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,
101 struct thandle *th, int ign)
103 return lod_sub_insert(env, dt_object_child(dt), rec, key, th, ign);
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 it is not a striped directory, then load nothing. */
780 if (magic != LMV_MAGIC_V1)
783 /* If it is in migration (or failure), then load nothing. */
784 if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
787 stripes = le32_to_cpu(lmv1->lmv_stripe_count);
791 rc = lmv_mds_md_size(stripes, magic);
795 if (buf->lb_len < lmv1_size) {
804 lu_buf_alloc(buf, lmv1_size);
809 memcpy(buf->lb_buf, tbuf.lb_buf, tbuf.lb_len);
812 if (unlikely(!dt_try_as_dir(env, obj)))
815 memset(&lmv1->lmv_stripe_fids[0], 0, stripes * sizeof(struct lu_fid));
816 iops = &obj->do_index_ops->dio_it;
817 it = iops->init(env, obj, LUDA_64BITHASH);
821 rc = iops->load(env, it, 0);
823 rc = iops->next(env, it);
828 char name[FID_LEN + 2] = "";
833 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
839 fid_le_to_cpu(&fid, &ent->lde_fid);
840 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
841 if (ent->lde_name[0] == '.') {
842 if (ent->lde_namelen == 1)
845 if (ent->lde_namelen == 2 && ent->lde_name[1] == '.')
849 len = snprintf(name, sizeof(name),
850 DFID":", PFID(&ent->lde_fid));
851 /* The ent->lde_name is composed of ${FID}:${index} */
852 if (ent->lde_namelen < len + 1 ||
853 memcmp(ent->lde_name, name, len) != 0) {
854 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
855 "%s: invalid shard name %.*s with the FID "DFID
856 " for the striped directory "DFID", %s\n",
857 lod2obd(lod)->obd_name, ent->lde_namelen,
858 ent->lde_name, PFID(&fid),
859 PFID(lu_object_fid(&obj->do_lu)),
860 lod->lod_lmv_failout ? "failout" : "skip");
862 if (lod->lod_lmv_failout)
870 if (ent->lde_name[len] < '0' ||
871 ent->lde_name[len] > '9') {
872 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
873 "%s: invalid shard name %.*s with the "
874 "FID "DFID" for the striped directory "
876 lod2obd(lod)->obd_name, ent->lde_namelen,
877 ent->lde_name, PFID(&fid),
878 PFID(lu_object_fid(&obj->do_lu)),
879 lod->lod_lmv_failout ?
882 if (lod->lod_lmv_failout)
888 index = index * 10 + ent->lde_name[len++] - '0';
889 } while (len < ent->lde_namelen);
891 if (len == ent->lde_namelen) {
892 /* Out of LMV EA range. */
893 if (index >= stripes) {
894 CERROR("%s: the shard %.*s for the striped "
895 "directory "DFID" is out of the known "
896 "LMV EA range [0 - %u], failout\n",
897 lod2obd(lod)->obd_name, ent->lde_namelen,
899 PFID(lu_object_fid(&obj->do_lu)),
905 /* The slot has been occupied. */
906 if (!fid_is_zero(&lmv1->lmv_stripe_fids[index])) {
910 &lmv1->lmv_stripe_fids[index]);
911 CERROR("%s: both the shard "DFID" and "DFID
912 " for the striped directory "DFID
913 " claim the same LMV EA slot at the "
914 "index %d, failout\n",
915 lod2obd(lod)->obd_name,
916 PFID(&fid0), PFID(&fid),
917 PFID(lu_object_fid(&obj->do_lu)), index);
922 /* stored as LE mode */
923 lmv1->lmv_stripe_fids[index] = ent->lde_fid;
926 rc = iops->next(env, it);
933 RETURN(rc > 0 ? lmv_mds_md_size(stripes, magic) : rc);
937 * Implementation of dt_object_operations::do_index_try.
939 * \see dt_object_operations::do_index_try() in the API description for details.
941 static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
942 const struct dt_index_features *feat)
944 struct lod_object *lo = lod_dt_obj(dt);
945 struct dt_object *next = dt_object_child(dt);
949 LASSERT(next->do_ops);
950 LASSERT(next->do_ops->do_index_try);
952 rc = lod_load_striping_locked(env, lo);
956 rc = next->do_ops->do_index_try(env, next, feat);
960 if (lo->ldo_dir_stripe_count > 0) {
963 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
964 if (dt_object_exists(lo->ldo_stripe[i]) == 0)
966 rc = lo->ldo_stripe[i]->do_ops->do_index_try(env,
967 lo->ldo_stripe[i], feat);
971 dt->do_index_ops = &lod_striped_index_ops;
973 dt->do_index_ops = &lod_index_ops;
980 * Implementation of dt_object_operations::do_read_lock.
982 * \see dt_object_operations::do_read_lock() in the API description for details.
984 static void lod_read_lock(const struct lu_env *env, struct dt_object *dt,
987 dt_read_lock(env, dt_object_child(dt), role);
991 * Implementation of dt_object_operations::do_write_lock.
993 * \see dt_object_operations::do_write_lock() in the API description for
996 static void lod_write_lock(const struct lu_env *env, struct dt_object *dt,
999 dt_write_lock(env, dt_object_child(dt), role);
1003 * Implementation of dt_object_operations::do_read_unlock.
1005 * \see dt_object_operations::do_read_unlock() in the API description for
1008 static void lod_read_unlock(const struct lu_env *env, struct dt_object *dt)
1010 dt_read_unlock(env, dt_object_child(dt));
1014 * Implementation of dt_object_operations::do_write_unlock.
1016 * \see dt_object_operations::do_write_unlock() in the API description for
1019 static void lod_write_unlock(const struct lu_env *env, struct dt_object *dt)
1021 dt_write_unlock(env, dt_object_child(dt));
1025 * Implementation of dt_object_operations::do_write_locked.
1027 * \see dt_object_operations::do_write_locked() in the API description for
1030 static int lod_write_locked(const struct lu_env *env, struct dt_object *dt)
1032 return dt_write_locked(env, dt_object_child(dt));
1036 * Implementation of dt_object_operations::do_attr_get.
1038 * \see dt_object_operations::do_attr_get() in the API description for details.
1040 static int lod_attr_get(const struct lu_env *env,
1041 struct dt_object *dt,
1042 struct lu_attr *attr)
1044 /* Note: for striped directory, client will merge attributes
1045 * from all of the sub-stripes see lmv_merge_attr(), and there
1046 * no MDD logic depend on directory nlink/size/time, so we can
1047 * always use master inode nlink and size for now. */
1048 return dt_attr_get(env, dt_object_child(dt), attr);
1051 static inline void lod_adjust_stripe_info(struct lod_layout_component *comp,
1052 struct lov_desc *desc)
1054 if (comp->llc_pattern != LOV_PATTERN_MDT) {
1055 if (!comp->llc_stripe_count)
1056 comp->llc_stripe_count =
1057 desc->ld_default_stripe_count;
1059 if (comp->llc_stripe_size <= 0)
1060 comp->llc_stripe_size = desc->ld_default_stripe_size;
1063 int lod_obj_for_each_stripe(const struct lu_env *env, struct lod_object *lo,
1065 struct lod_obj_stripe_cb_data *data)
1067 struct lod_layout_component *lod_comp;
1071 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1072 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1073 lod_comp = &lo->ldo_comp_entries[i];
1075 if (lod_comp->llc_stripe == NULL)
1078 /* has stripe but not inited yet, this component has been
1079 * declared to be created, but hasn't created yet.
1081 if (!lod_comp_inited(lod_comp))
1084 if (data->locd_comp_skip_cb &&
1085 data->locd_comp_skip_cb(env, lo, i, data))
1088 LASSERT(lod_comp->llc_stripe_count > 0);
1089 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
1090 struct dt_object *dt = lod_comp->llc_stripe[j];
1094 rc = data->locd_stripe_cb(env, lo, dt, th, i, j, data);
1102 static bool lod_obj_attr_set_comp_skip_cb(const struct lu_env *env,
1103 struct lod_object *lo, int comp_idx,
1104 struct lod_obj_stripe_cb_data *data)
1106 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[comp_idx];
1107 bool skipped = false;
1109 if (!(data->locd_attr->la_valid & LA_LAYOUT_VERSION))
1112 switch (lo->ldo_flr_state) {
1113 case LCM_FL_WRITE_PENDING: {
1116 /* skip stale components */
1117 if (lod_comp->llc_flags & LCME_FL_STALE) {
1122 /* skip valid and overlapping components, therefore any
1123 * attempts to write overlapped components will never succeed
1124 * because client will get EINPROGRESS. */
1125 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1129 if (lo->ldo_comp_entries[i].llc_flags & LCME_FL_STALE)
1132 if (lu_extent_is_overlapped(&lod_comp->llc_extent,
1133 &lo->ldo_comp_entries[i].llc_extent)) {
1141 LASSERTF(0, "impossible: %d\n", lo->ldo_flr_state);
1142 case LCM_FL_SYNC_PENDING:
1146 CDEBUG(D_LAYOUT, DFID": %s to set component %x to version: %u\n",
1147 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
1148 skipped ? "skipped" : "chose", lod_comp->llc_id,
1149 data->locd_attr->la_layout_version);
1155 lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
1156 struct dt_object *dt, struct thandle *th,
1157 int comp_idx, int stripe_idx,
1158 struct lod_obj_stripe_cb_data *data)
1160 if (data->locd_declare)
1161 return lod_sub_declare_attr_set(env, dt, data->locd_attr, th);
1163 if (data->locd_attr->la_valid & LA_LAYOUT_VERSION) {
1164 CDEBUG(D_LAYOUT, DFID": set layout version: %u, comp_idx: %d\n",
1165 PFID(lu_object_fid(&dt->do_lu)),
1166 data->locd_attr->la_layout_version, comp_idx);
1169 return lod_sub_attr_set(env, dt, data->locd_attr, th);
1173 * Implementation of dt_object_operations::do_declare_attr_set.
1175 * If the object is striped, then apply the changes to all the stripes.
1177 * \see dt_object_operations::do_declare_attr_set() in the API description
1180 static int lod_declare_attr_set(const struct lu_env *env,
1181 struct dt_object *dt,
1182 const struct lu_attr *attr,
1185 struct dt_object *next = dt_object_child(dt);
1186 struct lod_object *lo = lod_dt_obj(dt);
1191 * declare setattr on the local object
1193 rc = lod_sub_declare_attr_set(env, next, attr, th);
1197 /* osp_declare_attr_set() ignores all attributes other than
1198 * UID, GID, PROJID, and size, and osp_attr_set() ignores all
1199 * but UID, GID and PROJID. Declaration of size attr setting
1200 * happens through lod_declare_init_size(), and not through
1201 * this function. Therefore we need not load striping unless
1202 * ownership is changing. This should save memory and (we hope)
1203 * speed up rename().
1205 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1206 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1209 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1212 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID | LA_MODE |
1213 LA_ATIME | LA_MTIME | LA_CTIME |
1218 * load striping information, notice we don't do this when object
1219 * is being initialized as we don't need this information till
1220 * few specific cases like destroy, chown
1222 rc = lod_load_striping(env, lo);
1226 if (!lod_obj_is_striped(dt))
1230 * if object is striped declare changes on the stripes
1232 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1233 LASSERT(lo->ldo_stripe);
1234 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1235 if (lo->ldo_stripe[i] == NULL)
1237 rc = lod_sub_declare_attr_set(env, lo->ldo_stripe[i],
1243 struct lod_obj_stripe_cb_data data = { { 0 } };
1245 data.locd_attr = attr;
1246 data.locd_declare = true;
1247 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1248 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1254 if (!dt_object_exists(next) || dt_object_remote(next) ||
1255 !S_ISREG(attr->la_mode))
1258 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1259 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
1263 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) ||
1264 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1265 struct lod_thread_info *info = lod_env_info(env);
1266 struct lu_buf *buf = &info->lti_buf;
1268 buf->lb_buf = info->lti_ea_store;
1269 buf->lb_len = info->lti_ea_store_size;
1270 rc = lod_sub_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1271 LU_XATTR_REPLACE, th);
1278 * Implementation of dt_object_operations::do_attr_set.
1280 * If the object is striped, then apply the changes to all or subset of
1281 * the stripes depending on the object type and specific attributes.
1283 * \see dt_object_operations::do_attr_set() in the API description for details.
1285 static int lod_attr_set(const struct lu_env *env,
1286 struct dt_object *dt,
1287 const struct lu_attr *attr,
1290 struct dt_object *next = dt_object_child(dt);
1291 struct lod_object *lo = lod_dt_obj(dt);
1296 * apply changes to the local object
1298 rc = lod_sub_attr_set(env, next, attr, th);
1302 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1303 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1306 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1309 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE | LA_PROJID |
1310 LA_ATIME | LA_MTIME | LA_CTIME |
1315 /* FIXME: a tricky case in the code path of mdd_layout_change():
1316 * the in-memory striping information has been freed in lod_xattr_set()
1317 * due to layout change. It has to load stripe here again. It only
1318 * changes flags of layout so declare_attr_set() is still accurate */
1319 rc = lod_load_striping_locked(env, lo);
1323 if (!lod_obj_is_striped(dt))
1327 * if object is striped, apply changes to all the stripes
1329 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1330 LASSERT(lo->ldo_stripe);
1331 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1332 if (unlikely(lo->ldo_stripe[i] == NULL))
1335 if ((dt_object_exists(lo->ldo_stripe[i]) == 0))
1338 rc = lod_sub_attr_set(env, lo->ldo_stripe[i], attr, th);
1343 struct lod_obj_stripe_cb_data data = { { 0 } };
1345 data.locd_attr = attr;
1346 data.locd_declare = false;
1347 data.locd_comp_skip_cb = lod_obj_attr_set_comp_skip_cb;
1348 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1349 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1355 if (!dt_object_exists(next) || dt_object_remote(next) ||
1356 !S_ISREG(attr->la_mode))
1359 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1360 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
1364 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE)) {
1365 struct lod_thread_info *info = lod_env_info(env);
1366 struct lu_buf *buf = &info->lti_buf;
1367 struct ost_id *oi = &info->lti_ostid;
1368 struct lu_fid *fid = &info->lti_fid;
1369 struct lov_mds_md_v1 *lmm;
1370 struct lov_ost_data_v1 *objs;
1373 rc = lod_get_lov_ea(env, lo);
1377 buf->lb_buf = info->lti_ea_store;
1378 buf->lb_len = info->lti_ea_store_size;
1379 lmm = info->lti_ea_store;
1380 magic = le32_to_cpu(lmm->lmm_magic);
1381 if (magic == LOV_MAGIC_COMP_V1) {
1382 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1383 struct lov_comp_md_entry_v1 *lcme =
1384 &lcm->lcm_entries[0];
1386 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
1387 magic = le32_to_cpu(lmm->lmm_magic);
1390 if (magic == LOV_MAGIC_V1)
1391 objs = &(lmm->lmm_objects[0]);
1393 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1394 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1395 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1397 fid_to_ostid(fid, oi);
1398 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1400 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1401 LU_XATTR_REPLACE, th);
1402 } else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1403 struct lod_thread_info *info = lod_env_info(env);
1404 struct lu_buf *buf = &info->lti_buf;
1405 struct lov_comp_md_v1 *lcm;
1406 struct lov_comp_md_entry_v1 *lcme;
1408 rc = lod_get_lov_ea(env, lo);
1412 buf->lb_buf = info->lti_ea_store;
1413 buf->lb_len = info->lti_ea_store_size;
1415 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
1418 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1419 lcme = &lcm->lcm_entries[0];
1420 le64_add_cpu(&lcme->lcme_extent.e_start, 1);
1421 le64_add_cpu(&lcme->lcme_extent.e_end, -1);
1423 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1424 LU_XATTR_REPLACE, th);
1431 * Implementation of dt_object_operations::do_xattr_get.
1433 * If LOV EA is requested from the root object and it's not
1434 * found, then return default striping for the filesystem.
1436 * \see dt_object_operations::do_xattr_get() in the API description for details.
1438 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1439 struct lu_buf *buf, const char *name)
1441 struct lod_thread_info *info = lod_env_info(env);
1442 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1447 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1448 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1449 struct lmv_mds_md_v1 *lmv1;
1452 if (rc > (typeof(rc))sizeof(*lmv1))
1455 if (rc < (typeof(rc))sizeof(*lmv1))
1456 RETURN(rc = rc > 0 ? -EINVAL : rc);
1458 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1459 CLASSERT(sizeof(*lmv1) <= sizeof(info->lti_key));
1461 info->lti_buf.lb_buf = info->lti_key;
1462 info->lti_buf.lb_len = sizeof(*lmv1);
1463 rc = dt_xattr_get(env, dt_object_child(dt),
1464 &info->lti_buf, name);
1465 if (unlikely(rc != sizeof(*lmv1)))
1466 RETURN(rc = rc > 0 ? -EINVAL : rc);
1468 lmv1 = info->lti_buf.lb_buf;
1469 /* The on-disk LMV EA only contains header, but the
1470 * returned LMV EA size should contain the space for
1471 * the FIDs of all shards of the striped directory. */
1472 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1473 rc = lmv_mds_md_size(
1474 le32_to_cpu(lmv1->lmv_stripe_count),
1477 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1481 RETURN(rc = rc1 != 0 ? rc1 : rc);
1484 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1488 * XXX: Only used by lfsck
1490 * lod returns default striping on the real root of the device
1491 * this is like the root stores default striping for the whole
1492 * filesystem. historically we've been using a different approach
1493 * and store it in the config.
1495 dt_root_get(env, dev->lod_child, &info->lti_fid);
1496 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1498 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1499 struct lov_user_md *lum = buf->lb_buf;
1500 struct lov_desc *desc = &dev->lod_desc;
1502 if (buf->lb_buf == NULL) {
1504 } else if (buf->lb_len >= sizeof(*lum)) {
1505 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1506 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1507 lmm_oi_set_id(&lum->lmm_oi, 0);
1508 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1509 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1510 lum->lmm_stripe_size = cpu_to_le32(
1511 desc->ld_default_stripe_size);
1512 lum->lmm_stripe_count = cpu_to_le16(
1513 desc->ld_default_stripe_count);
1514 lum->lmm_stripe_offset = cpu_to_le16(
1515 desc->ld_default_stripe_offset);
1528 * Checks that the magic of the stripe is sane.
1530 * \param[in] lod lod device
1531 * \param[in] lum a buffer storing LMV EA to verify
1533 * \retval 0 if the EA is sane
1534 * \retval negative otherwise
1536 static int lod_verify_md_striping(struct lod_device *lod,
1537 const struct lmv_user_md_v1 *lum)
1539 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1540 CERROR("%s: invalid lmv_user_md: magic = %x, "
1541 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1542 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1543 (int)le32_to_cpu(lum->lum_stripe_offset),
1544 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1552 * Initialize LMV EA for a slave.
1554 * Initialize slave's LMV EA from the master's LMV EA.
1556 * \param[in] master_lmv a buffer containing master's EA
1557 * \param[out] slave_lmv a buffer where slave's EA will be stored
1560 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1561 const struct lmv_mds_md_v1 *master_lmv)
1563 *slave_lmv = *master_lmv;
1564 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1570 * Generate LMV EA from the object passed as \a dt. The object must have
1571 * the stripes created and initialized.
1573 * \param[in] env execution environment
1574 * \param[in] dt object
1575 * \param[out] lmv_buf buffer storing generated LMV EA
1577 * \retval 0 on success
1578 * \retval negative if failed
1580 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1581 struct lu_buf *lmv_buf)
1583 struct lod_thread_info *info = lod_env_info(env);
1584 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1585 struct lod_object *lo = lod_dt_obj(dt);
1586 struct lmv_mds_md_v1 *lmm1;
1588 int type = LU_SEQ_RANGE_ANY;
1593 LASSERT(lo->ldo_dir_striped != 0);
1594 LASSERT(lo->ldo_dir_stripe_count > 0);
1595 stripe_count = lo->ldo_dir_stripe_count;
1596 /* Only store the LMV EA heahder on the disk. */
1597 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1598 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1602 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1605 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1606 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1607 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1608 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1609 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1614 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1615 lmv_buf->lb_buf = info->lti_ea_store;
1616 lmv_buf->lb_len = sizeof(*lmm1);
1622 * Create in-core represenation for a striped directory.
1624 * Parse the buffer containing LMV EA and instantiate LU objects
1625 * representing the stripe objects. The pointers to the objects are
1626 * stored in ldo_stripe field of \a lo. This function is used when
1627 * we need to access an already created object (i.e. load from a disk).
1629 * \param[in] env execution environment
1630 * \param[in] lo lod object
1631 * \param[in] buf buffer containing LMV EA
1633 * \retval 0 on success
1634 * \retval negative if failed
1636 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1637 const struct lu_buf *buf)
1639 struct lod_thread_info *info = lod_env_info(env);
1640 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1641 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1642 struct dt_object **stripe;
1643 union lmv_mds_md *lmm = buf->lb_buf;
1644 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1645 struct lu_fid *fid = &info->lti_fid;
1650 if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
1653 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1654 lo->ldo_dir_slave_stripe = 1;
1658 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1661 if (le32_to_cpu(lmv1->lmv_stripe_count) < 1)
1664 LASSERT(lo->ldo_stripe == NULL);
1665 OBD_ALLOC(stripe, sizeof(stripe[0]) *
1666 (le32_to_cpu(lmv1->lmv_stripe_count)));
1670 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1671 struct dt_device *tgt_dt;
1672 struct dt_object *dto;
1673 int type = LU_SEQ_RANGE_ANY;
1676 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1677 if (!fid_is_sane(fid))
1678 GOTO(out, rc = -ESTALE);
1680 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1684 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1685 tgt_dt = lod->lod_child;
1687 struct lod_tgt_desc *tgt;
1689 tgt = LTD_TGT(ltd, idx);
1691 GOTO(out, rc = -ESTALE);
1692 tgt_dt = tgt->ltd_tgt;
1695 dto = dt_locate_at(env, tgt_dt, fid,
1696 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1699 GOTO(out, rc = PTR_ERR(dto));
1704 lo->ldo_stripe = stripe;
1705 lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1706 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1708 lod_object_free_striping(env, lo);
1714 * Declare create a striped directory.
1716 * Declare creating a striped directory with a given stripe pattern on the
1717 * specified MDTs. A striped directory is represented as a regular directory
1718 * - an index listing all the stripes. The stripes point back to the master
1719 * object with ".." and LinkEA. The master object gets LMV EA which
1720 * identifies it as a striped directory. The function allocates FIDs
1723 * \param[in] env execution environment
1724 * \param[in] dt object
1725 * \param[in] attr attributes to initialize the objects with
1726 * \param[in] dof type of objects to be created
1727 * \param[in] th transaction handle
1729 * \retval 0 on success
1730 * \retval negative if failed
1732 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1733 struct dt_object *dt,
1734 struct lu_attr *attr,
1735 struct dt_object_format *dof,
1738 struct lod_thread_info *info = lod_env_info(env);
1739 struct lu_buf lmv_buf;
1740 struct lu_buf slave_lmv_buf;
1741 struct lmv_mds_md_v1 *lmm;
1742 struct lmv_mds_md_v1 *slave_lmm = NULL;
1743 struct dt_insert_rec *rec = &info->lti_dt_rec;
1744 struct lod_object *lo = lod_dt_obj(dt);
1749 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1752 lmm = lmv_buf.lb_buf;
1754 OBD_ALLOC_PTR(slave_lmm);
1755 if (slave_lmm == NULL)
1756 GOTO(out, rc = -ENOMEM);
1758 lod_prep_slave_lmv_md(slave_lmm, lmm);
1759 slave_lmv_buf.lb_buf = slave_lmm;
1760 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1762 if (!dt_try_as_dir(env, dt_object_child(dt)))
1763 GOTO(out, rc = -EINVAL);
1765 rec->rec_type = S_IFDIR;
1766 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1767 struct dt_object *dto = lo->ldo_stripe[i];
1768 char *stripe_name = info->lti_key;
1769 struct lu_name *sname;
1770 struct linkea_data ldata = { NULL };
1771 struct lu_buf linkea_buf;
1773 rc = lod_sub_declare_create(env, dto, attr, NULL, dof, th);
1777 if (!dt_try_as_dir(env, dto))
1778 GOTO(out, rc = -EINVAL);
1780 rc = lod_sub_declare_ref_add(env, dto, th);
1784 rec->rec_fid = lu_object_fid(&dto->do_lu);
1785 rc = lod_sub_declare_insert(env, dto,
1786 (const struct dt_rec *)rec,
1787 (const struct dt_key *)dot, th);
1791 /* master stripe FID will be put to .. */
1792 rec->rec_fid = lu_object_fid(&dt->do_lu);
1793 rc = lod_sub_declare_insert(env, dto,
1794 (const struct dt_rec *)rec,
1795 (const struct dt_key *)dotdot, th);
1799 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1800 cfs_fail_val != i) {
1801 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1803 slave_lmm->lmv_master_mdt_index =
1806 slave_lmm->lmv_master_mdt_index =
1808 rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
1809 XATTR_NAME_LMV, 0, th);
1814 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1816 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1817 PFID(lu_object_fid(&dto->do_lu)), i + 1);
1819 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1820 PFID(lu_object_fid(&dto->do_lu)), i);
1822 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
1823 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1824 sname, lu_object_fid(&dt->do_lu));
1828 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1829 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1830 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
1831 XATTR_NAME_LINK, 0, th);
1835 rec->rec_fid = lu_object_fid(&dto->do_lu);
1836 rc = lod_sub_declare_insert(env, dt_object_child(dt),
1837 (const struct dt_rec *)rec,
1838 (const struct dt_key *)stripe_name,
1843 rc = lod_sub_declare_ref_add(env, dt_object_child(dt), th);
1848 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
1849 &lmv_buf, XATTR_NAME_LMV, 0, th);
1853 if (slave_lmm != NULL)
1854 OBD_FREE_PTR(slave_lmm);
1859 static int lod_prep_md_striped_create(const struct lu_env *env,
1860 struct dt_object *dt,
1861 struct lu_attr *attr,
1862 const struct lmv_user_md_v1 *lum,
1863 struct dt_object_format *dof,
1866 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1867 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1868 struct lod_object *lo = lod_dt_obj(dt);
1869 struct dt_object **stripe;
1876 bool is_specific = false;
1879 /* The lum has been verifed in lod_verify_md_striping */
1880 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1881 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
1882 LASSERT(le32_to_cpu(lum->lum_stripe_count) > 0);
1884 stripe_count = le32_to_cpu(lum->lum_stripe_count);
1886 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
1887 if (idx_array == NULL)
1890 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_count);
1892 GOTO(out_free, rc = -ENOMEM);
1894 /* Start index must be the master MDT */
1895 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1896 idx_array[0] = master_index;
1897 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
1899 for (i = 1; i < stripe_count; i++)
1900 idx_array[i] = le32_to_cpu(lum->lum_objects[i].lum_mds);
1903 for (i = 0; i < stripe_count; i++) {
1904 struct lod_tgt_desc *tgt = NULL;
1905 struct dt_object *dto;
1906 struct lu_fid fid = { 0 };
1908 struct lu_object_conf conf = { 0 };
1909 struct dt_device *tgt_dt = NULL;
1911 /* Try to find next avaible target */
1913 for (j = 0; j < lod->lod_remote_mdt_count;
1914 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
1915 bool already_allocated = false;
1918 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
1919 idx, lod->lod_remote_mdt_count + 1, i);
1921 if (likely(!is_specific &&
1922 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
1923 /* check whether the idx already exists
1924 * in current allocated array */
1925 for (k = 0; k < i; k++) {
1926 if (idx_array[k] == idx) {
1927 already_allocated = true;
1932 if (already_allocated)
1936 /* Sigh, this index is not in the bitmap, let's check
1937 * next available target */
1938 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) &&
1939 idx != master_index)
1942 if (idx == master_index) {
1943 /* Allocate the FID locally */
1944 rc = obd_fid_alloc(env, lod->lod_child_exp,
1948 tgt_dt = lod->lod_child;
1952 /* check the status of the OSP */
1953 tgt = LTD_TGT(ltd, idx);
1957 tgt_dt = tgt->ltd_tgt;
1958 rc = dt_statfs(env, tgt_dt, NULL);
1960 /* this OSP doesn't feel well */
1965 rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
1974 /* Can not allocate more stripes */
1975 if (j == lod->lod_remote_mdt_count) {
1976 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
1977 lod2obd(lod)->obd_name, stripe_count, i);
1981 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
1982 idx, i, PFID(&fid));
1984 /* Set the start index for next stripe allocation */
1985 if (!is_specific && i < stripe_count - 1) {
1987 * for large dir test, put all other slaves on one
1988 * remote MDT, otherwise we may save too many local
1989 * slave locks which will exceed RS_MAX_LOCKS.
1991 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)))
1993 idx_array[i + 1] = (idx + 1) %
1994 (lod->lod_remote_mdt_count + 1);
1996 /* tgt_dt and fid must be ready after search avaible OSP
1997 * in the above loop */
1998 LASSERT(tgt_dt != NULL);
1999 LASSERT(fid_is_sane(&fid));
2000 conf.loc_flags = LOC_F_NEW;
2001 dto = dt_locate_at(env, tgt_dt, &fid,
2002 dt->do_lu.lo_dev->ld_site->ls_top_dev,
2005 GOTO(out_put, rc = PTR_ERR(dto));
2009 lo->ldo_dir_stripe_loaded = 1;
2010 lo->ldo_dir_striped = 1;
2011 lo->ldo_stripe = stripe;
2012 lo->ldo_dir_stripe_count = i;
2013 lo->ldo_dir_stripes_allocated = stripe_count;
2015 if (lo->ldo_dir_stripe_count == 0)
2016 GOTO(out_put, rc = -ENOSPC);
2018 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2024 for (i = 0; i < stripe_count; i++)
2025 if (stripe[i] != NULL)
2026 dt_object_put(env, stripe[i]);
2027 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_count);
2028 lo->ldo_dir_stripe_count = 0;
2029 lo->ldo_dir_stripes_allocated = 0;
2030 lo->ldo_stripe = NULL;
2034 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
2040 * Declare create striped md object.
2042 * The function declares intention to create a striped directory. This is a
2043 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2044 * is to verify pattern \a lum_buf is good. Check that function for the details.
2046 * \param[in] env execution environment
2047 * \param[in] dt object
2048 * \param[in] attr attributes to initialize the objects with
2049 * \param[in] lum_buf a pattern specifying the number of stripes and
2051 * \param[in] dof type of objects to be created
2052 * \param[in] th transaction handle
2054 * \retval 0 on success
2055 * \retval negative if failed
2058 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2059 struct dt_object *dt,
2060 struct lu_attr *attr,
2061 const struct lu_buf *lum_buf,
2062 struct dt_object_format *dof,
2065 struct lod_object *lo = lod_dt_obj(dt);
2066 struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
2070 LASSERT(lum != NULL);
2072 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
2073 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
2074 (int)le32_to_cpu(lum->lum_stripe_offset));
2076 if (lo->ldo_dir_stripe_count == 0)
2079 /* prepare dir striped objects */
2080 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2082 /* failed to create striping, let's reset
2083 * config so that others don't get confused */
2084 lod_object_free_striping(env, lo);
2092 * Implementation of dt_object_operations::do_declare_xattr_set.
2094 * Used with regular (non-striped) objects. Basically it
2095 * initializes the striping information and applies the
2096 * change to all the stripes.
2098 * \see dt_object_operations::do_declare_xattr_set() in the API description
2101 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2102 struct dt_object *dt,
2103 const struct lu_buf *buf,
2104 const char *name, int fl,
2107 struct dt_object *next = dt_object_child(dt);
2108 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2109 struct lod_object *lo = lod_dt_obj(dt);
2114 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2115 struct lmv_user_md_v1 *lum;
2117 LASSERT(buf != NULL && buf->lb_buf != NULL);
2119 rc = lod_verify_md_striping(d, lum);
2122 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2123 rc = lod_verify_striping(d, lo, buf, false);
2128 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2132 /* Note: Do not set LinkEA on sub-stripes, otherwise
2133 * it will confuse the fid2path process(see mdt_path_current()).
2134 * The linkEA between master and sub-stripes is set in
2135 * lod_xattr_set_lmv(). */
2136 if (strcmp(name, XATTR_NAME_LINK) == 0)
2139 /* set xattr to each stripes, if needed */
2140 rc = lod_load_striping(env, lo);
2144 if (lo->ldo_dir_stripe_count == 0)
2147 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2148 LASSERT(lo->ldo_stripe[i]);
2150 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2160 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2161 struct lod_object *lo,
2162 struct dt_object *dt, struct thandle *th,
2163 int comp_idx, int stripe_idx,
2164 struct lod_obj_stripe_cb_data *data)
2166 struct lod_thread_info *info = lod_env_info(env);
2167 struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
2168 struct filter_fid *ff = &info->lti_ff;
2169 struct lu_buf *buf = &info->lti_buf;
2173 buf->lb_len = sizeof(*ff);
2174 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2181 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
2182 if (lu_fid_eq(lu_object_fid(&lo->ldo_obj.do_lu), &ff->ff_parent) &&
2183 ff->ff_layout.ol_comp_id == comp->llc_id)
2186 /* rewrite filter_fid */
2187 memset(ff, 0, sizeof(*ff));
2188 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2189 ff->ff_parent.f_ver = stripe_idx;
2190 ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
2191 ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
2192 ff->ff_layout.ol_comp_id = comp->llc_id;
2193 ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
2194 ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
2195 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
2197 if (data->locd_declare)
2198 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2199 LU_XATTR_REPLACE, th);
2201 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2202 LU_XATTR_REPLACE, th);
2208 * Reset parent FID on OST object
2210 * Replace parent FID with @dt object FID, which is only called during migration
2211 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2212 * the FID is changed.
2214 * \param[in] env execution environment
2215 * \param[in] dt dt_object whose stripes's parent FID will be reset
2216 * \parem[in] th thandle
2217 * \param[in] declare if it is declare
2219 * \retval 0 if reset succeeds
2220 * \retval negative errno if reset fails
2222 static int lod_replace_parent_fid(const struct lu_env *env,
2223 struct dt_object *dt,
2224 struct thandle *th, bool declare)
2226 struct lod_object *lo = lod_dt_obj(dt);
2227 struct lod_thread_info *info = lod_env_info(env);
2228 struct lu_buf *buf = &info->lti_buf;
2229 struct filter_fid *ff;
2230 struct lod_obj_stripe_cb_data data = { { 0 } };
2234 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2236 /* set xattr to each stripes, if needed */
2237 rc = lod_load_striping(env, lo);
2241 if (!lod_obj_is_striped(dt))
2244 if (info->lti_ea_store_size < sizeof(*ff)) {
2245 rc = lod_ea_store_resize(info, sizeof(*ff));
2250 buf->lb_buf = info->lti_ea_store;
2251 buf->lb_len = info->lti_ea_store_size;
2253 data.locd_declare = declare;
2254 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2255 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2260 inline __u16 lod_comp_entry_stripe_count(struct lod_object *lo,
2261 struct lod_layout_component *entry,
2264 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2268 else if (lod_comp_inited(entry))
2269 return entry->llc_stripe_count;
2270 else if ((__u16)-1 == entry->llc_stripe_count)
2271 return lod->lod_desc.ld_tgt_count;
2273 return lod_get_stripe_count(lod, lo, entry->llc_stripe_count);
2276 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2278 int magic, size = 0, i;
2279 struct lod_layout_component *comp_entries;
2284 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2285 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2287 lo->ldo_def_striping->lds_def_striping_is_composite;
2289 comp_cnt = lo->ldo_comp_cnt;
2290 comp_entries = lo->ldo_comp_entries;
2291 is_composite = lo->ldo_is_composite;
2295 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2297 size = sizeof(struct lov_comp_md_v1) +
2298 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2299 LASSERT(size % sizeof(__u64) == 0);
2302 for (i = 0; i < comp_cnt; i++) {
2305 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2306 stripe_count = lod_comp_entry_stripe_count(lo, &comp_entries[i],
2308 if (!is_dir && is_composite)
2309 lod_comp_shrink_stripe_count(&comp_entries[i],
2312 size += lov_user_md_size(stripe_count, magic);
2313 LASSERT(size % sizeof(__u64) == 0);
2319 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2320 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2323 * \param[in] env execution environment
2324 * \param[in] dt dt_object to add components on
2325 * \param[in] buf buffer contains components to be added
2326 * \parem[in] th thandle
2328 * \retval 0 on success
2329 * \retval negative errno on failure
2331 static int lod_declare_layout_add(const struct lu_env *env,
2332 struct dt_object *dt,
2333 const struct lu_buf *buf,
2336 struct lod_thread_info *info = lod_env_info(env);
2337 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2338 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2339 struct dt_object *next = dt_object_child(dt);
2340 struct lov_desc *desc = &d->lod_desc;
2341 struct lod_object *lo = lod_dt_obj(dt);
2342 struct lov_user_md_v3 *v3;
2343 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2345 int i, rc, array_cnt, old_array_cnt;
2348 LASSERT(lo->ldo_is_composite);
2350 if (lo->ldo_flr_state != LCM_FL_NONE)
2353 rc = lod_verify_striping(d, lo, buf, false);
2357 magic = comp_v1->lcm_magic;
2358 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2359 lustre_swab_lov_comp_md_v1(comp_v1);
2360 magic = comp_v1->lcm_magic;
2363 if (magic != LOV_USER_MAGIC_COMP_V1)
2366 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2367 OBD_ALLOC(comp_array, sizeof(*comp_array) * array_cnt);
2368 if (comp_array == NULL)
2371 memcpy(comp_array, lo->ldo_comp_entries,
2372 sizeof(*comp_array) * lo->ldo_comp_cnt);
2374 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2375 struct lov_user_md_v1 *v1;
2376 struct lu_extent *ext;
2378 v1 = (struct lov_user_md *)((char *)comp_v1 +
2379 comp_v1->lcm_entries[i].lcme_offset);
2380 ext = &comp_v1->lcm_entries[i].lcme_extent;
2382 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2383 lod_comp->llc_extent.e_start = ext->e_start;
2384 lod_comp->llc_extent.e_end = ext->e_end;
2385 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2386 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2388 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2389 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2390 lod_adjust_stripe_info(lod_comp, desc);
2392 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2393 v3 = (struct lov_user_md_v3 *) v1;
2394 if (v3->lmm_pool_name[0] != '\0') {
2395 rc = lod_set_pool(&lod_comp->llc_pool,
2403 old_array = lo->ldo_comp_entries;
2404 old_array_cnt = lo->ldo_comp_cnt;
2406 lo->ldo_comp_entries = comp_array;
2407 lo->ldo_comp_cnt = array_cnt;
2409 /* No need to increase layout generation here, it will be increased
2410 * later when generating component ID for the new components */
2412 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2413 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2414 XATTR_NAME_LOV, 0, th);
2416 lo->ldo_comp_entries = old_array;
2417 lo->ldo_comp_cnt = old_array_cnt;
2421 OBD_FREE(old_array, sizeof(*lod_comp) * old_array_cnt);
2423 LASSERT(lo->ldo_mirror_count == 1);
2424 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2429 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2430 lod_comp = &comp_array[i];
2431 if (lod_comp->llc_pool != NULL) {
2432 OBD_FREE(lod_comp->llc_pool,
2433 strlen(lod_comp->llc_pool) + 1);
2434 lod_comp->llc_pool = NULL;
2437 OBD_FREE(comp_array, sizeof(*comp_array) * array_cnt);
2442 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2443 * the '$field' can only be 'flags' now. The xattr value is binary
2444 * lov_comp_md_v1 which contains the component ID(s) and the value of
2445 * the field to be modified.
2447 * \param[in] env execution environment
2448 * \param[in] dt dt_object to be modified
2449 * \param[in] op operation string, like "set.flags"
2450 * \param[in] buf buffer contains components to be set
2451 * \parem[in] th thandle
2453 * \retval 0 on success
2454 * \retval negative errno on failure
2456 static int lod_declare_layout_set(const struct lu_env *env,
2457 struct dt_object *dt,
2458 char *op, const struct lu_buf *buf,
2461 struct lod_layout_component *lod_comp;
2462 struct lod_thread_info *info = lod_env_info(env);
2463 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2464 struct lod_object *lo = lod_dt_obj(dt);
2465 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2468 bool changed = false;
2471 if (strcmp(op, "set.flags") != 0) {
2472 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2473 lod2obd(d)->obd_name, op);
2477 magic = comp_v1->lcm_magic;
2478 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2479 lustre_swab_lov_comp_md_v1(comp_v1);
2480 magic = comp_v1->lcm_magic;
2483 if (magic != LOV_USER_MAGIC_COMP_V1)
2486 if (comp_v1->lcm_entry_count == 0) {
2487 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2488 lod2obd(d)->obd_name);
2492 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2493 __u32 id = comp_v1->lcm_entries[i].lcme_id;
2494 __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
2496 if (flags & LCME_FL_INIT) {
2498 lod_object_free_striping(env, lo);
2502 for (j = 0; j < lo->ldo_comp_cnt; j++) {
2503 lod_comp = &lo->ldo_comp_entries[j];
2504 if (id != lod_comp->llc_id)
2507 if (flags & LCME_FL_NEG) {
2508 flags &= ~LCME_FL_NEG;
2509 lod_comp->llc_flags &= ~flags;
2511 lod_comp->llc_flags |= flags;
2518 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
2519 lod2obd(d)->obd_name);
2523 lod_obj_inc_layout_gen(lo);
2525 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2526 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
2527 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
2532 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
2533 * and the xattr value is a unique component ID or a special lcme_id.
2535 * \param[in] env execution environment
2536 * \param[in] dt dt_object to be operated on
2537 * \param[in] buf buffer contains component ID or lcme_id
2538 * \parem[in] th thandle
2540 * \retval 0 on success
2541 * \retval negative errno on failure
2543 static int lod_declare_layout_del(const struct lu_env *env,
2544 struct dt_object *dt,
2545 const struct lu_buf *buf,
2548 struct lod_thread_info *info = lod_env_info(env);
2549 struct dt_object *next = dt_object_child(dt);
2550 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2551 struct lod_object *lo = lod_dt_obj(dt);
2552 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2553 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2554 __u32 magic, id, flags, neg_flags = 0;
2558 LASSERT(lo->ldo_is_composite);
2560 if (lo->ldo_flr_state != LCM_FL_NONE)
2563 magic = comp_v1->lcm_magic;
2564 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2565 lustre_swab_lov_comp_md_v1(comp_v1);
2566 magic = comp_v1->lcm_magic;
2569 if (magic != LOV_USER_MAGIC_COMP_V1)
2572 id = comp_v1->lcm_entries[0].lcme_id;
2573 flags = comp_v1->lcm_entries[0].lcme_flags;
2575 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
2576 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
2577 lod2obd(d)->obd_name, id, flags);
2581 if (id != LCME_ID_INVAL && flags != 0) {
2582 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
2583 lod2obd(d)->obd_name);
2587 if (id == LCME_ID_INVAL && !flags) {
2588 CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
2589 lod2obd(d)->obd_name);
2593 if (flags & LCME_FL_NEG) {
2594 neg_flags = flags & ~LCME_FL_NEG;
2598 left = lo->ldo_comp_cnt;
2602 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
2603 struct lod_layout_component *lod_comp;
2605 lod_comp = &lo->ldo_comp_entries[i];
2607 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
2609 else if (flags && !(flags & lod_comp->llc_flags))
2611 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
2614 if (left != (i + 1)) {
2615 CDEBUG(D_LAYOUT, "%s: this deletion will create "
2616 "a hole.\n", lod2obd(d)->obd_name);
2621 /* Mark the component as deleted */
2622 lod_comp->llc_id = LCME_ID_INVAL;
2624 /* Not instantiated component */
2625 if (lod_comp->llc_stripe == NULL)
2628 LASSERT(lod_comp->llc_stripe_count > 0);
2629 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
2630 struct dt_object *obj = lod_comp->llc_stripe[j];
2634 rc = lod_sub_declare_destroy(env, obj, th);
2640 LASSERTF(left >= 0, "left = %d\n", left);
2641 if (left == lo->ldo_comp_cnt) {
2642 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
2643 lod2obd(d)->obd_name, id);
2647 memset(attr, 0, sizeof(*attr));
2648 attr->la_valid = LA_SIZE;
2649 rc = lod_sub_declare_attr_set(env, next, attr, th);
2654 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2655 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2656 XATTR_NAME_LOV, 0, th);
2658 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
2665 * Declare layout add/set/del operations issued by special xattr names:
2667 * XATTR_LUSTRE_LOV.add add component(s) to existing file
2668 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
2669 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
2671 * \param[in] env execution environment
2672 * \param[in] dt object
2673 * \param[in] name name of xattr
2674 * \param[in] buf lu_buf contains xattr value
2675 * \param[in] th transaction handle
2677 * \retval 0 on success
2678 * \retval negative if failed
2680 static int lod_declare_modify_layout(const struct lu_env *env,
2681 struct dt_object *dt,
2683 const struct lu_buf *buf,
2686 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2687 struct lod_object *lo = lod_dt_obj(dt);
2688 struct dt_object *next = dt_object_child(&lo->ldo_obj);
2690 int rc, len = strlen(XATTR_LUSTRE_LOV);
2693 LASSERT(dt_object_exists(dt));
2695 if (strlen(name) <= len || name[len] != '.') {
2696 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
2697 lod2obd(d)->obd_name, name);
2702 dt_write_lock(env, next, 0);
2703 rc = lod_load_striping_locked(env, lo);
2707 /* the layout to be modified must be a composite layout */
2708 if (!lo->ldo_is_composite) {
2709 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
2710 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
2711 GOTO(unlock, rc = -EINVAL);
2714 op = (char *)name + len;
2715 if (strcmp(op, "add") == 0) {
2716 rc = lod_declare_layout_add(env, dt, buf, th);
2717 } else if (strcmp(op, "del") == 0) {
2718 rc = lod_declare_layout_del(env, dt, buf, th);
2719 } else if (strncmp(op, "set", strlen("set")) == 0) {
2720 rc = lod_declare_layout_set(env, dt, op, buf, th);
2722 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
2723 lod2obd(d)->obd_name, name);
2724 GOTO(unlock, rc = -ENOTSUPP);
2728 lod_object_free_striping(env, lo);
2729 dt_write_unlock(env, next);
2735 * Convert a plain file lov_mds_md to a composite layout.
2737 * \param[in,out] info the thread info::lti_ea_store buffer contains little
2738 * endian plain file layout
2740 * \retval 0 on success, <0 on failure
2742 static int lod_layout_convert(struct lod_thread_info *info)
2744 struct lov_mds_md *lmm = info->lti_ea_store;
2745 struct lov_mds_md *lmm_save;
2746 struct lov_comp_md_v1 *lcm;
2747 struct lov_comp_md_entry_v1 *lcme;
2753 /* realloc buffer to a composite layout which contains one component */
2754 blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
2755 le32_to_cpu(lmm->lmm_magic));
2756 size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
2758 OBD_ALLOC_LARGE(lmm_save, blob_size);
2760 GOTO(out, rc = -ENOMEM);
2762 memcpy(lmm_save, lmm, blob_size);
2764 if (info->lti_ea_store_size < size) {
2765 rc = lod_ea_store_resize(info, size);
2770 lcm = info->lti_ea_store;
2771 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
2772 lcm->lcm_size = cpu_to_le32(size);
2773 lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
2774 lmm_save->lmm_layout_gen));
2775 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
2776 lcm->lcm_entry_count = cpu_to_le16(1);
2777 lcm->lcm_mirror_count = 0;
2779 lcme = &lcm->lcm_entries[0];
2780 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
2781 lcme->lcme_extent.e_start = 0;
2782 lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
2783 lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
2784 lcme->lcme_size = cpu_to_le32(blob_size);
2786 memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
2791 OBD_FREE_LARGE(lmm_save, blob_size);
2796 * Merge layouts to form a mirrored file.
2798 static int lod_declare_layout_merge(const struct lu_env *env,
2799 struct dt_object *dt, const struct lu_buf *mbuf,
2802 struct lod_thread_info *info = lod_env_info(env);
2803 struct lu_buf *buf = &info->lti_buf;
2804 struct lod_object *lo = lod_dt_obj(dt);
2805 struct lov_comp_md_v1 *lcm;
2806 struct lov_comp_md_v1 *cur_lcm;
2807 struct lov_comp_md_v1 *merge_lcm;
2808 struct lov_comp_md_entry_v1 *lcme;
2811 __u16 cur_entry_count;
2812 __u16 merge_entry_count;
2814 __u16 mirror_id = 0;
2819 merge_lcm = mbuf->lb_buf;
2820 if (mbuf->lb_len < sizeof(*merge_lcm))
2823 /* must be an existing layout from disk */
2824 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
2827 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
2829 /* do not allow to merge two mirrored files */
2830 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
2833 /* verify the target buffer */
2834 rc = lod_get_lov_ea(env, lo);
2836 RETURN(rc ? : -ENODATA);
2838 cur_lcm = info->lti_ea_store;
2839 switch (le32_to_cpu(cur_lcm->lcm_magic)) {
2842 rc = lod_layout_convert(info);
2844 case LOV_MAGIC_COMP_V1:
2853 /* info->lti_ea_store could be reallocated in lod_layout_convert() */
2854 cur_lcm = info->lti_ea_store;
2855 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
2857 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
2858 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
2859 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
2862 /* size of new layout */
2863 size = le32_to_cpu(cur_lcm->lcm_size) +
2864 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
2866 memset(buf, 0, sizeof(*buf));
2867 lu_buf_alloc(buf, size);
2868 if (buf->lb_buf == NULL)
2872 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
2874 offset = sizeof(*lcm) +
2875 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
2876 for (i = 0; i < cur_entry_count; i++) {
2877 struct lov_comp_md_entry_v1 *cur_lcme;
2879 lcme = &lcm->lcm_entries[i];
2880 cur_lcme = &cur_lcm->lcm_entries[i];
2882 lcme->lcme_offset = cpu_to_le32(offset);
2883 memcpy((char *)lcm + offset,
2884 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
2885 le32_to_cpu(lcme->lcme_size));
2887 offset += le32_to_cpu(lcme->lcme_size);
2889 if (mirror_count == 1) {
2890 /* new mirrored file, create new mirror ID */
2891 id = pflr_id(1, i + 1);
2892 lcme->lcme_id = cpu_to_le32(id);
2895 id = MAX(le32_to_cpu(lcme->lcme_id), id);
2898 mirror_id = mirror_id_of(id) + 1;
2899 for (i = 0; i < merge_entry_count; i++) {
2900 struct lov_comp_md_entry_v1 *merge_lcme;
2902 merge_lcme = &merge_lcm->lcm_entries[i];
2903 lcme = &lcm->lcm_entries[cur_entry_count + i];
2905 *lcme = *merge_lcme;
2906 lcme->lcme_offset = cpu_to_le32(offset);
2908 id = pflr_id(mirror_id, i + 1);
2909 lcme->lcme_id = cpu_to_le32(id);
2911 memcpy((char *)lcm + offset,
2912 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
2913 le32_to_cpu(lcme->lcme_size));
2915 offset += le32_to_cpu(lcme->lcme_size);
2918 /* fixup layout information */
2919 lod_obj_inc_layout_gen(lo);
2920 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
2921 lcm->lcm_size = cpu_to_le32(size);
2922 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
2923 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
2924 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NONE)
2925 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
2927 LASSERT(dt_write_locked(env, dt_object_child(dt)));
2928 lod_object_free_striping(env, lo);
2929 rc = lod_parse_striping(env, lo, buf);
2933 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
2934 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
2942 * Split layouts, just set the LOVEA with the layout from mbuf.
2944 static int lod_declare_layout_split(const struct lu_env *env,
2945 struct dt_object *dt, const struct lu_buf *mbuf,
2948 struct lod_object *lo = lod_dt_obj(dt);
2949 struct lov_comp_md_v1 *lcm = mbuf->lb_buf;
2953 lod_obj_inc_layout_gen(lo);
2954 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
2956 lod_object_free_striping(env, lo);
2957 rc = lod_parse_striping(env, lo, mbuf);
2961 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), mbuf,
2962 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
2967 * Implementation of dt_object_operations::do_declare_xattr_set.
2969 * \see dt_object_operations::do_declare_xattr_set() in the API description
2972 * the extension to the API:
2973 * - declaring LOVEA requests striping creation
2974 * - LU_XATTR_REPLACE means layout swap
2976 static int lod_declare_xattr_set(const struct lu_env *env,
2977 struct dt_object *dt,
2978 const struct lu_buf *buf,
2979 const char *name, int fl,
2982 struct dt_object *next = dt_object_child(dt);
2983 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2988 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
2989 if ((S_ISREG(mode) || mode == 0) &&
2990 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE | LU_XATTR_SPLIT)) &&
2991 (strcmp(name, XATTR_NAME_LOV) == 0 ||
2992 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
2994 * this is a request to create object's striping.
2996 * allow to declare predefined striping on a new (!mode) object
2997 * which is supposed to be replay of regular file creation
2998 * (when LOV setting is declared)
3000 * LU_XATTR_REPLACE is set to indicate a layout swap
3002 if (dt_object_exists(dt)) {
3003 rc = dt_attr_get(env, next, attr);
3007 memset(attr, 0, sizeof(*attr));
3008 attr->la_valid = LA_TYPE | LA_MODE;
3009 attr->la_mode = S_IFREG;
3011 rc = lod_declare_striped_create(env, dt, attr, buf, th);
3012 } else if (fl & LU_XATTR_MERGE) {
3013 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3014 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3015 rc = lod_declare_layout_merge(env, dt, buf, th);
3016 } else if (fl & LU_XATTR_SPLIT) {
3017 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3018 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3019 rc = lod_declare_layout_split(env, dt, buf, th);
3020 } else if (S_ISREG(mode) &&
3021 strlen(name) > strlen(XATTR_LUSTRE_LOV) + 1 &&
3022 strncmp(name, XATTR_LUSTRE_LOV,
3023 strlen(XATTR_LUSTRE_LOV)) == 0) {
3025 * this is a request to modify object's striping.
3026 * add/set/del component(s).
3028 if (!dt_object_exists(dt))
3031 rc = lod_declare_modify_layout(env, dt, name, buf, th);
3032 } else if (S_ISDIR(mode)) {
3033 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
3034 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3035 rc = lod_replace_parent_fid(env, dt, th, true);
3037 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
3044 * Apply xattr changes to the object.
3046 * Applies xattr changes to the object and the stripes if the latter exist.
3048 * \param[in] env execution environment
3049 * \param[in] dt object
3050 * \param[in] buf buffer pointing to the new value of xattr
3051 * \param[in] name name of xattr
3052 * \param[in] fl flags
3053 * \param[in] th transaction handle
3055 * \retval 0 on success
3056 * \retval negative if failed
3058 static int lod_xattr_set_internal(const struct lu_env *env,
3059 struct dt_object *dt,
3060 const struct lu_buf *buf,
3061 const char *name, int fl,
3064 struct dt_object *next = dt_object_child(dt);
3065 struct lod_object *lo = lod_dt_obj(dt);
3070 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3071 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3074 /* Note: Do not set LinkEA on sub-stripes, otherwise
3075 * it will confuse the fid2path process(see mdt_path_current()).
3076 * The linkEA between master and sub-stripes is set in
3077 * lod_xattr_set_lmv(). */
3078 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
3081 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3082 LASSERT(lo->ldo_stripe[i]);
3084 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
3094 * Delete an extended attribute.
3096 * Deletes specified xattr from the object and the stripes if the latter exist.
3098 * \param[in] env execution environment
3099 * \param[in] dt object
3100 * \param[in] name name of xattr
3101 * \param[in] th transaction handle
3103 * \retval 0 on success
3104 * \retval negative if failed
3106 static int lod_xattr_del_internal(const struct lu_env *env,
3107 struct dt_object *dt,
3108 const char *name, struct thandle *th)
3110 struct dt_object *next = dt_object_child(dt);
3111 struct lod_object *lo = lod_dt_obj(dt);
3116 rc = lod_sub_xattr_del(env, next, name, th);
3117 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3120 if (lo->ldo_dir_stripe_count == 0)
3123 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3124 LASSERT(lo->ldo_stripe[i]);
3126 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3135 * Set default striping on a directory.
3137 * Sets specified striping on a directory object unless it matches the default
3138 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3139 * EA. This striping will be used when regular file is being created in this
3142 * \param[in] env execution environment
3143 * \param[in] dt the striped object
3144 * \param[in] buf buffer with the striping
3145 * \param[in] name name of EA
3146 * \param[in] fl xattr flag (see OSD API description)
3147 * \param[in] th transaction handle
3149 * \retval 0 on success
3150 * \retval negative if failed
3152 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
3153 struct dt_object *dt,
3154 const struct lu_buf *buf,
3155 const char *name, int fl,
3158 struct lov_user_md_v1 *lum;
3159 struct lov_user_md_v3 *v3 = NULL;
3160 const char *pool_name = NULL;
3165 LASSERT(buf != NULL && buf->lb_buf != NULL);
3168 switch (lum->lmm_magic) {
3169 case LOV_USER_MAGIC_V3:
3171 if (v3->lmm_pool_name[0] != '\0')
3172 pool_name = v3->lmm_pool_name;
3174 case LOV_USER_MAGIC_V1:
3175 /* if { size, offset, count } = { 0, -1, 0 } and no pool
3176 * (i.e. all default values specified) then delete default
3177 * striping from dir. */
3179 "set default striping: sz %u # %u offset %d %s %s\n",
3180 (unsigned)lum->lmm_stripe_size,
3181 (unsigned)lum->lmm_stripe_count,
3182 (int)lum->lmm_stripe_offset,
3183 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
3185 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
3186 lum->lmm_stripe_count,
3187 lum->lmm_stripe_offset,
3190 case LOV_USER_MAGIC_COMP_V1:
3194 CERROR("Invalid magic %x\n", lum->lmm_magic);
3199 rc = lod_xattr_del_internal(env, dt, name, th);
3203 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3210 * Set default striping on a directory object.
3212 * Sets specified striping on a directory object unless it matches the default
3213 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3214 * EA. This striping will be used when a new directory is being created in the
3217 * \param[in] env execution environment
3218 * \param[in] dt the striped object
3219 * \param[in] buf buffer with the striping
3220 * \param[in] name name of EA
3221 * \param[in] fl xattr flag (see OSD API description)
3222 * \param[in] th transaction handle
3224 * \retval 0 on success
3225 * \retval negative if failed
3227 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
3228 struct dt_object *dt,
3229 const struct lu_buf *buf,
3230 const char *name, int fl,
3233 struct lmv_user_md_v1 *lum;
3237 LASSERT(buf != NULL && buf->lb_buf != NULL);
3240 CDEBUG(D_OTHER, "set default stripe_count # %u stripe_offset %d\n",
3241 le32_to_cpu(lum->lum_stripe_count),
3242 (int)le32_to_cpu(lum->lum_stripe_offset));
3244 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
3245 le32_to_cpu(lum->lum_stripe_offset)) &&
3246 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
3247 rc = lod_xattr_del_internal(env, dt, name, th);
3251 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3260 * Turn directory into a striped directory.
3262 * During replay the client sends the striping created before MDT
3263 * failure, then the layer above LOD sends this defined striping
3264 * using ->do_xattr_set(), so LOD uses this method to replay creation
3265 * of the stripes. Notice the original information for the striping
3266 * (#stripes, FIDs, etc) was transferred in declare path.
3268 * \param[in] env execution environment
3269 * \param[in] dt the striped object
3270 * \param[in] buf not used currently
3271 * \param[in] name not used currently
3272 * \param[in] fl xattr flag (see OSD API description)
3273 * \param[in] th transaction handle
3275 * \retval 0 on success
3276 * \retval negative if failed
3278 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
3279 const struct lu_buf *buf, const char *name,
3280 int fl, struct thandle *th)
3282 struct lod_object *lo = lod_dt_obj(dt);
3283 struct lod_thread_info *info = lod_env_info(env);
3284 struct lu_attr *attr = &info->lti_attr;
3285 struct dt_object_format *dof = &info->lti_format;
3286 struct lu_buf lmv_buf;
3287 struct lu_buf slave_lmv_buf;
3288 struct lmv_mds_md_v1 *lmm;
3289 struct lmv_mds_md_v1 *slave_lmm = NULL;
3290 struct dt_insert_rec *rec = &info->lti_dt_rec;
3295 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3298 /* The stripes are supposed to be allocated in declare phase,
3299 * if there are no stripes being allocated, it will skip */
3300 if (lo->ldo_dir_stripe_count == 0)
3303 rc = dt_attr_get(env, dt_object_child(dt), attr);
3307 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME |
3308 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
3309 dof->dof_type = DFT_DIR;
3311 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
3314 lmm = lmv_buf.lb_buf;
3316 OBD_ALLOC_PTR(slave_lmm);
3317 if (slave_lmm == NULL)
3320 lod_prep_slave_lmv_md(slave_lmm, lmm);
3321 slave_lmv_buf.lb_buf = slave_lmm;
3322 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
3324 rec->rec_type = S_IFDIR;
3325 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3326 struct dt_object *dto;
3327 char *stripe_name = info->lti_key;
3328 struct lu_name *sname;
3329 struct linkea_data ldata = { NULL };
3330 struct lu_buf linkea_buf;
3332 dto = lo->ldo_stripe[i];
3334 dt_write_lock(env, dto, MOR_TGT_CHILD);
3335 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
3337 dt_write_unlock(env, dto);
3341 rc = lod_sub_ref_add(env, dto, th);
3342 dt_write_unlock(env, dto);
3346 rec->rec_fid = lu_object_fid(&dto->do_lu);
3347 rc = lod_sub_insert(env, dto, (const struct dt_rec *)rec,
3348 (const struct dt_key *)dot, th, 0);
3352 rec->rec_fid = lu_object_fid(&dt->do_lu);
3353 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
3354 (const struct dt_key *)dotdot, th, 0);
3358 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
3359 cfs_fail_val != i) {
3360 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
3362 slave_lmm->lmv_master_mdt_index =
3365 slave_lmm->lmv_master_mdt_index =
3368 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
3369 XATTR_NAME_LMV, fl, th);
3374 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
3376 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3377 PFID(lu_object_fid(&dto->do_lu)), i + 1);
3379 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3380 PFID(lu_object_fid(&dto->do_lu)), i);
3382 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
3383 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
3384 sname, lu_object_fid(&dt->do_lu));
3388 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
3389 linkea_buf.lb_len = ldata.ld_leh->leh_len;
3390 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
3391 XATTR_NAME_LINK, 0, th);
3395 rec->rec_fid = lu_object_fid(&dto->do_lu);
3396 rc = lod_sub_insert(env, dt_object_child(dt),
3397 (const struct dt_rec *)rec,
3398 (const struct dt_key *)stripe_name, th, 0);
3402 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
3407 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
3408 rc = lod_sub_xattr_set(env, dt_object_child(dt),
3409 &lmv_buf, XATTR_NAME_LMV, fl, th);
3411 if (slave_lmm != NULL)
3412 OBD_FREE_PTR(slave_lmm);
3418 * Helper function to declare/execute creation of a striped directory
3420 * Called in declare/create object path, prepare striping for a directory
3421 * and prepare defaults data striping for the objects to be created in
3422 * that directory. Notice the function calls "declaration" or "execution"
3423 * methods depending on \a declare param. This is a consequence of the
3424 * current approach while we don't have natural distributed transactions:
3425 * we basically execute non-local updates in the declare phase. So, the
3426 * arguments for the both phases are the same and this is the reason for
3427 * this function to exist.
3429 * \param[in] env execution environment
3430 * \param[in] dt object
3431 * \param[in] attr attributes the stripes will be created with
3432 * \param[in] lmu lmv_user_md if MDT indices are specified
3433 * \param[in] dof format of stripes (see OSD API description)
3434 * \param[in] th transaction handle
3435 * \param[in] declare where to call "declare" or "execute" methods
3437 * \retval 0 on success
3438 * \retval negative if failed
3440 static int lod_dir_striping_create_internal(const struct lu_env *env,
3441 struct dt_object *dt,
3442 struct lu_attr *attr,
3443 const struct lu_buf *lmu,
3444 struct dt_object_format *dof,
3448 struct lod_thread_info *info = lod_env_info(env);
3449 struct lod_object *lo = lod_dt_obj(dt);
3450 const struct lod_default_striping *lds = lo->ldo_def_striping;
3454 LASSERT(ergo(lds != NULL,
3455 lds->lds_def_striping_set ||
3456 lds->lds_dir_def_striping_set));
3458 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
3459 lo->ldo_dir_stripe_offset)) {
3461 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3462 int stripe_count = lo->ldo_dir_stripe_count;
3464 if (info->lti_ea_store_size < sizeof(*v1)) {
3465 rc = lod_ea_store_resize(info, sizeof(*v1));
3468 v1 = info->lti_ea_store;
3471 memset(v1, 0, sizeof(*v1));
3472 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3473 v1->lum_stripe_count = cpu_to_le32(stripe_count);
3474 v1->lum_stripe_offset =
3475 cpu_to_le32(lo->ldo_dir_stripe_offset);
3477 info->lti_buf.lb_buf = v1;
3478 info->lti_buf.lb_len = sizeof(*v1);
3479 lmu = &info->lti_buf;
3483 rc = lod_declare_xattr_set_lmv(env, dt, attr, lmu, dof,
3486 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
3492 /* Transfer default LMV striping from the parent */
3493 if (lds != NULL && lds->lds_dir_def_striping_set &&
3494 !LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
3495 lds->lds_dir_def_stripe_offset)) {
3496 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3498 if (info->lti_ea_store_size < sizeof(*v1)) {
3499 rc = lod_ea_store_resize(info, sizeof(*v1));
3502 v1 = info->lti_ea_store;
3505 memset(v1, 0, sizeof(*v1));
3506 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3507 v1->lum_stripe_count =
3508 cpu_to_le32(lds->lds_dir_def_stripe_count);
3509 v1->lum_stripe_offset =
3510 cpu_to_le32(lds->lds_dir_def_stripe_offset);
3512 cpu_to_le32(lds->lds_dir_def_hash_type);
3514 info->lti_buf.lb_buf = v1;
3515 info->lti_buf.lb_len = sizeof(*v1);
3517 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3518 XATTR_NAME_DEFAULT_LMV,
3521 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
3523 XATTR_NAME_DEFAULT_LMV, 0,
3529 /* Transfer default LOV striping from the parent */
3530 if (lds != NULL && lds->lds_def_striping_set &&
3531 lds->lds_def_comp_cnt != 0) {
3532 struct lov_mds_md *lmm;
3533 int lmm_size = lod_comp_md_size(lo, true);
3535 if (info->lti_ea_store_size < lmm_size) {
3536 rc = lod_ea_store_resize(info, lmm_size);
3540 lmm = info->lti_ea_store;
3542 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
3546 info->lti_buf.lb_buf = lmm;
3547 info->lti_buf.lb_len = lmm_size;
3550 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
3551 XATTR_NAME_LOV, 0, th);
3553 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
3554 XATTR_NAME_LOV, 0, th);
3562 static int lod_declare_dir_striping_create(const struct lu_env *env,
3563 struct dt_object *dt,
3564 struct lu_attr *attr,
3566 struct dt_object_format *dof,
3569 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
3573 static int lod_dir_striping_create(const struct lu_env *env,
3574 struct dt_object *dt,
3575 struct lu_attr *attr,
3576 struct dt_object_format *dof,
3579 return lod_dir_striping_create_internal(env, dt, attr, NULL, dof, th,
3584 * Make LOV EA for striped object.
3586 * Generate striping information and store it in the LOV EA of the given
3587 * object. The caller must ensure nobody else is calling the function
3588 * against the object concurrently. The transaction must be started.
3589 * FLDB service must be running as well; it's used to map FID to the target,
3590 * which is stored in LOV EA.
3592 * \param[in] env execution environment for this thread
3593 * \param[in] lo LOD object
3594 * \param[in] th transaction handle
3596 * \retval 0 if LOV EA is stored successfully
3597 * \retval negative error number on failure
3599 static int lod_generate_and_set_lovea(const struct lu_env *env,
3600 struct lod_object *lo,
3603 struct lod_thread_info *info = lod_env_info(env);
3604 struct dt_object *next = dt_object_child(&lo->ldo_obj);
3605 struct lov_mds_md_v1 *lmm;
3611 if (lo->ldo_comp_cnt == 0) {
3612 lod_object_free_striping(env, lo);
3613 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
3617 lmm_size = lod_comp_md_size(lo, false);
3618 if (info->lti_ea_store_size < lmm_size) {
3619 rc = lod_ea_store_resize(info, lmm_size);
3623 lmm = info->lti_ea_store;
3625 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
3629 info->lti_buf.lb_buf = lmm;
3630 info->lti_buf.lb_len = lmm_size;
3631 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
3632 XATTR_NAME_LOV, 0, th);
3637 * Delete layout component(s)
3639 * \param[in] env execution environment for this thread
3640 * \param[in] dt object
3641 * \param[in] th transaction handle
3643 * \retval 0 on success
3644 * \retval negative error number on failure
3646 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
3649 struct lod_layout_component *lod_comp;
3650 struct lod_object *lo = lod_dt_obj(dt);
3651 struct dt_object *next = dt_object_child(dt);
3652 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3655 LASSERT(lo->ldo_is_composite);
3656 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
3658 left = lo->ldo_comp_cnt;
3659 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3660 lod_comp = &lo->ldo_comp_entries[i];
3662 if (lod_comp->llc_id != LCME_ID_INVAL)
3666 /* Not instantiated component */
3667 if (lod_comp->llc_stripe == NULL)
3670 LASSERT(lod_comp->llc_stripe_count > 0);
3671 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
3672 struct dt_object *obj = lod_comp->llc_stripe[j];
3676 rc = lod_sub_destroy(env, obj, th);
3680 lu_object_put(env, &obj->do_lu);
3681 lod_comp->llc_stripe[j] = NULL;
3683 OBD_FREE(lod_comp->llc_stripe, sizeof(struct dt_object *) *
3684 lod_comp->llc_stripes_allocated);
3685 lod_comp->llc_stripe = NULL;
3686 OBD_FREE(lod_comp->llc_ost_indices,
3687 sizeof(__u32) * lod_comp->llc_stripes_allocated);
3688 lod_comp->llc_ost_indices = NULL;
3689 lod_comp->llc_stripes_allocated = 0;
3690 lod_obj_set_pool(lo, i, NULL);
3691 if (lod_comp->llc_ostlist.op_array) {
3692 OBD_FREE(lod_comp->llc_ostlist.op_array,
3693 lod_comp->llc_ostlist.op_size);
3694 lod_comp->llc_ostlist.op_array = NULL;
3695 lod_comp->llc_ostlist.op_size = 0;
3699 LASSERTF(left >= 0 && left < lo->ldo_comp_cnt, "left = %d\n", left);
3701 struct lod_layout_component *comp_array;
3703 OBD_ALLOC(comp_array, sizeof(*comp_array) * left);
3704 if (comp_array == NULL)
3705 GOTO(out, rc = -ENOMEM);
3707 memcpy(&comp_array[0], &lo->ldo_comp_entries[0],
3708 sizeof(*comp_array) * left);
3710 OBD_FREE(lo->ldo_comp_entries,
3711 sizeof(*comp_array) * lo->ldo_comp_cnt);
3712 lo->ldo_comp_entries = comp_array;
3713 lo->ldo_comp_cnt = left;
3715 LASSERT(lo->ldo_mirror_count == 1);
3716 lo->ldo_mirrors[0].lme_end = left - 1;
3717 lod_obj_inc_layout_gen(lo);
3719 lod_free_comp_entries(lo);
3722 LASSERT(dt_object_exists(dt));
3723 rc = dt_attr_get(env, next, attr);
3727 if (attr->la_size > 0) {
3729 attr->la_valid = LA_SIZE;
3730 rc = lod_sub_attr_set(env, next, attr, th);
3735 rc = lod_generate_and_set_lovea(env, lo, th);
3739 lod_object_free_striping(env, lo);
3744 static int lod_get_default_lov_striping(const struct lu_env *env,
3745 struct lod_object *lo,
3746 struct lod_default_striping *lds);
3748 * Implementation of dt_object_operations::do_xattr_set.
3750 * Sets specified extended attribute on the object. Three types of EAs are
3752 * LOV EA - stores striping for a regular file or default striping (when set
3754 * LMV EA - stores a marker for the striped directories
3755 * DMV EA - stores default directory striping
3757 * When striping is applied to a non-striped existing object (this is called
3758 * late striping), then LOD notices the caller wants to turn the object into a
3759 * striped one. The stripe objects are created and appropriate EA is set:
3760 * LOV EA storing all the stripes directly or LMV EA storing just a small header
3761 * with striping configuration.
3763 * \see dt_object_operations::do_xattr_set() in the API description for details.
3765 static int lod_xattr_set(const struct lu_env *env,
3766 struct dt_object *dt, const struct lu_buf *buf,
3767 const char *name, int fl, struct thandle *th)
3769 struct dt_object *next = dt_object_child(dt);
3773 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
3774 strcmp(name, XATTR_NAME_LMV) == 0) {
3775 struct lmv_mds_md_v1 *lmm = buf->lb_buf;
3777 if (lmm != NULL && le32_to_cpu(lmm->lmv_hash_type) &
3778 LMV_HASH_FLAG_MIGRATION)
3779 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3781 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
3786 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
3787 strcmp(name, XATTR_NAME_LOV) == 0) {
3788 struct lod_thread_info *info = lod_env_info(env);
3789 struct lod_default_striping *lds = &info->lti_def_striping;
3790 struct lov_user_md_v1 *v1 = buf->lb_buf;
3791 char pool[LOV_MAXPOOLNAME + 1];
3794 /* get existing striping config */
3795 rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds);
3799 memset(pool, 0, sizeof(pool));
3800 if (lds->lds_def_striping_set == 1)
3801 lod_layout_get_pool(lds->lds_def_comp_entries,
3802 lds->lds_def_comp_cnt, pool,
3805 is_del = LOVEA_DELETE_VALUES(v1->lmm_stripe_size,
3806 v1->lmm_stripe_count,
3807 v1->lmm_stripe_offset,
3810 /* Retain the pool name if it is not given */
3811 if (v1->lmm_magic == LOV_USER_MAGIC_V1 && pool[0] != '\0' &&
3813 struct lod_thread_info *info = lod_env_info(env);
3814 struct lov_user_md_v3 *v3 = info->lti_ea_store;
3816 memset(v3, 0, sizeof(*v3));
3817 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
3818 v3->lmm_pattern = cpu_to_le32(v1->lmm_pattern);
3819 v3->lmm_stripe_count =
3820 cpu_to_le32(v1->lmm_stripe_count);
3821 v3->lmm_stripe_offset =
3822 cpu_to_le32(v1->lmm_stripe_offset);
3823 v3->lmm_stripe_size = cpu_to_le32(v1->lmm_stripe_size);
3825 strlcpy(v3->lmm_pool_name, pool,
3826 sizeof(v3->lmm_pool_name));
3828 info->lti_buf.lb_buf = v3;
3829 info->lti_buf.lb_len = sizeof(*v3);
3830 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
3833 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name,
3837 if (lds->lds_def_striping_set == 1 &&
3838 lds->lds_def_comp_entries != NULL)
3839 lod_free_def_comp_entries(lds);
3842 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
3843 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
3845 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
3848 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
3849 (!strcmp(name, XATTR_NAME_LOV) ||
3850 !strncmp(name, XATTR_LUSTRE_LOV,
3851 strlen(XATTR_LUSTRE_LOV)))) {
3852 /* in case of lov EA swap, just set it
3853 * if not, it is a replay so check striping match what we
3854 * already have during req replay, declare_xattr_set()
3855 * defines striping, then create() does the work */
3856 if (fl & LU_XATTR_REPLACE) {
3857 /* free stripes, then update disk */
3858 lod_object_free_striping(env, lod_dt_obj(dt));
3860 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3861 } else if (dt_object_remote(dt)) {
3862 /* This only happens during migration, see
3863 * mdd_migrate_create(), in which Master MDT will
3864 * create a remote target object, and only set
3865 * (migrating) stripe EA on the remote object,
3866 * and does not need creating each stripes. */
3867 rc = lod_sub_xattr_set(env, next, buf, name,
3869 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
3870 /* delete component(s) */
3871 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
3872 rc = lod_layout_del(env, dt, th);
3875 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
3876 * it's going to create create file with specified
3877 * component(s), the striping must have not being
3878 * cached in this case;
3880 * Otherwise, it's going to add/change component(s) to
3881 * an existing file, the striping must have been cached
3884 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
3885 !strcmp(name, XATTR_NAME_LOV),
3886 !lod_dt_obj(dt)->ldo_comp_cached));
3888 rc = lod_striped_create(env, dt, NULL, NULL, th);
3891 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3892 rc = lod_replace_parent_fid(env, dt, th, false);
3897 /* then all other xattr */
3898 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3904 * Implementation of dt_object_operations::do_declare_xattr_del.
3906 * \see dt_object_operations::do_declare_xattr_del() in the API description
3909 static int lod_declare_xattr_del(const struct lu_env *env,
3910 struct dt_object *dt, const char *name,
3913 struct lod_object *lo = lod_dt_obj(dt);
3918 rc = lod_sub_declare_xattr_del(env, dt_object_child(dt), name, th);
3922 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3925 /* set xattr to each stripes, if needed */
3926 rc = lod_load_striping(env, lo);
3930 if (lo->ldo_dir_stripe_count == 0)
3933 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3934 LASSERT(lo->ldo_stripe[i]);
3935 rc = lod_sub_declare_xattr_del(env, lo->ldo_stripe[i],
3945 * Implementation of dt_object_operations::do_xattr_del.
3947 * If EA storing a regular striping is being deleted, then release
3948 * all the references to the stripe objects in core.
3950 * \see dt_object_operations::do_xattr_del() in the API description for details.
3952 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
3953 const char *name, struct thandle *th)
3955 struct dt_object *next = dt_object_child(dt);
3956 struct lod_object *lo = lod_dt_obj(dt);
3961 if (!strcmp(name, XATTR_NAME_LOV))
3962 lod_object_free_striping(env, lod_dt_obj(dt));
3964 rc = lod_sub_xattr_del(env, next, name, th);
3965 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3968 if (lo->ldo_dir_stripe_count == 0)
3971 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3972 LASSERT(lo->ldo_stripe[i]);
3974 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3983 * Implementation of dt_object_operations::do_xattr_list.
3985 * \see dt_object_operations::do_xattr_list() in the API description
3988 static int lod_xattr_list(const struct lu_env *env,
3989 struct dt_object *dt, const struct lu_buf *buf)
3991 return dt_xattr_list(env, dt_object_child(dt), buf);
3994 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
3996 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
4001 * Get default striping.
4003 * \param[in] env execution environment
4004 * \param[in] lo object
4005 * \param[out] lds default striping
4007 * \retval 0 on success
4008 * \retval negative if failed
4010 static int lod_get_default_lov_striping(const struct lu_env *env,
4011 struct lod_object *lo,
4012 struct lod_default_striping *lds)
4014 struct lod_thread_info *info = lod_env_info(env);
4015 struct lov_user_md_v1 *v1 = NULL;
4016 struct lov_user_md_v3 *v3 = NULL;
4017 struct lov_comp_md_v1 *comp_v1 = NULL;
4024 lds->lds_def_striping_set = 0;
4026 rc = lod_get_lov_ea(env, lo);
4030 if (rc < (typeof(rc))sizeof(struct lov_user_md))
4033 v1 = info->lti_ea_store;
4034 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
4035 lustre_swab_lov_user_md_v1(v1);
4036 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
4037 v3 = (struct lov_user_md_v3 *)v1;
4038 lustre_swab_lov_user_md_v3(v3);
4039 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
4040 comp_v1 = (struct lov_comp_md_v1 *)v1;
4041 lustre_swab_lov_comp_md_v1(comp_v1);
4044 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1 &&
4045 v1->lmm_magic != LOV_MAGIC_COMP_V1)
4048 if (v1->lmm_magic == LOV_MAGIC_COMP_V1) {
4049 comp_v1 = (struct lov_comp_md_v1 *)v1;
4050 comp_cnt = comp_v1->lcm_entry_count;
4053 mirror_cnt = comp_v1->lcm_mirror_count + 1;
4061 /* realloc default comp entries if necessary */
4062 rc = lod_def_striping_comp_resize(lds, comp_cnt);
4066 lds->lds_def_comp_cnt = comp_cnt;
4067 lds->lds_def_striping_is_composite = composite;
4068 lds->lds_def_mirror_cnt = mirror_cnt;
4070 for (i = 0; i < comp_cnt; i++) {
4071 struct lod_layout_component *lod_comp;
4072 struct lu_extent *ext;
4075 lod_comp = &lds->lds_def_comp_entries[i];
4077 * reset lod_comp values, llc_stripes is always NULL in
4078 * the default striping template, llc_pool will be reset
4081 memset(lod_comp, 0, offsetof(typeof(*lod_comp), llc_pool));
4084 v1 = (struct lov_user_md *)((char *)comp_v1 +
4085 comp_v1->lcm_entries[i].lcme_offset);
4086 ext = &comp_v1->lcm_entries[i].lcme_extent;
4087 lod_comp->llc_extent = *ext;
4090 if (v1->lmm_pattern != LOV_PATTERN_RAID0 &&
4091 v1->lmm_pattern != LOV_PATTERN_MDT &&
4092 v1->lmm_pattern != 0) {
4093 lod_free_def_comp_entries(lds);
4097 CDEBUG(D_LAYOUT, DFID" stripe_count=%d stripe_size=%d "
4098 "stripe_offset=%d\n",
4099 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
4100 (int)v1->lmm_stripe_count, (int)v1->lmm_stripe_size,
4101 (int)v1->lmm_stripe_offset);
4103 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
4104 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
4105 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
4106 lod_comp->llc_pattern = v1->lmm_pattern;
4109 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
4110 /* XXX: sanity check here */
4111 v3 = (struct lov_user_md_v3 *) v1;
4112 if (v3->lmm_pool_name[0] != '\0')
4113 pool = v3->lmm_pool_name;
4115 lod_set_def_pool(lds, i, pool);
4118 lds->lds_def_striping_set = 1;
4123 * Get default directory striping.
4125 * \param[in] env execution environment
4126 * \param[in] lo object
4127 * \param[out] lds default striping
4129 * \retval 0 on success
4130 * \retval negative if failed
4132 static int lod_get_default_lmv_striping(const struct lu_env *env,
4133 struct lod_object *lo,
4134 struct lod_default_striping *lds)
4136 struct lod_thread_info *info = lod_env_info(env);
4137 struct lmv_user_md_v1 *v1 = NULL;
4141 lds->lds_dir_def_striping_set = 0;
4142 rc = lod_get_default_lmv_ea(env, lo);
4146 if (rc < (typeof(rc))sizeof(struct lmv_user_md))
4149 v1 = info->lti_ea_store;
4151 lds->lds_dir_def_stripe_count = le32_to_cpu(v1->lum_stripe_count);
4152 lds->lds_dir_def_stripe_offset = le32_to_cpu(v1->lum_stripe_offset);
4153 lds->lds_dir_def_hash_type = le32_to_cpu(v1->lum_hash_type);
4154 lds->lds_dir_def_striping_set = 1;
4160 * Get default striping in the object.
4162 * Get object default striping and default directory striping.
4164 * \param[in] env execution environment
4165 * \param[in] lo object
4166 * \param[out] lds default striping
4168 * \retval 0 on success
4169 * \retval negative if failed
4171 static int lod_get_default_striping(const struct lu_env *env,
4172 struct lod_object *lo,
4173 struct lod_default_striping *lds)
4177 rc = lod_get_default_lov_striping(env, lo, lds);
4178 rc1 = lod_get_default_lmv_striping(env, lo, lds);
4179 if (rc == 0 && rc1 < 0)
4186 * Apply default striping on object.
4188 * If object striping pattern is not set, set to the one in default striping.
4189 * The default striping is from parent or fs.
4191 * \param[in] lo new object
4192 * \param[in] lds default striping
4193 * \param[in] mode new object's mode
4195 static void lod_striping_from_default(struct lod_object *lo,
4196 const struct lod_default_striping *lds,
4199 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
4200 struct lov_desc *desc = &d->lod_desc;
4203 if (lds->lds_def_striping_set && S_ISREG(mode)) {
4204 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
4205 lds->lds_def_comp_cnt);
4209 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
4210 if (lds->lds_def_mirror_cnt > 1)
4211 lo->ldo_flr_state = LCM_FL_RDONLY;
4213 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4214 struct lod_layout_component *obj_comp =
4215 &lo->ldo_comp_entries[i];
4216 struct lod_layout_component *def_comp =
4217 &lds->lds_def_comp_entries[i];
4219 CDEBUG(D_LAYOUT, "Inherite from default: size:%hu "
4220 "nr:%u offset:%u pattern %#x %s\n",
4221 def_comp->llc_stripe_size,
4222 def_comp->llc_stripe_count,
4223 def_comp->llc_stripe_offset,
4224 def_comp->llc_pattern,
4225 def_comp->llc_pool ?: "");
4227 *obj_comp = *def_comp;
4228 if (def_comp->llc_pool != NULL) {
4229 /* pointer was copied from def_comp */
4230 obj_comp->llc_pool = NULL;
4231 lod_obj_set_pool(lo, i, def_comp->llc_pool);
4235 * Don't initialize these fields for plain layout
4236 * (v1/v3) here, they are inherited in the order of
4237 * 'parent' -> 'fs default (root)' -> 'global default
4238 * values for stripe_count & stripe_size'.
4240 * see lod_ah_init().
4242 if (!lo->ldo_is_composite)
4245 lod_adjust_stripe_info(obj_comp, desc);
4247 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
4248 if (lo->ldo_dir_stripe_count == 0)
4249 lo->ldo_dir_stripe_count =
4250 lds->lds_dir_def_stripe_count;
4251 if (lo->ldo_dir_stripe_offset == -1)
4252 lo->ldo_dir_stripe_offset =
4253 lds->lds_dir_def_stripe_offset;
4254 if (lo->ldo_dir_hash_type == 0)
4255 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
4257 CDEBUG(D_LAYOUT, "striping from default dir: count:%hu, "
4258 "offset:%u, hash_type:%u\n",
4259 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
4260 lo->ldo_dir_hash_type);
4264 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root)
4266 struct lod_layout_component *lod_comp;
4268 if (lo->ldo_comp_cnt == 0)
4271 if (lo->ldo_is_composite)
4274 lod_comp = &lo->ldo_comp_entries[0];
4276 if (lod_comp->llc_stripe_count <= 0 ||
4277 lod_comp->llc_stripe_size <= 0)
4280 if (from_root && (lod_comp->llc_pool == NULL ||
4281 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
4288 * Implementation of dt_object_operations::do_ah_init.
4290 * This method is used to make a decision on the striping configuration for the
4291 * object being created. It can be taken from the \a parent object if it exists,
4292 * or filesystem's default. The resulting configuration (number of stripes,
4293 * stripe size/offset, pool name, etc) is stored in the object itself and will
4294 * be used by the methods like ->doo_declare_create().
4296 * \see dt_object_operations::do_ah_init() in the API description for details.
4298 static void lod_ah_init(const struct lu_env *env,
4299 struct dt_allocation_hint *ah,
4300 struct dt_object *parent,
4301 struct dt_object *child,
4304 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
4305 struct lod_thread_info *info = lod_env_info(env);
4306 struct lod_default_striping *lds = &info->lti_def_striping;
4307 struct dt_object *nextp = NULL;
4308 struct dt_object *nextc;
4309 struct lod_object *lp = NULL;
4310 struct lod_object *lc;
4311 struct lov_desc *desc;
4312 struct lod_layout_component *lod_comp;
4318 if (likely(parent)) {
4319 nextp = dt_object_child(parent);
4320 lp = lod_dt_obj(parent);
4323 nextc = dt_object_child(child);
4324 lc = lod_dt_obj(child);
4326 LASSERT(!lod_obj_is_striped(child));
4327 /* default layout template may have been set on the regular file
4328 * when this is called from mdd_create_data() */
4329 if (S_ISREG(child_mode))
4330 lod_free_comp_entries(lc);
4332 if (!dt_object_exists(nextc))
4333 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
4335 if (S_ISDIR(child_mode)) {
4336 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
4338 /* other default values are 0 */
4339 lc->ldo_dir_stripe_offset = -1;
4341 /* get default striping from parent object */
4342 if (likely(lp != NULL))
4343 lod_get_default_striping(env, lp, lds);
4345 /* set child default striping info, default value is NULL */
4346 if (lds->lds_def_striping_set || lds->lds_dir_def_striping_set)
4347 lc->ldo_def_striping = lds;
4349 /* It should always honour the specified stripes */
4350 /* Note: old client (< 2.7)might also do lfs mkdir, whose EA
4351 * will have old magic. In this case, we should ignore the
4352 * stripe count and try to create dir by default stripe.
4354 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
4355 (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
4356 le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC)) {
4357 lc->ldo_dir_stripe_count =
4358 le32_to_cpu(lum1->lum_stripe_count);
4359 lc->ldo_dir_stripe_offset =
4360 le32_to_cpu(lum1->lum_stripe_offset);
4361 lc->ldo_dir_hash_type =
4362 le32_to_cpu(lum1->lum_hash_type);
4364 "set dirstripe: count %hu, offset %d, hash %u\n",
4365 lc->ldo_dir_stripe_count,
4366 (int)lc->ldo_dir_stripe_offset,
4367 lc->ldo_dir_hash_type);
4369 /* transfer defaults LMV to new directory */
4370 lod_striping_from_default(lc, lds, child_mode);
4373 /* shrink the stripe_count to the avaible MDT count */
4374 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
4375 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))
4376 lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
4378 /* Directory will be striped only if stripe_count > 1, if
4379 * stripe_count == 1, let's reset stripe_count = 0 to avoid
4380 * create single master stripe and also help to unify the
4381 * stripe handling of directories and files */
4382 if (lc->ldo_dir_stripe_count == 1)
4383 lc->ldo_dir_stripe_count = 0;
4385 CDEBUG(D_INFO, "final dir stripe [%hu %d %u]\n",
4386 lc->ldo_dir_stripe_count,
4387 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
4392 /* child object regular file*/
4394 if (!lod_object_will_be_striped(S_ISREG(child_mode),
4395 lu_object_fid(&child->do_lu)))
4398 /* If object is going to be striped over OSTs, transfer default
4399 * striping information to the child, so that we can use it
4400 * during declaration and creation.
4402 * Try from the parent first.
4404 if (likely(lp != NULL)) {
4405 rc = lod_get_default_lov_striping(env, lp, lds);
4407 lod_striping_from_default(lc, lds, child_mode);
4410 /* Initialize lod_device::lod_md_root object reference */
4411 if (d->lod_md_root == NULL) {
4412 struct dt_object *root;
4413 struct lod_object *lroot;
4415 lu_root_fid(&info->lti_fid);
4416 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
4417 if (!IS_ERR(root)) {
4418 lroot = lod_dt_obj(root);
4420 spin_lock(&d->lod_lock);
4421 if (d->lod_md_root != NULL)
4422 dt_object_put(env, &d->lod_md_root->ldo_obj);
4423 d->lod_md_root = lroot;
4424 spin_unlock(&d->lod_lock);
4428 /* try inherit layout from the root object (fs default) when:
4429 * - parent does not have default layout; or
4430 * - parent has plain(v1/v3) default layout, and some attributes
4431 * are not specified in the default layout;
4433 if (d->lod_md_root != NULL && lod_need_inherit_more(lc, true)) {
4434 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds);
4437 if (lc->ldo_comp_cnt == 0) {
4438 lod_striping_from_default(lc, lds, child_mode);
4439 } else if (!lds->lds_def_striping_is_composite) {
4440 struct lod_layout_component *def_comp;
4442 LASSERT(!lc->ldo_is_composite);
4443 lod_comp = &lc->ldo_comp_entries[0];
4444 def_comp = &lds->lds_def_comp_entries[0];
4446 if (lod_comp->llc_stripe_count <= 0)
4447 lod_comp->llc_stripe_count =
4448 def_comp->llc_stripe_count;
4449 if (lod_comp->llc_stripe_size <= 0)
4450 lod_comp->llc_stripe_size =
4451 def_comp->llc_stripe_size;
4452 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT)
4453 lod_comp->llc_stripe_offset =
4454 def_comp->llc_stripe_offset;
4455 if (lod_comp->llc_pool == NULL)
4456 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
4461 * fs default striping may not be explicitly set, or historically set
4462 * in config log, use them.
4464 if (lod_need_inherit_more(lc, false)) {
4465 if (lc->ldo_comp_cnt == 0) {
4466 rc = lod_alloc_comp_entries(lc, 0, 1);
4468 /* fail to allocate memory, will create a
4469 * non-striped file. */
4471 lc->ldo_is_composite = 0;
4472 lod_comp = &lc->ldo_comp_entries[0];
4473 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4475 LASSERT(!lc->ldo_is_composite);
4476 lod_comp = &lc->ldo_comp_entries[0];
4477 desc = &d->lod_desc;
4478 lod_adjust_stripe_info(lod_comp, desc);
4484 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
4486 * Size initialization on late striping.
4488 * Propagate the size of a truncated object to a deferred striping.
4489 * This function handles a special case when truncate was done on a
4490 * non-striped object and now while the striping is being created
4491 * we can't lose that size, so we have to propagate it to the stripes
4494 * \param[in] env execution environment
4495 * \param[in] dt object
4496 * \param[in] th transaction handle
4498 * \retval 0 on success
4499 * \retval negative if failed
4501 static int lod_declare_init_size(const struct lu_env *env,
4502 struct dt_object *dt, struct thandle *th)
4504 struct dt_object *next = dt_object_child(dt);
4505 struct lod_object *lo = lod_dt_obj(dt);
4506 struct dt_object **objects = NULL;
4507 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4508 uint64_t size, offs;
4509 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
4510 struct lu_extent size_ext;
4513 if (!lod_obj_is_striped(dt))
4516 rc = dt_attr_get(env, next, attr);
4517 LASSERT(attr->la_valid & LA_SIZE);
4521 size = attr->la_size;
4525 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
4526 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4527 struct lod_layout_component *lod_comp;
4528 struct lu_extent *extent;
4530 lod_comp = &lo->ldo_comp_entries[i];
4532 if (lod_comp->llc_stripe == NULL)
4535 extent = &lod_comp->llc_extent;
4536 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
4537 if (!lo->ldo_is_composite ||
4538 lu_extent_is_overlapped(extent, &size_ext)) {
4539 objects = lod_comp->llc_stripe;
4540 stripe_count = lod_comp->llc_stripe_count;
4541 stripe_size = lod_comp->llc_stripe_size;
4544 if (stripe_count == 0)
4547 LASSERT(objects != NULL && stripe_size != 0);
4548 /* ll_do_div64(a, b) returns a % b, and a = a / b */
4549 ll_do_div64(size, (__u64)stripe_size);
4550 stripe = ll_do_div64(size, (__u64)stripe_count);
4551 LASSERT(objects[stripe] != NULL);
4553 size = size * stripe_size;
4554 offs = attr->la_size;
4555 size += ll_do_div64(offs, stripe_size);
4557 attr->la_valid = LA_SIZE;
4558 attr->la_size = size;
4560 rc = lod_sub_declare_attr_set(env, objects[stripe],
4569 * Declare creation of striped object.
4571 * The function declares creation stripes for a regular object. The function
4572 * also declares whether the stripes will be created with non-zero size if
4573 * previously size was set non-zero on the master object. If object \a dt is
4574 * not local, then only fully defined striping can be applied in \a lovea.
4575 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
4578 * \param[in] env execution environment
4579 * \param[in] dt object
4580 * \param[in] attr attributes the stripes will be created with
4581 * \param[in] lovea a buffer containing striping description
4582 * \param[in] th transaction handle
4584 * \retval 0 on success
4585 * \retval negative if failed
4587 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
4588 struct lu_attr *attr,
4589 const struct lu_buf *lovea, struct thandle *th)
4591 struct lod_thread_info *info = lod_env_info(env);
4592 struct dt_object *next = dt_object_child(dt);
4593 struct lod_object *lo = lod_dt_obj(dt);
4597 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
4598 GOTO(out, rc = -ENOMEM);
4600 if (!dt_object_remote(next)) {
4601 /* choose OST and generate appropriate objects */
4602 rc = lod_prepare_create(env, lo, attr, lovea, th);
4607 * declare storage for striping data
4609 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
4611 /* LOD can not choose OST objects for remote objects, i.e.
4612 * stripes must be ready before that. Right now, it can only
4613 * happen during migrate, i.e. migrate process needs to create
4614 * remote regular file (mdd_migrate_create), then the migrate
4615 * process will provide stripeEA. */
4616 LASSERT(lovea != NULL);
4617 info->lti_buf = *lovea;
4620 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
4621 XATTR_NAME_LOV, 0, th);
4626 * if striping is created with local object's size > 0,
4627 * we have to propagate this size to specific object
4628 * the case is possible only when local object was created previously
4630 if (dt_object_exists(next))
4631 rc = lod_declare_init_size(env, dt, th);
4634 /* failed to create striping or to set initial size, let's reset
4635 * config so that others don't get confused */
4637 lod_object_free_striping(env, lo);
4643 * Implementation of dt_object_operations::do_declare_create.
4645 * The method declares creation of a new object. If the object will be striped,
4646 * then helper functions are called to find FIDs for the stripes, declare
4647 * creation of the stripes and declare initialization of the striping
4648 * information to be stored in the master object.
4650 * \see dt_object_operations::do_declare_create() in the API description
4653 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
4654 struct lu_attr *attr,
4655 struct dt_allocation_hint *hint,
4656 struct dt_object_format *dof, struct thandle *th)
4658 struct dt_object *next = dt_object_child(dt);
4659 struct lod_object *lo = lod_dt_obj(dt);
4668 * first of all, we declare creation of local object
4670 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
4675 * it's lod_ah_init() that has decided the object will be striped
4677 if (dof->dof_type == DFT_REGULAR) {
4678 /* callers don't want stripes */
4679 /* XXX: all tricky interactions with ->ah_make_hint() decided
4680 * to use striping, then ->declare_create() behaving differently
4681 * should be cleaned */
4682 if (dof->u.dof_reg.striped != 0)
4683 rc = lod_declare_striped_create(env, dt, attr,
4685 } else if (dof->dof_type == DFT_DIR) {
4686 struct seq_server_site *ss;
4687 struct lu_buf buf = { NULL };
4688 struct lu_buf *lmu = NULL;
4690 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
4692 /* If the parent has default stripeEA, and client
4693 * did not find it before sending create request,
4694 * then MDT will return -EREMOTE, and client will
4695 * retrieve the default stripeEA and re-create the
4698 * Note: if dah_eadata != NULL, it means creating the
4699 * striped directory with specified stripeEA, then it
4700 * should ignore the default stripeEA */
4701 if (hint != NULL && hint->dah_eadata == NULL) {
4702 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
4703 GOTO(out, rc = -EREMOTE);
4705 if (lo->ldo_dir_stripe_offset == -1) {
4706 /* child and parent should be in the same MDT */
4707 if (hint->dah_parent != NULL &&
4708 dt_object_remote(hint->dah_parent))
4709 GOTO(out, rc = -EREMOTE);
4710 } else if (lo->ldo_dir_stripe_offset !=
4712 struct lod_device *lod;
4713 struct lod_tgt_descs *ltd;
4714 struct lod_tgt_desc *tgt = NULL;
4715 bool found_mdt = false;
4718 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
4719 ltd = &lod->lod_mdt_descs;
4720 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
4721 tgt = LTD_TGT(ltd, i);
4722 if (tgt->ltd_index ==
4723 lo->ldo_dir_stripe_offset) {
4729 /* If the MDT indicated by stripe_offset can be
4730 * found, then tell client to resend the create
4731 * request to the correct MDT, otherwise return
4732 * error to client */
4734 GOTO(out, rc = -EREMOTE);
4736 GOTO(out, rc = -EINVAL);
4738 } else if (hint && hint->dah_eadata) {
4740 lmu->lb_buf = (void *)hint->dah_eadata;
4741 lmu->lb_len = hint->dah_eadata_len;
4744 rc = lod_declare_dir_striping_create(env, dt, attr, lmu, dof,
4748 /* failed to create striping or to set initial size, let's reset
4749 * config so that others don't get confused */
4751 lod_object_free_striping(env, lo);
4756 * Generate component ID for new created component.
4758 * \param[in] lo LOD object
4759 * \param[in] comp_idx index of ldo_comp_entries
4761 * \retval component ID on success
4762 * \retval LCME_ID_INVAL on failure
4764 static __u32 lod_gen_component_id(struct lod_object *lo,
4765 int mirror_id, int comp_idx)
4767 struct lod_layout_component *lod_comp;
4768 __u32 id, start, end;
4771 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
4773 lod_obj_inc_layout_gen(lo);
4774 id = lo->ldo_layout_gen;
4775 if (likely(id <= SEQ_ID_MAX))
4776 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
4778 /* Layout generation wraps, need to check collisions. */
4779 start = id & SEQ_ID_MASK;
4782 for (id = start; id <= end; id++) {
4783 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4784 lod_comp = &lo->ldo_comp_entries[i];
4785 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
4788 /* Found the ununsed ID */
4789 if (i == lo->ldo_comp_cnt)
4790 RETURN(pflr_id(mirror_id, id));
4792 if (end == LCME_ID_MAX) {
4794 end = min(lo->ldo_layout_gen & LCME_ID_MASK,
4795 (__u32)(LCME_ID_MAX - 1));
4799 RETURN(LCME_ID_INVAL);
4803 * Creation of a striped regular object.
4805 * The function is called to create the stripe objects for a regular
4806 * striped file. This can happen at the initial object creation or
4807 * when the caller asks LOD to do so using ->do_xattr_set() method
4808 * (so called late striping). Notice all the information are already
4809 * prepared in the form of the list of objects (ldo_stripe field).
4810 * This is done during declare phase.
4812 * \param[in] env execution environment
4813 * \param[in] dt object
4814 * \param[in] attr attributes the stripes will be created with
4815 * \param[in] dof format of stripes (see OSD API description)
4816 * \param[in] th transaction handle
4818 * \retval 0 on success
4819 * \retval negative if failed
4821 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
4822 struct lu_attr *attr, struct dt_object_format *dof,
4825 struct lod_layout_component *lod_comp;
4826 struct lod_object *lo = lod_dt_obj(dt);
4831 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
4833 mirror_id = lo->ldo_mirror_count > 1 ? 1 : 0;
4835 /* create all underlying objects */
4836 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4837 lod_comp = &lo->ldo_comp_entries[i];
4839 if (lod_comp->llc_extent.e_start == 0 && i > 0) /* new mirror */
4842 if (lod_comp->llc_id == LCME_ID_INVAL) {
4843 lod_comp->llc_id = lod_gen_component_id(lo,
4845 if (lod_comp->llc_id == LCME_ID_INVAL)
4846 GOTO(out, rc = -ERANGE);
4849 if (lod_comp_inited(lod_comp))
4852 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
4853 lod_comp_set_init(lod_comp);
4855 if (lov_pattern(lod_comp->llc_pattern) == LOV_PATTERN_MDT)
4856 lod_comp_set_init(lod_comp);
4858 if (lod_comp->llc_stripe == NULL)
4861 LASSERT(lod_comp->llc_stripe_count);
4862 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4863 struct dt_object *object = lod_comp->llc_stripe[j];
4864 LASSERT(object != NULL);
4865 rc = lod_sub_create(env, object, attr, NULL, dof, th);
4869 lod_comp_set_init(lod_comp);
4872 rc = lod_fill_mirrors(lo);
4876 rc = lod_generate_and_set_lovea(env, lo, th);
4880 lo->ldo_comp_cached = 1;
4884 lod_object_free_striping(env, lo);
4889 * Implementation of dt_object_operations::do_create.
4891 * If any of preceeding methods (like ->do_declare_create(),
4892 * ->do_ah_init(), etc) chose to create a striped object,
4893 * then this method will create the master and the stripes.
4895 * \see dt_object_operations::do_create() in the API description for details.
4897 static int lod_create(const struct lu_env *env, struct dt_object *dt,
4898 struct lu_attr *attr, struct dt_allocation_hint *hint,
4899 struct dt_object_format *dof, struct thandle *th)
4904 /* create local object */
4905 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
4909 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
4910 lod_obj_is_striped(dt) && dof->u.dof_reg.striped != 0) {
4911 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
4912 rc = lod_striped_create(env, dt, attr, dof, th);
4919 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
4920 struct dt_object *dt, struct thandle *th,
4921 int comp_idx, int stripe_idx,
4922 struct lod_obj_stripe_cb_data *data)
4924 if (data->locd_declare)
4925 return lod_sub_declare_destroy(env, dt, th);
4926 else if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
4927 stripe_idx == cfs_fail_val)
4928 return lod_sub_destroy(env, dt, th);
4934 * Implementation of dt_object_operations::do_declare_destroy.
4936 * If the object is a striped directory, then the function declares reference
4937 * removal from the master object (this is an index) to the stripes and declares
4938 * destroy of all the stripes. In all the cases, it declares an intention to
4939 * destroy the object itself.
4941 * \see dt_object_operations::do_declare_destroy() in the API description
4944 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
4947 struct dt_object *next = dt_object_child(dt);
4948 struct lod_object *lo = lod_dt_obj(dt);
4949 struct lod_thread_info *info = lod_env_info(env);
4950 char *stripe_name = info->lti_key;
4955 * load striping information, notice we don't do this when object
4956 * is being initialized as we don't need this information till
4957 * few specific cases like destroy, chown
4959 rc = lod_load_striping(env, lo);
4963 /* declare destroy for all underlying objects */
4964 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
4965 rc = next->do_ops->do_index_try(env, next,
4966 &dt_directory_features);
4970 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4971 rc = lod_sub_declare_ref_del(env, next, th);
4975 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
4976 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
4978 rc = lod_sub_declare_delete(env, next,
4979 (const struct dt_key *)stripe_name, th);
4986 * we declare destroy for the local object
4988 rc = lod_sub_declare_destroy(env, next, th);
4992 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
4993 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
4996 if (!lod_obj_is_striped(dt))
4999 /* declare destroy all striped objects */
5000 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5001 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5002 if (lo->ldo_stripe[i] == NULL)
5005 rc = lod_sub_declare_ref_del(env, lo->ldo_stripe[i],
5008 rc = lod_sub_declare_destroy(env, lo->ldo_stripe[i],
5014 struct lod_obj_stripe_cb_data data = { { 0 } };
5016 data.locd_declare = true;
5017 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
5018 rc = lod_obj_for_each_stripe(env, lo, th, &data);
5025 * Implementation of dt_object_operations::do_destroy.
5027 * If the object is a striped directory, then the function removes references
5028 * from the master object (this is an index) to the stripes and destroys all
5029 * the stripes. In all the cases, the function destroys the object itself.
5031 * \see dt_object_operations::do_destroy() in the API description for details.
5033 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
5036 struct dt_object *next = dt_object_child(dt);
5037 struct lod_object *lo = lod_dt_obj(dt);
5038 struct lod_thread_info *info = lod_env_info(env);
5039 char *stripe_name = info->lti_key;
5044 /* destroy sub-stripe of master object */
5045 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5046 rc = next->do_ops->do_index_try(env, next,
5047 &dt_directory_features);
5051 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5052 rc = lod_sub_ref_del(env, next, th);
5056 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
5057 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
5060 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
5061 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
5062 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)));
5064 rc = lod_sub_delete(env, next,
5065 (const struct dt_key *)stripe_name, th);
5071 rc = lod_sub_destroy(env, next, th);
5075 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
5076 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
5079 if (!lod_obj_is_striped(dt))
5082 /* destroy all striped objects */
5083 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5084 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5085 if (lo->ldo_stripe[i] == NULL)
5087 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
5088 i == cfs_fail_val) {
5089 dt_write_lock(env, lo->ldo_stripe[i],
5091 rc = lod_sub_ref_del(env, lo->ldo_stripe[i],
5093 dt_write_unlock(env, lo->ldo_stripe[i]);
5097 rc = lod_sub_destroy(env, lo->ldo_stripe[i],
5104 struct lod_obj_stripe_cb_data data = { { 0 } };
5106 data.locd_declare = false;
5107 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
5108 rc = lod_obj_for_each_stripe(env, lo, th, &data);
5115 * Implementation of dt_object_operations::do_declare_ref_add.
5117 * \see dt_object_operations::do_declare_ref_add() in the API description
5120 static int lod_declare_ref_add(const struct lu_env *env,
5121 struct dt_object *dt, struct thandle *th)
5123 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
5127 * Implementation of dt_object_operations::do_ref_add.
5129 * \see dt_object_operations::do_ref_add() in the API description for details.
5131 static int lod_ref_add(const struct lu_env *env,
5132 struct dt_object *dt, struct thandle *th)
5134 return lod_sub_ref_add(env, dt_object_child(dt), th);
5138 * Implementation of dt_object_operations::do_declare_ref_del.
5140 * \see dt_object_operations::do_declare_ref_del() in the API description
5143 static int lod_declare_ref_del(const struct lu_env *env,
5144 struct dt_object *dt, struct thandle *th)
5146 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
5150 * Implementation of dt_object_operations::do_ref_del
5152 * \see dt_object_operations::do_ref_del() in the API description for details.
5154 static int lod_ref_del(const struct lu_env *env,
5155 struct dt_object *dt, struct thandle *th)
5157 return lod_sub_ref_del(env, dt_object_child(dt), th);
5161 * Implementation of dt_object_operations::do_object_sync.
5163 * \see dt_object_operations::do_object_sync() in the API description
5166 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
5167 __u64 start, __u64 end)
5169 return dt_object_sync(env, dt_object_child(dt), start, end);
5173 * Implementation of dt_object_operations::do_object_unlock.
5175 * Used to release LDLM lock(s).
5177 * \see dt_object_operations::do_object_unlock() in the API description
5180 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
5181 struct ldlm_enqueue_info *einfo,
5182 union ldlm_policy_data *policy)
5184 struct lod_object *lo = lod_dt_obj(dt);
5185 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
5186 int slave_locks_size;
5190 if (slave_locks == NULL)
5193 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
5194 LASSERT(lo->ldo_dir_stripe_count > 1);
5195 /* Note: for remote lock for single stripe dir, MDT will cancel
5196 * the lock by lockh directly */
5197 LASSERT(!dt_object_remote(dt_object_child(dt)));
5199 /* locks were unlocked in MDT layer */
5200 for (i = 0; i < slave_locks->ha_count; i++)
5201 LASSERT(!lustre_handle_is_used(&slave_locks->ha_handles[i]));
5204 * NB, ha_count may not equal to ldo_dir_stripe_count, because dir
5205 * layout may change, e.g., shrink dir layout after migration.
5207 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
5208 dt_invalidate(env, lo->ldo_stripe[i]);
5210 slave_locks_size = offsetof(typeof(*slave_locks),
5211 ha_handles[slave_locks->ha_count]);
5212 OBD_FREE(slave_locks, slave_locks_size);
5213 einfo->ei_cbdata = NULL;
5219 * Implementation of dt_object_operations::do_object_lock.
5221 * Used to get LDLM lock on the non-striped and striped objects.
5223 * \see dt_object_operations::do_object_lock() in the API description
5226 static int lod_object_lock(const struct lu_env *env,
5227 struct dt_object *dt,
5228 struct lustre_handle *lh,
5229 struct ldlm_enqueue_info *einfo,
5230 union ldlm_policy_data *policy)
5232 struct lod_object *lo = lod_dt_obj(dt);
5233 int slave_locks_size;
5234 struct lustre_handle_array *slave_locks = NULL;
5239 /* remote object lock */
5240 if (!einfo->ei_enq_slave) {
5241 LASSERT(dt_object_remote(dt));
5242 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
5246 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
5249 rc = lod_load_striping(env, lo);
5254 if (lo->ldo_dir_stripe_count <= 1)
5257 slave_locks_size = offsetof(typeof(*slave_locks),
5258 ha_handles[lo->ldo_dir_stripe_count]);
5259 /* Freed in lod_object_unlock */
5260 OBD_ALLOC(slave_locks, slave_locks_size);
5263 slave_locks->ha_count = lo->ldo_dir_stripe_count;
5265 /* striped directory lock */
5266 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5267 struct lustre_handle lockh;
5268 struct ldlm_res_id *res_id;
5270 res_id = &lod_env_info(env)->lti_res_id;
5271 fid_build_reg_res_name(lu_object_fid(&lo->ldo_stripe[i]->do_lu),
5273 einfo->ei_res_id = res_id;
5275 LASSERT(lo->ldo_stripe[i] != NULL);
5276 if (dt_object_remote(lo->ldo_stripe[i])) {
5277 set_bit(i, (void *)slave_locks->ha_map);
5278 rc = dt_object_lock(env, lo->ldo_stripe[i], &lockh,
5281 struct ldlm_namespace *ns = einfo->ei_namespace;
5282 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
5283 ldlm_completion_callback completion = einfo->ei_cb_cp;
5284 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
5286 if (einfo->ei_mode == LCK_PW ||
5287 einfo->ei_mode == LCK_EX)
5288 dlmflags |= LDLM_FL_COS_INCOMPAT;
5290 LASSERT(ns != NULL);
5291 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS,
5292 policy, einfo->ei_mode,
5293 &dlmflags, blocking,
5295 NULL, 0, LVB_T_NONE,
5300 ldlm_lock_decref_and_cancel(
5301 &slave_locks->ha_handles[i],
5303 OBD_FREE(slave_locks, slave_locks_size);
5306 slave_locks->ha_handles[i] = lockh;
5308 einfo->ei_cbdata = slave_locks;
5314 * Implementation of dt_object_operations::do_invalidate.
5316 * \see dt_object_operations::do_invalidate() in the API description for details
5318 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
5320 return dt_invalidate(env, dt_object_child(dt));
5323 static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
5327 /* clear memory region that will be used for layout change */
5328 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
5329 info->lti_count = 0;
5331 if (info->lti_comp_size >= comp_cnt)
5334 if (info->lti_comp_size > 0) {
5335 OBD_FREE(info->lti_comp_idx,
5336 info->lti_comp_size * sizeof(__u32));
5337 info->lti_comp_size = 0;
5340 OBD_ALLOC(info->lti_comp_idx, comp_cnt * sizeof(__u32));
5341 if (!info->lti_comp_idx)
5344 info->lti_comp_size = comp_cnt;
5348 static int lod_declare_instantiate_components(const struct lu_env *env,
5349 struct lod_object *lo, struct thandle *th)
5351 struct lod_thread_info *info = lod_env_info(env);
5352 struct ost_pool *inuse = &info->lti_inuse_osts;
5357 LASSERT(info->lti_count < lo->ldo_comp_cnt);
5358 if (info->lti_count > 0) {
5359 /* Prepare inuse array for composite file */
5360 rc = lod_prepare_inuse(env, lo);
5365 for (i = 0; i < info->lti_count; i++) {
5366 rc = lod_qos_prep_create(env, lo, NULL, th,
5367 info->lti_comp_idx[i], inuse);
5373 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
5374 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
5375 &info->lti_buf, XATTR_NAME_LOV, 0, th);
5381 static int lod_declare_update_plain(const struct lu_env *env,
5382 struct lod_object *lo, struct layout_intent *layout,
5383 const struct lu_buf *buf, struct thandle *th)
5385 struct lod_thread_info *info = lod_env_info(env);
5386 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5387 struct lod_layout_component *lod_comp;
5388 struct lov_comp_md_v1 *comp_v1 = NULL;
5389 bool replay = false;
5393 LASSERT(lo->ldo_flr_state == LCM_FL_NONE);
5396 * In case the client is passing lovea, which only happens during
5397 * the replay of layout intent write RPC for now, we may need to
5398 * parse the lovea and apply new layout configuration.
5400 if (buf && buf->lb_len) {
5401 struct lov_user_md_v1 *v1 = buf->lb_buf;
5403 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
5404 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
5405 LOV_MAGIC_COMP_V1)) {
5406 CERROR("%s: the replay buffer of layout extend "
5407 "(magic %#x) does not contain expected "
5408 "composite layout.\n",
5409 lod2obd(d)->obd_name, v1->lmm_magic);
5410 GOTO(out, rc = -EINVAL);
5413 lod_object_free_striping(env, lo);
5414 rc = lod_use_defined_striping(env, lo, buf);
5418 rc = lod_get_lov_ea(env, lo);
5421 /* old on-disk EA is stored in info->lti_buf */
5422 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
5425 /* non replay path */
5426 rc = lod_load_striping_locked(env, lo);
5431 /* Make sure defined layout covers the requested write range. */
5432 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
5433 if (lo->ldo_comp_cnt > 1 &&
5434 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
5435 lod_comp->llc_extent.e_end < layout->li_extent.e_end) {
5436 CDEBUG(replay ? D_ERROR : D_LAYOUT,
5437 "%s: the defined layout [0, %#llx) does not covers "
5438 "the write range "DEXT"\n",
5439 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
5440 PEXT(&layout->li_extent));
5441 GOTO(out, rc = -EINVAL);
5444 CDEBUG(D_LAYOUT, "%s: "DFID": instantiate components "DEXT"\n",
5445 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
5446 PEXT(&layout->li_extent));
5449 * Iterate ld->ldo_comp_entries, find the component whose extent under
5450 * the write range and not instantianted.
5452 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5453 lod_comp = &lo->ldo_comp_entries[i];
5455 if (lod_comp->llc_extent.e_start >= layout->li_extent.e_end)
5459 if (lod_comp_inited(lod_comp))
5463 * In replay path, lod_comp is the EA passed by
5464 * client replay buffer, comp_v1 is the pre-recovery
5465 * on-disk EA, we'd sift out those components which
5466 * were init-ed in the on-disk EA.
5468 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
5473 * this component hasn't instantiated in normal path, or during
5474 * replay it needs replay the instantiation.
5477 /* A released component is being extended */
5478 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
5479 GOTO(out, rc = -EINVAL);
5481 LASSERT(info->lti_comp_idx != NULL);
5482 info->lti_comp_idx[info->lti_count++] = i;
5485 if (info->lti_count == 0)
5488 lod_obj_inc_layout_gen(lo);
5489 rc = lod_declare_instantiate_components(env, lo, th);
5492 lod_object_free_striping(env, lo);
5496 static inline int lod_comp_index(struct lod_object *lo,
5497 struct lod_layout_component *lod_comp)
5499 LASSERT(lod_comp >= lo->ldo_comp_entries &&
5500 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
5502 return lod_comp - lo->ldo_comp_entries;
5506 * Stale other mirrors by writing extent.
5508 static void lod_stale_components(struct lod_object *lo, int primary,
5509 struct lu_extent *extent)
5511 struct lod_layout_component *pri_comp, *lod_comp;
5514 /* The writing extent decides which components in the primary
5515 * are affected... */
5516 CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
5517 lod_foreach_mirror_comp(pri_comp, lo, primary) {
5518 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
5521 CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
5522 lod_comp_index(lo, pri_comp),
5523 PEXT(&pri_comp->llc_extent));
5525 for (i = 0; i < lo->ldo_mirror_count; i++) {
5529 /* ... and then stale other components that are
5530 * overlapping with primary components */
5531 lod_foreach_mirror_comp(lod_comp, lo, i) {
5532 if (!lu_extent_is_overlapped(
5533 &pri_comp->llc_extent,
5534 &lod_comp->llc_extent))
5537 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
5538 i, lod_comp_index(lo, lod_comp));
5540 lod_comp->llc_flags |= LCME_FL_STALE;
5541 lo->ldo_mirrors[i].lme_stale = 1;
5548 * check an OST's availability
5549 * \param[in] env execution environment
5550 * \param[in] lo lod object
5551 * \param[in] dt dt object
5552 * \param[in] index mirror index
5554 * \retval negative if failed
5555 * \retval 1 if \a dt is available
5556 * \retval 0 if \a dt is not available
5558 static inline int lod_check_ost_avail(const struct lu_env *env,
5559 struct lod_object *lo,
5560 struct dt_object *dt, int index)
5562 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5563 struct lod_tgt_desc *ost;
5565 int type = LU_SEQ_RANGE_OST;
5568 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &idx, &type);
5570 CERROR("%s: can't locate "DFID":rc = %d\n",
5571 lod2obd(lod)->obd_name, PFID(lu_object_fid(&dt->do_lu)),
5576 ost = OST_TGT(lod, idx);
5577 if (ost->ltd_statfs.os_state &
5578 (OS_STATE_READONLY | OS_STATE_ENOSPC | OS_STATE_ENOINO |
5579 OS_STATE_NOPRECREATE) ||
5580 ost->ltd_active == 0) {
5581 CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail, rc = %d\n",
5582 PFID(lod_object_fid(lo)), index, idx, rc);
5590 * Pick primary mirror for write
5591 * \param[in] env execution environment
5592 * \param[in] lo object
5593 * \param[in] extent write range
5595 static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo,
5596 struct lu_extent *extent)
5598 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5599 unsigned int seq = 0;
5600 struct lod_layout_component *lod_comp;
5602 int picked = -1, second_pick = -1, third_pick = -1;
5605 if (OBD_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
5606 get_random_bytes(&seq, sizeof(seq));
5607 seq %= lo->ldo_mirror_count;
5611 * Pick a mirror as the primary, and check the availability of OSTs.
5613 * This algo can be revised later after knowing the topology of
5616 lod_qos_statfs_update(env, lod);
5617 for (i = 0; i < lo->ldo_mirror_count; i++) {
5618 bool ost_avail = true;
5619 int index = (i + seq) % lo->ldo_mirror_count;
5621 if (lo->ldo_mirrors[index].lme_stale) {
5622 CDEBUG(D_LAYOUT, DFID": mirror %d stale\n",
5623 PFID(lod_object_fid(lo)), index);
5627 /* 2nd pick is for the primary mirror containing unavail OST */
5628 if (lo->ldo_mirrors[index].lme_primary && second_pick < 0)
5629 second_pick = index;
5631 /* 3rd pick is for non-primary mirror containing unavail OST */
5632 if (second_pick < 0 && third_pick < 0)
5636 * we found a non-primary 1st pick, we'd like to find a
5637 * potential pirmary mirror.
5639 if (picked >= 0 && !lo->ldo_mirrors[index].lme_primary)
5642 /* check the availability of OSTs */
5643 lod_foreach_mirror_comp(lod_comp, lo, index) {
5644 if (!lod_comp_inited(lod_comp) || !lod_comp->llc_stripe)
5647 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
5648 struct dt_object *dt = lod_comp->llc_stripe[j];
5650 rc = lod_check_ost_avail(env, lo, dt, index);
5657 } /* for all dt object in one component */
5660 } /* for all components in a mirror */
5663 * the OSTs where allocated objects locates in the components
5664 * of the mirror are available.
5669 /* this mirror has all OSTs available */
5673 * primary with all OSTs are available, this is the perfect
5676 if (lo->ldo_mirrors[index].lme_primary)
5678 } /* for all mirrors */
5680 /* failed to pick a sound mirror, lower our expectation */
5682 picked = second_pick;
5684 picked = third_pick;
5692 * figure out the components should be instantiated for resync.
5694 static int lod_prepare_resync(const struct lu_env *env, struct lod_object *lo,
5695 struct lu_extent *extent)
5697 struct lod_thread_info *info = lod_env_info(env);
5698 struct lod_layout_component *lod_comp;
5699 unsigned int need_sync = 0;
5703 DFID": instantiate all stale components in "DEXT"\n",
5704 PFID(lod_object_fid(lo)), PEXT(extent));
5707 * instantiate all components within this extent, even non-stale
5710 for (i = 0; i < lo->ldo_mirror_count; i++) {
5711 if (!lo->ldo_mirrors[i].lme_stale)
5714 lod_foreach_mirror_comp(lod_comp, lo, i) {
5715 if (!lu_extent_is_overlapped(extent,
5716 &lod_comp->llc_extent))
5721 if (lod_comp_inited(lod_comp))
5724 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
5725 i, lod_comp_index(lo, lod_comp));
5726 info->lti_comp_idx[info->lti_count++] =
5727 lod_comp_index(lo, lod_comp);
5731 return need_sync ? 0 : -EALREADY;
5734 static int lod_declare_update_rdonly(const struct lu_env *env,
5735 struct lod_object *lo, struct md_layout_change *mlc,
5738 struct lod_thread_info *info = lod_env_info(env);
5739 struct lu_attr *layout_attr = &info->lti_layout_attr;
5740 struct lod_layout_component *lod_comp;
5741 struct lu_extent extent = { 0 };
5745 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
5746 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
5747 mlc->mlc_opc == MD_LAYOUT_RESYNC);
5748 LASSERT(lo->ldo_mirror_count > 0);
5750 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
5751 struct layout_intent *layout = mlc->mlc_intent;
5754 extent = layout->li_extent;
5755 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
5756 PFID(lod_object_fid(lo)), PEXT(&extent));
5758 picked = lod_primary_pick(env, lo, &extent);
5762 CDEBUG(D_LAYOUT, DFID": picked mirror id %u as primary\n",
5763 PFID(lod_object_fid(lo)),
5764 lo->ldo_mirrors[picked].lme_id);
5766 if (layout->li_opc == LAYOUT_INTENT_TRUNC) {
5768 * trunc transfers [0, size) in the intent extent, we'd
5769 * stale components overlapping [size, eof).
5771 extent.e_start = extent.e_end;
5772 extent.e_end = OBD_OBJECT_EOF;
5775 /* stale overlapping components from other mirrors */
5776 lod_stale_components(lo, picked, &extent);
5778 /* restore truncate intent extent */
5779 if (layout->li_opc == LAYOUT_INTENT_TRUNC)
5780 extent.e_end = extent.e_start;
5782 /* instantiate components for the picked mirror, start from 0 */
5785 lod_foreach_mirror_comp(lod_comp, lo, picked) {
5786 if (!lu_extent_is_overlapped(&extent,
5787 &lod_comp->llc_extent))
5790 if (lod_comp_inited(lod_comp))
5793 info->lti_comp_idx[info->lti_count++] =
5794 lod_comp_index(lo, lod_comp);
5797 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
5798 } else { /* MD_LAYOUT_RESYNC */
5802 * could contain multiple non-stale mirrors, so we need to
5803 * prep uninited all components assuming any non-stale mirror
5804 * could be picked as the primary mirror.
5806 for (i = 0; i < lo->ldo_mirror_count; i++) {
5807 if (lo->ldo_mirrors[i].lme_stale)
5810 lod_foreach_mirror_comp(lod_comp, lo, i) {
5811 if (!lod_comp_inited(lod_comp))
5814 if (extent.e_end < lod_comp->llc_extent.e_end)
5816 lod_comp->llc_extent.e_end;
5820 rc = lod_prepare_resync(env, lo, &extent);
5823 /* change the file state to SYNC_PENDING */
5824 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
5827 /* Reset the layout version once it's becoming too large.
5828 * This way it can make sure that the layout version is
5829 * monotonously increased in this writing era. */
5830 lod_obj_inc_layout_gen(lo);
5831 if (lo->ldo_layout_gen > (LCME_ID_MAX >> 1)) {
5832 __u32 layout_version;
5834 cfs_get_random_bytes(&layout_version, sizeof(layout_version));
5835 lo->ldo_layout_gen = layout_version & 0xffff;
5838 rc = lod_declare_instantiate_components(env, lo, th);
5842 layout_attr->la_valid = LA_LAYOUT_VERSION;
5843 layout_attr->la_layout_version = 0; /* set current version */
5844 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
5845 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
5846 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
5852 lod_object_free_striping(env, lo);
5856 static int lod_declare_update_write_pending(const struct lu_env *env,
5857 struct lod_object *lo, struct md_layout_change *mlc,
5860 struct lod_thread_info *info = lod_env_info(env);
5861 struct lu_attr *layout_attr = &info->lti_layout_attr;
5862 struct lod_layout_component *lod_comp;
5863 struct lu_extent extent = { 0 };
5869 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
5870 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
5871 mlc->mlc_opc == MD_LAYOUT_RESYNC);
5873 /* look for the primary mirror */
5874 for (i = 0; i < lo->ldo_mirror_count; i++) {
5875 if (lo->ldo_mirrors[i].lme_stale)
5878 LASSERTF(primary < 0, DFID " has multiple primary: %u / %u",
5879 PFID(lod_object_fid(lo)),
5880 lo->ldo_mirrors[i].lme_id,
5881 lo->ldo_mirrors[primary].lme_id);
5886 CERROR(DFID ": doesn't have a primary mirror\n",
5887 PFID(lod_object_fid(lo)));
5888 GOTO(out, rc = -ENODATA);
5891 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
5892 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
5894 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
5896 /* for LAYOUT_WRITE opc, it has to do the following operations:
5897 * 1. stale overlapping componets from stale mirrors;
5898 * 2. instantiate components of the primary mirror;
5899 * 3. transfter layout version to all objects of the primary;
5901 * for LAYOUT_RESYNC opc, it will do:
5902 * 1. instantiate components of all stale mirrors;
5903 * 2. transfer layout version to all objects to close write era. */
5905 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
5906 LASSERT(mlc->mlc_intent != NULL);
5908 extent = mlc->mlc_intent->li_extent;
5910 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
5911 PFID(lod_object_fid(lo)), PEXT(&extent));
5913 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC) {
5915 * trunc transfers [0, size) in the intent extent, we'd
5916 * stale components overlapping [size, eof).
5918 extent.e_start = extent.e_end;
5919 extent.e_end = OBD_OBJECT_EOF;
5921 /* 1. stale overlapping components */
5922 lod_stale_components(lo, primary, &extent);
5924 /* 2. find out the components need instantiating.
5925 * instantiate [0, mlc->mlc_intent->e_end) */
5927 /* restore truncate intent extent */
5928 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC)
5929 extent.e_end = extent.e_start;
5932 lod_foreach_mirror_comp(lod_comp, lo, primary) {
5933 if (!lu_extent_is_overlapped(&extent,
5934 &lod_comp->llc_extent))
5937 if (lod_comp_inited(lod_comp))
5940 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
5941 primary, lod_comp_index(lo, lod_comp));
5942 info->lti_comp_idx[info->lti_count++] =
5943 lod_comp_index(lo, lod_comp);
5945 } else { /* MD_LAYOUT_RESYNC */
5946 lod_foreach_mirror_comp(lod_comp, lo, primary) {
5947 if (!lod_comp_inited(lod_comp))
5950 extent.e_end = lod_comp->llc_extent.e_end;
5953 rc = lod_prepare_resync(env, lo, &extent);
5956 /* change the file state to SYNC_PENDING */
5957 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
5960 rc = lod_declare_instantiate_components(env, lo, th);
5964 /* 3. transfer layout version to OST objects.
5965 * transfer new layout version to OST objects so that stale writes
5966 * can be denied. It also ends an era of writing by setting
5967 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
5968 * send write RPC; only resync RPCs could do it. */
5969 layout_attr->la_valid = LA_LAYOUT_VERSION;
5970 layout_attr->la_layout_version = 0; /* set current version */
5971 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
5972 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
5973 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
5977 lod_obj_inc_layout_gen(lo);
5980 lod_object_free_striping(env, lo);
5984 static int lod_declare_update_sync_pending(const struct lu_env *env,
5985 struct lod_object *lo, struct md_layout_change *mlc,
5988 struct lod_thread_info *info = lod_env_info(env);
5989 unsigned sync_components = 0;
5990 unsigned resync_components = 0;
5995 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
5996 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
5997 mlc->mlc_opc == MD_LAYOUT_WRITE);
5999 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
6000 PFID(lod_object_fid(lo)), mlc->mlc_opc);
6002 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
6003 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
6004 PFID(lod_object_fid(lo)));
6006 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
6007 return lod_declare_update_write_pending(env, lo, mlc, th);
6010 /* MD_LAYOUT_RESYNC_DONE */
6012 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6013 struct lod_layout_component *lod_comp;
6016 lod_comp = &lo->ldo_comp_entries[i];
6018 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
6023 for (j = 0; j < mlc->mlc_resync_count; j++) {
6024 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
6027 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
6028 lod_comp->llc_flags &= ~LCME_FL_STALE;
6029 resync_components++;
6035 for (i = 0; i < mlc->mlc_resync_count; i++) {
6036 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
6039 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
6040 "or already synced\n", PFID(lod_object_fid(lo)),
6041 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
6042 GOTO(out, rc = -EINVAL);
6045 if (!sync_components || (mlc->mlc_resync_count && !resync_components)) {
6046 CDEBUG(D_LAYOUT, DFID": no mirror in sync\n",
6047 PFID(lod_object_fid(lo)));
6049 /* tend to return an error code here to prevent
6050 * the MDT from setting SoM attribute */
6051 GOTO(out, rc = -EINVAL);
6054 CDEBUG(D_LAYOUT, DFID": resynced %u/%zu components\n",
6055 PFID(lod_object_fid(lo)),
6056 resync_components, mlc->mlc_resync_count);
6058 lo->ldo_flr_state = LCM_FL_RDONLY;
6059 lod_obj_inc_layout_gen(lo);
6061 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6062 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
6063 &info->lti_buf, XATTR_NAME_LOV, 0, th);
6068 lod_object_free_striping(env, lo);
6072 static int lod_declare_layout_change(const struct lu_env *env,
6073 struct dt_object *dt, struct md_layout_change *mlc,
6076 struct lod_thread_info *info = lod_env_info(env);
6077 struct lod_object *lo = lod_dt_obj(dt);
6081 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
6082 dt_object_remote(dt_object_child(dt)))
6085 lod_write_lock(env, dt, 0);
6086 rc = lod_load_striping_locked(env, lo);
6090 LASSERT(lo->ldo_comp_cnt > 0);
6092 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
6096 switch (lo->ldo_flr_state) {
6098 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
6102 rc = lod_declare_update_rdonly(env, lo, mlc, th);
6104 case LCM_FL_WRITE_PENDING:
6105 rc = lod_declare_update_write_pending(env, lo, mlc, th);
6107 case LCM_FL_SYNC_PENDING:
6108 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
6115 dt_write_unlock(env, dt);
6120 * Instantiate layout component objects which covers the intent write offset.
6122 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
6123 struct md_layout_change *mlc, struct thandle *th)
6125 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
6126 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
6127 struct lod_object *lo = lod_dt_obj(dt);
6130 rc = lod_striped_create(env, dt, attr, NULL, th);
6131 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
6132 layout_attr->la_layout_version |= lo->ldo_layout_gen;
6133 rc = lod_attr_set(env, dt, layout_attr, th);
6139 struct dt_object_operations lod_obj_ops = {
6140 .do_read_lock = lod_read_lock,
6141 .do_write_lock = lod_write_lock,
6142 .do_read_unlock = lod_read_unlock,
6143 .do_write_unlock = lod_write_unlock,
6144 .do_write_locked = lod_write_locked,
6145 .do_attr_get = lod_attr_get,
6146 .do_declare_attr_set = lod_declare_attr_set,
6147 .do_attr_set = lod_attr_set,
6148 .do_xattr_get = lod_xattr_get,
6149 .do_declare_xattr_set = lod_declare_xattr_set,
6150 .do_xattr_set = lod_xattr_set,
6151 .do_declare_xattr_del = lod_declare_xattr_del,
6152 .do_xattr_del = lod_xattr_del,
6153 .do_xattr_list = lod_xattr_list,
6154 .do_ah_init = lod_ah_init,
6155 .do_declare_create = lod_declare_create,
6156 .do_create = lod_create,
6157 .do_declare_destroy = lod_declare_destroy,
6158 .do_destroy = lod_destroy,
6159 .do_index_try = lod_index_try,
6160 .do_declare_ref_add = lod_declare_ref_add,
6161 .do_ref_add = lod_ref_add,
6162 .do_declare_ref_del = lod_declare_ref_del,
6163 .do_ref_del = lod_ref_del,
6164 .do_object_sync = lod_object_sync,
6165 .do_object_lock = lod_object_lock,
6166 .do_object_unlock = lod_object_unlock,
6167 .do_invalidate = lod_invalidate,
6168 .do_declare_layout_change = lod_declare_layout_change,
6169 .do_layout_change = lod_layout_change,
6173 * Implementation of dt_body_operations::dbo_read.
6175 * \see dt_body_operations::dbo_read() in the API description for details.
6177 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
6178 struct lu_buf *buf, loff_t *pos)
6180 struct dt_object *next = dt_object_child(dt);
6182 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
6183 S_ISLNK(dt->do_lu.lo_header->loh_attr));
6184 return next->do_body_ops->dbo_read(env, next, buf, pos);
6188 * Implementation of dt_body_operations::dbo_declare_write.
6190 * \see dt_body_operations::dbo_declare_write() in the API description
6193 static ssize_t lod_declare_write(const struct lu_env *env,
6194 struct dt_object *dt,
6195 const struct lu_buf *buf, loff_t pos,
6198 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
6202 * Implementation of dt_body_operations::dbo_write.
6204 * \see dt_body_operations::dbo_write() in the API description for details.
6206 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
6207 const struct lu_buf *buf, loff_t *pos,
6208 struct thandle *th, int iq)
6210 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
6211 S_ISLNK(dt->do_lu.lo_header->loh_attr));
6212 return lod_sub_write(env, dt_object_child(dt), buf, pos, th, iq);
6215 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
6216 __u64 start, __u64 end, struct thandle *th)
6218 if (dt_object_remote(dt))
6221 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
6224 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
6225 __u64 start, __u64 end, struct thandle *th)
6227 if (dt_object_remote(dt))
6230 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
6231 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
6235 * different type of files use the same body_ops because object may be created
6236 * in OUT, where there is no chance to set correct body_ops for each type, so
6237 * body_ops themselves will check file type inside, see lod_read/write/punch for
6240 const struct dt_body_operations lod_body_ops = {
6241 .dbo_read = lod_read,
6242 .dbo_declare_write = lod_declare_write,
6243 .dbo_write = lod_write,
6244 .dbo_declare_punch = lod_declare_punch,
6245 .dbo_punch = lod_punch,
6249 * Implementation of lu_object_operations::loo_object_init.
6251 * The function determines the type and the index of the target device using
6252 * sequence of the object's FID. Then passes control down to the
6253 * corresponding device:
6254 * OSD for the local objects, OSP for remote
6256 * \see lu_object_operations::loo_object_init() in the API description
6259 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
6260 const struct lu_object_conf *conf)
6262 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
6263 struct lu_device *cdev = NULL;
6264 struct lu_object *cobj;
6265 struct lod_tgt_descs *ltd = NULL;
6266 struct lod_tgt_desc *tgt;
6268 int type = LU_SEQ_RANGE_ANY;
6272 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
6274 /* Note: Sometimes, it will Return EAGAIN here, see
6275 * ptrlpc_import_delay_req(), which might confuse
6276 * lu_object_find_at() and make it wait there incorrectly.
6277 * so we convert it to EIO here.*/
6284 if (type == LU_SEQ_RANGE_MDT &&
6285 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
6286 cdev = &lod->lod_child->dd_lu_dev;
6287 } else if (type == LU_SEQ_RANGE_MDT) {
6288 ltd = &lod->lod_mdt_descs;
6290 } else if (type == LU_SEQ_RANGE_OST) {
6291 ltd = &lod->lod_ost_descs;
6298 if (ltd->ltd_tgts_size > idx &&
6299 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
6300 tgt = LTD_TGT(ltd, idx);
6302 LASSERT(tgt != NULL);
6303 LASSERT(tgt->ltd_tgt != NULL);
6305 cdev = &(tgt->ltd_tgt->dd_lu_dev);
6307 lod_putref(lod, ltd);
6310 if (unlikely(cdev == NULL))
6313 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
6314 if (unlikely(cobj == NULL))
6317 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
6319 lu_object_add(lo, cobj);
6326 * Release resources associated with striping.
6328 * If the object is striped (regular or directory), then release
6329 * the stripe objects references and free the ldo_stripe array.
6331 * \param[in] env execution environment
6332 * \param[in] lo object
6334 void lod_object_free_striping(const struct lu_env *env, struct lod_object *lo)
6336 struct lod_layout_component *lod_comp;
6339 if (lo->ldo_stripe != NULL) {
6340 LASSERT(lo->ldo_comp_entries == NULL);
6341 LASSERT(lo->ldo_dir_stripes_allocated > 0);
6343 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6344 if (lo->ldo_stripe[i])
6345 dt_object_put(env, lo->ldo_stripe[i]);
6348 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
6349 OBD_FREE(lo->ldo_stripe, j);
6350 lo->ldo_stripe = NULL;
6351 lo->ldo_dir_stripes_allocated = 0;
6352 lo->ldo_dir_stripe_loaded = 0;
6353 lo->ldo_dir_stripe_count = 0;
6354 } else if (lo->ldo_comp_entries != NULL) {
6355 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6356 /* free lod_layout_component::llc_stripe array */
6357 lod_comp = &lo->ldo_comp_entries[i];
6359 if (lod_comp->llc_stripe == NULL)
6361 LASSERT(lod_comp->llc_stripes_allocated != 0);
6362 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
6363 if (lod_comp->llc_stripe[j] != NULL)
6365 &lod_comp->llc_stripe[j]->do_lu);
6367 OBD_FREE(lod_comp->llc_stripe,
6368 sizeof(struct dt_object *) *
6369 lod_comp->llc_stripes_allocated);
6370 lod_comp->llc_stripe = NULL;
6371 OBD_FREE(lod_comp->llc_ost_indices,
6373 lod_comp->llc_stripes_allocated);
6374 lod_comp->llc_ost_indices = NULL;
6375 lod_comp->llc_stripes_allocated = 0;
6377 lod_free_comp_entries(lo);
6378 lo->ldo_comp_cached = 0;
6383 * Implementation of lu_object_operations::loo_object_free.
6385 * \see lu_object_operations::loo_object_free() in the API description
6388 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
6390 struct lod_object *lo = lu2lod_obj(o);
6392 /* release all underlying object pinned */
6393 lod_object_free_striping(env, lo);
6395 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
6399 * Implementation of lu_object_operations::loo_object_release.
6401 * \see lu_object_operations::loo_object_release() in the API description
6404 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
6406 /* XXX: shouldn't we release everything here in case if object
6407 * creation failed before? */
6411 * Implementation of lu_object_operations::loo_object_print.
6413 * \see lu_object_operations::loo_object_print() in the API description
6416 static int lod_object_print(const struct lu_env *env, void *cookie,
6417 lu_printer_t p, const struct lu_object *l)
6419 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
6421 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
6424 struct lu_object_operations lod_lu_obj_ops = {
6425 .loo_object_init = lod_object_init,
6426 .loo_object_free = lod_object_free,
6427 .loo_object_release = lod_object_release,
6428 .loo_object_print = lod_object_print,