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, 2017, 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
43 #include <linux/random.h>
46 #include <obd_class.h>
47 #include <obd_support.h>
49 #include <lustre_fid.h>
50 #include <lustre_linkea.h>
51 #include <lustre_lmv.h>
52 #include <uapi/linux/lustre/lustre_param.h>
53 #include <lustre_swab.h>
54 #include <uapi/linux/lustre/lustre_ver.h>
55 #include <lprocfs_status.h>
56 #include <md_object.h>
58 #include "lod_internal.h"
60 static const char dot[] = ".";
61 static const char dotdot[] = "..";
64 * Implementation of dt_index_operations::dio_lookup
66 * Used with regular (non-striped) objects.
68 * \see dt_index_operations::dio_lookup() in the API description for details.
70 static int lod_lookup(const struct lu_env *env, struct dt_object *dt,
71 struct dt_rec *rec, const struct dt_key *key)
73 struct dt_object *next = dt_object_child(dt);
74 return next->do_index_ops->dio_lookup(env, next, rec, key);
78 * Implementation of dt_index_operations::dio_declare_insert.
80 * Used with regular (non-striped) objects.
82 * \see dt_index_operations::dio_declare_insert() in the API description
85 static int lod_declare_insert(const struct lu_env *env, struct dt_object *dt,
86 const struct dt_rec *rec,
87 const struct dt_key *key, struct thandle *th)
89 return lod_sub_declare_insert(env, dt_object_child(dt), rec, key, th);
93 * Implementation of dt_index_operations::dio_insert.
95 * Used with regular (non-striped) objects
97 * \see dt_index_operations::dio_insert() in the API description for details.
99 static int lod_insert(const struct lu_env *env, struct dt_object *dt,
100 const struct dt_rec *rec, const struct dt_key *key,
103 return lod_sub_insert(env, dt_object_child(dt), rec, key, th);
107 * Implementation of dt_index_operations::dio_declare_delete.
109 * Used with regular (non-striped) objects.
111 * \see dt_index_operations::dio_declare_delete() in the API description
114 static int lod_declare_delete(const struct lu_env *env, struct dt_object *dt,
115 const struct dt_key *key, struct thandle *th)
117 return lod_sub_declare_delete(env, dt_object_child(dt), key, th);
121 * Implementation of dt_index_operations::dio_delete.
123 * Used with regular (non-striped) objects.
125 * \see dt_index_operations::dio_delete() in the API description for details.
127 static int lod_delete(const struct lu_env *env, struct dt_object *dt,
128 const struct dt_key *key, struct thandle *th)
130 return lod_sub_delete(env, dt_object_child(dt), key, th);
134 * Implementation of dt_it_ops::init.
136 * Used with regular (non-striped) objects.
138 * \see dt_it_ops::init() in the API description for details.
140 static struct dt_it *lod_it_init(const struct lu_env *env,
141 struct dt_object *dt, __u32 attr)
143 struct dt_object *next = dt_object_child(dt);
144 struct lod_it *it = &lod_env_info(env)->lti_it;
145 struct dt_it *it_next;
147 it_next = next->do_index_ops->dio_it.init(env, next, attr);
151 /* currently we do not use more than one iterator per thread
152 * so we store it in thread info. if at some point we need
153 * more active iterators in a single thread, we can allocate
155 LASSERT(it->lit_obj == NULL);
157 it->lit_it = it_next;
160 return (struct dt_it *)it;
163 #define LOD_CHECK_IT(env, it) \
165 LASSERT((it)->lit_obj != NULL); \
166 LASSERT((it)->lit_it != NULL); \
170 * Implementation of dt_index_operations::dio_it.fini.
172 * Used with regular (non-striped) objects.
174 * \see dt_index_operations::dio_it.fini() in the API description for details.
176 static void lod_it_fini(const struct lu_env *env, struct dt_it *di)
178 struct lod_it *it = (struct lod_it *)di;
180 LOD_CHECK_IT(env, it);
181 it->lit_obj->do_index_ops->dio_it.fini(env, it->lit_it);
183 /* the iterator not in use any more */
189 * Implementation of dt_it_ops::get.
191 * Used with regular (non-striped) objects.
193 * \see dt_it_ops::get() in the API description for details.
195 static int lod_it_get(const struct lu_env *env, struct dt_it *di,
196 const struct dt_key *key)
198 const struct lod_it *it = (const struct lod_it *)di;
200 LOD_CHECK_IT(env, it);
201 return it->lit_obj->do_index_ops->dio_it.get(env, it->lit_it, key);
205 * Implementation of dt_it_ops::put.
207 * Used with regular (non-striped) objects.
209 * \see dt_it_ops::put() in the API description for details.
211 static void lod_it_put(const struct lu_env *env, struct dt_it *di)
213 struct lod_it *it = (struct lod_it *)di;
215 LOD_CHECK_IT(env, it);
216 return it->lit_obj->do_index_ops->dio_it.put(env, it->lit_it);
220 * Implementation of dt_it_ops::next.
222 * Used with regular (non-striped) objects
224 * \see dt_it_ops::next() in the API description for details.
226 static int lod_it_next(const struct lu_env *env, struct dt_it *di)
228 struct lod_it *it = (struct lod_it *)di;
230 LOD_CHECK_IT(env, it);
231 return it->lit_obj->do_index_ops->dio_it.next(env, it->lit_it);
235 * Implementation of dt_it_ops::key.
237 * Used with regular (non-striped) objects.
239 * \see dt_it_ops::key() in the API description for details.
241 static struct dt_key *lod_it_key(const struct lu_env *env,
242 const struct dt_it *di)
244 const struct lod_it *it = (const struct lod_it *)di;
246 LOD_CHECK_IT(env, it);
247 return it->lit_obj->do_index_ops->dio_it.key(env, it->lit_it);
251 * Implementation of dt_it_ops::key_size.
253 * Used with regular (non-striped) objects.
255 * \see dt_it_ops::key_size() in the API description for details.
257 static int lod_it_key_size(const struct lu_env *env, const struct dt_it *di)
259 struct lod_it *it = (struct lod_it *)di;
261 LOD_CHECK_IT(env, it);
262 return it->lit_obj->do_index_ops->dio_it.key_size(env, it->lit_it);
266 * Implementation of dt_it_ops::rec.
268 * Used with regular (non-striped) objects.
270 * \see dt_it_ops::rec() in the API description for details.
272 static int lod_it_rec(const struct lu_env *env, const struct dt_it *di,
273 struct dt_rec *rec, __u32 attr)
275 const struct lod_it *it = (const struct lod_it *)di;
277 LOD_CHECK_IT(env, it);
278 return it->lit_obj->do_index_ops->dio_it.rec(env, it->lit_it, rec,
283 * Implementation of dt_it_ops::rec_size.
285 * Used with regular (non-striped) objects.
287 * \see dt_it_ops::rec_size() in the API description for details.
289 static int lod_it_rec_size(const struct lu_env *env, const struct dt_it *di,
292 const struct lod_it *it = (const struct lod_it *)di;
294 LOD_CHECK_IT(env, it);
295 return it->lit_obj->do_index_ops->dio_it.rec_size(env, it->lit_it,
300 * Implementation of dt_it_ops::store.
302 * Used with regular (non-striped) objects.
304 * \see dt_it_ops::store() in the API description for details.
306 static __u64 lod_it_store(const struct lu_env *env, const struct dt_it *di)
308 const struct lod_it *it = (const struct lod_it *)di;
310 LOD_CHECK_IT(env, it);
311 return it->lit_obj->do_index_ops->dio_it.store(env, it->lit_it);
315 * Implementation of dt_it_ops::load.
317 * Used with regular (non-striped) objects.
319 * \see dt_it_ops::load() in the API description for details.
321 static int lod_it_load(const struct lu_env *env, const struct dt_it *di,
324 const struct lod_it *it = (const struct lod_it *)di;
326 LOD_CHECK_IT(env, it);
327 return it->lit_obj->do_index_ops->dio_it.load(env, it->lit_it, hash);
331 * Implementation of dt_it_ops::key_rec.
333 * Used with regular (non-striped) objects.
335 * \see dt_it_ops::rec() in the API description for details.
337 static int lod_it_key_rec(const struct lu_env *env, const struct dt_it *di,
340 const struct lod_it *it = (const struct lod_it *)di;
342 LOD_CHECK_IT(env, it);
343 return it->lit_obj->do_index_ops->dio_it.key_rec(env, it->lit_it,
347 static struct dt_index_operations lod_index_ops = {
348 .dio_lookup = lod_lookup,
349 .dio_declare_insert = lod_declare_insert,
350 .dio_insert = lod_insert,
351 .dio_declare_delete = lod_declare_delete,
352 .dio_delete = lod_delete,
360 .key_size = lod_it_key_size,
362 .rec_size = lod_it_rec_size,
363 .store = lod_it_store,
365 .key_rec = lod_it_key_rec,
370 * Implementation of dt_it_ops::init.
372 * Used with striped objects. Internally just initializes the iterator
373 * on the first stripe.
375 * \see dt_it_ops::init() in the API description for details.
377 static struct dt_it *lod_striped_it_init(const struct lu_env *env,
378 struct dt_object *dt, __u32 attr)
380 struct lod_object *lo = lod_dt_obj(dt);
381 struct dt_object *next;
382 struct lod_it *it = &lod_env_info(env)->lti_it;
383 struct dt_it *it_next;
386 LASSERT(lo->ldo_dir_stripe_count > 0);
387 next = lo->ldo_stripe[0];
388 LASSERT(next != NULL);
389 LASSERT(next->do_index_ops != NULL);
391 it_next = next->do_index_ops->dio_it.init(env, next, attr);
395 /* currently we do not use more than one iterator per thread
396 * so we store it in thread info. if at some point we need
397 * more active iterators in a single thread, we can allocate
399 LASSERT(it->lit_obj == NULL);
401 it->lit_stripe_index = 0;
403 it->lit_it = it_next;
406 return (struct dt_it *)it;
409 #define LOD_CHECK_STRIPED_IT(env, it, lo) \
411 LASSERT((it)->lit_obj != NULL); \
412 LASSERT((it)->lit_it != NULL); \
413 LASSERT((lo)->ldo_dir_stripe_count > 0); \
414 LASSERT((it)->lit_stripe_index < (lo)->ldo_dir_stripe_count); \
418 * Implementation of dt_it_ops::fini.
420 * Used with striped objects.
422 * \see dt_it_ops::fini() in the API description for details.
424 static void lod_striped_it_fini(const struct lu_env *env, struct dt_it *di)
426 struct lod_it *it = (struct lod_it *)di;
427 struct lod_object *lo = lod_dt_obj(it->lit_obj);
428 struct dt_object *next;
430 /* If lit_it == NULL, then it means the sub_it has been finished,
431 * which only happens in failure cases, see lod_striped_it_next() */
432 if (it->lit_it != NULL) {
433 LOD_CHECK_STRIPED_IT(env, it, lo);
435 next = lo->ldo_stripe[it->lit_stripe_index];
436 LASSERT(next != NULL);
437 LASSERT(next->do_index_ops != NULL);
439 next->do_index_ops->dio_it.fini(env, it->lit_it);
442 /* the iterator not in use any more */
445 it->lit_stripe_index = 0;
449 * Implementation of dt_it_ops::get.
451 * Right now it's not used widely, only to reset the iterator to the
452 * initial position. It should be possible to implement a full version
453 * which chooses a correct stripe to be able to position with any key.
455 * \see dt_it_ops::get() in the API description for details.
457 static int lod_striped_it_get(const struct lu_env *env, struct dt_it *di,
458 const struct dt_key *key)
460 const struct lod_it *it = (const struct lod_it *)di;
461 struct lod_object *lo = lod_dt_obj(it->lit_obj);
462 struct dt_object *next;
465 LOD_CHECK_STRIPED_IT(env, it, lo);
467 next = lo->ldo_stripe[it->lit_stripe_index];
468 LASSERT(next != NULL);
469 LASSERT(next->do_index_ops != NULL);
471 return next->do_index_ops->dio_it.get(env, it->lit_it, key);
475 * Implementation of dt_it_ops::put.
477 * Used with striped objects.
479 * \see dt_it_ops::put() in the API description for details.
481 static void lod_striped_it_put(const struct lu_env *env, struct dt_it *di)
483 struct lod_it *it = (struct lod_it *)di;
484 struct lod_object *lo = lod_dt_obj(it->lit_obj);
485 struct dt_object *next;
487 LOD_CHECK_STRIPED_IT(env, it, lo);
489 next = lo->ldo_stripe[it->lit_stripe_index];
490 LASSERT(next != NULL);
491 LASSERT(next->do_index_ops != NULL);
493 return next->do_index_ops->dio_it.put(env, it->lit_it);
497 * Implementation of dt_it_ops::next.
499 * Used with striped objects. When the end of the current stripe is
500 * reached, the method takes the next stripe's iterator.
502 * \see dt_it_ops::next() in the API description for details.
504 static int lod_striped_it_next(const struct lu_env *env, struct dt_it *di)
506 struct lod_it *it = (struct lod_it *)di;
507 struct lod_object *lo = lod_dt_obj(it->lit_obj);
508 struct dt_object *next;
509 struct dt_it *it_next;
513 LOD_CHECK_STRIPED_IT(env, it, lo);
515 next = lo->ldo_stripe[it->lit_stripe_index];
516 LASSERT(next != NULL);
517 LASSERT(next->do_index_ops != NULL);
519 rc = next->do_index_ops->dio_it.next(env, it->lit_it);
523 if (rc == 0 && it->lit_stripe_index == 0)
526 if (rc == 0 && it->lit_stripe_index > 0) {
527 struct lu_dirent *ent;
529 ent = (struct lu_dirent *)lod_env_info(env)->lti_key;
531 rc = next->do_index_ops->dio_it.rec(env, it->lit_it,
532 (struct dt_rec *)ent,
537 /* skip . and .. for slave stripe */
538 if ((strncmp(ent->lde_name, ".",
539 le16_to_cpu(ent->lde_namelen)) == 0 &&
540 le16_to_cpu(ent->lde_namelen) == 1) ||
541 (strncmp(ent->lde_name, "..",
542 le16_to_cpu(ent->lde_namelen)) == 0 &&
543 le16_to_cpu(ent->lde_namelen) == 2))
549 /* go to next stripe */
550 if (it->lit_stripe_index + 1 >= lo->ldo_dir_stripe_count)
553 it->lit_stripe_index++;
555 next->do_index_ops->dio_it.put(env, it->lit_it);
556 next->do_index_ops->dio_it.fini(env, it->lit_it);
559 next = lo->ldo_stripe[it->lit_stripe_index];
560 LASSERT(next != NULL);
561 rc = next->do_ops->do_index_try(env, next, &dt_directory_features);
565 LASSERT(next->do_index_ops != NULL);
567 it_next = next->do_index_ops->dio_it.init(env, next, it->lit_attr);
568 if (!IS_ERR(it_next)) {
569 it->lit_it = it_next;
572 rc = PTR_ERR(it_next);
579 * Implementation of dt_it_ops::key.
581 * Used with striped objects.
583 * \see dt_it_ops::key() in the API description for details.
585 static struct dt_key *lod_striped_it_key(const struct lu_env *env,
586 const struct dt_it *di)
588 const struct lod_it *it = (const struct lod_it *)di;
589 struct lod_object *lo = lod_dt_obj(it->lit_obj);
590 struct dt_object *next;
592 LOD_CHECK_STRIPED_IT(env, it, lo);
594 next = lo->ldo_stripe[it->lit_stripe_index];
595 LASSERT(next != NULL);
596 LASSERT(next->do_index_ops != NULL);
598 return next->do_index_ops->dio_it.key(env, it->lit_it);
602 * Implementation of dt_it_ops::key_size.
604 * Used with striped objects.
606 * \see dt_it_ops::size() in the API description for details.
608 static int lod_striped_it_key_size(const struct lu_env *env,
609 const struct dt_it *di)
611 struct lod_it *it = (struct lod_it *)di;
612 struct lod_object *lo = lod_dt_obj(it->lit_obj);
613 struct dt_object *next;
615 LOD_CHECK_STRIPED_IT(env, it, lo);
617 next = lo->ldo_stripe[it->lit_stripe_index];
618 LASSERT(next != NULL);
619 LASSERT(next->do_index_ops != NULL);
621 return next->do_index_ops->dio_it.key_size(env, it->lit_it);
625 * Implementation of dt_it_ops::rec.
627 * Used with striped objects.
629 * \see dt_it_ops::rec() in the API description for details.
631 static int lod_striped_it_rec(const struct lu_env *env, const struct dt_it *di,
632 struct dt_rec *rec, __u32 attr)
634 const struct lod_it *it = (const struct lod_it *)di;
635 struct lod_object *lo = lod_dt_obj(it->lit_obj);
636 struct dt_object *next;
638 LOD_CHECK_STRIPED_IT(env, it, lo);
640 next = lo->ldo_stripe[it->lit_stripe_index];
641 LASSERT(next != NULL);
642 LASSERT(next->do_index_ops != NULL);
644 return next->do_index_ops->dio_it.rec(env, it->lit_it, rec, attr);
648 * Implementation of dt_it_ops::rec_size.
650 * Used with striped objects.
652 * \see dt_it_ops::rec_size() in the API description for details.
654 static int lod_striped_it_rec_size(const struct lu_env *env,
655 const struct dt_it *di, __u32 attr)
657 struct lod_it *it = (struct lod_it *)di;
658 struct lod_object *lo = lod_dt_obj(it->lit_obj);
659 struct dt_object *next;
661 LOD_CHECK_STRIPED_IT(env, it, lo);
663 next = lo->ldo_stripe[it->lit_stripe_index];
664 LASSERT(next != NULL);
665 LASSERT(next->do_index_ops != NULL);
667 return next->do_index_ops->dio_it.rec_size(env, it->lit_it, attr);
671 * Implementation of dt_it_ops::store.
673 * Used with striped objects.
675 * \see dt_it_ops::store() in the API description for details.
677 static __u64 lod_striped_it_store(const struct lu_env *env,
678 const struct dt_it *di)
680 const struct lod_it *it = (const struct lod_it *)di;
681 struct lod_object *lo = lod_dt_obj(it->lit_obj);
682 struct dt_object *next;
684 LOD_CHECK_STRIPED_IT(env, it, lo);
686 next = lo->ldo_stripe[it->lit_stripe_index];
687 LASSERT(next != NULL);
688 LASSERT(next->do_index_ops != NULL);
690 return next->do_index_ops->dio_it.store(env, it->lit_it);
694 * Implementation of dt_it_ops::load.
696 * Used with striped objects.
698 * \see dt_it_ops::load() in the API description for details.
700 static int lod_striped_it_load(const struct lu_env *env,
701 const struct dt_it *di, __u64 hash)
703 const struct lod_it *it = (const struct lod_it *)di;
704 struct lod_object *lo = lod_dt_obj(it->lit_obj);
705 struct dt_object *next;
707 LOD_CHECK_STRIPED_IT(env, it, lo);
709 next = lo->ldo_stripe[it->lit_stripe_index];
710 LASSERT(next != NULL);
711 LASSERT(next->do_index_ops != NULL);
713 return next->do_index_ops->dio_it.load(env, it->lit_it, hash);
716 static struct dt_index_operations lod_striped_index_ops = {
717 .dio_lookup = lod_lookup,
718 .dio_declare_insert = lod_declare_insert,
719 .dio_insert = lod_insert,
720 .dio_declare_delete = lod_declare_delete,
721 .dio_delete = lod_delete,
723 .init = lod_striped_it_init,
724 .fini = lod_striped_it_fini,
725 .get = lod_striped_it_get,
726 .put = lod_striped_it_put,
727 .next = lod_striped_it_next,
728 .key = lod_striped_it_key,
729 .key_size = lod_striped_it_key_size,
730 .rec = lod_striped_it_rec,
731 .rec_size = lod_striped_it_rec_size,
732 .store = lod_striped_it_store,
733 .load = lod_striped_it_load,
738 * Append the FID for each shard of the striped directory after the
739 * given LMV EA header.
741 * To simplify striped directory and the consistency verification,
742 * we only store the LMV EA header on disk, for both master object
743 * and slave objects. When someone wants to know the whole LMV EA,
744 * such as client readdir(), we can build the entrie LMV EA on the
745 * MDT side (in RAM) via iterating the sub-directory entries that
746 * are contained in the master object of the stripe directory.
748 * For the master object of the striped directroy, the valid name
749 * for each shard is composed of the ${shard_FID}:${shard_idx}.
751 * There may be holes in the LMV EA if some shards' name entries
752 * are corrupted or lost.
754 * \param[in] env pointer to the thread context
755 * \param[in] lo pointer to the master object of the striped directory
756 * \param[in] buf pointer to the lu_buf which will hold the LMV EA
757 * \param[in] resize whether re-allocate the buffer if it is not big enough
759 * \retval positive size of the LMV EA
760 * \retval 0 for nothing to be loaded
761 * \retval negative error number on failure
763 int lod_load_lmv_shards(const struct lu_env *env, struct lod_object *lo,
764 struct lu_buf *buf, bool resize)
766 struct lu_dirent *ent =
767 (struct lu_dirent *)lod_env_info(env)->lti_key;
768 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
769 struct dt_object *obj = dt_object_child(&lo->ldo_obj);
770 struct lmv_mds_md_v1 *lmv1 = buf->lb_buf;
772 const struct dt_it_ops *iops;
774 __u32 magic = le32_to_cpu(lmv1->lmv_magic);
779 if (magic != LMV_MAGIC_V1)
782 stripes = le32_to_cpu(lmv1->lmv_stripe_count);
786 rc = lmv_mds_md_size(stripes, magic);
790 if (buf->lb_len < lmv1_size) {
799 lu_buf_alloc(buf, lmv1_size);
804 memcpy(buf->lb_buf, tbuf.lb_buf, tbuf.lb_len);
807 if (unlikely(!dt_try_as_dir(env, obj)))
810 memset(&lmv1->lmv_stripe_fids[0], 0, stripes * sizeof(struct lu_fid));
811 iops = &obj->do_index_ops->dio_it;
812 it = iops->init(env, obj, LUDA_64BITHASH);
816 rc = iops->load(env, it, 0);
818 rc = iops->next(env, it);
823 char name[FID_LEN + 2] = "";
828 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
834 fid_le_to_cpu(&fid, &ent->lde_fid);
835 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
836 if (ent->lde_name[0] == '.') {
837 if (ent->lde_namelen == 1)
840 if (ent->lde_namelen == 2 && ent->lde_name[1] == '.')
844 len = snprintf(name, sizeof(name),
845 DFID":", PFID(&ent->lde_fid));
846 /* The ent->lde_name is composed of ${FID}:${index} */
847 if (ent->lde_namelen < len + 1 ||
848 memcmp(ent->lde_name, name, len) != 0) {
849 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
850 "%s: invalid shard name %.*s with the FID "DFID
851 " for the striped directory "DFID", %s\n",
852 lod2obd(lod)->obd_name, ent->lde_namelen,
853 ent->lde_name, PFID(&fid),
854 PFID(lu_object_fid(&obj->do_lu)),
855 lod->lod_lmv_failout ? "failout" : "skip");
857 if (lod->lod_lmv_failout)
865 if (ent->lde_name[len] < '0' ||
866 ent->lde_name[len] > '9') {
867 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
868 "%s: invalid shard name %.*s with the "
869 "FID "DFID" for the striped directory "
871 lod2obd(lod)->obd_name, ent->lde_namelen,
872 ent->lde_name, PFID(&fid),
873 PFID(lu_object_fid(&obj->do_lu)),
874 lod->lod_lmv_failout ?
877 if (lod->lod_lmv_failout)
883 index = index * 10 + ent->lde_name[len++] - '0';
884 } while (len < ent->lde_namelen);
886 if (len == ent->lde_namelen) {
887 /* Out of LMV EA range. */
888 if (index >= stripes) {
889 CERROR("%s: the shard %.*s for the striped "
890 "directory "DFID" is out of the known "
891 "LMV EA range [0 - %u], failout\n",
892 lod2obd(lod)->obd_name, ent->lde_namelen,
894 PFID(lu_object_fid(&obj->do_lu)),
900 /* The slot has been occupied. */
901 if (!fid_is_zero(&lmv1->lmv_stripe_fids[index])) {
905 &lmv1->lmv_stripe_fids[index]);
906 CERROR("%s: both the shard "DFID" and "DFID
907 " for the striped directory "DFID
908 " claim the same LMV EA slot at the "
909 "index %d, failout\n",
910 lod2obd(lod)->obd_name,
911 PFID(&fid0), PFID(&fid),
912 PFID(lu_object_fid(&obj->do_lu)), index);
917 /* stored as LE mode */
918 lmv1->lmv_stripe_fids[index] = ent->lde_fid;
921 rc = iops->next(env, it);
928 RETURN(rc > 0 ? lmv_mds_md_size(stripes, magic) : rc);
932 * Implementation of dt_object_operations::do_index_try.
934 * \see dt_object_operations::do_index_try() in the API description for details.
936 static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
937 const struct dt_index_features *feat)
939 struct lod_object *lo = lod_dt_obj(dt);
940 struct dt_object *next = dt_object_child(dt);
944 LASSERT(next->do_ops);
945 LASSERT(next->do_ops->do_index_try);
947 rc = lod_striping_load(env, lo);
951 rc = next->do_ops->do_index_try(env, next, feat);
955 if (lo->ldo_dir_stripe_count > 0) {
958 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
959 if (dt_object_exists(lo->ldo_stripe[i]) == 0)
961 rc = lo->ldo_stripe[i]->do_ops->do_index_try(env,
962 lo->ldo_stripe[i], feat);
966 dt->do_index_ops = &lod_striped_index_ops;
968 dt->do_index_ops = &lod_index_ops;
975 * Implementation of dt_object_operations::do_read_lock.
977 * \see dt_object_operations::do_read_lock() in the API description for details.
979 static void lod_read_lock(const struct lu_env *env, struct dt_object *dt,
982 dt_read_lock(env, dt_object_child(dt), role);
986 * Implementation of dt_object_operations::do_write_lock.
988 * \see dt_object_operations::do_write_lock() in the API description for
991 static void lod_write_lock(const struct lu_env *env, struct dt_object *dt,
994 dt_write_lock(env, dt_object_child(dt), role);
998 * Implementation of dt_object_operations::do_read_unlock.
1000 * \see dt_object_operations::do_read_unlock() in the API description for
1003 static void lod_read_unlock(const struct lu_env *env, struct dt_object *dt)
1005 dt_read_unlock(env, dt_object_child(dt));
1009 * Implementation of dt_object_operations::do_write_unlock.
1011 * \see dt_object_operations::do_write_unlock() in the API description for
1014 static void lod_write_unlock(const struct lu_env *env, struct dt_object *dt)
1016 dt_write_unlock(env, dt_object_child(dt));
1020 * Implementation of dt_object_operations::do_write_locked.
1022 * \see dt_object_operations::do_write_locked() in the API description for
1025 static int lod_write_locked(const struct lu_env *env, struct dt_object *dt)
1027 return dt_write_locked(env, dt_object_child(dt));
1031 * Implementation of dt_object_operations::do_attr_get.
1033 * \see dt_object_operations::do_attr_get() in the API description for details.
1035 static int lod_attr_get(const struct lu_env *env,
1036 struct dt_object *dt,
1037 struct lu_attr *attr)
1039 /* Note: for striped directory, client will merge attributes
1040 * from all of the sub-stripes see lmv_merge_attr(), and there
1041 * no MDD logic depend on directory nlink/size/time, so we can
1042 * always use master inode nlink and size for now. */
1043 return dt_attr_get(env, dt_object_child(dt), attr);
1046 static inline void lod_adjust_stripe_info(struct lod_layout_component *comp,
1047 struct lov_desc *desc)
1049 if (comp->llc_pattern != LOV_PATTERN_MDT) {
1050 if (!comp->llc_stripe_count)
1051 comp->llc_stripe_count =
1052 desc->ld_default_stripe_count;
1054 if (comp->llc_stripe_size <= 0)
1055 comp->llc_stripe_size = desc->ld_default_stripe_size;
1058 int lod_obj_for_each_stripe(const struct lu_env *env, struct lod_object *lo,
1060 struct lod_obj_stripe_cb_data *data)
1062 struct lod_layout_component *lod_comp;
1066 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1067 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1068 lod_comp = &lo->ldo_comp_entries[i];
1070 if (lod_comp->llc_stripe == NULL)
1073 /* has stripe but not inited yet, this component has been
1074 * declared to be created, but hasn't created yet.
1076 if (!lod_comp_inited(lod_comp))
1079 if (data->locd_comp_skip_cb &&
1080 data->locd_comp_skip_cb(env, lo, i, data))
1083 if (data->locd_comp_cb) {
1084 rc = data->locd_comp_cb(env, lo, i, data);
1089 /* could used just to do sth about component, not each
1092 if (!data->locd_stripe_cb)
1095 LASSERT(lod_comp->llc_stripe_count > 0);
1096 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
1097 struct dt_object *dt = lod_comp->llc_stripe[j];
1101 rc = data->locd_stripe_cb(env, lo, dt, th, i, j, data);
1109 static bool lod_obj_attr_set_comp_skip_cb(const struct lu_env *env,
1110 struct lod_object *lo, int comp_idx,
1111 struct lod_obj_stripe_cb_data *data)
1113 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[comp_idx];
1114 bool skipped = false;
1116 if (!(data->locd_attr->la_valid & LA_LAYOUT_VERSION))
1119 switch (lo->ldo_flr_state) {
1120 case LCM_FL_WRITE_PENDING: {
1123 /* skip stale components */
1124 if (lod_comp->llc_flags & LCME_FL_STALE) {
1129 /* skip valid and overlapping components, therefore any
1130 * attempts to write overlapped components will never succeed
1131 * because client will get EINPROGRESS. */
1132 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1136 if (lo->ldo_comp_entries[i].llc_flags & LCME_FL_STALE)
1139 if (lu_extent_is_overlapped(&lod_comp->llc_extent,
1140 &lo->ldo_comp_entries[i].llc_extent)) {
1148 LASSERTF(0, "impossible: %d\n", lo->ldo_flr_state);
1149 case LCM_FL_SYNC_PENDING:
1153 CDEBUG(D_LAYOUT, DFID": %s to set component %x to version: %u\n",
1154 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
1155 skipped ? "skipped" : "chose", lod_comp->llc_id,
1156 data->locd_attr->la_layout_version);
1162 lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
1163 struct dt_object *dt, struct thandle *th,
1164 int comp_idx, int stripe_idx,
1165 struct lod_obj_stripe_cb_data *data)
1167 if (data->locd_declare)
1168 return lod_sub_declare_attr_set(env, dt, data->locd_attr, th);
1170 if (data->locd_attr->la_valid & LA_LAYOUT_VERSION) {
1171 CDEBUG(D_LAYOUT, DFID": set layout version: %u, comp_idx: %d\n",
1172 PFID(lu_object_fid(&dt->do_lu)),
1173 data->locd_attr->la_layout_version, comp_idx);
1176 return lod_sub_attr_set(env, dt, data->locd_attr, th);
1180 * Implementation of dt_object_operations::do_declare_attr_set.
1182 * If the object is striped, then apply the changes to all the stripes.
1184 * \see dt_object_operations::do_declare_attr_set() in the API description
1187 static int lod_declare_attr_set(const struct lu_env *env,
1188 struct dt_object *dt,
1189 const struct lu_attr *attr,
1192 struct dt_object *next = dt_object_child(dt);
1193 struct lod_object *lo = lod_dt_obj(dt);
1198 * declare setattr on the local object
1200 rc = lod_sub_declare_attr_set(env, next, attr, th);
1204 /* osp_declare_attr_set() ignores all attributes other than
1205 * UID, GID, PROJID, and size, and osp_attr_set() ignores all
1206 * but UID, GID and PROJID. Declaration of size attr setting
1207 * happens through lod_declare_init_size(), and not through
1208 * this function. Therefore we need not load striping unless
1209 * ownership is changing. This should save memory and (we hope)
1210 * speed up rename().
1212 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1213 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1216 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1219 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID | LA_MODE |
1220 LA_ATIME | LA_MTIME | LA_CTIME |
1225 * load striping information, notice we don't do this when object
1226 * is being initialized as we don't need this information till
1227 * few specific cases like destroy, chown
1229 rc = lod_striping_load(env, lo);
1233 if (!lod_obj_is_striped(dt))
1237 * if object is striped declare changes on the stripes
1239 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1240 LASSERT(lo->ldo_stripe);
1241 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1242 if (lo->ldo_stripe[i] == NULL)
1244 rc = lod_sub_declare_attr_set(env, lo->ldo_stripe[i],
1250 struct lod_obj_stripe_cb_data data = { { 0 } };
1252 data.locd_attr = attr;
1253 data.locd_declare = true;
1254 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1255 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1261 if (!dt_object_exists(next) || dt_object_remote(next) ||
1262 !S_ISREG(attr->la_mode))
1265 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1266 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
1270 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) ||
1271 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1272 struct lod_thread_info *info = lod_env_info(env);
1273 struct lu_buf *buf = &info->lti_buf;
1275 buf->lb_buf = info->lti_ea_store;
1276 buf->lb_len = info->lti_ea_store_size;
1277 rc = lod_sub_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1278 LU_XATTR_REPLACE, th);
1285 * Implementation of dt_object_operations::do_attr_set.
1287 * If the object is striped, then apply the changes to all or subset of
1288 * the stripes depending on the object type and specific attributes.
1290 * \see dt_object_operations::do_attr_set() in the API description for details.
1292 static int lod_attr_set(const struct lu_env *env,
1293 struct dt_object *dt,
1294 const struct lu_attr *attr,
1297 struct dt_object *next = dt_object_child(dt);
1298 struct lod_object *lo = lod_dt_obj(dt);
1303 * apply changes to the local object
1305 rc = lod_sub_attr_set(env, next, attr, th);
1309 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1310 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1313 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1316 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE | LA_PROJID |
1317 LA_ATIME | LA_MTIME | LA_CTIME |
1322 /* FIXME: a tricky case in the code path of mdd_layout_change():
1323 * the in-memory striping information has been freed in lod_xattr_set()
1324 * due to layout change. It has to load stripe here again. It only
1325 * changes flags of layout so declare_attr_set() is still accurate */
1326 rc = lod_striping_load(env, lo);
1330 if (!lod_obj_is_striped(dt))
1334 * if object is striped, apply changes to all the stripes
1336 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1337 LASSERT(lo->ldo_stripe);
1338 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1339 if (unlikely(lo->ldo_stripe[i] == NULL))
1342 if ((dt_object_exists(lo->ldo_stripe[i]) == 0))
1345 rc = lod_sub_attr_set(env, lo->ldo_stripe[i], attr, th);
1350 struct lod_obj_stripe_cb_data data = { { 0 } };
1352 data.locd_attr = attr;
1353 data.locd_declare = false;
1354 data.locd_comp_skip_cb = lod_obj_attr_set_comp_skip_cb;
1355 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1356 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1362 if (!dt_object_exists(next) || dt_object_remote(next) ||
1363 !S_ISREG(attr->la_mode))
1366 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1367 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
1371 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE)) {
1372 struct lod_thread_info *info = lod_env_info(env);
1373 struct lu_buf *buf = &info->lti_buf;
1374 struct ost_id *oi = &info->lti_ostid;
1375 struct lu_fid *fid = &info->lti_fid;
1376 struct lov_mds_md_v1 *lmm;
1377 struct lov_ost_data_v1 *objs;
1380 rc = lod_get_lov_ea(env, lo);
1384 buf->lb_buf = info->lti_ea_store;
1385 buf->lb_len = info->lti_ea_store_size;
1386 lmm = info->lti_ea_store;
1387 magic = le32_to_cpu(lmm->lmm_magic);
1388 if (magic == LOV_MAGIC_COMP_V1) {
1389 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1390 struct lov_comp_md_entry_v1 *lcme =
1391 &lcm->lcm_entries[0];
1393 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
1394 magic = le32_to_cpu(lmm->lmm_magic);
1397 if (magic == LOV_MAGIC_V1)
1398 objs = &(lmm->lmm_objects[0]);
1400 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1401 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1402 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1404 fid_to_ostid(fid, oi);
1405 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1407 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1408 LU_XATTR_REPLACE, th);
1409 } else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1410 struct lod_thread_info *info = lod_env_info(env);
1411 struct lu_buf *buf = &info->lti_buf;
1412 struct lov_comp_md_v1 *lcm;
1413 struct lov_comp_md_entry_v1 *lcme;
1415 rc = lod_get_lov_ea(env, lo);
1419 buf->lb_buf = info->lti_ea_store;
1420 buf->lb_len = info->lti_ea_store_size;
1422 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
1425 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1426 lcme = &lcm->lcm_entries[0];
1427 le64_add_cpu(&lcme->lcme_extent.e_start, 1);
1428 le64_add_cpu(&lcme->lcme_extent.e_end, -1);
1430 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1431 LU_XATTR_REPLACE, th);
1438 * Implementation of dt_object_operations::do_xattr_get.
1440 * If LOV EA is requested from the root object and it's not
1441 * found, then return default striping for the filesystem.
1443 * \see dt_object_operations::do_xattr_get() in the API description for details.
1445 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1446 struct lu_buf *buf, const char *name)
1448 struct lod_thread_info *info = lod_env_info(env);
1449 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1454 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1455 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1456 struct lmv_mds_md_v1 *lmv1;
1459 if (rc > (typeof(rc))sizeof(*lmv1))
1462 if (rc < (typeof(rc))sizeof(*lmv1))
1463 RETURN(rc = rc > 0 ? -EINVAL : rc);
1465 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1466 CLASSERT(sizeof(*lmv1) <= sizeof(info->lti_key));
1468 info->lti_buf.lb_buf = info->lti_key;
1469 info->lti_buf.lb_len = sizeof(*lmv1);
1470 rc = dt_xattr_get(env, dt_object_child(dt),
1471 &info->lti_buf, name);
1472 if (unlikely(rc != sizeof(*lmv1)))
1473 RETURN(rc = rc > 0 ? -EINVAL : rc);
1475 lmv1 = info->lti_buf.lb_buf;
1476 /* The on-disk LMV EA only contains header, but the
1477 * returned LMV EA size should contain the space for
1478 * the FIDs of all shards of the striped directory. */
1479 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1480 rc = lmv_mds_md_size(
1481 le32_to_cpu(lmv1->lmv_stripe_count),
1484 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1488 RETURN(rc = rc1 != 0 ? rc1 : rc);
1491 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1495 * XXX: Only used by lfsck
1497 * lod returns default striping on the real root of the device
1498 * this is like the root stores default striping for the whole
1499 * filesystem. historically we've been using a different approach
1500 * and store it in the config.
1502 dt_root_get(env, dev->lod_child, &info->lti_fid);
1503 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1505 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1506 struct lov_user_md *lum = buf->lb_buf;
1507 struct lov_desc *desc = &dev->lod_desc;
1509 if (buf->lb_buf == NULL) {
1511 } else if (buf->lb_len >= sizeof(*lum)) {
1512 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1513 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1514 lmm_oi_set_id(&lum->lmm_oi, 0);
1515 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1516 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1517 lum->lmm_stripe_size = cpu_to_le32(
1518 desc->ld_default_stripe_size);
1519 lum->lmm_stripe_count = cpu_to_le16(
1520 desc->ld_default_stripe_count);
1521 lum->lmm_stripe_offset = cpu_to_le16(
1522 desc->ld_default_stripe_offset);
1535 * Checks that the magic of the stripe is sane.
1537 * \param[in] lod lod device
1538 * \param[in] lum a buffer storing LMV EA to verify
1540 * \retval 0 if the EA is sane
1541 * \retval negative otherwise
1543 static int lod_verify_md_striping(struct lod_device *lod,
1544 const struct lmv_user_md_v1 *lum)
1546 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1547 CERROR("%s: invalid lmv_user_md: magic = %x, "
1548 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1549 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1550 (int)le32_to_cpu(lum->lum_stripe_offset),
1551 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1559 * Initialize LMV EA for a slave.
1561 * Initialize slave's LMV EA from the master's LMV EA.
1563 * \param[in] master_lmv a buffer containing master's EA
1564 * \param[out] slave_lmv a buffer where slave's EA will be stored
1567 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1568 const struct lmv_mds_md_v1 *master_lmv)
1570 *slave_lmv = *master_lmv;
1571 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1577 * Generate LMV EA from the object passed as \a dt. The object must have
1578 * the stripes created and initialized.
1580 * \param[in] env execution environment
1581 * \param[in] dt object
1582 * \param[out] lmv_buf buffer storing generated LMV EA
1584 * \retval 0 on success
1585 * \retval negative if failed
1587 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1588 struct lu_buf *lmv_buf)
1590 struct lod_thread_info *info = lod_env_info(env);
1591 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1592 struct lod_object *lo = lod_dt_obj(dt);
1593 struct lmv_mds_md_v1 *lmm1;
1595 int type = LU_SEQ_RANGE_ANY;
1600 LASSERT(lo->ldo_dir_striped != 0);
1601 LASSERT(lo->ldo_dir_stripe_count > 0);
1602 stripe_count = lo->ldo_dir_stripe_count;
1603 /* Only store the LMV EA heahder on the disk. */
1604 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1605 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1609 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1612 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1613 memset(lmm1, 0, sizeof(*lmm1));
1614 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1615 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1616 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1617 if (lo->ldo_dir_hash_type & LMV_HASH_FLAG_MIGRATION) {
1618 lmm1->lmv_migrate_hash = cpu_to_le32(lo->ldo_dir_migrate_hash);
1619 lmm1->lmv_migrate_offset =
1620 cpu_to_le32(lo->ldo_dir_migrate_offset);
1622 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1627 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1628 lmv_buf->lb_buf = info->lti_ea_store;
1629 lmv_buf->lb_len = sizeof(*lmm1);
1635 * Create in-core represenation for a striped directory.
1637 * Parse the buffer containing LMV EA and instantiate LU objects
1638 * representing the stripe objects. The pointers to the objects are
1639 * stored in ldo_stripe field of \a lo. This function is used when
1640 * we need to access an already created object (i.e. load from a disk).
1642 * \param[in] env execution environment
1643 * \param[in] lo lod object
1644 * \param[in] buf buffer containing LMV EA
1646 * \retval 0 on success
1647 * \retval negative if failed
1649 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1650 const struct lu_buf *buf)
1652 struct lod_thread_info *info = lod_env_info(env);
1653 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1654 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1655 struct dt_object **stripe;
1656 union lmv_mds_md *lmm = buf->lb_buf;
1657 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1658 struct lu_fid *fid = &info->lti_fid;
1663 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1665 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1666 lo->ldo_dir_slave_stripe = 1;
1670 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1673 if (le32_to_cpu(lmv1->lmv_stripe_count) < 1)
1676 LASSERT(lo->ldo_stripe == NULL);
1677 OBD_ALLOC(stripe, sizeof(stripe[0]) *
1678 (le32_to_cpu(lmv1->lmv_stripe_count)));
1682 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1683 struct dt_device *tgt_dt;
1684 struct dt_object *dto;
1685 int type = LU_SEQ_RANGE_ANY;
1688 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1689 if (!fid_is_sane(fid))
1690 GOTO(out, rc = -ESTALE);
1692 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1696 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1697 tgt_dt = lod->lod_child;
1699 struct lod_tgt_desc *tgt;
1701 tgt = LTD_TGT(ltd, idx);
1703 GOTO(out, rc = -ESTALE);
1704 tgt_dt = tgt->ltd_tgt;
1707 dto = dt_locate_at(env, tgt_dt, fid,
1708 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1711 GOTO(out, rc = PTR_ERR(dto));
1716 lo->ldo_stripe = stripe;
1717 lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1718 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1720 lod_striping_free_nolock(env, lo);
1726 * Declare create a striped directory.
1728 * Declare creating a striped directory with a given stripe pattern on the
1729 * specified MDTs. A striped directory is represented as a regular directory
1730 * - an index listing all the stripes. The stripes point back to the master
1731 * object with ".." and LinkEA. The master object gets LMV EA which
1732 * identifies it as a striped directory. The function allocates FIDs
1735 * \param[in] env execution environment
1736 * \param[in] dt object
1737 * \param[in] attr attributes to initialize the objects with
1738 * \param[in] dof type of objects to be created
1739 * \param[in] th transaction handle
1741 * \retval 0 on success
1742 * \retval negative if failed
1744 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1745 struct dt_object *dt,
1746 struct lu_attr *attr,
1747 struct dt_object_format *dof,
1750 struct lod_thread_info *info = lod_env_info(env);
1751 struct lu_buf lmv_buf;
1752 struct lu_buf slave_lmv_buf;
1753 struct lmv_mds_md_v1 *lmm;
1754 struct lmv_mds_md_v1 *slave_lmm = NULL;
1755 struct dt_insert_rec *rec = &info->lti_dt_rec;
1756 struct lod_object *lo = lod_dt_obj(dt);
1761 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1764 lmm = lmv_buf.lb_buf;
1766 OBD_ALLOC_PTR(slave_lmm);
1767 if (slave_lmm == NULL)
1768 GOTO(out, rc = -ENOMEM);
1770 lod_prep_slave_lmv_md(slave_lmm, lmm);
1771 slave_lmv_buf.lb_buf = slave_lmm;
1772 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1774 if (!dt_try_as_dir(env, dt_object_child(dt)))
1775 GOTO(out, rc = -EINVAL);
1777 rec->rec_type = S_IFDIR;
1778 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1779 struct dt_object *dto = lo->ldo_stripe[i];
1780 char *stripe_name = info->lti_key;
1781 struct lu_name *sname;
1782 struct linkea_data ldata = { NULL };
1783 struct lu_buf linkea_buf;
1785 rc = lod_sub_declare_create(env, dto, attr, NULL, dof, th);
1789 if (!dt_try_as_dir(env, dto))
1790 GOTO(out, rc = -EINVAL);
1792 rc = lod_sub_declare_ref_add(env, dto, th);
1796 rec->rec_fid = lu_object_fid(&dto->do_lu);
1797 rc = lod_sub_declare_insert(env, dto,
1798 (const struct dt_rec *)rec,
1799 (const struct dt_key *)dot, th);
1803 /* master stripe FID will be put to .. */
1804 rec->rec_fid = lu_object_fid(&dt->do_lu);
1805 rc = lod_sub_declare_insert(env, dto,
1806 (const struct dt_rec *)rec,
1807 (const struct dt_key *)dotdot, th);
1811 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1812 cfs_fail_val != i) {
1813 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1815 slave_lmm->lmv_master_mdt_index =
1818 slave_lmm->lmv_master_mdt_index =
1820 rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
1821 XATTR_NAME_LMV, 0, th);
1826 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1828 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1829 PFID(lu_object_fid(&dto->do_lu)), i + 1);
1831 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1832 PFID(lu_object_fid(&dto->do_lu)), i);
1834 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
1835 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1836 sname, lu_object_fid(&dt->do_lu));
1840 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1841 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1842 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
1843 XATTR_NAME_LINK, 0, th);
1847 rec->rec_fid = lu_object_fid(&dto->do_lu);
1848 rc = lod_sub_declare_insert(env, dt_object_child(dt),
1849 (const struct dt_rec *)rec,
1850 (const struct dt_key *)stripe_name,
1855 rc = lod_sub_declare_ref_add(env, dt_object_child(dt), th);
1860 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
1861 &lmv_buf, XATTR_NAME_LMV, 0, th);
1865 if (slave_lmm != NULL)
1866 OBD_FREE_PTR(slave_lmm);
1871 static int lod_prep_md_striped_create(const struct lu_env *env,
1872 struct dt_object *dt,
1873 struct lu_attr *attr,
1874 const struct lmv_user_md_v1 *lum,
1875 struct dt_object_format *dof,
1878 struct lod_thread_info *info = lod_env_info(env);
1879 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1880 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1881 struct lod_object *lo = lod_dt_obj(dt);
1882 struct dt_object **stripe;
1889 bool is_specific = false;
1892 /* The lum has been verifed in lod_verify_md_striping */
1893 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1894 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
1896 stripe_count = lo->ldo_dir_stripe_count;
1898 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
1899 if (idx_array == NULL)
1902 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_count);
1904 GOTO(out_free, rc = -ENOMEM);
1906 /* Start index must be the master MDT */
1907 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1908 idx_array[0] = master_index;
1909 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
1911 for (i = 1; i < stripe_count; i++)
1912 idx_array[i] = le32_to_cpu(lum->lum_objects[i].lum_mds);
1915 for (i = 0; i < stripe_count; i++) {
1916 struct lod_tgt_desc *tgt = NULL;
1917 struct dt_object *dto;
1918 struct lu_fid fid = { 0 };
1920 struct lu_object_conf conf = { 0 };
1921 struct dt_device *tgt_dt = NULL;
1923 /* Try to find next avaible target */
1925 for (j = 0; j < lod->lod_remote_mdt_count;
1926 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
1927 bool already_allocated = false;
1930 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
1931 idx, lod->lod_remote_mdt_count + 1, i);
1933 if (likely(!is_specific &&
1934 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
1935 /* check whether the idx already exists
1936 * in current allocated array */
1937 for (k = 0; k < i; k++) {
1938 if (idx_array[k] == idx) {
1939 already_allocated = true;
1944 if (already_allocated)
1948 /* Sigh, this index is not in the bitmap, let's check
1949 * next available target */
1950 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) &&
1951 idx != master_index)
1954 if (idx == master_index) {
1955 /* Allocate the FID locally */
1956 rc = obd_fid_alloc(env, lod->lod_child_exp,
1960 tgt_dt = lod->lod_child;
1964 /* check the status of the OSP */
1965 tgt = LTD_TGT(ltd, idx);
1969 tgt_dt = tgt->ltd_tgt;
1970 rc = dt_statfs(env, tgt_dt, &info->lti_osfs);
1972 /* this OSP doesn't feel well */
1977 rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
1986 /* Can not allocate more stripes */
1987 if (j == lod->lod_remote_mdt_count) {
1988 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
1989 lod2obd(lod)->obd_name, stripe_count, i);
1993 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
1994 idx, i, PFID(&fid));
1996 /* Set the start index for next stripe allocation */
1997 if (!is_specific && i < stripe_count - 1) {
1999 * for large dir test, put all other slaves on one
2000 * remote MDT, otherwise we may save too many local
2001 * slave locks which will exceed RS_MAX_LOCKS.
2003 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)))
2005 idx_array[i + 1] = (idx + 1) %
2006 (lod->lod_remote_mdt_count + 1);
2008 /* tgt_dt and fid must be ready after search avaible OSP
2009 * in the above loop */
2010 LASSERT(tgt_dt != NULL);
2011 LASSERT(fid_is_sane(&fid));
2012 conf.loc_flags = LOC_F_NEW;
2013 dto = dt_locate_at(env, tgt_dt, &fid,
2014 dt->do_lu.lo_dev->ld_site->ls_top_dev,
2017 GOTO(out_put, rc = PTR_ERR(dto));
2021 lo->ldo_dir_striped = 1;
2022 lo->ldo_stripe = stripe;
2023 lo->ldo_dir_stripe_count = i;
2024 lo->ldo_dir_stripes_allocated = stripe_count;
2026 lo->ldo_dir_stripe_loaded = 1;
2028 if (lo->ldo_dir_stripe_count == 0)
2029 GOTO(out_put, rc = -ENOSPC);
2031 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2037 for (i = 0; i < stripe_count; i++)
2038 if (stripe[i] != NULL)
2039 dt_object_put(env, stripe[i]);
2040 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_count);
2041 lo->ldo_dir_stripe_count = 0;
2042 lo->ldo_dir_stripes_allocated = 0;
2043 lo->ldo_stripe = NULL;
2047 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
2053 * Declare create striped md object.
2055 * The function declares intention to create a striped directory. This is a
2056 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2057 * is to verify pattern \a lum_buf is good. Check that function for the details.
2059 * \param[in] env execution environment
2060 * \param[in] dt object
2061 * \param[in] attr attributes to initialize the objects with
2062 * \param[in] lum_buf a pattern specifying the number of stripes and
2064 * \param[in] dof type of objects to be created
2065 * \param[in] th transaction handle
2067 * \retval 0 on success
2068 * \retval negative if failed
2071 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2072 struct dt_object *dt,
2073 struct lu_attr *attr,
2074 const struct lu_buf *lum_buf,
2075 struct dt_object_format *dof,
2078 struct lod_object *lo = lod_dt_obj(dt);
2079 struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
2083 LASSERT(lum != NULL);
2085 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
2086 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
2087 (int)le32_to_cpu(lum->lum_stripe_offset));
2089 if (lo->ldo_dir_stripe_count == 0)
2092 /* prepare dir striped objects */
2093 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2095 /* failed to create striping, let's reset
2096 * config so that others don't get confused */
2097 lod_striping_free(env, lo);
2105 * Append source stripes after target stripes for migrating directory. NB, we
2106 * only need to declare this, the append is done inside lod_xattr_set_lmv().
2108 * \param[in] env execution environment
2109 * \param[in] dt target object
2110 * \param[in] buf LMV buf which contains source stripe fids
2111 * \param[in] th transaction handle
2113 * \retval 0 on success
2114 * \retval negative if failed
2116 static int lod_dir_declare_layout_add(const struct lu_env *env,
2117 struct dt_object *dt,
2118 const struct lu_buf *buf,
2121 struct lod_thread_info *info = lod_env_info(env);
2122 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
2123 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
2124 struct lod_object *lo = lod_dt_obj(dt);
2125 struct dt_object *next = dt_object_child(dt);
2126 struct dt_object_format *dof = &info->lti_format;
2127 struct lmv_mds_md_v1 *lmv = buf->lb_buf;
2128 struct dt_object **stripe;
2129 __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
2130 struct lu_fid *fid = &info->lti_fid;
2131 struct lod_tgt_desc *tgt;
2132 struct dt_object *dto;
2133 struct dt_device *tgt_dt;
2134 int type = LU_SEQ_RANGE_ANY;
2135 struct dt_insert_rec *rec = &info->lti_dt_rec;
2136 char *stripe_name = info->lti_key;
2137 struct lu_name *sname;
2138 struct linkea_data ldata = { NULL };
2139 struct lu_buf linkea_buf;
2146 if (le32_to_cpu(lmv->lmv_magic) != LMV_MAGIC_V1)
2149 if (stripe_count == 0)
2152 dof->dof_type = DFT_DIR;
2155 sizeof(*stripe) * (lo->ldo_dir_stripe_count + stripe_count));
2159 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
2160 stripe[i] = lo->ldo_stripe[i];
2162 for (i = 0; i < stripe_count; i++) {
2164 &lmv->lmv_stripe_fids[i]);
2165 if (!fid_is_sane(fid))
2166 GOTO(out, rc = -ESTALE);
2168 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
2172 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
2173 tgt_dt = lod->lod_child;
2175 tgt = LTD_TGT(ltd, idx);
2177 GOTO(out, rc = -ESTALE);
2178 tgt_dt = tgt->ltd_tgt;
2181 dto = dt_locate_at(env, tgt_dt, fid,
2182 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
2185 GOTO(out, rc = PTR_ERR(dto));
2187 stripe[i + lo->ldo_dir_stripe_count] = dto;
2189 if (!dt_try_as_dir(env, dto))
2190 GOTO(out, rc = -ENOTDIR);
2192 rc = lod_sub_declare_ref_add(env, dto, th);
2196 rc = lod_sub_declare_insert(env, dto,
2197 (const struct dt_rec *)rec,
2198 (const struct dt_key *)dot, th);
2202 rc = lod_sub_declare_insert(env, dto,
2203 (const struct dt_rec *)rec,
2204 (const struct dt_key *)dotdot, th);
2208 rc = lod_sub_declare_xattr_set(env, dto, buf,
2209 XATTR_NAME_LMV, 0, th);
2213 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
2214 PFID(lu_object_fid(&dto->do_lu)),
2215 i + lo->ldo_dir_stripe_count);
2217 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
2218 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
2219 sname, lu_object_fid(&dt->do_lu));
2223 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
2224 linkea_buf.lb_len = ldata.ld_leh->leh_len;
2225 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
2226 XATTR_NAME_LINK, 0, th);
2230 rc = lod_sub_declare_insert(env, next,
2231 (const struct dt_rec *)rec,
2232 (const struct dt_key *)stripe_name,
2237 rc = lod_sub_declare_ref_add(env, next, th);
2243 OBD_FREE(lo->ldo_stripe,
2244 sizeof(*stripe) * lo->ldo_dir_stripes_allocated);
2245 lo->ldo_stripe = stripe;
2246 lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count;
2247 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type);
2248 lo->ldo_dir_stripe_count += stripe_count;
2249 lo->ldo_dir_stripes_allocated += stripe_count;
2250 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_MIGRATION;
2254 i = lo->ldo_dir_stripe_count;
2255 while (i < lo->ldo_dir_stripe_count + stripe_count && stripe[i])
2256 dt_object_put(env, stripe[i++]);
2259 sizeof(*stripe) * (stripe_count + lo->ldo_dir_stripe_count));
2263 static int lod_dir_declare_layout_delete(const struct lu_env *env,
2264 struct dt_object *dt,
2265 const struct lu_buf *buf,
2268 struct lod_thread_info *info = lod_env_info(env);
2269 struct lod_object *lo = lod_dt_obj(dt);
2270 struct dt_object *next = dt_object_child(dt);
2271 struct lmv_user_md *lmu = buf->lb_buf;
2272 __u32 final_stripe_count;
2273 char *stripe_name = info->lti_key;
2274 struct dt_object *dto;
2281 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
2282 if (final_stripe_count >= lo->ldo_dir_stripe_count)
2285 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
2286 dto = lo->ldo_stripe[i];
2289 if (!dt_try_as_dir(env, dto))
2292 rc = lod_sub_declare_delete(env, dto,
2293 (const struct dt_key *)dot, th);
2297 rc = lod_sub_declare_ref_del(env, dto, th);
2301 rc = lod_sub_declare_delete(env, dto,
2302 (const struct dt_key *)dotdot, th);
2306 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2307 PFID(lu_object_fid(&dto->do_lu)), i);
2309 rc = lod_sub_declare_delete(env, next,
2310 (const struct dt_key *)stripe_name, th);
2314 rc = lod_sub_declare_ref_del(env, next, th);
2323 * delete stripes from dir master object, the lum_stripe_count in argument is
2324 * the final stripe count, the stripes after that will be deleted, NB, they
2325 * are not destroyed, but deleted from it's parent namespace, this function
2326 * will be called in two places:
2327 * 1. mdd_migrate_create() delete stripes from source, and append them to
2329 * 2. mdd_dir_layout_shrink() delete stripes from source, and destroy them.
2331 static int lod_dir_layout_delete(const struct lu_env *env,
2332 struct dt_object *dt,
2333 const struct lu_buf *buf,
2336 struct lod_thread_info *info = lod_env_info(env);
2337 struct lod_object *lo = lod_dt_obj(dt);
2338 struct dt_object *next = dt_object_child(dt);
2339 struct lmv_user_md *lmu = buf->lb_buf;
2340 __u32 final_stripe_count;
2341 char *stripe_name = info->lti_key;
2342 struct dt_object *dto;
2351 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
2352 if (final_stripe_count >= lo->ldo_dir_stripe_count)
2355 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
2356 dto = lo->ldo_stripe[i];
2359 rc = lod_sub_delete(env, dto,
2360 (const struct dt_key *)dotdot, th);
2364 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2365 PFID(lu_object_fid(&dto->do_lu)), i);
2367 rc = lod_sub_delete(env, next,
2368 (const struct dt_key *)stripe_name, th);
2372 rc = lod_sub_ref_del(env, next, th);
2377 lod_striping_free(env, lod_dt_obj(dt));
2383 * Implementation of dt_object_operations::do_declare_xattr_set.
2385 * Used with regular (non-striped) objects. Basically it
2386 * initializes the striping information and applies the
2387 * change to all the stripes.
2389 * \see dt_object_operations::do_declare_xattr_set() in the API description
2392 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2393 struct dt_object *dt,
2394 const struct lu_buf *buf,
2395 const char *name, int fl,
2398 struct dt_object *next = dt_object_child(dt);
2399 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2400 struct lod_object *lo = lod_dt_obj(dt);
2405 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2406 struct lmv_user_md_v1 *lum;
2408 LASSERT(buf != NULL && buf->lb_buf != NULL);
2410 rc = lod_verify_md_striping(d, lum);
2413 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2414 rc = lod_verify_striping(d, lo, buf, false);
2419 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2423 /* Note: Do not set LinkEA on sub-stripes, otherwise
2424 * it will confuse the fid2path process(see mdt_path_current()).
2425 * The linkEA between master and sub-stripes is set in
2426 * lod_xattr_set_lmv(). */
2427 if (strcmp(name, XATTR_NAME_LINK) == 0)
2430 /* set xattr to each stripes, if needed */
2431 rc = lod_striping_load(env, lo);
2435 if (lo->ldo_dir_stripe_count == 0)
2438 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2439 LASSERT(lo->ldo_stripe[i]);
2441 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2451 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2452 struct lod_object *lo,
2453 struct dt_object *dt, struct thandle *th,
2454 int comp_idx, int stripe_idx,
2455 struct lod_obj_stripe_cb_data *data)
2457 struct lod_thread_info *info = lod_env_info(env);
2458 struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
2459 struct filter_fid *ff = &info->lti_ff;
2460 struct lu_buf *buf = &info->lti_buf;
2464 buf->lb_len = sizeof(*ff);
2465 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2473 * locd_buf is set if it's called by dir migration, which doesn't check
2476 if (data->locd_buf) {
2477 memset(ff, 0, sizeof(*ff));
2478 ff->ff_parent = *(struct lu_fid *)data->locd_buf->lb_buf;
2480 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
2482 if (lu_fid_eq(lod_object_fid(lo), &ff->ff_parent) &&
2483 ff->ff_layout.ol_comp_id == comp->llc_id)
2486 memset(ff, 0, sizeof(*ff));
2487 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2490 /* rewrite filter_fid */
2491 ff->ff_parent.f_ver = stripe_idx;
2492 ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
2493 ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
2494 ff->ff_layout.ol_comp_id = comp->llc_id;
2495 ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
2496 ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
2497 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
2499 if (data->locd_declare)
2500 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2501 LU_XATTR_REPLACE, th);
2503 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2504 LU_XATTR_REPLACE, th);
2510 * Reset parent FID on OST object
2512 * Replace parent FID with @dt object FID, which is only called during migration
2513 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2514 * the FID is changed.
2516 * \param[in] env execution environment
2517 * \param[in] dt dt_object whose stripes's parent FID will be reset
2518 * \parem[in] th thandle
2519 * \param[in] declare if it is declare
2521 * \retval 0 if reset succeeds
2522 * \retval negative errno if reset fails
2524 static int lod_replace_parent_fid(const struct lu_env *env,
2525 struct dt_object *dt,
2526 const struct lu_buf *buf,
2527 struct thandle *th, bool declare)
2529 struct lod_object *lo = lod_dt_obj(dt);
2530 struct lod_thread_info *info = lod_env_info(env);
2531 struct filter_fid *ff;
2532 struct lod_obj_stripe_cb_data data = { { 0 } };
2536 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2538 /* set xattr to each stripes, if needed */
2539 rc = lod_striping_load(env, lo);
2543 if (!lod_obj_is_striped(dt))
2546 if (info->lti_ea_store_size < sizeof(*ff)) {
2547 rc = lod_ea_store_resize(info, sizeof(*ff));
2552 data.locd_declare = declare;
2553 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2554 data.locd_buf = buf;
2555 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2560 inline __u16 lod_comp_entry_stripe_count(struct lod_object *lo,
2561 struct lod_layout_component *entry,
2564 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2568 else if (lod_comp_inited(entry))
2569 return entry->llc_stripe_count;
2570 else if ((__u16)-1 == entry->llc_stripe_count)
2571 return lod->lod_desc.ld_tgt_count;
2573 return lod_get_stripe_count(lod, lo, entry->llc_stripe_count);
2576 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2578 int magic, size = 0, i;
2579 struct lod_layout_component *comp_entries;
2581 bool is_composite, is_foreign = false;
2584 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2585 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2587 lo->ldo_def_striping->lds_def_striping_is_composite;
2589 comp_cnt = lo->ldo_comp_cnt;
2590 comp_entries = lo->ldo_comp_entries;
2591 is_composite = lo->ldo_is_composite;
2592 is_foreign = lo->ldo_is_foreign;
2596 return lo->ldo_foreign_lov_size;
2598 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2600 size = sizeof(struct lov_comp_md_v1) +
2601 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2602 LASSERT(size % sizeof(__u64) == 0);
2605 for (i = 0; i < comp_cnt; i++) {
2608 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2609 stripe_count = lod_comp_entry_stripe_count(lo, &comp_entries[i],
2611 if (!is_dir && is_composite)
2612 lod_comp_shrink_stripe_count(&comp_entries[i],
2615 size += lov_user_md_size(stripe_count, magic);
2616 LASSERT(size % sizeof(__u64) == 0);
2622 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2623 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2626 * \param[in] env execution environment
2627 * \param[in] dt dt_object to add components on
2628 * \param[in] buf buffer contains components to be added
2629 * \parem[in] th thandle
2631 * \retval 0 on success
2632 * \retval negative errno on failure
2634 static int lod_declare_layout_add(const struct lu_env *env,
2635 struct dt_object *dt,
2636 const struct lu_buf *buf,
2639 struct lod_thread_info *info = lod_env_info(env);
2640 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2641 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2642 struct dt_object *next = dt_object_child(dt);
2643 struct lov_desc *desc = &d->lod_desc;
2644 struct lod_object *lo = lod_dt_obj(dt);
2645 struct lov_user_md_v3 *v3;
2646 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2648 int i, rc, array_cnt, old_array_cnt;
2651 LASSERT(lo->ldo_is_composite);
2653 if (lo->ldo_flr_state != LCM_FL_NONE)
2656 rc = lod_verify_striping(d, lo, buf, false);
2660 magic = comp_v1->lcm_magic;
2661 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2662 lustre_swab_lov_comp_md_v1(comp_v1);
2663 magic = comp_v1->lcm_magic;
2666 if (magic != LOV_USER_MAGIC_COMP_V1)
2669 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2670 OBD_ALLOC(comp_array, sizeof(*comp_array) * array_cnt);
2671 if (comp_array == NULL)
2674 memcpy(comp_array, lo->ldo_comp_entries,
2675 sizeof(*comp_array) * lo->ldo_comp_cnt);
2677 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2678 struct lov_user_md_v1 *v1;
2679 struct lu_extent *ext;
2681 v1 = (struct lov_user_md *)((char *)comp_v1 +
2682 comp_v1->lcm_entries[i].lcme_offset);
2683 ext = &comp_v1->lcm_entries[i].lcme_extent;
2685 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2686 lod_comp->llc_extent.e_start = ext->e_start;
2687 lod_comp->llc_extent.e_end = ext->e_end;
2688 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2689 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2691 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2692 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2693 lod_adjust_stripe_info(lod_comp, desc);
2695 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2696 v3 = (struct lov_user_md_v3 *) v1;
2697 if (v3->lmm_pool_name[0] != '\0') {
2698 rc = lod_set_pool(&lod_comp->llc_pool,
2706 old_array = lo->ldo_comp_entries;
2707 old_array_cnt = lo->ldo_comp_cnt;
2709 lo->ldo_comp_entries = comp_array;
2710 lo->ldo_comp_cnt = array_cnt;
2712 /* No need to increase layout generation here, it will be increased
2713 * later when generating component ID for the new components */
2715 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2716 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2717 XATTR_NAME_LOV, 0, th);
2719 lo->ldo_comp_entries = old_array;
2720 lo->ldo_comp_cnt = old_array_cnt;
2724 OBD_FREE(old_array, sizeof(*lod_comp) * old_array_cnt);
2726 LASSERT(lo->ldo_mirror_count == 1);
2727 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2732 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2733 lod_comp = &comp_array[i];
2734 if (lod_comp->llc_pool != NULL) {
2735 OBD_FREE(lod_comp->llc_pool,
2736 strlen(lod_comp->llc_pool) + 1);
2737 lod_comp->llc_pool = NULL;
2740 OBD_FREE(comp_array, sizeof(*comp_array) * array_cnt);
2745 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2746 * the '$field' can only be 'flags' now. The xattr value is binary
2747 * lov_comp_md_v1 which contains the component ID(s) and the value of
2748 * the field to be modified.
2750 * \param[in] env execution environment
2751 * \param[in] dt dt_object to be modified
2752 * \param[in] op operation string, like "set.flags"
2753 * \param[in] buf buffer contains components to be set
2754 * \parem[in] th thandle
2756 * \retval 0 on success
2757 * \retval negative errno on failure
2759 static int lod_declare_layout_set(const struct lu_env *env,
2760 struct dt_object *dt,
2761 char *op, const struct lu_buf *buf,
2764 struct lod_layout_component *lod_comp;
2765 struct lod_thread_info *info = lod_env_info(env);
2766 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2767 struct lod_object *lo = lod_dt_obj(dt);
2768 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2771 bool changed = false;
2774 if (strcmp(op, "set.flags") != 0) {
2775 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2776 lod2obd(d)->obd_name, op);
2780 magic = comp_v1->lcm_magic;
2781 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2782 lustre_swab_lov_comp_md_v1(comp_v1);
2783 magic = comp_v1->lcm_magic;
2786 if (magic != LOV_USER_MAGIC_COMP_V1)
2789 if (comp_v1->lcm_entry_count == 0) {
2790 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2791 lod2obd(d)->obd_name);
2795 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2796 __u32 id = comp_v1->lcm_entries[i].lcme_id;
2797 __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
2798 __u32 mirror_flag = flags & LCME_MIRROR_FLAGS;
2799 bool neg = flags & LCME_FL_NEG;
2801 if (flags & LCME_FL_INIT) {
2803 lod_striping_free(env, lo);
2807 flags &= ~(LCME_MIRROR_FLAGS | LCME_FL_NEG);
2808 for (j = 0; j < lo->ldo_comp_cnt; j++) {
2809 lod_comp = &lo->ldo_comp_entries[j];
2811 /* lfs only put one flag in each entry */
2812 if ((flags && id != lod_comp->llc_id) ||
2813 (mirror_flag && mirror_id_of(id) !=
2814 mirror_id_of(lod_comp->llc_id)))
2819 lod_comp->llc_flags &= ~flags;
2821 lod_comp->llc_flags &= ~mirror_flag;
2824 lod_comp->llc_flags |= flags;
2826 lod_comp->llc_flags |= mirror_flag;
2827 if (mirror_flag & LCME_FL_NOSYNC)
2828 lod_comp->llc_timestamp =
2829 ktime_get_real_seconds();
2837 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
2838 lod2obd(d)->obd_name);
2842 lod_obj_inc_layout_gen(lo);
2844 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2845 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
2846 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
2851 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
2852 * and the xattr value is a unique component ID or a special lcme_id.
2854 * \param[in] env execution environment
2855 * \param[in] dt dt_object to be operated on
2856 * \param[in] buf buffer contains component ID or lcme_id
2857 * \parem[in] th thandle
2859 * \retval 0 on success
2860 * \retval negative errno on failure
2862 static int lod_declare_layout_del(const struct lu_env *env,
2863 struct dt_object *dt,
2864 const struct lu_buf *buf,
2867 struct lod_thread_info *info = lod_env_info(env);
2868 struct dt_object *next = dt_object_child(dt);
2869 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2870 struct lod_object *lo = lod_dt_obj(dt);
2871 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2872 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2873 __u32 magic, id, flags, neg_flags = 0;
2877 LASSERT(lo->ldo_is_composite);
2879 if (lo->ldo_flr_state != LCM_FL_NONE)
2882 magic = comp_v1->lcm_magic;
2883 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2884 lustre_swab_lov_comp_md_v1(comp_v1);
2885 magic = comp_v1->lcm_magic;
2888 if (magic != LOV_USER_MAGIC_COMP_V1)
2891 id = comp_v1->lcm_entries[0].lcme_id;
2892 flags = comp_v1->lcm_entries[0].lcme_flags;
2894 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
2895 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
2896 lod2obd(d)->obd_name, id, flags);
2900 if (id != LCME_ID_INVAL && flags != 0) {
2901 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
2902 lod2obd(d)->obd_name);
2906 if (id == LCME_ID_INVAL && !flags) {
2907 CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
2908 lod2obd(d)->obd_name);
2912 if (flags & LCME_FL_NEG) {
2913 neg_flags = flags & ~LCME_FL_NEG;
2917 left = lo->ldo_comp_cnt;
2921 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
2922 struct lod_layout_component *lod_comp;
2924 lod_comp = &lo->ldo_comp_entries[i];
2926 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
2928 else if (flags && !(flags & lod_comp->llc_flags))
2930 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
2933 if (left != (i + 1)) {
2934 CDEBUG(D_LAYOUT, "%s: this deletion will create "
2935 "a hole.\n", lod2obd(d)->obd_name);
2940 /* Mark the component as deleted */
2941 lod_comp->llc_id = LCME_ID_INVAL;
2943 /* Not instantiated component */
2944 if (lod_comp->llc_stripe == NULL)
2947 LASSERT(lod_comp->llc_stripe_count > 0);
2948 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
2949 struct dt_object *obj = lod_comp->llc_stripe[j];
2953 rc = lod_sub_declare_destroy(env, obj, th);
2959 LASSERTF(left >= 0, "left = %d\n", left);
2960 if (left == lo->ldo_comp_cnt) {
2961 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
2962 lod2obd(d)->obd_name, id);
2966 memset(attr, 0, sizeof(*attr));
2967 attr->la_valid = LA_SIZE;
2968 rc = lod_sub_declare_attr_set(env, next, attr, th);
2973 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2974 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2975 XATTR_NAME_LOV, 0, th);
2977 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
2984 * Declare layout add/set/del operations issued by special xattr names:
2986 * XATTR_LUSTRE_LOV.add add component(s) to existing file
2987 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
2988 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
2990 * \param[in] env execution environment
2991 * \param[in] dt object
2992 * \param[in] name name of xattr
2993 * \param[in] buf lu_buf contains xattr value
2994 * \param[in] th transaction handle
2996 * \retval 0 on success
2997 * \retval negative if failed
2999 static int lod_declare_modify_layout(const struct lu_env *env,
3000 struct dt_object *dt,
3002 const struct lu_buf *buf,
3005 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3006 struct lod_object *lo = lod_dt_obj(dt);
3008 int rc, len = strlen(XATTR_LUSTRE_LOV);
3011 LASSERT(dt_object_exists(dt));
3013 if (strlen(name) <= len || name[len] != '.') {
3014 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
3015 lod2obd(d)->obd_name, name);
3020 rc = lod_striping_load(env, lo);
3024 /* the layout to be modified must be a composite layout */
3025 if (!lo->ldo_is_composite) {
3026 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
3027 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3028 GOTO(unlock, rc = -EINVAL);
3031 op = (char *)name + len;
3032 if (strcmp(op, "add") == 0) {
3033 rc = lod_declare_layout_add(env, dt, buf, th);
3034 } else if (strcmp(op, "del") == 0) {
3035 rc = lod_declare_layout_del(env, dt, buf, th);
3036 } else if (strncmp(op, "set", strlen("set")) == 0) {
3037 rc = lod_declare_layout_set(env, dt, op, buf, th);
3039 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
3040 lod2obd(d)->obd_name, name);
3041 GOTO(unlock, rc = -ENOTSUPP);
3045 lod_striping_free(env, lo);
3051 * Convert a plain file lov_mds_md to a composite layout.
3053 * \param[in,out] info the thread info::lti_ea_store buffer contains little
3054 * endian plain file layout
3056 * \retval 0 on success, <0 on failure
3058 static int lod_layout_convert(struct lod_thread_info *info)
3060 struct lov_mds_md *lmm = info->lti_ea_store;
3061 struct lov_mds_md *lmm_save;
3062 struct lov_comp_md_v1 *lcm;
3063 struct lov_comp_md_entry_v1 *lcme;
3069 /* realloc buffer to a composite layout which contains one component */
3070 blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3071 le32_to_cpu(lmm->lmm_magic));
3072 size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
3074 OBD_ALLOC_LARGE(lmm_save, blob_size);
3076 GOTO(out, rc = -ENOMEM);
3078 memcpy(lmm_save, lmm, blob_size);
3080 if (info->lti_ea_store_size < size) {
3081 rc = lod_ea_store_resize(info, size);
3086 lcm = info->lti_ea_store;
3087 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
3088 lcm->lcm_size = cpu_to_le32(size);
3089 lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
3090 lmm_save->lmm_layout_gen));
3091 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
3092 lcm->lcm_entry_count = cpu_to_le16(1);
3093 lcm->lcm_mirror_count = 0;
3095 lcme = &lcm->lcm_entries[0];
3096 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
3097 lcme->lcme_extent.e_start = 0;
3098 lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
3099 lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
3100 lcme->lcme_size = cpu_to_le32(blob_size);
3102 memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
3107 OBD_FREE_LARGE(lmm_save, blob_size);
3112 * Merge layouts to form a mirrored file.
3114 static int lod_declare_layout_merge(const struct lu_env *env,
3115 struct dt_object *dt, const struct lu_buf *mbuf,
3118 struct lod_thread_info *info = lod_env_info(env);
3119 struct lu_buf *buf = &info->lti_buf;
3120 struct lod_object *lo = lod_dt_obj(dt);
3121 struct lov_comp_md_v1 *lcm;
3122 struct lov_comp_md_v1 *cur_lcm;
3123 struct lov_comp_md_v1 *merge_lcm;
3124 struct lov_comp_md_entry_v1 *lcme;
3127 __u16 cur_entry_count;
3128 __u16 merge_entry_count;
3130 __u16 mirror_id = 0;
3135 merge_lcm = mbuf->lb_buf;
3136 if (mbuf->lb_len < sizeof(*merge_lcm))
3139 /* must be an existing layout from disk */
3140 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
3143 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
3145 /* do not allow to merge two mirrored files */
3146 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
3149 /* verify the target buffer */
3150 rc = lod_get_lov_ea(env, lo);
3152 RETURN(rc ? : -ENODATA);
3154 cur_lcm = info->lti_ea_store;
3155 switch (le32_to_cpu(cur_lcm->lcm_magic)) {
3158 rc = lod_layout_convert(info);
3160 case LOV_MAGIC_COMP_V1:
3169 /* info->lti_ea_store could be reallocated in lod_layout_convert() */
3170 cur_lcm = info->lti_ea_store;
3171 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
3173 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
3174 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
3175 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
3178 /* size of new layout */
3179 size = le32_to_cpu(cur_lcm->lcm_size) +
3180 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
3182 memset(buf, 0, sizeof(*buf));
3183 lu_buf_alloc(buf, size);
3184 if (buf->lb_buf == NULL)
3188 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
3190 offset = sizeof(*lcm) +
3191 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
3192 for (i = 0; i < cur_entry_count; i++) {
3193 struct lov_comp_md_entry_v1 *cur_lcme;
3195 lcme = &lcm->lcm_entries[i];
3196 cur_lcme = &cur_lcm->lcm_entries[i];
3198 lcme->lcme_offset = cpu_to_le32(offset);
3199 memcpy((char *)lcm + offset,
3200 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
3201 le32_to_cpu(lcme->lcme_size));
3203 offset += le32_to_cpu(lcme->lcme_size);
3205 if (mirror_count == 1 &&
3206 mirror_id_of(le32_to_cpu(lcme->lcme_id)) == 0) {
3207 /* Add mirror from a non-flr file, create new mirror ID.
3208 * Otherwise, keep existing mirror's component ID, used
3209 * for mirror extension.
3211 id = pflr_id(1, i + 1);
3212 lcme->lcme_id = cpu_to_le32(id);
3215 id = MAX(le32_to_cpu(lcme->lcme_id), id);
3218 mirror_id = mirror_id_of(id) + 1;
3219 for (i = 0; i < merge_entry_count; i++) {
3220 struct lov_comp_md_entry_v1 *merge_lcme;
3222 merge_lcme = &merge_lcm->lcm_entries[i];
3223 lcme = &lcm->lcm_entries[cur_entry_count + i];
3225 *lcme = *merge_lcme;
3226 lcme->lcme_offset = cpu_to_le32(offset);
3228 id = pflr_id(mirror_id, i + 1);
3229 lcme->lcme_id = cpu_to_le32(id);
3231 memcpy((char *)lcm + offset,
3232 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
3233 le32_to_cpu(lcme->lcme_size));
3235 offset += le32_to_cpu(lcme->lcme_size);
3238 /* fixup layout information */
3239 lod_obj_inc_layout_gen(lo);
3240 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3241 lcm->lcm_size = cpu_to_le32(size);
3242 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
3243 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
3244 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NONE)
3245 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
3247 rc = lod_striping_reload(env, lo, buf);
3251 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
3252 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3260 * Split layouts, just set the LOVEA with the layout from mbuf.
3262 static int lod_declare_layout_split(const struct lu_env *env,
3263 struct dt_object *dt, const struct lu_buf *mbuf,
3266 struct lod_object *lo = lod_dt_obj(dt);
3267 struct lov_comp_md_v1 *lcm = mbuf->lb_buf;
3271 lod_obj_inc_layout_gen(lo);
3272 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3274 rc = lod_striping_reload(env, lo, mbuf);
3278 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), mbuf,
3279 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3284 * Implementation of dt_object_operations::do_declare_xattr_set.
3286 * \see dt_object_operations::do_declare_xattr_set() in the API description
3289 * the extension to the API:
3290 * - declaring LOVEA requests striping creation
3291 * - LU_XATTR_REPLACE means layout swap
3293 static int lod_declare_xattr_set(const struct lu_env *env,
3294 struct dt_object *dt,
3295 const struct lu_buf *buf,
3296 const char *name, int fl,
3299 struct dt_object *next = dt_object_child(dt);
3300 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3305 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
3306 if ((S_ISREG(mode) || mode == 0) &&
3307 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE | LU_XATTR_SPLIT)) &&
3308 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3309 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
3311 * this is a request to create object's striping.
3313 * allow to declare predefined striping on a new (!mode) object
3314 * which is supposed to be replay of regular file creation
3315 * (when LOV setting is declared)
3317 * LU_XATTR_REPLACE is set to indicate a layout swap
3319 if (dt_object_exists(dt)) {
3320 rc = dt_attr_get(env, next, attr);
3324 memset(attr, 0, sizeof(*attr));
3325 attr->la_valid = LA_TYPE | LA_MODE;
3326 attr->la_mode = S_IFREG;
3328 rc = lod_declare_striped_create(env, dt, attr, buf, th);
3329 } else if (fl & LU_XATTR_MERGE) {
3330 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3331 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3332 rc = lod_declare_layout_merge(env, dt, buf, th);
3333 } else if (fl & LU_XATTR_SPLIT) {
3334 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3335 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3336 rc = lod_declare_layout_split(env, dt, buf, th);
3337 } else if (S_ISREG(mode) &&
3338 strlen(name) > strlen(XATTR_LUSTRE_LOV) + 1 &&
3339 strncmp(name, XATTR_LUSTRE_LOV,
3340 strlen(XATTR_LUSTRE_LOV)) == 0) {
3342 * this is a request to modify object's striping.
3343 * add/set/del component(s).
3345 if (!dt_object_exists(dt))
3348 rc = lod_declare_modify_layout(env, dt, name, buf, th);
3349 } else if (strncmp(name, XATTR_NAME_LMV, strlen(XATTR_NAME_LMV)) == 0 &&
3350 strlen(name) > strlen(XATTR_NAME_LMV) + 1) {
3351 const char *op = name + strlen(XATTR_NAME_LMV) + 1;
3354 if (strcmp(op, "add") == 0)
3355 rc = lod_dir_declare_layout_add(env, dt, buf, th);
3356 else if (strcmp(op, "del") == 0)
3357 rc = lod_dir_declare_layout_delete(env, dt, buf, th);
3358 else if (strcmp(op, "set") == 0)
3359 rc = lod_sub_declare_xattr_set(env, next, buf,
3360 XATTR_NAME_LMV, fl, th);
3363 } else if (S_ISDIR(mode)) {
3364 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
3365 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3366 rc = lod_replace_parent_fid(env, dt, buf, th, true);
3368 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
3375 * Apply xattr changes to the object.
3377 * Applies xattr changes to the object and the stripes if the latter exist.
3379 * \param[in] env execution environment
3380 * \param[in] dt object
3381 * \param[in] buf buffer pointing to the new value of xattr
3382 * \param[in] name name of xattr
3383 * \param[in] fl flags
3384 * \param[in] th transaction handle
3386 * \retval 0 on success
3387 * \retval negative if failed
3389 static int lod_xattr_set_internal(const struct lu_env *env,
3390 struct dt_object *dt,
3391 const struct lu_buf *buf,
3392 const char *name, int fl,
3395 struct dt_object *next = dt_object_child(dt);
3396 struct lod_object *lo = lod_dt_obj(dt);
3401 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3402 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3405 /* Note: Do not set LinkEA on sub-stripes, otherwise
3406 * it will confuse the fid2path process(see mdt_path_current()).
3407 * The linkEA between master and sub-stripes is set in
3408 * lod_xattr_set_lmv(). */
3409 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
3412 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3413 LASSERT(lo->ldo_stripe[i]);
3415 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
3425 * Delete an extended attribute.
3427 * Deletes specified xattr from the object and the stripes if the latter exist.
3429 * \param[in] env execution environment
3430 * \param[in] dt object
3431 * \param[in] name name of xattr
3432 * \param[in] th transaction handle
3434 * \retval 0 on success
3435 * \retval negative if failed
3437 static int lod_xattr_del_internal(const struct lu_env *env,
3438 struct dt_object *dt,
3439 const char *name, struct thandle *th)
3441 struct dt_object *next = dt_object_child(dt);
3442 struct lod_object *lo = lod_dt_obj(dt);
3447 rc = lod_sub_xattr_del(env, next, name, th);
3448 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3451 if (lo->ldo_dir_stripe_count == 0)
3454 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3455 LASSERT(lo->ldo_stripe[i]);
3457 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3466 * Set default striping on a directory.
3468 * Sets specified striping on a directory object unless it matches the default
3469 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3470 * EA. This striping will be used when regular file is being created in this
3473 * \param[in] env execution environment
3474 * \param[in] dt the striped object
3475 * \param[in] buf buffer with the striping
3476 * \param[in] name name of EA
3477 * \param[in] fl xattr flag (see OSD API description)
3478 * \param[in] th transaction handle
3480 * \retval 0 on success
3481 * \retval negative if failed
3483 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
3484 struct dt_object *dt,
3485 const struct lu_buf *buf,
3486 const char *name, int fl,
3489 struct lov_user_md_v1 *lum;
3490 struct lov_user_md_v3 *v3 = NULL;
3491 const char *pool_name = NULL;
3496 LASSERT(buf != NULL && buf->lb_buf != NULL);
3499 switch (lum->lmm_magic) {
3500 case LOV_USER_MAGIC_SPECIFIC:
3501 case LOV_USER_MAGIC_V3:
3503 if (v3->lmm_pool_name[0] != '\0')
3504 pool_name = v3->lmm_pool_name;
3506 case LOV_USER_MAGIC_V1:
3507 /* if { size, offset, count } = { 0, -1, 0 } and no pool
3508 * (i.e. all default values specified) then delete default
3509 * striping from dir. */
3511 "set default striping: sz %u # %u offset %d %s %s\n",
3512 (unsigned)lum->lmm_stripe_size,
3513 (unsigned)lum->lmm_stripe_count,
3514 (int)lum->lmm_stripe_offset,
3515 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
3517 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
3518 lum->lmm_stripe_count,
3519 lum->lmm_stripe_offset,
3522 case LOV_USER_MAGIC_COMP_V1:
3526 CERROR("Invalid magic %x\n", lum->lmm_magic);
3531 rc = lod_xattr_del_internal(env, dt, name, th);
3535 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3542 * Set default striping on a directory object.
3544 * Sets specified striping on a directory object unless it matches the default
3545 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3546 * EA. This striping will be used when a new directory is being created in the
3549 * \param[in] env execution environment
3550 * \param[in] dt the striped object
3551 * \param[in] buf buffer with the striping
3552 * \param[in] name name of EA
3553 * \param[in] fl xattr flag (see OSD API description)
3554 * \param[in] th transaction handle
3556 * \retval 0 on success
3557 * \retval negative if failed
3559 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
3560 struct dt_object *dt,
3561 const struct lu_buf *buf,
3562 const char *name, int fl,
3565 struct lmv_user_md_v1 *lum;
3569 LASSERT(buf != NULL && buf->lb_buf != NULL);
3572 CDEBUG(D_OTHER, "set default stripe_count # %u stripe_offset %d\n",
3573 le32_to_cpu(lum->lum_stripe_count),
3574 (int)le32_to_cpu(lum->lum_stripe_offset));
3576 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
3577 le32_to_cpu(lum->lum_stripe_offset)) &&
3578 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
3579 rc = lod_xattr_del_internal(env, dt, name, th);
3583 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3592 * Turn directory into a striped directory.
3594 * During replay the client sends the striping created before MDT
3595 * failure, then the layer above LOD sends this defined striping
3596 * using ->do_xattr_set(), so LOD uses this method to replay creation
3597 * of the stripes. Notice the original information for the striping
3598 * (#stripes, FIDs, etc) was transferred in declare path.
3600 * \param[in] env execution environment
3601 * \param[in] dt the striped object
3602 * \param[in] buf not used currently
3603 * \param[in] name not used currently
3604 * \param[in] fl xattr flag (see OSD API description)
3605 * \param[in] th transaction handle
3607 * \retval 0 on success
3608 * \retval negative if failed
3610 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
3611 const struct lu_buf *buf, const char *name,
3612 int fl, struct thandle *th)
3614 struct lod_object *lo = lod_dt_obj(dt);
3615 struct lod_thread_info *info = lod_env_info(env);
3616 struct lu_attr *attr = &info->lti_attr;
3617 struct dt_object_format *dof = &info->lti_format;
3618 struct lu_buf lmv_buf;
3619 struct lu_buf slave_lmv_buf;
3620 struct lmv_mds_md_v1 *lmm;
3621 struct lmv_mds_md_v1 *slave_lmm = NULL;
3622 struct dt_insert_rec *rec = &info->lti_dt_rec;
3627 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3630 /* The stripes are supposed to be allocated in declare phase,
3631 * if there are no stripes being allocated, it will skip */
3632 if (lo->ldo_dir_stripe_count == 0)
3635 rc = dt_attr_get(env, dt_object_child(dt), attr);
3639 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME |
3640 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
3641 dof->dof_type = DFT_DIR;
3643 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
3646 lmm = lmv_buf.lb_buf;
3648 OBD_ALLOC_PTR(slave_lmm);
3649 if (slave_lmm == NULL)
3652 lod_prep_slave_lmv_md(slave_lmm, lmm);
3653 slave_lmv_buf.lb_buf = slave_lmm;
3654 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
3656 rec->rec_type = S_IFDIR;
3657 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3658 struct dt_object *dto = lo->ldo_stripe[i];
3659 char *stripe_name = info->lti_key;
3660 struct lu_name *sname;
3661 struct linkea_data ldata = { NULL };
3662 struct lu_buf linkea_buf;
3664 /* if it's source stripe of migrating directory, don't create */
3665 if (!((lo->ldo_dir_hash_type & LMV_HASH_FLAG_MIGRATION) &&
3666 i >= lo->ldo_dir_migrate_offset)) {
3667 dt_write_lock(env, dto, MOR_TGT_CHILD);
3668 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
3670 dt_write_unlock(env, dto);
3674 rc = lod_sub_ref_add(env, dto, th);
3675 dt_write_unlock(env, dto);
3679 rec->rec_fid = lu_object_fid(&dto->do_lu);
3680 rc = lod_sub_insert(env, dto,
3681 (const struct dt_rec *)rec,
3682 (const struct dt_key *)dot, th);
3687 rec->rec_fid = lu_object_fid(&dt->do_lu);
3688 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
3689 (const struct dt_key *)dotdot, th);
3693 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
3694 cfs_fail_val != i) {
3695 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
3697 slave_lmm->lmv_master_mdt_index =
3700 slave_lmm->lmv_master_mdt_index =
3703 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
3704 XATTR_NAME_LMV, fl, th);
3709 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
3711 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3712 PFID(lu_object_fid(&dto->do_lu)), i + 1);
3714 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3715 PFID(lu_object_fid(&dto->do_lu)), i);
3717 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
3718 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
3719 sname, lu_object_fid(&dt->do_lu));
3723 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
3724 linkea_buf.lb_len = ldata.ld_leh->leh_len;
3725 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
3726 XATTR_NAME_LINK, 0, th);
3730 rec->rec_fid = lu_object_fid(&dto->do_lu);
3731 rc = lod_sub_insert(env, dt_object_child(dt),
3732 (const struct dt_rec *)rec,
3733 (const struct dt_key *)stripe_name, th);
3737 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
3742 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
3743 rc = lod_sub_xattr_set(env, dt_object_child(dt),
3744 &lmv_buf, XATTR_NAME_LMV, fl, th);
3746 if (slave_lmm != NULL)
3747 OBD_FREE_PTR(slave_lmm);
3753 * Helper function to declare/execute creation of a striped directory
3755 * Called in declare/create object path, prepare striping for a directory
3756 * and prepare defaults data striping for the objects to be created in
3757 * that directory. Notice the function calls "declaration" or "execution"
3758 * methods depending on \a declare param. This is a consequence of the
3759 * current approach while we don't have natural distributed transactions:
3760 * we basically execute non-local updates in the declare phase. So, the
3761 * arguments for the both phases are the same and this is the reason for
3762 * this function to exist.
3764 * \param[in] env execution environment
3765 * \param[in] dt object
3766 * \param[in] attr attributes the stripes will be created with
3767 * \param[in] lmu lmv_user_md if MDT indices are specified
3768 * \param[in] dof format of stripes (see OSD API description)
3769 * \param[in] th transaction handle
3770 * \param[in] declare where to call "declare" or "execute" methods
3772 * \retval 0 on success
3773 * \retval negative if failed
3775 static int lod_dir_striping_create_internal(const struct lu_env *env,
3776 struct dt_object *dt,
3777 struct lu_attr *attr,
3778 const struct lu_buf *lmu,
3779 struct dt_object_format *dof,
3783 struct lod_thread_info *info = lod_env_info(env);
3784 struct lod_object *lo = lod_dt_obj(dt);
3785 const struct lod_default_striping *lds = lo->ldo_def_striping;
3789 LASSERT(ergo(lds != NULL,
3790 lds->lds_def_striping_set ||
3791 lds->lds_dir_def_striping_set));
3793 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
3794 lo->ldo_dir_stripe_offset)) {
3796 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3797 int stripe_count = lo->ldo_dir_stripe_count;
3799 if (info->lti_ea_store_size < sizeof(*v1)) {
3800 rc = lod_ea_store_resize(info, sizeof(*v1));
3803 v1 = info->lti_ea_store;
3806 memset(v1, 0, sizeof(*v1));
3807 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3808 v1->lum_stripe_count = cpu_to_le32(stripe_count);
3809 v1->lum_stripe_offset =
3810 cpu_to_le32(lo->ldo_dir_stripe_offset);
3812 info->lti_buf.lb_buf = v1;
3813 info->lti_buf.lb_len = sizeof(*v1);
3814 lmu = &info->lti_buf;
3818 rc = lod_declare_xattr_set_lmv(env, dt, attr, lmu, dof,
3821 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
3827 /* Transfer default LMV striping from the parent */
3828 if (lds != NULL && lds->lds_dir_def_striping_set &&
3829 !LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
3830 lds->lds_dir_def_stripe_offset)) {
3831 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3833 if (info->lti_ea_store_size < sizeof(*v1)) {
3834 rc = lod_ea_store_resize(info, sizeof(*v1));
3837 v1 = info->lti_ea_store;
3840 memset(v1, 0, sizeof(*v1));
3841 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3842 v1->lum_stripe_count =
3843 cpu_to_le32(lds->lds_dir_def_stripe_count);
3844 v1->lum_stripe_offset =
3845 cpu_to_le32(lds->lds_dir_def_stripe_offset);
3847 cpu_to_le32(lds->lds_dir_def_hash_type);
3849 info->lti_buf.lb_buf = v1;
3850 info->lti_buf.lb_len = sizeof(*v1);
3852 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3853 XATTR_NAME_DEFAULT_LMV,
3856 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
3858 XATTR_NAME_DEFAULT_LMV, 0,
3864 /* Transfer default LOV striping from the parent */
3865 if (lds != NULL && lds->lds_def_striping_set &&
3866 lds->lds_def_comp_cnt != 0) {
3867 struct lov_mds_md *lmm;
3868 int lmm_size = lod_comp_md_size(lo, true);
3870 if (info->lti_ea_store_size < lmm_size) {
3871 rc = lod_ea_store_resize(info, lmm_size);
3875 lmm = info->lti_ea_store;
3877 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
3881 info->lti_buf.lb_buf = lmm;
3882 info->lti_buf.lb_len = lmm_size;
3885 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3886 XATTR_NAME_LOV, 0, th);
3888 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
3889 XATTR_NAME_LOV, 0, th);
3897 static int lod_declare_dir_striping_create(const struct lu_env *env,
3898 struct dt_object *dt,
3899 struct lu_attr *attr,
3901 struct dt_object_format *dof,
3904 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
3908 static int lod_dir_striping_create(const struct lu_env *env,
3909 struct dt_object *dt,
3910 struct lu_attr *attr,
3911 struct dt_object_format *dof,
3914 return lod_dir_striping_create_internal(env, dt, attr, NULL, dof, th,
3919 * Make LOV EA for striped object.
3921 * Generate striping information and store it in the LOV EA of the given
3922 * object. The caller must ensure nobody else is calling the function
3923 * against the object concurrently. The transaction must be started.
3924 * FLDB service must be running as well; it's used to map FID to the target,
3925 * which is stored in LOV EA.
3927 * \param[in] env execution environment for this thread
3928 * \param[in] lo LOD object
3929 * \param[in] th transaction handle
3931 * \retval 0 if LOV EA is stored successfully
3932 * \retval negative error number on failure
3934 static int lod_generate_and_set_lovea(const struct lu_env *env,
3935 struct lod_object *lo,
3938 struct lod_thread_info *info = lod_env_info(env);
3939 struct dt_object *next = dt_object_child(&lo->ldo_obj);
3940 struct lov_mds_md_v1 *lmm;
3946 if (lo->ldo_comp_cnt == 0 && !lo->ldo_is_foreign) {
3947 lod_striping_free(env, lo);
3948 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
3952 lmm_size = lod_comp_md_size(lo, false);
3953 if (info->lti_ea_store_size < lmm_size) {
3954 rc = lod_ea_store_resize(info, lmm_size);
3958 lmm = info->lti_ea_store;
3960 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
3964 info->lti_buf.lb_buf = lmm;
3965 info->lti_buf.lb_len = lmm_size;
3966 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
3967 XATTR_NAME_LOV, 0, th);
3972 * Delete layout component(s)
3974 * \param[in] env execution environment for this thread
3975 * \param[in] dt object
3976 * \param[in] th transaction handle
3978 * \retval 0 on success
3979 * \retval negative error number on failure
3981 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
3984 struct lod_layout_component *lod_comp;
3985 struct lod_object *lo = lod_dt_obj(dt);
3986 struct dt_object *next = dt_object_child(dt);
3987 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3990 LASSERT(lo->ldo_is_composite);
3991 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
3993 left = lo->ldo_comp_cnt;
3994 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3995 lod_comp = &lo->ldo_comp_entries[i];
3997 if (lod_comp->llc_id != LCME_ID_INVAL)
4001 /* Not instantiated component */
4002 if (lod_comp->llc_stripe == NULL)
4005 LASSERT(lod_comp->llc_stripe_count > 0);
4006 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4007 struct dt_object *obj = lod_comp->llc_stripe[j];
4011 rc = lod_sub_destroy(env, obj, th);
4015 lu_object_put(env, &obj->do_lu);
4016 lod_comp->llc_stripe[j] = NULL;
4018 OBD_FREE(lod_comp->llc_stripe, sizeof(struct dt_object *) *
4019 lod_comp->llc_stripes_allocated);
4020 lod_comp->llc_stripe = NULL;
4021 OBD_FREE(lod_comp->llc_ost_indices,
4022 sizeof(__u32) * lod_comp->llc_stripes_allocated);
4023 lod_comp->llc_ost_indices = NULL;
4024 lod_comp->llc_stripes_allocated = 0;
4025 lod_obj_set_pool(lo, i, NULL);
4026 if (lod_comp->llc_ostlist.op_array) {
4027 OBD_FREE(lod_comp->llc_ostlist.op_array,
4028 lod_comp->llc_ostlist.op_size);
4029 lod_comp->llc_ostlist.op_array = NULL;
4030 lod_comp->llc_ostlist.op_size = 0;
4034 LASSERTF(left >= 0 && left < lo->ldo_comp_cnt, "left = %d\n", left);
4036 struct lod_layout_component *comp_array;
4038 OBD_ALLOC(comp_array, sizeof(*comp_array) * left);
4039 if (comp_array == NULL)
4040 GOTO(out, rc = -ENOMEM);
4042 memcpy(&comp_array[0], &lo->ldo_comp_entries[0],
4043 sizeof(*comp_array) * left);
4045 OBD_FREE(lo->ldo_comp_entries,
4046 sizeof(*comp_array) * lo->ldo_comp_cnt);
4047 lo->ldo_comp_entries = comp_array;
4048 lo->ldo_comp_cnt = left;
4050 LASSERT(lo->ldo_mirror_count == 1);
4051 lo->ldo_mirrors[0].lme_end = left - 1;
4052 lod_obj_inc_layout_gen(lo);
4054 lod_free_comp_entries(lo);
4057 LASSERT(dt_object_exists(dt));
4058 rc = dt_attr_get(env, next, attr);
4062 if (attr->la_size > 0) {
4064 attr->la_valid = LA_SIZE;
4065 rc = lod_sub_attr_set(env, next, attr, th);
4070 rc = lod_generate_and_set_lovea(env, lo, th);
4074 lod_striping_free(env, lo);
4079 static int lod_get_default_lov_striping(const struct lu_env *env,
4080 struct lod_object *lo,
4081 struct lod_default_striping *lds);
4083 * Implementation of dt_object_operations::do_xattr_set.
4085 * Sets specified extended attribute on the object. Three types of EAs are
4087 * LOV EA - stores striping for a regular file or default striping (when set
4089 * LMV EA - stores a marker for the striped directories
4090 * DMV EA - stores default directory striping
4092 * When striping is applied to a non-striped existing object (this is called
4093 * late striping), then LOD notices the caller wants to turn the object into a
4094 * striped one. The stripe objects are created and appropriate EA is set:
4095 * LOV EA storing all the stripes directly or LMV EA storing just a small header
4096 * with striping configuration.
4098 * \see dt_object_operations::do_xattr_set() in the API description for details.
4100 static int lod_xattr_set(const struct lu_env *env,
4101 struct dt_object *dt, const struct lu_buf *buf,
4102 const char *name, int fl, struct thandle *th)
4104 struct dt_object *next = dt_object_child(dt);
4108 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4109 strcmp(name, XATTR_NAME_LMV) == 0) {
4110 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
4112 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4113 strncmp(name, XATTR_NAME_LMV, strlen(XATTR_NAME_LMV)) == 0 &&
4114 strlen(name) > strlen(XATTR_NAME_LMV) + 1) {
4115 const char *op = name + strlen(XATTR_NAME_LMV) + 1;
4119 * XATTR_NAME_LMV".add" is never called, but only declared,
4120 * because lod_xattr_set_lmv() will do the addition.
4122 if (strcmp(op, "del") == 0)
4123 rc = lod_dir_layout_delete(env, dt, buf, th);
4124 else if (strcmp(op, "set") == 0)
4125 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LMV,
4129 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4130 strcmp(name, XATTR_NAME_LOV) == 0) {
4131 struct lod_default_striping *lds = lod_lds_buf_get(env);
4132 struct lov_user_md_v1 *v1 = buf->lb_buf;
4133 char pool[LOV_MAXPOOLNAME + 1];
4136 /* get existing striping config */
4137 rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds);
4141 memset(pool, 0, sizeof(pool));
4142 if (lds->lds_def_striping_set == 1)
4143 lod_layout_get_pool(lds->lds_def_comp_entries,
4144 lds->lds_def_comp_cnt, pool,
4147 is_del = LOVEA_DELETE_VALUES(v1->lmm_stripe_size,
4148 v1->lmm_stripe_count,
4149 v1->lmm_stripe_offset,
4152 /* Retain the pool name if it is not given */
4153 if (v1->lmm_magic == LOV_USER_MAGIC_V1 && pool[0] != '\0' &&
4155 struct lod_thread_info *info = lod_env_info(env);
4156 struct lov_user_md_v3 *v3 = info->lti_ea_store;
4158 memset(v3, 0, sizeof(*v3));
4159 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4160 v3->lmm_pattern = cpu_to_le32(v1->lmm_pattern);
4161 v3->lmm_stripe_count =
4162 cpu_to_le32(v1->lmm_stripe_count);
4163 v3->lmm_stripe_offset =
4164 cpu_to_le32(v1->lmm_stripe_offset);
4165 v3->lmm_stripe_size = cpu_to_le32(v1->lmm_stripe_size);
4167 strlcpy(v3->lmm_pool_name, pool,
4168 sizeof(v3->lmm_pool_name));
4170 info->lti_buf.lb_buf = v3;
4171 info->lti_buf.lb_len = sizeof(*v3);
4172 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4175 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name,
4179 if (lds->lds_def_striping_set == 1 &&
4180 lds->lds_def_comp_entries != NULL)
4181 lod_free_def_comp_entries(lds);
4184 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4185 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
4187 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
4190 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
4191 (!strcmp(name, XATTR_NAME_LOV) ||
4192 !strncmp(name, XATTR_LUSTRE_LOV,
4193 strlen(XATTR_LUSTRE_LOV)))) {
4194 /* in case of lov EA swap, just set it
4195 * if not, it is a replay so check striping match what we
4196 * already have during req replay, declare_xattr_set()
4197 * defines striping, then create() does the work */
4198 if (fl & LU_XATTR_REPLACE) {
4199 /* free stripes, then update disk */
4200 lod_striping_free(env, lod_dt_obj(dt));
4202 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
4203 } else if (dt_object_remote(dt)) {
4204 /* This only happens during migration, see
4205 * mdd_migrate_create(), in which Master MDT will
4206 * create a remote target object, and only set
4207 * (migrating) stripe EA on the remote object,
4208 * and does not need creating each stripes. */
4209 rc = lod_sub_xattr_set(env, next, buf, name,
4211 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
4212 /* delete component(s) */
4213 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
4214 rc = lod_layout_del(env, dt, th);
4217 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
4218 * it's going to create create file with specified
4219 * component(s), the striping must have not being
4220 * cached in this case;
4222 * Otherwise, it's going to add/change component(s) to
4223 * an existing file, the striping must have been cached
4226 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
4227 !strcmp(name, XATTR_NAME_LOV),
4228 !lod_dt_obj(dt)->ldo_comp_cached));
4230 rc = lod_striped_create(env, dt, NULL, NULL, th);
4233 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4234 rc = lod_replace_parent_fid(env, dt, buf, th, false);
4239 /* then all other xattr */
4240 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4246 * Implementation of dt_object_operations::do_declare_xattr_del.
4248 * \see dt_object_operations::do_declare_xattr_del() in the API description
4251 static int lod_declare_xattr_del(const struct lu_env *env,
4252 struct dt_object *dt, const char *name,
4255 struct lod_object *lo = lod_dt_obj(dt);
4256 struct dt_object *next = dt_object_child(dt);
4261 rc = lod_sub_declare_xattr_del(env, next, name, th);
4265 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4268 /* set xattr to each stripes, if needed */
4269 rc = lod_striping_load(env, lo);
4273 if (lo->ldo_dir_stripe_count == 0)
4276 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4277 struct dt_object *dto = lo->ldo_stripe[i];
4280 rc = lod_sub_declare_xattr_del(env, dto, name, th);
4289 * Implementation of dt_object_operations::do_xattr_del.
4291 * If EA storing a regular striping is being deleted, then release
4292 * all the references to the stripe objects in core.
4294 * \see dt_object_operations::do_xattr_del() in the API description for details.
4296 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
4297 const char *name, struct thandle *th)
4299 struct dt_object *next = dt_object_child(dt);
4300 struct lod_object *lo = lod_dt_obj(dt);
4305 if (!strcmp(name, XATTR_NAME_LOV) || !strcmp(name, XATTR_NAME_LMV))
4306 lod_striping_free(env, lod_dt_obj(dt));
4308 rc = lod_sub_xattr_del(env, next, name, th);
4309 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
4312 if (lo->ldo_dir_stripe_count == 0)
4315 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4316 struct dt_object *dto = lo->ldo_stripe[i];
4320 rc = lod_sub_xattr_del(env, dto, name, th);
4329 * Implementation of dt_object_operations::do_xattr_list.
4331 * \see dt_object_operations::do_xattr_list() in the API description
4334 static int lod_xattr_list(const struct lu_env *env,
4335 struct dt_object *dt, const struct lu_buf *buf)
4337 return dt_xattr_list(env, dt_object_child(dt), buf);
4340 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
4342 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
4346 * Copy OST list from layout provided by user.
4348 * \param[in] lod_comp layout_component to be filled
4349 * \param[in] v3 LOV EA V3 user data
4351 * \retval 0 on success
4352 * \retval negative if failed
4354 int lod_comp_copy_ost_lists(struct lod_layout_component *lod_comp,
4355 struct lov_user_md_v3 *v3)
4361 if (v3->lmm_stripe_offset == LOV_OFFSET_DEFAULT)
4362 v3->lmm_stripe_offset = v3->lmm_objects[0].l_ost_idx;
4364 if (lod_comp->llc_ostlist.op_array) {
4365 if (lod_comp->llc_ostlist.op_size >=
4366 v3->lmm_stripe_count * sizeof(__u32)) {
4367 lod_comp->llc_ostlist.op_count =
4368 v3->lmm_stripe_count;
4371 OBD_FREE(lod_comp->llc_ostlist.op_array,
4372 lod_comp->llc_ostlist.op_size);
4375 /* copy ost list from lmm */
4376 lod_comp->llc_ostlist.op_count = v3->lmm_stripe_count;
4377 lod_comp->llc_ostlist.op_size = v3->lmm_stripe_count * sizeof(__u32);
4378 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
4379 lod_comp->llc_ostlist.op_size);
4380 if (!lod_comp->llc_ostlist.op_array)
4383 for (j = 0; j < v3->lmm_stripe_count; j++) {
4384 lod_comp->llc_ostlist.op_array[j] =
4385 v3->lmm_objects[j].l_ost_idx;
4393 * Get default striping.
4395 * \param[in] env execution environment
4396 * \param[in] lo object
4397 * \param[out] lds default striping
4399 * \retval 0 on success
4400 * \retval negative if failed
4402 static int lod_get_default_lov_striping(const struct lu_env *env,
4403 struct lod_object *lo,
4404 struct lod_default_striping *lds)
4406 struct lod_thread_info *info = lod_env_info(env);
4407 struct lov_user_md_v1 *v1 = NULL;
4408 struct lov_user_md_v3 *v3 = NULL;
4409 struct lov_comp_md_v1 *comp_v1 = NULL;
4416 lds->lds_def_striping_set = 0;
4418 rc = lod_get_lov_ea(env, lo);
4422 if (rc < (typeof(rc))sizeof(struct lov_user_md))
4425 v1 = info->lti_ea_store;
4426 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
4427 lustre_swab_lov_user_md_v1(v1);
4428 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
4429 v3 = (struct lov_user_md_v3 *)v1;
4430 lustre_swab_lov_user_md_v3(v3);
4431 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_SPECIFIC)) {
4432 v3 = (struct lov_user_md_v3 *)v1;
4433 lustre_swab_lov_user_md_v3(v3);
4434 lustre_swab_lov_user_md_objects(v3->lmm_objects,
4435 v3->lmm_stripe_count);
4436 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
4437 comp_v1 = (struct lov_comp_md_v1 *)v1;
4438 lustre_swab_lov_comp_md_v1(comp_v1);
4441 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1 &&
4442 v1->lmm_magic != LOV_MAGIC_COMP_V1 &&
4443 v1->lmm_magic != LOV_USER_MAGIC_SPECIFIC)
4446 if (v1->lmm_magic == LOV_MAGIC_COMP_V1) {
4447 comp_v1 = (struct lov_comp_md_v1 *)v1;
4448 comp_cnt = comp_v1->lcm_entry_count;
4451 mirror_cnt = comp_v1->lcm_mirror_count + 1;
4459 /* realloc default comp entries if necessary */
4460 rc = lod_def_striping_comp_resize(lds, comp_cnt);
4464 lds->lds_def_comp_cnt = comp_cnt;
4465 lds->lds_def_striping_is_composite = composite;
4466 lds->lds_def_mirror_cnt = mirror_cnt;
4468 for (i = 0; i < comp_cnt; i++) {
4469 struct lod_layout_component *lod_comp;
4472 lod_comp = &lds->lds_def_comp_entries[i];
4474 * reset lod_comp values, llc_stripes is always NULL in
4475 * the default striping template, llc_pool will be reset
4478 memset(lod_comp, 0, offsetof(typeof(*lod_comp), llc_pool));
4481 v1 = (struct lov_user_md *)((char *)comp_v1 +
4482 comp_v1->lcm_entries[i].lcme_offset);
4483 lod_comp->llc_extent =
4484 comp_v1->lcm_entries[i].lcme_extent;
4485 /* We only inherit certain flags from the layout */
4486 lod_comp->llc_flags =
4487 comp_v1->lcm_entries[i].lcme_flags &
4488 LCME_TEMPLATE_FLAGS;
4491 if (v1->lmm_pattern != LOV_PATTERN_RAID0 &&
4492 v1->lmm_pattern != LOV_PATTERN_MDT &&
4493 v1->lmm_pattern != 0) {
4494 lod_free_def_comp_entries(lds);
4498 CDEBUG(D_LAYOUT, DFID" stripe_count=%d stripe_size=%d "
4499 "stripe_offset=%d\n",
4500 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
4501 (int)v1->lmm_stripe_count, (int)v1->lmm_stripe_size,
4502 (int)v1->lmm_stripe_offset);
4504 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
4505 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
4506 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
4507 lod_comp->llc_pattern = v1->lmm_pattern;
4510 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
4511 /* XXX: sanity check here */
4512 v3 = (struct lov_user_md_v3 *) v1;
4513 if (v3->lmm_pool_name[0] != '\0')
4514 pool = v3->lmm_pool_name;
4516 lod_set_def_pool(lds, i, pool);
4517 if (v1->lmm_magic == LOV_USER_MAGIC_SPECIFIC) {
4518 v3 = (struct lov_user_md_v3 *)v1;
4519 rc = lod_comp_copy_ost_lists(lod_comp, v3);
4522 } else if (lod_comp->llc_ostlist.op_array &&
4523 lod_comp->llc_ostlist.op_count) {
4524 for (j = 0; j < lod_comp->llc_ostlist.op_count; j++)
4525 lod_comp->llc_ostlist.op_array[j] = -1;
4526 lod_comp->llc_ostlist.op_count = 0;
4530 lds->lds_def_striping_set = 1;
4535 * Get default directory striping.
4537 * \param[in] env execution environment
4538 * \param[in] lo object
4539 * \param[out] lds default striping
4541 * \retval 0 on success
4542 * \retval negative if failed
4544 static int lod_get_default_lmv_striping(const struct lu_env *env,
4545 struct lod_object *lo,
4546 struct lod_default_striping *lds)
4548 struct lod_thread_info *info = lod_env_info(env);
4549 struct lmv_user_md_v1 *v1 = NULL;
4553 lds->lds_dir_def_striping_set = 0;
4554 rc = lod_get_default_lmv_ea(env, lo);
4558 if (rc < (typeof(rc))sizeof(struct lmv_user_md))
4561 v1 = info->lti_ea_store;
4563 lds->lds_dir_def_stripe_count = le32_to_cpu(v1->lum_stripe_count);
4564 lds->lds_dir_def_stripe_offset = le32_to_cpu(v1->lum_stripe_offset);
4565 lds->lds_dir_def_hash_type = le32_to_cpu(v1->lum_hash_type);
4566 lds->lds_dir_def_striping_set = 1;
4572 * Get default striping in the object.
4574 * Get object default striping and default directory striping.
4576 * \param[in] env execution environment
4577 * \param[in] lo object
4578 * \param[out] lds default striping
4580 * \retval 0 on success
4581 * \retval negative if failed
4583 static int lod_get_default_striping(const struct lu_env *env,
4584 struct lod_object *lo,
4585 struct lod_default_striping *lds)
4589 rc = lod_get_default_lov_striping(env, lo, lds);
4590 rc1 = lod_get_default_lmv_striping(env, lo, lds);
4591 if (rc == 0 && rc1 < 0)
4598 * Apply default striping on object.
4600 * If object striping pattern is not set, set to the one in default striping.
4601 * The default striping is from parent or fs.
4603 * \param[in] lo new object
4604 * \param[in] lds default striping
4605 * \param[in] mode new object's mode
4607 static void lod_striping_from_default(struct lod_object *lo,
4608 const struct lod_default_striping *lds,
4611 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
4612 struct lov_desc *desc = &d->lod_desc;
4615 if (lds->lds_def_striping_set && S_ISREG(mode)) {
4616 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
4617 lds->lds_def_comp_cnt);
4621 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
4622 if (lds->lds_def_mirror_cnt > 1)
4623 lo->ldo_flr_state = LCM_FL_RDONLY;
4625 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4626 struct lod_layout_component *obj_comp =
4627 &lo->ldo_comp_entries[i];
4628 struct lod_layout_component *def_comp =
4629 &lds->lds_def_comp_entries[i];
4631 CDEBUG(D_LAYOUT, "Inherit from default: flags=%#x "
4632 "size=%hu nr=%u offset=%u pattern=%#x pool=%s\n",
4633 def_comp->llc_flags,
4634 def_comp->llc_stripe_size,
4635 def_comp->llc_stripe_count,
4636 def_comp->llc_stripe_offset,
4637 def_comp->llc_pattern,
4638 def_comp->llc_pool ?: "");
4640 *obj_comp = *def_comp;
4641 if (def_comp->llc_pool != NULL) {
4642 /* pointer was copied from def_comp */
4643 obj_comp->llc_pool = NULL;
4644 lod_obj_set_pool(lo, i, def_comp->llc_pool);
4648 if (def_comp->llc_ostlist.op_array &&
4649 def_comp->llc_ostlist.op_count) {
4650 OBD_ALLOC(obj_comp->llc_ostlist.op_array,
4651 obj_comp->llc_ostlist.op_size);
4652 if (!obj_comp->llc_ostlist.op_array)
4654 memcpy(obj_comp->llc_ostlist.op_array,
4655 def_comp->llc_ostlist.op_array,
4656 obj_comp->llc_ostlist.op_size);
4657 } else if (def_comp->llc_ostlist.op_array) {
4658 obj_comp->llc_ostlist.op_array = NULL;
4662 * Don't initialize these fields for plain layout
4663 * (v1/v3) here, they are inherited in the order of
4664 * 'parent' -> 'fs default (root)' -> 'global default
4665 * values for stripe_count & stripe_size'.
4667 * see lod_ah_init().
4669 if (!lo->ldo_is_composite)
4672 lod_adjust_stripe_info(obj_comp, desc);
4674 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
4675 if (lo->ldo_dir_stripe_count == 0)
4676 lo->ldo_dir_stripe_count =
4677 lds->lds_dir_def_stripe_count;
4678 if (lo->ldo_dir_stripe_offset == -1)
4679 lo->ldo_dir_stripe_offset =
4680 lds->lds_dir_def_stripe_offset;
4681 if (lo->ldo_dir_hash_type == 0)
4682 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
4684 CDEBUG(D_LAYOUT, "striping from default dir: count:%hu, "
4685 "offset:%u, hash_type:%u\n",
4686 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
4687 lo->ldo_dir_hash_type);
4691 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root)
4693 struct lod_layout_component *lod_comp;
4695 if (lo->ldo_comp_cnt == 0)
4698 if (lo->ldo_is_composite)
4701 lod_comp = &lo->ldo_comp_entries[0];
4703 if (lod_comp->llc_stripe_count <= 0 ||
4704 lod_comp->llc_stripe_size <= 0)
4707 if (from_root && (lod_comp->llc_pool == NULL ||
4708 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
4715 * Implementation of dt_object_operations::do_ah_init.
4717 * This method is used to make a decision on the striping configuration for the
4718 * object being created. It can be taken from the \a parent object if it exists,
4719 * or filesystem's default. The resulting configuration (number of stripes,
4720 * stripe size/offset, pool name, etc) is stored in the object itself and will
4721 * be used by the methods like ->doo_declare_create().
4723 * \see dt_object_operations::do_ah_init() in the API description for details.
4725 static void lod_ah_init(const struct lu_env *env,
4726 struct dt_allocation_hint *ah,
4727 struct dt_object *parent,
4728 struct dt_object *child,
4731 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
4732 struct lod_thread_info *info = lod_env_info(env);
4733 struct lod_default_striping *lds = lod_lds_buf_get(env);
4734 struct dt_object *nextp = NULL;
4735 struct dt_object *nextc;
4736 struct lod_object *lp = NULL;
4737 struct lod_object *lc;
4738 struct lov_desc *desc;
4739 struct lod_layout_component *lod_comp;
4745 if (likely(parent)) {
4746 nextp = dt_object_child(parent);
4747 lp = lod_dt_obj(parent);
4750 nextc = dt_object_child(child);
4751 lc = lod_dt_obj(child);
4753 LASSERT(!lod_obj_is_striped(child));
4754 /* default layout template may have been set on the regular file
4755 * when this is called from mdd_create_data() */
4756 if (S_ISREG(child_mode))
4757 lod_free_comp_entries(lc);
4759 if (!dt_object_exists(nextc))
4760 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
4762 if (S_ISDIR(child_mode)) {
4763 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
4765 /* other default values are 0 */
4766 lc->ldo_dir_stripe_offset = -1;
4769 * If parent object is not root directory,
4770 * then get default striping from parent object.
4772 if (likely(lp != NULL) && !fid_is_root(lod_object_fid(lp)))
4773 lod_get_default_striping(env, lp, lds);
4775 /* set child default striping info, default value is NULL */
4776 if (lds->lds_def_striping_set || lds->lds_dir_def_striping_set)
4777 lc->ldo_def_striping = lds;
4779 /* It should always honour the specified stripes */
4780 /* Note: old client (< 2.7)might also do lfs mkdir, whose EA
4781 * will have old magic. In this case, we should ignore the
4782 * stripe count and try to create dir by default stripe.
4784 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
4785 (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
4786 le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC)) {
4787 lc->ldo_dir_stripe_count =
4788 le32_to_cpu(lum1->lum_stripe_count);
4789 lc->ldo_dir_stripe_offset =
4790 le32_to_cpu(lum1->lum_stripe_offset);
4791 lc->ldo_dir_hash_type =
4792 le32_to_cpu(lum1->lum_hash_type);
4794 "set dirstripe: count %hu, offset %d, hash %u\n",
4795 lc->ldo_dir_stripe_count,
4796 (int)lc->ldo_dir_stripe_offset,
4797 lc->ldo_dir_hash_type);
4799 /* transfer defaults LMV to new directory */
4800 lod_striping_from_default(lc, lds, child_mode);
4802 /* set count 0 to create normal directory */
4803 if (lc->ldo_dir_stripe_count == 1)
4804 lc->ldo_dir_stripe_count = 0;
4807 /* shrink the stripe_count to the avaible MDT count */
4808 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
4809 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)) {
4810 lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
4811 if (lc->ldo_dir_stripe_count == 1)
4812 lc->ldo_dir_stripe_count = 0;
4815 CDEBUG(D_INFO, "final dir stripe [%hu %d %u]\n",
4816 lc->ldo_dir_stripe_count,
4817 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
4822 /* child object regular file*/
4824 if (!lod_object_will_be_striped(S_ISREG(child_mode),
4825 lu_object_fid(&child->do_lu)))
4828 /* If object is going to be striped over OSTs, transfer default
4829 * striping information to the child, so that we can use it
4830 * during declaration and creation.
4832 * Try from the parent first.
4834 if (likely(lp != NULL)) {
4835 rc = lod_get_default_lov_striping(env, lp, lds);
4837 lod_striping_from_default(lc, lds, child_mode);
4840 /* Initialize lod_device::lod_md_root object reference */
4841 if (d->lod_md_root == NULL) {
4842 struct dt_object *root;
4843 struct lod_object *lroot;
4845 lu_root_fid(&info->lti_fid);
4846 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
4847 if (!IS_ERR(root)) {
4848 lroot = lod_dt_obj(root);
4850 spin_lock(&d->lod_lock);
4851 if (d->lod_md_root != NULL)
4852 dt_object_put(env, &d->lod_md_root->ldo_obj);
4853 d->lod_md_root = lroot;
4854 spin_unlock(&d->lod_lock);
4858 /* try inherit layout from the root object (fs default) when:
4859 * - parent does not have default layout; or
4860 * - parent has plain(v1/v3) default layout, and some attributes
4861 * are not specified in the default layout;
4863 if (d->lod_md_root != NULL && lod_need_inherit_more(lc, true)) {
4864 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds);
4867 if (lc->ldo_comp_cnt == 0) {
4868 lod_striping_from_default(lc, lds, child_mode);
4869 } else if (!lds->lds_def_striping_is_composite) {
4870 struct lod_layout_component *def_comp;
4872 LASSERT(!lc->ldo_is_composite);
4873 lod_comp = &lc->ldo_comp_entries[0];
4874 def_comp = &lds->lds_def_comp_entries[0];
4876 if (lod_comp->llc_stripe_count <= 0)
4877 lod_comp->llc_stripe_count =
4878 def_comp->llc_stripe_count;
4879 if (lod_comp->llc_stripe_size <= 0)
4880 lod_comp->llc_stripe_size =
4881 def_comp->llc_stripe_size;
4882 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT &&
4883 (!lod_comp->llc_pool || !lod_comp->llc_pool[0]))
4884 lod_comp->llc_stripe_offset =
4885 def_comp->llc_stripe_offset;
4886 if (lod_comp->llc_pool == NULL)
4887 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
4892 * fs default striping may not be explicitly set, or historically set
4893 * in config log, use them.
4895 if (lod_need_inherit_more(lc, false)) {
4896 if (lc->ldo_comp_cnt == 0) {
4897 rc = lod_alloc_comp_entries(lc, 0, 1);
4899 /* fail to allocate memory, will create a
4900 * non-striped file. */
4902 lc->ldo_is_composite = 0;
4903 lod_comp = &lc->ldo_comp_entries[0];
4904 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4906 LASSERT(!lc->ldo_is_composite);
4907 lod_comp = &lc->ldo_comp_entries[0];
4908 desc = &d->lod_desc;
4909 lod_adjust_stripe_info(lod_comp, desc);
4915 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
4917 * Size initialization on late striping.
4919 * Propagate the size of a truncated object to a deferred striping.
4920 * This function handles a special case when truncate was done on a
4921 * non-striped object and now while the striping is being created
4922 * we can't lose that size, so we have to propagate it to the stripes
4925 * \param[in] env execution environment
4926 * \param[in] dt object
4927 * \param[in] th transaction handle
4929 * \retval 0 on success
4930 * \retval negative if failed
4932 static int lod_declare_init_size(const struct lu_env *env,
4933 struct dt_object *dt, struct thandle *th)
4935 struct dt_object *next = dt_object_child(dt);
4936 struct lod_object *lo = lod_dt_obj(dt);
4937 struct dt_object **objects = NULL;
4938 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4939 uint64_t size, offs;
4940 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
4941 struct lu_extent size_ext;
4944 if (!lod_obj_is_striped(dt))
4947 rc = dt_attr_get(env, next, attr);
4948 LASSERT(attr->la_valid & LA_SIZE);
4952 size = attr->la_size;
4956 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
4957 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4958 struct lod_layout_component *lod_comp;
4959 struct lu_extent *extent;
4961 lod_comp = &lo->ldo_comp_entries[i];
4963 if (lod_comp->llc_stripe == NULL)
4966 extent = &lod_comp->llc_extent;
4967 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
4968 if (!lo->ldo_is_composite ||
4969 lu_extent_is_overlapped(extent, &size_ext)) {
4970 objects = lod_comp->llc_stripe;
4971 stripe_count = lod_comp->llc_stripe_count;
4972 stripe_size = lod_comp->llc_stripe_size;
4975 if (stripe_count == 0)
4978 LASSERT(objects != NULL && stripe_size != 0);
4979 /* ll_do_div64(a, b) returns a % b, and a = a / b */
4980 ll_do_div64(size, (__u64)stripe_size);
4981 stripe = ll_do_div64(size, (__u64)stripe_count);
4982 LASSERT(objects[stripe] != NULL);
4984 size = size * stripe_size;
4985 offs = attr->la_size;
4986 size += ll_do_div64(offs, stripe_size);
4988 attr->la_valid = LA_SIZE;
4989 attr->la_size = size;
4991 rc = lod_sub_declare_attr_set(env, objects[stripe],
5000 * Declare creation of striped object.
5002 * The function declares creation stripes for a regular object. The function
5003 * also declares whether the stripes will be created with non-zero size if
5004 * previously size was set non-zero on the master object. If object \a dt is
5005 * not local, then only fully defined striping can be applied in \a lovea.
5006 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
5009 * \param[in] env execution environment
5010 * \param[in] dt object
5011 * \param[in] attr attributes the stripes will be created with
5012 * \param[in] lovea a buffer containing striping description
5013 * \param[in] th transaction handle
5015 * \retval 0 on success
5016 * \retval negative if failed
5018 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
5019 struct lu_attr *attr,
5020 const struct lu_buf *lovea, struct thandle *th)
5022 struct lod_thread_info *info = lod_env_info(env);
5023 struct dt_object *next = dt_object_child(dt);
5024 struct lod_object *lo = lod_dt_obj(dt);
5028 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
5029 GOTO(out, rc = -ENOMEM);
5031 if (!dt_object_remote(next)) {
5032 /* choose OST and generate appropriate objects */
5033 rc = lod_prepare_create(env, lo, attr, lovea, th);
5038 * declare storage for striping data
5040 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
5042 /* LOD can not choose OST objects for remote objects, i.e.
5043 * stripes must be ready before that. Right now, it can only
5044 * happen during migrate, i.e. migrate process needs to create
5045 * remote regular file (mdd_migrate_create), then the migrate
5046 * process will provide stripeEA. */
5047 LASSERT(lovea != NULL);
5048 info->lti_buf = *lovea;
5051 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
5052 XATTR_NAME_LOV, 0, th);
5057 * if striping is created with local object's size > 0,
5058 * we have to propagate this size to specific object
5059 * the case is possible only when local object was created previously
5061 if (dt_object_exists(next))
5062 rc = lod_declare_init_size(env, dt, th);
5065 /* failed to create striping or to set initial size, let's reset
5066 * config so that others don't get confused */
5068 lod_striping_free(env, lo);
5074 * Implementation of dt_object_operations::do_declare_create.
5076 * The method declares creation of a new object. If the object will be striped,
5077 * then helper functions are called to find FIDs for the stripes, declare
5078 * creation of the stripes and declare initialization of the striping
5079 * information to be stored in the master object.
5081 * \see dt_object_operations::do_declare_create() in the API description
5084 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
5085 struct lu_attr *attr,
5086 struct dt_allocation_hint *hint,
5087 struct dt_object_format *dof, struct thandle *th)
5089 struct dt_object *next = dt_object_child(dt);
5090 struct lod_object *lo = lod_dt_obj(dt);
5099 * first of all, we declare creation of local object
5101 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
5106 * it's lod_ah_init() that has decided the object will be striped
5108 if (dof->dof_type == DFT_REGULAR) {
5109 /* callers don't want stripes */
5110 /* XXX: all tricky interactions with ->ah_make_hint() decided
5111 * to use striping, then ->declare_create() behaving differently
5112 * should be cleaned */
5113 if (dof->u.dof_reg.striped != 0)
5114 rc = lod_declare_striped_create(env, dt, attr,
5116 } else if (dof->dof_type == DFT_DIR) {
5117 struct seq_server_site *ss;
5118 struct lu_buf buf = { NULL };
5119 struct lu_buf *lmu = NULL;
5121 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
5123 /* If the parent has default stripeEA, and client
5124 * did not find it before sending create request,
5125 * then MDT will return -EREMOTE, and client will
5126 * retrieve the default stripeEA and re-create the
5129 * Note: if dah_eadata != NULL, it means creating the
5130 * striped directory with specified stripeEA, then it
5131 * should ignore the default stripeEA */
5132 if (hint != NULL && hint->dah_eadata == NULL) {
5133 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
5134 GOTO(out, rc = -EREMOTE);
5136 if (lo->ldo_dir_stripe_offset == -1) {
5137 /* child and parent should be in the same MDT */
5138 if (hint->dah_parent != NULL &&
5139 dt_object_remote(hint->dah_parent))
5140 GOTO(out, rc = -EREMOTE);
5141 } else if (lo->ldo_dir_stripe_offset !=
5143 struct lod_device *lod;
5144 struct lod_tgt_descs *ltd;
5145 struct lod_tgt_desc *tgt = NULL;
5146 bool found_mdt = false;
5149 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5150 ltd = &lod->lod_mdt_descs;
5151 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
5152 tgt = LTD_TGT(ltd, i);
5153 if (tgt->ltd_index ==
5154 lo->ldo_dir_stripe_offset) {
5160 /* If the MDT indicated by stripe_offset can be
5161 * found, then tell client to resend the create
5162 * request to the correct MDT, otherwise return
5163 * error to client */
5165 GOTO(out, rc = -EREMOTE);
5167 GOTO(out, rc = -EINVAL);
5169 } else if (hint && hint->dah_eadata) {
5171 lmu->lb_buf = (void *)hint->dah_eadata;
5172 lmu->lb_len = hint->dah_eadata_len;
5175 rc = lod_declare_dir_striping_create(env, dt, attr, lmu, dof,
5179 /* failed to create striping or to set initial size, let's reset
5180 * config so that others don't get confused */
5182 lod_striping_free(env, lo);
5187 * Generate component ID for new created component.
5189 * \param[in] lo LOD object
5190 * \param[in] comp_idx index of ldo_comp_entries
5192 * \retval component ID on success
5193 * \retval LCME_ID_INVAL on failure
5195 static __u32 lod_gen_component_id(struct lod_object *lo,
5196 int mirror_id, int comp_idx)
5198 struct lod_layout_component *lod_comp;
5199 __u32 id, start, end;
5202 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
5204 lod_obj_inc_layout_gen(lo);
5205 id = lo->ldo_layout_gen;
5206 if (likely(id <= SEQ_ID_MAX))
5207 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
5209 /* Layout generation wraps, need to check collisions. */
5210 start = id & SEQ_ID_MASK;
5213 for (id = start; id <= end; id++) {
5214 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5215 lod_comp = &lo->ldo_comp_entries[i];
5216 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
5219 /* Found the ununsed ID */
5220 if (i == lo->ldo_comp_cnt)
5221 RETURN(pflr_id(mirror_id, id));
5223 if (end == LCME_ID_MAX) {
5225 end = min(lo->ldo_layout_gen & LCME_ID_MASK,
5226 (__u32)(LCME_ID_MAX - 1));
5230 RETURN(LCME_ID_INVAL);
5234 * Creation of a striped regular object.
5236 * The function is called to create the stripe objects for a regular
5237 * striped file. This can happen at the initial object creation or
5238 * when the caller asks LOD to do so using ->do_xattr_set() method
5239 * (so called late striping). Notice all the information are already
5240 * prepared in the form of the list of objects (ldo_stripe field).
5241 * This is done during declare phase.
5243 * \param[in] env execution environment
5244 * \param[in] dt object
5245 * \param[in] attr attributes the stripes will be created with
5246 * \param[in] dof format of stripes (see OSD API description)
5247 * \param[in] th transaction handle
5249 * \retval 0 on success
5250 * \retval negative if failed
5252 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
5253 struct lu_attr *attr, struct dt_object_format *dof,
5256 struct lod_layout_component *lod_comp;
5257 struct lod_object *lo = lod_dt_obj(dt);
5262 LASSERT((lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL) ||
5263 lo->ldo_is_foreign);
5265 mirror_id = 0; /* non-flr file's mirror_id is 0 */
5266 if (lo->ldo_mirror_count > 1) {
5267 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5268 lod_comp = &lo->ldo_comp_entries[i];
5269 if (lod_comp->llc_id != LCME_ID_INVAL &&
5270 mirror_id_of(lod_comp->llc_id) > mirror_id)
5271 mirror_id = mirror_id_of(lod_comp->llc_id);
5275 /* create all underlying objects */
5276 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5277 lod_comp = &lo->ldo_comp_entries[i];
5279 if (lod_comp->llc_id == LCME_ID_INVAL) {
5280 /* only the component of FLR layout with more than 1
5281 * mirror has mirror ID in its component ID.
5283 if (lod_comp->llc_extent.e_start == 0 &&
5284 lo->ldo_mirror_count > 1)
5287 lod_comp->llc_id = lod_gen_component_id(lo,
5289 if (lod_comp->llc_id == LCME_ID_INVAL)
5290 GOTO(out, rc = -ERANGE);
5293 if (lod_comp_inited(lod_comp))
5296 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
5297 lod_comp_set_init(lod_comp);
5299 if (lov_pattern(lod_comp->llc_pattern) == LOV_PATTERN_MDT)
5300 lod_comp_set_init(lod_comp);
5302 if (lod_comp->llc_stripe == NULL)
5305 LASSERT(lod_comp->llc_stripe_count);
5306 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
5307 struct dt_object *object = lod_comp->llc_stripe[j];
5308 LASSERT(object != NULL);
5309 rc = lod_sub_create(env, object, attr, NULL, dof, th);
5313 lod_comp_set_init(lod_comp);
5316 rc = lod_fill_mirrors(lo);
5320 rc = lod_generate_and_set_lovea(env, lo, th);
5324 lo->ldo_comp_cached = 1;
5328 lod_striping_free(env, lo);
5332 static inline bool lod_obj_is_dom(struct dt_object *dt)
5334 struct lod_object *lo = lod_dt_obj(dt);
5336 if (!dt_object_exists(dt_object_child(dt)))
5339 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
5342 if (!lo->ldo_comp_cnt)
5345 return (lov_pattern(lo->ldo_comp_entries[0].llc_pattern) ==
5350 * Implementation of dt_object_operations::do_create.
5352 * If any of preceeding methods (like ->do_declare_create(),
5353 * ->do_ah_init(), etc) chose to create a striped object,
5354 * then this method will create the master and the stripes.
5356 * \see dt_object_operations::do_create() in the API description for details.
5358 static int lod_create(const struct lu_env *env, struct dt_object *dt,
5359 struct lu_attr *attr, struct dt_allocation_hint *hint,
5360 struct dt_object_format *dof, struct thandle *th)
5365 /* create local object */
5366 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
5370 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
5371 (lod_obj_is_striped(dt) || lod_obj_is_dom(dt)) &&
5372 dof->u.dof_reg.striped != 0) {
5373 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
5374 rc = lod_striped_create(env, dt, attr, dof, th);
5381 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
5382 struct dt_object *dt, struct thandle *th,
5383 int comp_idx, int stripe_idx,
5384 struct lod_obj_stripe_cb_data *data)
5386 if (data->locd_declare)
5387 return lod_sub_declare_destroy(env, dt, th);
5388 else if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
5389 stripe_idx == cfs_fail_val)
5390 return lod_sub_destroy(env, dt, th);
5396 * Implementation of dt_object_operations::do_declare_destroy.
5398 * If the object is a striped directory, then the function declares reference
5399 * removal from the master object (this is an index) to the stripes and declares
5400 * destroy of all the stripes. In all the cases, it declares an intention to
5401 * destroy the object itself.
5403 * \see dt_object_operations::do_declare_destroy() in the API description
5406 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
5409 struct dt_object *next = dt_object_child(dt);
5410 struct lod_object *lo = lod_dt_obj(dt);
5411 struct lod_thread_info *info = lod_env_info(env);
5412 char *stripe_name = info->lti_key;
5417 * load striping information, notice we don't do this when object
5418 * is being initialized as we don't need this information till
5419 * few specific cases like destroy, chown
5421 rc = lod_striping_load(env, lo);
5425 /* declare destroy for all underlying objects */
5426 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5427 rc = next->do_ops->do_index_try(env, next,
5428 &dt_directory_features);
5432 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5433 rc = lod_sub_declare_ref_del(env, next, th);
5437 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
5438 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
5440 rc = lod_sub_declare_delete(env, next,
5441 (const struct dt_key *)stripe_name, th);
5448 * we declare destroy for the local object
5450 rc = lod_sub_declare_destroy(env, next, th);
5454 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
5455 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
5458 if (!lod_obj_is_striped(dt))
5461 /* declare destroy all striped objects */
5462 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5463 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5464 if (lo->ldo_stripe[i] == NULL)
5467 rc = lod_sub_declare_ref_del(env, lo->ldo_stripe[i],
5470 rc = lod_sub_declare_destroy(env, lo->ldo_stripe[i],
5476 struct lod_obj_stripe_cb_data data = { { 0 } };
5478 data.locd_declare = true;
5479 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
5480 rc = lod_obj_for_each_stripe(env, lo, th, &data);
5487 * Implementation of dt_object_operations::do_destroy.
5489 * If the object is a striped directory, then the function removes references
5490 * from the master object (this is an index) to the stripes and destroys all
5491 * the stripes. In all the cases, the function destroys the object itself.
5493 * \see dt_object_operations::do_destroy() in the API description for details.
5495 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
5498 struct dt_object *next = dt_object_child(dt);
5499 struct lod_object *lo = lod_dt_obj(dt);
5500 struct lod_thread_info *info = lod_env_info(env);
5501 char *stripe_name = info->lti_key;
5506 /* destroy sub-stripe of master object */
5507 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5508 rc = next->do_ops->do_index_try(env, next,
5509 &dt_directory_features);
5513 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5514 rc = lod_sub_ref_del(env, next, th);
5518 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
5519 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
5522 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
5523 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
5524 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)));
5526 rc = lod_sub_delete(env, next,
5527 (const struct dt_key *)stripe_name, th);
5533 rc = lod_sub_destroy(env, next, th);
5537 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
5538 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
5541 if (!lod_obj_is_striped(dt))
5544 /* destroy all striped objects */
5545 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5546 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5547 if (lo->ldo_stripe[i] == NULL)
5549 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
5550 i == cfs_fail_val) {
5551 dt_write_lock(env, lo->ldo_stripe[i],
5553 rc = lod_sub_ref_del(env, lo->ldo_stripe[i],
5555 dt_write_unlock(env, lo->ldo_stripe[i]);
5559 rc = lod_sub_destroy(env, lo->ldo_stripe[i],
5566 struct lod_obj_stripe_cb_data data = { { 0 } };
5568 data.locd_declare = false;
5569 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
5570 rc = lod_obj_for_each_stripe(env, lo, th, &data);
5577 * Implementation of dt_object_operations::do_declare_ref_add.
5579 * \see dt_object_operations::do_declare_ref_add() in the API description
5582 static int lod_declare_ref_add(const struct lu_env *env,
5583 struct dt_object *dt, struct thandle *th)
5585 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
5589 * Implementation of dt_object_operations::do_ref_add.
5591 * \see dt_object_operations::do_ref_add() in the API description for details.
5593 static int lod_ref_add(const struct lu_env *env,
5594 struct dt_object *dt, struct thandle *th)
5596 return lod_sub_ref_add(env, dt_object_child(dt), th);
5600 * Implementation of dt_object_operations::do_declare_ref_del.
5602 * \see dt_object_operations::do_declare_ref_del() in the API description
5605 static int lod_declare_ref_del(const struct lu_env *env,
5606 struct dt_object *dt, struct thandle *th)
5608 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
5612 * Implementation of dt_object_operations::do_ref_del
5614 * \see dt_object_operations::do_ref_del() in the API description for details.
5616 static int lod_ref_del(const struct lu_env *env,
5617 struct dt_object *dt, struct thandle *th)
5619 return lod_sub_ref_del(env, dt_object_child(dt), th);
5623 * Implementation of dt_object_operations::do_object_sync.
5625 * \see dt_object_operations::do_object_sync() in the API description
5628 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
5629 __u64 start, __u64 end)
5631 return dt_object_sync(env, dt_object_child(dt), start, end);
5635 * Implementation of dt_object_operations::do_object_unlock.
5637 * Used to release LDLM lock(s).
5639 * \see dt_object_operations::do_object_unlock() in the API description
5642 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
5643 struct ldlm_enqueue_info *einfo,
5644 union ldlm_policy_data *policy)
5646 struct lod_object *lo = lod_dt_obj(dt);
5647 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
5648 int slave_locks_size;
5652 if (slave_locks == NULL)
5655 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
5656 /* Note: for remote lock for single stripe dir, MDT will cancel
5657 * the lock by lockh directly */
5658 LASSERT(!dt_object_remote(dt_object_child(dt)));
5660 /* locks were unlocked in MDT layer */
5661 for (i = 0; i < slave_locks->ha_count; i++)
5662 LASSERT(!lustre_handle_is_used(&slave_locks->ha_handles[i]));
5665 * NB, ha_count may not equal to ldo_dir_stripe_count, because dir
5666 * layout may change, e.g., shrink dir layout after migration.
5668 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
5669 dt_invalidate(env, lo->ldo_stripe[i]);
5671 slave_locks_size = offsetof(typeof(*slave_locks),
5672 ha_handles[slave_locks->ha_count]);
5673 OBD_FREE(slave_locks, slave_locks_size);
5674 einfo->ei_cbdata = NULL;
5680 * Implementation of dt_object_operations::do_object_lock.
5682 * Used to get LDLM lock on the non-striped and striped objects.
5684 * \see dt_object_operations::do_object_lock() in the API description
5687 static int lod_object_lock(const struct lu_env *env,
5688 struct dt_object *dt,
5689 struct lustre_handle *lh,
5690 struct ldlm_enqueue_info *einfo,
5691 union ldlm_policy_data *policy)
5693 struct lod_object *lo = lod_dt_obj(dt);
5694 int slave_locks_size;
5695 struct lustre_handle_array *slave_locks = NULL;
5700 /* remote object lock */
5701 if (!einfo->ei_enq_slave) {
5702 LASSERT(dt_object_remote(dt));
5703 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
5707 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
5710 rc = lod_striping_load(env, lo);
5715 if (lo->ldo_dir_stripe_count <= 1)
5718 slave_locks_size = offsetof(typeof(*slave_locks),
5719 ha_handles[lo->ldo_dir_stripe_count]);
5720 /* Freed in lod_object_unlock */
5721 OBD_ALLOC(slave_locks, slave_locks_size);
5724 slave_locks->ha_count = lo->ldo_dir_stripe_count;
5726 /* striped directory lock */
5727 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5728 struct lustre_handle lockh;
5729 struct ldlm_res_id *res_id;
5731 res_id = &lod_env_info(env)->lti_res_id;
5732 fid_build_reg_res_name(lu_object_fid(&lo->ldo_stripe[i]->do_lu),
5734 einfo->ei_res_id = res_id;
5736 LASSERT(lo->ldo_stripe[i] != NULL);
5737 if (dt_object_remote(lo->ldo_stripe[i])) {
5738 set_bit(i, (void *)slave_locks->ha_map);
5739 rc = dt_object_lock(env, lo->ldo_stripe[i], &lockh,
5742 struct ldlm_namespace *ns = einfo->ei_namespace;
5743 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
5744 ldlm_completion_callback completion = einfo->ei_cb_cp;
5745 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
5747 if (einfo->ei_mode == LCK_PW ||
5748 einfo->ei_mode == LCK_EX)
5749 dlmflags |= LDLM_FL_COS_INCOMPAT;
5751 LASSERT(ns != NULL);
5752 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS,
5753 policy, einfo->ei_mode,
5754 &dlmflags, blocking,
5756 NULL, 0, LVB_T_NONE,
5761 ldlm_lock_decref_and_cancel(
5762 &slave_locks->ha_handles[i],
5764 OBD_FREE(slave_locks, slave_locks_size);
5767 slave_locks->ha_handles[i] = lockh;
5769 einfo->ei_cbdata = slave_locks;
5775 * Implementation of dt_object_operations::do_invalidate.
5777 * \see dt_object_operations::do_invalidate() in the API description for details
5779 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
5781 return dt_invalidate(env, dt_object_child(dt));
5784 static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
5788 /* clear memory region that will be used for layout change */
5789 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
5790 info->lti_count = 0;
5792 if (info->lti_comp_size >= comp_cnt)
5795 if (info->lti_comp_size > 0) {
5796 OBD_FREE(info->lti_comp_idx,
5797 info->lti_comp_size * sizeof(__u32));
5798 info->lti_comp_size = 0;
5801 OBD_ALLOC(info->lti_comp_idx, comp_cnt * sizeof(__u32));
5802 if (!info->lti_comp_idx)
5805 info->lti_comp_size = comp_cnt;
5809 static int lod_declare_instantiate_components(const struct lu_env *env,
5810 struct lod_object *lo, struct thandle *th)
5812 struct lod_thread_info *info = lod_env_info(env);
5817 LASSERT(info->lti_count < lo->ldo_comp_cnt);
5819 for (i = 0; i < info->lti_count; i++) {
5820 rc = lod_qos_prep_create(env, lo, NULL, th,
5821 info->lti_comp_idx[i]);
5827 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
5828 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
5829 &info->lti_buf, XATTR_NAME_LOV, 0, th);
5835 static int lod_declare_update_plain(const struct lu_env *env,
5836 struct lod_object *lo, struct layout_intent *layout,
5837 const struct lu_buf *buf, struct thandle *th)
5839 struct lod_thread_info *info = lod_env_info(env);
5840 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5841 struct lod_layout_component *lod_comp;
5842 struct lov_comp_md_v1 *comp_v1 = NULL;
5843 bool replay = false;
5847 LASSERT(lo->ldo_flr_state == LCM_FL_NONE);
5850 * In case the client is passing lovea, which only happens during
5851 * the replay of layout intent write RPC for now, we may need to
5852 * parse the lovea and apply new layout configuration.
5854 if (buf && buf->lb_len) {
5855 struct lov_user_md_v1 *v1 = buf->lb_buf;
5857 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
5858 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
5859 LOV_MAGIC_COMP_V1)) {
5860 CERROR("%s: the replay buffer of layout extend "
5861 "(magic %#x) does not contain expected "
5862 "composite layout.\n",
5863 lod2obd(d)->obd_name, v1->lmm_magic);
5864 GOTO(out, rc = -EINVAL);
5867 rc = lod_use_defined_striping(env, lo, buf);
5870 lo->ldo_comp_cached = 1;
5872 rc = lod_get_lov_ea(env, lo);
5875 /* old on-disk EA is stored in info->lti_buf */
5876 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
5879 /* non replay path */
5880 rc = lod_striping_load(env, lo);
5885 /* Make sure defined layout covers the requested write range. */
5886 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
5887 if (lo->ldo_comp_cnt > 1 &&
5888 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
5889 lod_comp->llc_extent.e_end < layout->li_extent.e_end) {
5890 CDEBUG(replay ? D_ERROR : D_LAYOUT,
5891 "%s: the defined layout [0, %#llx) does not covers "
5892 "the write range "DEXT"\n",
5893 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
5894 PEXT(&layout->li_extent));
5895 GOTO(out, rc = -EINVAL);
5898 CDEBUG(D_LAYOUT, "%s: "DFID": instantiate components "DEXT"\n",
5899 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
5900 PEXT(&layout->li_extent));
5903 * Iterate ld->ldo_comp_entries, find the component whose extent under
5904 * the write range and not instantianted.
5906 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5907 lod_comp = &lo->ldo_comp_entries[i];
5909 if (lod_comp->llc_extent.e_start >= layout->li_extent.e_end)
5913 if (lod_comp_inited(lod_comp))
5917 * In replay path, lod_comp is the EA passed by
5918 * client replay buffer, comp_v1 is the pre-recovery
5919 * on-disk EA, we'd sift out those components which
5920 * were init-ed in the on-disk EA.
5922 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
5927 * this component hasn't instantiated in normal path, or during
5928 * replay it needs replay the instantiation.
5931 /* A released component is being extended */
5932 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
5933 GOTO(out, rc = -EINVAL);
5935 LASSERT(info->lti_comp_idx != NULL);
5936 info->lti_comp_idx[info->lti_count++] = i;
5939 if (info->lti_count == 0)
5942 lod_obj_inc_layout_gen(lo);
5943 rc = lod_declare_instantiate_components(env, lo, th);
5946 lod_striping_free(env, lo);
5950 static inline int lod_comp_index(struct lod_object *lo,
5951 struct lod_layout_component *lod_comp)
5953 LASSERT(lod_comp >= lo->ldo_comp_entries &&
5954 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
5956 return lod_comp - lo->ldo_comp_entries;
5960 * Stale other mirrors by writing extent.
5962 static void lod_stale_components(struct lod_object *lo, int primary,
5963 struct lu_extent *extent)
5965 struct lod_layout_component *pri_comp, *lod_comp;
5968 /* The writing extent decides which components in the primary
5969 * are affected... */
5970 CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
5971 lod_foreach_mirror_comp(pri_comp, lo, primary) {
5972 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
5975 CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
5976 lod_comp_index(lo, pri_comp),
5977 PEXT(&pri_comp->llc_extent));
5979 for (i = 0; i < lo->ldo_mirror_count; i++) {
5983 /* ... and then stale other components that are
5984 * overlapping with primary components */
5985 lod_foreach_mirror_comp(lod_comp, lo, i) {
5986 if (!lu_extent_is_overlapped(
5987 &pri_comp->llc_extent,
5988 &lod_comp->llc_extent))
5991 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
5992 i, lod_comp_index(lo, lod_comp));
5994 lod_comp->llc_flags |= LCME_FL_STALE;
5995 lo->ldo_mirrors[i].lme_stale = 1;
6002 * check an OST's availability
6003 * \param[in] env execution environment
6004 * \param[in] lo lod object
6005 * \param[in] dt dt object
6006 * \param[in] index mirror index
6008 * \retval negative if failed
6009 * \retval 1 if \a dt is available
6010 * \retval 0 if \a dt is not available
6012 static inline int lod_check_ost_avail(const struct lu_env *env,
6013 struct lod_object *lo,
6014 struct dt_object *dt, int index)
6016 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6017 struct lod_tgt_desc *ost;
6019 int type = LU_SEQ_RANGE_OST;
6022 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &idx, &type);
6024 CERROR("%s: can't locate "DFID":rc = %d\n",
6025 lod2obd(lod)->obd_name, PFID(lu_object_fid(&dt->do_lu)),
6030 ost = OST_TGT(lod, idx);
6031 if (ost->ltd_statfs.os_state &
6032 (OS_STATE_READONLY | OS_STATE_ENOSPC | OS_STATE_ENOINO |
6033 OS_STATE_NOPRECREATE) ||
6034 ost->ltd_active == 0) {
6035 CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail, rc = %d\n",
6036 PFID(lod_object_fid(lo)), index, idx, rc);
6044 * Pick primary mirror for write
6045 * \param[in] env execution environment
6046 * \param[in] lo object
6047 * \param[in] extent write range
6049 static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo,
6050 struct lu_extent *extent)
6052 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6053 unsigned int seq = 0;
6054 struct lod_layout_component *lod_comp;
6056 int picked = -1, second_pick = -1, third_pick = -1;
6059 if (OBD_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
6060 get_random_bytes(&seq, sizeof(seq));
6061 seq %= lo->ldo_mirror_count;
6065 * Pick a mirror as the primary, and check the availability of OSTs.
6067 * This algo can be revised later after knowing the topology of
6070 lod_qos_statfs_update(env, lod);
6071 for (i = 0; i < lo->ldo_mirror_count; i++) {
6072 bool ost_avail = true;
6073 int index = (i + seq) % lo->ldo_mirror_count;
6075 if (lo->ldo_mirrors[index].lme_stale) {
6076 CDEBUG(D_LAYOUT, DFID": mirror %d stale\n",
6077 PFID(lod_object_fid(lo)), index);
6081 /* 2nd pick is for the primary mirror containing unavail OST */
6082 if (lo->ldo_mirrors[index].lme_primary && second_pick < 0)
6083 second_pick = index;
6085 /* 3rd pick is for non-primary mirror containing unavail OST */
6086 if (second_pick < 0 && third_pick < 0)
6090 * we found a non-primary 1st pick, we'd like to find a
6091 * potential pirmary mirror.
6093 if (picked >= 0 && !lo->ldo_mirrors[index].lme_primary)
6096 /* check the availability of OSTs */
6097 lod_foreach_mirror_comp(lod_comp, lo, index) {
6098 if (!lod_comp_inited(lod_comp) || !lod_comp->llc_stripe)
6101 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6102 struct dt_object *dt = lod_comp->llc_stripe[j];
6104 rc = lod_check_ost_avail(env, lo, dt, index);
6111 } /* for all dt object in one component */
6114 } /* for all components in a mirror */
6117 * the OSTs where allocated objects locates in the components
6118 * of the mirror are available.
6123 /* this mirror has all OSTs available */
6127 * primary with all OSTs are available, this is the perfect
6130 if (lo->ldo_mirrors[index].lme_primary)
6132 } /* for all mirrors */
6134 /* failed to pick a sound mirror, lower our expectation */
6136 picked = second_pick;
6138 picked = third_pick;
6145 static int lod_prepare_resync_mirror(const struct lu_env *env,
6146 struct lod_object *lo,
6149 struct lod_thread_info *info = lod_env_info(env);
6150 struct lod_layout_component *lod_comp;
6151 bool neg = !!(MIRROR_ID_NEG & mirror_id);
6154 mirror_id &= ~MIRROR_ID_NEG;
6156 for (i = 0; i < lo->ldo_mirror_count; i++) {
6157 if ((!neg && lo->ldo_mirrors[i].lme_id != mirror_id) ||
6158 (neg && lo->ldo_mirrors[i].lme_id == mirror_id))
6161 lod_foreach_mirror_comp(lod_comp, lo, i) {
6162 if (lod_comp_inited(lod_comp))
6165 info->lti_comp_idx[info->lti_count++] =
6166 lod_comp_index(lo, lod_comp);
6174 * figure out the components should be instantiated for resync.
6176 static int lod_prepare_resync(const struct lu_env *env, struct lod_object *lo,
6177 struct lu_extent *extent)
6179 struct lod_thread_info *info = lod_env_info(env);
6180 struct lod_layout_component *lod_comp;
6181 unsigned int need_sync = 0;
6185 DFID": instantiate all stale components in "DEXT"\n",
6186 PFID(lod_object_fid(lo)), PEXT(extent));
6189 * instantiate all components within this extent, even non-stale
6192 for (i = 0; i < lo->ldo_mirror_count; i++) {
6193 if (!lo->ldo_mirrors[i].lme_stale)
6196 lod_foreach_mirror_comp(lod_comp, lo, i) {
6197 if (!lu_extent_is_overlapped(extent,
6198 &lod_comp->llc_extent))
6203 if (lod_comp_inited(lod_comp))
6206 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
6207 i, lod_comp_index(lo, lod_comp));
6208 info->lti_comp_idx[info->lti_count++] =
6209 lod_comp_index(lo, lod_comp);
6213 return need_sync ? 0 : -EALREADY;
6216 static int lod_declare_update_rdonly(const struct lu_env *env,
6217 struct lod_object *lo, struct md_layout_change *mlc,
6220 struct lod_thread_info *info = lod_env_info(env);
6221 struct lu_attr *layout_attr = &info->lti_layout_attr;
6222 struct lod_layout_component *lod_comp;
6223 struct lu_extent extent = { 0 };
6227 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
6228 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
6229 mlc->mlc_opc == MD_LAYOUT_RESYNC);
6230 LASSERT(lo->ldo_mirror_count > 0);
6232 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
6233 struct layout_intent *layout = mlc->mlc_intent;
6236 extent = layout->li_extent;
6237 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
6238 PFID(lod_object_fid(lo)), PEXT(&extent));
6240 picked = lod_primary_pick(env, lo, &extent);
6244 CDEBUG(D_LAYOUT, DFID": picked mirror id %u as primary\n",
6245 PFID(lod_object_fid(lo)),
6246 lo->ldo_mirrors[picked].lme_id);
6248 if (layout->li_opc == LAYOUT_INTENT_TRUNC) {
6250 * trunc transfers [0, size) in the intent extent, we'd
6251 * stale components overlapping [size, eof).
6253 extent.e_start = extent.e_end;
6254 extent.e_end = OBD_OBJECT_EOF;
6257 /* stale overlapping components from other mirrors */
6258 lod_stale_components(lo, picked, &extent);
6260 /* restore truncate intent extent */
6261 if (layout->li_opc == LAYOUT_INTENT_TRUNC)
6262 extent.e_end = extent.e_start;
6264 /* instantiate components for the picked mirror, start from 0 */
6267 lod_foreach_mirror_comp(lod_comp, lo, picked) {
6268 if (!lu_extent_is_overlapped(&extent,
6269 &lod_comp->llc_extent))
6272 if (lod_comp_inited(lod_comp))
6275 info->lti_comp_idx[info->lti_count++] =
6276 lod_comp_index(lo, lod_comp);
6279 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
6280 } else { /* MD_LAYOUT_RESYNC */
6284 * could contain multiple non-stale mirrors, so we need to
6285 * prep uninited all components assuming any non-stale mirror
6286 * could be picked as the primary mirror.
6288 if (mlc->mlc_mirror_id == 0) {
6290 for (i = 0; i < lo->ldo_mirror_count; i++) {
6291 if (lo->ldo_mirrors[i].lme_stale)
6294 lod_foreach_mirror_comp(lod_comp, lo, i) {
6295 if (!lod_comp_inited(lod_comp))
6299 lod_comp->llc_extent.e_end)
6301 lod_comp->llc_extent.e_end;
6304 rc = lod_prepare_resync(env, lo, &extent);
6308 /* mirror write, try to init its all components */
6309 rc = lod_prepare_resync_mirror(env, lo,
6310 mlc->mlc_mirror_id);
6315 /* change the file state to SYNC_PENDING */
6316 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
6319 /* Reset the layout version once it's becoming too large.
6320 * This way it can make sure that the layout version is
6321 * monotonously increased in this writing era. */
6322 lod_obj_inc_layout_gen(lo);
6323 if (lo->ldo_layout_gen > (LCME_ID_MAX >> 1)) {
6324 __u32 layout_version;
6326 cfs_get_random_bytes(&layout_version, sizeof(layout_version));
6327 lo->ldo_layout_gen = layout_version & 0xffff;
6330 rc = lod_declare_instantiate_components(env, lo, th);
6334 layout_attr->la_valid = LA_LAYOUT_VERSION;
6335 layout_attr->la_layout_version = 0; /* set current version */
6336 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
6337 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
6338 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
6344 lod_striping_free(env, lo);
6348 static int lod_declare_update_write_pending(const struct lu_env *env,
6349 struct lod_object *lo, struct md_layout_change *mlc,
6352 struct lod_thread_info *info = lod_env_info(env);
6353 struct lu_attr *layout_attr = &info->lti_layout_attr;
6354 struct lod_layout_component *lod_comp;
6355 struct lu_extent extent = { 0 };
6361 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
6362 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
6363 mlc->mlc_opc == MD_LAYOUT_RESYNC);
6365 /* look for the primary mirror */
6366 for (i = 0; i < lo->ldo_mirror_count; i++) {
6367 if (lo->ldo_mirrors[i].lme_stale)
6370 LASSERTF(primary < 0, DFID " has multiple primary: %u / %u",
6371 PFID(lod_object_fid(lo)),
6372 lo->ldo_mirrors[i].lme_id,
6373 lo->ldo_mirrors[primary].lme_id);
6378 CERROR(DFID ": doesn't have a primary mirror\n",
6379 PFID(lod_object_fid(lo)));
6380 GOTO(out, rc = -ENODATA);
6383 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
6384 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
6386 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
6388 /* for LAYOUT_WRITE opc, it has to do the following operations:
6389 * 1. stale overlapping componets from stale mirrors;
6390 * 2. instantiate components of the primary mirror;
6391 * 3. transfter layout version to all objects of the primary;
6393 * for LAYOUT_RESYNC opc, it will do:
6394 * 1. instantiate components of all stale mirrors;
6395 * 2. transfer layout version to all objects to close write era. */
6397 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
6398 LASSERT(mlc->mlc_intent != NULL);
6400 extent = mlc->mlc_intent->li_extent;
6402 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
6403 PFID(lod_object_fid(lo)), PEXT(&extent));
6405 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC) {
6407 * trunc transfers [0, size) in the intent extent, we'd
6408 * stale components overlapping [size, eof).
6410 extent.e_start = extent.e_end;
6411 extent.e_end = OBD_OBJECT_EOF;
6413 /* 1. stale overlapping components */
6414 lod_stale_components(lo, primary, &extent);
6416 /* 2. find out the components need instantiating.
6417 * instantiate [0, mlc->mlc_intent->e_end) */
6419 /* restore truncate intent extent */
6420 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC)
6421 extent.e_end = extent.e_start;
6424 lod_foreach_mirror_comp(lod_comp, lo, primary) {
6425 if (!lu_extent_is_overlapped(&extent,
6426 &lod_comp->llc_extent))
6429 if (lod_comp_inited(lod_comp))
6432 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
6433 primary, lod_comp_index(lo, lod_comp));
6434 info->lti_comp_idx[info->lti_count++] =
6435 lod_comp_index(lo, lod_comp);
6437 } else { /* MD_LAYOUT_RESYNC */
6438 if (mlc->mlc_mirror_id == 0) {
6440 lod_foreach_mirror_comp(lod_comp, lo, primary) {
6441 if (!lod_comp_inited(lod_comp))
6444 extent.e_end = lod_comp->llc_extent.e_end;
6447 rc = lod_prepare_resync(env, lo, &extent);
6451 /* mirror write, try to init its all components */
6452 rc = lod_prepare_resync_mirror(env, lo,
6453 mlc->mlc_mirror_id);
6458 /* change the file state to SYNC_PENDING */
6459 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
6462 rc = lod_declare_instantiate_components(env, lo, th);
6466 /* 3. transfer layout version to OST objects.
6467 * transfer new layout version to OST objects so that stale writes
6468 * can be denied. It also ends an era of writing by setting
6469 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
6470 * send write RPC; only resync RPCs could do it. */
6471 layout_attr->la_valid = LA_LAYOUT_VERSION;
6472 layout_attr->la_layout_version = 0; /* set current version */
6473 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
6474 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
6475 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
6479 lod_obj_inc_layout_gen(lo);
6482 lod_striping_free(env, lo);
6486 static int lod_declare_update_sync_pending(const struct lu_env *env,
6487 struct lod_object *lo, struct md_layout_change *mlc,
6490 struct lod_thread_info *info = lod_env_info(env);
6491 unsigned sync_components = 0;
6492 unsigned resync_components = 0;
6497 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
6498 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
6499 mlc->mlc_opc == MD_LAYOUT_WRITE);
6501 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
6502 PFID(lod_object_fid(lo)), mlc->mlc_opc);
6504 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
6505 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
6506 PFID(lod_object_fid(lo)));
6508 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
6509 return lod_declare_update_write_pending(env, lo, mlc, th);
6512 /* MD_LAYOUT_RESYNC_DONE */
6514 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6515 struct lod_layout_component *lod_comp;
6518 lod_comp = &lo->ldo_comp_entries[i];
6520 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
6525 for (j = 0; j < mlc->mlc_resync_count; j++) {
6526 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
6529 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
6530 lod_comp->llc_flags &= ~LCME_FL_STALE;
6531 resync_components++;
6537 for (i = 0; i < mlc->mlc_resync_count; i++) {
6538 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
6541 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
6542 "or already synced\n", PFID(lod_object_fid(lo)),
6543 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
6544 GOTO(out, rc = -EINVAL);
6547 if (!sync_components || (mlc->mlc_resync_count && !resync_components)) {
6548 CDEBUG(D_LAYOUT, DFID": no mirror in sync\n",
6549 PFID(lod_object_fid(lo)));
6551 /* tend to return an error code here to prevent
6552 * the MDT from setting SoM attribute */
6553 GOTO(out, rc = -EINVAL);
6556 CDEBUG(D_LAYOUT, DFID": resynced %u/%zu components\n",
6557 PFID(lod_object_fid(lo)),
6558 resync_components, mlc->mlc_resync_count);
6560 lo->ldo_flr_state = LCM_FL_RDONLY;
6561 lod_obj_inc_layout_gen(lo);
6563 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6564 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
6565 &info->lti_buf, XATTR_NAME_LOV, 0, th);
6570 lod_striping_free(env, lo);
6574 static int lod_declare_layout_change(const struct lu_env *env,
6575 struct dt_object *dt, struct md_layout_change *mlc,
6578 struct lod_thread_info *info = lod_env_info(env);
6579 struct lod_object *lo = lod_dt_obj(dt);
6583 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
6584 dt_object_remote(dt_object_child(dt)))
6587 rc = lod_striping_load(env, lo);
6591 LASSERT(lo->ldo_comp_cnt > 0);
6593 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
6597 switch (lo->ldo_flr_state) {
6599 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
6603 rc = lod_declare_update_rdonly(env, lo, mlc, th);
6605 case LCM_FL_WRITE_PENDING:
6606 rc = lod_declare_update_write_pending(env, lo, mlc, th);
6608 case LCM_FL_SYNC_PENDING:
6609 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
6620 * Instantiate layout component objects which covers the intent write offset.
6622 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
6623 struct md_layout_change *mlc, struct thandle *th)
6625 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
6626 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
6627 struct lod_object *lo = lod_dt_obj(dt);
6630 rc = lod_striped_create(env, dt, attr, NULL, th);
6631 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
6632 layout_attr->la_layout_version |= lo->ldo_layout_gen;
6633 rc = lod_attr_set(env, dt, layout_attr, th);
6639 struct dt_object_operations lod_obj_ops = {
6640 .do_read_lock = lod_read_lock,
6641 .do_write_lock = lod_write_lock,
6642 .do_read_unlock = lod_read_unlock,
6643 .do_write_unlock = lod_write_unlock,
6644 .do_write_locked = lod_write_locked,
6645 .do_attr_get = lod_attr_get,
6646 .do_declare_attr_set = lod_declare_attr_set,
6647 .do_attr_set = lod_attr_set,
6648 .do_xattr_get = lod_xattr_get,
6649 .do_declare_xattr_set = lod_declare_xattr_set,
6650 .do_xattr_set = lod_xattr_set,
6651 .do_declare_xattr_del = lod_declare_xattr_del,
6652 .do_xattr_del = lod_xattr_del,
6653 .do_xattr_list = lod_xattr_list,
6654 .do_ah_init = lod_ah_init,
6655 .do_declare_create = lod_declare_create,
6656 .do_create = lod_create,
6657 .do_declare_destroy = lod_declare_destroy,
6658 .do_destroy = lod_destroy,
6659 .do_index_try = lod_index_try,
6660 .do_declare_ref_add = lod_declare_ref_add,
6661 .do_ref_add = lod_ref_add,
6662 .do_declare_ref_del = lod_declare_ref_del,
6663 .do_ref_del = lod_ref_del,
6664 .do_object_sync = lod_object_sync,
6665 .do_object_lock = lod_object_lock,
6666 .do_object_unlock = lod_object_unlock,
6667 .do_invalidate = lod_invalidate,
6668 .do_declare_layout_change = lod_declare_layout_change,
6669 .do_layout_change = lod_layout_change,
6673 * Implementation of dt_body_operations::dbo_read.
6675 * \see dt_body_operations::dbo_read() in the API description for details.
6677 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
6678 struct lu_buf *buf, loff_t *pos)
6680 struct dt_object *next = dt_object_child(dt);
6682 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
6683 S_ISLNK(dt->do_lu.lo_header->loh_attr));
6684 return next->do_body_ops->dbo_read(env, next, buf, pos);
6688 * Implementation of dt_body_operations::dbo_declare_write.
6690 * \see dt_body_operations::dbo_declare_write() in the API description
6693 static ssize_t lod_declare_write(const struct lu_env *env,
6694 struct dt_object *dt,
6695 const struct lu_buf *buf, loff_t pos,
6698 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
6702 * Implementation of dt_body_operations::dbo_write.
6704 * \see dt_body_operations::dbo_write() in the API description for details.
6706 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
6707 const struct lu_buf *buf, loff_t *pos,
6710 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
6711 S_ISLNK(dt->do_lu.lo_header->loh_attr));
6712 return lod_sub_write(env, dt_object_child(dt), buf, pos, th);
6715 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
6716 __u64 start, __u64 end, struct thandle *th)
6718 if (dt_object_remote(dt))
6721 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
6724 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
6725 __u64 start, __u64 end, struct thandle *th)
6727 if (dt_object_remote(dt))
6730 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
6731 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
6735 * different type of files use the same body_ops because object may be created
6736 * in OUT, where there is no chance to set correct body_ops for each type, so
6737 * body_ops themselves will check file type inside, see lod_read/write/punch for
6740 const struct dt_body_operations lod_body_ops = {
6741 .dbo_read = lod_read,
6742 .dbo_declare_write = lod_declare_write,
6743 .dbo_write = lod_write,
6744 .dbo_declare_punch = lod_declare_punch,
6745 .dbo_punch = lod_punch,
6749 * Implementation of lu_object_operations::loo_object_init.
6751 * The function determines the type and the index of the target device using
6752 * sequence of the object's FID. Then passes control down to the
6753 * corresponding device:
6754 * OSD for the local objects, OSP for remote
6756 * \see lu_object_operations::loo_object_init() in the API description
6759 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
6760 const struct lu_object_conf *conf)
6762 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
6763 struct lu_device *cdev = NULL;
6764 struct lu_object *cobj;
6765 struct lod_tgt_descs *ltd = NULL;
6766 struct lod_tgt_desc *tgt;
6768 int type = LU_SEQ_RANGE_ANY;
6772 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
6774 /* Note: Sometimes, it will Return EAGAIN here, see
6775 * ptrlpc_import_delay_req(), which might confuse
6776 * lu_object_find_at() and make it wait there incorrectly.
6777 * so we convert it to EIO here.*/
6784 if (type == LU_SEQ_RANGE_MDT &&
6785 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
6786 cdev = &lod->lod_child->dd_lu_dev;
6787 } else if (type == LU_SEQ_RANGE_MDT) {
6788 ltd = &lod->lod_mdt_descs;
6790 } else if (type == LU_SEQ_RANGE_OST) {
6791 ltd = &lod->lod_ost_descs;
6798 if (ltd->ltd_tgts_size > idx &&
6799 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
6800 tgt = LTD_TGT(ltd, idx);
6802 LASSERT(tgt != NULL);
6803 LASSERT(tgt->ltd_tgt != NULL);
6805 cdev = &(tgt->ltd_tgt->dd_lu_dev);
6807 lod_putref(lod, ltd);
6810 if (unlikely(cdev == NULL))
6813 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
6814 if (unlikely(cobj == NULL))
6817 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
6819 lu_object_add(lo, cobj);
6826 * Alloc cached foreign LOV
6828 * \param[in] lo object
6829 * \param[in] size size of foreign LOV
6831 * \retval 0 on success
6832 * \retval negative if failed
6834 int lod_alloc_foreign_lov(struct lod_object *lo, size_t size)
6836 OBD_ALLOC_LARGE(lo->ldo_foreign_lov, size);
6837 if (lo->ldo_foreign_lov == NULL)
6839 lo->ldo_foreign_lov_size = size;
6840 lo->ldo_is_foreign = 1;
6846 * Free cached foreign LOV
6848 * \param[in] lo object
6850 void lod_free_foreign_lov(struct lod_object *lo)
6852 if (lo->ldo_foreign_lov != NULL)
6853 OBD_FREE_LARGE(lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
6854 lo->ldo_foreign_lov = NULL;
6855 lo->ldo_foreign_lov_size = 0;
6856 lo->ldo_is_foreign = 0;
6861 * Release resources associated with striping.
6863 * If the object is striped (regular or directory), then release
6864 * the stripe objects references and free the ldo_stripe array.
6866 * \param[in] env execution environment
6867 * \param[in] lo object
6869 void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo)
6871 struct lod_layout_component *lod_comp;
6874 if (unlikely(lo->ldo_is_foreign)) {
6875 lod_free_foreign_lov(lo);
6876 lo->ldo_comp_cached = 0;
6877 } else if (lo->ldo_stripe != NULL) {
6878 LASSERT(lo->ldo_comp_entries == NULL);
6879 LASSERT(lo->ldo_dir_stripes_allocated > 0);
6881 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6882 if (lo->ldo_stripe[i])
6883 dt_object_put(env, lo->ldo_stripe[i]);
6886 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
6887 OBD_FREE(lo->ldo_stripe, j);
6888 lo->ldo_stripe = NULL;
6889 lo->ldo_dir_stripes_allocated = 0;
6890 lo->ldo_dir_stripe_loaded = 0;
6891 lo->ldo_dir_stripe_count = 0;
6892 } else if (lo->ldo_comp_entries != NULL) {
6893 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6894 /* free lod_layout_component::llc_stripe array */
6895 lod_comp = &lo->ldo_comp_entries[i];
6897 if (lod_comp->llc_stripe == NULL)
6899 LASSERT(lod_comp->llc_stripes_allocated != 0);
6900 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
6901 if (lod_comp->llc_stripe[j] != NULL)
6903 &lod_comp->llc_stripe[j]->do_lu);
6905 OBD_FREE(lod_comp->llc_stripe,
6906 sizeof(struct dt_object *) *
6907 lod_comp->llc_stripes_allocated);
6908 lod_comp->llc_stripe = NULL;
6909 OBD_FREE(lod_comp->llc_ost_indices,
6911 lod_comp->llc_stripes_allocated);
6912 lod_comp->llc_ost_indices = NULL;
6913 lod_comp->llc_stripes_allocated = 0;
6915 lod_free_comp_entries(lo);
6916 lo->ldo_comp_cached = 0;
6920 void lod_striping_free(const struct lu_env *env, struct lod_object *lo)
6922 mutex_lock(&lo->ldo_layout_mutex);
6923 lod_striping_free_nolock(env, lo);
6924 mutex_unlock(&lo->ldo_layout_mutex);
6928 * Implementation of lu_object_operations::loo_object_free.
6930 * \see lu_object_operations::loo_object_free() in the API description
6933 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
6935 struct lod_object *lo = lu2lod_obj(o);
6937 /* release all underlying object pinned */
6938 lod_striping_free(env, lo);
6940 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
6944 * Implementation of lu_object_operations::loo_object_release.
6946 * \see lu_object_operations::loo_object_release() in the API description
6949 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
6951 /* XXX: shouldn't we release everything here in case if object
6952 * creation failed before? */
6956 * Implementation of lu_object_operations::loo_object_print.
6958 * \see lu_object_operations::loo_object_print() in the API description
6961 static int lod_object_print(const struct lu_env *env, void *cookie,
6962 lu_printer_t p, const struct lu_object *l)
6964 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
6966 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
6969 struct lu_object_operations lod_lu_obj_ops = {
6970 .loo_object_init = lod_object_init,
6971 .loo_object_free = lod_object_free,
6972 .loo_object_release = lod_object_release,
6973 .loo_object_print = lod_object_print,