocation */
1884 if (i < stripe_count - 1)
1885 idx_array[i + 1] = (idx + 1) %
1886 (lod->lod_remote_mdt_count + 1);
1887 /* tgt_dt and fid must be ready after search avaible OSP
1888 * in the above loop */
1889 LASSERT(tgt_dt != NULL);
1890 LASSERT(fid_is_sane(&fid));
1891 conf.loc_flags = LOC_F_NEW;
1892 dto = dt_locate_at(env, tgt_dt, &fid,
1893 dt->do_lu.lo_dev->ld_site->ls_top_dev,
1896 GOTO(out_put, rc = PTR_ERR(dto));
1900 lo->ldo_dir_stripe_loaded = 1;
1901 lo->ldo_dir_striped = 1;
1902 lo->ldo_stripe = stripe;
1903 lo->ldo_dir_stripe_count = i;
1904 lo->ldo_dir_stripes_allocated = stripe_count;
1906 if (lo->ldo_dir_stripe_count == 0)
1907 GOTO(out_put, rc = -ENOSPC);
1909 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
1915 for (i = 0; i < stripe_count; i++)
1916 if (stripe[i] != NULL)
1917 dt_object_put(env, stripe[i]);
1918 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_count);
1919 lo->ldo_dir_stripe_count = 0;
1920 lo->ldo_dir_stripes_allocated = 0;
1921 lo->ldo_stripe = NULL;
1925 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
1931 * Declare create striped md object.
1933 * The function declares intention to create a striped directory. This is a
1934 * wrapper for lod_prep_md_striped_create(). The only additional functionality
1935 * is to verify pattern \a lum_buf is good. Check that function for the details.
1937 * \param[in] env execution environment
1938 * \param[in] dt object
1939 * \param[in] attr attributes to initialize the objects with
1940 * \param[in] lum_buf a pattern specifying the number of stripes and
1942 * \param[in] dof type of objects to be created
1943 * \param[in] th transaction handle
1945 * \retval 0 on success
1946 * \retval negative if failed
1949 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
1950 struct dt_object *dt,
1951 struct lu_attr *attr,
1952 const struct lu_buf *lum_buf,
1953 struct dt_object_format *dof,
1956 struct lod_object *lo = lod_dt_obj(dt);
1957 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1958 struct lmv_user_md_v1 *lum;
1962 lum = lum_buf->lb_buf;
1963 LASSERT(lum != NULL);
1965 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
1966 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
1967 (int)le32_to_cpu(lum->lum_stripe_offset));
1969 if (le32_to_cpu(lum->lum_stripe_count) == 0)
1972 rc = lod_verify_md_striping(lod, lum);
1976 /* prepare dir striped objects */
1977 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
1979 /* failed to create striping, let's reset
1980 * config so that others don't get confused */
1981 lod_object_free_striping(env, lo);
1989 * Implementation of dt_object_operations::do_declare_xattr_set.
1991 * Used with regular (non-striped) objects. Basically it
1992 * initializes the striping information and applies the
1993 * change to all the stripes.
1995 * \see dt_object_operations::do_declare_xattr_set() in the API description
1998 static int lod_dir_declare_xattr_set(const struct lu_env *env,
1999 struct dt_object *dt,
2000 const struct lu_buf *buf,
2001 const char *name, int fl,
2004 struct dt_object *next = dt_object_child(dt);
2005 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2006 struct lod_object *lo = lod_dt_obj(dt);
2011 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2012 struct lmv_user_md_v1 *lum;
2014 LASSERT(buf != NULL && buf->lb_buf != NULL);
2016 rc = lod_verify_md_striping(d, lum);
2019 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2020 rc = lod_verify_striping(d, buf, false, 0);
2025 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2029 /* Note: Do not set LinkEA on sub-stripes, otherwise
2030 * it will confuse the fid2path process(see mdt_path_current()).
2031 * The linkEA between master and sub-stripes is set in
2032 * lod_xattr_set_lmv(). */
2033 if (strcmp(name, XATTR_NAME_LINK) == 0)
2036 /* set xattr to each stripes, if needed */
2037 rc = lod_load_striping(env, lo);
2041 if (lo->ldo_dir_stripe_count == 0)
2044 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2045 LASSERT(lo->ldo_stripe[i]);
2047 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2057 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2058 struct lod_object *lo,
2059 struct dt_object *dt, struct thandle *th,
2061 struct lod_obj_stripe_cb_data *data)
2063 struct lod_thread_info *info = lod_env_info(env);
2064 struct filter_fid *ff = &info->lti_ff;
2065 struct lu_buf *buf = &info->lti_buf;
2069 buf->lb_len = sizeof(*ff);
2070 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2077 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2078 ff->ff_parent.f_ver = stripe_idx;
2079 fid_cpu_to_le(&ff->ff_parent, &ff->ff_parent);
2080 if (data->locd_declare)
2081 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2082 LU_XATTR_REPLACE, th);
2084 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2085 LU_XATTR_REPLACE, th);
2091 * Reset parent FID on OST object
2093 * Replace parent FID with @dt object FID, which is only called during migration
2094 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2095 * the FID is changed.
2097 * \param[in] env execution environment
2098 * \param[in] dt dt_object whose stripes's parent FID will be reset
2099 * \parem[in] th thandle
2100 * \param[in] declare if it is declare
2102 * \retval 0 if reset succeeds
2103 * \retval negative errno if reset fails
2105 static int lod_replace_parent_fid(const struct lu_env *env,
2106 struct dt_object *dt,
2107 struct thandle *th, bool declare)
2109 struct lod_object *lo = lod_dt_obj(dt);
2110 struct lod_thread_info *info = lod_env_info(env);
2111 struct lu_buf *buf = &info->lti_buf;
2112 struct filter_fid *ff;
2113 struct lod_obj_stripe_cb_data data;
2117 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2119 /* set xattr to each stripes, if needed */
2120 rc = lod_load_striping(env, lo);
2124 if (!lod_obj_is_striped(dt))
2127 if (info->lti_ea_store_size < sizeof(*ff)) {
2128 rc = lod_ea_store_resize(info, sizeof(*ff));
2133 buf->lb_buf = info->lti_ea_store;
2134 buf->lb_len = info->lti_ea_store_size;
2136 data.locd_declare = declare;
2137 rc = lod_obj_for_each_stripe(env, lo, th,
2138 lod_obj_stripe_replace_parent_fid_cb,
2144 inline __u16 lod_comp_entry_stripe_count(struct lod_object *lo,
2145 struct lod_layout_component *entry,
2148 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2152 else if (lod_comp_inited(entry))
2153 return entry->llc_stripe_count;
2154 else if ((__u16)-1 == entry->llc_stripe_count)
2155 return lod->lod_desc.ld_tgt_count;
2157 return lod_get_stripe_count(lod, lo, entry->llc_stripe_count);
2160 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2162 int magic, size = 0, i;
2163 struct lod_layout_component *comp_entries;
2168 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2169 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2171 lo->ldo_def_striping->lds_def_striping_is_composite;
2173 comp_cnt = lo->ldo_comp_cnt;
2174 comp_entries = lo->ldo_comp_entries;
2175 is_composite = lo->ldo_is_composite;
2179 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2181 size = sizeof(struct lov_comp_md_v1) +
2182 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2183 LASSERT(size % sizeof(__u64) == 0);
2186 for (i = 0; i < comp_cnt; i++) {
2189 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2190 stripe_count = lod_comp_entry_stripe_count(lo, &comp_entries[i],
2192 if (!is_dir && is_composite)
2193 lod_comp_shrink_stripe_count(&comp_entries[i],
2196 size += lov_user_md_size(stripe_count, magic);
2197 LASSERT(size % sizeof(__u64) == 0);
2203 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2204 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2207 * \param[in] env execution environment
2208 * \param[in] dt dt_object to add components on
2209 * \param[in] buf buffer contains components to be added
2210 * \parem[in] th thandle
2212 * \retval 0 on success
2213 * \retval negative errno on failure
2215 static int lod_declare_layout_add(const struct lu_env *env,
2216 struct dt_object *dt,
2217 const struct lu_buf *buf,
2220 struct lod_thread_info *info = lod_env_info(env);
2221 struct lod_layout_component *comp_array, *lod_comp;
2222 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2223 struct dt_object *next = dt_object_child(dt);
2224 struct lov_desc *desc = &d->lod_desc;
2225 struct lod_object *lo = lod_dt_obj(dt);
2226 struct lov_user_md_v3 *v3;
2227 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2230 int i, rc, array_cnt;
2233 LASSERT(lo->ldo_is_composite);
2235 prev_end = lo->ldo_comp_entries[lo->ldo_comp_cnt - 1].llc_extent.e_end;
2236 rc = lod_verify_striping(d, buf, false, prev_end);
2240 magic = comp_v1->lcm_magic;
2241 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2242 lustre_swab_lov_comp_md_v1(comp_v1);
2243 magic = comp_v1->lcm_magic;
2246 if (magic != LOV_USER_MAGIC_COMP_V1)
2249 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2250 OBD_ALLOC(comp_array, sizeof(*comp_array) * array_cnt);
2251 if (comp_array == NULL)
2254 memcpy(comp_array, lo->ldo_comp_entries,
2255 sizeof(*comp_array) * lo->ldo_comp_cnt);
2257 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2258 struct lov_user_md_v1 *v1;
2259 struct lu_extent *ext;
2261 v1 = (struct lov_user_md *)((char *)comp_v1 +
2262 comp_v1->lcm_entries[i].lcme_offset);
2263 ext = &comp_v1->lcm_entries[i].lcme_extent;
2265 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2266 lod_comp->llc_extent.e_start = ext->e_start;
2267 lod_comp->llc_extent.e_end = ext->e_end;
2268 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2270 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2271 if (!lod_comp->llc_stripe_count ||
2272 lod_comp->llc_stripe_count == (__u16)-1)
2273 lod_comp->llc_stripe_count =
2274 desc->ld_default_stripe_count;
2275 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2276 if (!lod_comp->llc_stripe_size)
2277 lod_comp->llc_stripe_size =
2278 desc->ld_default_stripe_size;
2280 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2281 v3 = (struct lov_user_md_v3 *) v1;
2282 if (v3->lmm_pool_name[0] != '\0') {
2283 rc = lod_set_pool(&lod_comp->llc_pool,
2291 OBD_FREE(lo->ldo_comp_entries, sizeof(*lod_comp) * lo->ldo_comp_cnt);
2292 lo->ldo_comp_entries = comp_array;
2293 lo->ldo_comp_cnt = array_cnt;
2294 /* No need to increase layout generation here, it will be increased
2295 * later when generating component ID for the new components */
2297 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2298 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2299 XATTR_NAME_LOV, 0, th);
2306 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2307 lod_comp = &comp_array[i];
2308 if (lod_comp->llc_pool != NULL) {
2309 OBD_FREE(lod_comp->llc_pool,
2310 strlen(lod_comp->llc_pool) + 1);
2311 lod_comp->llc_pool = NULL;
2314 OBD_FREE(comp_array, sizeof(*comp_array) * array_cnt);
2319 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2320 * the '$field' can only be 'flags' now. The xattr value is binary
2321 * lov_comp_md_v1 which contains the component ID(s) and the value of
2322 * the field to be modified.
2324 * \param[in] env execution environment
2325 * \param[in] dt dt_object to be modified
2326 * \param[in] op operation string, like "set.flags"
2327 * \param[in] buf buffer contains components to be set
2328 * \parem[in] th thandle
2330 * \retval 0 on success
2331 * \retval negative errno on failure
2333 static int lod_declare_layout_set(const struct lu_env *env,
2334 struct dt_object *dt,
2335 char *op, const struct lu_buf *buf,
2338 struct lod_layout_component *lod_comp;
2339 struct lod_thread_info *info = lod_env_info(env);
2340 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2341 struct lod_object *lo = lod_dt_obj(dt);
2342 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2345 bool changed = false;
2348 if (strcmp(op, "set.flags") != 0) {
2349 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2350 lod2obd(d)->obd_name, op);
2354 magic = comp_v1->lcm_magic;
2355 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2356 lustre_swab_lov_comp_md_v1(comp_v1);
2357 magic = comp_v1->lcm_magic;
2360 if (magic != LOV_USER_MAGIC_COMP_V1)
2363 if (comp_v1->lcm_entry_count == 0) {
2364 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2365 lod2obd(d)->obd_name);
2369 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2370 id = comp_v1->lcm_entries[i].lcme_id;
2372 for (j = 0; j < lo->ldo_comp_cnt; j++) {
2373 lod_comp = &lo->ldo_comp_entries[j];
2374 if (id == lod_comp->llc_id || id == LCME_ID_ALL) {
2375 lod_comp->llc_flags =
2376 comp_v1->lcm_entries[i].lcme_flags;
2383 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
2384 lod2obd(d)->obd_name);
2388 lod_obj_inc_layout_gen(lo);
2390 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2391 rc = lod_sub_declare_xattr_set(env, dt, &info->lti_buf,
2392 XATTR_NAME_LOV, 0, th);
2397 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
2398 * and the xattr value is a unique component ID or a special lcme_id.
2400 * \param[in] env execution environment
2401 * \param[in] dt dt_object to be operated on
2402 * \param[in] buf buffer contains component ID or lcme_id
2403 * \parem[in] th thandle
2405 * \retval 0 on success
2406 * \retval negative errno on failure
2408 static int lod_declare_layout_del(const struct lu_env *env,
2409 struct dt_object *dt,
2410 const struct lu_buf *buf,
2413 struct lod_thread_info *info = lod_env_info(env);
2414 struct dt_object *next = dt_object_child(dt);
2415 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2416 struct lod_object *lo = lod_dt_obj(dt);
2417 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2418 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2419 __u32 magic, id, flags, neg_flags = 0;
2423 LASSERT(lo->ldo_is_composite);
2425 rc = lod_verify_striping(d, buf, false, 0);
2429 magic = comp_v1->lcm_magic;
2430 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2431 lustre_swab_lov_comp_md_v1(comp_v1);
2432 magic = comp_v1->lcm_magic;
2435 if (magic != LOV_USER_MAGIC_COMP_V1)
2438 id = comp_v1->lcm_entries[0].lcme_id;
2439 flags = comp_v1->lcm_entries[0].lcme_flags;
2441 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
2442 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
2443 lod2obd(d)->obd_name, id, flags);
2447 if (id != LCME_ID_INVAL && flags != 0) {
2448 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
2449 lod2obd(d)->obd_name);
2453 if (flags & LCME_FL_NEG) {
2454 neg_flags = flags & ~LCME_FL_NEG;
2458 left = lo->ldo_comp_cnt;
2462 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
2463 struct lod_layout_component *lod_comp;
2465 lod_comp = &lo->ldo_comp_entries[i];
2467 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
2469 else if (flags && !(flags & lod_comp->llc_flags))
2471 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
2474 if (left != (i + 1)) {
2475 CDEBUG(D_LAYOUT, "%s: this deletion will create "
2476 "a hole.\n", lod2obd(d)->obd_name);
2481 /* Mark the component as deleted */
2482 lod_comp->llc_id = LCME_ID_INVAL;
2484 /* Not instantiated component */
2485 if (lod_comp->llc_stripe == NULL)
2488 LASSERT(lod_comp->llc_stripe_count > 0);
2489 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
2490 struct dt_object *obj = lod_comp->llc_stripe[j];
2494 rc = lod_sub_declare_destroy(env, obj, th);
2500 LASSERTF(left >= 0, "left = %d\n", left);
2501 if (left == lo->ldo_comp_cnt) {
2502 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
2503 lod2obd(d)->obd_name, id);
2507 memset(attr, 0, sizeof(*attr));
2508 attr->la_valid = LA_SIZE;
2509 rc = lod_sub_declare_attr_set(env, next, attr, th);
2514 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2515 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2516 XATTR_NAME_LOV, 0, th);
2518 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
2525 * Declare layout add/set/del operations issued by special xattr names:
2527 * XATTR_LUSTRE_LOV.add add component(s) to existing file
2528 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
2529 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
2531 * \param[in] env execution environment
2532 * \param[in] dt object
2533 * \param[in] name name of xattr
2534 * \param[in] buf lu_buf contains xattr value
2535 * \param[in] th transaction handle
2537 * \retval 0 on success
2538 * \retval negative if failed
2540 static int lod_declare_modify_layout(const struct lu_env *env,
2541 struct dt_object *dt,
2543 const struct lu_buf *buf,
2546 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2547 struct lod_object *lo = lod_dt_obj(dt);
2548 struct dt_object *next = dt_object_child(&lo->ldo_obj);
2550 int rc, len = strlen(XATTR_LUSTRE_LOV);
2553 LASSERT(dt_object_exists(dt));
2555 if (strlen(name) <= len || name[len] != '.') {
2556 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
2557 lod2obd(d)->obd_name, name);
2562 dt_write_lock(env, next, 0);
2563 rc = lod_load_striping_locked(env, lo);
2567 /* the layout to be modified must be a composite layout */
2568 if (!lo->ldo_is_composite) {
2569 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
2570 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
2571 GOTO(unlock, rc = -EINVAL);
2574 op = (char *)name + len;
2575 if (strcmp(op, "add") == 0) {
2576 rc = lod_declare_layout_add(env, dt, buf, th);
2577 } else if (strcmp(op, "del") == 0) {
2578 rc = lod_declare_layout_del(env, dt, buf, th);
2579 } else if (strncmp(op, "set", strlen("set")) == 0) {
2580 rc = lod_declare_layout_set(env, dt, op, buf, th);
2582 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
2583 lod2obd(d)->obd_name, name);
2584 GOTO(unlock, rc = -ENOTSUPP);
2588 lod_object_free_striping(env, lo);
2589 dt_write_unlock(env, next);
2595 * Implementation of dt_object_operations::do_declare_xattr_set.
2597 * \see dt_object_operations::do_declare_xattr_set() in the API description
2600 * the extension to the API:
2601 * - declaring LOVEA requests striping creation
2602 * - LU_XATTR_REPLACE means layout swap
2604 static int lod_declare_xattr_set(const struct lu_env *env,
2605 struct dt_object *dt,
2606 const struct lu_buf *buf,
2607 const char *name, int fl,
2610 struct dt_object *next = dt_object_child(dt);
2611 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2616 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
2617 if ((S_ISREG(mode) || mode == 0) && !(fl & LU_XATTR_REPLACE) &&
2618 (strcmp(name, XATTR_NAME_LOV) == 0 ||
2619 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
2621 * this is a request to create object's striping.
2623 * allow to declare predefined striping on a new (!mode) object
2624 * which is supposed to be replay of regular file creation
2625 * (when LOV setting is declared)
2627 * LU_XATTR_REPLACE is set to indicate a layout swap
2629 if (dt_object_exists(dt)) {
2630 rc = dt_attr_get(env, next, attr);
2634 memset(attr, 0, sizeof(*attr));
2635 attr->la_valid = LA_TYPE | LA_MODE;
2636 attr->la_mode = S_IFREG;
2638 rc = lod_declare_striped_create(env, dt, attr, buf, th);
2639 } else if (S_ISREG(mode) &&
2640 strlen(name) > strlen(XATTR_LUSTRE_LOV) + 1 &&
2641 strncmp(name, XATTR_LUSTRE_LOV,
2642 strlen(XATTR_LUSTRE_LOV)) == 0) {
2644 * this is a request to modify object's striping.
2645 * add/set/del component(s).
2647 if (!dt_object_exists(dt))
2650 rc = lod_declare_modify_layout(env, dt, name, buf, th);
2651 } else if (S_ISDIR(mode)) {
2652 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
2653 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
2654 rc = lod_replace_parent_fid(env, dt, th, true);
2656 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2663 * Apply xattr changes to the object.
2665 * Applies xattr changes to the object and the stripes if the latter exist.
2667 * \param[in] env execution environment
2668 * \param[in] dt object
2669 * \param[in] buf buffer pointing to the new value of xattr
2670 * \param[in] name name of xattr
2671 * \param[in] fl flags
2672 * \param[in] th transaction handle
2674 * \retval 0 on success
2675 * \retval negative if failed
2677 static int lod_xattr_set_internal(const struct lu_env *env,
2678 struct dt_object *dt,
2679 const struct lu_buf *buf,
2680 const char *name, int fl,
2683 struct dt_object *next = dt_object_child(dt);
2684 struct lod_object *lo = lod_dt_obj(dt);
2689 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
2690 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2693 /* Note: Do not set LinkEA on sub-stripes, otherwise
2694 * it will confuse the fid2path process(see mdt_path_current()).
2695 * The linkEA between master and sub-stripes is set in
2696 * lod_xattr_set_lmv(). */
2697 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
2700 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2701 LASSERT(lo->ldo_stripe[i]);
2703 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
2713 * Delete an extended attribute.
2715 * Deletes specified xattr from the object and the stripes if the latter exist.
2717 * \param[in] env execution environment
2718 * \param[in] dt object
2719 * \param[in] name name of xattr
2720 * \param[in] th transaction handle
2722 * \retval 0 on success
2723 * \retval negative if failed
2725 static int lod_xattr_del_internal(const struct lu_env *env,
2726 struct dt_object *dt,
2727 const char *name, struct thandle *th)
2729 struct dt_object *next = dt_object_child(dt);
2730 struct lod_object *lo = lod_dt_obj(dt);
2735 rc = lod_sub_xattr_del(env, next, name, th);
2736 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2739 if (lo->ldo_dir_stripe_count == 0)
2742 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2743 LASSERT(lo->ldo_stripe[i]);
2745 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
2754 * Set default striping on a directory.
2756 * Sets specified striping on a directory object unless it matches the default
2757 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
2758 * EA. This striping will be used when regular file is being created in this
2761 * \param[in] env execution environment
2762 * \param[in] dt the striped object
2763 * \param[in] buf buffer with the striping
2764 * \param[in] name name of EA
2765 * \param[in] fl xattr flag (see OSD API description)
2766 * \param[in] th transaction handle
2768 * \retval 0 on success
2769 * \retval negative if failed
2771 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
2772 struct dt_object *dt,
2773 const struct lu_buf *buf,
2774 const char *name, int fl,
2777 struct lov_user_md_v1 *lum;
2778 struct lov_user_md_v3 *v3 = NULL;
2779 const char *pool_name = NULL;
2784 LASSERT(buf != NULL && buf->lb_buf != NULL);
2787 switch (lum->lmm_magic) {
2788 case LOV_USER_MAGIC_V3:
2790 if (v3->lmm_pool_name[0] != '\0')
2791 pool_name = v3->lmm_pool_name;
2793 case LOV_USER_MAGIC_V1:
2794 /* if { size, offset, count } = { 0, -1, 0 } and no pool
2795 * (i.e. all default values specified) then delete default
2796 * striping from dir. */
2798 "set default striping: sz %u # %u offset %d %s %s\n",
2799 (unsigned)lum->lmm_stripe_size,
2800 (unsigned)lum->lmm_stripe_count,
2801 (int)lum->lmm_stripe_offset,
2802 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
2804 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
2805 lum->lmm_stripe_count,
2806 lum->lmm_stripe_offset,
2809 case LOV_USER_MAGIC_COMP_V1:
2813 CERROR("Invalid magic %x\n", lum->lmm_magic);
2818 rc = lod_xattr_del_internal(env, dt, name, th);
2822 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
2829 * Set default striping on a directory object.
2831 * Sets specified striping on a directory object unless it matches the default
2832 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
2833 * EA. This striping will be used when a new directory is being created in the
2836 * \param[in] env execution environment
2837 * \param[in] dt the striped object
2838 * \param[in] buf buffer with the striping
2839 * \param[in] name name of EA
2840 * \param[in] fl xattr flag (see OSD API description)
2841 * \param[in] th transaction handle
2843 * \retval 0 on success
2844 * \retval negative if failed
2846 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
2847 struct dt_object *dt,
2848 const struct lu_buf *buf,
2849 const char *name, int fl,
2852 struct lmv_user_md_v1 *lum;
2856 LASSERT(buf != NULL && buf->lb_buf != NULL);
2859 CDEBUG(D_OTHER, "set default stripe_count # %u stripe_offset %d\n",
2860 le32_to_cpu(lum->lum_stripe_count),
2861 (int)le32_to_cpu(lum->lum_stripe_offset));
2863 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
2864 le32_to_cpu(lum->lum_stripe_offset)) &&
2865 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
2866 rc = lod_xattr_del_internal(env, dt, name, th);
2870 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
2879 * Turn directory into a striped directory.
2881 * During replay the client sends the striping created before MDT
2882 * failure, then the layer above LOD sends this defined striping
2883 * using ->do_xattr_set(), so LOD uses this method to replay creation
2884 * of the stripes. Notice the original information for the striping
2885 * (#stripes, FIDs, etc) was transferred in declare path.
2887 * \param[in] env execution environment
2888 * \param[in] dt the striped object
2889 * \param[in] buf not used currently
2890 * \param[in] name not used currently
2891 * \param[in] fl xattr flag (see OSD API description)
2892 * \param[in] th transaction handle
2894 * \retval 0 on success
2895 * \retval negative if failed
2897 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
2898 const struct lu_buf *buf, const char *name,
2899 int fl, struct thandle *th)
2901 struct lod_object *lo = lod_dt_obj(dt);
2902 struct lod_thread_info *info = lod_env_info(env);
2903 struct lu_attr *attr = &info->lti_attr;
2904 struct dt_object_format *dof = &info->lti_format;
2905 struct lu_buf lmv_buf;
2906 struct lu_buf slave_lmv_buf;
2907 struct lmv_mds_md_v1 *lmm;
2908 struct lmv_mds_md_v1 *slave_lmm = NULL;
2909 struct dt_insert_rec *rec = &info->lti_dt_rec;
2914 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
2917 /* The stripes are supposed to be allocated in declare phase,
2918 * if there are no stripes being allocated, it will skip */
2919 if (lo->ldo_dir_stripe_count == 0)
2922 rc = dt_attr_get(env, dt_object_child(dt), attr);
2926 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME |
2927 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
2928 dof->dof_type = DFT_DIR;
2930 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
2933 lmm = lmv_buf.lb_buf;
2935 OBD_ALLOC_PTR(slave_lmm);
2936 if (slave_lmm == NULL)
2939 lod_prep_slave_lmv_md(slave_lmm, lmm);
2940 slave_lmv_buf.lb_buf = slave_lmm;
2941 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
2943 rec->rec_type = S_IFDIR;
2944 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2945 struct dt_object *dto;
2946 char *stripe_name = info->lti_key;
2947 struct lu_name *sname;
2948 struct linkea_data ldata = { NULL };
2949 struct lu_buf linkea_buf;
2951 dto = lo->ldo_stripe[i];
2953 dt_write_lock(env, dto, MOR_TGT_CHILD);
2954 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
2956 dt_write_unlock(env, dto);
2960 rc = lod_sub_ref_add(env, dto, th);
2961 dt_write_unlock(env, dto);
2965 rec->rec_fid = lu_object_fid(&dto->do_lu);
2966 rc = lod_sub_insert(env, dto, (const struct dt_rec *)rec,
2967 (const struct dt_key *)dot, th, 0);
2971 rec->rec_fid = lu_object_fid(&dt->do_lu);
2972 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
2973 (const struct dt_key *)dotdot, th, 0);
2977 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
2978 cfs_fail_val != i) {
2979 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
2981 slave_lmm->lmv_master_mdt_index =
2984 slave_lmm->lmv_master_mdt_index =
2987 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
2988 XATTR_NAME_LMV, fl, th);
2993 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
2995 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2996 PFID(lu_object_fid(&dto->do_lu)), i + 1);
2998 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2999 PFID(lu_object_fid(&dto->do_lu)), i);
3001 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
3002 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
3003 sname, lu_object_fid(&dt->do_lu));
3007 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
3008 linkea_buf.lb_len = ldata.ld_leh->leh_len;
3009 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
3010 XATTR_NAME_LINK, 0, th);
3014 rec->rec_fid = lu_object_fid(&dto->do_lu);
3015 rc = lod_sub_insert(env, dt_object_child(dt),
3016 (const struct dt_rec *)rec,
3017 (const struct dt_key *)stripe_name, th, 0);
3021 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
3026 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
3027 rc = lod_sub_xattr_set(env, dt_object_child(dt),
3028 &lmv_buf, XATTR_NAME_LMV, fl, th);
3030 if (slave_lmm != NULL)
3031 OBD_FREE_PTR(slave_lmm);
3037 * Helper function to declare/execute creation of a striped directory
3039 * Called in declare/create object path, prepare striping for a directory
3040 * and prepare defaults data striping for the objects to be created in
3041 * that directory. Notice the function calls "declaration" or "execution"
3042 * methods depending on \a declare param. This is a consequence of the
3043 * current approach while we don't have natural distributed transactions:
3044 * we basically execute non-local updates in the declare phase. So, the
3045 * arguments for the both phases are the same and this is the reason for
3046 * this function to exist.
3048 * \param[in] env execution environment
3049 * \param[in] dt object
3050 * \param[in] attr attributes the stripes will be created with
3051 * \param[in] dof format of stripes (see OSD API description)
3052 * \param[in] th transaction handle
3053 * \param[in] declare where to call "declare" or "execute" methods
3055 * \retval 0 on success
3056 * \retval negative if failed
3058 static int lod_dir_striping_create_internal(const struct lu_env *env,
3059 struct dt_object *dt,
3060 struct lu_attr *attr,
3061 struct dt_object_format *dof,
3065 struct lod_thread_info *info = lod_env_info(env);
3066 struct lod_object *lo = lod_dt_obj(dt);
3067 const struct lod_default_striping *lds = lo->ldo_def_striping;
3071 LASSERT(ergo(lds != NULL,
3072 lds->lds_def_striping_set ||
3073 lds->lds_dir_def_striping_set));
3075 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
3076 lo->ldo_dir_stripe_offset)) {
3077 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3078 int stripe_count = lo->ldo_dir_stripe_count;
3080 if (info->lti_ea_store_size < sizeof(*v1)) {
3081 rc = lod_ea_store_resize(info, sizeof(*v1));
3084 v1 = info->lti_ea_store;
3087 memset(v1, 0, sizeof(*v1));
3088 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3089 v1->lum_stripe_count = cpu_to_le32(stripe_count);
3090 v1->lum_stripe_offset =
3091 cpu_to_le32(lo->ldo_dir_stripe_offset);
3093 info->lti_buf.lb_buf = v1;
3094 info->lti_buf.lb_len = sizeof(*v1);
3097 rc = lod_declare_xattr_set_lmv(env, dt, attr,
3098 &info->lti_buf, dof, th);
3100 rc = lod_xattr_set_lmv(env, dt, &info->lti_buf,
3101 XATTR_NAME_LMV, 0, th);
3106 /* Transfer default LMV striping from the parent */
3107 if (lds != NULL && lds->lds_dir_def_striping_set &&
3108 !LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
3109 lds->lds_dir_def_stripe_offset)) {
3110 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3112 if (info->lti_ea_store_size < sizeof(*v1)) {
3113 rc = lod_ea_store_resize(info, sizeof(*v1));
3116 v1 = info->lti_ea_store;
3119 memset(v1, 0, sizeof(*v1));
3120 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3121 v1->lum_stripe_count =
3122 cpu_to_le32(lds->lds_dir_def_stripe_count);
3123 v1->lum_stripe_offset =
3124 cpu_to_le32(lds->lds_dir_def_stripe_offset);
3126 cpu_to_le32(lds->lds_dir_def_hash_type);
3128 info->lti_buf.lb_buf = v1;
3129 info->lti_buf.lb_len = sizeof(*v1);
3131 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3132 XATTR_NAME_DEFAULT_LMV,
3135 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
3137 XATTR_NAME_DEFAULT_LMV, 0,
3143 /* Transfer default LOV striping from the parent */
3144 if (lds != NULL && lds->lds_def_striping_set &&
3145 lds->lds_def_comp_cnt != 0) {
3146 struct lov_mds_md *lmm;
3147 int lmm_size = lod_comp_md_size(lo, true);
3149 if (info->lti_ea_store_size < lmm_size) {
3150 rc = lod_ea_store_resize(info, lmm_size);
3154 lmm = info->lti_ea_store;
3156 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
3160 info->lti_buf.lb_buf = lmm;
3161 info->lti_buf.lb_len = lmm_size;
3164 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3165 XATTR_NAME_LOV, 0, th);
3167 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
3168 XATTR_NAME_LOV, 0, th);
3176 static int lod_declare_dir_striping_create(const struct lu_env *env,
3177 struct dt_object *dt,
3178 struct lu_attr *attr,
3179 struct dt_object_format *dof,
3182 return lod_dir_striping_create_internal(env, dt, attr, dof, th, true);
3185 static int lod_dir_striping_create(const struct lu_env *env,
3186 struct dt_object *dt,
3187 struct lu_attr *attr,
3188 struct dt_object_format *dof,
3191 return lod_dir_striping_create_internal(env, dt, attr, dof, th, false);
3195 * Make LOV EA for striped object.
3197 * Generate striping information and store it in the LOV EA of the given
3198 * object. The caller must ensure nobody else is calling the function
3199 * against the object concurrently. The transaction must be started.
3200 * FLDB service must be running as well; it's used to map FID to the target,
3201 * which is stored in LOV EA.
3203 * \param[in] env execution environment for this thread
3204 * \param[in] lo LOD object
3205 * \param[in] th transaction handle
3207 * \retval 0 if LOV EA is stored successfully
3208 * \retval negative error number on failure
3210 static int lod_generate_and_set_lovea(const struct lu_env *env,
3211 struct lod_object *lo,
3214 struct lod_thread_info *info = lod_env_info(env);
3215 struct dt_object *next = dt_object_child(&lo->ldo_obj);
3216 struct lov_mds_md_v1 *lmm;
3222 if (lo->ldo_comp_cnt == 0) {
3223 lod_object_free_striping(env, lo);
3224 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
3228 lmm_size = lod_comp_md_size(lo, false);
3229 if (info->lti_ea_store_size < lmm_size) {
3230 rc = lod_ea_store_resize(info, lmm_size);
3234 lmm = info->lti_ea_store;
3236 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
3240 info->lti_buf.lb_buf = lmm;
3241 info->lti_buf.lb_len = lmm_size;
3242 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
3243 XATTR_NAME_LOV, 0, th);
3248 * Delete layout component(s)
3250 * \param[in] env execution environment for this thread
3251 * \param[in] dt object
3252 * \param[in] th transaction handle
3254 * \retval 0 on success
3255 * \retval negative error number on failure
3257 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
3260 struct lod_layout_component *lod_comp;
3261 struct lod_object *lo = lod_dt_obj(dt);
3262 struct dt_object *next = dt_object_child(dt);
3263 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3266 LASSERT(lo->ldo_is_composite);
3267 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
3269 left = lo->ldo_comp_cnt;
3270 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3271 lod_comp = &lo->ldo_comp_entries[i];
3273 if (lod_comp->llc_id != LCME_ID_INVAL)
3277 /* Not instantiated component */
3278 if (lod_comp->llc_stripe == NULL)
3281 LASSERT(lod_comp->llc_stripe_count > 0);
3282 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
3283 struct dt_object *obj = lod_comp->llc_stripe[j];
3287 rc = lod_sub_destroy(env, obj, th);
3291 lu_object_put(env, &obj->do_lu);
3292 lod_comp->llc_stripe[j] = NULL;
3294 OBD_FREE(lod_comp->llc_stripe, sizeof(struct dt_object *) *
3295 lod_comp->llc_stripes_allocated);
3296 lod_comp->llc_stripe = NULL;
3297 lod_comp->llc_stripes_allocated = 0;
3298 lod_obj_set_pool(lo, i, NULL);
3299 if (lod_comp->llc_ostlist.op_array) {
3300 OBD_FREE(lod_comp->llc_ostlist.op_array,
3301 lod_comp->llc_ostlist.op_size);
3302 lod_comp->llc_ostlist.op_array = NULL;
3303 lod_comp->llc_ostlist.op_size = 0;
3307 LASSERTF(left >= 0 && left < lo->ldo_comp_cnt, "left = %d\n", left);
3309 struct lod_layout_component *comp_array;
3311 OBD_ALLOC(comp_array, sizeof(*comp_array) * left);
3312 if (comp_array == NULL)
3313 GOTO(out, rc = -ENOMEM);
3315 memcpy(&comp_array[0], &lo->ldo_comp_entries[0],
3316 sizeof(*comp_array) * left);
3318 OBD_FREE(lo->ldo_comp_entries,
3319 sizeof(*comp_array) * lo->ldo_comp_cnt);
3320 lo->ldo_comp_entries = comp_array;
3321 lo->ldo_comp_cnt = left;
3322 lod_obj_inc_layout_gen(lo);
3324 lod_free_comp_entries(lo);
3327 LASSERT(dt_object_exists(dt));
3328 rc = dt_attr_get(env, next, attr);
3332 if (attr->la_size > 0) {
3334 attr->la_valid = LA_SIZE;
3335 rc = lod_sub_attr_set(env, next, attr, th);
3340 rc = lod_generate_and_set_lovea(env, lo, th);
3344 lod_object_free_striping(env, lo);
3349 * Implementation of dt_object_operations::do_xattr_set.
3351 * Sets specified extended attribute on the object. Three types of EAs are
3353 * LOV EA - stores striping for a regular file or default striping (when set
3355 * LMV EA - stores a marker for the striped directories
3356 * DMV EA - stores default directory striping
3358 * When striping is applied to a non-striped existing object (this is called
3359 * late striping), then LOD notices the caller wants to turn the object into a
3360 * striped one. The stripe objects are created and appropriate EA is set:
3361 * LOV EA storing all the stripes directly or LMV EA storing just a small header
3362 * with striping configuration.
3364 * \see dt_object_operations::do_xattr_set() in the API description for details.
3366 static int lod_xattr_set(const struct lu_env *env,
3367 struct dt_object *dt, const struct lu_buf *buf,
3368 const char *name, int fl, struct thandle *th)
3370 struct dt_object *next = dt_object_child(dt);
3374 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
3375 strcmp(name, XATTR_NAME_LMV) == 0) {
3376 struct lmv_mds_md_v1 *lmm = buf->lb_buf;
3378 if (lmm != NULL && le32_to_cpu(lmm->lmv_hash_type) &
3379 LMV_HASH_FLAG_MIGRATION)
3380 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3382 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
3387 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
3388 strcmp(name, XATTR_NAME_LOV) == 0) {
3390 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl, th);
3392 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
3393 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
3395 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
3398 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
3399 (!strcmp(name, XATTR_NAME_LOV) ||
3400 !strncmp(name, XATTR_LUSTRE_LOV,
3401 strlen(XATTR_LUSTRE_LOV)))) {
3402 /* in case of lov EA swap, just set it
3403 * if not, it is a replay so check striping match what we
3404 * already have during req replay, declare_xattr_set()
3405 * defines striping, then create() does the work */
3406 if (fl & LU_XATTR_REPLACE) {
3407 /* free stripes, then update disk */
3408 lod_object_free_striping(env, lod_dt_obj(dt));
3410 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3411 } else if (dt_object_remote(dt)) {
3412 /* This only happens during migration, see
3413 * mdd_migrate_create(), in which Master MDT will
3414 * create a remote target object, and only set
3415 * (migrating) stripe EA on the remote object,
3416 * and does not need creating each stripes. */
3417 rc = lod_sub_xattr_set(env, next, buf, name,
3419 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
3420 /* delete component(s) */
3421 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
3422 rc = lod_layout_del(env, dt, th);
3425 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
3426 * it's going to create create file with specified
3427 * component(s), the striping must have not being
3428 * cached in this case;
3430 * Otherwise, it's going to add/change component(s) to
3431 * an existing file, the striping must have been cached
3434 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
3435 !strcmp(name, XATTR_NAME_LOV),
3436 !lod_dt_obj(dt)->ldo_comp_cached));
3438 rc = lod_striped_create(env, dt, NULL, NULL, th);
3441 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3442 rc = lod_replace_parent_fid(env, dt, th, false);
3447 /* then all other xattr */
3448 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3454 * Implementation of dt_object_operations::do_declare_xattr_del.
3456 * \see dt_object_operations::do_declare_xattr_del() in the API description
3459 static int lod_declare_xattr_del(const struct lu_env *env,
3460 struct dt_object *dt, const char *name,
3463 struct lod_object *lo = lod_dt_obj(dt);
3468 rc = lod_sub_declare_xattr_del(env, dt_object_child(dt), name, th);
3472 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3475 /* set xattr to each stripes, if needed */
3476 rc = lod_load_striping(env, lo);
3480 if (lo->ldo_dir_stripe_count == 0)
3483 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3484 LASSERT(lo->ldo_stripe[i]);
3485 rc = lod_sub_declare_xattr_del(env, lo->ldo_stripe[i],
3495 * Implementation of dt_object_operations::do_xattr_del.
3497 * If EA storing a regular striping is being deleted, then release
3498 * all the references to the stripe objects in core.
3500 * \see dt_object_operations::do_xattr_del() in the API description for details.
3502 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
3503 const char *name, struct thandle *th)
3505 struct dt_object *next = dt_object_child(dt);
3506 struct lod_object *lo = lod_dt_obj(dt);
3511 if (!strcmp(name, XATTR_NAME_LOV))
3512 lod_object_free_striping(env, lod_dt_obj(dt));
3514 rc = lod_sub_xattr_del(env, next, name, th);
3515 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3518 if (lo->ldo_dir_stripe_count == 0)
3521 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3522 LASSERT(lo->ldo_stripe[i]);
3524 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3533 * Implementation of dt_object_operations::do_xattr_list.
3535 * \see dt_object_operations::do_xattr_list() in the API description
3538 static int lod_xattr_list(const struct lu_env *env,
3539 struct dt_object *dt, const struct lu_buf *buf)
3541 return dt_xattr_list(env, dt_object_child(dt), buf);
3544 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
3546 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
3551 * Get default striping.
3553 * \param[in] env execution environment
3554 * \param[in] lo object
3555 * \param[out] lds default striping
3557 * \retval 0 on success
3558 * \retval negative if failed
3560 static int lod_get_default_lov_striping(const struct lu_env *env,
3561 struct lod_object *lo,
3562 struct lod_default_striping *lds)
3564 struct lod_thread_info *info = lod_env_info(env);
3565 struct lov_user_md_v1 *v1 = NULL;
3566 struct lov_user_md_v3 *v3 = NULL;
3567 struct lov_comp_md_v1 *comp_v1 = NULL;
3573 lds->lds_def_striping_set = 0;
3575 rc = lod_get_lov_ea(env, lo);
3579 if (rc < (typeof(rc))sizeof(struct lov_user_md))
3582 v1 = info->lti_ea_store;
3583 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
3584 lustre_swab_lov_user_md_v1(v1);
3585 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
3586 v3 = (struct lov_user_md_v3 *)v1;
3587 lustre_swab_lov_user_md_v3(v3);
3588 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
3589 comp_v1 = (struct lov_comp_md_v1 *)v1;
3590 lustre_swab_lov_comp_md_v1(comp_v1);
3593 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1 &&
3594 v1->lmm_magic != LOV_MAGIC_COMP_V1)
3597 if (v1->lmm_magic == LOV_MAGIC_COMP_V1) {
3598 comp_v1 = (struct lov_comp_md_v1 *)v1;
3599 comp_cnt = comp_v1->lcm_entry_count;
3608 /* realloc default comp entries if necessary */
3609 rc = lod_def_striping_comp_resize(lds, comp_cnt);
3613 lds->lds_def_comp_cnt = comp_cnt;
3614 lds->lds_def_striping_is_composite = composite ? 1 : 0;
3616 for (i = 0; i < comp_cnt; i++) {
3617 struct lod_layout_component *lod_comp;
3618 struct lu_extent *ext;
3621 lod_comp = &lds->lds_def_comp_entries[i];
3623 * reset lod_comp values, llc_stripes is always NULL in
3624 * the default striping template, llc_pool will be reset
3627 memset(lod_comp, 0, offsetof(typeof(*lod_comp), llc_pool));
3630 v1 = (struct lov_user_md *)((char *)comp_v1 +
3631 comp_v1->lcm_entries[i].lcme_offset);
3632 ext = &comp_v1->lcm_entries[i].lcme_extent;
3633 lod_comp->llc_extent = *ext;
3636 if (v1->lmm_pattern != LOV_PATTERN_RAID0 &&
3637 v1->lmm_pattern != 0) {
3638 lod_free_def_comp_entries(lds);
3642 CDEBUG(D_LAYOUT, DFID" stripe_count=%d stripe_size=%d "
3643 "stripe_offset=%d\n",
3644 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
3645 (int)v1->lmm_stripe_count, (int)v1->lmm_stripe_size,
3646 (int)v1->lmm_stripe_offset);
3648 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
3649 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
3650 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
3653 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
3654 /* XXX: sanity check here */
3655 v3 = (struct lov_user_md_v3 *) v1;
3656 if (v3->lmm_pool_name[0] != '\0')
3657 pool = v3->lmm_pool_name;
3659 lod_set_def_pool(lds, i, pool);
3662 lds->lds_def_striping_set = 1;
3667 * Get default directory striping.
3669 * \param[in] env execution environment
3670 * \param[in] lo object
3671 * \param[out] lds default striping
3673 * \retval 0 on success
3674 * \retval negative if failed
3676 static int lod_get_default_lmv_striping(const struct lu_env *env,
3677 struct lod_object *lo,
3678 struct lod_default_striping *lds)
3680 struct lod_thread_info *info = lod_env_info(env);
3681 struct lmv_user_md_v1 *v1 = NULL;
3685 lds->lds_dir_def_striping_set = 0;
3686 rc = lod_get_default_lmv_ea(env, lo);
3690 if (rc < (typeof(rc))sizeof(struct lmv_user_md))
3693 v1 = info->lti_ea_store;
3695 lds->lds_dir_def_stripe_count = le32_to_cpu(v1->lum_stripe_count);
3696 lds->lds_dir_def_stripe_offset = le32_to_cpu(v1->lum_stripe_offset);
3697 lds->lds_dir_def_hash_type = le32_to_cpu(v1->lum_hash_type);
3698 lds->lds_dir_def_striping_set = 1;
3704 * Get default striping in the object.
3706 * Get object default striping and default directory striping.
3708 * \param[in] env execution environment
3709 * \param[in] lo object
3710 * \param[out] lds default striping
3712 * \retval 0 on success
3713 * \retval negative if failed
3715 static int lod_get_default_striping(const struct lu_env *env,
3716 struct lod_object *lo,
3717 struct lod_default_striping *lds)
3721 rc = lod_get_default_lov_striping(env, lo, lds);
3722 rc1 = lod_get_default_lmv_striping(env, lo, lds);
3723 if (rc == 0 && rc1 < 0)
3730 * Apply default striping on object.
3732 * If object striping pattern is not set, set to the one in default striping.
3733 * The default striping is from parent or fs.
3735 * \param[in] lo new object
3736 * \param[in] lds default striping
3737 * \param[in] mode new object's mode
3739 static void lod_striping_from_default(struct lod_object *lo,
3740 const struct lod_default_striping *lds,
3743 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
3744 struct lov_desc *desc = &d->lod_desc;
3747 if (lds->lds_def_striping_set && S_ISREG(mode)) {
3748 rc = lod_alloc_comp_entries(lo, lds->lds_def_comp_cnt);
3752 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
3754 for (i = 0; i < lo->ldo_comp_cnt; i++) {
3755 struct lod_layout_component *obj_comp =
3756 &lo->ldo_comp_entries[i];
3757 struct lod_layout_component *def_comp =
3758 &lds->lds_def_comp_entries[i];
3760 CDEBUG(D_LAYOUT, "Inherite from default: size:%hu "
3761 "nr:%u offset:%u %s\n",
3762 def_comp->llc_stripe_size,
3763 def_comp->llc_stripe_count,
3764 def_comp->llc_stripe_offset,
3765 def_comp->llc_pool ?: "");
3767 *obj_comp = *def_comp;
3768 if (def_comp->llc_pool != NULL) {
3769 /* pointer was copied from def_comp */
3770 obj_comp->llc_pool = NULL;
3771 lod_obj_set_pool(lo, i, def_comp->llc_pool);
3775 * Don't initialize these fields for plain layout
3776 * (v1/v3) here, they are inherited in the order of
3777 * 'parent' -> 'fs default (root)' -> 'global default
3778 * values for stripe_count & stripe_size'.
3780 * see lod_ah_init().
3782 if (!lo->ldo_is_composite)
3785 if (obj_comp->llc_stripe_count <= 0)
3786 obj_comp->llc_stripe_count =
3787 desc->ld_default_stripe_count;
3788 if (obj_comp->llc_stripe_size <= 0)
3789 obj_comp->llc_stripe_size =
3790 desc->ld_default_stripe_size;
3792 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
3793 if (lo->ldo_dir_stripe_count == 0)
3794 lo->ldo_dir_stripe_count =
3795 lds->lds_dir_def_stripe_count;
3796 if (lo->ldo_dir_stripe_offset == -1)
3797 lo->ldo_dir_stripe_offset =
3798 lds->lds_dir_def_stripe_offset;
3799 if (lo->ldo_dir_hash_type == 0)
3800 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
3802 CDEBUG(D_LAYOUT, "striping from default dir: count:%hu, "
3803 "offset:%u, hash_type:%u\n",
3804 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
3805 lo->ldo_dir_hash_type);
3809 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root)
3811 struct lod_layout_component *lod_comp;
3813 if (lo->ldo_comp_cnt == 0)
3816 if (lo->ldo_is_composite)
3819 lod_comp = &lo->ldo_comp_entries[0];
3821 if (lod_comp->llc_stripe_count <= 0 ||
3822 lod_comp->llc_stripe_size <= 0)
3825 if (from_root && (lod_comp->llc_pool == NULL ||
3826 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
3833 * Implementation of dt_object_operations::do_ah_init.
3835 * This method is used to make a decision on the striping configuration for the
3836 * object being created. It can be taken from the \a parent object if it exists,
3837 * or filesystem's default. The resulting configuration (number of stripes,
3838 * stripe size/offset, pool name, etc) is stored in the object itself and will
3839 * be used by the methods like ->doo_declare_create().
3841 * \see dt_object_operations::do_ah_init() in the API description for details.
3843 static void lod_ah_init(const struct lu_env *env,
3844 struct dt_allocation_hint *ah,
3845 struct dt_object *parent,
3846 struct dt_object *child,
3849 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
3850 struct lod_thread_info *info = lod_env_info(env);
3851 struct lod_default_striping *lds = &info->lti_def_striping;
3852 struct dt_object *nextp = NULL;
3853 struct dt_object *nextc;
3854 struct lod_object *lp = NULL;
3855 struct lod_object *lc;
3856 struct lov_desc *desc;
3857 struct lod_layout_component *lod_comp;
3863 if (likely(parent)) {
3864 nextp = dt_object_child(parent);
3865 lp = lod_dt_obj(parent);
3868 nextc = dt_object_child(child);
3869 lc = lod_dt_obj(child);
3871 LASSERT(!lod_obj_is_striped(child));
3872 /* default layout template may have been set on the regular file
3873 * when this is called from mdd_create_data() */
3874 if (S_ISREG(child_mode))
3875 lod_free_comp_entries(lc);
3877 if (!dt_object_exists(nextc))
3878 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
3880 if (S_ISDIR(child_mode)) {
3881 /* other default values are 0 */
3882 lc->ldo_dir_stripe_offset = -1;
3884 /* get default striping from parent object */
3885 if (likely(lp != NULL))
3886 lod_get_default_striping(env, lp, lds);
3888 /* set child default striping info, default value is NULL */
3889 if (lds->lds_def_striping_set || lds->lds_dir_def_striping_set)
3890 lc->ldo_def_striping = lds;
3892 /* It should always honour the specified stripes */
3893 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
3894 lod_verify_md_striping(d, ah->dah_eadata) == 0) {
3895 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
3897 lc->ldo_dir_stripe_count =
3898 le32_to_cpu(lum1->lum_stripe_count);
3899 lc->ldo_dir_stripe_offset =
3900 le32_to_cpu(lum1->lum_stripe_offset);
3901 lc->ldo_dir_hash_type =
3902 le32_to_cpu(lum1->lum_hash_type);
3904 "set dirstripe: count %hu, offset %d, hash %u\n",
3905 lc->ldo_dir_stripe_count,
3906 (int)lc->ldo_dir_stripe_offset,
3907 lc->ldo_dir_hash_type);
3909 /* transfer defaults LMV to new directory */
3910 lod_striping_from_default(lc, lds, child_mode);
3913 /* shrink the stripe_count to the avaible MDT count */
3914 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
3915 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))
3916 lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
3918 /* Directory will be striped only if stripe_count > 1, if
3919 * stripe_count == 1, let's reset stripe_count = 0 to avoid
3920 * create single master stripe and also help to unify the
3921 * stripe handling of directories and files */
3922 if (lc->ldo_dir_stripe_count == 1)
3923 lc->ldo_dir_stripe_count = 0;
3925 CDEBUG(D_INFO, "final dir stripe [%hu %d %u]\n",
3926 lc->ldo_dir_stripe_count,
3927 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
3932 /* child object regular file*/
3934 if (!lod_object_will_be_striped(S_ISREG(child_mode),
3935 lu_object_fid(&child->do_lu)))
3938 /* If object is going to be striped over OSTs, transfer default
3939 * striping information to the child, so that we can use it
3940 * during declaration and creation.
3942 * Try from the parent first.
3944 if (likely(lp != NULL)) {
3945 rc = lod_get_default_lov_striping(env, lp, lds);
3947 lod_striping_from_default(lc, lds, child_mode);
3950 /* Initialize lod_device::lod_md_root object reference */
3951 if (d->lod_md_root == NULL) {
3952 struct dt_object *root;
3953 struct lod_object *lroot;
3955 lu_root_fid(&info->lti_fid);
3956 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
3957 if (!IS_ERR(root)) {
3958 lroot = lod_dt_obj(root);
3960 spin_lock(&d->lod_lock);
3961 if (d->lod_md_root != NULL)
3962 dt_object_put(env, &d->lod_md_root->ldo_obj);
3963 d->lod_md_root = lroot;
3964 spin_unlock(&d->lod_lock);
3968 /* try inherit layout from the root object (fs default) when:
3969 * - parent does not have default layout; or
3970 * - parent has plain(v1/v3) default layout, and some attributes
3971 * are not specified in the default layout;
3973 if (d->lod_md_root != NULL && lod_need_inherit_more(lc, true)) {
3974 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds);
3977 if (lc->ldo_comp_cnt == 0) {
3978 lod_striping_from_default(lc, lds, child_mode);
3979 } else if (!lds->lds_def_striping_is_composite) {
3980 struct lod_layout_component *def_comp;
3982 LASSERT(!lc->ldo_is_composite);
3983 lod_comp = &lc->ldo_comp_entries[0];
3984 def_comp = &lds->lds_def_comp_entries[0];
3986 if (lod_comp->llc_stripe_count <= 0)
3987 lod_comp->llc_stripe_count =
3988 def_comp->llc_stripe_count;
3989 if (lod_comp->llc_stripe_size <= 0)
3990 lod_comp->llc_stripe_size =
3991 def_comp->llc_stripe_size;
3992 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT)
3993 lod_comp->llc_stripe_offset =
3994 def_comp->llc_stripe_offset;
3995 if (lod_comp->llc_pool == NULL)
3996 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
4001 * fs default striping may not be explicitly set, or historically set
4002 * in config log, use them.
4004 if (lod_need_inherit_more(lc, false)) {
4006 if (lc->ldo_comp_cnt == 0) {
4007 rc = lod_alloc_comp_entries(lc, 1);
4009 /* fail to allocate memory, will create a
4010 * non-striped file. */
4012 lc->ldo_is_composite = 0;
4013 lod_comp = &lc->ldo_comp_entries[0];
4014 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4016 LASSERT(!lc->ldo_is_composite);
4017 lod_comp = &lc->ldo_comp_entries[0];
4018 desc = &d->lod_desc;
4019 if (lod_comp->llc_stripe_count <= 0)
4020 lod_comp->llc_stripe_count =
4021 desc->ld_default_stripe_count;
4022 if (lod_comp->llc_stripe_size <= 0)
4023 lod_comp->llc_stripe_size =
4024 desc->ld_default_stripe_size;
4030 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
4032 * Size initialization on late striping.
4034 * Propagate the size of a truncated object to a deferred striping.
4035 * This function handles a special case when truncate was done on a
4036 * non-striped object and now while the striping is being created
4037 * we can't lose that size, so we have to propagate it to the stripes
4040 * \param[in] env execution environment
4041 * \param[in] dt object
4042 * \param[in] th transaction handle
4044 * \retval 0 on success
4045 * \retval negative if failed
4047 static int lod_declare_init_size(const struct lu_env *env,
4048 struct dt_object *dt, struct thandle *th)
4050 struct dt_object *next = dt_object_child(dt);
4051 struct lod_object *lo = lod_dt_obj(dt);
4052 struct dt_object **objects = NULL;
4053 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4054 uint64_t size, offs;
4055 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
4058 if (!lod_obj_is_striped(dt))
4061 rc = dt_attr_get(env, next, attr);
4062 LASSERT(attr->la_valid & LA_SIZE);
4066 size = attr->la_size;
4070 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4071 struct lod_layout_component *lod_comp;
4072 struct lu_extent *extent;
4074 lod_comp = &lo->ldo_comp_entries[i];
4076 if (lod_comp->llc_stripe == NULL)
4079 extent = &lod_comp->llc_extent;
4080 CDEBUG(D_INFO, "%lld [%lld, %lld)\n",
4081 size, extent->e_start, extent->e_end);
4082 if (!lo->ldo_is_composite ||
4083 (size >= extent->e_start && size < extent->e_end)) {
4084 objects = lod_comp->llc_stripe;
4085 stripe_count = lod_comp->llc_stripe_count;
4086 stripe_size = lod_comp->llc_stripe_size;
4091 if (stripe_count == 0)
4094 LASSERT(objects != NULL && stripe_size != 0);
4096 /* ll_do_div64(a, b) returns a % b, and a = a / b */
4097 ll_do_div64(size, (__u64)stripe_size);
4098 stripe = ll_do_div64(size, (__u64)stripe_count);
4099 LASSERT(objects[stripe] != NULL);
4101 size = size * stripe_size;
4102 offs = attr->la_size;
4103 size += ll_do_div64(offs, stripe_size);
4105 attr->la_valid = LA_SIZE;
4106 attr->la_size = size;
4108 rc = lod_sub_declare_attr_set(env, objects[stripe], attr, th);
4114 * Declare creation of striped object.
4116 * The function declares creation stripes for a regular object. The function
4117 * also declares whether the stripes will be created with non-zero size if
4118 * previously size was set non-zero on the master object. If object \a dt is
4119 * not local, then only fully defined striping can be applied in \a lovea.
4120 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
4123 * \param[in] env execution environment
4124 * \param[in] dt object
4125 * \param[in] attr attributes the stripes will be created with
4126 * \param[in] lovea a buffer containing striping description
4127 * \param[in] th transaction handle
4129 * \retval 0 on success
4130 * \retval negative if failed
4132 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
4133 struct lu_attr *attr,
4134 const struct lu_buf *lovea, struct thandle *th)
4136 struct lod_thread_info *info = lod_env_info(env);
4137 struct dt_object *next = dt_object_child(dt);
4138 struct lod_object *lo = lod_dt_obj(dt);
4142 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
4143 GOTO(out, rc = -ENOMEM);
4145 if (!dt_object_remote(next)) {
4146 /* choose OST and generate appropriate objects */
4147 rc = lod_prepare_create(env, lo, attr, lovea, th);
4152 * declare storage for striping data
4154 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
4156 /* LOD can not choose OST objects for remote objects, i.e.
4157 * stripes must be ready before that. Right now, it can only
4158 * happen during migrate, i.e. migrate process needs to create
4159 * remote regular file (mdd_migrate_create), then the migrate
4160 * process will provide stripeEA. */
4161 LASSERT(lovea != NULL);
4162 info->lti_buf = *lovea;
4165 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
4166 XATTR_NAME_LOV, 0, th);
4171 * if striping is created with local object's size > 0,
4172 * we have to propagate this size to specific object
4173 * the case is possible only when local object was created previously
4175 if (dt_object_exists(next))
4176 rc = lod_declare_init_size(env, dt, th);
4179 /* failed to create striping or to set initial size, let's reset
4180 * config so that others don't get confused */
4182 lod_object_free_striping(env, lo);
4188 * Implementation of dt_object_operations::do_declare_create.
4190 * The method declares creation of a new object. If the object will be striped,
4191 * then helper functions are called to find FIDs for the stripes, declare
4192 * creation of the stripes and declare initialization of the striping
4193 * information to be stored in the master object.
4195 * \see dt_object_operations::do_declare_create() in the API description
4198 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
4199 struct lu_attr *attr,
4200 struct dt_allocation_hint *hint,
4201 struct dt_object_format *dof, struct thandle *th)
4203 struct dt_object *next = dt_object_child(dt);
4204 struct lod_object *lo = lod_dt_obj(dt);
4213 * first of all, we declare creation of local object
4215 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
4220 * it's lod_ah_init() that has decided the object will be striped
4222 if (dof->dof_type == DFT_REGULAR) {
4223 /* callers don't want stripes */
4224 /* XXX: all tricky interactions with ->ah_make_hint() decided
4225 * to use striping, then ->declare_create() behaving differently
4226 * should be cleaned */
4227 if (dof->u.dof_reg.striped != 0)
4228 rc = lod_declare_striped_create(env, dt, attr,
4230 } else if (dof->dof_type == DFT_DIR) {
4231 struct seq_server_site *ss;
4233 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
4235 /* If the parent has default stripeEA, and client
4236 * did not find it before sending create request,
4237 * then MDT will return -EREMOTE, and client will
4238 * retrieve the default stripeEA and re-create the
4241 * Note: if dah_eadata != NULL, it means creating the
4242 * striped directory with specified stripeEA, then it
4243 * should ignore the default stripeEA */
4244 if (hint != NULL && hint->dah_eadata == NULL) {
4245 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
4246 GOTO(out, rc = -EREMOTE);
4248 if (lo->ldo_dir_stripe_offset == -1) {
4249 /* child and parent should be in the same MDT */
4250 if (hint->dah_parent != NULL &&
4251 dt_object_remote(hint->dah_parent))
4252 GOTO(out, rc = -EREMOTE);
4253 } else if (lo->ldo_dir_stripe_offset !=
4255 struct lod_device *lod;
4256 struct lod_tgt_descs *ltd;
4257 struct lod_tgt_desc *tgt = NULL;
4258 bool found_mdt = false;
4261 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
4262 ltd = &lod->lod_mdt_descs;
4263 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
4264 tgt = LTD_TGT(ltd, i);
4265 if (tgt->ltd_index ==
4266 lo->ldo_dir_stripe_offset) {
4272 /* If the MDT indicated by stripe_offset can be
4273 * found, then tell client to resend the create
4274 * request to the correct MDT, otherwise return
4275 * error to client */
4277 GOTO(out, rc = -EREMOTE);
4279 GOTO(out, rc = -EINVAL);
4283 rc = lod_declare_dir_striping_create(env, dt, attr, dof, th);
4286 /* failed to create striping or to set initial size, let's reset
4287 * config so that others don't get confused */
4289 lod_object_free_striping(env, lo);
4294 * Creation of a striped regular object.
4296 * The function is called to create the stripe objects for a regular
4297 * striped file. This can happen at the initial object creation or
4298 * when the caller asks LOD to do so using ->do_xattr_set() method
4299 * (so called late striping). Notice all the information are already
4300 * prepared in the form of the list of objects (ldo_stripe field).
4301 * This is done during declare phase.
4303 * \param[in] env execution environment
4304 * \param[in] dt object
4305 * \param[in] attr attributes the stripes will be created with
4306 * \param[in] dof format of stripes (see OSD API description)
4307 * \param[in] th transaction handle
4309 * \retval 0 on success
4310 * \retval negative if failed
4312 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
4313 struct lu_attr *attr, struct dt_object_format *dof,
4316 struct lod_layout_component *lod_comp;
4317 struct lod_object *lo = lod_dt_obj(dt);
4321 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
4323 /* create all underlying objects */
4324 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4325 lod_comp = &lo->ldo_comp_entries[i];
4327 if (lod_comp_inited(lod_comp))
4330 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
4331 lod_comp_set_init(lod_comp);
4333 if (lod_comp->llc_stripe == NULL)
4336 LASSERT(lod_comp->llc_stripe_count);
4337 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4338 struct dt_object *object = lod_comp->llc_stripe[j];
4339 LASSERT(object != NULL);
4340 rc = lod_sub_create(env, object, attr, NULL, dof, th);
4344 lod_comp_set_init(lod_comp);
4348 rc = lod_generate_and_set_lovea(env, lo, th);
4351 lo->ldo_comp_cached = 1;
4353 lod_object_free_striping(env, lo);
4359 * Implementation of dt_object_operations::do_create.
4361 * If any of preceeding methods (like ->do_declare_create(),
4362 * ->do_ah_init(), etc) chose to create a striped object,
4363 * then this method will create the master and the stripes.
4365 * \see dt_object_operations::do_create() in the API description for details.
4367 static int lod_create(const struct lu_env *env, struct dt_object *dt,
4368 struct lu_attr *attr, struct dt_allocation_hint *hint,
4369 struct dt_object_format *dof, struct thandle *th)
4374 /* create local object */
4375 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
4379 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
4380 lod_obj_is_striped(dt) && dof->u.dof_reg.striped != 0) {
4381 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
4382 rc = lod_striped_create(env, dt, attr, dof, th);
4389 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
4390 struct dt_object *dt, struct thandle *th,
4391 int stripe_idx, struct lod_obj_stripe_cb_data *data)
4393 if (data->locd_declare)
4394 return lod_sub_declare_destroy(env, dt, th);
4395 else if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
4396 stripe_idx == cfs_fail_val)
4397 return lod_sub_destroy(env, dt, th);
4403 * Implementation of dt_object_operations::do_declare_destroy.
4405 * If the object is a striped directory, then the function declares reference
4406 * removal from the master object (this is an index) to the stripes and declares
4407 * destroy of all the stripes. In all the cases, it declares an intention to
4408 * destroy the object itself.
4410 * \see dt_object_operations::do_declare_destroy() in the API description
4413 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
4416 struct dt_object *next = dt_object_child(dt);
4417 struct lod_object *lo = lod_dt_obj(dt);
4418 struct lod_thread_info *info = lod_env_info(env);
4419 char *stripe_name = info->lti_key;
4424 * load striping information, notice we don't do this when object
4425 * is being initialized as we don't need this information till
4426 * few specific cases like destroy, chown
4428 rc = lod_load_striping(env, lo);
4432 /* declare destroy for all underlying objects */
4433 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4434 rc = next->do_ops->do_index_try(env, next,
4435 &dt_directory_features);
4439 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4440 rc = lod_sub_declare_ref_del(env, next, th);
4444 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4445 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
4447 rc = lod_sub_declare_delete(env, next,
4448 (const struct dt_key *)stripe_name, th);
4455 * we declare destroy for the local object
4457 rc = lod_sub_declare_destroy(env, next, th);
4461 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
4462 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
4465 if (!lod_obj_is_striped(dt))
4468 /* declare destroy all striped objects */
4469 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4470 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4471 if (lo->ldo_stripe[i] == NULL)
4474 rc = lod_sub_declare_ref_del(env, lo->ldo_stripe[i],
4477 rc = lod_sub_declare_destroy(env, lo->ldo_stripe[i],
4483 struct lod_obj_stripe_cb_data data;
4485 data.locd_declare = true;
4486 rc = lod_obj_for_each_stripe(env, lo, th,
4487 lod_obj_stripe_destroy_cb, &data);
4494 * Implementation of dt_object_operations::do_destroy.
4496 * If the object is a striped directory, then the function removes references
4497 * from the master object (this is an index) to the stripes and destroys all
4498 * the stripes. In all the cases, the function destroys the object itself.
4500 * \see dt_object_operations::do_destroy() in the API description for details.
4502 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
4505 struct dt_object *next = dt_object_child(dt);
4506 struct lod_object *lo = lod_dt_obj(dt);
4507 struct lod_thread_info *info = lod_env_info(env);
4508 char *stripe_name = info->lti_key;
4513 /* destroy sub-stripe of master object */
4514 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4515 rc = next->do_ops->do_index_try(env, next,
4516 &dt_directory_features);
4520 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4521 rc = lod_sub_ref_del(env, next, th);
4525 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4526 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
4529 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
4530 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
4531 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)));
4533 rc = lod_sub_delete(env, next,
4534 (const struct dt_key *)stripe_name, th);
4540 rc = lod_sub_destroy(env, next, th);
4544 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
4545 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
4548 if (!lod_obj_is_striped(dt))
4551 /* destroy all striped objects */
4552 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4553 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4554 if (lo->ldo_stripe[i] == NULL)
4556 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
4557 i == cfs_fail_val) {
4558 dt_write_lock(env, lo->ldo_stripe[i],
4560 rc = lod_sub_ref_del(env, lo->ldo_stripe[i],
4562 dt_write_unlock(env, lo->ldo_stripe[i]);
4566 rc = lod_sub_destroy(env, lo->ldo_stripe[i],
4573 struct lod_obj_stripe_cb_data data;
4575 data.locd_declare = false;
4576 rc = lod_obj_for_each_stripe(env, lo, th,
4577 lod_obj_stripe_destroy_cb, &data);
4584 * Implementation of dt_object_operations::do_declare_ref_add.
4586 * \see dt_object_operations::do_declare_ref_add() in the API description
4589 static int lod_declare_ref_add(const struct lu_env *env,
4590 struct dt_object *dt, struct thandle *th)
4592 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
4596 * Implementation of dt_object_operations::do_ref_add.
4598 * \see dt_object_operations::do_ref_add() in the API description for details.
4600 static int lod_ref_add(const struct lu_env *env,
4601 struct dt_object *dt, struct thandle *th)
4603 return lod_sub_ref_add(env, dt_object_child(dt), th);
4607 * Implementation of dt_object_operations::do_declare_ref_del.
4609 * \see dt_object_operations::do_declare_ref_del() in the API description
4612 static int lod_declare_ref_del(const struct lu_env *env,
4613 struct dt_object *dt, struct thandle *th)
4615 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
4619 * Implementation of dt_object_operations::do_ref_del
4621 * \see dt_object_operations::do_ref_del() in the API description for details.
4623 static int lod_ref_del(const struct lu_env *env,
4624 struct dt_object *dt, struct thandle *th)
4626 return lod_sub_ref_del(env, dt_object_child(dt), th);
4630 * Implementation of dt_object_operations::do_object_sync.
4632 * \see dt_object_operations::do_object_sync() in the API description
4635 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
4636 __u64 start, __u64 end)
4638 return dt_object_sync(env, dt_object_child(dt), start, end);
4642 * Release LDLM locks on the stripes of a striped directory.
4644 * Iterates over all the locks taken on the stripe objects and
4647 * \param[in] env execution environment
4648 * \param[in] dt striped object
4649 * \param[in] einfo lock description
4650 * \param[in] policy data describing requested lock
4652 * \retval 0 on success
4653 * \retval negative if failed
4655 static int lod_object_unlock_internal(const struct lu_env *env,
4656 struct dt_object *dt,
4657 struct ldlm_enqueue_info *einfo,
4658 union ldlm_policy_data *policy)
4660 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
4665 if (slave_locks == NULL)
4668 for (i = 1; i < slave_locks->count; i++) {
4669 if (lustre_handle_is_used(&slave_locks->handles[i]))
4670 ldlm_lock_decref_and_cancel(&slave_locks->handles[i],
4678 * Implementation of dt_object_operations::do_object_unlock.
4680 * Used to release LDLM lock(s).
4682 * \see dt_object_operations::do_object_unlock() in the API description
4685 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
4686 struct ldlm_enqueue_info *einfo,
4687 union ldlm_policy_data *policy)
4689 struct lod_object *lo = lod_dt_obj(dt);
4690 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
4691 int slave_locks_size;
4695 if (slave_locks == NULL)
4698 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
4699 LASSERT(lo->ldo_dir_stripe_count > 1);
4700 /* Note: for remote lock for single stripe dir, MDT will cancel
4701 * the lock by lockh directly */
4702 LASSERT(!dt_object_remote(dt_object_child(dt)));
4704 /* locks were unlocked in MDT layer */
4705 for (i = 1; i < slave_locks->count; i++) {
4706 LASSERT(!lustre_handle_is_used(&slave_locks->handles[i]));
4707 dt_invalidate(env, lo->ldo_stripe[i]);
4710 slave_locks_size = sizeof(*slave_locks) + slave_locks->count *
4711 sizeof(slave_locks->handles[0]);
4712 OBD_FREE(slave_locks, slave_locks_size);
4713 einfo->ei_cbdata = NULL;
4719 * Implementation of dt_object_operations::do_object_lock.
4721 * Used to get LDLM lock on the non-striped and striped objects.
4723 * \see dt_object_operations::do_object_lock() in the API description
4726 static int lod_object_lock(const struct lu_env *env,
4727 struct dt_object *dt,
4728 struct lustre_handle *lh,
4729 struct ldlm_enqueue_info *einfo,
4730 union ldlm_policy_data *policy)
4732 struct lod_object *lo = lod_dt_obj(dt);
4735 int slave_locks_size;
4736 struct lustre_handle_array *slave_locks = NULL;
4739 /* remote object lock */
4740 if (!einfo->ei_enq_slave) {
4741 LASSERT(dt_object_remote(dt));
4742 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
4746 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4747 GOTO(out, rc = -ENOTDIR);
4749 rc = lod_load_striping(env, lo);
4754 if (lo->ldo_dir_stripe_count <= 1) {
4756 * NB, ei_cbdata stores pointer to slave locks, if no locks
4757 * taken, make sure it's set to NULL, otherwise MDT will try to
4760 einfo->ei_cbdata = NULL;
4764 slave_locks_size = sizeof(*slave_locks) + lo->ldo_dir_stripe_count *
4765 sizeof(slave_locks->handles[0]);
4766 /* Freed in lod_object_unlock */
4767 OBD_ALLOC(slave_locks, slave_locks_size);
4768 if (slave_locks == NULL)
4769 GOTO(out, rc = -ENOMEM);
4770 slave_locks->count = lo->ldo_dir_stripe_count;
4772 /* striped directory lock */
4773 for (i = 1; i < lo->ldo_dir_stripe_count; i++) {
4774 struct lustre_handle lockh;
4775 struct ldlm_res_id *res_id;
4777 res_id = &lod_env_info(env)->lti_res_id;
4778 fid_build_reg_res_name(lu_object_fid(&lo->ldo_stripe[i]->do_lu),
4780 einfo->ei_res_id = res_id;
4782 LASSERT(lo->ldo_stripe[i] != NULL);
4783 if (likely(dt_object_remote(lo->ldo_stripe[i]))) {
4784 rc = dt_object_lock(env, lo->ldo_stripe[i], &lockh,
4787 struct ldlm_namespace *ns = einfo->ei_namespace;
4788 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
4789 ldlm_completion_callback completion = einfo->ei_cb_cp;
4790 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
4792 if (einfo->ei_mode == LCK_PW ||
4793 einfo->ei_mode == LCK_EX)
4794 dlmflags |= LDLM_FL_COS_INCOMPAT;
4796 /* This only happens if there are mulitple stripes
4797 * on the master MDT, i.e. except stripe0, there are
4798 * other stripes on the Master MDT as well, Only
4799 * happens in the test case right now. */
4800 LASSERT(ns != NULL);
4801 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS,
4802 policy, einfo->ei_mode,
4803 &dlmflags, blocking,
4805 NULL, 0, LVB_T_NONE,
4810 slave_locks->handles[i] = lockh;
4812 einfo->ei_cbdata = slave_locks;
4814 if (rc != 0 && slave_locks != NULL) {
4815 lod_object_unlock_internal(env, dt, einfo, policy);
4816 OBD_FREE(slave_locks, slave_locks_size);
4821 einfo->ei_cbdata = NULL;
4826 * Implementation of dt_object_operations::do_invalidate.
4828 * \see dt_object_operations::do_invalidate() in the API description for details
4830 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
4832 return dt_invalidate(env, dt_object_child(dt));
4835 static int lod_declare_layout_change(const struct lu_env *env,
4836 struct dt_object *dt,
4837 struct layout_intent *layout,
4838 const struct lu_buf *buf,
4841 struct lod_thread_info *info = lod_env_info(env);
4842 struct lod_object *lo = lod_dt_obj(dt);
4843 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
4844 struct dt_object *next = dt_object_child(dt);
4845 struct ost_pool *inuse = &info->lti_inuse_osts;
4846 struct lod_layout_component *lod_comp;
4847 struct lov_comp_md_v1 *comp_v1 = NULL;
4848 bool replay = false;
4849 bool need_create = false;
4853 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
4854 dt_object_remote(next))
4857 dt_write_lock(env, next, 0);
4859 * In case the client is passing lovea, which only happens during
4860 * the replay of layout intent write RPC for now, we may need to
4861 * parse the lovea and apply new layout configuration.
4863 if (buf && buf->lb_len) {
4864 struct lov_user_md_v1 *v1 = buf->lb_buf;
4866 if (v1->lmm_magic != (LOV_MAGIC_DEF | LOV_MAGIC_COMP_V1) &&
4868 __swab32(LOV_MAGIC_DEF | LOV_MAGIC_COMP_V1)) {
4869 CERROR("%s: the replay buffer of layout extend "
4870 "(magic %#x) does not contain expected "
4871 "composite layout.\n",
4872 lod2obd(d)->obd_name, v1->lmm_magic);
4873 GOTO(out, rc = -EINVAL);
4876 lod_object_free_striping(env, lo);
4877 rc = lod_use_defined_striping(env, lo, buf);
4881 rc = lod_get_lov_ea(env, lo);
4884 /* old on-disk EA is stored in info->lti_buf */
4885 comp_v1 = (struct lov_comp_md_v1 *)&info->lti_buf.lb_buf;
4888 /* non replay path */
4889 rc = lod_load_striping_locked(env, lo);
4893 /* Prepare inuse array for composite file */
4894 rc = lod_prepare_inuse(env, lo);
4899 /* Make sure defined layout covers the requested write range. */
4900 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
4901 if (lo->ldo_comp_cnt > 1 &&
4902 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
4903 lod_comp->llc_extent.e_end < layout->li_end) {
4904 CDEBUG(replay ? D_ERROR : D_LAYOUT,
4905 "%s: the defined layout [0, %#llx) does not covers "
4906 "the write range [%#llx, %#llx).\n",
4907 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
4908 layout->li_start, layout->li_end);
4909 GOTO(out, rc = -EINVAL);
4913 * Iterate ld->ldo_comp_entries, find the component whose extent under
4914 * the write range and not instantianted.
4916 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4917 lod_comp = &lo->ldo_comp_entries[i];
4919 if (lod_comp->llc_extent.e_start >= layout->li_end)
4923 if (lod_comp_inited(lod_comp))
4927 * In replay path, lod_comp is the EA passed by
4928 * client replay buffer, comp_v1 is the pre-recovery
4929 * on-disk EA, we'd sift out those components which
4930 * were init-ed in the on-disk EA.
4932 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
4937 * this component hasn't instantiated in normal path, or during
4938 * replay it needs replay the instantiation.
4941 /* A released component is being extended */
4942 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
4943 GOTO(out, rc = -EINVAL);
4947 rc = lod_qos_prep_create(env, lo, NULL, th, i, inuse);
4953 lod_obj_inc_layout_gen(lo);
4955 GOTO(unlock, rc = -EALREADY);
4958 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
4959 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
4960 XATTR_NAME_LOV, 0, th);
4964 lod_object_free_striping(env, lo);
4967 dt_write_unlock(env, next);
4973 * Instantiate layout component objects which covers the intent write offset.
4975 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
4976 struct layout_intent *layout,
4977 const struct lu_buf *buf, struct thandle *th)
4979 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4981 RETURN(lod_striped_create(env, dt, attr, NULL, th));
4984 struct dt_object_operations lod_obj_ops = {
4985 .do_read_lock = lod_read_lock,
4986 .do_write_lock = lod_write_lock,
4987 .do_read_unlock = lod_read_unlock,
4988 .do_write_unlock = lod_write_unlock,
4989 .do_write_locked = lod_write_locked,
4990 .do_attr_get = lod_attr_get,
4991 .do_declare_attr_set = lod_declare_attr_set,
4992 .do_attr_set = lod_attr_set,
4993 .do_xattr_get = lod_xattr_get,
4994 .do_declare_xattr_set = lod_declare_xattr_set,
4995 .do_xattr_set = lod_xattr_set,
4996 .do_declare_xattr_del = lod_declare_xattr_del,
4997 .do_xattr_del = lod_xattr_del,
4998 .do_xattr_list = lod_xattr_list,
4999 .do_ah_init = lod_ah_init,
5000 .do_declare_create = lod_declare_create,
5001 .do_create = lod_create,
5002 .do_declare_destroy = lod_declare_destroy,
5003 .do_destroy = lod_destroy,
5004 .do_index_try = lod_index_try,
5005 .do_declare_ref_add = lod_declare_ref_add,
5006 .do_ref_add = lod_ref_add,
5007 .do_declare_ref_del = lod_declare_ref_del,
5008 .do_ref_del = lod_ref_del,
5009 .do_object_sync = lod_object_sync,
5010 .do_object_lock = lod_object_lock,
5011 .do_object_unlock = lod_object_unlock,
5012 .do_invalidate = lod_invalidate,
5013 .do_declare_layout_change = lod_declare_layout_change,
5014 .do_layout_change = lod_layout_change,
5018 * Implementation of dt_body_operations::dbo_read.
5020 * \see dt_body_operations::dbo_read() in the API description for details.
5022 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
5023 struct lu_buf *buf, loff_t *pos)
5025 struct dt_object *next = dt_object_child(dt);
5027 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
5028 S_ISLNK(dt->do_lu.lo_header->loh_attr));
5029 return next->do_body_ops->dbo_read(env, next, buf, pos);
5033 * Implementation of dt_body_operations::dbo_declare_write.
5035 * \see dt_body_operations::dbo_declare_write() in the API description
5038 static ssize_t lod_declare_write(const struct lu_env *env,
5039 struct dt_object *dt,
5040 const struct lu_buf *buf, loff_t pos,
5043 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
5047 * Implementation of dt_body_operations::dbo_write.
5049 * \see dt_body_operations::dbo_write() in the API description for details.
5051 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
5052 const struct lu_buf *buf, loff_t *pos,
5053 struct thandle *th, int iq)
5055 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
5056 S_ISLNK(dt->do_lu.lo_header->loh_attr));
5057 return lod_sub_write(env, dt_object_child(dt), buf, pos, th, iq);
5060 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
5061 __u64 start, __u64 end, struct thandle *th)
5063 if (dt_object_remote(dt))
5066 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
5069 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
5070 __u64 start, __u64 end, struct thandle *th)
5072 if (dt_object_remote(dt))
5075 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
5076 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
5080 * different type of files use the same body_ops because object may be created
5081 * in OUT, where there is no chance to set correct body_ops for each type, so
5082 * body_ops themselves will check file type inside, see lod_read/write/punch for
5085 const struct dt_body_operations lod_body_ops = {
5086 .dbo_read = lod_read,
5087 .dbo_declare_write = lod_declare_write,
5088 .dbo_write = lod_write,
5089 .dbo_declare_punch = lod_declare_punch,
5090 .dbo_punch = lod_punch,
5094 * Implementation of lu_object_operations::loo_object_init.
5096 * The function determines the type and the index of the target device using
5097 * sequence of the object's FID. Then passes control down to the
5098 * corresponding device:
5099 * OSD for the local objects, OSP for remote
5101 * \see lu_object_operations::loo_object_init() in the API description
5104 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
5105 const struct lu_object_conf *conf)
5107 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
5108 struct lu_device *cdev = NULL;
5109 struct lu_object *cobj;
5110 struct lod_tgt_descs *ltd = NULL;
5111 struct lod_tgt_desc *tgt;
5113 int type = LU_SEQ_RANGE_ANY;
5117 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
5119 /* Note: Sometimes, it will Return EAGAIN here, see
5120 * ptrlpc_import_delay_req(), which might confuse
5121 * lu_object_find_at() and make it wait there incorrectly.
5122 * so we convert it to EIO here.*/
5129 if (type == LU_SEQ_RANGE_MDT &&
5130 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
5131 cdev = &lod->lod_child->dd_lu_dev;
5132 } else if (type == LU_SEQ_RANGE_MDT) {
5133 ltd = &lod->lod_mdt_descs;
5135 } else if (type == LU_SEQ_RANGE_OST) {
5136 ltd = &lod->lod_ost_descs;
5143 if (ltd->ltd_tgts_size > idx &&
5144 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
5145 tgt = LTD_TGT(ltd, idx);
5147 LASSERT(tgt != NULL);
5148 LASSERT(tgt->ltd_tgt != NULL);
5150 cdev = &(tgt->ltd_tgt->dd_lu_dev);
5152 lod_putref(lod, ltd);
5155 if (unlikely(cdev == NULL))
5158 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
5159 if (unlikely(cobj == NULL))
5162 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
5164 lu_object_add(lo, cobj);
5171 * Release resources associated with striping.
5173 * If the object is striped (regular or directory), then release
5174 * the stripe objects references and free the ldo_stripe array.
5176 * \param[in] env execution environment
5177 * \param[in] lo object
5179 void lod_object_free_striping(const struct lu_env *env, struct lod_object *lo)
5181 struct lod_layout_component *lod_comp;
5184 if (lo->ldo_stripe != NULL) {
5185 LASSERT(lo->ldo_comp_entries == NULL);
5186 LASSERT(lo->ldo_dir_stripes_allocated > 0);
5188 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5189 if (lo->ldo_stripe[i])
5190 dt_object_put(env, lo->ldo_stripe[i]);
5193 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
5194 OBD_FREE(lo->ldo_stripe, j);
5195 lo->ldo_stripe = NULL;
5196 lo->ldo_dir_stripes_allocated = 0;
5197 lo->ldo_dir_stripe_loaded = 0;
5198 lo->ldo_dir_stripe_count = 0;
5199 } else if (lo->ldo_comp_entries != NULL) {
5200 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5201 /* free lod_layout_component::llc_stripe array */
5202 lod_comp = &lo->ldo_comp_entries[i];
5204 if (lod_comp->llc_stripe == NULL)
5206 LASSERT(lod_comp->llc_stripes_allocated != 0);
5207 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
5208 if (lod_comp->llc_stripe[j] != NULL)
5210 &lod_comp->llc_stripe[j]->do_lu);
5212 OBD_FREE(lod_comp->llc_stripe,
5213 sizeof(struct dt_object *) *
5214 lod_comp->llc_stripes_allocated);
5215 lod_comp->llc_stripe = NULL;
5216 lod_comp->llc_stripes_allocated = 0;
5218 lod_free_comp_entries(lo);
5219 lo->ldo_comp_cached = 0;
5224 * Implementation of lu_object_operations::loo_object_free.
5226 * \see lu_object_operations::loo_object_free() in the API description
5229 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
5231 struct lod_object *lo = lu2lod_obj(o);
5233 /* release all underlying object pinned */
5234 lod_object_free_striping(env, lo);
5236 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
5240 * Implementation of lu_object_operations::loo_object_release.
5242 * \see lu_object_operations::loo_object_release() in the API description
5245 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
5247 /* XXX: shouldn't we release everything here in case if object
5248 * creation failed before? */
5252 * Implementation of lu_object_operations::loo_object_print.
5254 * \see lu_object_operations::loo_object_print() in the API description
5257 static int lod_object_print(const struct lu_env *env, void *cookie,
5258 lu_printer_t p, const struct lu_object *l)
5260 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
5262 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
5265 struct lu_object_operations lod_lu_obj_ops = {
5266 .loo_object_init = lod_object_init,
5267 .loo_object_free = lod_object_free,
5268 .loo_object_release = lod_object_release,
5269 .loo_object_print = lod_object_print,