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;
1457 struct lmv_foreign_md *lfm;
1460 if (rc > (typeof(rc))sizeof(*lmv1))
1463 /* short (<= sizeof(struct lmv_mds_md_v1)) foreign LMV case */
1464 /* XXX empty foreign LMV is not allowed */
1465 if (rc <= offsetof(typeof(*lfm), lfm_value))
1466 RETURN(rc = rc > 0 ? -EINVAL : rc);
1468 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1469 CLASSERT(sizeof(*lmv1) <= sizeof(info->lti_key));
1471 /* lti_buf is large enough for *lmv1 or a short
1472 * (<= sizeof(struct lmv_mds_md_v1)) foreign LMV
1474 info->lti_buf.lb_buf = info->lti_key;
1475 info->lti_buf.lb_len = sizeof(*lmv1);
1476 rc = dt_xattr_get(env, dt_object_child(dt),
1477 &info->lti_buf, name);
1478 if (unlikely(rc <= offsetof(typeof(*lfm),
1480 RETURN(rc = rc > 0 ? -EINVAL : rc);
1482 lfm = info->lti_buf.lb_buf;
1483 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN)
1486 if (unlikely(rc != sizeof(*lmv1)))
1487 RETURN(rc = rc > 0 ? -EINVAL : rc);
1489 lmv1 = info->lti_buf.lb_buf;
1490 /* The on-disk LMV EA only contains header, but the
1491 * returned LMV EA size should contain the space for
1492 * the FIDs of all shards of the striped directory. */
1493 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1494 rc = lmv_mds_md_size(
1495 le32_to_cpu(lmv1->lmv_stripe_count),
1499 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN)
1502 if (rc != sizeof(*lmv1))
1503 RETURN(rc = rc > 0 ? -EINVAL : rc);
1505 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1509 RETURN(rc = rc1 != 0 ? rc1 : rc);
1512 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1516 * XXX: Only used by lfsck
1518 * lod returns default striping on the real root of the device
1519 * this is like the root stores default striping for the whole
1520 * filesystem. historically we've been using a different approach
1521 * and store it in the config.
1523 dt_root_get(env, dev->lod_child, &info->lti_fid);
1524 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1526 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1527 struct lov_user_md *lum = buf->lb_buf;
1528 struct lov_desc *desc = &dev->lod_desc;
1530 if (buf->lb_buf == NULL) {
1532 } else if (buf->lb_len >= sizeof(*lum)) {
1533 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1534 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1535 lmm_oi_set_id(&lum->lmm_oi, 0);
1536 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1537 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1538 lum->lmm_stripe_size = cpu_to_le32(
1539 desc->ld_default_stripe_size);
1540 lum->lmm_stripe_count = cpu_to_le16(
1541 desc->ld_default_stripe_count);
1542 lum->lmm_stripe_offset = cpu_to_le16(
1543 desc->ld_default_stripe_offset);
1556 * Checks that the magic of the stripe is sane.
1558 * \param[in] lod lod device
1559 * \param[in] lum a buffer storing LMV EA to verify
1561 * \retval 0 if the EA is sane
1562 * \retval negative otherwise
1564 static int lod_verify_md_striping(struct lod_device *lod,
1565 const struct lmv_user_md_v1 *lum)
1567 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1568 CERROR("%s: invalid lmv_user_md: magic = %x, "
1569 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1570 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1571 (int)le32_to_cpu(lum->lum_stripe_offset),
1572 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1580 * Initialize LMV EA for a slave.
1582 * Initialize slave's LMV EA from the master's LMV EA.
1584 * \param[in] master_lmv a buffer containing master's EA
1585 * \param[out] slave_lmv a buffer where slave's EA will be stored
1588 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1589 const struct lmv_mds_md_v1 *master_lmv)
1591 *slave_lmv = *master_lmv;
1592 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1598 * Generate LMV EA from the object passed as \a dt. The object must have
1599 * the stripes created and initialized.
1601 * \param[in] env execution environment
1602 * \param[in] dt object
1603 * \param[out] lmv_buf buffer storing generated LMV EA
1605 * \retval 0 on success
1606 * \retval negative if failed
1608 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1609 struct lu_buf *lmv_buf)
1611 struct lod_thread_info *info = lod_env_info(env);
1612 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1613 struct lod_object *lo = lod_dt_obj(dt);
1614 struct lmv_mds_md_v1 *lmm1;
1616 int type = LU_SEQ_RANGE_ANY;
1621 LASSERT(lo->ldo_dir_striped != 0);
1622 LASSERT(lo->ldo_dir_stripe_count > 0);
1623 stripe_count = lo->ldo_dir_stripe_count;
1624 /* Only store the LMV EA heahder on the disk. */
1625 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1626 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1630 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1633 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1634 memset(lmm1, 0, sizeof(*lmm1));
1635 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1636 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1637 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1638 if (lo->ldo_dir_hash_type & LMV_HASH_FLAG_MIGRATION) {
1639 lmm1->lmv_migrate_hash = cpu_to_le32(lo->ldo_dir_migrate_hash);
1640 lmm1->lmv_migrate_offset =
1641 cpu_to_le32(lo->ldo_dir_migrate_offset);
1643 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1648 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1649 lmv_buf->lb_buf = info->lti_ea_store;
1650 lmv_buf->lb_len = sizeof(*lmm1);
1656 * Create in-core represenation for a striped directory.
1658 * Parse the buffer containing LMV EA and instantiate LU objects
1659 * representing the stripe objects. The pointers to the objects are
1660 * stored in ldo_stripe field of \a lo. This function is used when
1661 * we need to access an already created object (i.e. load from a disk).
1663 * \param[in] env execution environment
1664 * \param[in] lo lod object
1665 * \param[in] buf buffer containing LMV EA
1667 * \retval 0 on success
1668 * \retval negative if failed
1670 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1671 const struct lu_buf *buf)
1673 struct lod_thread_info *info = lod_env_info(env);
1674 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1675 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1676 struct dt_object **stripe;
1677 union lmv_mds_md *lmm = buf->lb_buf;
1678 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1679 struct lu_fid *fid = &info->lti_fid;
1684 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1686 /* XXX may be useless as not called for foreign LMV ?? */
1687 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_FOREIGN)
1690 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1691 lo->ldo_dir_slave_stripe = 1;
1695 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1698 if (le32_to_cpu(lmv1->lmv_stripe_count) < 1)
1701 LASSERT(lo->ldo_stripe == NULL);
1702 OBD_ALLOC(stripe, sizeof(stripe[0]) *
1703 (le32_to_cpu(lmv1->lmv_stripe_count)));
1707 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1708 struct dt_device *tgt_dt;
1709 struct dt_object *dto;
1710 int type = LU_SEQ_RANGE_ANY;
1713 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1714 if (!fid_is_sane(fid))
1715 GOTO(out, rc = -ESTALE);
1717 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1721 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1722 tgt_dt = lod->lod_child;
1724 struct lod_tgt_desc *tgt;
1726 tgt = LTD_TGT(ltd, idx);
1728 GOTO(out, rc = -ESTALE);
1729 tgt_dt = tgt->ltd_tgt;
1732 dto = dt_locate_at(env, tgt_dt, fid,
1733 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1736 GOTO(out, rc = PTR_ERR(dto));
1741 lo->ldo_stripe = stripe;
1742 lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1743 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1745 lod_striping_free_nolock(env, lo);
1751 * Declare create a striped directory.
1753 * Declare creating a striped directory with a given stripe pattern on the
1754 * specified MDTs. A striped directory is represented as a regular directory
1755 * - an index listing all the stripes. The stripes point back to the master
1756 * object with ".." and LinkEA. The master object gets LMV EA which
1757 * identifies it as a striped directory. The function allocates FIDs
1760 * \param[in] env execution environment
1761 * \param[in] dt object
1762 * \param[in] attr attributes to initialize the objects with
1763 * \param[in] dof type of objects to be created
1764 * \param[in] th transaction handle
1766 * \retval 0 on success
1767 * \retval negative if failed
1769 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1770 struct dt_object *dt,
1771 struct lu_attr *attr,
1772 struct dt_object_format *dof,
1775 struct lod_thread_info *info = lod_env_info(env);
1776 struct lu_buf lmv_buf;
1777 struct lu_buf slave_lmv_buf;
1778 struct lmv_mds_md_v1 *lmm;
1779 struct lmv_mds_md_v1 *slave_lmm = NULL;
1780 struct dt_insert_rec *rec = &info->lti_dt_rec;
1781 struct lod_object *lo = lod_dt_obj(dt);
1786 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1789 lmm = lmv_buf.lb_buf;
1791 OBD_ALLOC_PTR(slave_lmm);
1792 if (slave_lmm == NULL)
1793 GOTO(out, rc = -ENOMEM);
1795 lod_prep_slave_lmv_md(slave_lmm, lmm);
1796 slave_lmv_buf.lb_buf = slave_lmm;
1797 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1799 if (!dt_try_as_dir(env, dt_object_child(dt)))
1800 GOTO(out, rc = -EINVAL);
1802 rec->rec_type = S_IFDIR;
1803 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1804 struct dt_object *dto = lo->ldo_stripe[i];
1805 char *stripe_name = info->lti_key;
1806 struct lu_name *sname;
1807 struct linkea_data ldata = { NULL };
1808 struct lu_buf linkea_buf;
1810 rc = lod_sub_declare_create(env, dto, attr, NULL, dof, th);
1814 if (!dt_try_as_dir(env, dto))
1815 GOTO(out, rc = -EINVAL);
1817 rc = lod_sub_declare_ref_add(env, dto, th);
1821 rec->rec_fid = lu_object_fid(&dto->do_lu);
1822 rc = lod_sub_declare_insert(env, dto,
1823 (const struct dt_rec *)rec,
1824 (const struct dt_key *)dot, th);
1828 /* master stripe FID will be put to .. */
1829 rec->rec_fid = lu_object_fid(&dt->do_lu);
1830 rc = lod_sub_declare_insert(env, dto,
1831 (const struct dt_rec *)rec,
1832 (const struct dt_key *)dotdot, th);
1836 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1837 cfs_fail_val != i) {
1838 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1840 slave_lmm->lmv_master_mdt_index =
1843 slave_lmm->lmv_master_mdt_index =
1845 rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
1846 XATTR_NAME_LMV, 0, th);
1851 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1853 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1854 PFID(lu_object_fid(&dto->do_lu)), i + 1);
1856 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1857 PFID(lu_object_fid(&dto->do_lu)), i);
1859 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
1860 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1861 sname, lu_object_fid(&dt->do_lu));
1865 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1866 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1867 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
1868 XATTR_NAME_LINK, 0, th);
1872 rec->rec_fid = lu_object_fid(&dto->do_lu);
1873 rc = lod_sub_declare_insert(env, dt_object_child(dt),
1874 (const struct dt_rec *)rec,
1875 (const struct dt_key *)stripe_name,
1880 rc = lod_sub_declare_ref_add(env, dt_object_child(dt), th);
1885 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
1886 &lmv_buf, XATTR_NAME_LMV, 0, th);
1890 if (slave_lmm != NULL)
1891 OBD_FREE_PTR(slave_lmm);
1896 static int lod_prep_md_striped_create(const struct lu_env *env,
1897 struct dt_object *dt,
1898 struct lu_attr *attr,
1899 const struct lmv_user_md_v1 *lum,
1900 struct dt_object_format *dof,
1903 struct lod_thread_info *info = lod_env_info(env);
1904 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1905 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1906 struct lod_object *lo = lod_dt_obj(dt);
1907 struct dt_object **stripe;
1914 bool is_specific = false;
1917 /* The lum has been verifed in lod_verify_md_striping */
1918 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1919 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
1921 stripe_count = lo->ldo_dir_stripe_count;
1923 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
1924 if (idx_array == NULL)
1927 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_count);
1929 GOTO(out_free, rc = -ENOMEM);
1931 /* Start index must be the master MDT */
1932 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1933 idx_array[0] = master_index;
1934 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
1936 for (i = 1; i < stripe_count; i++)
1937 idx_array[i] = le32_to_cpu(lum->lum_objects[i].lum_mds);
1940 for (i = 0; i < stripe_count; i++) {
1941 struct lod_tgt_desc *tgt = NULL;
1942 struct dt_object *dto;
1943 struct lu_fid fid = { 0 };
1945 struct lu_object_conf conf = { 0 };
1946 struct dt_device *tgt_dt = NULL;
1948 /* Try to find next avaible target */
1950 for (j = 0; j < lod->lod_remote_mdt_count;
1951 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
1952 bool already_allocated = false;
1955 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
1956 idx, lod->lod_remote_mdt_count + 1, i);
1958 if (likely(!is_specific &&
1959 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
1960 /* check whether the idx already exists
1961 * in current allocated array */
1962 for (k = 0; k < i; k++) {
1963 if (idx_array[k] == idx) {
1964 already_allocated = true;
1969 if (already_allocated)
1973 /* Sigh, this index is not in the bitmap, let's check
1974 * next available target */
1975 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) &&
1976 idx != master_index)
1979 if (idx == master_index) {
1980 /* Allocate the FID locally */
1981 rc = obd_fid_alloc(env, lod->lod_child_exp,
1985 tgt_dt = lod->lod_child;
1989 /* check the status of the OSP */
1990 tgt = LTD_TGT(ltd, idx);
1994 tgt_dt = tgt->ltd_tgt;
1995 rc = dt_statfs(env, tgt_dt, &info->lti_osfs);
1997 /* this OSP doesn't feel well */
2002 rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
2011 /* Can not allocate more stripes */
2012 if (j == lod->lod_remote_mdt_count) {
2013 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
2014 lod2obd(lod)->obd_name, stripe_count, i);
2018 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
2019 idx, i, PFID(&fid));
2021 /* Set the start index for next stripe allocation */
2022 if (!is_specific && i < stripe_count - 1) {
2024 * for large dir test, put all other slaves on one
2025 * remote MDT, otherwise we may save too many local
2026 * slave locks which will exceed RS_MAX_LOCKS.
2028 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)))
2030 idx_array[i + 1] = (idx + 1) %
2031 (lod->lod_remote_mdt_count + 1);
2033 /* tgt_dt and fid must be ready after search avaible OSP
2034 * in the above loop */
2035 LASSERT(tgt_dt != NULL);
2036 LASSERT(fid_is_sane(&fid));
2037 conf.loc_flags = LOC_F_NEW;
2038 dto = dt_locate_at(env, tgt_dt, &fid,
2039 dt->do_lu.lo_dev->ld_site->ls_top_dev,
2042 GOTO(out_put, rc = PTR_ERR(dto));
2046 lo->ldo_dir_striped = 1;
2047 lo->ldo_stripe = stripe;
2048 lo->ldo_dir_stripe_count = i;
2049 lo->ldo_dir_stripes_allocated = stripe_count;
2051 lo->ldo_dir_stripe_loaded = 1;
2053 if (lo->ldo_dir_stripe_count == 0)
2054 GOTO(out_put, rc = -ENOSPC);
2056 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2062 for (i = 0; i < stripe_count; i++)
2063 if (stripe[i] != NULL)
2064 dt_object_put(env, stripe[i]);
2065 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_count);
2066 lo->ldo_dir_stripe_count = 0;
2067 lo->ldo_dir_stripes_allocated = 0;
2068 lo->ldo_stripe = NULL;
2072 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
2079 * Alloc cached foreign LMV
2081 * \param[in] lo object
2082 * \param[in] size size of foreign LMV
2084 * \retval 0 on success
2085 * \retval negative if failed
2087 int lod_alloc_foreign_lmv(struct lod_object *lo, size_t size)
2089 OBD_ALLOC_LARGE(lo->ldo_foreign_lmv, size);
2090 if (lo->ldo_foreign_lmv == NULL)
2092 lo->ldo_foreign_lmv_size = size;
2093 lo->ldo_dir_is_foreign = 1;
2099 * Declare create striped md object.
2101 * The function declares intention to create a striped directory. This is a
2102 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2103 * is to verify pattern \a lum_buf is good. Check that function for the details.
2105 * \param[in] env execution environment
2106 * \param[in] dt object
2107 * \param[in] attr attributes to initialize the objects with
2108 * \param[in] lum_buf a pattern specifying the number of stripes and
2110 * \param[in] dof type of objects to be created
2111 * \param[in] th transaction handle
2113 * \retval 0 on success
2114 * \retval negative if failed
2117 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2118 struct dt_object *dt,
2119 struct lu_attr *attr,
2120 const struct lu_buf *lum_buf,
2121 struct dt_object_format *dof,
2124 struct lod_object *lo = lod_dt_obj(dt);
2125 struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
2129 LASSERT(lum != NULL);
2131 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
2132 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
2133 (int)le32_to_cpu(lum->lum_stripe_offset));
2135 if (lo->ldo_dir_stripe_count == 0) {
2136 if (lo->ldo_dir_is_foreign) {
2137 rc = lod_alloc_foreign_lmv(lo, lum_buf->lb_len);
2140 memcpy(lo->ldo_foreign_lmv, lum, lum_buf->lb_len);
2141 lo->ldo_dir_stripe_loaded = 1;
2146 /* prepare dir striped objects */
2147 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2149 /* failed to create striping, let's reset
2150 * config so that others don't get confused */
2151 lod_striping_free(env, lo);
2159 * Append source stripes after target stripes for migrating directory. NB, we
2160 * only need to declare this, the append is done inside lod_xattr_set_lmv().
2162 * \param[in] env execution environment
2163 * \param[in] dt target object
2164 * \param[in] buf LMV buf which contains source stripe fids
2165 * \param[in] th transaction handle
2167 * \retval 0 on success
2168 * \retval negative if failed
2170 static int lod_dir_declare_layout_add(const struct lu_env *env,
2171 struct dt_object *dt,
2172 const struct lu_buf *buf,
2175 struct lod_thread_info *info = lod_env_info(env);
2176 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
2177 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
2178 struct lod_object *lo = lod_dt_obj(dt);
2179 struct dt_object *next = dt_object_child(dt);
2180 struct dt_object_format *dof = &info->lti_format;
2181 struct lmv_mds_md_v1 *lmv = buf->lb_buf;
2182 struct dt_object **stripe;
2183 __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
2184 struct lu_fid *fid = &info->lti_fid;
2185 struct lod_tgt_desc *tgt;
2186 struct dt_object *dto;
2187 struct dt_device *tgt_dt;
2188 int type = LU_SEQ_RANGE_ANY;
2189 struct dt_insert_rec *rec = &info->lti_dt_rec;
2190 char *stripe_name = info->lti_key;
2191 struct lu_name *sname;
2192 struct linkea_data ldata = { NULL };
2193 struct lu_buf linkea_buf;
2200 if (le32_to_cpu(lmv->lmv_magic) != LMV_MAGIC_V1)
2203 if (stripe_count == 0)
2206 dof->dof_type = DFT_DIR;
2209 sizeof(*stripe) * (lo->ldo_dir_stripe_count + stripe_count));
2213 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
2214 stripe[i] = lo->ldo_stripe[i];
2216 for (i = 0; i < stripe_count; i++) {
2218 &lmv->lmv_stripe_fids[i]);
2219 if (!fid_is_sane(fid))
2220 GOTO(out, rc = -ESTALE);
2222 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
2226 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
2227 tgt_dt = lod->lod_child;
2229 tgt = LTD_TGT(ltd, idx);
2231 GOTO(out, rc = -ESTALE);
2232 tgt_dt = tgt->ltd_tgt;
2235 dto = dt_locate_at(env, tgt_dt, fid,
2236 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
2239 GOTO(out, rc = PTR_ERR(dto));
2241 stripe[i + lo->ldo_dir_stripe_count] = dto;
2243 if (!dt_try_as_dir(env, dto))
2244 GOTO(out, rc = -ENOTDIR);
2246 rc = lod_sub_declare_ref_add(env, dto, th);
2250 rc = lod_sub_declare_insert(env, dto,
2251 (const struct dt_rec *)rec,
2252 (const struct dt_key *)dot, th);
2256 rc = lod_sub_declare_insert(env, dto,
2257 (const struct dt_rec *)rec,
2258 (const struct dt_key *)dotdot, th);
2262 rc = lod_sub_declare_xattr_set(env, dto, buf,
2263 XATTR_NAME_LMV, 0, th);
2267 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
2268 PFID(lu_object_fid(&dto->do_lu)),
2269 i + lo->ldo_dir_stripe_count);
2271 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
2272 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
2273 sname, lu_object_fid(&dt->do_lu));
2277 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
2278 linkea_buf.lb_len = ldata.ld_leh->leh_len;
2279 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
2280 XATTR_NAME_LINK, 0, th);
2284 rc = lod_sub_declare_insert(env, next,
2285 (const struct dt_rec *)rec,
2286 (const struct dt_key *)stripe_name,
2291 rc = lod_sub_declare_ref_add(env, next, th);
2297 OBD_FREE(lo->ldo_stripe,
2298 sizeof(*stripe) * lo->ldo_dir_stripes_allocated);
2299 lo->ldo_stripe = stripe;
2300 lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count;
2301 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type);
2302 lo->ldo_dir_stripe_count += stripe_count;
2303 lo->ldo_dir_stripes_allocated += stripe_count;
2304 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_MIGRATION;
2308 i = lo->ldo_dir_stripe_count;
2309 while (i < lo->ldo_dir_stripe_count + stripe_count && stripe[i])
2310 dt_object_put(env, stripe[i++]);
2313 sizeof(*stripe) * (stripe_count + lo->ldo_dir_stripe_count));
2317 static int lod_dir_declare_layout_delete(const struct lu_env *env,
2318 struct dt_object *dt,
2319 const struct lu_buf *buf,
2322 struct lod_thread_info *info = lod_env_info(env);
2323 struct lod_object *lo = lod_dt_obj(dt);
2324 struct dt_object *next = dt_object_child(dt);
2325 struct lmv_user_md *lmu = buf->lb_buf;
2326 __u32 final_stripe_count;
2327 char *stripe_name = info->lti_key;
2328 struct dt_object *dto;
2335 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
2336 if (final_stripe_count >= lo->ldo_dir_stripe_count)
2339 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
2340 dto = lo->ldo_stripe[i];
2343 if (!dt_try_as_dir(env, dto))
2346 rc = lod_sub_declare_delete(env, dto,
2347 (const struct dt_key *)dot, th);
2351 rc = lod_sub_declare_ref_del(env, dto, th);
2355 rc = lod_sub_declare_delete(env, dto,
2356 (const struct dt_key *)dotdot, th);
2360 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2361 PFID(lu_object_fid(&dto->do_lu)), i);
2363 rc = lod_sub_declare_delete(env, next,
2364 (const struct dt_key *)stripe_name, th);
2368 rc = lod_sub_declare_ref_del(env, next, th);
2377 * delete stripes from dir master object, the lum_stripe_count in argument is
2378 * the final stripe count, the stripes after that will be deleted, NB, they
2379 * are not destroyed, but deleted from it's parent namespace, this function
2380 * will be called in two places:
2381 * 1. mdd_migrate_create() delete stripes from source, and append them to
2383 * 2. mdd_dir_layout_shrink() delete stripes from source, and destroy them.
2385 static int lod_dir_layout_delete(const struct lu_env *env,
2386 struct dt_object *dt,
2387 const struct lu_buf *buf,
2390 struct lod_thread_info *info = lod_env_info(env);
2391 struct lod_object *lo = lod_dt_obj(dt);
2392 struct dt_object *next = dt_object_child(dt);
2393 struct lmv_user_md *lmu = buf->lb_buf;
2394 __u32 final_stripe_count;
2395 char *stripe_name = info->lti_key;
2396 struct dt_object *dto;
2405 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
2406 if (final_stripe_count >= lo->ldo_dir_stripe_count)
2409 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
2410 dto = lo->ldo_stripe[i];
2413 rc = lod_sub_delete(env, dto,
2414 (const struct dt_key *)dotdot, th);
2418 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2419 PFID(lu_object_fid(&dto->do_lu)), i);
2421 rc = lod_sub_delete(env, next,
2422 (const struct dt_key *)stripe_name, th);
2426 rc = lod_sub_ref_del(env, next, th);
2431 lod_striping_free(env, lod_dt_obj(dt));
2437 * Implementation of dt_object_operations::do_declare_xattr_set.
2439 * Used with regular (non-striped) objects. Basically it
2440 * initializes the striping information and applies the
2441 * change to all the stripes.
2443 * \see dt_object_operations::do_declare_xattr_set() in the API description
2446 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2447 struct dt_object *dt,
2448 const struct lu_buf *buf,
2449 const char *name, int fl,
2452 struct dt_object *next = dt_object_child(dt);
2453 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2454 struct lod_object *lo = lod_dt_obj(dt);
2459 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2460 struct lmv_user_md_v1 *lum;
2462 LASSERT(buf != NULL && buf->lb_buf != NULL);
2464 rc = lod_verify_md_striping(d, lum);
2467 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2468 rc = lod_verify_striping(d, lo, buf, false);
2473 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2477 /* Note: Do not set LinkEA on sub-stripes, otherwise
2478 * it will confuse the fid2path process(see mdt_path_current()).
2479 * The linkEA between master and sub-stripes is set in
2480 * lod_xattr_set_lmv(). */
2481 if (strcmp(name, XATTR_NAME_LINK) == 0)
2484 /* set xattr to each stripes, if needed */
2485 rc = lod_striping_load(env, lo);
2489 if (lo->ldo_dir_stripe_count == 0)
2492 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2493 LASSERT(lo->ldo_stripe[i]);
2495 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2505 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2506 struct lod_object *lo,
2507 struct dt_object *dt, struct thandle *th,
2508 int comp_idx, int stripe_idx,
2509 struct lod_obj_stripe_cb_data *data)
2511 struct lod_thread_info *info = lod_env_info(env);
2512 struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
2513 struct filter_fid *ff = &info->lti_ff;
2514 struct lu_buf *buf = &info->lti_buf;
2518 buf->lb_len = sizeof(*ff);
2519 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2527 * locd_buf is set if it's called by dir migration, which doesn't check
2530 if (data->locd_buf) {
2531 memset(ff, 0, sizeof(*ff));
2532 ff->ff_parent = *(struct lu_fid *)data->locd_buf->lb_buf;
2534 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
2536 if (lu_fid_eq(lod_object_fid(lo), &ff->ff_parent) &&
2537 ff->ff_layout.ol_comp_id == comp->llc_id)
2540 memset(ff, 0, sizeof(*ff));
2541 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2544 /* rewrite filter_fid */
2545 ff->ff_parent.f_ver = stripe_idx;
2546 ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
2547 ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
2548 ff->ff_layout.ol_comp_id = comp->llc_id;
2549 ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
2550 ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
2551 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
2553 if (data->locd_declare)
2554 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2555 LU_XATTR_REPLACE, th);
2557 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2558 LU_XATTR_REPLACE, th);
2564 * Reset parent FID on OST object
2566 * Replace parent FID with @dt object FID, which is only called during migration
2567 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2568 * the FID is changed.
2570 * \param[in] env execution environment
2571 * \param[in] dt dt_object whose stripes's parent FID will be reset
2572 * \parem[in] th thandle
2573 * \param[in] declare if it is declare
2575 * \retval 0 if reset succeeds
2576 * \retval negative errno if reset fails
2578 static int lod_replace_parent_fid(const struct lu_env *env,
2579 struct dt_object *dt,
2580 const struct lu_buf *buf,
2581 struct thandle *th, bool declare)
2583 struct lod_object *lo = lod_dt_obj(dt);
2584 struct lod_thread_info *info = lod_env_info(env);
2585 struct filter_fid *ff;
2586 struct lod_obj_stripe_cb_data data = { { 0 } };
2590 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2592 /* set xattr to each stripes, if needed */
2593 rc = lod_striping_load(env, lo);
2597 if (!lod_obj_is_striped(dt))
2600 if (info->lti_ea_store_size < sizeof(*ff)) {
2601 rc = lod_ea_store_resize(info, sizeof(*ff));
2606 data.locd_declare = declare;
2607 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2608 data.locd_buf = buf;
2609 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2614 inline __u16 lod_comp_entry_stripe_count(struct lod_object *lo,
2615 struct lod_layout_component *entry,
2618 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2622 else if (lod_comp_inited(entry))
2623 return entry->llc_stripe_count;
2624 else if ((__u16)-1 == entry->llc_stripe_count)
2625 return lod->lod_desc.ld_tgt_count;
2627 return lod_get_stripe_count(lod, lo, entry->llc_stripe_count);
2630 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2632 int magic, size = 0, i;
2633 struct lod_layout_component *comp_entries;
2635 bool is_composite, is_foreign = false;
2638 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2639 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2641 lo->ldo_def_striping->lds_def_striping_is_composite;
2643 comp_cnt = lo->ldo_comp_cnt;
2644 comp_entries = lo->ldo_comp_entries;
2645 is_composite = lo->ldo_is_composite;
2646 is_foreign = lo->ldo_is_foreign;
2650 return lo->ldo_foreign_lov_size;
2652 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2654 size = sizeof(struct lov_comp_md_v1) +
2655 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2656 LASSERT(size % sizeof(__u64) == 0);
2659 for (i = 0; i < comp_cnt; i++) {
2662 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2663 stripe_count = lod_comp_entry_stripe_count(lo, &comp_entries[i],
2665 if (!is_dir && is_composite)
2666 lod_comp_shrink_stripe_count(&comp_entries[i],
2669 size += lov_user_md_size(stripe_count, magic);
2670 LASSERT(size % sizeof(__u64) == 0);
2676 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2677 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2680 * \param[in] env execution environment
2681 * \param[in] dt dt_object to add components on
2682 * \param[in] buf buffer contains components to be added
2683 * \parem[in] th thandle
2685 * \retval 0 on success
2686 * \retval negative errno on failure
2688 static int lod_declare_layout_add(const struct lu_env *env,
2689 struct dt_object *dt,
2690 const struct lu_buf *buf,
2693 struct lod_thread_info *info = lod_env_info(env);
2694 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2695 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2696 struct dt_object *next = dt_object_child(dt);
2697 struct lov_desc *desc = &d->lod_desc;
2698 struct lod_object *lo = lod_dt_obj(dt);
2699 struct lov_user_md_v3 *v3;
2700 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2702 int i, rc, array_cnt, old_array_cnt;
2705 LASSERT(lo->ldo_is_composite);
2707 if (lo->ldo_flr_state != LCM_FL_NONE)
2710 rc = lod_verify_striping(d, lo, buf, false);
2714 magic = comp_v1->lcm_magic;
2715 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2716 lustre_swab_lov_comp_md_v1(comp_v1);
2717 magic = comp_v1->lcm_magic;
2720 if (magic != LOV_USER_MAGIC_COMP_V1)
2723 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2724 OBD_ALLOC(comp_array, sizeof(*comp_array) * array_cnt);
2725 if (comp_array == NULL)
2728 memcpy(comp_array, lo->ldo_comp_entries,
2729 sizeof(*comp_array) * lo->ldo_comp_cnt);
2731 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2732 struct lov_user_md_v1 *v1;
2733 struct lu_extent *ext;
2735 v1 = (struct lov_user_md *)((char *)comp_v1 +
2736 comp_v1->lcm_entries[i].lcme_offset);
2737 ext = &comp_v1->lcm_entries[i].lcme_extent;
2739 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2740 lod_comp->llc_extent.e_start = ext->e_start;
2741 lod_comp->llc_extent.e_end = ext->e_end;
2742 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2743 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2745 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2746 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2747 lod_adjust_stripe_info(lod_comp, desc);
2749 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2750 v3 = (struct lov_user_md_v3 *) v1;
2751 if (v3->lmm_pool_name[0] != '\0') {
2752 rc = lod_set_pool(&lod_comp->llc_pool,
2760 old_array = lo->ldo_comp_entries;
2761 old_array_cnt = lo->ldo_comp_cnt;
2763 lo->ldo_comp_entries = comp_array;
2764 lo->ldo_comp_cnt = array_cnt;
2766 /* No need to increase layout generation here, it will be increased
2767 * later when generating component ID for the new components */
2769 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2770 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2771 XATTR_NAME_LOV, 0, th);
2773 lo->ldo_comp_entries = old_array;
2774 lo->ldo_comp_cnt = old_array_cnt;
2778 OBD_FREE(old_array, sizeof(*lod_comp) * old_array_cnt);
2780 LASSERT(lo->ldo_mirror_count == 1);
2781 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2786 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2787 lod_comp = &comp_array[i];
2788 if (lod_comp->llc_pool != NULL) {
2789 OBD_FREE(lod_comp->llc_pool,
2790 strlen(lod_comp->llc_pool) + 1);
2791 lod_comp->llc_pool = NULL;
2794 OBD_FREE(comp_array, sizeof(*comp_array) * array_cnt);
2799 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2800 * the '$field' can only be 'flags' now. The xattr value is binary
2801 * lov_comp_md_v1 which contains the component ID(s) and the value of
2802 * the field to be modified.
2804 * \param[in] env execution environment
2805 * \param[in] dt dt_object to be modified
2806 * \param[in] op operation string, like "set.flags"
2807 * \param[in] buf buffer contains components to be set
2808 * \parem[in] th thandle
2810 * \retval 0 on success
2811 * \retval negative errno on failure
2813 static int lod_declare_layout_set(const struct lu_env *env,
2814 struct dt_object *dt,
2815 char *op, const struct lu_buf *buf,
2818 struct lod_layout_component *lod_comp;
2819 struct lod_thread_info *info = lod_env_info(env);
2820 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2821 struct lod_object *lo = lod_dt_obj(dt);
2822 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2825 bool changed = false;
2828 if (strcmp(op, "set.flags") != 0) {
2829 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2830 lod2obd(d)->obd_name, op);
2834 magic = comp_v1->lcm_magic;
2835 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2836 lustre_swab_lov_comp_md_v1(comp_v1);
2837 magic = comp_v1->lcm_magic;
2840 if (magic != LOV_USER_MAGIC_COMP_V1)
2843 if (comp_v1->lcm_entry_count == 0) {
2844 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2845 lod2obd(d)->obd_name);
2849 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2850 __u32 id = comp_v1->lcm_entries[i].lcme_id;
2851 __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
2852 __u32 mirror_flag = flags & LCME_MIRROR_FLAGS;
2853 bool neg = flags & LCME_FL_NEG;
2855 if (flags & LCME_FL_INIT) {
2857 lod_striping_free(env, lo);
2861 flags &= ~(LCME_MIRROR_FLAGS | LCME_FL_NEG);
2862 for (j = 0; j < lo->ldo_comp_cnt; j++) {
2863 lod_comp = &lo->ldo_comp_entries[j];
2865 /* lfs only put one flag in each entry */
2866 if ((flags && id != lod_comp->llc_id) ||
2867 (mirror_flag && mirror_id_of(id) !=
2868 mirror_id_of(lod_comp->llc_id)))
2873 lod_comp->llc_flags &= ~flags;
2875 lod_comp->llc_flags &= ~mirror_flag;
2878 lod_comp->llc_flags |= flags;
2880 lod_comp->llc_flags |= mirror_flag;
2881 if (mirror_flag & LCME_FL_NOSYNC)
2882 lod_comp->llc_timestamp =
2883 ktime_get_real_seconds();
2891 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
2892 lod2obd(d)->obd_name);
2896 lod_obj_inc_layout_gen(lo);
2898 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2899 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
2900 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
2905 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
2906 * and the xattr value is a unique component ID or a special lcme_id.
2908 * \param[in] env execution environment
2909 * \param[in] dt dt_object to be operated on
2910 * \param[in] buf buffer contains component ID or lcme_id
2911 * \parem[in] th thandle
2913 * \retval 0 on success
2914 * \retval negative errno on failure
2916 static int lod_declare_layout_del(const struct lu_env *env,
2917 struct dt_object *dt,
2918 const struct lu_buf *buf,
2921 struct lod_thread_info *info = lod_env_info(env);
2922 struct dt_object *next = dt_object_child(dt);
2923 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2924 struct lod_object *lo = lod_dt_obj(dt);
2925 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2926 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2927 __u32 magic, id, flags, neg_flags = 0;
2931 LASSERT(lo->ldo_is_composite);
2933 if (lo->ldo_flr_state != LCM_FL_NONE)
2936 magic = comp_v1->lcm_magic;
2937 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2938 lustre_swab_lov_comp_md_v1(comp_v1);
2939 magic = comp_v1->lcm_magic;
2942 if (magic != LOV_USER_MAGIC_COMP_V1)
2945 id = comp_v1->lcm_entries[0].lcme_id;
2946 flags = comp_v1->lcm_entries[0].lcme_flags;
2948 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
2949 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
2950 lod2obd(d)->obd_name, id, flags);
2954 if (id != LCME_ID_INVAL && flags != 0) {
2955 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
2956 lod2obd(d)->obd_name);
2960 if (id == LCME_ID_INVAL && !flags) {
2961 CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
2962 lod2obd(d)->obd_name);
2966 if (flags & LCME_FL_NEG) {
2967 neg_flags = flags & ~LCME_FL_NEG;
2971 left = lo->ldo_comp_cnt;
2975 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
2976 struct lod_layout_component *lod_comp;
2978 lod_comp = &lo->ldo_comp_entries[i];
2980 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
2982 else if (flags && !(flags & lod_comp->llc_flags))
2984 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
2987 if (left != (i + 1)) {
2988 CDEBUG(D_LAYOUT, "%s: this deletion will create "
2989 "a hole.\n", lod2obd(d)->obd_name);
2994 /* Mark the component as deleted */
2995 lod_comp->llc_id = LCME_ID_INVAL;
2997 /* Not instantiated component */
2998 if (lod_comp->llc_stripe == NULL)
3001 LASSERT(lod_comp->llc_stripe_count > 0);
3002 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
3003 struct dt_object *obj = lod_comp->llc_stripe[j];
3007 rc = lod_sub_declare_destroy(env, obj, th);
3013 LASSERTF(left >= 0, "left = %d\n", left);
3014 if (left == lo->ldo_comp_cnt) {
3015 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
3016 lod2obd(d)->obd_name, id);
3020 memset(attr, 0, sizeof(*attr));
3021 attr->la_valid = LA_SIZE;
3022 rc = lod_sub_declare_attr_set(env, next, attr, th);
3027 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3028 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
3029 XATTR_NAME_LOV, 0, th);
3031 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
3038 * Declare layout add/set/del operations issued by special xattr names:
3040 * XATTR_LUSTRE_LOV.add add component(s) to existing file
3041 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
3042 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
3044 * \param[in] env execution environment
3045 * \param[in] dt object
3046 * \param[in] name name of xattr
3047 * \param[in] buf lu_buf contains xattr value
3048 * \param[in] th transaction handle
3050 * \retval 0 on success
3051 * \retval negative if failed
3053 static int lod_declare_modify_layout(const struct lu_env *env,
3054 struct dt_object *dt,
3056 const struct lu_buf *buf,
3059 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3060 struct lod_object *lo = lod_dt_obj(dt);
3062 int rc, len = strlen(XATTR_LUSTRE_LOV);
3065 LASSERT(dt_object_exists(dt));
3067 if (strlen(name) <= len || name[len] != '.') {
3068 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
3069 lod2obd(d)->obd_name, name);
3074 rc = lod_striping_load(env, lo);
3078 /* the layout to be modified must be a composite layout */
3079 if (!lo->ldo_is_composite) {
3080 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
3081 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3082 GOTO(unlock, rc = -EINVAL);
3085 op = (char *)name + len;
3086 if (strcmp(op, "add") == 0) {
3087 rc = lod_declare_layout_add(env, dt, buf, th);
3088 } else if (strcmp(op, "del") == 0) {
3089 rc = lod_declare_layout_del(env, dt, buf, th);
3090 } else if (strncmp(op, "set", strlen("set")) == 0) {
3091 rc = lod_declare_layout_set(env, dt, op, buf, th);
3093 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
3094 lod2obd(d)->obd_name, name);
3095 GOTO(unlock, rc = -ENOTSUPP);
3099 lod_striping_free(env, lo);
3105 * Convert a plain file lov_mds_md to a composite layout.
3107 * \param[in,out] info the thread info::lti_ea_store buffer contains little
3108 * endian plain file layout
3110 * \retval 0 on success, <0 on failure
3112 static int lod_layout_convert(struct lod_thread_info *info)
3114 struct lov_mds_md *lmm = info->lti_ea_store;
3115 struct lov_mds_md *lmm_save;
3116 struct lov_comp_md_v1 *lcm;
3117 struct lov_comp_md_entry_v1 *lcme;
3123 /* realloc buffer to a composite layout which contains one component */
3124 blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3125 le32_to_cpu(lmm->lmm_magic));
3126 size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
3128 OBD_ALLOC_LARGE(lmm_save, blob_size);
3130 GOTO(out, rc = -ENOMEM);
3132 memcpy(lmm_save, lmm, blob_size);
3134 if (info->lti_ea_store_size < size) {
3135 rc = lod_ea_store_resize(info, size);
3140 lcm = info->lti_ea_store;
3141 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
3142 lcm->lcm_size = cpu_to_le32(size);
3143 lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
3144 lmm_save->lmm_layout_gen));
3145 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
3146 lcm->lcm_entry_count = cpu_to_le16(1);
3147 lcm->lcm_mirror_count = 0;
3149 lcme = &lcm->lcm_entries[0];
3150 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
3151 lcme->lcme_extent.e_start = 0;
3152 lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
3153 lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
3154 lcme->lcme_size = cpu_to_le32(blob_size);
3156 memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
3161 OBD_FREE_LARGE(lmm_save, blob_size);
3166 * Merge layouts to form a mirrored file.
3168 static int lod_declare_layout_merge(const struct lu_env *env,
3169 struct dt_object *dt, const struct lu_buf *mbuf,
3172 struct lod_thread_info *info = lod_env_info(env);
3173 struct lu_buf *buf = &info->lti_buf;
3174 struct lod_object *lo = lod_dt_obj(dt);
3175 struct lov_comp_md_v1 *lcm;
3176 struct lov_comp_md_v1 *cur_lcm;
3177 struct lov_comp_md_v1 *merge_lcm;
3178 struct lov_comp_md_entry_v1 *lcme;
3181 __u16 cur_entry_count;
3182 __u16 merge_entry_count;
3184 __u16 mirror_id = 0;
3189 merge_lcm = mbuf->lb_buf;
3190 if (mbuf->lb_len < sizeof(*merge_lcm))
3193 /* must be an existing layout from disk */
3194 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
3197 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
3199 /* do not allow to merge two mirrored files */
3200 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
3203 /* verify the target buffer */
3204 rc = lod_get_lov_ea(env, lo);
3206 RETURN(rc ? : -ENODATA);
3208 cur_lcm = info->lti_ea_store;
3209 switch (le32_to_cpu(cur_lcm->lcm_magic)) {
3212 rc = lod_layout_convert(info);
3214 case LOV_MAGIC_COMP_V1:
3223 /* info->lti_ea_store could be reallocated in lod_layout_convert() */
3224 cur_lcm = info->lti_ea_store;
3225 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
3227 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
3228 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
3229 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
3232 /* size of new layout */
3233 size = le32_to_cpu(cur_lcm->lcm_size) +
3234 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
3236 memset(buf, 0, sizeof(*buf));
3237 lu_buf_alloc(buf, size);
3238 if (buf->lb_buf == NULL)
3242 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
3244 offset = sizeof(*lcm) +
3245 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
3246 for (i = 0; i < cur_entry_count; i++) {
3247 struct lov_comp_md_entry_v1 *cur_lcme;
3249 lcme = &lcm->lcm_entries[i];
3250 cur_lcme = &cur_lcm->lcm_entries[i];
3252 lcme->lcme_offset = cpu_to_le32(offset);
3253 memcpy((char *)lcm + offset,
3254 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
3255 le32_to_cpu(lcme->lcme_size));
3257 offset += le32_to_cpu(lcme->lcme_size);
3259 if (mirror_count == 1 &&
3260 mirror_id_of(le32_to_cpu(lcme->lcme_id)) == 0) {
3261 /* Add mirror from a non-flr file, create new mirror ID.
3262 * Otherwise, keep existing mirror's component ID, used
3263 * for mirror extension.
3265 id = pflr_id(1, i + 1);
3266 lcme->lcme_id = cpu_to_le32(id);
3269 id = MAX(le32_to_cpu(lcme->lcme_id), id);
3272 mirror_id = mirror_id_of(id) + 1;
3273 for (i = 0; i < merge_entry_count; i++) {
3274 struct lov_comp_md_entry_v1 *merge_lcme;
3276 merge_lcme = &merge_lcm->lcm_entries[i];
3277 lcme = &lcm->lcm_entries[cur_entry_count + i];
3279 *lcme = *merge_lcme;
3280 lcme->lcme_offset = cpu_to_le32(offset);
3282 id = pflr_id(mirror_id, i + 1);
3283 lcme->lcme_id = cpu_to_le32(id);
3285 memcpy((char *)lcm + offset,
3286 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
3287 le32_to_cpu(lcme->lcme_size));
3289 offset += le32_to_cpu(lcme->lcme_size);
3292 /* fixup layout information */
3293 lod_obj_inc_layout_gen(lo);
3294 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3295 lcm->lcm_size = cpu_to_le32(size);
3296 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
3297 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
3298 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NONE)
3299 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
3301 rc = lod_striping_reload(env, lo, buf);
3305 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
3306 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3314 * Split layouts, just set the LOVEA with the layout from mbuf.
3316 static int lod_declare_layout_split(const struct lu_env *env,
3317 struct dt_object *dt, const struct lu_buf *mbuf,
3320 struct lod_object *lo = lod_dt_obj(dt);
3321 struct lov_comp_md_v1 *lcm = mbuf->lb_buf;
3325 lod_obj_inc_layout_gen(lo);
3326 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3328 rc = lod_striping_reload(env, lo, mbuf);
3332 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), mbuf,
3333 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3338 * Implementation of dt_object_operations::do_declare_xattr_set.
3340 * \see dt_object_operations::do_declare_xattr_set() in the API description
3343 * the extension to the API:
3344 * - declaring LOVEA requests striping creation
3345 * - LU_XATTR_REPLACE means layout swap
3347 static int lod_declare_xattr_set(const struct lu_env *env,
3348 struct dt_object *dt,
3349 const struct lu_buf *buf,
3350 const char *name, int fl,
3353 struct dt_object *next = dt_object_child(dt);
3354 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3359 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
3360 if ((S_ISREG(mode) || mode == 0) &&
3361 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE | LU_XATTR_SPLIT)) &&
3362 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3363 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
3365 * this is a request to create object's striping.
3367 * allow to declare predefined striping on a new (!mode) object
3368 * which is supposed to be replay of regular file creation
3369 * (when LOV setting is declared)
3371 * LU_XATTR_REPLACE is set to indicate a layout swap
3373 if (dt_object_exists(dt)) {
3374 rc = dt_attr_get(env, next, attr);
3378 memset(attr, 0, sizeof(*attr));
3379 attr->la_valid = LA_TYPE | LA_MODE;
3380 attr->la_mode = S_IFREG;
3382 rc = lod_declare_striped_create(env, dt, attr, buf, th);
3383 } else if (fl & LU_XATTR_MERGE) {
3384 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3385 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3386 rc = lod_declare_layout_merge(env, dt, buf, th);
3387 } else if (fl & LU_XATTR_SPLIT) {
3388 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3389 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3390 rc = lod_declare_layout_split(env, dt, buf, th);
3391 } else if (S_ISREG(mode) &&
3392 strlen(name) > strlen(XATTR_LUSTRE_LOV) + 1 &&
3393 strncmp(name, XATTR_LUSTRE_LOV,
3394 strlen(XATTR_LUSTRE_LOV)) == 0) {
3396 * this is a request to modify object's striping.
3397 * add/set/del component(s).
3399 if (!dt_object_exists(dt))
3402 rc = lod_declare_modify_layout(env, dt, name, buf, th);
3403 } else if (strncmp(name, XATTR_NAME_LMV, strlen(XATTR_NAME_LMV)) == 0 &&
3404 strlen(name) > strlen(XATTR_NAME_LMV) + 1) {
3405 const char *op = name + strlen(XATTR_NAME_LMV) + 1;
3408 if (strcmp(op, "add") == 0)
3409 rc = lod_dir_declare_layout_add(env, dt, buf, th);
3410 else if (strcmp(op, "del") == 0)
3411 rc = lod_dir_declare_layout_delete(env, dt, buf, th);
3412 else if (strcmp(op, "set") == 0)
3413 rc = lod_sub_declare_xattr_set(env, next, buf,
3414 XATTR_NAME_LMV, fl, th);
3417 } else if (S_ISDIR(mode)) {
3418 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
3419 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3420 rc = lod_replace_parent_fid(env, dt, buf, th, true);
3422 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
3429 * Apply xattr changes to the object.
3431 * Applies xattr changes to the object and the stripes if the latter exist.
3433 * \param[in] env execution environment
3434 * \param[in] dt object
3435 * \param[in] buf buffer pointing to the new value of xattr
3436 * \param[in] name name of xattr
3437 * \param[in] fl flags
3438 * \param[in] th transaction handle
3440 * \retval 0 on success
3441 * \retval negative if failed
3443 static int lod_xattr_set_internal(const struct lu_env *env,
3444 struct dt_object *dt,
3445 const struct lu_buf *buf,
3446 const char *name, int fl,
3449 struct dt_object *next = dt_object_child(dt);
3450 struct lod_object *lo = lod_dt_obj(dt);
3455 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3456 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3459 /* Note: Do not set LinkEA on sub-stripes, otherwise
3460 * it will confuse the fid2path process(see mdt_path_current()).
3461 * The linkEA between master and sub-stripes is set in
3462 * lod_xattr_set_lmv(). */
3463 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
3466 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3467 LASSERT(lo->ldo_stripe[i]);
3469 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
3479 * Delete an extended attribute.
3481 * Deletes specified xattr from the object and the stripes if the latter exist.
3483 * \param[in] env execution environment
3484 * \param[in] dt object
3485 * \param[in] name name of xattr
3486 * \param[in] th transaction handle
3488 * \retval 0 on success
3489 * \retval negative if failed
3491 static int lod_xattr_del_internal(const struct lu_env *env,
3492 struct dt_object *dt,
3493 const char *name, struct thandle *th)
3495 struct dt_object *next = dt_object_child(dt);
3496 struct lod_object *lo = lod_dt_obj(dt);
3501 rc = lod_sub_xattr_del(env, next, name, th);
3502 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3505 if (lo->ldo_dir_stripe_count == 0)
3508 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3509 LASSERT(lo->ldo_stripe[i]);
3511 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3520 * Set default striping on a directory.
3522 * Sets specified striping on a directory object unless it matches the default
3523 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3524 * EA. This striping will be used when regular file is being created in this
3527 * \param[in] env execution environment
3528 * \param[in] dt the striped object
3529 * \param[in] buf buffer with the striping
3530 * \param[in] name name of EA
3531 * \param[in] fl xattr flag (see OSD API description)
3532 * \param[in] th transaction handle
3534 * \retval 0 on success
3535 * \retval negative if failed
3537 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
3538 struct dt_object *dt,
3539 const struct lu_buf *buf,
3540 const char *name, int fl,
3543 struct lov_user_md_v1 *lum;
3544 struct lov_user_md_v3 *v3 = NULL;
3545 const char *pool_name = NULL;
3550 LASSERT(buf != NULL && buf->lb_buf != NULL);
3553 switch (lum->lmm_magic) {
3554 case LOV_USER_MAGIC_SPECIFIC:
3555 case LOV_USER_MAGIC_V3:
3557 if (v3->lmm_pool_name[0] != '\0')
3558 pool_name = v3->lmm_pool_name;
3560 case LOV_USER_MAGIC_V1:
3561 /* if { size, offset, count } = { 0, -1, 0 } and no pool
3562 * (i.e. all default values specified) then delete default
3563 * striping from dir. */
3565 "set default striping: sz %u # %u offset %d %s %s\n",
3566 (unsigned)lum->lmm_stripe_size,
3567 (unsigned)lum->lmm_stripe_count,
3568 (int)lum->lmm_stripe_offset,
3569 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
3571 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
3572 lum->lmm_stripe_count,
3573 lum->lmm_stripe_offset,
3576 case LOV_USER_MAGIC_COMP_V1:
3580 CERROR("Invalid magic %x\n", lum->lmm_magic);
3585 rc = lod_xattr_del_internal(env, dt, name, th);
3589 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3596 * Set default striping on a directory object.
3598 * Sets specified striping on a directory object unless it matches the default
3599 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3600 * EA. This striping will be used when a new directory is being created in the
3603 * \param[in] env execution environment
3604 * \param[in] dt the striped object
3605 * \param[in] buf buffer with the striping
3606 * \param[in] name name of EA
3607 * \param[in] fl xattr flag (see OSD API description)
3608 * \param[in] th transaction handle
3610 * \retval 0 on success
3611 * \retval negative if failed
3613 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
3614 struct dt_object *dt,
3615 const struct lu_buf *buf,
3616 const char *name, int fl,
3619 struct lmv_user_md_v1 *lum;
3623 LASSERT(buf != NULL && buf->lb_buf != NULL);
3626 CDEBUG(D_OTHER, "set default stripe_count # %u stripe_offset %d\n",
3627 le32_to_cpu(lum->lum_stripe_count),
3628 (int)le32_to_cpu(lum->lum_stripe_offset));
3630 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
3631 le32_to_cpu(lum->lum_stripe_offset)) &&
3632 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
3633 rc = lod_xattr_del_internal(env, dt, name, th);
3637 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3646 * Turn directory into a striped directory.
3648 * During replay the client sends the striping created before MDT
3649 * failure, then the layer above LOD sends this defined striping
3650 * using ->do_xattr_set(), so LOD uses this method to replay creation
3651 * of the stripes. Notice the original information for the striping
3652 * (#stripes, FIDs, etc) was transferred in declare path.
3654 * \param[in] env execution environment
3655 * \param[in] dt the striped object
3656 * \param[in] buf not used currently
3657 * \param[in] name not used currently
3658 * \param[in] fl xattr flag (see OSD API description)
3659 * \param[in] th transaction handle
3661 * \retval 0 on success
3662 * \retval negative if failed
3664 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
3665 const struct lu_buf *buf, const char *name,
3666 int fl, struct thandle *th)
3668 struct lod_object *lo = lod_dt_obj(dt);
3669 struct lod_thread_info *info = lod_env_info(env);
3670 struct lu_attr *attr = &info->lti_attr;
3671 struct dt_object_format *dof = &info->lti_format;
3672 struct lu_buf lmv_buf;
3673 struct lu_buf slave_lmv_buf;
3674 struct lmv_mds_md_v1 *lmm;
3675 struct lmv_mds_md_v1 *slave_lmm = NULL;
3676 struct dt_insert_rec *rec = &info->lti_dt_rec;
3681 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3684 /* The stripes are supposed to be allocated in declare phase,
3685 * if there are no stripes being allocated, it will skip */
3686 if (lo->ldo_dir_stripe_count == 0) {
3687 if (lo->ldo_dir_is_foreign) {
3688 rc = lod_sub_xattr_set(env, dt_object_child(dt), buf,
3689 XATTR_NAME_LMV, fl, th);
3696 rc = dt_attr_get(env, dt_object_child(dt), attr);
3700 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME |
3701 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
3702 dof->dof_type = DFT_DIR;
3704 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
3707 lmm = lmv_buf.lb_buf;
3709 OBD_ALLOC_PTR(slave_lmm);
3710 if (slave_lmm == NULL)
3713 lod_prep_slave_lmv_md(slave_lmm, lmm);
3714 slave_lmv_buf.lb_buf = slave_lmm;
3715 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
3717 rec->rec_type = S_IFDIR;
3718 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3719 struct dt_object *dto = lo->ldo_stripe[i];
3720 char *stripe_name = info->lti_key;
3721 struct lu_name *sname;
3722 struct linkea_data ldata = { NULL };
3723 struct lu_buf linkea_buf;
3725 /* if it's source stripe of migrating directory, don't create */
3726 if (!((lo->ldo_dir_hash_type & LMV_HASH_FLAG_MIGRATION) &&
3727 i >= lo->ldo_dir_migrate_offset)) {
3728 dt_write_lock(env, dto, MOR_TGT_CHILD);
3729 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
3731 dt_write_unlock(env, dto);
3735 rc = lod_sub_ref_add(env, dto, th);
3736 dt_write_unlock(env, dto);
3740 rec->rec_fid = lu_object_fid(&dto->do_lu);
3741 rc = lod_sub_insert(env, dto,
3742 (const struct dt_rec *)rec,
3743 (const struct dt_key *)dot, th);
3748 rec->rec_fid = lu_object_fid(&dt->do_lu);
3749 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
3750 (const struct dt_key *)dotdot, th);
3754 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
3755 cfs_fail_val != i) {
3756 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
3758 slave_lmm->lmv_master_mdt_index =
3761 slave_lmm->lmv_master_mdt_index =
3764 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
3765 XATTR_NAME_LMV, fl, th);
3770 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
3772 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3773 PFID(lu_object_fid(&dto->do_lu)), i + 1);
3775 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3776 PFID(lu_object_fid(&dto->do_lu)), i);
3778 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
3779 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
3780 sname, lu_object_fid(&dt->do_lu));
3784 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
3785 linkea_buf.lb_len = ldata.ld_leh->leh_len;
3786 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
3787 XATTR_NAME_LINK, 0, th);
3791 rec->rec_fid = lu_object_fid(&dto->do_lu);
3792 rc = lod_sub_insert(env, dt_object_child(dt),
3793 (const struct dt_rec *)rec,
3794 (const struct dt_key *)stripe_name, th);
3798 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
3803 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
3804 rc = lod_sub_xattr_set(env, dt_object_child(dt),
3805 &lmv_buf, XATTR_NAME_LMV, fl, th);
3807 if (slave_lmm != NULL)
3808 OBD_FREE_PTR(slave_lmm);
3814 * Helper function to declare/execute creation of a striped directory
3816 * Called in declare/create object path, prepare striping for a directory
3817 * and prepare defaults data striping for the objects to be created in
3818 * that directory. Notice the function calls "declaration" or "execution"
3819 * methods depending on \a declare param. This is a consequence of the
3820 * current approach while we don't have natural distributed transactions:
3821 * we basically execute non-local updates in the declare phase. So, the
3822 * arguments for the both phases are the same and this is the reason for
3823 * this function to exist.
3825 * \param[in] env execution environment
3826 * \param[in] dt object
3827 * \param[in] attr attributes the stripes will be created with
3828 * \param[in] lmu lmv_user_md if MDT indices are specified
3829 * \param[in] dof format of stripes (see OSD API description)
3830 * \param[in] th transaction handle
3831 * \param[in] declare where to call "declare" or "execute" methods
3833 * \retval 0 on success
3834 * \retval negative if failed
3836 static int lod_dir_striping_create_internal(const struct lu_env *env,
3837 struct dt_object *dt,
3838 struct lu_attr *attr,
3839 const struct lu_buf *lmu,
3840 struct dt_object_format *dof,
3844 struct lod_thread_info *info = lod_env_info(env);
3845 struct lod_object *lo = lod_dt_obj(dt);
3846 const struct lod_default_striping *lds = lo->ldo_def_striping;
3850 LASSERT(ergo(lds != NULL,
3851 lds->lds_def_striping_set ||
3852 lds->lds_dir_def_striping_set));
3854 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
3855 lo->ldo_dir_stripe_offset)) {
3857 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3858 int stripe_count = lo->ldo_dir_stripe_count;
3860 if (info->lti_ea_store_size < sizeof(*v1)) {
3861 rc = lod_ea_store_resize(info, sizeof(*v1));
3864 v1 = info->lti_ea_store;
3867 memset(v1, 0, sizeof(*v1));
3868 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3869 v1->lum_stripe_count = cpu_to_le32(stripe_count);
3870 v1->lum_stripe_offset =
3871 cpu_to_le32(lo->ldo_dir_stripe_offset);
3873 info->lti_buf.lb_buf = v1;
3874 info->lti_buf.lb_len = sizeof(*v1);
3875 lmu = &info->lti_buf;
3879 rc = lod_declare_xattr_set_lmv(env, dt, attr, lmu, dof,
3882 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
3887 /* foreign LMV EA case */
3889 struct lmv_foreign_md *lfm = lmu->lb_buf;
3891 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
3892 rc = lod_declare_xattr_set_lmv(env, dt, attr,
3896 if (lo->ldo_dir_is_foreign) {
3897 LASSERT(lo->ldo_foreign_lmv != NULL &&
3898 lo->ldo_foreign_lmv_size > 0);
3899 info->lti_buf.lb_buf = lo->ldo_foreign_lmv;
3900 info->lti_buf.lb_len = lo->ldo_foreign_lmv_size;
3901 lmu = &info->lti_buf;
3902 rc = lod_xattr_set_lmv(env, dt, lmu,
3903 XATTR_NAME_LMV, 0, th);
3908 /* Transfer default LMV striping from the parent */
3909 if (lds != NULL && lds->lds_dir_def_striping_set &&
3910 !LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
3911 lds->lds_dir_def_stripe_offset)) {
3912 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3914 if (info->lti_ea_store_size < sizeof(*v1)) {
3915 rc = lod_ea_store_resize(info, sizeof(*v1));
3918 v1 = info->lti_ea_store;
3921 memset(v1, 0, sizeof(*v1));
3922 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3923 v1->lum_stripe_count =
3924 cpu_to_le32(lds->lds_dir_def_stripe_count);
3925 v1->lum_stripe_offset =
3926 cpu_to_le32(lds->lds_dir_def_stripe_offset);
3928 cpu_to_le32(lds->lds_dir_def_hash_type);
3930 info->lti_buf.lb_buf = v1;
3931 info->lti_buf.lb_len = sizeof(*v1);
3933 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3934 XATTR_NAME_DEFAULT_LMV,
3937 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
3939 XATTR_NAME_DEFAULT_LMV, 0,
3945 /* Transfer default LOV striping from the parent */
3946 if (lds != NULL && lds->lds_def_striping_set &&
3947 lds->lds_def_comp_cnt != 0) {
3948 struct lov_mds_md *lmm;
3949 int lmm_size = lod_comp_md_size(lo, true);
3951 if (info->lti_ea_store_size < lmm_size) {
3952 rc = lod_ea_store_resize(info, lmm_size);
3956 lmm = info->lti_ea_store;
3958 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
3962 info->lti_buf.lb_buf = lmm;
3963 info->lti_buf.lb_len = lmm_size;
3966 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3967 XATTR_NAME_LOV, 0, th);
3969 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
3970 XATTR_NAME_LOV, 0, th);
3978 static int lod_declare_dir_striping_create(const struct lu_env *env,
3979 struct dt_object *dt,
3980 struct lu_attr *attr,
3982 struct dt_object_format *dof,
3985 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
3989 static int lod_dir_striping_create(const struct lu_env *env,
3990 struct dt_object *dt,
3991 struct lu_attr *attr,
3992 struct dt_object_format *dof,
3995 return lod_dir_striping_create_internal(env, dt, attr, NULL, dof, th,
4000 * Make LOV EA for striped object.
4002 * Generate striping information and store it in the LOV EA of the given
4003 * object. The caller must ensure nobody else is calling the function
4004 * against the object concurrently. The transaction must be started.
4005 * FLDB service must be running as well; it's used to map FID to the target,
4006 * which is stored in LOV EA.
4008 * \param[in] env execution environment for this thread
4009 * \param[in] lo LOD object
4010 * \param[in] th transaction handle
4012 * \retval 0 if LOV EA is stored successfully
4013 * \retval negative error number on failure
4015 static int lod_generate_and_set_lovea(const struct lu_env *env,
4016 struct lod_object *lo,
4019 struct lod_thread_info *info = lod_env_info(env);
4020 struct dt_object *next = dt_object_child(&lo->ldo_obj);
4021 struct lov_mds_md_v1 *lmm;
4027 if (lo->ldo_comp_cnt == 0 && !lo->ldo_is_foreign) {
4028 lod_striping_free(env, lo);
4029 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
4033 lmm_size = lod_comp_md_size(lo, false);
4034 if (info->lti_ea_store_size < lmm_size) {
4035 rc = lod_ea_store_resize(info, lmm_size);
4039 lmm = info->lti_ea_store;
4041 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
4045 info->lti_buf.lb_buf = lmm;
4046 info->lti_buf.lb_len = lmm_size;
4047 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
4048 XATTR_NAME_LOV, 0, th);
4053 * Delete layout component(s)
4055 * \param[in] env execution environment for this thread
4056 * \param[in] dt object
4057 * \param[in] th transaction handle
4059 * \retval 0 on success
4060 * \retval negative error number on failure
4062 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
4065 struct lod_layout_component *lod_comp;
4066 struct lod_object *lo = lod_dt_obj(dt);
4067 struct dt_object *next = dt_object_child(dt);
4068 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4071 LASSERT(lo->ldo_is_composite);
4072 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
4074 left = lo->ldo_comp_cnt;
4075 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
4076 lod_comp = &lo->ldo_comp_entries[i];
4078 if (lod_comp->llc_id != LCME_ID_INVAL)
4082 /* Not instantiated component */
4083 if (lod_comp->llc_stripe == NULL)
4086 LASSERT(lod_comp->llc_stripe_count > 0);
4087 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4088 struct dt_object *obj = lod_comp->llc_stripe[j];
4092 rc = lod_sub_destroy(env, obj, th);
4096 lu_object_put(env, &obj->do_lu);
4097 lod_comp->llc_stripe[j] = NULL;
4099 OBD_FREE(lod_comp->llc_stripe, sizeof(struct dt_object *) *
4100 lod_comp->llc_stripes_allocated);
4101 lod_comp->llc_stripe = NULL;
4102 OBD_FREE(lod_comp->llc_ost_indices,
4103 sizeof(__u32) * lod_comp->llc_stripes_allocated);
4104 lod_comp->llc_ost_indices = NULL;
4105 lod_comp->llc_stripes_allocated = 0;
4106 lod_obj_set_pool(lo, i, NULL);
4107 if (lod_comp->llc_ostlist.op_array) {
4108 OBD_FREE(lod_comp->llc_ostlist.op_array,
4109 lod_comp->llc_ostlist.op_size);
4110 lod_comp->llc_ostlist.op_array = NULL;
4111 lod_comp->llc_ostlist.op_size = 0;
4115 LASSERTF(left >= 0 && left < lo->ldo_comp_cnt, "left = %d\n", left);
4117 struct lod_layout_component *comp_array;
4119 OBD_ALLOC(comp_array, sizeof(*comp_array) * left);
4120 if (comp_array == NULL)
4121 GOTO(out, rc = -ENOMEM);
4123 memcpy(&comp_array[0], &lo->ldo_comp_entries[0],
4124 sizeof(*comp_array) * left);
4126 OBD_FREE(lo->ldo_comp_entries,
4127 sizeof(*comp_array) * lo->ldo_comp_cnt);
4128 lo->ldo_comp_entries = comp_array;
4129 lo->ldo_comp_cnt = left;
4131 LASSERT(lo->ldo_mirror_count == 1);
4132 lo->ldo_mirrors[0].lme_end = left - 1;
4133 lod_obj_inc_layout_gen(lo);
4135 lod_free_comp_entries(lo);
4138 LASSERT(dt_object_exists(dt));
4139 rc = dt_attr_get(env, next, attr);
4143 if (attr->la_size > 0) {
4145 attr->la_valid = LA_SIZE;
4146 rc = lod_sub_attr_set(env, next, attr, th);
4151 rc = lod_generate_and_set_lovea(env, lo, th);
4155 lod_striping_free(env, lo);
4160 static int lod_get_default_lov_striping(const struct lu_env *env,
4161 struct lod_object *lo,
4162 struct lod_default_striping *lds);
4164 * Implementation of dt_object_operations::do_xattr_set.
4166 * Sets specified extended attribute on the object. Three types of EAs are
4168 * LOV EA - stores striping for a regular file or default striping (when set
4170 * LMV EA - stores a marker for the striped directories
4171 * DMV EA - stores default directory striping
4173 * When striping is applied to a non-striped existing object (this is called
4174 * late striping), then LOD notices the caller wants to turn the object into a
4175 * striped one. The stripe objects are created and appropriate EA is set:
4176 * LOV EA storing all the stripes directly or LMV EA storing just a small header
4177 * with striping configuration.
4179 * \see dt_object_operations::do_xattr_set() in the API description for details.
4181 static int lod_xattr_set(const struct lu_env *env,
4182 struct dt_object *dt, const struct lu_buf *buf,
4183 const char *name, int fl, struct thandle *th)
4185 struct dt_object *next = dt_object_child(dt);
4189 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4190 strcmp(name, XATTR_NAME_LMV) == 0) {
4191 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
4193 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4194 strncmp(name, XATTR_NAME_LMV, strlen(XATTR_NAME_LMV)) == 0 &&
4195 strlen(name) > strlen(XATTR_NAME_LMV) + 1) {
4196 const char *op = name + strlen(XATTR_NAME_LMV) + 1;
4200 * XATTR_NAME_LMV".add" is never called, but only declared,
4201 * because lod_xattr_set_lmv() will do the addition.
4203 if (strcmp(op, "del") == 0)
4204 rc = lod_dir_layout_delete(env, dt, buf, th);
4205 else if (strcmp(op, "set") == 0)
4206 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LMV,
4210 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4211 strcmp(name, XATTR_NAME_LOV) == 0) {
4212 struct lod_default_striping *lds = lod_lds_buf_get(env);
4213 struct lov_user_md_v1 *v1 = buf->lb_buf;
4214 char pool[LOV_MAXPOOLNAME + 1];
4217 /* get existing striping config */
4218 rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds);
4222 memset(pool, 0, sizeof(pool));
4223 if (lds->lds_def_striping_set == 1)
4224 lod_layout_get_pool(lds->lds_def_comp_entries,
4225 lds->lds_def_comp_cnt, pool,
4228 is_del = LOVEA_DELETE_VALUES(v1->lmm_stripe_size,
4229 v1->lmm_stripe_count,
4230 v1->lmm_stripe_offset,
4233 /* Retain the pool name if it is not given */
4234 if (v1->lmm_magic == LOV_USER_MAGIC_V1 && pool[0] != '\0' &&
4236 struct lod_thread_info *info = lod_env_info(env);
4237 struct lov_user_md_v3 *v3 = info->lti_ea_store;
4239 memset(v3, 0, sizeof(*v3));
4240 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4241 v3->lmm_pattern = cpu_to_le32(v1->lmm_pattern);
4242 v3->lmm_stripe_count =
4243 cpu_to_le32(v1->lmm_stripe_count);
4244 v3->lmm_stripe_offset =
4245 cpu_to_le32(v1->lmm_stripe_offset);
4246 v3->lmm_stripe_size = cpu_to_le32(v1->lmm_stripe_size);
4248 strlcpy(v3->lmm_pool_name, pool,
4249 sizeof(v3->lmm_pool_name));
4251 info->lti_buf.lb_buf = v3;
4252 info->lti_buf.lb_len = sizeof(*v3);
4253 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4256 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name,
4260 if (lds->lds_def_striping_set == 1 &&
4261 lds->lds_def_comp_entries != NULL)
4262 lod_free_def_comp_entries(lds);
4265 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4266 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
4268 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
4271 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
4272 (!strcmp(name, XATTR_NAME_LOV) ||
4273 !strncmp(name, XATTR_LUSTRE_LOV,
4274 strlen(XATTR_LUSTRE_LOV)))) {
4275 /* in case of lov EA swap, just set it
4276 * if not, it is a replay so check striping match what we
4277 * already have during req replay, declare_xattr_set()
4278 * defines striping, then create() does the work */
4279 if (fl & LU_XATTR_REPLACE) {
4280 /* free stripes, then update disk */
4281 lod_striping_free(env, lod_dt_obj(dt));
4283 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
4284 } else if (dt_object_remote(dt)) {
4285 /* This only happens during migration, see
4286 * mdd_migrate_create(), in which Master MDT will
4287 * create a remote target object, and only set
4288 * (migrating) stripe EA on the remote object,
4289 * and does not need creating each stripes. */
4290 rc = lod_sub_xattr_set(env, next, buf, name,
4292 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
4293 /* delete component(s) */
4294 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
4295 rc = lod_layout_del(env, dt, th);
4298 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
4299 * it's going to create create file with specified
4300 * component(s), the striping must have not being
4301 * cached in this case;
4303 * Otherwise, it's going to add/change component(s) to
4304 * an existing file, the striping must have been cached
4307 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
4308 !strcmp(name, XATTR_NAME_LOV),
4309 !lod_dt_obj(dt)->ldo_comp_cached));
4311 rc = lod_striped_create(env, dt, NULL, NULL, th);
4314 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4315 rc = lod_replace_parent_fid(env, dt, buf, th, false);
4320 /* then all other xattr */
4321 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4327 * Implementation of dt_object_operations::do_declare_xattr_del.
4329 * \see dt_object_operations::do_declare_xattr_del() in the API description
4332 static int lod_declare_xattr_del(const struct lu_env *env,
4333 struct dt_object *dt, const char *name,
4336 struct lod_object *lo = lod_dt_obj(dt);
4337 struct dt_object *next = dt_object_child(dt);
4342 rc = lod_sub_declare_xattr_del(env, next, name, th);
4346 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4349 /* set xattr to each stripes, if needed */
4350 rc = lod_striping_load(env, lo);
4354 if (lo->ldo_dir_stripe_count == 0)
4357 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4358 struct dt_object *dto = lo->ldo_stripe[i];
4361 rc = lod_sub_declare_xattr_del(env, dto, name, th);
4370 * Implementation of dt_object_operations::do_xattr_del.
4372 * If EA storing a regular striping is being deleted, then release
4373 * all the references to the stripe objects in core.
4375 * \see dt_object_operations::do_xattr_del() in the API description for details.
4377 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
4378 const char *name, struct thandle *th)
4380 struct dt_object *next = dt_object_child(dt);
4381 struct lod_object *lo = lod_dt_obj(dt);
4386 if (!strcmp(name, XATTR_NAME_LOV) || !strcmp(name, XATTR_NAME_LMV))
4387 lod_striping_free(env, lod_dt_obj(dt));
4389 rc = lod_sub_xattr_del(env, next, name, th);
4390 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
4393 if (lo->ldo_dir_stripe_count == 0)
4396 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4397 struct dt_object *dto = lo->ldo_stripe[i];
4401 rc = lod_sub_xattr_del(env, dto, name, th);
4410 * Implementation of dt_object_operations::do_xattr_list.
4412 * \see dt_object_operations::do_xattr_list() in the API description
4415 static int lod_xattr_list(const struct lu_env *env,
4416 struct dt_object *dt, const struct lu_buf *buf)
4418 return dt_xattr_list(env, dt_object_child(dt), buf);
4421 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
4423 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
4427 * Copy OST list from layout provided by user.
4429 * \param[in] lod_comp layout_component to be filled
4430 * \param[in] v3 LOV EA V3 user data
4432 * \retval 0 on success
4433 * \retval negative if failed
4435 int lod_comp_copy_ost_lists(struct lod_layout_component *lod_comp,
4436 struct lov_user_md_v3 *v3)
4442 if (v3->lmm_stripe_offset == LOV_OFFSET_DEFAULT)
4443 v3->lmm_stripe_offset = v3->lmm_objects[0].l_ost_idx;
4445 if (lod_comp->llc_ostlist.op_array) {
4446 if (lod_comp->llc_ostlist.op_size >=
4447 v3->lmm_stripe_count * sizeof(__u32)) {
4448 lod_comp->llc_ostlist.op_count =
4449 v3->lmm_stripe_count;
4452 OBD_FREE(lod_comp->llc_ostlist.op_array,
4453 lod_comp->llc_ostlist.op_size);
4456 /* copy ost list from lmm */
4457 lod_comp->llc_ostlist.op_count = v3->lmm_stripe_count;
4458 lod_comp->llc_ostlist.op_size = v3->lmm_stripe_count * sizeof(__u32);
4459 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
4460 lod_comp->llc_ostlist.op_size);
4461 if (!lod_comp->llc_ostlist.op_array)
4464 for (j = 0; j < v3->lmm_stripe_count; j++) {
4465 lod_comp->llc_ostlist.op_array[j] =
4466 v3->lmm_objects[j].l_ost_idx;
4474 * Get default striping.
4476 * \param[in] env execution environment
4477 * \param[in] lo object
4478 * \param[out] lds default striping
4480 * \retval 0 on success
4481 * \retval negative if failed
4483 static int lod_get_default_lov_striping(const struct lu_env *env,
4484 struct lod_object *lo,
4485 struct lod_default_striping *lds)
4487 struct lod_thread_info *info = lod_env_info(env);
4488 struct lov_user_md_v1 *v1 = NULL;
4489 struct lov_user_md_v3 *v3 = NULL;
4490 struct lov_comp_md_v1 *comp_v1 = NULL;
4497 lds->lds_def_striping_set = 0;
4499 rc = lod_get_lov_ea(env, lo);
4503 if (rc < (typeof(rc))sizeof(struct lov_user_md))
4506 v1 = info->lti_ea_store;
4507 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
4508 lustre_swab_lov_user_md_v1(v1);
4509 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
4510 v3 = (struct lov_user_md_v3 *)v1;
4511 lustre_swab_lov_user_md_v3(v3);
4512 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_SPECIFIC)) {
4513 v3 = (struct lov_user_md_v3 *)v1;
4514 lustre_swab_lov_user_md_v3(v3);
4515 lustre_swab_lov_user_md_objects(v3->lmm_objects,
4516 v3->lmm_stripe_count);
4517 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
4518 comp_v1 = (struct lov_comp_md_v1 *)v1;
4519 lustre_swab_lov_comp_md_v1(comp_v1);
4522 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1 &&
4523 v1->lmm_magic != LOV_MAGIC_COMP_V1 &&
4524 v1->lmm_magic != LOV_USER_MAGIC_SPECIFIC)
4527 if (v1->lmm_magic == LOV_MAGIC_COMP_V1) {
4528 comp_v1 = (struct lov_comp_md_v1 *)v1;
4529 comp_cnt = comp_v1->lcm_entry_count;
4532 mirror_cnt = comp_v1->lcm_mirror_count + 1;
4540 /* realloc default comp entries if necessary */
4541 rc = lod_def_striping_comp_resize(lds, comp_cnt);
4545 lds->lds_def_comp_cnt = comp_cnt;
4546 lds->lds_def_striping_is_composite = composite;
4547 lds->lds_def_mirror_cnt = mirror_cnt;
4549 for (i = 0; i < comp_cnt; i++) {
4550 struct lod_layout_component *lod_comp;
4553 lod_comp = &lds->lds_def_comp_entries[i];
4555 * reset lod_comp values, llc_stripes is always NULL in
4556 * the default striping template, llc_pool will be reset
4559 memset(lod_comp, 0, offsetof(typeof(*lod_comp), llc_pool));
4562 v1 = (struct lov_user_md *)((char *)comp_v1 +
4563 comp_v1->lcm_entries[i].lcme_offset);
4564 lod_comp->llc_extent =
4565 comp_v1->lcm_entries[i].lcme_extent;
4566 /* We only inherit certain flags from the layout */
4567 lod_comp->llc_flags =
4568 comp_v1->lcm_entries[i].lcme_flags &
4569 LCME_TEMPLATE_FLAGS;
4572 if (v1->lmm_pattern != LOV_PATTERN_RAID0 &&
4573 v1->lmm_pattern != LOV_PATTERN_MDT &&
4574 v1->lmm_pattern != 0) {
4575 lod_free_def_comp_entries(lds);
4579 CDEBUG(D_LAYOUT, DFID" stripe_count=%d stripe_size=%d "
4580 "stripe_offset=%d\n",
4581 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
4582 (int)v1->lmm_stripe_count, (int)v1->lmm_stripe_size,
4583 (int)v1->lmm_stripe_offset);
4585 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
4586 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
4587 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
4588 lod_comp->llc_pattern = v1->lmm_pattern;
4591 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
4592 /* XXX: sanity check here */
4593 v3 = (struct lov_user_md_v3 *) v1;
4594 if (v3->lmm_pool_name[0] != '\0')
4595 pool = v3->lmm_pool_name;
4597 lod_set_def_pool(lds, i, pool);
4598 if (v1->lmm_magic == LOV_USER_MAGIC_SPECIFIC) {
4599 v3 = (struct lov_user_md_v3 *)v1;
4600 rc = lod_comp_copy_ost_lists(lod_comp, v3);
4603 } else if (lod_comp->llc_ostlist.op_array &&
4604 lod_comp->llc_ostlist.op_count) {
4605 for (j = 0; j < lod_comp->llc_ostlist.op_count; j++)
4606 lod_comp->llc_ostlist.op_array[j] = -1;
4607 lod_comp->llc_ostlist.op_count = 0;
4611 lds->lds_def_striping_set = 1;
4616 * Get default directory striping.
4618 * \param[in] env execution environment
4619 * \param[in] lo object
4620 * \param[out] lds default striping
4622 * \retval 0 on success
4623 * \retval negative if failed
4625 static int lod_get_default_lmv_striping(const struct lu_env *env,
4626 struct lod_object *lo,
4627 struct lod_default_striping *lds)
4629 struct lod_thread_info *info = lod_env_info(env);
4630 struct lmv_user_md_v1 *v1 = NULL;
4634 lds->lds_dir_def_striping_set = 0;
4635 rc = lod_get_default_lmv_ea(env, lo);
4639 if (rc < (typeof(rc))sizeof(struct lmv_user_md))
4642 v1 = info->lti_ea_store;
4644 lds->lds_dir_def_stripe_count = le32_to_cpu(v1->lum_stripe_count);
4645 lds->lds_dir_def_stripe_offset = le32_to_cpu(v1->lum_stripe_offset);
4646 lds->lds_dir_def_hash_type = le32_to_cpu(v1->lum_hash_type);
4647 lds->lds_dir_def_striping_set = 1;
4653 * Get default striping in the object.
4655 * Get object default striping and default directory striping.
4657 * \param[in] env execution environment
4658 * \param[in] lo object
4659 * \param[out] lds default striping
4661 * \retval 0 on success
4662 * \retval negative if failed
4664 static int lod_get_default_striping(const struct lu_env *env,
4665 struct lod_object *lo,
4666 struct lod_default_striping *lds)
4670 rc = lod_get_default_lov_striping(env, lo, lds);
4671 rc1 = lod_get_default_lmv_striping(env, lo, lds);
4672 if (rc == 0 && rc1 < 0)
4679 * Apply default striping on object.
4681 * If object striping pattern is not set, set to the one in default striping.
4682 * The default striping is from parent or fs.
4684 * \param[in] lo new object
4685 * \param[in] lds default striping
4686 * \param[in] mode new object's mode
4688 static void lod_striping_from_default(struct lod_object *lo,
4689 const struct lod_default_striping *lds,
4692 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
4693 struct lov_desc *desc = &d->lod_desc;
4696 if (lds->lds_def_striping_set && S_ISREG(mode)) {
4697 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
4698 lds->lds_def_comp_cnt);
4702 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
4703 if (lds->lds_def_mirror_cnt > 1)
4704 lo->ldo_flr_state = LCM_FL_RDONLY;
4706 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4707 struct lod_layout_component *obj_comp =
4708 &lo->ldo_comp_entries[i];
4709 struct lod_layout_component *def_comp =
4710 &lds->lds_def_comp_entries[i];
4712 CDEBUG(D_LAYOUT, "Inherit from default: flags=%#x "
4713 "size=%hu nr=%u offset=%u pattern=%#x pool=%s\n",
4714 def_comp->llc_flags,
4715 def_comp->llc_stripe_size,
4716 def_comp->llc_stripe_count,
4717 def_comp->llc_stripe_offset,
4718 def_comp->llc_pattern,
4719 def_comp->llc_pool ?: "");
4721 *obj_comp = *def_comp;
4722 if (def_comp->llc_pool != NULL) {
4723 /* pointer was copied from def_comp */
4724 obj_comp->llc_pool = NULL;
4725 lod_obj_set_pool(lo, i, def_comp->llc_pool);
4729 if (def_comp->llc_ostlist.op_array &&
4730 def_comp->llc_ostlist.op_count) {
4731 OBD_ALLOC(obj_comp->llc_ostlist.op_array,
4732 obj_comp->llc_ostlist.op_size);
4733 if (!obj_comp->llc_ostlist.op_array)
4735 memcpy(obj_comp->llc_ostlist.op_array,
4736 def_comp->llc_ostlist.op_array,
4737 obj_comp->llc_ostlist.op_size);
4738 } else if (def_comp->llc_ostlist.op_array) {
4739 obj_comp->llc_ostlist.op_array = NULL;
4743 * Don't initialize these fields for plain layout
4744 * (v1/v3) here, they are inherited in the order of
4745 * 'parent' -> 'fs default (root)' -> 'global default
4746 * values for stripe_count & stripe_size'.
4748 * see lod_ah_init().
4750 if (!lo->ldo_is_composite)
4753 lod_adjust_stripe_info(obj_comp, desc);
4755 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
4756 if (lo->ldo_dir_stripe_count == 0)
4757 lo->ldo_dir_stripe_count =
4758 lds->lds_dir_def_stripe_count;
4759 if (lo->ldo_dir_stripe_offset == -1)
4760 lo->ldo_dir_stripe_offset =
4761 lds->lds_dir_def_stripe_offset;
4762 if (lo->ldo_dir_hash_type == 0)
4763 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
4765 CDEBUG(D_LAYOUT, "striping from default dir: count:%hu, "
4766 "offset:%u, hash_type:%u\n",
4767 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
4768 lo->ldo_dir_hash_type);
4772 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root)
4774 struct lod_layout_component *lod_comp;
4776 if (lo->ldo_comp_cnt == 0)
4779 if (lo->ldo_is_composite)
4782 lod_comp = &lo->ldo_comp_entries[0];
4784 if (lod_comp->llc_stripe_count <= 0 ||
4785 lod_comp->llc_stripe_size <= 0)
4788 if (from_root && (lod_comp->llc_pool == NULL ||
4789 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
4796 * Implementation of dt_object_operations::do_ah_init.
4798 * This method is used to make a decision on the striping configuration for the
4799 * object being created. It can be taken from the \a parent object if it exists,
4800 * or filesystem's default. The resulting configuration (number of stripes,
4801 * stripe size/offset, pool name, etc) is stored in the object itself and will
4802 * be used by the methods like ->doo_declare_create().
4804 * \see dt_object_operations::do_ah_init() in the API description for details.
4806 static void lod_ah_init(const struct lu_env *env,
4807 struct dt_allocation_hint *ah,
4808 struct dt_object *parent,
4809 struct dt_object *child,
4812 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
4813 struct lod_thread_info *info = lod_env_info(env);
4814 struct lod_default_striping *lds = lod_lds_buf_get(env);
4815 struct dt_object *nextp = NULL;
4816 struct dt_object *nextc;
4817 struct lod_object *lp = NULL;
4818 struct lod_object *lc;
4819 struct lov_desc *desc;
4820 struct lod_layout_component *lod_comp;
4826 if (likely(parent)) {
4827 nextp = dt_object_child(parent);
4828 lp = lod_dt_obj(parent);
4831 nextc = dt_object_child(child);
4832 lc = lod_dt_obj(child);
4834 LASSERT(!lod_obj_is_striped(child));
4835 /* default layout template may have been set on the regular file
4836 * when this is called from mdd_create_data() */
4837 if (S_ISREG(child_mode))
4838 lod_free_comp_entries(lc);
4840 if (!dt_object_exists(nextc))
4841 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
4843 if (S_ISDIR(child_mode)) {
4844 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
4846 /* other default values are 0 */
4847 lc->ldo_dir_stripe_offset = -1;
4849 /* no default striping configuration is needed for
4852 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
4853 le32_to_cpu(lum1->lum_magic) == LMV_MAGIC_FOREIGN) {
4854 lc->ldo_dir_is_foreign = true;
4855 /* keep stripe_count 0 and stripe_offset -1 */
4856 CDEBUG(D_INFO, "no default striping for foreign dir\n");
4861 * If parent object is not root directory,
4862 * then get default striping from parent object.
4864 if (likely(lp != NULL) && !fid_is_root(lod_object_fid(lp)))
4865 lod_get_default_striping(env, lp, lds);
4867 /* set child default striping info, default value is NULL */
4868 if (lds->lds_def_striping_set || lds->lds_dir_def_striping_set)
4869 lc->ldo_def_striping = lds;
4871 /* It should always honour the specified stripes */
4872 /* Note: old client (< 2.7)might also do lfs mkdir, whose EA
4873 * will have old magic. In this case, we should ignore the
4874 * stripe count and try to create dir by default stripe.
4876 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
4877 (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
4878 le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC)) {
4879 lc->ldo_dir_stripe_count =
4880 le32_to_cpu(lum1->lum_stripe_count);
4881 lc->ldo_dir_stripe_offset =
4882 le32_to_cpu(lum1->lum_stripe_offset);
4883 lc->ldo_dir_hash_type =
4884 le32_to_cpu(lum1->lum_hash_type);
4886 "set dirstripe: count %hu, offset %d, hash %u\n",
4887 lc->ldo_dir_stripe_count,
4888 (int)lc->ldo_dir_stripe_offset,
4889 lc->ldo_dir_hash_type);
4891 /* transfer defaults LMV to new directory */
4892 lod_striping_from_default(lc, lds, child_mode);
4894 /* set count 0 to create normal directory */
4895 if (lc->ldo_dir_stripe_count == 1)
4896 lc->ldo_dir_stripe_count = 0;
4899 /* shrink the stripe_count to the avaible MDT count */
4900 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
4901 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)) {
4902 lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
4903 if (lc->ldo_dir_stripe_count == 1)
4904 lc->ldo_dir_stripe_count = 0;
4907 CDEBUG(D_INFO, "final dir stripe [%hu %d %u]\n",
4908 lc->ldo_dir_stripe_count,
4909 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
4914 /* child object regular file*/
4916 if (!lod_object_will_be_striped(S_ISREG(child_mode),
4917 lu_object_fid(&child->do_lu)))
4920 /* If object is going to be striped over OSTs, transfer default
4921 * striping information to the child, so that we can use it
4922 * during declaration and creation.
4924 * Try from the parent first.
4926 if (likely(lp != NULL)) {
4927 rc = lod_get_default_lov_striping(env, lp, lds);
4929 lod_striping_from_default(lc, lds, child_mode);
4932 /* Initialize lod_device::lod_md_root object reference */
4933 if (d->lod_md_root == NULL) {
4934 struct dt_object *root;
4935 struct lod_object *lroot;
4937 lu_root_fid(&info->lti_fid);
4938 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
4939 if (!IS_ERR(root)) {
4940 lroot = lod_dt_obj(root);
4942 spin_lock(&d->lod_lock);
4943 if (d->lod_md_root != NULL)
4944 dt_object_put(env, &d->lod_md_root->ldo_obj);
4945 d->lod_md_root = lroot;
4946 spin_unlock(&d->lod_lock);
4950 /* try inherit layout from the root object (fs default) when:
4951 * - parent does not have default layout; or
4952 * - parent has plain(v1/v3) default layout, and some attributes
4953 * are not specified in the default layout;
4955 if (d->lod_md_root != NULL && lod_need_inherit_more(lc, true)) {
4956 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds);
4959 if (lc->ldo_comp_cnt == 0) {
4960 lod_striping_from_default(lc, lds, child_mode);
4961 } else if (!lds->lds_def_striping_is_composite) {
4962 struct lod_layout_component *def_comp;
4964 LASSERT(!lc->ldo_is_composite);
4965 lod_comp = &lc->ldo_comp_entries[0];
4966 def_comp = &lds->lds_def_comp_entries[0];
4968 if (lod_comp->llc_stripe_count <= 0)
4969 lod_comp->llc_stripe_count =
4970 def_comp->llc_stripe_count;
4971 if (lod_comp->llc_stripe_size <= 0)
4972 lod_comp->llc_stripe_size =
4973 def_comp->llc_stripe_size;
4974 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT &&
4975 (!lod_comp->llc_pool || !lod_comp->llc_pool[0]))
4976 lod_comp->llc_stripe_offset =
4977 def_comp->llc_stripe_offset;
4978 if (lod_comp->llc_pool == NULL)
4979 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
4984 * fs default striping may not be explicitly set, or historically set
4985 * in config log, use them.
4987 if (lod_need_inherit_more(lc, false)) {
4988 if (lc->ldo_comp_cnt == 0) {
4989 rc = lod_alloc_comp_entries(lc, 0, 1);
4991 /* fail to allocate memory, will create a
4992 * non-striped file. */
4994 lc->ldo_is_composite = 0;
4995 lod_comp = &lc->ldo_comp_entries[0];
4996 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4998 LASSERT(!lc->ldo_is_composite);
4999 lod_comp = &lc->ldo_comp_entries[0];
5000 desc = &d->lod_desc;
5001 lod_adjust_stripe_info(lod_comp, desc);
5007 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
5009 * Size initialization on late striping.
5011 * Propagate the size of a truncated object to a deferred striping.
5012 * This function handles a special case when truncate was done on a
5013 * non-striped object and now while the striping is being created
5014 * we can't lose that size, so we have to propagate it to the stripes
5017 * \param[in] env execution environment
5018 * \param[in] dt object
5019 * \param[in] th transaction handle
5021 * \retval 0 on success
5022 * \retval negative if failed
5024 static int lod_declare_init_size(const struct lu_env *env,
5025 struct dt_object *dt, struct thandle *th)
5027 struct dt_object *next = dt_object_child(dt);
5028 struct lod_object *lo = lod_dt_obj(dt);
5029 struct dt_object **objects = NULL;
5030 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
5031 uint64_t size, offs;
5032 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
5033 struct lu_extent size_ext;
5036 if (!lod_obj_is_striped(dt))
5039 rc = dt_attr_get(env, next, attr);
5040 LASSERT(attr->la_valid & LA_SIZE);
5044 size = attr->la_size;
5048 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
5049 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5050 struct lod_layout_component *lod_comp;
5051 struct lu_extent *extent;
5053 lod_comp = &lo->ldo_comp_entries[i];
5055 if (lod_comp->llc_stripe == NULL)
5058 extent = &lod_comp->llc_extent;
5059 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
5060 if (!lo->ldo_is_composite ||
5061 lu_extent_is_overlapped(extent, &size_ext)) {
5062 objects = lod_comp->llc_stripe;
5063 stripe_count = lod_comp->llc_stripe_count;
5064 stripe_size = lod_comp->llc_stripe_size;
5067 if (stripe_count == 0)
5070 LASSERT(objects != NULL && stripe_size != 0);
5071 /* ll_do_div64(a, b) returns a % b, and a = a / b */
5072 ll_do_div64(size, (__u64)stripe_size);
5073 stripe = ll_do_div64(size, (__u64)stripe_count);
5074 LASSERT(objects[stripe] != NULL);
5076 size = size * stripe_size;
5077 offs = attr->la_size;
5078 size += ll_do_div64(offs, stripe_size);
5080 attr->la_valid = LA_SIZE;
5081 attr->la_size = size;
5083 rc = lod_sub_declare_attr_set(env, objects[stripe],
5092 * Declare creation of striped object.
5094 * The function declares creation stripes for a regular object. The function
5095 * also declares whether the stripes will be created with non-zero size if
5096 * previously size was set non-zero on the master object. If object \a dt is
5097 * not local, then only fully defined striping can be applied in \a lovea.
5098 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
5101 * \param[in] env execution environment
5102 * \param[in] dt object
5103 * \param[in] attr attributes the stripes will be created with
5104 * \param[in] lovea a buffer containing striping description
5105 * \param[in] th transaction handle
5107 * \retval 0 on success
5108 * \retval negative if failed
5110 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
5111 struct lu_attr *attr,
5112 const struct lu_buf *lovea, struct thandle *th)
5114 struct lod_thread_info *info = lod_env_info(env);
5115 struct dt_object *next = dt_object_child(dt);
5116 struct lod_object *lo = lod_dt_obj(dt);
5120 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
5121 GOTO(out, rc = -ENOMEM);
5123 if (!dt_object_remote(next)) {
5124 /* choose OST and generate appropriate objects */
5125 rc = lod_prepare_create(env, lo, attr, lovea, th);
5130 * declare storage for striping data
5132 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
5134 /* LOD can not choose OST objects for remote objects, i.e.
5135 * stripes must be ready before that. Right now, it can only
5136 * happen during migrate, i.e. migrate process needs to create
5137 * remote regular file (mdd_migrate_create), then the migrate
5138 * process will provide stripeEA. */
5139 LASSERT(lovea != NULL);
5140 info->lti_buf = *lovea;
5143 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
5144 XATTR_NAME_LOV, 0, th);
5149 * if striping is created with local object's size > 0,
5150 * we have to propagate this size to specific object
5151 * the case is possible only when local object was created previously
5153 if (dt_object_exists(next))
5154 rc = lod_declare_init_size(env, dt, th);
5157 /* failed to create striping or to set initial size, let's reset
5158 * config so that others don't get confused */
5160 lod_striping_free(env, lo);
5166 * Implementation of dt_object_operations::do_declare_create.
5168 * The method declares creation of a new object. If the object will be striped,
5169 * then helper functions are called to find FIDs for the stripes, declare
5170 * creation of the stripes and declare initialization of the striping
5171 * information to be stored in the master object.
5173 * \see dt_object_operations::do_declare_create() in the API description
5176 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
5177 struct lu_attr *attr,
5178 struct dt_allocation_hint *hint,
5179 struct dt_object_format *dof, struct thandle *th)
5181 struct dt_object *next = dt_object_child(dt);
5182 struct lod_object *lo = lod_dt_obj(dt);
5191 * first of all, we declare creation of local object
5193 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
5198 * it's lod_ah_init() that has decided the object will be striped
5200 if (dof->dof_type == DFT_REGULAR) {
5201 /* callers don't want stripes */
5202 /* XXX: all tricky interactions with ->ah_make_hint() decided
5203 * to use striping, then ->declare_create() behaving differently
5204 * should be cleaned */
5205 if (dof->u.dof_reg.striped != 0)
5206 rc = lod_declare_striped_create(env, dt, attr,
5208 } else if (dof->dof_type == DFT_DIR) {
5209 struct seq_server_site *ss;
5210 struct lu_buf buf = { NULL };
5211 struct lu_buf *lmu = NULL;
5213 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
5215 /* If the parent has default stripeEA, and client
5216 * did not find it before sending create request,
5217 * then MDT will return -EREMOTE, and client will
5218 * retrieve the default stripeEA and re-create the
5221 * Note: if dah_eadata != NULL, it means creating the
5222 * striped directory with specified stripeEA, then it
5223 * should ignore the default stripeEA */
5224 if (hint != NULL && hint->dah_eadata == NULL) {
5225 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
5226 GOTO(out, rc = -EREMOTE);
5228 if (lo->ldo_dir_stripe_offset == -1) {
5229 /* child and parent should be in the same MDT */
5230 if (hint->dah_parent != NULL &&
5231 dt_object_remote(hint->dah_parent))
5232 GOTO(out, rc = -EREMOTE);
5233 } else if (lo->ldo_dir_stripe_offset !=
5235 struct lod_device *lod;
5236 struct lod_tgt_descs *ltd;
5237 struct lod_tgt_desc *tgt = NULL;
5238 bool found_mdt = false;
5241 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5242 ltd = &lod->lod_mdt_descs;
5243 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
5244 tgt = LTD_TGT(ltd, i);
5245 if (tgt->ltd_index ==
5246 lo->ldo_dir_stripe_offset) {
5252 /* If the MDT indicated by stripe_offset can be
5253 * found, then tell client to resend the create
5254 * request to the correct MDT, otherwise return
5255 * error to client */
5257 GOTO(out, rc = -EREMOTE);
5259 GOTO(out, rc = -EINVAL);
5261 } else if (hint && hint->dah_eadata) {
5263 lmu->lb_buf = (void *)hint->dah_eadata;
5264 lmu->lb_len = hint->dah_eadata_len;
5267 rc = lod_declare_dir_striping_create(env, dt, attr, lmu, dof,
5271 /* failed to create striping or to set initial size, let's reset
5272 * config so that others don't get confused */
5274 lod_striping_free(env, lo);
5279 * Generate component ID for new created component.
5281 * \param[in] lo LOD object
5282 * \param[in] comp_idx index of ldo_comp_entries
5284 * \retval component ID on success
5285 * \retval LCME_ID_INVAL on failure
5287 static __u32 lod_gen_component_id(struct lod_object *lo,
5288 int mirror_id, int comp_idx)
5290 struct lod_layout_component *lod_comp;
5291 __u32 id, start, end;
5294 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
5296 lod_obj_inc_layout_gen(lo);
5297 id = lo->ldo_layout_gen;
5298 if (likely(id <= SEQ_ID_MAX))
5299 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
5301 /* Layout generation wraps, need to check collisions. */
5302 start = id & SEQ_ID_MASK;
5305 for (id = start; id <= end; id++) {
5306 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5307 lod_comp = &lo->ldo_comp_entries[i];
5308 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
5311 /* Found the ununsed ID */
5312 if (i == lo->ldo_comp_cnt)
5313 RETURN(pflr_id(mirror_id, id));
5315 if (end == LCME_ID_MAX) {
5317 end = min(lo->ldo_layout_gen & LCME_ID_MASK,
5318 (__u32)(LCME_ID_MAX - 1));
5322 RETURN(LCME_ID_INVAL);
5326 * Creation of a striped regular object.
5328 * The function is called to create the stripe objects for a regular
5329 * striped file. This can happen at the initial object creation or
5330 * when the caller asks LOD to do so using ->do_xattr_set() method
5331 * (so called late striping). Notice all the information are already
5332 * prepared in the form of the list of objects (ldo_stripe field).
5333 * This is done during declare phase.
5335 * \param[in] env execution environment
5336 * \param[in] dt object
5337 * \param[in] attr attributes the stripes will be created with
5338 * \param[in] dof format of stripes (see OSD API description)
5339 * \param[in] th transaction handle
5341 * \retval 0 on success
5342 * \retval negative if failed
5344 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
5345 struct lu_attr *attr, struct dt_object_format *dof,
5348 struct lod_layout_component *lod_comp;
5349 struct lod_object *lo = lod_dt_obj(dt);
5354 LASSERT((lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL) ||
5355 lo->ldo_is_foreign);
5357 mirror_id = 0; /* non-flr file's mirror_id is 0 */
5358 if (lo->ldo_mirror_count > 1) {
5359 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5360 lod_comp = &lo->ldo_comp_entries[i];
5361 if (lod_comp->llc_id != LCME_ID_INVAL &&
5362 mirror_id_of(lod_comp->llc_id) > mirror_id)
5363 mirror_id = mirror_id_of(lod_comp->llc_id);
5367 /* create all underlying objects */
5368 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5369 lod_comp = &lo->ldo_comp_entries[i];
5371 if (lod_comp->llc_id == LCME_ID_INVAL) {
5372 /* only the component of FLR layout with more than 1
5373 * mirror has mirror ID in its component ID.
5375 if (lod_comp->llc_extent.e_start == 0 &&
5376 lo->ldo_mirror_count > 1)
5379 lod_comp->llc_id = lod_gen_component_id(lo,
5381 if (lod_comp->llc_id == LCME_ID_INVAL)
5382 GOTO(out, rc = -ERANGE);
5385 if (lod_comp_inited(lod_comp))
5388 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
5389 lod_comp_set_init(lod_comp);
5391 if (lov_pattern(lod_comp->llc_pattern) == LOV_PATTERN_MDT)
5392 lod_comp_set_init(lod_comp);
5394 if (lod_comp->llc_stripe == NULL)
5397 LASSERT(lod_comp->llc_stripe_count);
5398 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
5399 struct dt_object *object = lod_comp->llc_stripe[j];
5400 LASSERT(object != NULL);
5401 rc = lod_sub_create(env, object, attr, NULL, dof, th);
5405 lod_comp_set_init(lod_comp);
5408 rc = lod_fill_mirrors(lo);
5412 rc = lod_generate_and_set_lovea(env, lo, th);
5416 lo->ldo_comp_cached = 1;
5420 lod_striping_free(env, lo);
5424 static inline bool lod_obj_is_dom(struct dt_object *dt)
5426 struct lod_object *lo = lod_dt_obj(dt);
5428 if (!dt_object_exists(dt_object_child(dt)))
5431 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
5434 if (!lo->ldo_comp_cnt)
5437 return (lov_pattern(lo->ldo_comp_entries[0].llc_pattern) ==
5442 * Implementation of dt_object_operations::do_create.
5444 * If any of preceeding methods (like ->do_declare_create(),
5445 * ->do_ah_init(), etc) chose to create a striped object,
5446 * then this method will create the master and the stripes.
5448 * \see dt_object_operations::do_create() in the API description for details.
5450 static int lod_create(const struct lu_env *env, struct dt_object *dt,
5451 struct lu_attr *attr, struct dt_allocation_hint *hint,
5452 struct dt_object_format *dof, struct thandle *th)
5457 /* create local object */
5458 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
5462 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
5463 (lod_obj_is_striped(dt) || lod_obj_is_dom(dt)) &&
5464 dof->u.dof_reg.striped != 0) {
5465 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
5466 rc = lod_striped_create(env, dt, attr, dof, th);
5473 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
5474 struct dt_object *dt, struct thandle *th,
5475 int comp_idx, int stripe_idx,
5476 struct lod_obj_stripe_cb_data *data)
5478 if (data->locd_declare)
5479 return lod_sub_declare_destroy(env, dt, th);
5480 else if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
5481 stripe_idx == cfs_fail_val)
5482 return lod_sub_destroy(env, dt, th);
5488 * Implementation of dt_object_operations::do_declare_destroy.
5490 * If the object is a striped directory, then the function declares reference
5491 * removal from the master object (this is an index) to the stripes and declares
5492 * destroy of all the stripes. In all the cases, it declares an intention to
5493 * destroy the object itself.
5495 * \see dt_object_operations::do_declare_destroy() in the API description
5498 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
5501 struct dt_object *next = dt_object_child(dt);
5502 struct lod_object *lo = lod_dt_obj(dt);
5503 struct lod_thread_info *info = lod_env_info(env);
5504 char *stripe_name = info->lti_key;
5509 * load striping information, notice we don't do this when object
5510 * is being initialized as we don't need this information till
5511 * few specific cases like destroy, chown
5513 rc = lod_striping_load(env, lo);
5517 /* declare destroy for all underlying objects */
5518 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5519 rc = next->do_ops->do_index_try(env, next,
5520 &dt_directory_features);
5524 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5525 rc = lod_sub_declare_ref_del(env, next, th);
5529 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
5530 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
5532 rc = lod_sub_declare_delete(env, next,
5533 (const struct dt_key *)stripe_name, th);
5540 * we declare destroy for the local object
5542 rc = lod_sub_declare_destroy(env, next, th);
5546 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
5547 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
5550 if (!lod_obj_is_striped(dt))
5553 /* declare destroy all striped objects */
5554 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5555 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5556 if (lo->ldo_stripe[i] == NULL)
5559 rc = lod_sub_declare_ref_del(env, lo->ldo_stripe[i],
5562 rc = lod_sub_declare_destroy(env, lo->ldo_stripe[i],
5568 struct lod_obj_stripe_cb_data data = { { 0 } };
5570 data.locd_declare = true;
5571 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
5572 rc = lod_obj_for_each_stripe(env, lo, th, &data);
5579 * Implementation of dt_object_operations::do_destroy.
5581 * If the object is a striped directory, then the function removes references
5582 * from the master object (this is an index) to the stripes and destroys all
5583 * the stripes. In all the cases, the function destroys the object itself.
5585 * \see dt_object_operations::do_destroy() in the API description for details.
5587 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
5590 struct dt_object *next = dt_object_child(dt);
5591 struct lod_object *lo = lod_dt_obj(dt);
5592 struct lod_thread_info *info = lod_env_info(env);
5593 char *stripe_name = info->lti_key;
5598 /* destroy sub-stripe of master object */
5599 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5600 rc = next->do_ops->do_index_try(env, next,
5601 &dt_directory_features);
5605 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5606 rc = lod_sub_ref_del(env, next, th);
5610 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
5611 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
5614 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
5615 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
5616 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)));
5618 rc = lod_sub_delete(env, next,
5619 (const struct dt_key *)stripe_name, th);
5625 rc = lod_sub_destroy(env, next, th);
5629 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
5630 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
5633 if (!lod_obj_is_striped(dt))
5636 /* destroy all striped objects */
5637 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5638 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5639 if (lo->ldo_stripe[i] == NULL)
5641 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
5642 i == cfs_fail_val) {
5643 dt_write_lock(env, lo->ldo_stripe[i],
5645 rc = lod_sub_ref_del(env, lo->ldo_stripe[i],
5647 dt_write_unlock(env, lo->ldo_stripe[i]);
5651 rc = lod_sub_destroy(env, lo->ldo_stripe[i],
5658 struct lod_obj_stripe_cb_data data = { { 0 } };
5660 data.locd_declare = false;
5661 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
5662 rc = lod_obj_for_each_stripe(env, lo, th, &data);
5669 * Implementation of dt_object_operations::do_declare_ref_add.
5671 * \see dt_object_operations::do_declare_ref_add() in the API description
5674 static int lod_declare_ref_add(const struct lu_env *env,
5675 struct dt_object *dt, struct thandle *th)
5677 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
5681 * Implementation of dt_object_operations::do_ref_add.
5683 * \see dt_object_operations::do_ref_add() in the API description for details.
5685 static int lod_ref_add(const struct lu_env *env,
5686 struct dt_object *dt, struct thandle *th)
5688 return lod_sub_ref_add(env, dt_object_child(dt), th);
5692 * Implementation of dt_object_operations::do_declare_ref_del.
5694 * \see dt_object_operations::do_declare_ref_del() in the API description
5697 static int lod_declare_ref_del(const struct lu_env *env,
5698 struct dt_object *dt, struct thandle *th)
5700 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
5704 * Implementation of dt_object_operations::do_ref_del
5706 * \see dt_object_operations::do_ref_del() in the API description for details.
5708 static int lod_ref_del(const struct lu_env *env,
5709 struct dt_object *dt, struct thandle *th)
5711 return lod_sub_ref_del(env, dt_object_child(dt), th);
5715 * Implementation of dt_object_operations::do_object_sync.
5717 * \see dt_object_operations::do_object_sync() in the API description
5720 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
5721 __u64 start, __u64 end)
5723 return dt_object_sync(env, dt_object_child(dt), start, end);
5727 * Implementation of dt_object_operations::do_object_unlock.
5729 * Used to release LDLM lock(s).
5731 * \see dt_object_operations::do_object_unlock() in the API description
5734 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
5735 struct ldlm_enqueue_info *einfo,
5736 union ldlm_policy_data *policy)
5738 struct lod_object *lo = lod_dt_obj(dt);
5739 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
5740 int slave_locks_size;
5744 if (slave_locks == NULL)
5747 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
5748 /* Note: for remote lock for single stripe dir, MDT will cancel
5749 * the lock by lockh directly */
5750 LASSERT(!dt_object_remote(dt_object_child(dt)));
5752 /* locks were unlocked in MDT layer */
5753 for (i = 0; i < slave_locks->ha_count; i++)
5754 LASSERT(!lustre_handle_is_used(&slave_locks->ha_handles[i]));
5757 * NB, ha_count may not equal to ldo_dir_stripe_count, because dir
5758 * layout may change, e.g., shrink dir layout after migration.
5760 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
5761 dt_invalidate(env, lo->ldo_stripe[i]);
5763 slave_locks_size = offsetof(typeof(*slave_locks),
5764 ha_handles[slave_locks->ha_count]);
5765 OBD_FREE(slave_locks, slave_locks_size);
5766 einfo->ei_cbdata = NULL;
5772 * Implementation of dt_object_operations::do_object_lock.
5774 * Used to get LDLM lock on the non-striped and striped objects.
5776 * \see dt_object_operations::do_object_lock() in the API description
5779 static int lod_object_lock(const struct lu_env *env,
5780 struct dt_object *dt,
5781 struct lustre_handle *lh,
5782 struct ldlm_enqueue_info *einfo,
5783 union ldlm_policy_data *policy)
5785 struct lod_object *lo = lod_dt_obj(dt);
5786 int slave_locks_size;
5787 struct lustre_handle_array *slave_locks = NULL;
5792 /* remote object lock */
5793 if (!einfo->ei_enq_slave) {
5794 LASSERT(dt_object_remote(dt));
5795 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
5799 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
5802 rc = lod_striping_load(env, lo);
5807 if (lo->ldo_dir_stripe_count <= 1)
5810 slave_locks_size = offsetof(typeof(*slave_locks),
5811 ha_handles[lo->ldo_dir_stripe_count]);
5812 /* Freed in lod_object_unlock */
5813 OBD_ALLOC(slave_locks, slave_locks_size);
5816 slave_locks->ha_count = lo->ldo_dir_stripe_count;
5818 /* striped directory lock */
5819 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5820 struct lustre_handle lockh;
5821 struct ldlm_res_id *res_id;
5823 res_id = &lod_env_info(env)->lti_res_id;
5824 fid_build_reg_res_name(lu_object_fid(&lo->ldo_stripe[i]->do_lu),
5826 einfo->ei_res_id = res_id;
5828 LASSERT(lo->ldo_stripe[i] != NULL);
5829 if (dt_object_remote(lo->ldo_stripe[i])) {
5830 set_bit(i, (void *)slave_locks->ha_map);
5831 rc = dt_object_lock(env, lo->ldo_stripe[i], &lockh,
5834 struct ldlm_namespace *ns = einfo->ei_namespace;
5835 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
5836 ldlm_completion_callback completion = einfo->ei_cb_cp;
5837 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
5839 if (einfo->ei_mode == LCK_PW ||
5840 einfo->ei_mode == LCK_EX)
5841 dlmflags |= LDLM_FL_COS_INCOMPAT;
5843 LASSERT(ns != NULL);
5844 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS,
5845 policy, einfo->ei_mode,
5846 &dlmflags, blocking,
5848 NULL, 0, LVB_T_NONE,
5853 ldlm_lock_decref_and_cancel(
5854 &slave_locks->ha_handles[i],
5856 OBD_FREE(slave_locks, slave_locks_size);
5859 slave_locks->ha_handles[i] = lockh;
5861 einfo->ei_cbdata = slave_locks;
5867 * Implementation of dt_object_operations::do_invalidate.
5869 * \see dt_object_operations::do_invalidate() in the API description for details
5871 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
5873 return dt_invalidate(env, dt_object_child(dt));
5876 static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
5880 /* clear memory region that will be used for layout change */
5881 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
5882 info->lti_count = 0;
5884 if (info->lti_comp_size >= comp_cnt)
5887 if (info->lti_comp_size > 0) {
5888 OBD_FREE(info->lti_comp_idx,
5889 info->lti_comp_size * sizeof(__u32));
5890 info->lti_comp_size = 0;
5893 OBD_ALLOC(info->lti_comp_idx, comp_cnt * sizeof(__u32));
5894 if (!info->lti_comp_idx)
5897 info->lti_comp_size = comp_cnt;
5901 static int lod_declare_instantiate_components(const struct lu_env *env,
5902 struct lod_object *lo, struct thandle *th)
5904 struct lod_thread_info *info = lod_env_info(env);
5909 LASSERT(info->lti_count < lo->ldo_comp_cnt);
5911 for (i = 0; i < info->lti_count; i++) {
5912 rc = lod_qos_prep_create(env, lo, NULL, th,
5913 info->lti_comp_idx[i]);
5919 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
5920 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
5921 &info->lti_buf, XATTR_NAME_LOV, 0, th);
5927 static int lod_declare_update_plain(const struct lu_env *env,
5928 struct lod_object *lo, struct layout_intent *layout,
5929 const struct lu_buf *buf, struct thandle *th)
5931 struct lod_thread_info *info = lod_env_info(env);
5932 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5933 struct lod_layout_component *lod_comp;
5934 struct lov_comp_md_v1 *comp_v1 = NULL;
5935 bool replay = false;
5939 LASSERT(lo->ldo_flr_state == LCM_FL_NONE);
5942 * In case the client is passing lovea, which only happens during
5943 * the replay of layout intent write RPC for now, we may need to
5944 * parse the lovea and apply new layout configuration.
5946 if (buf && buf->lb_len) {
5947 struct lov_user_md_v1 *v1 = buf->lb_buf;
5949 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
5950 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
5951 LOV_MAGIC_COMP_V1)) {
5952 CERROR("%s: the replay buffer of layout extend "
5953 "(magic %#x) does not contain expected "
5954 "composite layout.\n",
5955 lod2obd(d)->obd_name, v1->lmm_magic);
5956 GOTO(out, rc = -EINVAL);
5959 rc = lod_use_defined_striping(env, lo, buf);
5962 lo->ldo_comp_cached = 1;
5964 rc = lod_get_lov_ea(env, lo);
5967 /* old on-disk EA is stored in info->lti_buf */
5968 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
5971 /* non replay path */
5972 rc = lod_striping_load(env, lo);
5977 /* Make sure defined layout covers the requested write range. */
5978 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
5979 if (lo->ldo_comp_cnt > 1 &&
5980 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
5981 lod_comp->llc_extent.e_end < layout->li_extent.e_end) {
5982 CDEBUG(replay ? D_ERROR : D_LAYOUT,
5983 "%s: the defined layout [0, %#llx) does not covers "
5984 "the write range "DEXT"\n",
5985 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
5986 PEXT(&layout->li_extent));
5987 GOTO(out, rc = -EINVAL);
5990 CDEBUG(D_LAYOUT, "%s: "DFID": instantiate components "DEXT"\n",
5991 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
5992 PEXT(&layout->li_extent));
5995 * Iterate ld->ldo_comp_entries, find the component whose extent under
5996 * the write range and not instantianted.
5998 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5999 lod_comp = &lo->ldo_comp_entries[i];
6001 if (lod_comp->llc_extent.e_start >= layout->li_extent.e_end)
6005 if (lod_comp_inited(lod_comp))
6009 * In replay path, lod_comp is the EA passed by
6010 * client replay buffer, comp_v1 is the pre-recovery
6011 * on-disk EA, we'd sift out those components which
6012 * were init-ed in the on-disk EA.
6014 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
6019 * this component hasn't instantiated in normal path, or during
6020 * replay it needs replay the instantiation.
6023 /* A released component is being extended */
6024 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
6025 GOTO(out, rc = -EINVAL);
6027 LASSERT(info->lti_comp_idx != NULL);
6028 info->lti_comp_idx[info->lti_count++] = i;
6031 if (info->lti_count == 0)
6034 lod_obj_inc_layout_gen(lo);
6035 rc = lod_declare_instantiate_components(env, lo, th);
6038 lod_striping_free(env, lo);
6042 static inline int lod_comp_index(struct lod_object *lo,
6043 struct lod_layout_component *lod_comp)
6045 LASSERT(lod_comp >= lo->ldo_comp_entries &&
6046 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
6048 return lod_comp - lo->ldo_comp_entries;
6052 * Stale other mirrors by writing extent.
6054 static void lod_stale_components(struct lod_object *lo, int primary,
6055 struct lu_extent *extent)
6057 struct lod_layout_component *pri_comp, *lod_comp;
6060 /* The writing extent decides which components in the primary
6061 * are affected... */
6062 CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
6063 lod_foreach_mirror_comp(pri_comp, lo, primary) {
6064 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
6067 CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
6068 lod_comp_index(lo, pri_comp),
6069 PEXT(&pri_comp->llc_extent));
6071 for (i = 0; i < lo->ldo_mirror_count; i++) {
6075 /* ... and then stale other components that are
6076 * overlapping with primary components */
6077 lod_foreach_mirror_comp(lod_comp, lo, i) {
6078 if (!lu_extent_is_overlapped(
6079 &pri_comp->llc_extent,
6080 &lod_comp->llc_extent))
6083 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
6084 i, lod_comp_index(lo, lod_comp));
6086 lod_comp->llc_flags |= LCME_FL_STALE;
6087 lo->ldo_mirrors[i].lme_stale = 1;
6094 * check an OST's availability
6095 * \param[in] env execution environment
6096 * \param[in] lo lod object
6097 * \param[in] dt dt object
6098 * \param[in] index mirror index
6100 * \retval negative if failed
6101 * \retval 1 if \a dt is available
6102 * \retval 0 if \a dt is not available
6104 static inline int lod_check_ost_avail(const struct lu_env *env,
6105 struct lod_object *lo,
6106 struct dt_object *dt, int index)
6108 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6109 struct lod_tgt_desc *ost;
6111 int type = LU_SEQ_RANGE_OST;
6114 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &idx, &type);
6116 CERROR("%s: can't locate "DFID":rc = %d\n",
6117 lod2obd(lod)->obd_name, PFID(lu_object_fid(&dt->do_lu)),
6122 ost = OST_TGT(lod, idx);
6123 if (ost->ltd_statfs.os_state &
6124 (OS_STATE_READONLY | OS_STATE_ENOSPC | OS_STATE_ENOINO |
6125 OS_STATE_NOPRECREATE) ||
6126 ost->ltd_active == 0) {
6127 CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail, rc = %d\n",
6128 PFID(lod_object_fid(lo)), index, idx, rc);
6136 * Pick primary mirror for write
6137 * \param[in] env execution environment
6138 * \param[in] lo object
6139 * \param[in] extent write range
6141 static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo,
6142 struct lu_extent *extent)
6144 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6145 unsigned int seq = 0;
6146 struct lod_layout_component *lod_comp;
6148 int picked = -1, second_pick = -1, third_pick = -1;
6151 if (OBD_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
6152 get_random_bytes(&seq, sizeof(seq));
6153 seq %= lo->ldo_mirror_count;
6157 * Pick a mirror as the primary, and check the availability of OSTs.
6159 * This algo can be revised later after knowing the topology of
6162 lod_qos_statfs_update(env, lod);
6163 for (i = 0; i < lo->ldo_mirror_count; i++) {
6164 bool ost_avail = true;
6165 int index = (i + seq) % lo->ldo_mirror_count;
6167 if (lo->ldo_mirrors[index].lme_stale) {
6168 CDEBUG(D_LAYOUT, DFID": mirror %d stale\n",
6169 PFID(lod_object_fid(lo)), index);
6173 /* 2nd pick is for the primary mirror containing unavail OST */
6174 if (lo->ldo_mirrors[index].lme_primary && second_pick < 0)
6175 second_pick = index;
6177 /* 3rd pick is for non-primary mirror containing unavail OST */
6178 if (second_pick < 0 && third_pick < 0)
6182 * we found a non-primary 1st pick, we'd like to find a
6183 * potential pirmary mirror.
6185 if (picked >= 0 && !lo->ldo_mirrors[index].lme_primary)
6188 /* check the availability of OSTs */
6189 lod_foreach_mirror_comp(lod_comp, lo, index) {
6190 if (!lod_comp_inited(lod_comp) || !lod_comp->llc_stripe)
6193 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6194 struct dt_object *dt = lod_comp->llc_stripe[j];
6196 rc = lod_check_ost_avail(env, lo, dt, index);
6203 } /* for all dt object in one component */
6206 } /* for all components in a mirror */
6209 * the OSTs where allocated objects locates in the components
6210 * of the mirror are available.
6215 /* this mirror has all OSTs available */
6219 * primary with all OSTs are available, this is the perfect
6222 if (lo->ldo_mirrors[index].lme_primary)
6224 } /* for all mirrors */
6226 /* failed to pick a sound mirror, lower our expectation */
6228 picked = second_pick;
6230 picked = third_pick;
6237 static int lod_prepare_resync_mirror(const struct lu_env *env,
6238 struct lod_object *lo,
6241 struct lod_thread_info *info = lod_env_info(env);
6242 struct lod_layout_component *lod_comp;
6243 bool neg = !!(MIRROR_ID_NEG & mirror_id);
6246 mirror_id &= ~MIRROR_ID_NEG;
6248 for (i = 0; i < lo->ldo_mirror_count; i++) {
6249 if ((!neg && lo->ldo_mirrors[i].lme_id != mirror_id) ||
6250 (neg && lo->ldo_mirrors[i].lme_id == mirror_id))
6253 lod_foreach_mirror_comp(lod_comp, lo, i) {
6254 if (lod_comp_inited(lod_comp))
6257 info->lti_comp_idx[info->lti_count++] =
6258 lod_comp_index(lo, lod_comp);
6266 * figure out the components should be instantiated for resync.
6268 static int lod_prepare_resync(const struct lu_env *env, struct lod_object *lo,
6269 struct lu_extent *extent)
6271 struct lod_thread_info *info = lod_env_info(env);
6272 struct lod_layout_component *lod_comp;
6273 unsigned int need_sync = 0;
6277 DFID": instantiate all stale components in "DEXT"\n",
6278 PFID(lod_object_fid(lo)), PEXT(extent));
6281 * instantiate all components within this extent, even non-stale
6284 for (i = 0; i < lo->ldo_mirror_count; i++) {
6285 if (!lo->ldo_mirrors[i].lme_stale)
6288 lod_foreach_mirror_comp(lod_comp, lo, i) {
6289 if (!lu_extent_is_overlapped(extent,
6290 &lod_comp->llc_extent))
6295 if (lod_comp_inited(lod_comp))
6298 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
6299 i, lod_comp_index(lo, lod_comp));
6300 info->lti_comp_idx[info->lti_count++] =
6301 lod_comp_index(lo, lod_comp);
6305 return need_sync ? 0 : -EALREADY;
6308 static int lod_declare_update_rdonly(const struct lu_env *env,
6309 struct lod_object *lo, struct md_layout_change *mlc,
6312 struct lod_thread_info *info = lod_env_info(env);
6313 struct lu_attr *layout_attr = &info->lti_layout_attr;
6314 struct lod_layout_component *lod_comp;
6315 struct lu_extent extent = { 0 };
6319 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
6320 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
6321 mlc->mlc_opc == MD_LAYOUT_RESYNC);
6322 LASSERT(lo->ldo_mirror_count > 0);
6324 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
6325 struct layout_intent *layout = mlc->mlc_intent;
6328 extent = layout->li_extent;
6329 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
6330 PFID(lod_object_fid(lo)), PEXT(&extent));
6332 picked = lod_primary_pick(env, lo, &extent);
6336 CDEBUG(D_LAYOUT, DFID": picked mirror id %u as primary\n",
6337 PFID(lod_object_fid(lo)),
6338 lo->ldo_mirrors[picked].lme_id);
6340 if (layout->li_opc == LAYOUT_INTENT_TRUNC) {
6342 * trunc transfers [0, size) in the intent extent, we'd
6343 * stale components overlapping [size, eof).
6345 extent.e_start = extent.e_end;
6346 extent.e_end = OBD_OBJECT_EOF;
6349 /* stale overlapping components from other mirrors */
6350 lod_stale_components(lo, picked, &extent);
6352 /* restore truncate intent extent */
6353 if (layout->li_opc == LAYOUT_INTENT_TRUNC)
6354 extent.e_end = extent.e_start;
6356 /* instantiate components for the picked mirror, start from 0 */
6359 lod_foreach_mirror_comp(lod_comp, lo, picked) {
6360 if (!lu_extent_is_overlapped(&extent,
6361 &lod_comp->llc_extent))
6364 if (lod_comp_inited(lod_comp))
6367 info->lti_comp_idx[info->lti_count++] =
6368 lod_comp_index(lo, lod_comp);
6371 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
6372 } else { /* MD_LAYOUT_RESYNC */
6376 * could contain multiple non-stale mirrors, so we need to
6377 * prep uninited all components assuming any non-stale mirror
6378 * could be picked as the primary mirror.
6380 if (mlc->mlc_mirror_id == 0) {
6382 for (i = 0; i < lo->ldo_mirror_count; i++) {
6383 if (lo->ldo_mirrors[i].lme_stale)
6386 lod_foreach_mirror_comp(lod_comp, lo, i) {
6387 if (!lod_comp_inited(lod_comp))
6391 lod_comp->llc_extent.e_end)
6393 lod_comp->llc_extent.e_end;
6396 rc = lod_prepare_resync(env, lo, &extent);
6400 /* mirror write, try to init its all components */
6401 rc = lod_prepare_resync_mirror(env, lo,
6402 mlc->mlc_mirror_id);
6407 /* change the file state to SYNC_PENDING */
6408 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
6411 /* Reset the layout version once it's becoming too large.
6412 * This way it can make sure that the layout version is
6413 * monotonously increased in this writing era. */
6414 lod_obj_inc_layout_gen(lo);
6415 if (lo->ldo_layout_gen > (LCME_ID_MAX >> 1)) {
6416 __u32 layout_version;
6418 cfs_get_random_bytes(&layout_version, sizeof(layout_version));
6419 lo->ldo_layout_gen = layout_version & 0xffff;
6422 rc = lod_declare_instantiate_components(env, lo, th);
6426 layout_attr->la_valid = LA_LAYOUT_VERSION;
6427 layout_attr->la_layout_version = 0; /* set current version */
6428 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
6429 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
6430 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
6436 lod_striping_free(env, lo);
6440 static int lod_declare_update_write_pending(const struct lu_env *env,
6441 struct lod_object *lo, struct md_layout_change *mlc,
6444 struct lod_thread_info *info = lod_env_info(env);
6445 struct lu_attr *layout_attr = &info->lti_layout_attr;
6446 struct lod_layout_component *lod_comp;
6447 struct lu_extent extent = { 0 };
6453 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
6454 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
6455 mlc->mlc_opc == MD_LAYOUT_RESYNC);
6457 /* look for the primary mirror */
6458 for (i = 0; i < lo->ldo_mirror_count; i++) {
6459 if (lo->ldo_mirrors[i].lme_stale)
6462 LASSERTF(primary < 0, DFID " has multiple primary: %u / %u",
6463 PFID(lod_object_fid(lo)),
6464 lo->ldo_mirrors[i].lme_id,
6465 lo->ldo_mirrors[primary].lme_id);
6470 CERROR(DFID ": doesn't have a primary mirror\n",
6471 PFID(lod_object_fid(lo)));
6472 GOTO(out, rc = -ENODATA);
6475 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
6476 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
6478 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
6480 /* for LAYOUT_WRITE opc, it has to do the following operations:
6481 * 1. stale overlapping componets from stale mirrors;
6482 * 2. instantiate components of the primary mirror;
6483 * 3. transfter layout version to all objects of the primary;
6485 * for LAYOUT_RESYNC opc, it will do:
6486 * 1. instantiate components of all stale mirrors;
6487 * 2. transfer layout version to all objects to close write era. */
6489 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
6490 LASSERT(mlc->mlc_intent != NULL);
6492 extent = mlc->mlc_intent->li_extent;
6494 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
6495 PFID(lod_object_fid(lo)), PEXT(&extent));
6497 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC) {
6499 * trunc transfers [0, size) in the intent extent, we'd
6500 * stale components overlapping [size, eof).
6502 extent.e_start = extent.e_end;
6503 extent.e_end = OBD_OBJECT_EOF;
6505 /* 1. stale overlapping components */
6506 lod_stale_components(lo, primary, &extent);
6508 /* 2. find out the components need instantiating.
6509 * instantiate [0, mlc->mlc_intent->e_end) */
6511 /* restore truncate intent extent */
6512 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC)
6513 extent.e_end = extent.e_start;
6516 lod_foreach_mirror_comp(lod_comp, lo, primary) {
6517 if (!lu_extent_is_overlapped(&extent,
6518 &lod_comp->llc_extent))
6521 if (lod_comp_inited(lod_comp))
6524 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
6525 primary, lod_comp_index(lo, lod_comp));
6526 info->lti_comp_idx[info->lti_count++] =
6527 lod_comp_index(lo, lod_comp);
6529 } else { /* MD_LAYOUT_RESYNC */
6530 if (mlc->mlc_mirror_id == 0) {
6532 lod_foreach_mirror_comp(lod_comp, lo, primary) {
6533 if (!lod_comp_inited(lod_comp))
6536 extent.e_end = lod_comp->llc_extent.e_end;
6539 rc = lod_prepare_resync(env, lo, &extent);
6543 /* mirror write, try to init its all components */
6544 rc = lod_prepare_resync_mirror(env, lo,
6545 mlc->mlc_mirror_id);
6550 /* change the file state to SYNC_PENDING */
6551 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
6554 rc = lod_declare_instantiate_components(env, lo, th);
6558 /* 3. transfer layout version to OST objects.
6559 * transfer new layout version to OST objects so that stale writes
6560 * can be denied. It also ends an era of writing by setting
6561 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
6562 * send write RPC; only resync RPCs could do it. */
6563 layout_attr->la_valid = LA_LAYOUT_VERSION;
6564 layout_attr->la_layout_version = 0; /* set current version */
6565 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
6566 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
6567 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
6571 lod_obj_inc_layout_gen(lo);
6574 lod_striping_free(env, lo);
6578 static int lod_declare_update_sync_pending(const struct lu_env *env,
6579 struct lod_object *lo, struct md_layout_change *mlc,
6582 struct lod_thread_info *info = lod_env_info(env);
6583 unsigned sync_components = 0;
6584 unsigned resync_components = 0;
6589 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
6590 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
6591 mlc->mlc_opc == MD_LAYOUT_WRITE);
6593 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
6594 PFID(lod_object_fid(lo)), mlc->mlc_opc);
6596 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
6597 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
6598 PFID(lod_object_fid(lo)));
6600 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
6601 return lod_declare_update_write_pending(env, lo, mlc, th);
6604 /* MD_LAYOUT_RESYNC_DONE */
6606 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6607 struct lod_layout_component *lod_comp;
6610 lod_comp = &lo->ldo_comp_entries[i];
6612 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
6617 for (j = 0; j < mlc->mlc_resync_count; j++) {
6618 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
6621 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
6622 lod_comp->llc_flags &= ~LCME_FL_STALE;
6623 resync_components++;
6629 for (i = 0; i < mlc->mlc_resync_count; i++) {
6630 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
6633 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
6634 "or already synced\n", PFID(lod_object_fid(lo)),
6635 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
6636 GOTO(out, rc = -EINVAL);
6639 if (!sync_components || (mlc->mlc_resync_count && !resync_components)) {
6640 CDEBUG(D_LAYOUT, DFID": no mirror in sync\n",
6641 PFID(lod_object_fid(lo)));
6643 /* tend to return an error code here to prevent
6644 * the MDT from setting SoM attribute */
6645 GOTO(out, rc = -EINVAL);
6648 CDEBUG(D_LAYOUT, DFID": resynced %u/%zu components\n",
6649 PFID(lod_object_fid(lo)),
6650 resync_components, mlc->mlc_resync_count);
6652 lo->ldo_flr_state = LCM_FL_RDONLY;
6653 lod_obj_inc_layout_gen(lo);
6655 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6656 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
6657 &info->lti_buf, XATTR_NAME_LOV, 0, th);
6662 lod_striping_free(env, lo);
6666 static int lod_declare_layout_change(const struct lu_env *env,
6667 struct dt_object *dt, struct md_layout_change *mlc,
6670 struct lod_thread_info *info = lod_env_info(env);
6671 struct lod_object *lo = lod_dt_obj(dt);
6675 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
6676 dt_object_remote(dt_object_child(dt)))
6679 rc = lod_striping_load(env, lo);
6683 LASSERT(lo->ldo_comp_cnt > 0);
6685 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
6689 switch (lo->ldo_flr_state) {
6691 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
6695 rc = lod_declare_update_rdonly(env, lo, mlc, th);
6697 case LCM_FL_WRITE_PENDING:
6698 rc = lod_declare_update_write_pending(env, lo, mlc, th);
6700 case LCM_FL_SYNC_PENDING:
6701 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
6712 * Instantiate layout component objects which covers the intent write offset.
6714 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
6715 struct md_layout_change *mlc, struct thandle *th)
6717 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
6718 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
6719 struct lod_object *lo = lod_dt_obj(dt);
6722 rc = lod_striped_create(env, dt, attr, NULL, th);
6723 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
6724 layout_attr->la_layout_version |= lo->ldo_layout_gen;
6725 rc = lod_attr_set(env, dt, layout_attr, th);
6731 struct dt_object_operations lod_obj_ops = {
6732 .do_read_lock = lod_read_lock,
6733 .do_write_lock = lod_write_lock,
6734 .do_read_unlock = lod_read_unlock,
6735 .do_write_unlock = lod_write_unlock,
6736 .do_write_locked = lod_write_locked,
6737 .do_attr_get = lod_attr_get,
6738 .do_declare_attr_set = lod_declare_attr_set,
6739 .do_attr_set = lod_attr_set,
6740 .do_xattr_get = lod_xattr_get,
6741 .do_declare_xattr_set = lod_declare_xattr_set,
6742 .do_xattr_set = lod_xattr_set,
6743 .do_declare_xattr_del = lod_declare_xattr_del,
6744 .do_xattr_del = lod_xattr_del,
6745 .do_xattr_list = lod_xattr_list,
6746 .do_ah_init = lod_ah_init,
6747 .do_declare_create = lod_declare_create,
6748 .do_create = lod_create,
6749 .do_declare_destroy = lod_declare_destroy,
6750 .do_destroy = lod_destroy,
6751 .do_index_try = lod_index_try,
6752 .do_declare_ref_add = lod_declare_ref_add,
6753 .do_ref_add = lod_ref_add,
6754 .do_declare_ref_del = lod_declare_ref_del,
6755 .do_ref_del = lod_ref_del,
6756 .do_object_sync = lod_object_sync,
6757 .do_object_lock = lod_object_lock,
6758 .do_object_unlock = lod_object_unlock,
6759 .do_invalidate = lod_invalidate,
6760 .do_declare_layout_change = lod_declare_layout_change,
6761 .do_layout_change = lod_layout_change,
6765 * Implementation of dt_body_operations::dbo_read.
6767 * \see dt_body_operations::dbo_read() in the API description for details.
6769 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
6770 struct lu_buf *buf, loff_t *pos)
6772 struct dt_object *next = dt_object_child(dt);
6774 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
6775 S_ISLNK(dt->do_lu.lo_header->loh_attr));
6776 return next->do_body_ops->dbo_read(env, next, buf, pos);
6780 * Implementation of dt_body_operations::dbo_declare_write.
6782 * \see dt_body_operations::dbo_declare_write() in the API description
6785 static ssize_t lod_declare_write(const struct lu_env *env,
6786 struct dt_object *dt,
6787 const struct lu_buf *buf, loff_t pos,
6790 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
6794 * Implementation of dt_body_operations::dbo_write.
6796 * \see dt_body_operations::dbo_write() in the API description for details.
6798 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
6799 const struct lu_buf *buf, loff_t *pos,
6802 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
6803 S_ISLNK(dt->do_lu.lo_header->loh_attr));
6804 return lod_sub_write(env, dt_object_child(dt), buf, pos, th);
6807 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
6808 __u64 start, __u64 end, struct thandle *th)
6810 if (dt_object_remote(dt))
6813 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
6816 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
6817 __u64 start, __u64 end, struct thandle *th)
6819 if (dt_object_remote(dt))
6822 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
6823 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
6827 * different type of files use the same body_ops because object may be created
6828 * in OUT, where there is no chance to set correct body_ops for each type, so
6829 * body_ops themselves will check file type inside, see lod_read/write/punch for
6832 const struct dt_body_operations lod_body_ops = {
6833 .dbo_read = lod_read,
6834 .dbo_declare_write = lod_declare_write,
6835 .dbo_write = lod_write,
6836 .dbo_declare_punch = lod_declare_punch,
6837 .dbo_punch = lod_punch,
6841 * Implementation of lu_object_operations::loo_object_init.
6843 * The function determines the type and the index of the target device using
6844 * sequence of the object's FID. Then passes control down to the
6845 * corresponding device:
6846 * OSD for the local objects, OSP for remote
6848 * \see lu_object_operations::loo_object_init() in the API description
6851 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
6852 const struct lu_object_conf *conf)
6854 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
6855 struct lu_device *cdev = NULL;
6856 struct lu_object *cobj;
6857 struct lod_tgt_descs *ltd = NULL;
6858 struct lod_tgt_desc *tgt;
6860 int type = LU_SEQ_RANGE_ANY;
6864 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
6866 /* Note: Sometimes, it will Return EAGAIN here, see
6867 * ptrlpc_import_delay_req(), which might confuse
6868 * lu_object_find_at() and make it wait there incorrectly.
6869 * so we convert it to EIO here.*/
6876 if (type == LU_SEQ_RANGE_MDT &&
6877 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
6878 cdev = &lod->lod_child->dd_lu_dev;
6879 } else if (type == LU_SEQ_RANGE_MDT) {
6880 ltd = &lod->lod_mdt_descs;
6882 } else if (type == LU_SEQ_RANGE_OST) {
6883 ltd = &lod->lod_ost_descs;
6890 if (ltd->ltd_tgts_size > idx &&
6891 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
6892 tgt = LTD_TGT(ltd, idx);
6894 LASSERT(tgt != NULL);
6895 LASSERT(tgt->ltd_tgt != NULL);
6897 cdev = &(tgt->ltd_tgt->dd_lu_dev);
6899 lod_putref(lod, ltd);
6902 if (unlikely(cdev == NULL))
6905 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
6906 if (unlikely(cobj == NULL))
6909 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
6911 lu_object_add(lo, cobj);
6918 * Alloc cached foreign LOV
6920 * \param[in] lo object
6921 * \param[in] size size of foreign LOV
6923 * \retval 0 on success
6924 * \retval negative if failed
6926 int lod_alloc_foreign_lov(struct lod_object *lo, size_t size)
6928 OBD_ALLOC_LARGE(lo->ldo_foreign_lov, size);
6929 if (lo->ldo_foreign_lov == NULL)
6931 lo->ldo_foreign_lov_size = size;
6932 lo->ldo_is_foreign = 1;
6938 * Free cached foreign LOV
6940 * \param[in] lo object
6942 void lod_free_foreign_lov(struct lod_object *lo)
6944 if (lo->ldo_foreign_lov != NULL)
6945 OBD_FREE_LARGE(lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
6946 lo->ldo_foreign_lov = NULL;
6947 lo->ldo_foreign_lov_size = 0;
6948 lo->ldo_is_foreign = 0;
6953 * Free cached foreign LMV
6955 * \param[in] lo object
6957 void lod_free_foreign_lmv(struct lod_object *lo)
6959 if (lo->ldo_foreign_lmv != NULL)
6960 OBD_FREE_LARGE(lo->ldo_foreign_lmv, lo->ldo_foreign_lmv_size);
6961 lo->ldo_foreign_lmv = NULL;
6962 lo->ldo_foreign_lmv_size = 0;
6963 lo->ldo_dir_is_foreign = 0;
6968 * Release resources associated with striping.
6970 * If the object is striped (regular or directory), then release
6971 * the stripe objects references and free the ldo_stripe array.
6973 * \param[in] env execution environment
6974 * \param[in] lo object
6976 void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo)
6978 struct lod_layout_component *lod_comp;
6981 if (unlikely(lo->ldo_is_foreign)) {
6982 lod_free_foreign_lov(lo);
6983 lo->ldo_comp_cached = 0;
6984 } else if (unlikely(lo->ldo_dir_is_foreign)) {
6985 lod_free_foreign_lmv(lo);
6986 lo->ldo_dir_stripe_loaded = 0;
6987 } else if (lo->ldo_stripe != NULL) {
6988 LASSERT(lo->ldo_comp_entries == NULL);
6989 LASSERT(lo->ldo_dir_stripes_allocated > 0);
6991 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6992 if (lo->ldo_stripe[i])
6993 dt_object_put(env, lo->ldo_stripe[i]);
6996 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
6997 OBD_FREE(lo->ldo_stripe, j);
6998 lo->ldo_stripe = NULL;
6999 lo->ldo_dir_stripes_allocated = 0;
7000 lo->ldo_dir_stripe_loaded = 0;
7001 lo->ldo_dir_stripe_count = 0;
7002 } else if (lo->ldo_comp_entries != NULL) {
7003 for (i = 0; i < lo->ldo_comp_cnt; i++) {
7004 /* free lod_layout_component::llc_stripe array */
7005 lod_comp = &lo->ldo_comp_entries[i];
7007 if (lod_comp->llc_stripe == NULL)
7009 LASSERT(lod_comp->llc_stripes_allocated != 0);
7010 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
7011 if (lod_comp->llc_stripe[j] != NULL)
7013 &lod_comp->llc_stripe[j]->do_lu);
7015 OBD_FREE(lod_comp->llc_stripe,
7016 sizeof(struct dt_object *) *
7017 lod_comp->llc_stripes_allocated);
7018 lod_comp->llc_stripe = NULL;
7019 OBD_FREE(lod_comp->llc_ost_indices,
7021 lod_comp->llc_stripes_allocated);
7022 lod_comp->llc_ost_indices = NULL;
7023 lod_comp->llc_stripes_allocated = 0;
7025 lod_free_comp_entries(lo);
7026 lo->ldo_comp_cached = 0;
7030 void lod_striping_free(const struct lu_env *env, struct lod_object *lo)
7032 mutex_lock(&lo->ldo_layout_mutex);
7033 lod_striping_free_nolock(env, lo);
7034 mutex_unlock(&lo->ldo_layout_mutex);
7038 * Implementation of lu_object_operations::loo_object_free.
7040 * \see lu_object_operations::loo_object_free() in the API description
7043 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
7045 struct lod_object *lo = lu2lod_obj(o);
7047 /* release all underlying object pinned */
7048 lod_striping_free(env, lo);
7050 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
7054 * Implementation of lu_object_operations::loo_object_release.
7056 * \see lu_object_operations::loo_object_release() in the API description
7059 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
7061 /* XXX: shouldn't we release everything here in case if object
7062 * creation failed before? */
7066 * Implementation of lu_object_operations::loo_object_print.
7068 * \see lu_object_operations::loo_object_print() in the API description
7071 static int lod_object_print(const struct lu_env *env, void *cookie,
7072 lu_printer_t p, const struct lu_object *l)
7074 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
7076 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
7079 struct lu_object_operations lod_lu_obj_ops = {
7080 .loo_object_init = lod_object_init,
7081 .loo_object_free = lod_object_free,
7082 .loo_object_release = lod_object_release,
7083 .loo_object_print = lod_object_print,