4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * lustre/lod/lod_object.c
31 * This file contains implementations of methods for the OSD API
32 * for the Logical Object Device (LOD) layer, which provides a virtual
33 * local OSD object interface to the MDD layer, and abstracts the
34 * addressing of local (OSD) and remote (OSP) objects. The API is
35 * described in the file lustre/include/dt_object.h and in
36 * Documentation/osd-api.txt.
38 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
41 #define DEBUG_SUBSYSTEM S_MDS
43 #include <linux/random.h>
46 #include <obd_class.h>
47 #include <obd_support.h>
49 #include <lustre_fid.h>
50 #include <lustre_linkea.h>
51 #include <lustre_lmv.h>
52 #include <uapi/linux/lustre/lustre_param.h>
53 #include <lustre_swab.h>
54 #include <uapi/linux/lustre/lustre_ver.h>
55 #include <lprocfs_status.h>
56 #include <md_object.h>
58 #include "lod_internal.h"
60 static const char dot[] = ".";
61 static const char dotdot[] = "..";
64 * Implementation of dt_index_operations::dio_lookup
66 * Used with regular (non-striped) objects.
68 * \see dt_index_operations::dio_lookup() in the API description for details.
70 static int lod_lookup(const struct lu_env *env, struct dt_object *dt,
71 struct dt_rec *rec, const struct dt_key *key)
73 struct dt_object *next = dt_object_child(dt);
74 return next->do_index_ops->dio_lookup(env, next, rec, key);
78 * Implementation of dt_index_operations::dio_declare_insert.
80 * Used with regular (non-striped) objects.
82 * \see dt_index_operations::dio_declare_insert() in the API description
85 static int lod_declare_insert(const struct lu_env *env, struct dt_object *dt,
86 const struct dt_rec *rec,
87 const struct dt_key *key, struct thandle *th)
89 return lod_sub_declare_insert(env, dt_object_child(dt), rec, key, th);
93 * Implementation of dt_index_operations::dio_insert.
95 * Used with regular (non-striped) objects
97 * \see dt_index_operations::dio_insert() in the API description for details.
99 static int lod_insert(const struct lu_env *env, struct dt_object *dt,
100 const struct dt_rec *rec, const struct dt_key *key,
103 return lod_sub_insert(env, dt_object_child(dt), rec, key, th);
107 * Implementation of dt_index_operations::dio_declare_delete.
109 * Used with regular (non-striped) objects.
111 * \see dt_index_operations::dio_declare_delete() in the API description
114 static int lod_declare_delete(const struct lu_env *env, struct dt_object *dt,
115 const struct dt_key *key, struct thandle *th)
117 return lod_sub_declare_delete(env, dt_object_child(dt), key, th);
121 * Implementation of dt_index_operations::dio_delete.
123 * Used with regular (non-striped) objects.
125 * \see dt_index_operations::dio_delete() in the API description for details.
127 static int lod_delete(const struct lu_env *env, struct dt_object *dt,
128 const struct dt_key *key, struct thandle *th)
130 return lod_sub_delete(env, dt_object_child(dt), key, th);
134 * Implementation of dt_it_ops::init.
136 * Used with regular (non-striped) objects.
138 * \see dt_it_ops::init() in the API description for details.
140 static struct dt_it *lod_it_init(const struct lu_env *env,
141 struct dt_object *dt, __u32 attr)
143 struct dt_object *next = dt_object_child(dt);
144 struct lod_it *it = &lod_env_info(env)->lti_it;
145 struct dt_it *it_next;
147 it_next = next->do_index_ops->dio_it.init(env, next, attr);
151 /* currently we do not use more than one iterator per thread
152 * so we store it in thread info. if at some point we need
153 * more active iterators in a single thread, we can allocate
155 LASSERT(it->lit_obj == NULL);
157 it->lit_it = it_next;
160 return (struct dt_it *)it;
163 #define LOD_CHECK_IT(env, it) \
165 LASSERT((it)->lit_obj != NULL); \
166 LASSERT((it)->lit_it != NULL); \
170 * Implementation of dt_index_operations::dio_it.fini.
172 * Used with regular (non-striped) objects.
174 * \see dt_index_operations::dio_it.fini() in the API description for details.
176 static void lod_it_fini(const struct lu_env *env, struct dt_it *di)
178 struct lod_it *it = (struct lod_it *)di;
180 LOD_CHECK_IT(env, it);
181 it->lit_obj->do_index_ops->dio_it.fini(env, it->lit_it);
183 /* the iterator not in use any more */
189 * Implementation of dt_it_ops::get.
191 * Used with regular (non-striped) objects.
193 * \see dt_it_ops::get() in the API description for details.
195 static int lod_it_get(const struct lu_env *env, struct dt_it *di,
196 const struct dt_key *key)
198 const struct lod_it *it = (const struct lod_it *)di;
200 LOD_CHECK_IT(env, it);
201 return it->lit_obj->do_index_ops->dio_it.get(env, it->lit_it, key);
205 * Implementation of dt_it_ops::put.
207 * Used with regular (non-striped) objects.
209 * \see dt_it_ops::put() in the API description for details.
211 static void lod_it_put(const struct lu_env *env, struct dt_it *di)
213 struct lod_it *it = (struct lod_it *)di;
215 LOD_CHECK_IT(env, it);
216 return it->lit_obj->do_index_ops->dio_it.put(env, it->lit_it);
220 * Implementation of dt_it_ops::next.
222 * Used with regular (non-striped) objects
224 * \see dt_it_ops::next() in the API description for details.
226 static int lod_it_next(const struct lu_env *env, struct dt_it *di)
228 struct lod_it *it = (struct lod_it *)di;
230 LOD_CHECK_IT(env, it);
231 return it->lit_obj->do_index_ops->dio_it.next(env, it->lit_it);
235 * Implementation of dt_it_ops::key.
237 * Used with regular (non-striped) objects.
239 * \see dt_it_ops::key() in the API description for details.
241 static struct dt_key *lod_it_key(const struct lu_env *env,
242 const struct dt_it *di)
244 const struct lod_it *it = (const struct lod_it *)di;
246 LOD_CHECK_IT(env, it);
247 return it->lit_obj->do_index_ops->dio_it.key(env, it->lit_it);
251 * Implementation of dt_it_ops::key_size.
253 * Used with regular (non-striped) objects.
255 * \see dt_it_ops::key_size() in the API description for details.
257 static int lod_it_key_size(const struct lu_env *env, const struct dt_it *di)
259 struct lod_it *it = (struct lod_it *)di;
261 LOD_CHECK_IT(env, it);
262 return it->lit_obj->do_index_ops->dio_it.key_size(env, it->lit_it);
266 * Implementation of dt_it_ops::rec.
268 * Used with regular (non-striped) objects.
270 * \see dt_it_ops::rec() in the API description for details.
272 static int lod_it_rec(const struct lu_env *env, const struct dt_it *di,
273 struct dt_rec *rec, __u32 attr)
275 const struct lod_it *it = (const struct lod_it *)di;
277 LOD_CHECK_IT(env, it);
278 return it->lit_obj->do_index_ops->dio_it.rec(env, it->lit_it, rec,
283 * Implementation of dt_it_ops::rec_size.
285 * Used with regular (non-striped) objects.
287 * \see dt_it_ops::rec_size() in the API description for details.
289 static int lod_it_rec_size(const struct lu_env *env, const struct dt_it *di,
292 const struct lod_it *it = (const struct lod_it *)di;
294 LOD_CHECK_IT(env, it);
295 return it->lit_obj->do_index_ops->dio_it.rec_size(env, it->lit_it,
300 * Implementation of dt_it_ops::store.
302 * Used with regular (non-striped) objects.
304 * \see dt_it_ops::store() in the API description for details.
306 static __u64 lod_it_store(const struct lu_env *env, const struct dt_it *di)
308 const struct lod_it *it = (const struct lod_it *)di;
310 LOD_CHECK_IT(env, it);
311 return it->lit_obj->do_index_ops->dio_it.store(env, it->lit_it);
315 * Implementation of dt_it_ops::load.
317 * Used with regular (non-striped) objects.
319 * \see dt_it_ops::load() in the API description for details.
321 static int lod_it_load(const struct lu_env *env, const struct dt_it *di,
324 const struct lod_it *it = (const struct lod_it *)di;
326 LOD_CHECK_IT(env, it);
327 return it->lit_obj->do_index_ops->dio_it.load(env, it->lit_it, hash);
331 * Implementation of dt_it_ops::key_rec.
333 * Used with regular (non-striped) objects.
335 * \see dt_it_ops::rec() in the API description for details.
337 static int lod_it_key_rec(const struct lu_env *env, const struct dt_it *di,
340 const struct lod_it *it = (const struct lod_it *)di;
342 LOD_CHECK_IT(env, it);
343 return it->lit_obj->do_index_ops->dio_it.key_rec(env, it->lit_it,
347 static struct dt_index_operations lod_index_ops = {
348 .dio_lookup = lod_lookup,
349 .dio_declare_insert = lod_declare_insert,
350 .dio_insert = lod_insert,
351 .dio_declare_delete = lod_declare_delete,
352 .dio_delete = lod_delete,
360 .key_size = lod_it_key_size,
362 .rec_size = lod_it_rec_size,
363 .store = lod_it_store,
365 .key_rec = lod_it_key_rec,
370 * Implementation of dt_it_ops::init.
372 * Used with striped objects. Internally just initializes the iterator
373 * on the first stripe.
375 * \see dt_it_ops::init() in the API description for details.
377 static struct dt_it *lod_striped_it_init(const struct lu_env *env,
378 struct dt_object *dt, __u32 attr)
380 struct lod_object *lo = lod_dt_obj(dt);
381 struct dt_object *next;
382 struct lod_it *it = &lod_env_info(env)->lti_it;
383 struct dt_it *it_next;
386 LASSERT(lo->ldo_dir_stripe_count > 0);
389 next = lo->ldo_stripe[index];
390 if (next && dt_object_exists(next))
392 } while (++index < lo->ldo_dir_stripe_count);
394 /* no valid stripe */
395 if (!next || !dt_object_exists(next))
396 return ERR_PTR(-ENODEV);
398 LASSERT(next->do_index_ops != NULL);
400 it_next = next->do_index_ops->dio_it.init(env, next, attr);
404 /* currently we do not use more than one iterator per thread
405 * so we store it in thread info. if at some point we need
406 * more active iterators in a single thread, we can allocate
408 LASSERT(it->lit_obj == NULL);
410 it->lit_stripe_index = index;
412 it->lit_it = it_next;
415 return (struct dt_it *)it;
418 #define LOD_CHECK_STRIPED_IT(env, it, lo) \
420 LASSERT((it)->lit_obj != NULL); \
421 LASSERT((it)->lit_it != NULL); \
422 LASSERT((lo)->ldo_dir_stripe_count > 0); \
423 LASSERT((it)->lit_stripe_index < (lo)->ldo_dir_stripe_count); \
427 * Implementation of dt_it_ops::fini.
429 * Used with striped objects.
431 * \see dt_it_ops::fini() in the API description for details.
433 static void lod_striped_it_fini(const struct lu_env *env, struct dt_it *di)
435 struct lod_it *it = (struct lod_it *)di;
436 struct lod_object *lo = lod_dt_obj(it->lit_obj);
437 struct dt_object *next;
439 /* If lit_it == NULL, then it means the sub_it has been finished,
440 * which only happens in failure cases, see lod_striped_it_next() */
441 if (it->lit_it != NULL) {
442 LOD_CHECK_STRIPED_IT(env, it, lo);
444 next = lo->ldo_stripe[it->lit_stripe_index];
446 LASSERT(next->do_index_ops != NULL);
447 next->do_index_ops->dio_it.fini(env, it->lit_it);
451 /* the iterator not in use any more */
454 it->lit_stripe_index = 0;
458 * Implementation of dt_it_ops::get.
460 * Right now it's not used widely, only to reset the iterator to the
461 * initial position. It should be possible to implement a full version
462 * which chooses a correct stripe to be able to position with any key.
464 * \see dt_it_ops::get() in the API description for details.
466 static int lod_striped_it_get(const struct lu_env *env, struct dt_it *di,
467 const struct dt_key *key)
469 const struct lod_it *it = (const struct lod_it *)di;
470 struct lod_object *lo = lod_dt_obj(it->lit_obj);
471 struct dt_object *next;
473 LOD_CHECK_STRIPED_IT(env, it, lo);
475 next = lo->ldo_stripe[it->lit_stripe_index];
476 LASSERT(next != NULL);
477 LASSERT(dt_object_exists(next));
478 LASSERT(next->do_index_ops != NULL);
480 return next->do_index_ops->dio_it.get(env, it->lit_it, key);
484 * Implementation of dt_it_ops::put.
486 * Used with striped objects.
488 * \see dt_it_ops::put() in the API description for details.
490 static void lod_striped_it_put(const struct lu_env *env, struct dt_it *di)
492 struct lod_it *it = (struct lod_it *)di;
493 struct lod_object *lo = lod_dt_obj(it->lit_obj);
494 struct dt_object *next;
497 * If lit_it == NULL, then it means the sub_it has been finished,
498 * which only happens in failure cases, see lod_striped_it_next()
503 LOD_CHECK_STRIPED_IT(env, it, lo);
505 next = lo->ldo_stripe[it->lit_stripe_index];
506 LASSERT(next != NULL);
507 LASSERT(next->do_index_ops != NULL);
509 return next->do_index_ops->dio_it.put(env, it->lit_it);
513 * Implementation of dt_it_ops::next.
515 * Used with striped objects. When the end of the current stripe is
516 * reached, the method takes the next stripe's iterator.
518 * \see dt_it_ops::next() in the API description for details.
520 static int lod_striped_it_next(const struct lu_env *env, struct dt_it *di)
522 struct lod_it *it = (struct lod_it *)di;
523 struct lod_object *lo = lod_dt_obj(it->lit_obj);
524 struct dt_object *next;
525 struct dt_it *it_next;
531 LOD_CHECK_STRIPED_IT(env, it, lo);
533 next = lo->ldo_stripe[it->lit_stripe_index];
534 LASSERT(next != NULL);
535 LASSERT(dt_object_exists(next));
536 LASSERT(next->do_index_ops != NULL);
538 rc = next->do_index_ops->dio_it.next(env, it->lit_it);
542 if (rc == 0 && it->lit_stripe_index == 0)
545 if (rc == 0 && it->lit_stripe_index > 0) {
546 struct lu_dirent *ent;
548 ent = (struct lu_dirent *)lod_env_info(env)->lti_key;
550 rc = next->do_index_ops->dio_it.rec(env, it->lit_it,
551 (struct dt_rec *)ent,
556 /* skip . and .. for slave stripe */
557 if ((strncmp(ent->lde_name, ".",
558 le16_to_cpu(ent->lde_namelen)) == 0 &&
559 le16_to_cpu(ent->lde_namelen) == 1) ||
560 (strncmp(ent->lde_name, "..",
561 le16_to_cpu(ent->lde_namelen)) == 0 &&
562 le16_to_cpu(ent->lde_namelen) == 2))
568 next->do_index_ops->dio_it.put(env, it->lit_it);
569 next->do_index_ops->dio_it.fini(env, it->lit_it);
572 /* go to next stripe */
573 index = it->lit_stripe_index;
574 while (++index < lo->ldo_dir_stripe_count) {
575 next = lo->ldo_stripe[index];
579 if (!dt_object_exists(next))
582 rc = next->do_ops->do_index_try(env, next,
583 &dt_directory_features);
587 LASSERT(next->do_index_ops != NULL);
589 it_next = next->do_index_ops->dio_it.init(env, next,
592 RETURN(PTR_ERR(it_next));
594 rc = next->do_index_ops->dio_it.get(env, it_next,
595 (const struct dt_key *)"");
597 RETURN(rc == 0 ? -EIO : rc);
599 it->lit_it = it_next;
600 it->lit_stripe_index = index;
609 * Implementation of dt_it_ops::key.
611 * Used with striped objects.
613 * \see dt_it_ops::key() in the API description for details.
615 static struct dt_key *lod_striped_it_key(const struct lu_env *env,
616 const struct dt_it *di)
618 const struct lod_it *it = (const struct lod_it *)di;
619 struct lod_object *lo = lod_dt_obj(it->lit_obj);
620 struct dt_object *next;
622 LOD_CHECK_STRIPED_IT(env, it, lo);
624 next = lo->ldo_stripe[it->lit_stripe_index];
625 LASSERT(next != NULL);
626 LASSERT(next->do_index_ops != NULL);
628 return next->do_index_ops->dio_it.key(env, it->lit_it);
632 * Implementation of dt_it_ops::key_size.
634 * Used with striped objects.
636 * \see dt_it_ops::size() in the API description for details.
638 static int lod_striped_it_key_size(const struct lu_env *env,
639 const struct dt_it *di)
641 struct lod_it *it = (struct lod_it *)di;
642 struct lod_object *lo = lod_dt_obj(it->lit_obj);
643 struct dt_object *next;
645 LOD_CHECK_STRIPED_IT(env, it, lo);
647 next = lo->ldo_stripe[it->lit_stripe_index];
648 LASSERT(next != NULL);
649 LASSERT(next->do_index_ops != NULL);
651 return next->do_index_ops->dio_it.key_size(env, it->lit_it);
655 * Implementation of dt_it_ops::rec.
657 * Used with striped objects.
659 * \see dt_it_ops::rec() in the API description for details.
661 static int lod_striped_it_rec(const struct lu_env *env, const struct dt_it *di,
662 struct dt_rec *rec, __u32 attr)
664 const struct lod_it *it = (const struct lod_it *)di;
665 struct lod_object *lo = lod_dt_obj(it->lit_obj);
666 struct dt_object *next;
668 LOD_CHECK_STRIPED_IT(env, it, lo);
670 next = lo->ldo_stripe[it->lit_stripe_index];
671 LASSERT(next != NULL);
672 LASSERT(next->do_index_ops != NULL);
674 return next->do_index_ops->dio_it.rec(env, it->lit_it, rec, attr);
678 * Implementation of dt_it_ops::rec_size.
680 * Used with striped objects.
682 * \see dt_it_ops::rec_size() in the API description for details.
684 static int lod_striped_it_rec_size(const struct lu_env *env,
685 const struct dt_it *di, __u32 attr)
687 struct lod_it *it = (struct lod_it *)di;
688 struct lod_object *lo = lod_dt_obj(it->lit_obj);
689 struct dt_object *next;
691 LOD_CHECK_STRIPED_IT(env, it, lo);
693 next = lo->ldo_stripe[it->lit_stripe_index];
694 LASSERT(next != NULL);
695 LASSERT(next->do_index_ops != NULL);
697 return next->do_index_ops->dio_it.rec_size(env, it->lit_it, attr);
701 * Implementation of dt_it_ops::store.
703 * Used with striped objects.
705 * \see dt_it_ops::store() in the API description for details.
707 static __u64 lod_striped_it_store(const struct lu_env *env,
708 const struct dt_it *di)
710 const struct lod_it *it = (const struct lod_it *)di;
711 struct lod_object *lo = lod_dt_obj(it->lit_obj);
712 struct dt_object *next;
714 LOD_CHECK_STRIPED_IT(env, it, lo);
716 next = lo->ldo_stripe[it->lit_stripe_index];
717 LASSERT(next != NULL);
718 LASSERT(next->do_index_ops != NULL);
720 return next->do_index_ops->dio_it.store(env, it->lit_it);
724 * Implementation of dt_it_ops::load.
726 * Used with striped objects.
728 * \see dt_it_ops::load() in the API description for details.
730 static int lod_striped_it_load(const struct lu_env *env,
731 const struct dt_it *di, __u64 hash)
733 const struct lod_it *it = (const struct lod_it *)di;
734 struct lod_object *lo = lod_dt_obj(it->lit_obj);
735 struct dt_object *next;
737 LOD_CHECK_STRIPED_IT(env, it, lo);
739 next = lo->ldo_stripe[it->lit_stripe_index];
740 LASSERT(next != NULL);
741 LASSERT(next->do_index_ops != NULL);
743 return next->do_index_ops->dio_it.load(env, it->lit_it, hash);
746 static struct dt_index_operations lod_striped_index_ops = {
747 .dio_lookup = lod_lookup,
748 .dio_declare_insert = lod_declare_insert,
749 .dio_insert = lod_insert,
750 .dio_declare_delete = lod_declare_delete,
751 .dio_delete = lod_delete,
753 .init = lod_striped_it_init,
754 .fini = lod_striped_it_fini,
755 .get = lod_striped_it_get,
756 .put = lod_striped_it_put,
757 .next = lod_striped_it_next,
758 .key = lod_striped_it_key,
759 .key_size = lod_striped_it_key_size,
760 .rec = lod_striped_it_rec,
761 .rec_size = lod_striped_it_rec_size,
762 .store = lod_striped_it_store,
763 .load = lod_striped_it_load,
768 * Append the FID for each shard of the striped directory after the
769 * given LMV EA header.
771 * To simplify striped directory and the consistency verification,
772 * we only store the LMV EA header on disk, for both master object
773 * and slave objects. When someone wants to know the whole LMV EA,
774 * such as client readdir(), we can build the entrie LMV EA on the
775 * MDT side (in RAM) via iterating the sub-directory entries that
776 * are contained in the master object of the stripe directory.
778 * For the master object of the striped directroy, the valid name
779 * for each shard is composed of the ${shard_FID}:${shard_idx}.
781 * There may be holes in the LMV EA if some shards' name entries
782 * are corrupted or lost.
784 * \param[in] env pointer to the thread context
785 * \param[in] lo pointer to the master object of the striped directory
786 * \param[in] buf pointer to the lu_buf which will hold the LMV EA
787 * \param[in] resize whether re-allocate the buffer if it is not big enough
789 * \retval positive size of the LMV EA
790 * \retval 0 for nothing to be loaded
791 * \retval negative error number on failure
793 int lod_load_lmv_shards(const struct lu_env *env, struct lod_object *lo,
794 struct lu_buf *buf, bool resize)
796 struct lu_dirent *ent =
797 (struct lu_dirent *)lod_env_info(env)->lti_key;
798 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
799 struct dt_object *obj = dt_object_child(&lo->ldo_obj);
800 struct lmv_mds_md_v1 *lmv1 = buf->lb_buf;
802 const struct dt_it_ops *iops;
804 __u32 magic = le32_to_cpu(lmv1->lmv_magic);
809 if (magic != LMV_MAGIC_V1)
812 stripes = le32_to_cpu(lmv1->lmv_stripe_count);
816 rc = lmv_mds_md_size(stripes, magic);
820 if (buf->lb_len < lmv1_size) {
829 lu_buf_alloc(buf, lmv1_size);
834 memcpy(buf->lb_buf, tbuf.lb_buf, tbuf.lb_len);
837 if (unlikely(!dt_try_as_dir(env, obj)))
840 memset(&lmv1->lmv_stripe_fids[0], 0, stripes * sizeof(struct lu_fid));
841 iops = &obj->do_index_ops->dio_it;
842 it = iops->init(env, obj, LUDA_64BITHASH);
846 rc = iops->load(env, it, 0);
848 rc = iops->next(env, it);
853 char name[FID_LEN + 2] = "";
858 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
864 fid_le_to_cpu(&fid, &ent->lde_fid);
865 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
866 if (ent->lde_name[0] == '.') {
867 if (ent->lde_namelen == 1)
870 if (ent->lde_namelen == 2 && ent->lde_name[1] == '.')
874 len = scnprintf(name, sizeof(name),
875 DFID":", PFID(&ent->lde_fid));
876 /* The ent->lde_name is composed of ${FID}:${index} */
877 if (ent->lde_namelen < len + 1 ||
878 memcmp(ent->lde_name, name, len) != 0) {
879 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
880 "%s: invalid shard name %.*s with the FID "DFID
881 " for the striped directory "DFID", %s\n",
882 lod2obd(lod)->obd_name, ent->lde_namelen,
883 ent->lde_name, PFID(&fid),
884 PFID(lu_object_fid(&obj->do_lu)),
885 lod->lod_lmv_failout ? "failout" : "skip");
887 if (lod->lod_lmv_failout)
895 if (ent->lde_name[len] < '0' ||
896 ent->lde_name[len] > '9') {
897 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
898 "%s: invalid shard name %.*s with the "
899 "FID "DFID" for the striped directory "
901 lod2obd(lod)->obd_name, ent->lde_namelen,
902 ent->lde_name, PFID(&fid),
903 PFID(lu_object_fid(&obj->do_lu)),
904 lod->lod_lmv_failout ?
907 if (lod->lod_lmv_failout)
913 index = index * 10 + ent->lde_name[len++] - '0';
914 } while (len < ent->lde_namelen);
916 if (len == ent->lde_namelen) {
917 /* Out of LMV EA range. */
918 if (index >= stripes) {
919 CERROR("%s: the shard %.*s for the striped "
920 "directory "DFID" is out of the known "
921 "LMV EA range [0 - %u], failout\n",
922 lod2obd(lod)->obd_name, ent->lde_namelen,
924 PFID(lu_object_fid(&obj->do_lu)),
930 /* The slot has been occupied. */
931 if (!fid_is_zero(&lmv1->lmv_stripe_fids[index])) {
935 &lmv1->lmv_stripe_fids[index]);
936 CERROR("%s: both the shard "DFID" and "DFID
937 " for the striped directory "DFID
938 " claim the same LMV EA slot at the "
939 "index %d, failout\n",
940 lod2obd(lod)->obd_name,
941 PFID(&fid0), PFID(&fid),
942 PFID(lu_object_fid(&obj->do_lu)), index);
947 /* stored as LE mode */
948 lmv1->lmv_stripe_fids[index] = ent->lde_fid;
951 rc = iops->next(env, it);
958 RETURN(rc > 0 ? lmv_mds_md_size(stripes, magic) : rc);
962 * Implementation of dt_object_operations::do_index_try.
964 * \see dt_object_operations::do_index_try() in the API description for details.
966 static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
967 const struct dt_index_features *feat)
969 struct lod_object *lo = lod_dt_obj(dt);
970 struct dt_object *next = dt_object_child(dt);
974 LASSERT(next->do_ops);
975 LASSERT(next->do_ops->do_index_try);
977 rc = lod_striping_load(env, lo);
981 rc = next->do_ops->do_index_try(env, next, feat);
985 if (lo->ldo_dir_stripe_count > 0) {
988 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
989 if (!lo->ldo_stripe[i])
991 if (!dt_object_exists(lo->ldo_stripe[i]))
993 rc = lo->ldo_stripe[i]->do_ops->do_index_try(env,
994 lo->ldo_stripe[i], feat);
998 dt->do_index_ops = &lod_striped_index_ops;
1000 dt->do_index_ops = &lod_index_ops;
1007 * Implementation of dt_object_operations::do_read_lock.
1009 * \see dt_object_operations::do_read_lock() in the API description for details.
1011 static void lod_read_lock(const struct lu_env *env, struct dt_object *dt,
1014 dt_read_lock(env, dt_object_child(dt), role);
1018 * Implementation of dt_object_operations::do_write_lock.
1020 * \see dt_object_operations::do_write_lock() in the API description for
1023 static void lod_write_lock(const struct lu_env *env, struct dt_object *dt,
1026 dt_write_lock(env, dt_object_child(dt), role);
1030 * Implementation of dt_object_operations::do_read_unlock.
1032 * \see dt_object_operations::do_read_unlock() in the API description for
1035 static void lod_read_unlock(const struct lu_env *env, struct dt_object *dt)
1037 dt_read_unlock(env, dt_object_child(dt));
1041 * Implementation of dt_object_operations::do_write_unlock.
1043 * \see dt_object_operations::do_write_unlock() in the API description for
1046 static void lod_write_unlock(const struct lu_env *env, struct dt_object *dt)
1048 dt_write_unlock(env, dt_object_child(dt));
1052 * Implementation of dt_object_operations::do_write_locked.
1054 * \see dt_object_operations::do_write_locked() in the API description for
1057 static int lod_write_locked(const struct lu_env *env, struct dt_object *dt)
1059 return dt_write_locked(env, dt_object_child(dt));
1063 * Implementation of dt_object_operations::do_attr_get.
1065 * \see dt_object_operations::do_attr_get() in the API description for details.
1067 static int lod_attr_get(const struct lu_env *env,
1068 struct dt_object *dt,
1069 struct lu_attr *attr)
1071 /* Note: for striped directory, client will merge attributes
1072 * from all of the sub-stripes see lmv_merge_attr(), and there
1073 * no MDD logic depend on directory nlink/size/time, so we can
1074 * always use master inode nlink and size for now. */
1075 return dt_attr_get(env, dt_object_child(dt), attr);
1078 void lod_adjust_stripe_size(struct lod_layout_component *comp,
1079 __u32 def_stripe_size)
1081 __u64 comp_end = comp->llc_extent.e_end;
1083 /* Choose stripe size if not set. Note that default stripe size can't
1084 * be used as is, because it must be multiplier of given component end.
1085 * - first check if default stripe size can be used
1086 * - if not than select the lowest set bit from component end and use
1087 * that value as stripe size
1089 if (!comp->llc_stripe_size) {
1090 if (comp_end == LUSTRE_EOF || !(comp_end % def_stripe_size))
1091 comp->llc_stripe_size = def_stripe_size;
1093 comp->llc_stripe_size = comp_end & ~(comp_end - 1);
1095 /* check stripe size is multiplier of comp_end */
1096 if (comp_end != LUSTRE_EOF &&
1097 comp_end % comp->llc_stripe_size) {
1098 /* fix that even for defined stripe size but warn
1099 * about the problem, that must not happen
1101 CWARN("Component end %llu is not aligned by the stripe size %u\n",
1102 comp_end, comp->llc_stripe_size);
1104 comp->llc_stripe_size = comp_end & ~(comp_end - 1);
1109 static inline void lod_adjust_stripe_info(struct lod_layout_component *comp,
1110 struct lov_desc *desc,
1113 if (comp->llc_pattern != LOV_PATTERN_MDT) {
1114 if (append_stripes) {
1115 comp->llc_stripe_count = append_stripes;
1116 } else if (!comp->llc_stripe_count) {
1117 comp->llc_stripe_count =
1118 desc->ld_default_stripe_count;
1122 lod_adjust_stripe_size(comp, desc->ld_default_stripe_size);
1125 int lod_obj_for_each_stripe(const struct lu_env *env, struct lod_object *lo,
1127 struct lod_obj_stripe_cb_data *data)
1129 struct lod_layout_component *lod_comp;
1133 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1134 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1135 lod_comp = &lo->ldo_comp_entries[i];
1137 if (lod_comp->llc_stripe == NULL)
1140 /* has stripe but not inited yet, this component has been
1141 * declared to be created, but hasn't created yet.
1143 if (!lod_comp_inited(lod_comp))
1146 if (data->locd_comp_skip_cb &&
1147 data->locd_comp_skip_cb(env, lo, i, data))
1150 if (data->locd_comp_cb) {
1151 rc = data->locd_comp_cb(env, lo, i, data);
1156 /* could used just to do sth about component, not each
1159 if (!data->locd_stripe_cb)
1162 LASSERT(lod_comp->llc_stripe_count > 0);
1163 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
1164 struct dt_object *dt = lod_comp->llc_stripe[j];
1168 rc = data->locd_stripe_cb(env, lo, dt, th, i, j, data);
1176 static bool lod_obj_attr_set_comp_skip_cb(const struct lu_env *env,
1177 struct lod_object *lo, int comp_idx,
1178 struct lod_obj_stripe_cb_data *data)
1180 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[comp_idx];
1181 bool skipped = false;
1183 if (!(data->locd_attr->la_valid & LA_LAYOUT_VERSION))
1186 switch (lo->ldo_flr_state) {
1187 case LCM_FL_WRITE_PENDING: {
1190 /* skip stale components */
1191 if (lod_comp->llc_flags & LCME_FL_STALE) {
1196 /* skip valid and overlapping components, therefore any
1197 * attempts to write overlapped components will never succeed
1198 * because client will get EINPROGRESS. */
1199 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1203 if (lo->ldo_comp_entries[i].llc_flags & LCME_FL_STALE)
1206 if (lu_extent_is_overlapped(&lod_comp->llc_extent,
1207 &lo->ldo_comp_entries[i].llc_extent)) {
1215 LASSERTF(0, "impossible: %d\n", lo->ldo_flr_state);
1216 case LCM_FL_SYNC_PENDING:
1220 CDEBUG(D_LAYOUT, DFID": %s to set component %x to version: %u\n",
1221 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
1222 skipped ? "skipped" : "chose", lod_comp->llc_id,
1223 data->locd_attr->la_layout_version);
1229 lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
1230 struct dt_object *dt, struct thandle *th,
1231 int comp_idx, int stripe_idx,
1232 struct lod_obj_stripe_cb_data *data)
1234 if (data->locd_declare)
1235 return lod_sub_declare_attr_set(env, dt, data->locd_attr, th);
1237 if (data->locd_attr->la_valid & LA_LAYOUT_VERSION) {
1238 CDEBUG(D_LAYOUT, DFID": set layout version: %u, comp_idx: %d\n",
1239 PFID(lu_object_fid(&dt->do_lu)),
1240 data->locd_attr->la_layout_version, comp_idx);
1243 return lod_sub_attr_set(env, dt, data->locd_attr, th);
1247 * Implementation of dt_object_operations::do_declare_attr_set.
1249 * If the object is striped, then apply the changes to all the stripes.
1251 * \see dt_object_operations::do_declare_attr_set() in the API description
1254 static int lod_declare_attr_set(const struct lu_env *env,
1255 struct dt_object *dt,
1256 const struct lu_attr *attr,
1259 struct dt_object *next = dt_object_child(dt);
1260 struct lod_object *lo = lod_dt_obj(dt);
1265 * declare setattr on the local object
1267 rc = lod_sub_declare_attr_set(env, next, attr, th);
1271 /* osp_declare_attr_set() ignores all attributes other than
1272 * UID, GID, PROJID, and size, and osp_attr_set() ignores all
1273 * but UID, GID and PROJID. Declaration of size attr setting
1274 * happens through lod_declare_init_size(), and not through
1275 * this function. Therefore we need not load striping unless
1276 * ownership is changing. This should save memory and (we hope)
1277 * speed up rename().
1279 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1280 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1283 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1286 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID | LA_MODE |
1287 LA_ATIME | LA_MTIME | LA_CTIME |
1292 * load striping information, notice we don't do this when object
1293 * is being initialized as we don't need this information till
1294 * few specific cases like destroy, chown
1296 rc = lod_striping_load(env, lo);
1300 if (!lod_obj_is_striped(dt))
1304 * if object is striped declare changes on the stripes
1306 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1307 LASSERT(lo->ldo_stripe);
1308 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1309 if (lo->ldo_stripe[i] == NULL)
1311 if (!dt_object_exists(lo->ldo_stripe[i]))
1313 rc = lod_sub_declare_attr_set(env, lo->ldo_stripe[i],
1319 struct lod_obj_stripe_cb_data data = { { 0 } };
1321 data.locd_attr = attr;
1322 data.locd_declare = true;
1323 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1324 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1330 if (!dt_object_exists(next) || dt_object_remote(next) ||
1331 !S_ISREG(attr->la_mode))
1334 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1335 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
1339 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) ||
1340 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1341 struct lod_thread_info *info = lod_env_info(env);
1342 struct lu_buf *buf = &info->lti_buf;
1344 buf->lb_buf = info->lti_ea_store;
1345 buf->lb_len = info->lti_ea_store_size;
1346 rc = lod_sub_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1347 LU_XATTR_REPLACE, th);
1354 * Implementation of dt_object_operations::do_attr_set.
1356 * If the object is striped, then apply the changes to all or subset of
1357 * the stripes depending on the object type and specific attributes.
1359 * \see dt_object_operations::do_attr_set() in the API description for details.
1361 static int lod_attr_set(const struct lu_env *env,
1362 struct dt_object *dt,
1363 const struct lu_attr *attr,
1366 struct dt_object *next = dt_object_child(dt);
1367 struct lod_object *lo = lod_dt_obj(dt);
1372 * apply changes to the local object
1374 rc = lod_sub_attr_set(env, next, attr, th);
1378 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1379 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1382 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1385 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE | LA_PROJID |
1386 LA_ATIME | LA_MTIME | LA_CTIME |
1391 /* FIXME: a tricky case in the code path of mdd_layout_change():
1392 * the in-memory striping information has been freed in lod_xattr_set()
1393 * due to layout change. It has to load stripe here again. It only
1394 * changes flags of layout so declare_attr_set() is still accurate */
1395 rc = lod_striping_load(env, lo);
1399 if (!lod_obj_is_striped(dt))
1403 * if object is striped, apply changes to all the stripes
1405 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1406 LASSERT(lo->ldo_stripe);
1407 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1408 if (unlikely(lo->ldo_stripe[i] == NULL))
1411 if ((dt_object_exists(lo->ldo_stripe[i]) == 0))
1414 rc = lod_sub_attr_set(env, lo->ldo_stripe[i], attr, th);
1419 struct lod_obj_stripe_cb_data data = { { 0 } };
1421 data.locd_attr = attr;
1422 data.locd_declare = false;
1423 data.locd_comp_skip_cb = lod_obj_attr_set_comp_skip_cb;
1424 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1425 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1431 if (!dt_object_exists(next) || dt_object_remote(next) ||
1432 !S_ISREG(attr->la_mode))
1435 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1436 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
1440 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE)) {
1441 struct lod_thread_info *info = lod_env_info(env);
1442 struct lu_buf *buf = &info->lti_buf;
1443 struct ost_id *oi = &info->lti_ostid;
1444 struct lu_fid *fid = &info->lti_fid;
1445 struct lov_mds_md_v1 *lmm;
1446 struct lov_ost_data_v1 *objs;
1449 rc = lod_get_lov_ea(env, lo);
1453 buf->lb_buf = info->lti_ea_store;
1454 buf->lb_len = info->lti_ea_store_size;
1455 lmm = info->lti_ea_store;
1456 magic = le32_to_cpu(lmm->lmm_magic);
1457 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1458 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1459 struct lov_comp_md_entry_v1 *lcme =
1460 &lcm->lcm_entries[0];
1462 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
1463 magic = le32_to_cpu(lmm->lmm_magic);
1466 if (magic == LOV_MAGIC_V1)
1467 objs = &(lmm->lmm_objects[0]);
1469 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1470 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1471 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1473 fid_to_ostid(fid, oi);
1474 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1476 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1477 LU_XATTR_REPLACE, th);
1478 } else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1479 struct lod_thread_info *info = lod_env_info(env);
1480 struct lu_buf *buf = &info->lti_buf;
1481 struct lov_comp_md_v1 *lcm;
1482 struct lov_comp_md_entry_v1 *lcme;
1484 rc = lod_get_lov_ea(env, lo);
1488 buf->lb_buf = info->lti_ea_store;
1489 buf->lb_len = info->lti_ea_store_size;
1491 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1 &&
1492 le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_SEL)
1495 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1496 lcme = &lcm->lcm_entries[0];
1497 le64_add_cpu(&lcme->lcme_extent.e_start, 1);
1498 le64_add_cpu(&lcme->lcme_extent.e_end, -1);
1500 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1501 LU_XATTR_REPLACE, th);
1508 * Implementation of dt_object_operations::do_xattr_get.
1510 * If LOV EA is requested from the root object and it's not
1511 * found, then return default striping for the filesystem.
1513 * \see dt_object_operations::do_xattr_get() in the API description for details.
1515 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1516 struct lu_buf *buf, const char *name)
1518 struct lod_thread_info *info = lod_env_info(env);
1519 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1524 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1525 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1526 struct lmv_mds_md_v1 *lmv1;
1527 struct lmv_foreign_md *lfm;
1530 if (rc > (typeof(rc))sizeof(*lmv1))
1533 /* short (<= sizeof(struct lmv_mds_md_v1)) foreign LMV case */
1534 /* XXX empty foreign LMV is not allowed */
1535 if (rc <= offsetof(typeof(*lfm), lfm_value))
1536 RETURN(rc = rc > 0 ? -EINVAL : rc);
1538 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1539 BUILD_BUG_ON(sizeof(*lmv1) > sizeof(info->lti_key));
1541 /* lti_buf is large enough for *lmv1 or a short
1542 * (<= sizeof(struct lmv_mds_md_v1)) foreign LMV
1544 info->lti_buf.lb_buf = info->lti_key;
1545 info->lti_buf.lb_len = sizeof(*lmv1);
1546 rc = dt_xattr_get(env, dt_object_child(dt),
1547 &info->lti_buf, name);
1548 if (unlikely(rc <= offsetof(typeof(*lfm),
1550 RETURN(rc = rc > 0 ? -EINVAL : rc);
1552 lfm = info->lti_buf.lb_buf;
1553 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN)
1556 if (unlikely(rc != sizeof(*lmv1)))
1557 RETURN(rc = rc > 0 ? -EINVAL : rc);
1559 lmv1 = info->lti_buf.lb_buf;
1560 /* The on-disk LMV EA only contains header, but the
1561 * returned LMV EA size should contain the space for
1562 * the FIDs of all shards of the striped directory. */
1563 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1564 rc = lmv_mds_md_size(
1565 le32_to_cpu(lmv1->lmv_stripe_count),
1566 le32_to_cpu(lmv1->lmv_magic));
1569 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1572 if (rc != sizeof(*lmv1))
1573 RETURN(rc = rc > 0 ? -EINVAL : rc);
1575 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1579 RETURN(rc = rc1 != 0 ? rc1 : rc);
1582 if ((rc > 0) && buf->lb_buf && strcmp(name, XATTR_NAME_LOV) == 0) {
1583 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1585 if (lcm->lcm_magic == cpu_to_le32(LOV_MAGIC_SEL))
1586 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
1589 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1593 * XXX: Only used by lfsck
1595 * lod returns default striping on the real root of the device
1596 * this is like the root stores default striping for the whole
1597 * filesystem. historically we've been using a different approach
1598 * and store it in the config.
1600 dt_root_get(env, dev->lod_child, &info->lti_fid);
1601 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1603 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1604 struct lov_user_md *lum = buf->lb_buf;
1605 struct lov_desc *desc = &dev->lod_ost_descs.ltd_lov_desc;
1607 if (buf->lb_buf == NULL) {
1609 } else if (buf->lb_len >= sizeof(*lum)) {
1610 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1611 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1612 lmm_oi_set_id(&lum->lmm_oi, 0);
1613 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1614 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1615 lum->lmm_stripe_size = cpu_to_le32(
1616 desc->ld_default_stripe_size);
1617 lum->lmm_stripe_count = cpu_to_le16(
1618 desc->ld_default_stripe_count);
1619 lum->lmm_stripe_offset = cpu_to_le16(
1620 desc->ld_default_stripe_offset);
1633 * Checks that the magic of the stripe is sane.
1635 * \param[in] lod lod device
1636 * \param[in] lum a buffer storing LMV EA to verify
1638 * \retval 0 if the EA is sane
1639 * \retval negative otherwise
1641 static int lod_verify_md_striping(struct lod_device *lod,
1642 const struct lmv_user_md_v1 *lum)
1644 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1645 CERROR("%s: invalid lmv_user_md: magic = %x, "
1646 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1647 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1648 (int)le32_to_cpu(lum->lum_stripe_offset),
1649 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1657 * Initialize LMV EA for a slave.
1659 * Initialize slave's LMV EA from the master's LMV EA.
1661 * \param[in] master_lmv a buffer containing master's EA
1662 * \param[out] slave_lmv a buffer where slave's EA will be stored
1665 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1666 const struct lmv_mds_md_v1 *master_lmv)
1668 *slave_lmv = *master_lmv;
1669 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1675 * Generate LMV EA from the object passed as \a dt. The object must have
1676 * the stripes created and initialized.
1678 * \param[in] env execution environment
1679 * \param[in] dt object
1680 * \param[out] lmv_buf buffer storing generated LMV EA
1682 * \retval 0 on success
1683 * \retval negative if failed
1685 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1686 struct lu_buf *lmv_buf)
1688 struct lod_thread_info *info = lod_env_info(env);
1689 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1690 struct lod_object *lo = lod_dt_obj(dt);
1691 struct lmv_mds_md_v1 *lmm1;
1693 int type = LU_SEQ_RANGE_ANY;
1698 LASSERT(lo->ldo_dir_striped != 0);
1699 LASSERT(lo->ldo_dir_stripe_count > 0);
1700 stripe_count = lo->ldo_dir_stripe_count;
1701 /* Only store the LMV EA heahder on the disk. */
1702 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1703 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1707 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1710 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1711 memset(lmm1, 0, sizeof(*lmm1));
1712 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1713 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1714 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1715 lmm1->lmv_layout_version = cpu_to_le32(lo->ldo_dir_layout_version);
1716 if (lod_is_layout_changing(lo)) {
1717 lmm1->lmv_migrate_hash = cpu_to_le32(lo->ldo_dir_migrate_hash);
1718 lmm1->lmv_migrate_offset =
1719 cpu_to_le32(lo->ldo_dir_migrate_offset);
1721 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1726 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1727 lmv_buf->lb_buf = info->lti_ea_store;
1728 lmv_buf->lb_len = sizeof(*lmm1);
1734 * Create in-core represenation for a striped directory.
1736 * Parse the buffer containing LMV EA and instantiate LU objects
1737 * representing the stripe objects. The pointers to the objects are
1738 * stored in ldo_stripe field of \a lo. This function is used when
1739 * we need to access an already created object (i.e. load from a disk).
1741 * \param[in] env execution environment
1742 * \param[in] lo lod object
1743 * \param[in] buf buffer containing LMV EA
1745 * \retval 0 on success
1746 * \retval negative if failed
1748 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1749 const struct lu_buf *buf)
1751 struct lod_thread_info *info = lod_env_info(env);
1752 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1753 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1754 struct dt_object **stripe;
1755 union lmv_mds_md *lmm = buf->lb_buf;
1756 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1757 struct lu_fid *fid = &info->lti_fid;
1762 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1764 /* XXX may be useless as not called for foreign LMV ?? */
1765 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_FOREIGN)
1768 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1769 lo->ldo_dir_slave_stripe = 1;
1773 if (!lmv_is_sane(lmv1))
1776 LASSERT(lo->ldo_stripe == NULL);
1777 OBD_ALLOC(stripe, sizeof(stripe[0]) *
1778 (le32_to_cpu(lmv1->lmv_stripe_count)));
1782 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1783 struct dt_device *tgt_dt;
1784 struct dt_object *dto;
1785 int type = LU_SEQ_RANGE_ANY;
1788 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1789 if (!fid_is_sane(fid)) {
1794 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1798 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1799 tgt_dt = lod->lod_child;
1801 struct lod_tgt_desc *tgt;
1803 tgt = LTD_TGT(ltd, idx);
1805 GOTO(out, rc = -ESTALE);
1806 tgt_dt = tgt->ltd_tgt;
1809 dto = dt_locate_at(env, tgt_dt, fid,
1810 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1813 GOTO(out, rc = PTR_ERR(dto));
1818 lo->ldo_stripe = stripe;
1819 lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1820 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1821 lo->ldo_dir_layout_version = le32_to_cpu(lmv1->lmv_layout_version);
1822 lo->ldo_dir_migrate_offset = le32_to_cpu(lmv1->lmv_migrate_offset);
1823 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv1->lmv_migrate_hash);
1824 lo->ldo_dir_hash_type = le32_to_cpu(lmv1->lmv_hash_type);
1826 lod_striping_free_nolock(env, lo);
1832 * Declare create a striped directory.
1834 * Declare creating a striped directory with a given stripe pattern on the
1835 * specified MDTs. A striped directory is represented as a regular directory
1836 * - an index listing all the stripes. The stripes point back to the master
1837 * object with ".." and LinkEA. The master object gets LMV EA which
1838 * identifies it as a striped directory. The function allocates FIDs
1841 * \param[in] env execution environment
1842 * \param[in] dt object
1843 * \param[in] attr attributes to initialize the objects with
1844 * \param[in] dof type of objects to be created
1845 * \param[in] th transaction handle
1847 * \retval 0 on success
1848 * \retval negative if failed
1850 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1851 struct dt_object *dt,
1852 struct lu_attr *attr,
1853 struct dt_object_format *dof,
1856 struct lod_thread_info *info = lod_env_info(env);
1857 struct lu_buf lmv_buf;
1858 struct lu_buf slave_lmv_buf;
1859 struct lmv_mds_md_v1 *lmm;
1860 struct lmv_mds_md_v1 *slave_lmm = NULL;
1861 struct dt_insert_rec *rec = &info->lti_dt_rec;
1862 struct lod_object *lo = lod_dt_obj(dt);
1867 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1870 lmm = lmv_buf.lb_buf;
1872 OBD_ALLOC_PTR(slave_lmm);
1873 if (slave_lmm == NULL)
1874 GOTO(out, rc = -ENOMEM);
1876 lod_prep_slave_lmv_md(slave_lmm, lmm);
1877 slave_lmv_buf.lb_buf = slave_lmm;
1878 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1880 if (!dt_try_as_dir(env, dt_object_child(dt)))
1881 GOTO(out, rc = -EINVAL);
1883 rec->rec_type = S_IFDIR;
1884 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1885 struct dt_object *dto = lo->ldo_stripe[i];
1886 char *stripe_name = info->lti_key;
1887 struct lu_name *sname;
1888 struct linkea_data ldata = { NULL };
1889 struct lu_buf linkea_buf;
1891 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
1895 /* directory split skip create for existing stripes */
1896 if (!(lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) {
1897 rc = lod_sub_declare_create(env, dto, attr, NULL, dof,
1902 if (!dt_try_as_dir(env, dto))
1903 GOTO(out, rc = -EINVAL);
1905 rc = lod_sub_declare_ref_add(env, dto, th);
1909 rec->rec_fid = lu_object_fid(&dto->do_lu);
1910 rc = lod_sub_declare_insert(env, dto,
1911 (const struct dt_rec *)rec,
1912 (const struct dt_key *)dot,
1917 /* master stripe FID will be put to .. */
1918 rec->rec_fid = lu_object_fid(&dt->do_lu);
1919 rc = lod_sub_declare_insert(env, dto,
1920 (const struct dt_rec *)rec,
1921 (const struct dt_key *)dotdot,
1926 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1928 snprintf(stripe_name, sizeof(info->lti_key),
1930 PFID(lu_object_fid(&dto->do_lu)),
1933 snprintf(stripe_name, sizeof(info->lti_key),
1935 PFID(lu_object_fid(&dto->do_lu)), i);
1937 sname = lod_name_get(env, stripe_name,
1938 strlen(stripe_name));
1939 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1940 sname, lu_object_fid(&dt->do_lu));
1944 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1945 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1946 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
1947 XATTR_NAME_LINK, 0, th);
1951 rec->rec_fid = lu_object_fid(&dto->do_lu);
1952 rc = lod_sub_declare_insert(env, dt_object_child(dt),
1953 (const struct dt_rec *)rec,
1954 (const struct dt_key *)stripe_name, th);
1958 rc = lod_sub_declare_ref_add(env, dt_object_child(dt),
1964 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1965 cfs_fail_val != i) {
1966 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1968 slave_lmm->lmv_master_mdt_index =
1971 slave_lmm->lmv_master_mdt_index =
1973 rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
1974 XATTR_NAME_LMV, 0, th);
1980 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
1981 &lmv_buf, XATTR_NAME_LMV, 0, th);
1985 if (slave_lmm != NULL)
1986 OBD_FREE_PTR(slave_lmm);
1992 * Allocate a striping on a predefined set of MDTs.
1994 * Allocates new striping using the MDT index range provided by the data from
1995 * the lum_obejcts contained in the lmv_user_md passed to this method if
1996 * \a is_specific is true; or allocates new layout starting from MDT index in
1997 * lo->ldo_dir_stripe_offset. The exact order of MDTs is not important and
1998 * varies depending on MDT status. The number of stripes needed and stripe
1999 * offset are taken from the object. If that number cannot be met, then the
2000 * function returns an error and then it's the caller's responsibility to
2001 * release the stripes allocated. All the internal structures are protected,
2002 * but no concurrent allocation is allowed on the same objects.
2004 * \param[in] env execution environment for this thread
2005 * \param[in] lo LOD object
2006 * \param[out] stripes striping created
2007 * \param[out] mdt_indices MDT indices of striping created
2008 * \param[in] is_specific true if the MDTs are provided by lum; false if
2009 * only the starting MDT index is provided
2011 * \retval positive stripes allocated, including the first stripe allocated
2013 * \retval negative errno on failure
2015 static int lod_mdt_alloc_specific(const struct lu_env *env,
2016 struct lod_object *lo,
2017 struct dt_object **stripes,
2018 __u32 *mdt_indices, bool is_specific)
2020 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
2021 struct lu_tgt_descs *ltd = &lod->lod_mdt_descs;
2022 struct lu_tgt_desc *tgt = NULL;
2023 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2024 struct dt_device *tgt_dt = NULL;
2025 struct lu_fid fid = { 0 };
2026 struct dt_object *dto;
2028 u32 stripe_count = lo->ldo_dir_stripe_count;
2034 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
2035 if (stripe_count > 1)
2036 /* Set the start index for the 2nd stripe allocation */
2037 mdt_indices[1] = (mdt_indices[0] + 1) %
2038 (lod->lod_remote_mdt_count + 1);
2040 for (; stripe_idx < stripe_count; stripe_idx++) {
2041 /* Try to find next avaible target */
2042 idx = mdt_indices[stripe_idx];
2043 for (j = 0; j < lod->lod_remote_mdt_count;
2044 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
2045 bool already_allocated = false;
2048 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
2049 idx, lod->lod_remote_mdt_count + 1, stripe_idx);
2051 if (likely(!is_specific &&
2052 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
2053 /* check whether the idx already exists
2054 * in current allocated array */
2055 for (k = 0; k < stripe_idx; k++) {
2056 if (mdt_indices[k] == idx) {
2057 already_allocated = true;
2062 if (already_allocated)
2066 /* Sigh, this index is not in the bitmap, let's check
2067 * next available target */
2068 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) &&
2069 idx != master_index)
2072 if (idx == master_index) {
2073 /* Allocate the FID locally */
2074 tgt_dt = lod->lod_child;
2075 rc = dt_fid_alloc(env, tgt_dt, &fid, NULL,
2082 /* check the status of the OSP */
2083 tgt = LTD_TGT(ltd, idx);
2087 tgt_dt = tgt->ltd_tgt;
2088 if (!tgt->ltd_active)
2089 /* this OSP doesn't feel well */
2092 rc = dt_fid_alloc(env, tgt_dt, &fid, NULL, NULL);
2099 /* Can not allocate more stripes */
2100 if (j == lod->lod_remote_mdt_count) {
2101 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
2102 lod2obd(lod)->obd_name, stripe_count,
2107 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
2108 idx, stripe_idx, PFID(&fid));
2109 mdt_indices[stripe_idx] = idx;
2110 /* Set the start index for next stripe allocation */
2111 if (!is_specific && stripe_idx < stripe_count - 1) {
2113 * for large dir test, put all other slaves on one
2114 * remote MDT, otherwise we may save too many local
2115 * slave locks which will exceed RS_MAX_LOCKS.
2117 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)))
2119 mdt_indices[stripe_idx + 1] = (idx + 1) %
2120 (lod->lod_remote_mdt_count + 1);
2122 /* tgt_dt and fid must be ready after search avaible OSP
2123 * in the above loop */
2124 LASSERT(tgt_dt != NULL);
2125 LASSERT(fid_is_sane(&fid));
2127 /* fail a remote stripe FID allocation */
2128 if (stripe_idx && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_FID))
2131 dto = dt_locate_at(env, tgt_dt, &fid,
2132 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
2139 stripes[stripe_idx] = dto;
2145 for (j = 1; j < stripe_idx; j++) {
2146 LASSERT(stripes[j] != NULL);
2147 dt_object_put(env, stripes[j]);
2153 static int lod_prep_md_striped_create(const struct lu_env *env,
2154 struct dt_object *dt,
2155 struct lu_attr *attr,
2156 const struct lmv_user_md_v1 *lum,
2157 struct dt_object_format *dof,
2160 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
2161 struct lod_object *lo = lod_dt_obj(dt);
2162 struct dt_object **stripes;
2163 struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2164 struct lu_fid fid = { 0 };
2171 /* The lum has been verifed in lod_verify_md_striping */
2172 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
2173 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
2175 stripe_count = lo->ldo_dir_stripe_count;
2177 OBD_ALLOC(stripes, sizeof(stripes[0]) * stripe_count);
2181 /* Allocate the first stripe locally */
2182 rc = dt_fid_alloc(env, lod->lod_child, &fid, NULL, NULL);
2186 stripes[0] = dt_locate_at(env, lod->lod_child, &fid,
2187 dt->do_lu.lo_dev->ld_site->ls_top_dev, &conf);
2188 if (IS_ERR(stripes[0]))
2189 GOTO(out, rc = PTR_ERR(stripes[0]));
2191 if (lo->ldo_dir_stripe_offset == LMV_OFFSET_DEFAULT) {
2192 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
2193 rc = lod_mdt_alloc_qos(env, lo, stripes, 1, stripe_count);
2195 rc = lod_mdt_alloc_rr(env, lo, stripes, 1,
2199 bool is_specific = false;
2201 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
2203 GOTO(out, rc = -ENOMEM);
2205 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
2207 for (i = 0; i < stripe_count; i++)
2209 le32_to_cpu(lum->lum_objects[i].lum_mds);
2212 /* stripe 0 is local */
2214 lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
2215 rc = lod_mdt_alloc_specific(env, lo, stripes, idx_array,
2217 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
2225 lo->ldo_dir_striped = 1;
2226 lo->ldo_stripe = stripes;
2227 lo->ldo_dir_stripe_count = rc;
2228 lo->ldo_dir_stripes_allocated = stripe_count;
2230 lo->ldo_dir_stripe_loaded = 1;
2232 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2234 lod_striping_free(env, lo);
2240 if (!IS_ERR_OR_NULL(stripes[0]))
2241 dt_object_put(env, stripes[0]);
2242 for (i = 1; i < stripe_count; i++)
2243 LASSERT(!stripes[i]);
2244 OBD_FREE(stripes, sizeof(stripes[0]) * stripe_count);
2251 * Alloc cached foreign LMV
2253 * \param[in] lo object
2254 * \param[in] size size of foreign LMV
2256 * \retval 0 on success
2257 * \retval negative if failed
2259 int lod_alloc_foreign_lmv(struct lod_object *lo, size_t size)
2261 OBD_ALLOC_LARGE(lo->ldo_foreign_lmv, size);
2262 if (lo->ldo_foreign_lmv == NULL)
2264 lo->ldo_foreign_lmv_size = size;
2265 lo->ldo_dir_is_foreign = 1;
2271 * Declare create striped md object.
2273 * The function declares intention to create a striped directory. This is a
2274 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2275 * is to verify pattern \a lum_buf is good. Check that function for the details.
2277 * \param[in] env execution environment
2278 * \param[in] dt object
2279 * \param[in] attr attributes to initialize the objects with
2280 * \param[in] lum_buf a pattern specifying the number of stripes and
2282 * \param[in] dof type of objects to be created
2283 * \param[in] th transaction handle
2285 * \retval 0 on success
2286 * \retval negative if failed
2289 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2290 struct dt_object *dt,
2291 struct lu_attr *attr,
2292 const struct lu_buf *lum_buf,
2293 struct dt_object_format *dof,
2296 struct lod_object *lo = lod_dt_obj(dt);
2297 struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
2301 LASSERT(lum != NULL);
2303 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
2304 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
2305 (int)le32_to_cpu(lum->lum_stripe_offset));
2307 if (lo->ldo_dir_stripe_count == 0) {
2308 if (lo->ldo_dir_is_foreign) {
2309 rc = lod_alloc_foreign_lmv(lo, lum_buf->lb_len);
2312 memcpy(lo->ldo_foreign_lmv, lum, lum_buf->lb_len);
2313 lo->ldo_dir_stripe_loaded = 1;
2318 /* prepare dir striped objects */
2319 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2321 /* failed to create striping, let's reset
2322 * config so that others don't get confused */
2323 lod_striping_free(env, lo);
2331 * Set or replace striped directory layout, and LFSCK may set layout on a plain
2332 * directory, so don't check stripe count.
2334 * \param[in] env execution environment
2335 * \param[in] dt target object
2336 * \param[in] buf LMV buf which contains source stripe fids
2337 * \param[in] fl set or replace
2338 * \param[in] th transaction handle
2340 * \retval 0 on success
2341 * \retval negative if failed
2343 static int lod_dir_layout_set(const struct lu_env *env,
2344 struct dt_object *dt,
2345 const struct lu_buf *buf,
2349 struct dt_object *next = dt_object_child(dt);
2350 struct lod_object *lo = lod_dt_obj(dt);
2351 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2352 struct lmv_mds_md_v1 *lmv = buf->lb_buf;
2353 struct lmv_mds_md_v1 *slave_lmv;
2354 struct lu_buf slave_buf;
2360 if (!lmv_is_sane2(lmv))
2363 /* adjust hash for dir merge, which may not be set in user command */
2364 if (lmv_is_merging(lmv) && !lmv->lmv_migrate_hash)
2365 lmv->lmv_merge_hash =
2366 lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
2368 LMV_DEBUG(D_INFO, lmv, "set");
2370 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LMV, fl, th);
2374 /* directory restripe may update stripe LMV directly */
2375 if (!lo->ldo_dir_stripe_count)
2378 lo->ldo_dir_hash_type = le32_to_cpu(lmv->lmv_hash_type);
2379 lo->ldo_dir_migrate_offset = le32_to_cpu(lmv->lmv_migrate_offset);
2380 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_migrate_hash);
2381 lo->ldo_dir_layout_version = le32_to_cpu(lmv->lmv_layout_version);
2383 OBD_ALLOC_PTR(slave_lmv);
2387 lod_prep_slave_lmv_md(slave_lmv, lmv);
2388 slave_buf.lb_buf = slave_lmv;
2389 slave_buf.lb_len = sizeof(*slave_lmv);
2391 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2392 if (!lo->ldo_stripe[i])
2395 if (!dt_object_exists(lo->ldo_stripe[i]))
2398 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], &slave_buf,
2399 XATTR_NAME_LMV, fl, th);
2404 OBD_FREE_PTR(slave_lmv);
2410 * Implementation of dt_object_operations::do_declare_xattr_set.
2412 * Used with regular (non-striped) objects. Basically it
2413 * initializes the striping information and applies the
2414 * change to all the stripes.
2416 * \see dt_object_operations::do_declare_xattr_set() in the API description
2419 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2420 struct dt_object *dt,
2421 const struct lu_buf *buf,
2422 const char *name, int fl,
2425 struct dt_object *next = dt_object_child(dt);
2426 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2427 struct lod_object *lo = lod_dt_obj(dt);
2432 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2433 struct lmv_user_md_v1 *lum;
2435 LASSERT(buf != NULL && buf->lb_buf != NULL);
2437 rc = lod_verify_md_striping(d, lum);
2440 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2441 rc = lod_verify_striping(env, d, lo, buf, false);
2446 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2450 /* Note: Do not set LinkEA on sub-stripes, otherwise
2451 * it will confuse the fid2path process(see mdt_path_current()).
2452 * The linkEA between master and sub-stripes is set in
2453 * lod_xattr_set_lmv(). */
2454 if (strcmp(name, XATTR_NAME_LINK) == 0)
2457 /* set xattr to each stripes, if needed */
2458 rc = lod_striping_load(env, lo);
2462 if (lo->ldo_dir_stripe_count == 0)
2465 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2466 if (!lo->ldo_stripe[i])
2469 if (!dt_object_exists(lo->ldo_stripe[i]))
2472 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2482 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2483 struct lod_object *lo,
2484 struct dt_object *dt, struct thandle *th,
2485 int comp_idx, int stripe_idx,
2486 struct lod_obj_stripe_cb_data *data)
2488 struct lod_thread_info *info = lod_env_info(env);
2489 struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
2490 struct filter_fid *ff = &info->lti_ff;
2491 struct lu_buf *buf = &info->lti_buf;
2495 buf->lb_len = sizeof(*ff);
2496 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2504 * locd_buf is set if it's called by dir migration, which doesn't check
2507 if (data->locd_buf) {
2508 memset(ff, 0, sizeof(*ff));
2509 ff->ff_parent = *(struct lu_fid *)data->locd_buf->lb_buf;
2511 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
2513 if (lu_fid_eq(lod_object_fid(lo), &ff->ff_parent) &&
2514 ff->ff_layout.ol_comp_id == comp->llc_id)
2517 memset(ff, 0, sizeof(*ff));
2518 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2521 /* rewrite filter_fid */
2522 ff->ff_parent.f_ver = stripe_idx;
2523 ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
2524 ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
2525 ff->ff_layout.ol_comp_id = comp->llc_id;
2526 ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
2527 ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
2528 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
2530 if (data->locd_declare)
2531 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2532 LU_XATTR_REPLACE, th);
2534 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2535 LU_XATTR_REPLACE, th);
2541 * Reset parent FID on OST object
2543 * Replace parent FID with @dt object FID, which is only called during migration
2544 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2545 * the FID is changed.
2547 * \param[in] env execution environment
2548 * \param[in] dt dt_object whose stripes's parent FID will be reset
2549 * \parem[in] th thandle
2550 * \param[in] declare if it is declare
2552 * \retval 0 if reset succeeds
2553 * \retval negative errno if reset fails
2555 static int lod_replace_parent_fid(const struct lu_env *env,
2556 struct dt_object *dt,
2557 const struct lu_buf *buf,
2558 struct thandle *th, bool declare)
2560 struct lod_object *lo = lod_dt_obj(dt);
2561 struct lod_thread_info *info = lod_env_info(env);
2562 struct filter_fid *ff;
2563 struct lod_obj_stripe_cb_data data = { { 0 } };
2567 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2569 /* set xattr to each stripes, if needed */
2570 rc = lod_striping_load(env, lo);
2574 if (!lod_obj_is_striped(dt))
2577 if (info->lti_ea_store_size < sizeof(*ff)) {
2578 rc = lod_ea_store_resize(info, sizeof(*ff));
2583 data.locd_declare = declare;
2584 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2585 data.locd_buf = buf;
2586 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2591 inline __u16 lod_comp_entry_stripe_count(struct lod_object *lo,
2592 struct lod_layout_component *entry,
2595 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2599 else if (lod_comp_inited(entry))
2600 return entry->llc_stripe_count;
2601 else if ((__u16)-1 == entry->llc_stripe_count)
2602 return lod->lod_ost_count;
2604 return lod_get_stripe_count(lod, lo,
2605 entry->llc_stripe_count, false);
2608 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2610 int magic, size = 0, i;
2611 struct lod_layout_component *comp_entries;
2613 bool is_composite, is_foreign = false;
2616 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2617 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2619 lo->ldo_def_striping->lds_def_striping_is_composite;
2621 comp_cnt = lo->ldo_comp_cnt;
2622 comp_entries = lo->ldo_comp_entries;
2623 is_composite = lo->ldo_is_composite;
2624 is_foreign = lo->ldo_is_foreign;
2628 return lo->ldo_foreign_lov_size;
2630 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2632 size = sizeof(struct lov_comp_md_v1) +
2633 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2634 LASSERT(size % sizeof(__u64) == 0);
2637 for (i = 0; i < comp_cnt; i++) {
2640 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2641 stripe_count = lod_comp_entry_stripe_count(lo, &comp_entries[i],
2643 if (!is_dir && is_composite)
2644 lod_comp_shrink_stripe_count(&comp_entries[i],
2647 size += lov_user_md_size(stripe_count, magic);
2648 LASSERT(size % sizeof(__u64) == 0);
2654 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2655 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2658 * \param[in] env execution environment
2659 * \param[in] dt dt_object to add components on
2660 * \param[in] buf buffer contains components to be added
2661 * \parem[in] th thandle
2663 * \retval 0 on success
2664 * \retval negative errno on failure
2666 static int lod_declare_layout_add(const struct lu_env *env,
2667 struct dt_object *dt,
2668 const struct lu_buf *buf,
2671 struct lod_thread_info *info = lod_env_info(env);
2672 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2673 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2674 struct dt_object *next = dt_object_child(dt);
2675 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
2676 struct lod_object *lo = lod_dt_obj(dt);
2677 struct lov_user_md_v3 *v3;
2678 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2680 int i, rc, array_cnt, old_array_cnt;
2683 LASSERT(lo->ldo_is_composite);
2685 if (lo->ldo_flr_state != LCM_FL_NONE)
2688 rc = lod_verify_striping(env, d, lo, buf, false);
2692 magic = comp_v1->lcm_magic;
2693 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2694 lustre_swab_lov_comp_md_v1(comp_v1);
2695 magic = comp_v1->lcm_magic;
2698 if (magic != LOV_USER_MAGIC_COMP_V1)
2701 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2702 OBD_ALLOC(comp_array, sizeof(*comp_array) * array_cnt);
2703 if (comp_array == NULL)
2706 memcpy(comp_array, lo->ldo_comp_entries,
2707 sizeof(*comp_array) * lo->ldo_comp_cnt);
2709 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2710 struct lov_user_md_v1 *v1;
2711 struct lu_extent *ext;
2713 v1 = (struct lov_user_md *)((char *)comp_v1 +
2714 comp_v1->lcm_entries[i].lcme_offset);
2715 ext = &comp_v1->lcm_entries[i].lcme_extent;
2717 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2718 lod_comp->llc_extent.e_start = ext->e_start;
2719 lod_comp->llc_extent.e_end = ext->e_end;
2720 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2721 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2723 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2724 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2725 lod_adjust_stripe_info(lod_comp, desc, 0);
2727 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2728 v3 = (struct lov_user_md_v3 *) v1;
2729 if (v3->lmm_pool_name[0] != '\0') {
2730 rc = lod_set_pool(&lod_comp->llc_pool,
2738 old_array = lo->ldo_comp_entries;
2739 old_array_cnt = lo->ldo_comp_cnt;
2741 lo->ldo_comp_entries = comp_array;
2742 lo->ldo_comp_cnt = array_cnt;
2744 /* No need to increase layout generation here, it will be increased
2745 * later when generating component ID for the new components */
2747 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2748 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2749 XATTR_NAME_LOV, 0, th);
2751 lo->ldo_comp_entries = old_array;
2752 lo->ldo_comp_cnt = old_array_cnt;
2756 OBD_FREE(old_array, sizeof(*lod_comp) * old_array_cnt);
2758 LASSERT(lo->ldo_mirror_count == 1);
2759 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2764 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2765 lod_comp = &comp_array[i];
2766 if (lod_comp->llc_pool != NULL) {
2767 OBD_FREE(lod_comp->llc_pool,
2768 strlen(lod_comp->llc_pool) + 1);
2769 lod_comp->llc_pool = NULL;
2772 OBD_FREE(comp_array, sizeof(*comp_array) * array_cnt);
2777 * lod_last_non_stale_mirror() - Check if a mirror is the last non-stale mirror.
2778 * @mirror_id: Mirror id to be checked.
2781 * This function checks if a mirror with specified @mirror_id is the last
2782 * non-stale mirror of a LOD object @lo.
2784 * Return: true or false.
2787 bool lod_last_non_stale_mirror(__u16 mirror_id, struct lod_object *lo)
2789 struct lod_layout_component *lod_comp;
2790 bool has_stale_flag;
2793 for (i = 0; i < lo->ldo_mirror_count; i++) {
2794 if (lo->ldo_mirrors[i].lme_id == mirror_id ||
2795 lo->ldo_mirrors[i].lme_stale)
2798 has_stale_flag = false;
2799 lod_foreach_mirror_comp(lod_comp, lo, i) {
2800 if (lod_comp->llc_flags & LCME_FL_STALE) {
2801 has_stale_flag = true;
2805 if (!has_stale_flag)
2813 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2814 * the '$field' can only be 'flags' now. The xattr value is binary
2815 * lov_comp_md_v1 which contains the component ID(s) and the value of
2816 * the field to be modified.
2817 * Please update allowed_lustre_lov macro if $field groks more values
2820 * \param[in] env execution environment
2821 * \param[in] dt dt_object to be modified
2822 * \param[in] op operation string, like "set.flags"
2823 * \param[in] buf buffer contains components to be set
2824 * \parem[in] th thandle
2826 * \retval 0 on success
2827 * \retval negative errno on failure
2829 static int lod_declare_layout_set(const struct lu_env *env,
2830 struct dt_object *dt,
2831 char *op, const struct lu_buf *buf,
2834 struct lod_layout_component *lod_comp;
2835 struct lod_thread_info *info = lod_env_info(env);
2836 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2837 struct lod_object *lo = lod_dt_obj(dt);
2838 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2841 bool changed = false;
2844 /* Please update allowed_lustre_lov macro if op
2845 * groks more values in the future
2847 if (strcmp(op, "set.flags") != 0) {
2848 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2849 lod2obd(d)->obd_name, op);
2853 magic = comp_v1->lcm_magic;
2854 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2855 lustre_swab_lov_comp_md_v1(comp_v1);
2856 magic = comp_v1->lcm_magic;
2859 if (magic != LOV_USER_MAGIC_COMP_V1)
2862 if (comp_v1->lcm_entry_count == 0) {
2863 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2864 lod2obd(d)->obd_name);
2868 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2869 __u32 id = comp_v1->lcm_entries[i].lcme_id;
2870 __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
2871 __u32 mirror_flag = flags & LCME_MIRROR_FLAGS;
2872 __u16 mirror_id = mirror_id_of(id);
2873 bool neg = flags & LCME_FL_NEG;
2875 if (flags & LCME_FL_INIT) {
2877 lod_striping_free(env, lo);
2881 flags &= ~(LCME_MIRROR_FLAGS | LCME_FL_NEG);
2882 for (j = 0; j < lo->ldo_comp_cnt; j++) {
2883 lod_comp = &lo->ldo_comp_entries[j];
2885 /* lfs only put one flag in each entry */
2886 if ((flags && id != lod_comp->llc_id) ||
2887 (mirror_flag && mirror_id !=
2888 mirror_id_of(lod_comp->llc_id)))
2893 lod_comp->llc_flags &= ~flags;
2895 lod_comp->llc_flags &= ~mirror_flag;
2898 if ((flags & LCME_FL_STALE) &&
2899 lod_last_non_stale_mirror(mirror_id,
2902 lod_comp->llc_flags |= flags;
2905 lod_comp->llc_flags |= mirror_flag;
2906 if (mirror_flag & LCME_FL_NOSYNC)
2907 lod_comp->llc_timestamp =
2908 ktime_get_real_seconds();
2916 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
2917 lod2obd(d)->obd_name);
2921 lod_obj_inc_layout_gen(lo);
2923 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2924 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
2925 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
2930 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
2931 * and the xattr value is a unique component ID or a special lcme_id.
2933 * \param[in] env execution environment
2934 * \param[in] dt dt_object to be operated on
2935 * \param[in] buf buffer contains component ID or lcme_id
2936 * \parem[in] th thandle
2938 * \retval 0 on success
2939 * \retval negative errno on failure
2941 static int lod_declare_layout_del(const struct lu_env *env,
2942 struct dt_object *dt,
2943 const struct lu_buf *buf,
2946 struct lod_thread_info *info = lod_env_info(env);
2947 struct dt_object *next = dt_object_child(dt);
2948 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2949 struct lod_object *lo = lod_dt_obj(dt);
2950 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2951 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2952 __u32 magic, id, flags, neg_flags = 0;
2956 LASSERT(lo->ldo_is_composite);
2958 if (lo->ldo_flr_state != LCM_FL_NONE)
2961 magic = comp_v1->lcm_magic;
2962 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2963 lustre_swab_lov_comp_md_v1(comp_v1);
2964 magic = comp_v1->lcm_magic;
2967 if (magic != LOV_USER_MAGIC_COMP_V1)
2970 id = comp_v1->lcm_entries[0].lcme_id;
2971 flags = comp_v1->lcm_entries[0].lcme_flags;
2973 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
2974 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
2975 lod2obd(d)->obd_name, id, flags);
2979 if (id != LCME_ID_INVAL && flags != 0) {
2980 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
2981 lod2obd(d)->obd_name);
2985 if (id == LCME_ID_INVAL && !flags) {
2986 CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
2987 lod2obd(d)->obd_name);
2991 if (flags & LCME_FL_NEG) {
2992 neg_flags = flags & ~LCME_FL_NEG;
2996 left = lo->ldo_comp_cnt;
3000 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3001 struct lod_layout_component *lod_comp;
3003 lod_comp = &lo->ldo_comp_entries[i];
3005 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
3007 else if (flags && !(flags & lod_comp->llc_flags))
3009 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
3012 if (left != (i + 1)) {
3013 CDEBUG(D_LAYOUT, "%s: this deletion will create "
3014 "a hole.\n", lod2obd(d)->obd_name);
3019 /* Mark the component as deleted */
3020 lod_comp->llc_id = LCME_ID_INVAL;
3022 /* Not instantiated component */
3023 if (lod_comp->llc_stripe == NULL)
3026 LASSERT(lod_comp->llc_stripe_count > 0);
3027 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
3028 struct dt_object *obj = lod_comp->llc_stripe[j];
3032 rc = lod_sub_declare_destroy(env, obj, th);
3038 LASSERTF(left >= 0, "left = %d\n", left);
3039 if (left == lo->ldo_comp_cnt) {
3040 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
3041 lod2obd(d)->obd_name, id);
3045 memset(attr, 0, sizeof(*attr));
3046 attr->la_valid = LA_SIZE;
3047 rc = lod_sub_declare_attr_set(env, next, attr, th);
3052 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3053 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
3054 XATTR_NAME_LOV, 0, th);
3056 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
3063 * Declare layout add/set/del operations issued by special xattr names:
3065 * XATTR_LUSTRE_LOV.add add component(s) to existing file
3066 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
3067 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
3069 * \param[in] env execution environment
3070 * \param[in] dt object
3071 * \param[in] name name of xattr
3072 * \param[in] buf lu_buf contains xattr value
3073 * \param[in] th transaction handle
3075 * \retval 0 on success
3076 * \retval negative if failed
3078 static int lod_declare_modify_layout(const struct lu_env *env,
3079 struct dt_object *dt,
3081 const struct lu_buf *buf,
3084 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3085 struct lod_object *lo = lod_dt_obj(dt);
3087 int rc, len = strlen(XATTR_LUSTRE_LOV);
3090 LASSERT(dt_object_exists(dt));
3092 if (strlen(name) <= len || name[len] != '.') {
3093 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
3094 lod2obd(d)->obd_name, name);
3099 rc = lod_striping_load(env, lo);
3103 /* the layout to be modified must be a composite layout */
3104 if (!lo->ldo_is_composite) {
3105 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
3106 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3107 GOTO(unlock, rc = -EINVAL);
3110 op = (char *)name + len;
3111 if (strcmp(op, "add") == 0) {
3112 rc = lod_declare_layout_add(env, dt, buf, th);
3113 } else if (strcmp(op, "del") == 0) {
3114 rc = lod_declare_layout_del(env, dt, buf, th);
3115 } else if (strncmp(op, "set", strlen("set")) == 0) {
3116 rc = lod_declare_layout_set(env, dt, op, buf, th);
3118 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
3119 lod2obd(d)->obd_name, name);
3120 GOTO(unlock, rc = -ENOTSUPP);
3124 lod_striping_free(env, lo);
3130 * Convert a plain file lov_mds_md to a composite layout.
3132 * \param[in,out] info the thread info::lti_ea_store buffer contains little
3133 * endian plain file layout
3135 * \retval 0 on success, <0 on failure
3137 static int lod_layout_convert(struct lod_thread_info *info)
3139 struct lov_mds_md *lmm = info->lti_ea_store;
3140 struct lov_mds_md *lmm_save;
3141 struct lov_comp_md_v1 *lcm;
3142 struct lov_comp_md_entry_v1 *lcme;
3148 /* realloc buffer to a composite layout which contains one component */
3149 blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3150 le32_to_cpu(lmm->lmm_magic));
3151 size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
3153 OBD_ALLOC_LARGE(lmm_save, blob_size);
3155 GOTO(out, rc = -ENOMEM);
3157 memcpy(lmm_save, lmm, blob_size);
3159 if (info->lti_ea_store_size < size) {
3160 rc = lod_ea_store_resize(info, size);
3165 lcm = info->lti_ea_store;
3166 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
3167 lcm->lcm_size = cpu_to_le32(size);
3168 lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
3169 lmm_save->lmm_layout_gen));
3170 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
3171 lcm->lcm_entry_count = cpu_to_le16(1);
3172 lcm->lcm_mirror_count = 0;
3174 lcme = &lcm->lcm_entries[0];
3175 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
3176 lcme->lcme_extent.e_start = 0;
3177 lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
3178 lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
3179 lcme->lcme_size = cpu_to_le32(blob_size);
3181 memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
3186 OBD_FREE_LARGE(lmm_save, blob_size);
3191 * Merge layouts to form a mirrored file.
3193 static int lod_declare_layout_merge(const struct lu_env *env,
3194 struct dt_object *dt, const struct lu_buf *mbuf,
3197 struct lod_thread_info *info = lod_env_info(env);
3198 struct lu_buf *buf = &info->lti_buf;
3199 struct lod_object *lo = lod_dt_obj(dt);
3200 struct lov_comp_md_v1 *lcm;
3201 struct lov_comp_md_v1 *cur_lcm;
3202 struct lov_comp_md_v1 *merge_lcm;
3203 struct lov_comp_md_entry_v1 *lcme;
3204 struct lov_mds_md_v1 *lmm;
3207 __u16 cur_entry_count;
3208 __u16 merge_entry_count;
3210 __u16 mirror_id = 0;
3217 merge_lcm = mbuf->lb_buf;
3218 if (mbuf->lb_len < sizeof(*merge_lcm))
3221 /* must be an existing layout from disk */
3222 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
3225 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
3227 /* do not allow to merge two mirrored files */
3228 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
3231 /* verify the target buffer */
3232 rc = lod_get_lov_ea(env, lo);
3234 RETURN(rc ? : -ENODATA);
3236 cur_lcm = info->lti_ea_store;
3237 switch (le32_to_cpu(cur_lcm->lcm_magic)) {
3240 rc = lod_layout_convert(info);
3242 case LOV_MAGIC_COMP_V1:
3252 /* info->lti_ea_store could be reallocated in lod_layout_convert() */
3253 cur_lcm = info->lti_ea_store;
3254 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
3256 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
3257 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
3258 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
3261 /* size of new layout */
3262 size = le32_to_cpu(cur_lcm->lcm_size) +
3263 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
3265 memset(buf, 0, sizeof(*buf));
3266 lu_buf_alloc(buf, size);
3267 if (buf->lb_buf == NULL)
3271 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
3273 offset = sizeof(*lcm) +
3274 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
3275 for (i = 0; i < cur_entry_count; i++) {
3276 struct lov_comp_md_entry_v1 *cur_lcme;
3278 lcme = &lcm->lcm_entries[i];
3279 cur_lcme = &cur_lcm->lcm_entries[i];
3281 lcme->lcme_offset = cpu_to_le32(offset);
3282 memcpy((char *)lcm + offset,
3283 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
3284 le32_to_cpu(lcme->lcme_size));
3286 offset += le32_to_cpu(lcme->lcme_size);
3288 if (mirror_count == 1 &&
3289 mirror_id_of(le32_to_cpu(lcme->lcme_id)) == 0) {
3290 /* Add mirror from a non-flr file, create new mirror ID.
3291 * Otherwise, keep existing mirror's component ID, used
3292 * for mirror extension.
3294 id = pflr_id(1, i + 1);
3295 lcme->lcme_id = cpu_to_le32(id);
3298 id = max(le32_to_cpu(lcme->lcme_id), id);
3301 mirror_id = mirror_id_of(id) + 1;
3303 /* check if first entry in new layout is DOM */
3304 lmm = (struct lov_mds_md_v1 *)((char *)merge_lcm +
3305 merge_lcm->lcm_entries[0].lcme_offset);
3306 merge_has_dom = lov_pattern(le32_to_cpu(lmm->lmm_pattern)) ==
3309 for (i = 0; i < merge_entry_count; i++) {
3310 struct lov_comp_md_entry_v1 *merge_lcme;
3312 merge_lcme = &merge_lcm->lcm_entries[i];
3313 lcme = &lcm->lcm_entries[cur_entry_count + i];
3315 *lcme = *merge_lcme;
3316 lcme->lcme_offset = cpu_to_le32(offset);
3317 if (merge_has_dom && i == 0)
3318 lcme->lcme_flags |= cpu_to_le32(LCME_FL_STALE);
3320 id = pflr_id(mirror_id, i + 1);
3321 lcme->lcme_id = cpu_to_le32(id);
3323 memcpy((char *)lcm + offset,
3324 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
3325 le32_to_cpu(lcme->lcme_size));
3327 offset += le32_to_cpu(lcme->lcme_size);
3330 /* fixup layout information */
3331 lod_obj_inc_layout_gen(lo);
3332 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3333 lcm->lcm_size = cpu_to_le32(size);
3334 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
3335 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
3336 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NONE)
3337 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
3339 rc = lod_striping_reload(env, lo, buf);
3343 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
3344 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3352 * Split layouts, just set the LOVEA with the layout from mbuf.
3354 static int lod_declare_layout_split(const struct lu_env *env,
3355 struct dt_object *dt, const struct lu_buf *mbuf,
3358 struct lod_object *lo = lod_dt_obj(dt);
3359 struct lov_comp_md_v1 *lcm = mbuf->lb_buf;
3363 lod_obj_inc_layout_gen(lo);
3364 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3366 rc = lod_striping_reload(env, lo, mbuf);
3370 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), mbuf,
3371 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3376 * Implementation of dt_object_operations::do_declare_xattr_set.
3378 * \see dt_object_operations::do_declare_xattr_set() in the API description
3381 * the extension to the API:
3382 * - declaring LOVEA requests striping creation
3383 * - LU_XATTR_REPLACE means layout swap
3385 static int lod_declare_xattr_set(const struct lu_env *env,
3386 struct dt_object *dt,
3387 const struct lu_buf *buf,
3388 const char *name, int fl,
3391 struct dt_object *next = dt_object_child(dt);
3392 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3397 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
3398 if ((S_ISREG(mode) || mode == 0) &&
3399 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE | LU_XATTR_SPLIT)) &&
3400 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3401 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
3403 * this is a request to create object's striping.
3405 * allow to declare predefined striping on a new (!mode) object
3406 * which is supposed to be replay of regular file creation
3407 * (when LOV setting is declared)
3409 * LU_XATTR_REPLACE is set to indicate a layout swap
3411 if (dt_object_exists(dt)) {
3412 rc = dt_attr_get(env, next, attr);
3416 memset(attr, 0, sizeof(*attr));
3417 attr->la_valid = LA_TYPE | LA_MODE;
3418 attr->la_mode = S_IFREG;
3420 rc = lod_declare_striped_create(env, dt, attr, buf, th);
3421 } else if (fl & LU_XATTR_MERGE) {
3422 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3423 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3424 rc = lod_declare_layout_merge(env, dt, buf, th);
3425 } else if (fl & LU_XATTR_SPLIT) {
3426 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3427 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3428 rc = lod_declare_layout_split(env, dt, buf, th);
3429 } else if (S_ISREG(mode) &&
3430 strlen(name) >= sizeof(XATTR_LUSTRE_LOV) + 3 &&
3431 allowed_lustre_lov(name)) {
3433 * this is a request to modify object's striping.
3434 * add/set/del component(s).
3436 if (!dt_object_exists(dt))
3439 rc = lod_declare_modify_layout(env, dt, name, buf, th);
3440 } else if (S_ISDIR(mode)) {
3441 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
3442 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3443 rc = lod_replace_parent_fid(env, dt, buf, th, true);
3445 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
3452 * Apply xattr changes to the object.
3454 * Applies xattr changes to the object and the stripes if the latter exist.
3456 * \param[in] env execution environment
3457 * \param[in] dt object
3458 * \param[in] buf buffer pointing to the new value of xattr
3459 * \param[in] name name of xattr
3460 * \param[in] fl flags
3461 * \param[in] th transaction handle
3463 * \retval 0 on success
3464 * \retval negative if failed
3466 static int lod_xattr_set_internal(const struct lu_env *env,
3467 struct dt_object *dt,
3468 const struct lu_buf *buf,
3469 const char *name, int fl,
3472 struct dt_object *next = dt_object_child(dt);
3473 struct lod_object *lo = lod_dt_obj(dt);
3478 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3479 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3482 /* Note: Do not set LinkEA on sub-stripes, otherwise
3483 * it will confuse the fid2path process(see mdt_path_current()).
3484 * The linkEA between master and sub-stripes is set in
3485 * lod_xattr_set_lmv(). */
3486 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
3489 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3490 if (!lo->ldo_stripe[i])
3493 if (!dt_object_exists(lo->ldo_stripe[i]))
3496 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
3506 * Delete an extended attribute.
3508 * Deletes specified xattr from the object and the stripes if the latter exist.
3510 * \param[in] env execution environment
3511 * \param[in] dt object
3512 * \param[in] name name of xattr
3513 * \param[in] th transaction handle
3515 * \retval 0 on success
3516 * \retval negative if failed
3518 static int lod_xattr_del_internal(const struct lu_env *env,
3519 struct dt_object *dt,
3520 const char *name, struct thandle *th)
3522 struct dt_object *next = dt_object_child(dt);
3523 struct lod_object *lo = lod_dt_obj(dt);
3528 rc = lod_sub_xattr_del(env, next, name, th);
3529 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3532 if (lo->ldo_dir_stripe_count == 0)
3535 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3536 LASSERT(lo->ldo_stripe[i]);
3538 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3547 * Set default striping on a directory.
3549 * Sets specified striping on a directory object unless it matches the default
3550 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3551 * EA. This striping will be used when regular file is being created in this
3554 * \param[in] env execution environment
3555 * \param[in] dt the striped object
3556 * \param[in] buf buffer with the striping
3557 * \param[in] name name of EA
3558 * \param[in] fl xattr flag (see OSD API description)
3559 * \param[in] th transaction handle
3561 * \retval 0 on success
3562 * \retval negative if failed
3564 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
3565 struct dt_object *dt,
3566 const struct lu_buf *buf,
3567 const char *name, int fl,
3570 struct lov_user_md_v1 *lum;
3571 struct lov_user_md_v3 *v3 = NULL;
3572 const char *pool_name = NULL;
3577 LASSERT(buf != NULL && buf->lb_buf != NULL);
3580 switch (lum->lmm_magic) {
3581 case LOV_USER_MAGIC_SPECIFIC:
3582 case LOV_USER_MAGIC_V3:
3584 if (v3->lmm_pool_name[0] != '\0')
3585 pool_name = v3->lmm_pool_name;
3587 case LOV_USER_MAGIC_V1:
3588 /* if { size, offset, count } = { 0, -1, 0 } and no pool
3589 * (i.e. all default values specified) then delete default
3590 * striping from dir. */
3592 "set default striping: sz %u # %u offset %d %s %s\n",
3593 (unsigned)lum->lmm_stripe_size,
3594 (unsigned)lum->lmm_stripe_count,
3595 (int)lum->lmm_stripe_offset,
3596 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
3598 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
3599 lum->lmm_stripe_count,
3600 lum->lmm_stripe_offset,
3603 case LOV_USER_MAGIC_COMP_V1:
3605 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lum;
3606 struct lov_comp_md_entry_v1 *lcme;
3609 comp_cnt = le16_to_cpu(lcm->lcm_entry_count);
3610 for (i = 0; i < comp_cnt; i++) {
3611 lcme = &lcm->lcm_entries[i];
3612 if (lcme->lcme_flags & cpu_to_le32(LCME_FL_EXTENSION)) {
3613 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
3622 CERROR("Invalid magic %x\n", lum->lmm_magic);
3627 rc = lod_xattr_del_internal(env, dt, name, th);
3631 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3638 * Set default striping on a directory object.
3640 * Sets specified striping on a directory object unless it matches the default
3641 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3642 * EA. This striping will be used when a new directory is being created in the
3645 * \param[in] env execution environment
3646 * \param[in] dt the striped object
3647 * \param[in] buf buffer with the striping
3648 * \param[in] name name of EA
3649 * \param[in] fl xattr flag (see OSD API description)
3650 * \param[in] th transaction handle
3652 * \retval 0 on success
3653 * \retval negative if failed
3655 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
3656 struct dt_object *dt,
3657 const struct lu_buf *buf,
3658 const char *name, int fl,
3661 struct lmv_user_md_v1 *lum;
3666 LASSERT(buf != NULL && buf->lb_buf != NULL);
3670 "set default stripe_count # %u stripe_offset %d hash %u\n",
3671 le32_to_cpu(lum->lum_stripe_count),
3672 (int)le32_to_cpu(lum->lum_stripe_offset),
3673 le32_to_cpu(lum->lum_hash_type));
3675 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
3676 le32_to_cpu(lum->lum_stripe_offset)) &&
3677 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
3678 rc = lod_xattr_del_internal(env, dt, name, th);
3682 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3691 * Turn directory into a striped directory.
3693 * During replay the client sends the striping created before MDT
3694 * failure, then the layer above LOD sends this defined striping
3695 * using ->do_xattr_set(), so LOD uses this method to replay creation
3696 * of the stripes. Notice the original information for the striping
3697 * (#stripes, FIDs, etc) was transferred in declare path.
3699 * \param[in] env execution environment
3700 * \param[in] dt the striped object
3701 * \param[in] buf not used currently
3702 * \param[in] name not used currently
3703 * \param[in] fl xattr flag (see OSD API description)
3704 * \param[in] th transaction handle
3706 * \retval 0 on success
3707 * \retval negative if failed
3709 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
3710 const struct lu_buf *buf, const char *name,
3711 int fl, struct thandle *th)
3713 struct lod_object *lo = lod_dt_obj(dt);
3714 struct lod_thread_info *info = lod_env_info(env);
3715 struct lu_attr *attr = &info->lti_attr;
3716 struct dt_object_format *dof = &info->lti_format;
3717 struct lu_buf lmv_buf;
3718 struct lu_buf slave_lmv_buf;
3719 struct lmv_mds_md_v1 *lmm;
3720 struct lmv_mds_md_v1 *slave_lmm = NULL;
3721 struct dt_insert_rec *rec = &info->lti_dt_rec;
3726 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3729 /* The stripes are supposed to be allocated in declare phase,
3730 * if there are no stripes being allocated, it will skip */
3731 if (lo->ldo_dir_stripe_count == 0) {
3732 if (lo->ldo_dir_is_foreign) {
3733 rc = lod_sub_xattr_set(env, dt_object_child(dt), buf,
3734 XATTR_NAME_LMV, fl, th);
3741 rc = dt_attr_get(env, dt_object_child(dt), attr);
3745 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME |
3746 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
3747 dof->dof_type = DFT_DIR;
3749 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
3752 lmm = lmv_buf.lb_buf;
3754 OBD_ALLOC_PTR(slave_lmm);
3755 if (slave_lmm == NULL)
3758 lod_prep_slave_lmv_md(slave_lmm, lmm);
3759 slave_lmv_buf.lb_buf = slave_lmm;
3760 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
3762 rec->rec_type = S_IFDIR;
3763 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3764 struct dt_object *dto = lo->ldo_stripe[i];
3765 char *stripe_name = info->lti_key;
3766 struct lu_name *sname;
3767 struct linkea_data ldata = { NULL };
3768 struct lu_buf linkea_buf;
3770 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
3774 /* fail a remote stripe creation */
3775 if (i && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_CREATE))
3778 /* don't create stripe if:
3779 * 1. it's source stripe of migrating directory
3780 * 2. it's existed stripe of splitting directory
3782 if ((lod_is_migrating(lo) && i >= lo->ldo_dir_migrate_offset) ||
3783 (lod_is_splitting(lo) && i < lo->ldo_dir_split_offset)) {
3784 if (!dt_object_exists(dto))
3785 GOTO(out, rc = -EINVAL);
3787 dt_write_lock(env, dto, DT_TGT_CHILD);
3788 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
3790 dt_write_unlock(env, dto);
3794 rc = lod_sub_ref_add(env, dto, th);
3795 dt_write_unlock(env, dto);
3799 rec->rec_fid = lu_object_fid(&dto->do_lu);
3800 rc = lod_sub_insert(env, dto,
3801 (const struct dt_rec *)rec,
3802 (const struct dt_key *)dot, th);
3807 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
3808 cfs_fail_val != i) {
3809 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
3811 slave_lmm->lmv_master_mdt_index =
3814 slave_lmm->lmv_master_mdt_index =
3817 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
3818 XATTR_NAME_LMV, 0, th);
3823 /* don't insert stripe if it's existed stripe of splitting
3824 * directory (this directory is striped).
3825 * NB, plain directory will insert itself as the first
3828 if (lod_is_splitting(lo) && lo->ldo_dir_split_offset > 1 &&
3829 lo->ldo_dir_split_offset > i)
3832 rec->rec_fid = lu_object_fid(&dt->do_lu);
3833 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
3834 (const struct dt_key *)dotdot, th);
3838 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
3840 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3841 PFID(lu_object_fid(&dto->do_lu)), i + 1);
3843 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3844 PFID(lu_object_fid(&dto->do_lu)), i);
3846 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
3847 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
3848 sname, lu_object_fid(&dt->do_lu));
3852 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
3853 linkea_buf.lb_len = ldata.ld_leh->leh_len;
3854 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
3855 XATTR_NAME_LINK, 0, th);
3859 rec->rec_fid = lu_object_fid(&dto->do_lu);
3860 rc = lod_sub_insert(env, dt_object_child(dt),
3861 (const struct dt_rec *)rec,
3862 (const struct dt_key *)stripe_name, th);
3866 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
3871 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
3872 rc = lod_sub_xattr_set(env, dt_object_child(dt),
3873 &lmv_buf, XATTR_NAME_LMV, fl, th);
3875 if (slave_lmm != NULL)
3876 OBD_FREE_PTR(slave_lmm);
3882 * Helper function to declare/execute creation of a striped directory
3884 * Called in declare/create object path, prepare striping for a directory
3885 * and prepare defaults data striping for the objects to be created in
3886 * that directory. Notice the function calls "declaration" or "execution"
3887 * methods depending on \a declare param. This is a consequence of the
3888 * current approach while we don't have natural distributed transactions:
3889 * we basically execute non-local updates in the declare phase. So, the
3890 * arguments for the both phases are the same and this is the reason for
3891 * this function to exist.
3893 * \param[in] env execution environment
3894 * \param[in] dt object
3895 * \param[in] attr attributes the stripes will be created with
3896 * \param[in] lmu lmv_user_md if MDT indices are specified
3897 * \param[in] dof format of stripes (see OSD API description)
3898 * \param[in] th transaction handle
3899 * \param[in] declare where to call "declare" or "execute" methods
3901 * \retval 0 on success
3902 * \retval negative if failed
3904 static int lod_dir_striping_create_internal(const struct lu_env *env,
3905 struct dt_object *dt,
3906 struct lu_attr *attr,
3907 const struct lu_buf *lmu,
3908 struct dt_object_format *dof,
3912 struct lod_thread_info *info = lod_env_info(env);
3913 struct lod_object *lo = lod_dt_obj(dt);
3914 const struct lod_default_striping *lds = lo->ldo_def_striping;
3918 LASSERT(ergo(lds != NULL,
3919 lds->lds_def_striping_set ||
3920 lds->lds_dir_def_striping_set));
3922 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
3923 lo->ldo_dir_stripe_offset)) {
3925 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3926 int stripe_count = lo->ldo_dir_stripe_count;
3928 if (info->lti_ea_store_size < sizeof(*v1)) {
3929 rc = lod_ea_store_resize(info, sizeof(*v1));
3932 v1 = info->lti_ea_store;
3935 memset(v1, 0, sizeof(*v1));
3936 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3937 v1->lum_stripe_count = cpu_to_le32(stripe_count);
3938 v1->lum_stripe_offset =
3939 cpu_to_le32(lo->ldo_dir_stripe_offset);
3941 info->lti_buf.lb_buf = v1;
3942 info->lti_buf.lb_len = sizeof(*v1);
3943 lmu = &info->lti_buf;
3947 rc = lod_declare_xattr_set_lmv(env, dt, attr, lmu, dof,
3950 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
3955 /* foreign LMV EA case */
3957 struct lmv_foreign_md *lfm = lmu->lb_buf;
3959 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
3960 rc = lod_declare_xattr_set_lmv(env, dt, attr,
3964 if (lo->ldo_dir_is_foreign) {
3965 LASSERT(lo->ldo_foreign_lmv != NULL &&
3966 lo->ldo_foreign_lmv_size > 0);
3967 info->lti_buf.lb_buf = lo->ldo_foreign_lmv;
3968 info->lti_buf.lb_len = lo->ldo_foreign_lmv_size;
3969 lmu = &info->lti_buf;
3970 rc = lod_xattr_set_lmv(env, dt, lmu,
3971 XATTR_NAME_LMV, 0, th);
3976 /* Transfer default LMV striping from the parent */
3977 if (lds != NULL && lds->lds_dir_def_striping_set &&
3978 !(LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
3979 lds->lds_dir_def_stripe_offset) &&
3980 le32_to_cpu(lds->lds_dir_def_hash_type) !=
3981 LMV_HASH_TYPE_UNKNOWN)) {
3982 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3984 if (info->lti_ea_store_size < sizeof(*v1)) {
3985 rc = lod_ea_store_resize(info, sizeof(*v1));
3988 v1 = info->lti_ea_store;
3991 memset(v1, 0, sizeof(*v1));
3992 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3993 v1->lum_stripe_count =
3994 cpu_to_le32(lds->lds_dir_def_stripe_count);
3995 v1->lum_stripe_offset =
3996 cpu_to_le32(lds->lds_dir_def_stripe_offset);
3998 cpu_to_le32(lds->lds_dir_def_hash_type);
4000 info->lti_buf.lb_buf = v1;
4001 info->lti_buf.lb_len = sizeof(*v1);
4003 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4004 XATTR_NAME_DEFAULT_LMV,
4007 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
4009 XATTR_NAME_DEFAULT_LMV, 0,
4015 /* Transfer default LOV striping from the parent */
4016 if (lds != NULL && lds->lds_def_striping_set &&
4017 lds->lds_def_comp_cnt != 0) {
4018 struct lov_mds_md *lmm;
4019 int lmm_size = lod_comp_md_size(lo, true);
4021 if (info->lti_ea_store_size < lmm_size) {
4022 rc = lod_ea_store_resize(info, lmm_size);
4026 lmm = info->lti_ea_store;
4028 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
4032 info->lti_buf.lb_buf = lmm;
4033 info->lti_buf.lb_len = lmm_size;
4036 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4037 XATTR_NAME_LOV, 0, th);
4039 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4040 XATTR_NAME_LOV, 0, th);
4048 static int lod_declare_dir_striping_create(const struct lu_env *env,
4049 struct dt_object *dt,
4050 struct lu_attr *attr,
4052 struct dt_object_format *dof,
4055 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
4059 static int lod_dir_striping_create(const struct lu_env *env,
4060 struct dt_object *dt,
4061 struct lu_attr *attr,
4062 struct dt_object_format *dof,
4065 return lod_dir_striping_create_internal(env, dt, attr, NULL, dof, th,
4070 * Make LOV EA for striped object.
4072 * Generate striping information and store it in the LOV EA of the given
4073 * object. The caller must ensure nobody else is calling the function
4074 * against the object concurrently. The transaction must be started.
4075 * FLDB service must be running as well; it's used to map FID to the target,
4076 * which is stored in LOV EA.
4078 * \param[in] env execution environment for this thread
4079 * \param[in] lo LOD object
4080 * \param[in] th transaction handle
4082 * \retval 0 if LOV EA is stored successfully
4083 * \retval negative error number on failure
4085 static int lod_generate_and_set_lovea(const struct lu_env *env,
4086 struct lod_object *lo,
4089 struct lod_thread_info *info = lod_env_info(env);
4090 struct dt_object *next = dt_object_child(&lo->ldo_obj);
4091 struct lov_mds_md_v1 *lmm;
4097 if (lo->ldo_comp_cnt == 0 && !lo->ldo_is_foreign) {
4098 lod_striping_free(env, lo);
4099 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
4103 lmm_size = lod_comp_md_size(lo, false);
4104 if (info->lti_ea_store_size < lmm_size) {
4105 rc = lod_ea_store_resize(info, lmm_size);
4109 lmm = info->lti_ea_store;
4111 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
4115 info->lti_buf.lb_buf = lmm;
4116 info->lti_buf.lb_len = lmm_size;
4117 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
4118 XATTR_NAME_LOV, 0, th);
4122 static __u32 lod_gen_component_id(struct lod_object *lo,
4123 int mirror_id, int comp_idx);
4126 * Repeat an existing component
4128 * Creates a new layout by replicating an existing component. Uses striping
4129 * policy from previous component as a template for the striping for the new
4132 * New component starts with zero length, will be extended (or removed) before
4133 * returning layout to client.
4135 * NB: Reallocates layout components array (lo->ldo_comp_entries), invalidating
4136 * any pre-existing pointers to components. Handle with care.
4138 * \param[in] env execution environment for this thread
4139 * \param[in,out] lo object to update the layout of
4140 * \param[in] index index of component to copy
4142 * \retval 0 on success
4143 * \retval negative errno on error
4145 static int lod_layout_repeat_comp(const struct lu_env *env,
4146 struct lod_object *lo, int index)
4148 struct lod_layout_component *lod_comp;
4149 struct lod_layout_component *new_comp = NULL;
4150 struct lod_layout_component *comp_array;
4151 int rc = 0, i, new_cnt = lo->ldo_comp_cnt + 1;
4156 lod_comp = &lo->ldo_comp_entries[index];
4157 LASSERT(lod_comp_inited(lod_comp) && lod_comp->llc_id != LCME_ID_INVAL);
4159 CDEBUG(D_LAYOUT, "repeating component %d\n", index);
4161 OBD_ALLOC(comp_array, sizeof(*comp_array) * new_cnt);
4162 if (comp_array == NULL)
4163 GOTO(out, rc = -ENOMEM);
4165 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4166 memcpy(&comp_array[i + offset], &lo->ldo_comp_entries[i],
4167 sizeof(*comp_array));
4169 /* Duplicate this component in to the next slot */
4171 new_comp = &comp_array[i + 1];
4172 memcpy(&comp_array[i + 1], &lo->ldo_comp_entries[i],
4173 sizeof(*comp_array));
4174 /* We must now skip this new component when copying */
4179 /* Set up copied component */
4180 new_comp->llc_flags &= ~LCME_FL_INIT;
4181 new_comp->llc_stripe = NULL;
4182 new_comp->llc_stripes_allocated = 0;
4183 new_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4184 /* for uninstantiated components, layout gen stores default stripe
4186 new_comp->llc_layout_gen = lod_comp->llc_stripe_offset;
4187 /* This makes the repeated component zero-length, placed at the end of
4188 * the preceding component */
4189 new_comp->llc_extent.e_start = new_comp->llc_extent.e_end;
4190 new_comp->llc_timestamp = lod_comp->llc_timestamp;
4191 new_comp->llc_pool = NULL;
4193 rc = lod_set_pool(&new_comp->llc_pool, lod_comp->llc_pool);
4197 if (new_comp->llc_ostlist.op_array) {
4198 __u32 *op_array = NULL;
4200 OBD_ALLOC(op_array, new_comp->llc_ostlist.op_size);
4202 GOTO(out, rc = -ENOMEM);
4203 memcpy(op_array, &new_comp->llc_ostlist.op_array,
4204 new_comp->llc_ostlist.op_size);
4205 new_comp->llc_ostlist.op_array = op_array;
4208 OBD_FREE(lo->ldo_comp_entries,
4209 sizeof(*comp_array) * lo->ldo_comp_cnt);
4210 lo->ldo_comp_entries = comp_array;
4211 lo->ldo_comp_cnt = new_cnt;
4213 /* Generate an id for the new component */
4214 mirror_id = mirror_id_of(new_comp->llc_id);
4215 new_comp->llc_id = LCME_ID_INVAL;
4216 new_comp->llc_id = lod_gen_component_id(lo, mirror_id, index + 1);
4217 if (new_comp->llc_id == LCME_ID_INVAL)
4218 GOTO(out, rc = -ERANGE);
4223 OBD_FREE(comp_array, sizeof(*comp_array) * new_cnt);
4228 static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
4232 /* clear memory region that will be used for layout change */
4233 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
4234 info->lti_count = 0;
4236 if (info->lti_comp_size >= comp_cnt)
4239 if (info->lti_comp_size > 0) {
4240 OBD_FREE(info->lti_comp_idx,
4241 info->lti_comp_size * sizeof(__u32));
4242 info->lti_comp_size = 0;
4245 OBD_ALLOC(info->lti_comp_idx, comp_cnt * sizeof(__u32));
4246 if (!info->lti_comp_idx)
4249 info->lti_comp_size = comp_cnt;
4254 * Prepare new layout minus deleted components
4256 * Removes components marked for deletion (LCME_ID_INVAL) by copying to a new
4257 * layout and skipping those components. Removes stripe objects if any exist.
4260 * Reallocates layout components array (lo->ldo_comp_entries), invalidating
4261 * any pre-existing pointers to components.
4263 * Caller is responsible for updating mirror end (ldo_mirror[].lme_end).
4265 * \param[in] env execution environment for this thread
4266 * \param[in,out] lo object to update the layout of
4267 * \param[in] th transaction handle for this operation
4269 * \retval # of components deleted
4270 * \retval negative errno on error
4272 static int lod_layout_del_prep_layout(const struct lu_env *env,
4273 struct lod_object *lo,
4276 struct lod_layout_component *lod_comp;
4277 struct lod_thread_info *info = lod_env_info(env);
4278 int rc = 0, i, j, deleted = 0;
4282 LASSERT(lo->ldo_is_composite);
4283 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
4285 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
4289 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4290 lod_comp = &lo->ldo_comp_entries[i];
4292 if (lod_comp->llc_id != LCME_ID_INVAL) {
4293 /* Build array of things to keep */
4294 info->lti_comp_idx[info->lti_count++] = i;
4298 lod_obj_set_pool(lo, i, NULL);
4299 if (lod_comp->llc_ostlist.op_array) {
4300 OBD_FREE(lod_comp->llc_ostlist.op_array,
4301 lod_comp->llc_ostlist.op_size);
4302 lod_comp->llc_ostlist.op_array = NULL;
4303 lod_comp->llc_ostlist.op_size = 0;
4307 CDEBUG(D_LAYOUT, "deleting comp %d, left %d\n", i,
4308 lo->ldo_comp_cnt - deleted);
4310 /* No striping info for this component */
4311 if (lod_comp->llc_stripe == NULL)
4314 LASSERT(lod_comp->llc_stripe_count > 0);
4315 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4316 struct dt_object *obj = lod_comp->llc_stripe[j];
4321 /* components which are not init have no sub objects
4323 if (lod_comp_inited(lod_comp)) {
4324 rc = lod_sub_destroy(env, obj, th);
4329 lu_object_put(env, &obj->do_lu);
4330 lod_comp->llc_stripe[j] = NULL;
4332 OBD_FREE(lod_comp->llc_stripe, sizeof(*lod_comp->llc_stripe) *
4333 lod_comp->llc_stripes_allocated);
4334 lod_comp->llc_stripe = NULL;
4335 OBD_FREE(lod_comp->llc_ost_indices,
4336 sizeof(__u32) * lod_comp->llc_stripes_allocated);
4337 lod_comp->llc_ost_indices = NULL;
4338 lod_comp->llc_stripes_allocated = 0;
4341 /* info->lti_count has the amount of left components */
4342 LASSERTF(info->lti_count >= 0 && info->lti_count < lo->ldo_comp_cnt,
4343 "left = %d, lo->ldo_comp_cnt %d\n", (int)info->lti_count,
4344 (int)lo->ldo_comp_cnt);
4346 if (info->lti_count > 0) {
4347 struct lod_layout_component *comp_array;
4349 OBD_ALLOC(comp_array, sizeof(*comp_array) * info->lti_count);
4350 if (comp_array == NULL)
4351 GOTO(out, rc = -ENOMEM);
4353 for (i = 0; i < info->lti_count; i++) {
4354 memcpy(&comp_array[i],
4355 &lo->ldo_comp_entries[info->lti_comp_idx[i]],
4356 sizeof(*comp_array));
4359 OBD_FREE(lo->ldo_comp_entries,
4360 sizeof(*comp_array) * lo->ldo_comp_cnt);
4361 lo->ldo_comp_entries = comp_array;
4362 lo->ldo_comp_cnt = info->lti_count;
4364 lod_free_comp_entries(lo);
4369 return rc ? rc : deleted;
4373 * Delete layout component(s)
4375 * This function sets up the layout data in the env and does the setattrs
4376 * required to write out the new layout. The layout itself is modified in
4377 * lod_layout_del_prep_layout.
4379 * \param[in] env execution environment for this thread
4380 * \param[in] dt object
4381 * \param[in] th transaction handle
4383 * \retval 0 on success
4384 * \retval negative error number on failure
4386 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
4389 struct lod_object *lo = lod_dt_obj(dt);
4390 struct dt_object *next = dt_object_child(dt);
4391 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4394 LASSERT(lo->ldo_mirror_count == 1);
4396 rc = lod_layout_del_prep_layout(env, lo, th);
4400 /* Only do this if we didn't delete all components */
4401 if (lo->ldo_comp_cnt > 0) {
4402 lo->ldo_mirrors[0].lme_end = lo->ldo_comp_cnt - 1;
4403 lod_obj_inc_layout_gen(lo);
4406 LASSERT(dt_object_exists(dt));
4407 rc = dt_attr_get(env, next, attr);
4411 if (attr->la_size > 0) {
4413 attr->la_valid = LA_SIZE;
4414 rc = lod_sub_attr_set(env, next, attr, th);
4419 rc = lod_generate_and_set_lovea(env, lo, th);
4423 lod_striping_free(env, lo);
4428 static int lod_get_default_lov_striping(const struct lu_env *env,
4429 struct lod_object *lo,
4430 struct lod_default_striping *lds,
4431 struct dt_allocation_hint *ah);
4433 * Implementation of dt_object_operations::do_xattr_set.
4435 * Sets specified extended attribute on the object. Three types of EAs are
4437 * LOV EA - stores striping for a regular file or default striping (when set
4439 * LMV EA - stores a marker for the striped directories
4440 * DMV EA - stores default directory striping
4442 * When striping is applied to a non-striped existing object (this is called
4443 * late striping), then LOD notices the caller wants to turn the object into a
4444 * striped one. The stripe objects are created and appropriate EA is set:
4445 * LOV EA storing all the stripes directly or LMV EA storing just a small header
4446 * with striping configuration.
4448 * \see dt_object_operations::do_xattr_set() in the API description for details.
4450 static int lod_xattr_set(const struct lu_env *env,
4451 struct dt_object *dt, const struct lu_buf *buf,
4452 const char *name, int fl, struct thandle *th)
4454 struct dt_object *next = dt_object_child(dt);
4459 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4460 !strcmp(name, XATTR_NAME_LMV)) {
4462 case LU_XATTR_CREATE:
4463 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
4466 case LU_XATTR_REPLACE:
4467 rc = lod_dir_layout_set(env, dt, buf, fl, th);
4474 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4475 strcmp(name, XATTR_NAME_LOV) == 0) {
4476 struct lod_default_striping *lds = lod_lds_buf_get(env);
4477 struct lov_user_md_v1 *v1 = buf->lb_buf;
4478 char pool[LOV_MAXPOOLNAME + 1];
4481 /* get existing striping config */
4482 rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds,
4487 memset(pool, 0, sizeof(pool));
4488 if (lds->lds_def_striping_set == 1)
4489 lod_layout_get_pool(lds->lds_def_comp_entries,
4490 lds->lds_def_comp_cnt, pool,
4493 is_del = LOVEA_DELETE_VALUES(v1->lmm_stripe_size,
4494 v1->lmm_stripe_count,
4495 v1->lmm_stripe_offset,
4498 /* Retain the pool name if it is not given */
4499 if (v1->lmm_magic == LOV_USER_MAGIC_V1 && pool[0] != '\0' &&
4501 struct lod_thread_info *info = lod_env_info(env);
4502 struct lov_user_md_v3 *v3 = info->lti_ea_store;
4504 memset(v3, 0, sizeof(*v3));
4505 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4506 v3->lmm_pattern = cpu_to_le32(v1->lmm_pattern);
4507 v3->lmm_stripe_count =
4508 cpu_to_le32(v1->lmm_stripe_count);
4509 v3->lmm_stripe_offset =
4510 cpu_to_le32(v1->lmm_stripe_offset);
4511 v3->lmm_stripe_size = cpu_to_le32(v1->lmm_stripe_size);
4513 strlcpy(v3->lmm_pool_name, pool,
4514 sizeof(v3->lmm_pool_name));
4516 info->lti_buf.lb_buf = v3;
4517 info->lti_buf.lb_len = sizeof(*v3);
4518 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4521 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name,
4525 if (lds->lds_def_striping_set == 1 &&
4526 lds->lds_def_comp_entries != NULL)
4527 lod_free_def_comp_entries(lds);
4530 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4531 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
4533 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
4536 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
4537 (strcmp(name, XATTR_NAME_LOV) == 0 ||
4538 strcmp(name, XATTR_LUSTRE_LOV) == 0 ||
4539 allowed_lustre_lov(name))) {
4540 /* in case of lov EA swap, just set it
4541 * if not, it is a replay so check striping match what we
4542 * already have during req replay, declare_xattr_set()
4543 * defines striping, then create() does the work */
4544 if (fl & LU_XATTR_REPLACE) {
4545 /* free stripes, then update disk */
4546 lod_striping_free(env, lod_dt_obj(dt));
4548 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
4549 } else if (dt_object_remote(dt)) {
4550 /* This only happens during migration, see
4551 * mdd_migrate_create(), in which Master MDT will
4552 * create a remote target object, and only set
4553 * (migrating) stripe EA on the remote object,
4554 * and does not need creating each stripes. */
4555 rc = lod_sub_xattr_set(env, next, buf, name,
4557 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
4558 /* delete component(s) */
4559 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
4560 rc = lod_layout_del(env, dt, th);
4563 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
4564 * it's going to create create file with specified
4565 * component(s), the striping must have not being
4566 * cached in this case;
4568 * Otherwise, it's going to add/change component(s) to
4569 * an existing file, the striping must have been cached
4572 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
4573 !strcmp(name, XATTR_NAME_LOV),
4574 !lod_dt_obj(dt)->ldo_comp_cached));
4576 rc = lod_striped_create(env, dt, NULL, NULL, th);
4579 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4580 rc = lod_replace_parent_fid(env, dt, buf, th, false);
4585 /* then all other xattr */
4586 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4592 * Implementation of dt_object_operations::do_declare_xattr_del.
4594 * \see dt_object_operations::do_declare_xattr_del() in the API description
4597 static int lod_declare_xattr_del(const struct lu_env *env,
4598 struct dt_object *dt, const char *name,
4601 struct lod_object *lo = lod_dt_obj(dt);
4602 struct dt_object *next = dt_object_child(dt);
4607 rc = lod_sub_declare_xattr_del(env, next, name, th);
4611 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4614 /* NB: don't delete stripe LMV, because when we do this, normally we
4615 * will remove stripes, besides, if directory LMV is corrupt, this will
4616 * prevent deleting its LMV and fixing it (via LFSCK).
4618 if (!strcmp(name, XATTR_NAME_LMV))
4621 rc = lod_striping_load(env, lo);
4625 if (lo->ldo_dir_stripe_count == 0)
4628 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4629 struct dt_object *dto = lo->ldo_stripe[i];
4634 rc = lod_sub_declare_xattr_del(env, dto, name, th);
4643 * Implementation of dt_object_operations::do_xattr_del.
4645 * If EA storing a regular striping is being deleted, then release
4646 * all the references to the stripe objects in core.
4648 * \see dt_object_operations::do_xattr_del() in the API description for details.
4650 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
4651 const char *name, struct thandle *th)
4653 struct dt_object *next = dt_object_child(dt);
4654 struct lod_object *lo = lod_dt_obj(dt);
4659 if (!strcmp(name, XATTR_NAME_LOV) || !strcmp(name, XATTR_NAME_LMV))
4660 lod_striping_free(env, lod_dt_obj(dt));
4662 rc = lod_sub_xattr_del(env, next, name, th);
4663 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
4666 if (!strcmp(name, XATTR_NAME_LMV))
4669 if (lo->ldo_dir_stripe_count == 0)
4672 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4673 struct dt_object *dto = lo->ldo_stripe[i];
4678 rc = lod_sub_xattr_del(env, dto, name, th);
4687 * Implementation of dt_object_operations::do_xattr_list.
4689 * \see dt_object_operations::do_xattr_list() in the API description
4692 static int lod_xattr_list(const struct lu_env *env,
4693 struct dt_object *dt, const struct lu_buf *buf)
4695 return dt_xattr_list(env, dt_object_child(dt), buf);
4698 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
4700 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
4704 * Copy OST list from layout provided by user.
4706 * \param[in] lod_comp layout_component to be filled
4707 * \param[in] v3 LOV EA V3 user data
4709 * \retval 0 on success
4710 * \retval negative if failed
4712 int lod_comp_copy_ost_lists(struct lod_layout_component *lod_comp,
4713 struct lov_user_md_v3 *v3)
4719 if (v3->lmm_stripe_offset == LOV_OFFSET_DEFAULT)
4720 v3->lmm_stripe_offset = v3->lmm_objects[0].l_ost_idx;
4722 if (lod_comp->llc_ostlist.op_array) {
4723 if (lod_comp->llc_ostlist.op_size >=
4724 v3->lmm_stripe_count * sizeof(__u32)) {
4725 lod_comp->llc_ostlist.op_count =
4726 v3->lmm_stripe_count;
4729 OBD_FREE(lod_comp->llc_ostlist.op_array,
4730 lod_comp->llc_ostlist.op_size);
4733 /* copy ost list from lmm */
4734 lod_comp->llc_ostlist.op_count = v3->lmm_stripe_count;
4735 lod_comp->llc_ostlist.op_size = v3->lmm_stripe_count * sizeof(__u32);
4736 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
4737 lod_comp->llc_ostlist.op_size);
4738 if (!lod_comp->llc_ostlist.op_array)
4741 for (j = 0; j < v3->lmm_stripe_count; j++) {
4742 lod_comp->llc_ostlist.op_array[j] =
4743 v3->lmm_objects[j].l_ost_idx;
4751 * Get default striping.
4753 * \param[in] env execution environment
4754 * \param[in] lo object
4755 * \param[out] lds default striping
4757 * \retval 0 on success
4758 * \retval negative if failed
4760 static int lod_get_default_lov_striping(const struct lu_env *env,
4761 struct lod_object *lo,
4762 struct lod_default_striping *lds,
4763 struct dt_allocation_hint *ah)
4765 struct lod_thread_info *info = lod_env_info(env);
4766 struct lov_user_md_v1 *v1 = NULL;
4767 struct lov_user_md_v3 *v3 = NULL;
4768 struct lov_comp_md_v1 *comp_v1 = NULL;
4776 rc = lod_get_lov_ea(env, lo);
4780 if (rc < (typeof(rc))sizeof(struct lov_user_md))
4783 v1 = info->lti_ea_store;
4784 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
4785 lustre_swab_lov_user_md_v1(v1);
4786 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
4787 v3 = (struct lov_user_md_v3 *)v1;
4788 lustre_swab_lov_user_md_v3(v3);
4789 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_SPECIFIC)) {
4790 v3 = (struct lov_user_md_v3 *)v1;
4791 lustre_swab_lov_user_md_v3(v3);
4792 lustre_swab_lov_user_md_objects(v3->lmm_objects,
4793 v3->lmm_stripe_count);
4794 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_COMP_V1) ||
4795 v1->lmm_magic == __swab32(LOV_USER_MAGIC_SEL)) {
4796 comp_v1 = (struct lov_comp_md_v1 *)v1;
4797 lustre_swab_lov_comp_md_v1(comp_v1);
4800 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1 &&
4801 v1->lmm_magic != LOV_MAGIC_COMP_V1 &&
4802 v1->lmm_magic != LOV_MAGIC_SEL &&
4803 v1->lmm_magic != LOV_USER_MAGIC_SPECIFIC)
4806 if ((v1->lmm_magic == LOV_MAGIC_COMP_V1 ||
4807 v1->lmm_magic == LOV_MAGIC_SEL) &&
4808 !(ah && ah->dah_append_stripes)) {
4809 comp_v1 = (struct lov_comp_md_v1 *)v1;
4810 comp_cnt = comp_v1->lcm_entry_count;
4813 mirror_cnt = comp_v1->lcm_mirror_count + 1;
4821 /* realloc default comp entries if necessary */
4822 rc = lod_def_striping_comp_resize(lds, comp_cnt);
4826 lds->lds_def_comp_cnt = comp_cnt;
4827 lds->lds_def_striping_is_composite = composite;
4828 lds->lds_def_mirror_cnt = mirror_cnt;
4830 for (i = 0; i < comp_cnt; i++) {
4831 struct lod_layout_component *lod_comp;
4834 lod_comp = &lds->lds_def_comp_entries[i];
4836 * reset lod_comp values, llc_stripes is always NULL in
4837 * the default striping template, llc_pool will be reset
4840 memset(lod_comp, 0, offsetof(typeof(*lod_comp), llc_pool));
4843 v1 = (struct lov_user_md *)((char *)comp_v1 +
4844 comp_v1->lcm_entries[i].lcme_offset);
4845 lod_comp->llc_extent =
4846 comp_v1->lcm_entries[i].lcme_extent;
4847 /* We only inherit certain flags from the layout */
4848 lod_comp->llc_flags =
4849 comp_v1->lcm_entries[i].lcme_flags &
4850 LCME_TEMPLATE_FLAGS;
4853 if (!lov_pattern_supported(v1->lmm_pattern) &&
4854 !(v1->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
4855 lod_free_def_comp_entries(lds);
4859 CDEBUG(D_LAYOUT, DFID" stripe_count=%d stripe_size=%d stripe_offset=%d append_stripes=%d\n",
4860 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
4861 (int)v1->lmm_stripe_count, (int)v1->lmm_stripe_size,
4862 (int)v1->lmm_stripe_offset,
4863 ah ? ah->dah_append_stripes : 0);
4865 if (ah && ah->dah_append_stripes)
4866 lod_comp->llc_stripe_count = ah->dah_append_stripes;
4868 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
4869 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
4870 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
4871 lod_comp->llc_pattern = v1->lmm_pattern;
4874 if (ah && ah->dah_append_pool && ah->dah_append_pool[0]) {
4875 pool = ah->dah_append_pool;
4876 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
4877 /* XXX: sanity check here */
4878 v3 = (struct lov_user_md_v3 *) v1;
4879 if (v3->lmm_pool_name[0] != '\0')
4880 pool = v3->lmm_pool_name;
4882 lod_set_def_pool(lds, i, pool);
4883 if (v1->lmm_magic == LOV_USER_MAGIC_SPECIFIC) {
4884 v3 = (struct lov_user_md_v3 *)v1;
4885 rc = lod_comp_copy_ost_lists(lod_comp, v3);
4888 } else if (lod_comp->llc_ostlist.op_array &&
4889 lod_comp->llc_ostlist.op_count) {
4890 for (j = 0; j < lod_comp->llc_ostlist.op_count; j++)
4891 lod_comp->llc_ostlist.op_array[j] = -1;
4892 lod_comp->llc_ostlist.op_count = 0;
4896 lds->lds_def_striping_set = 1;
4901 * Get default directory striping.
4903 * \param[in] env execution environment
4904 * \param[in] lo object
4905 * \param[out] lds default striping
4907 * \retval 0 on success
4908 * \retval negative if failed
4910 static int lod_get_default_lmv_striping(const struct lu_env *env,
4911 struct lod_object *lo,
4912 struct lod_default_striping *lds)
4914 struct lmv_user_md *lmu;
4917 lds->lds_dir_def_striping_set = 0;
4919 rc = lod_get_default_lmv_ea(env, lo);
4923 if (rc >= (int)sizeof(*lmu)) {
4924 struct lod_thread_info *info = lod_env_info(env);
4926 lmu = info->lti_ea_store;
4928 lds->lds_dir_def_stripe_count =
4929 le32_to_cpu(lmu->lum_stripe_count);
4930 lds->lds_dir_def_stripe_offset =
4931 le32_to_cpu(lmu->lum_stripe_offset);
4932 lds->lds_dir_def_hash_type =
4933 le32_to_cpu(lmu->lum_hash_type);
4934 lds->lds_dir_def_striping_set = 1;
4941 * Get default striping in the object.
4943 * Get object default striping and default directory striping.
4945 * \param[in] env execution environment
4946 * \param[in] lo object
4947 * \param[out] lds default striping
4949 * \retval 0 on success
4950 * \retval negative if failed
4952 static int lod_get_default_striping(const struct lu_env *env,
4953 struct lod_object *lo,
4954 struct lod_default_striping *lds)
4958 rc = lod_get_default_lov_striping(env, lo, lds, NULL);
4959 rc1 = lod_get_default_lmv_striping(env, lo, lds);
4960 if (rc == 0 && rc1 < 0)
4967 * Apply default striping on object.
4969 * If object striping pattern is not set, set to the one in default striping.
4970 * The default striping is from parent or fs.
4972 * \param[in] lo new object
4973 * \param[in] lds default striping
4974 * \param[in] mode new object's mode
4976 static void lod_striping_from_default(struct lod_object *lo,
4977 const struct lod_default_striping *lds,
4980 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
4983 if (lds->lds_def_striping_set && S_ISREG(mode)) {
4984 struct lov_desc *desc = &d->lod_ost_descs.ltd_lov_desc;
4986 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
4987 lds->lds_def_comp_cnt);
4991 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
4992 if (lds->lds_def_mirror_cnt > 1)
4993 lo->ldo_flr_state = LCM_FL_RDONLY;
4995 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4996 struct lod_layout_component *obj_comp =
4997 &lo->ldo_comp_entries[i];
4998 struct lod_layout_component *def_comp =
4999 &lds->lds_def_comp_entries[i];
5001 CDEBUG(D_LAYOUT, "Inherit from default: flags=%#x "
5002 "size=%hu nr=%u offset=%u pattern=%#x pool=%s\n",
5003 def_comp->llc_flags,
5004 def_comp->llc_stripe_size,
5005 def_comp->llc_stripe_count,
5006 def_comp->llc_stripe_offset,
5007 def_comp->llc_pattern,
5008 def_comp->llc_pool ?: "");
5010 *obj_comp = *def_comp;
5011 if (def_comp->llc_pool != NULL) {
5012 /* pointer was copied from def_comp */
5013 obj_comp->llc_pool = NULL;
5014 lod_obj_set_pool(lo, i, def_comp->llc_pool);
5018 if (def_comp->llc_ostlist.op_array &&
5019 def_comp->llc_ostlist.op_count) {
5020 OBD_ALLOC(obj_comp->llc_ostlist.op_array,
5021 obj_comp->llc_ostlist.op_size);
5022 if (!obj_comp->llc_ostlist.op_array)
5024 memcpy(obj_comp->llc_ostlist.op_array,
5025 def_comp->llc_ostlist.op_array,
5026 obj_comp->llc_ostlist.op_size);
5027 } else if (def_comp->llc_ostlist.op_array) {
5028 obj_comp->llc_ostlist.op_array = NULL;
5032 * Don't initialize these fields for plain layout
5033 * (v1/v3) here, they are inherited in the order of
5034 * 'parent' -> 'fs default (root)' -> 'global default
5035 * values for stripe_count & stripe_size'.
5037 * see lod_ah_init().
5039 if (!lo->ldo_is_composite)
5042 lod_adjust_stripe_info(obj_comp, desc, 0);
5044 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
5045 if (lo->ldo_dir_stripe_count == 0)
5046 lo->ldo_dir_stripe_count =
5047 lds->lds_dir_def_stripe_count;
5048 if (lo->ldo_dir_stripe_offset == -1)
5049 lo->ldo_dir_stripe_offset =
5050 lds->lds_dir_def_stripe_offset;
5051 if (lo->ldo_dir_hash_type == 0)
5052 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type;
5054 CDEBUG(D_LAYOUT, "striping from default dir: count:%hu, "
5055 "offset:%u, hash_type:%u\n",
5056 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
5057 lo->ldo_dir_hash_type);
5061 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root,
5064 struct lod_layout_component *lod_comp;
5066 if (lo->ldo_comp_cnt == 0)
5069 if (lo->ldo_is_composite)
5072 lod_comp = &lo->ldo_comp_entries[0];
5074 if (lod_comp->llc_stripe_count <= 0 ||
5075 lod_comp->llc_stripe_size <= 0)
5078 if (from_root && (lod_comp->llc_pool == NULL ||
5079 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
5082 if (append_pool && append_pool[0])
5089 * Implementation of dt_object_operations::do_ah_init.
5091 * This method is used to make a decision on the striping configuration for the
5092 * object being created. It can be taken from the \a parent object if it exists,
5093 * or filesystem's default. The resulting configuration (number of stripes,
5094 * stripe size/offset, pool name, etc) is stored in the object itself and will
5095 * be used by the methods like ->doo_declare_create().
5097 * \see dt_object_operations::do_ah_init() in the API description for details.
5099 static void lod_ah_init(const struct lu_env *env,
5100 struct dt_allocation_hint *ah,
5101 struct dt_object *parent,
5102 struct dt_object *child,
5105 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
5106 struct lod_thread_info *info = lod_env_info(env);
5107 struct lod_default_striping *lds = lod_lds_buf_get(env);
5108 struct dt_object *nextp = NULL;
5109 struct dt_object *nextc;
5110 struct lod_object *lp = NULL;
5111 struct lod_object *lc;
5112 struct lov_desc *desc;
5113 struct lod_layout_component *lod_comp;
5119 if (ah->dah_append_stripes == -1)
5120 ah->dah_append_stripes =
5121 d->lod_ost_descs.ltd_lov_desc.ld_tgt_count;
5123 if (likely(parent)) {
5124 nextp = dt_object_child(parent);
5125 lp = lod_dt_obj(parent);
5128 nextc = dt_object_child(child);
5129 lc = lod_dt_obj(child);
5131 LASSERT(!lod_obj_is_striped(child));
5132 /* default layout template may have been set on the regular file
5133 * when this is called from mdd_create_data() */
5134 if (S_ISREG(child_mode))
5135 lod_free_comp_entries(lc);
5137 if (!dt_object_exists(nextc))
5138 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
5140 if (S_ISDIR(child_mode)) {
5141 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
5143 /* other default values are 0 */
5144 lc->ldo_dir_stripe_offset = -1;
5146 /* no default striping configuration is needed for
5149 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5150 le32_to_cpu(lum1->lum_magic) == LMV_MAGIC_FOREIGN) {
5151 lc->ldo_dir_is_foreign = true;
5152 /* keep stripe_count 0 and stripe_offset -1 */
5153 CDEBUG(D_INFO, "no default striping for foreign dir\n");
5158 * If parent object is not root directory,
5159 * then get default striping from parent object.
5161 if (likely(lp != NULL)) {
5162 lod_get_default_striping(env, lp, lds);
5164 /* inherit default striping except ROOT */
5165 if ((lds->lds_def_striping_set ||
5166 lds->lds_dir_def_striping_set) &&
5167 !fid_is_root(lod_object_fid(lp)))
5168 lc->ldo_def_striping = lds;
5171 /* It should always honour the specified stripes */
5172 /* Note: old client (< 2.7)might also do lfs mkdir, whose EA
5173 * will have old magic. In this case, we should ignore the
5174 * stripe count and try to create dir by default stripe.
5176 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5177 (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
5178 le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC)) {
5179 lc->ldo_dir_stripe_count =
5180 le32_to_cpu(lum1->lum_stripe_count);
5181 lc->ldo_dir_stripe_offset =
5182 le32_to_cpu(lum1->lum_stripe_offset);
5183 lc->ldo_dir_hash_type =
5184 le32_to_cpu(lum1->lum_hash_type);
5186 "set dirstripe: count %hu, offset %d, hash %u\n",
5187 lc->ldo_dir_stripe_count,
5188 (int)lc->ldo_dir_stripe_offset,
5189 lc->ldo_dir_hash_type);
5191 /* transfer defaults LMV to new directory */
5192 lod_striping_from_default(lc, lds, child_mode);
5194 /* set count 0 to create normal directory */
5195 if (lc->ldo_dir_stripe_count == 1)
5196 lc->ldo_dir_stripe_count = 0;
5199 /* shrink the stripe_count to the avaible MDT count */
5200 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
5201 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)) {
5202 lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
5203 if (lc->ldo_dir_stripe_count == 1)
5204 lc->ldo_dir_stripe_count = 0;
5207 if (!(lc->ldo_dir_hash_type & LMV_HASH_TYPE_MASK))
5208 lc->ldo_dir_hash_type |=
5209 d->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
5211 CDEBUG(D_INFO, "final dir stripe [%hu %d %u]\n",
5212 lc->ldo_dir_stripe_count,
5213 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
5218 /* child object regular file*/
5220 if (!lod_object_will_be_striped(S_ISREG(child_mode),
5221 lu_object_fid(&child->do_lu)))
5224 /* If object is going to be striped over OSTs, transfer default
5225 * striping information to the child, so that we can use it
5226 * during declaration and creation.
5228 * Try from the parent first.
5230 if (likely(lp != NULL)) {
5231 rc = lod_get_default_lov_striping(env, lp, lds, ah);
5233 lod_striping_from_default(lc, lds, child_mode);
5236 /* Initialize lod_device::lod_md_root object reference */
5237 if (d->lod_md_root == NULL) {
5238 struct dt_object *root;
5239 struct lod_object *lroot;
5241 lu_root_fid(&info->lti_fid);
5242 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
5243 if (!IS_ERR(root)) {
5244 lroot = lod_dt_obj(root);
5246 spin_lock(&d->lod_lock);
5247 if (d->lod_md_root != NULL)
5248 dt_object_put(env, &d->lod_md_root->ldo_obj);
5249 d->lod_md_root = lroot;
5250 spin_unlock(&d->lod_lock);
5254 /* try inherit layout from the root object (fs default) when:
5255 * - parent does not have default layout; or
5256 * - parent has plain(v1/v3) default layout, and some attributes
5257 * are not specified in the default layout;
5259 if (d->lod_md_root != NULL &&
5260 lod_need_inherit_more(lc, true, ah->dah_append_pool)) {
5261 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds,
5265 if (lc->ldo_comp_cnt == 0) {
5266 lod_striping_from_default(lc, lds, child_mode);
5267 } else if (!lds->lds_def_striping_is_composite) {
5268 struct lod_layout_component *def_comp;
5270 LASSERT(!lc->ldo_is_composite);
5271 lod_comp = &lc->ldo_comp_entries[0];
5272 def_comp = &lds->lds_def_comp_entries[0];
5274 if (lod_comp->llc_stripe_count <= 0)
5275 lod_comp->llc_stripe_count =
5276 def_comp->llc_stripe_count;
5277 if (lod_comp->llc_stripe_size <= 0)
5278 lod_comp->llc_stripe_size =
5279 def_comp->llc_stripe_size;
5280 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT &&
5281 (!lod_comp->llc_pool || !lod_comp->llc_pool[0]))
5282 lod_comp->llc_stripe_offset =
5283 def_comp->llc_stripe_offset;
5284 if (lod_comp->llc_pool == NULL)
5285 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
5290 * fs default striping may not be explicitly set, or historically set
5291 * in config log, use them.
5293 if (lod_need_inherit_more(lc, false, ah->dah_append_pool)) {
5294 if (lc->ldo_comp_cnt == 0) {
5295 rc = lod_alloc_comp_entries(lc, 0, 1);
5297 /* fail to allocate memory, will create a
5298 * non-striped file. */
5300 lc->ldo_is_composite = 0;
5301 lod_comp = &lc->ldo_comp_entries[0];
5302 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
5304 LASSERT(!lc->ldo_is_composite);
5305 lod_comp = &lc->ldo_comp_entries[0];
5306 desc = &d->lod_ost_descs.ltd_lov_desc;
5307 lod_adjust_stripe_info(lod_comp, desc, ah->dah_append_stripes);
5308 if (ah->dah_append_pool && ah->dah_append_pool[0])
5309 lod_obj_set_pool(lc, 0, ah->dah_append_pool);
5316 * Size initialization on late striping.
5318 * Propagate the size of a truncated object to a deferred striping.
5319 * This function handles a special case when truncate was done on a
5320 * non-striped object and now while the striping is being created
5321 * we can't lose that size, so we have to propagate it to the stripes
5324 * \param[in] env execution environment
5325 * \param[in] dt object
5326 * \param[in] th transaction handle
5328 * \retval 0 on success
5329 * \retval negative if failed
5331 static int lod_declare_init_size(const struct lu_env *env,
5332 struct dt_object *dt, struct thandle *th)
5334 struct dt_object *next = dt_object_child(dt);
5335 struct lod_object *lo = lod_dt_obj(dt);
5336 struct dt_object **objects = NULL;
5337 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
5338 uint64_t size, offs;
5339 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
5340 struct lu_extent size_ext;
5343 if (!lod_obj_is_striped(dt))
5346 rc = dt_attr_get(env, next, attr);
5347 LASSERT(attr->la_valid & LA_SIZE);
5351 size = attr->la_size;
5355 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
5356 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5357 struct lod_layout_component *lod_comp;
5358 struct lu_extent *extent;
5360 lod_comp = &lo->ldo_comp_entries[i];
5362 if (lod_comp->llc_stripe == NULL)
5365 extent = &lod_comp->llc_extent;
5366 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
5367 if (!lo->ldo_is_composite ||
5368 lu_extent_is_overlapped(extent, &size_ext)) {
5369 objects = lod_comp->llc_stripe;
5370 stripe_count = lod_comp->llc_stripe_count;
5371 stripe_size = lod_comp->llc_stripe_size;
5374 if (stripe_count == 0)
5377 LASSERT(objects != NULL && stripe_size != 0);
5378 do_div(size, stripe_size);
5379 stripe = do_div(size, stripe_count);
5380 LASSERT(objects[stripe] != NULL);
5382 size = size * stripe_size;
5383 offs = attr->la_size;
5384 size += do_div(offs, stripe_size);
5386 attr->la_valid = LA_SIZE;
5387 attr->la_size = size;
5389 rc = lod_sub_declare_attr_set(env, objects[stripe],
5398 * Declare creation of striped object.
5400 * The function declares creation stripes for a regular object. The function
5401 * also declares whether the stripes will be created with non-zero size if
5402 * previously size was set non-zero on the master object. If object \a dt is
5403 * not local, then only fully defined striping can be applied in \a lovea.
5404 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
5407 * \param[in] env execution environment
5408 * \param[in] dt object
5409 * \param[in] attr attributes the stripes will be created with
5410 * \param[in] lovea a buffer containing striping description
5411 * \param[in] th transaction handle
5413 * \retval 0 on success
5414 * \retval negative if failed
5416 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
5417 struct lu_attr *attr,
5418 const struct lu_buf *lovea, struct thandle *th)
5420 struct lod_thread_info *info = lod_env_info(env);
5421 struct dt_object *next = dt_object_child(dt);
5422 struct lod_object *lo = lod_dt_obj(dt);
5426 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
5427 GOTO(out, rc = -ENOMEM);
5429 if (!dt_object_remote(next)) {
5430 /* choose OST and generate appropriate objects */
5431 rc = lod_prepare_create(env, lo, attr, lovea, th);
5436 * declare storage for striping data
5438 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
5440 /* LOD can not choose OST objects for remote objects, i.e.
5441 * stripes must be ready before that. Right now, it can only
5442 * happen during migrate, i.e. migrate process needs to create
5443 * remote regular file (mdd_migrate_create), then the migrate
5444 * process will provide stripeEA. */
5445 LASSERT(lovea != NULL);
5446 info->lti_buf = *lovea;
5449 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
5450 XATTR_NAME_LOV, 0, th);
5455 * if striping is created with local object's size > 0,
5456 * we have to propagate this size to specific object
5457 * the case is possible only when local object was created previously
5459 if (dt_object_exists(next))
5460 rc = lod_declare_init_size(env, dt, th);
5463 /* failed to create striping or to set initial size, let's reset
5464 * config so that others don't get confused */
5466 lod_striping_free(env, lo);
5472 * Whether subdirectories under \a dt should be created on MDTs by space QoS
5474 * If LMV_HASH_FLAG_SPACE is set on directory default layout, its subdirectories
5475 * should be created on MDT by space QoS.
5477 * \param[in] env execution environment
5478 * \param[in] dev lu device
5479 * \param[in] dt object
5481 * \retval 1 if directory should create subdir by space usage
5483 * \retval -ev if failed
5485 static inline int dt_object_qos_mkdir(const struct lu_env *env,
5486 struct lu_device *dev,
5487 struct dt_object *dt)
5489 struct lod_thread_info *info = lod_env_info(env);
5490 struct lu_object *obj;
5491 struct lod_object *lo;
5492 struct lmv_user_md *lmu;
5495 obj = lu_object_find_slice(env, dev, lu_object_fid(&dt->do_lu), NULL);
5497 return PTR_ERR(obj);
5499 lo = lu2lod_obj(obj);
5501 rc = lod_get_default_lmv_ea(env, lo);
5502 dt_object_put(env, dt);
5506 if (rc < (int)sizeof(*lmu))
5509 lmu = info->lti_ea_store;
5510 return le32_to_cpu(lmu->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
5514 * Implementation of dt_object_operations::do_declare_create.
5516 * The method declares creation of a new object. If the object will be striped,
5517 * then helper functions are called to find FIDs for the stripes, declare
5518 * creation of the stripes and declare initialization of the striping
5519 * information to be stored in the master object.
5521 * \see dt_object_operations::do_declare_create() in the API description
5524 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
5525 struct lu_attr *attr,
5526 struct dt_allocation_hint *hint,
5527 struct dt_object_format *dof, struct thandle *th)
5529 struct dt_object *next = dt_object_child(dt);
5530 struct lod_object *lo = lod_dt_obj(dt);
5539 * first of all, we declare creation of local object
5541 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
5546 * it's lod_ah_init() that has decided the object will be striped
5548 if (dof->dof_type == DFT_REGULAR) {
5549 /* callers don't want stripes */
5550 /* XXX: all tricky interactions with ->ah_make_hint() decided
5551 * to use striping, then ->declare_create() behaving differently
5552 * should be cleaned */
5553 if (dof->u.dof_reg.striped != 0)
5554 rc = lod_declare_striped_create(env, dt, attr,
5556 } else if (dof->dof_type == DFT_DIR) {
5557 struct seq_server_site *ss;
5558 struct lu_buf buf = { NULL };
5559 struct lu_buf *lmu = NULL;
5561 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
5563 /* If the parent has default stripeEA, and client
5564 * did not find it before sending create request,
5565 * then MDT will return -EREMOTE, and client will
5566 * retrieve the default stripeEA and re-create the
5569 * Note: if dah_eadata != NULL, it means creating the
5570 * striped directory with specified stripeEA, then it
5571 * should ignore the default stripeEA */
5572 if (hint != NULL && hint->dah_eadata == NULL) {
5573 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
5574 GOTO(out, rc = -EREMOTE);
5576 if (lo->ldo_dir_stripe_offset == LMV_OFFSET_DEFAULT) {
5577 struct lod_default_striping *lds;
5579 lds = lo->ldo_def_striping;
5581 * child and parent should be on the same MDT,
5582 * but if parent has default LMV, and the start
5583 * MDT offset is -1, it's allowed. This check
5584 * is not necessary after 2.12.22 because client
5585 * follows this already, but old client may not.
5587 if (hint->dah_parent &&
5588 dt_object_remote(hint->dah_parent) && lds &&
5589 lds->lds_dir_def_stripe_offset !=
5591 GOTO(out, rc = -EREMOTE);
5592 } else if (lo->ldo_dir_stripe_offset !=
5594 struct lod_device *lod;
5595 struct lu_tgt_desc *mdt = NULL;
5596 bool found_mdt = false;
5598 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5599 lod_foreach_mdt(lod, mdt) {
5600 if (mdt->ltd_index ==
5601 lo->ldo_dir_stripe_offset) {
5607 /* If the MDT indicated by stripe_offset can be
5608 * found, then tell client to resend the create
5609 * request to the correct MDT, otherwise return
5610 * error to client */
5612 GOTO(out, rc = -EREMOTE);
5614 GOTO(out, rc = -EINVAL);
5616 } else if (hint && hint->dah_eadata) {
5618 lmu->lb_buf = (void *)hint->dah_eadata;
5619 lmu->lb_len = hint->dah_eadata_len;
5622 rc = lod_declare_dir_striping_create(env, dt, attr, lmu, dof,
5626 /* failed to create striping or to set initial size, let's reset
5627 * config so that others don't get confused */
5629 lod_striping_free(env, lo);
5634 * Generate component ID for new created component.
5636 * \param[in] lo LOD object
5637 * \param[in] comp_idx index of ldo_comp_entries
5639 * \retval component ID on success
5640 * \retval LCME_ID_INVAL on failure
5642 static __u32 lod_gen_component_id(struct lod_object *lo,
5643 int mirror_id, int comp_idx)
5645 struct lod_layout_component *lod_comp;
5646 __u32 id, start, end;
5649 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
5651 lod_obj_inc_layout_gen(lo);
5652 id = lo->ldo_layout_gen;
5653 if (likely(id <= SEQ_ID_MAX))
5654 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
5656 /* Layout generation wraps, need to check collisions. */
5657 start = id & SEQ_ID_MASK;
5660 for (id = start; id <= end; id++) {
5661 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5662 lod_comp = &lo->ldo_comp_entries[i];
5663 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
5666 /* Found the ununsed ID */
5667 if (i == lo->ldo_comp_cnt)
5668 RETURN(pflr_id(mirror_id, id));
5670 if (end == LCME_ID_MAX) {
5672 end = min(lo->ldo_layout_gen & LCME_ID_MASK,
5673 (__u32)(LCME_ID_MAX - 1));
5677 RETURN(LCME_ID_INVAL);
5681 * Creation of a striped regular object.
5683 * The function is called to create the stripe objects for a regular
5684 * striped file. This can happen at the initial object creation or
5685 * when the caller asks LOD to do so using ->do_xattr_set() method
5686 * (so called late striping). Notice all the information are already
5687 * prepared in the form of the list of objects (ldo_stripe field).
5688 * This is done during declare phase.
5690 * \param[in] env execution environment
5691 * \param[in] dt object
5692 * \param[in] attr attributes the stripes will be created with
5693 * \param[in] dof format of stripes (see OSD API description)
5694 * \param[in] th transaction handle
5696 * \retval 0 on success
5697 * \retval negative if failed
5699 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
5700 struct lu_attr *attr, struct dt_object_format *dof,
5703 struct lod_layout_component *lod_comp;
5704 struct lod_object *lo = lod_dt_obj(dt);
5709 LASSERT((lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL) ||
5710 lo->ldo_is_foreign);
5712 mirror_id = 0; /* non-flr file's mirror_id is 0 */
5713 if (lo->ldo_mirror_count > 1) {
5714 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5715 lod_comp = &lo->ldo_comp_entries[i];
5716 if (lod_comp->llc_id != LCME_ID_INVAL &&
5717 mirror_id_of(lod_comp->llc_id) > mirror_id)
5718 mirror_id = mirror_id_of(lod_comp->llc_id);
5722 /* create all underlying objects */
5723 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5724 lod_comp = &lo->ldo_comp_entries[i];
5726 if (lod_comp->llc_id == LCME_ID_INVAL) {
5727 /* only the component of FLR layout with more than 1
5728 * mirror has mirror ID in its component ID.
5730 if (lod_comp->llc_extent.e_start == 0 &&
5731 lo->ldo_mirror_count > 1)
5734 lod_comp->llc_id = lod_gen_component_id(lo,
5736 if (lod_comp->llc_id == LCME_ID_INVAL)
5737 GOTO(out, rc = -ERANGE);
5740 if (lod_comp_inited(lod_comp))
5743 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
5744 lod_comp_set_init(lod_comp);
5746 if (lov_pattern(lod_comp->llc_pattern) == LOV_PATTERN_MDT)
5747 lod_comp_set_init(lod_comp);
5749 if (lod_comp->llc_stripe == NULL)
5752 LASSERT(lod_comp->llc_stripe_count);
5753 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
5754 struct dt_object *object = lod_comp->llc_stripe[j];
5755 LASSERT(object != NULL);
5756 rc = lod_sub_create(env, object, attr, NULL, dof, th);
5760 lod_comp_set_init(lod_comp);
5763 rc = lod_fill_mirrors(lo);
5767 rc = lod_generate_and_set_lovea(env, lo, th);
5771 lo->ldo_comp_cached = 1;
5775 lod_striping_free(env, lo);
5779 static inline bool lod_obj_is_dom(struct dt_object *dt)
5781 struct lod_object *lo = lod_dt_obj(dt);
5783 if (!dt_object_exists(dt_object_child(dt)))
5786 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
5789 if (!lo->ldo_comp_cnt)
5792 return (lov_pattern(lo->ldo_comp_entries[0].llc_pattern) ==
5797 * Implementation of dt_object_operations::do_create.
5799 * If any of preceeding methods (like ->do_declare_create(),
5800 * ->do_ah_init(), etc) chose to create a striped object,
5801 * then this method will create the master and the stripes.
5803 * \see dt_object_operations::do_create() in the API description for details.
5805 static int lod_create(const struct lu_env *env, struct dt_object *dt,
5806 struct lu_attr *attr, struct dt_allocation_hint *hint,
5807 struct dt_object_format *dof, struct thandle *th)
5812 /* create local object */
5813 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
5817 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
5818 (lod_obj_is_striped(dt) || lod_obj_is_dom(dt)) &&
5819 dof->u.dof_reg.striped != 0) {
5820 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
5821 rc = lod_striped_create(env, dt, attr, dof, th);
5828 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
5829 struct dt_object *dt, struct thandle *th,
5830 int comp_idx, int stripe_idx,
5831 struct lod_obj_stripe_cb_data *data)
5833 if (data->locd_declare)
5834 return lod_sub_declare_destroy(env, dt, th);
5835 else if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
5836 stripe_idx == cfs_fail_val)
5837 return lod_sub_destroy(env, dt, th);
5843 * Implementation of dt_object_operations::do_declare_destroy.
5845 * If the object is a striped directory, then the function declares reference
5846 * removal from the master object (this is an index) to the stripes and declares
5847 * destroy of all the stripes. In all the cases, it declares an intention to
5848 * destroy the object itself.
5850 * \see dt_object_operations::do_declare_destroy() in the API description
5853 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
5856 struct dt_object *next = dt_object_child(dt);
5857 struct lod_object *lo = lod_dt_obj(dt);
5858 struct lod_thread_info *info = lod_env_info(env);
5859 struct dt_object *stripe;
5860 char *stripe_name = info->lti_key;
5866 * load striping information, notice we don't do this when object
5867 * is being initialized as we don't need this information till
5868 * few specific cases like destroy, chown
5870 rc = lod_striping_load(env, lo);
5874 /* declare destroy for all underlying objects */
5875 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5876 rc = next->do_ops->do_index_try(env, next,
5877 &dt_directory_features);
5881 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5882 stripe = lo->ldo_stripe[i];
5886 rc = lod_sub_declare_ref_del(env, next, th);
5890 snprintf(stripe_name, sizeof(info->lti_key),
5892 PFID(lu_object_fid(&stripe->do_lu)), i);
5893 rc = lod_sub_declare_delete(env, next,
5894 (const struct dt_key *)stripe_name, th);
5901 * we declare destroy for the local object
5903 rc = lod_sub_declare_destroy(env, next, th);
5907 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
5908 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
5911 if (!lod_obj_is_striped(dt))
5914 /* declare destroy all striped objects */
5915 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5916 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5917 stripe = lo->ldo_stripe[i];
5921 if (!dt_object_exists(stripe))
5924 rc = lod_sub_declare_ref_del(env, stripe, th);
5928 rc = lod_sub_declare_destroy(env, stripe, th);
5933 struct lod_obj_stripe_cb_data data = { { 0 } };
5935 data.locd_declare = true;
5936 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
5937 rc = lod_obj_for_each_stripe(env, lo, th, &data);
5944 * Implementation of dt_object_operations::do_destroy.
5946 * If the object is a striped directory, then the function removes references
5947 * from the master object (this is an index) to the stripes and destroys all
5948 * the stripes. In all the cases, the function destroys the object itself.
5950 * \see dt_object_operations::do_destroy() in the API description for details.
5952 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
5955 struct dt_object *next = dt_object_child(dt);
5956 struct lod_object *lo = lod_dt_obj(dt);
5957 struct lod_thread_info *info = lod_env_info(env);
5958 char *stripe_name = info->lti_key;
5959 struct dt_object *stripe;
5965 /* destroy sub-stripe of master object */
5966 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5967 rc = next->do_ops->do_index_try(env, next,
5968 &dt_directory_features);
5972 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5973 stripe = lo->ldo_stripe[i];
5977 rc = lod_sub_ref_del(env, next, th);
5981 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
5982 PFID(lu_object_fid(&stripe->do_lu)), i);
5984 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
5985 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
5986 PFID(lu_object_fid(&stripe->do_lu)));
5988 rc = lod_sub_delete(env, next,
5989 (const struct dt_key *)stripe_name, th);
5995 rc = lod_sub_destroy(env, next, th);
5999 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
6000 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
6003 if (!lod_obj_is_striped(dt))
6006 /* destroy all striped objects */
6007 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6008 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6009 stripe = lo->ldo_stripe[i];
6013 if (!dt_object_exists(stripe))
6016 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
6017 i == cfs_fail_val) {
6018 dt_write_lock(env, stripe, DT_TGT_CHILD);
6019 rc = lod_sub_ref_del(env, stripe, th);
6020 dt_write_unlock(env, stripe);
6024 rc = lod_sub_destroy(env, stripe, th);
6030 struct lod_obj_stripe_cb_data data = { { 0 } };
6032 data.locd_declare = false;
6033 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
6034 rc = lod_obj_for_each_stripe(env, lo, th, &data);
6041 * Implementation of dt_object_operations::do_declare_ref_add.
6043 * \see dt_object_operations::do_declare_ref_add() in the API description
6046 static int lod_declare_ref_add(const struct lu_env *env,
6047 struct dt_object *dt, struct thandle *th)
6049 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
6053 * Implementation of dt_object_operations::do_ref_add.
6055 * \see dt_object_operations::do_ref_add() in the API description for details.
6057 static int lod_ref_add(const struct lu_env *env,
6058 struct dt_object *dt, struct thandle *th)
6060 return lod_sub_ref_add(env, dt_object_child(dt), th);
6064 * Implementation of dt_object_operations::do_declare_ref_del.
6066 * \see dt_object_operations::do_declare_ref_del() in the API description
6069 static int lod_declare_ref_del(const struct lu_env *env,
6070 struct dt_object *dt, struct thandle *th)
6072 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
6076 * Implementation of dt_object_operations::do_ref_del
6078 * \see dt_object_operations::do_ref_del() in the API description for details.
6080 static int lod_ref_del(const struct lu_env *env,
6081 struct dt_object *dt, struct thandle *th)
6083 return lod_sub_ref_del(env, dt_object_child(dt), th);
6087 * Implementation of dt_object_operations::do_object_sync.
6089 * \see dt_object_operations::do_object_sync() in the API description
6092 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
6093 __u64 start, __u64 end)
6095 return dt_object_sync(env, dt_object_child(dt), start, end);
6099 * Implementation of dt_object_operations::do_object_unlock.
6101 * Used to release LDLM lock(s).
6103 * \see dt_object_operations::do_object_unlock() in the API description
6106 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
6107 struct ldlm_enqueue_info *einfo,
6108 union ldlm_policy_data *policy)
6110 struct lod_object *lo = lod_dt_obj(dt);
6111 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
6112 int slave_locks_size;
6116 if (slave_locks == NULL)
6119 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
6120 /* Note: for remote lock for single stripe dir, MDT will cancel
6121 * the lock by lockh directly */
6122 LASSERT(!dt_object_remote(dt_object_child(dt)));
6124 /* locks were unlocked in MDT layer */
6125 for (i = 0; i < slave_locks->ha_count; i++)
6126 LASSERT(!lustre_handle_is_used(&slave_locks->ha_handles[i]));
6129 * NB, ha_count may not equal to ldo_dir_stripe_count, because dir
6130 * layout may change, e.g., shrink dir layout after migration.
6132 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6133 if (lo->ldo_stripe[i])
6134 dt_invalidate(env, lo->ldo_stripe[i]);
6137 slave_locks_size = offsetof(typeof(*slave_locks),
6138 ha_handles[slave_locks->ha_count]);
6139 OBD_FREE(slave_locks, slave_locks_size);
6140 einfo->ei_cbdata = NULL;
6146 * Implementation of dt_object_operations::do_object_lock.
6148 * Used to get LDLM lock on the non-striped and striped objects.
6150 * \see dt_object_operations::do_object_lock() in the API description
6153 static int lod_object_lock(const struct lu_env *env,
6154 struct dt_object *dt,
6155 struct lustre_handle *lh,
6156 struct ldlm_enqueue_info *einfo,
6157 union ldlm_policy_data *policy)
6159 struct lod_object *lo = lod_dt_obj(dt);
6160 int slave_locks_size;
6161 struct lustre_handle_array *slave_locks = NULL;
6166 /* remote object lock */
6167 if (!einfo->ei_enq_slave) {
6168 LASSERT(dt_object_remote(dt));
6169 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
6173 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
6176 rc = lod_striping_load(env, lo);
6181 if (lo->ldo_dir_stripe_count <= 1)
6184 slave_locks_size = offsetof(typeof(*slave_locks),
6185 ha_handles[lo->ldo_dir_stripe_count]);
6186 /* Freed in lod_object_unlock */
6187 OBD_ALLOC(slave_locks, slave_locks_size);
6190 slave_locks->ha_count = lo->ldo_dir_stripe_count;
6192 /* striped directory lock */
6193 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6194 struct lustre_handle lockh;
6195 struct ldlm_res_id *res_id;
6196 struct dt_object *stripe;
6198 stripe = lo->ldo_stripe[i];
6202 res_id = &lod_env_info(env)->lti_res_id;
6203 fid_build_reg_res_name(lu_object_fid(&stripe->do_lu), res_id);
6204 einfo->ei_res_id = res_id;
6206 if (dt_object_remote(stripe)) {
6207 set_bit(i, (void *)slave_locks->ha_map);
6208 rc = dt_object_lock(env, stripe, &lockh, einfo, policy);
6210 struct ldlm_namespace *ns = einfo->ei_namespace;
6211 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
6212 ldlm_completion_callback completion = einfo->ei_cb_cp;
6213 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
6215 if (einfo->ei_mode == LCK_PW ||
6216 einfo->ei_mode == LCK_EX)
6217 dlmflags |= LDLM_FL_COS_INCOMPAT;
6219 LASSERT(ns != NULL);
6220 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS,
6221 policy, einfo->ei_mode,
6222 &dlmflags, blocking,
6224 NULL, 0, LVB_T_NONE,
6229 ldlm_lock_decref_and_cancel(
6230 &slave_locks->ha_handles[i],
6232 OBD_FREE(slave_locks, slave_locks_size);
6235 slave_locks->ha_handles[i] = lockh;
6237 einfo->ei_cbdata = slave_locks;
6243 * Implementation of dt_object_operations::do_invalidate.
6245 * \see dt_object_operations::do_invalidate() in the API description for details
6247 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
6249 return dt_invalidate(env, dt_object_child(dt));
6252 static int lod_declare_instantiate_components(const struct lu_env *env,
6253 struct lod_object *lo, struct thandle *th)
6255 struct lod_thread_info *info = lod_env_info(env);
6260 LASSERT(info->lti_count < lo->ldo_comp_cnt);
6262 for (i = 0; i < info->lti_count; i++) {
6263 rc = lod_qos_prep_create(env, lo, NULL, th,
6264 info->lti_comp_idx[i]);
6270 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6271 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
6272 &info->lti_buf, XATTR_NAME_LOV, 0, th);
6279 * Check OSTs for an existing component for further extension
6281 * Checks if OSTs are still healthy and not out of space. Gets free space
6282 * on OSTs (relative to allocation watermark rmb_low) and compares to
6283 * the proposed new_end for this component.
6285 * Decides whether or not to extend a component on its current OSTs.
6287 * \param[in] env execution environment for this thread
6288 * \param[in] lo object we're checking
6289 * \param[in] index index of this component
6290 * \param[in] extension_size extension size for this component
6291 * \param[in] extent layout extent for requested operation
6292 * \param[in] comp_extent extension component extent
6293 * \param[in] write if this is write operation
6295 * \retval true - OK to extend on current OSTs
6296 * \retval false - do not extend on current OSTs
6298 static bool lod_sel_osts_allowed(const struct lu_env *env,
6299 struct lod_object *lo,
6300 int index, __u64 extension_size,
6301 struct lu_extent *extent,
6302 struct lu_extent *comp_extent, int write)
6304 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[index];
6305 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6306 struct obd_statfs *sfs = &lod_env_info(env)->lti_osfs;
6307 __u64 available = 0;
6314 LASSERT(lod_comp->llc_stripe_count != 0);
6317 (extent->e_start == 0 && extent->e_end == OBD_OBJECT_EOF)) {
6318 /* truncate or append */
6319 size = extension_size;
6321 /* In case of write op, check the real write extent,
6322 * it may be larger than the extension_size */
6323 size = roundup(min(extent->e_end, comp_extent->e_end) -
6324 max(extent->e_start, comp_extent->e_start),
6327 /* extension_size is file level, so we must divide by stripe count to
6328 * compare it to available space on a single OST */
6329 size /= lod_comp->llc_stripe_count;
6331 lod_getref(&lod->lod_ost_descs);
6332 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
6333 int index = lod_comp->llc_ost_indices[i];
6334 struct lod_tgt_desc *ost = OST_TGT(lod, index);
6335 struct obd_statfs_info info = { 0 };
6336 int j, repeated = 0;
6340 /* Get the number of times this OST repeats in this component.
6341 * Note: inter-component repeats are not counted as this is
6342 * considered as a rare case: we try to not repeat OST in other
6343 * components if possible. */
6344 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6345 if (index != lod_comp->llc_ost_indices[j])
6348 /* already handled */
6354 if (j < lod_comp->llc_stripe_count)
6357 if (!cfs_bitmap_check(lod->lod_ost_bitmap, index)) {
6358 CDEBUG(D_LAYOUT, "ost %d no longer present\n", index);
6363 rc = dt_statfs_info(env, ost->ltd_tgt, sfs, &info);
6365 CDEBUG(D_LAYOUT, "statfs failed for ost %d, error %d\n",
6371 if (sfs->os_state & OS_STATE_ENOSPC ||
6372 sfs->os_state & OS_STATE_READONLY ||
6373 sfs->os_state & OS_STATE_DEGRADED) {
6374 CDEBUG(D_LAYOUT, "ost %d is not availble for SEL "
6375 "extension, state %u\n", index, sfs->os_state);
6381 available = sfs->os_bavail * sfs->os_bsize;
6382 /* 'available' is relative to the allocation threshold */
6383 available -= (__u64) info.os_reserved_mb_low << 20;
6385 CDEBUG(D_LAYOUT, "ost %d lowwm: %d highwm: %d, "
6386 "%llu %% blocks available, %llu %% blocks free\n",
6387 index, info.os_reserved_mb_low, info.os_reserved_mb_high,
6388 (100ull * sfs->os_bavail) / sfs->os_blocks,
6389 (100ull * sfs->os_bfree) / sfs->os_blocks);
6391 if (size * repeated > available) {
6393 CDEBUG(D_LAYOUT, "low space on ost %d, available %llu "
6394 "< extension size %llu\n", index, available,
6399 lod_putref(lod, &lod->lod_ost_descs);
6405 * Adjust extents after component removal
6407 * When we remove an extension component, we move the start of the next
6408 * component to match the start of the extension component, so no space is left
6411 * \param[in] env execution environment for this thread
6412 * \param[in] lo object
6413 * \param[in] max_comp layout component
6414 * \param[in] index index of this component
6416 * \retval 0 on success
6417 * \retval negative errno on error
6419 static void lod_sel_adjust_extents(const struct lu_env *env,
6420 struct lod_object *lo,
6421 int max_comp, int index)
6423 struct lod_layout_component *lod_comp = NULL;
6424 struct lod_layout_component *next = NULL;
6425 struct lod_layout_component *prev = NULL;
6426 __u64 new_start = 0;
6430 /* Extension space component */
6431 lod_comp = &lo->ldo_comp_entries[index];
6432 next = &lo->ldo_comp_entries[index + 1];
6433 prev = &lo->ldo_comp_entries[index - 1];
6435 LASSERT(lod_comp != NULL && prev != NULL && next != NULL);
6436 LASSERT(lod_comp->llc_flags & LCME_FL_EXTENSION);
6438 /* Previous is being removed */
6439 if (prev && prev->llc_id == LCME_ID_INVAL)
6440 new_start = prev->llc_extent.e_start;
6442 new_start = lod_comp->llc_extent.e_start;
6444 for (i = index + 1; i < max_comp; i++) {
6445 lod_comp = &lo->ldo_comp_entries[i];
6447 start = lod_comp->llc_extent.e_start;
6448 lod_comp->llc_extent.e_start = new_start;
6450 /* We only move zero length extendable components */
6451 if (!(start == lod_comp->llc_extent.e_end))
6454 LASSERT(!(lod_comp->llc_flags & LCME_FL_INIT));
6456 lod_comp->llc_extent.e_end = new_start;
6460 /* Calculate the proposed 'new end' for a component we're extending */
6461 static __u64 lod_extension_new_end(__u64 extension_size, __u64 extent_end,
6462 __u32 stripe_size, __u64 component_end,
6463 __u64 extension_end)
6467 LASSERT(extension_size != 0 && stripe_size != 0);
6469 /* Round up to extension size */
6470 if (extent_end == OBD_OBJECT_EOF) {
6471 new_end = OBD_OBJECT_EOF;
6473 /* Add at least extension_size to the previous component_end,
6474 * covering the req layout extent */
6475 new_end = max(extent_end - component_end, extension_size);
6476 new_end = roundup(new_end, extension_size);
6477 new_end += component_end;
6479 /* Component end must be min stripe size aligned */
6480 if (new_end % stripe_size) {
6481 CDEBUG(D_LAYOUT, "new component end is not aligned "
6482 "by the stripe size %u: [%llu, %llu) ext size "
6483 "%llu new end %llu, aligning\n",
6484 stripe_size, component_end, extent_end,
6485 extension_size, new_end);
6486 new_end = roundup(new_end, stripe_size);
6490 if (new_end < extent_end)
6491 new_end = OBD_OBJECT_EOF;
6494 /* Don't extend past the end of the extension component */
6495 if (new_end > extension_end)
6496 new_end = extension_end;
6501 /* As lod_sel_handler() could be re-entered for the same component several
6502 * times, this is the data for the next call. Fields could be changed to
6503 * component indexes when needed, (e.g. if there is no need to instantiate
6504 * all the previous components up to the current position) to tell the caller
6505 * where to start over from. */
6512 * Process extent updates for a particular layout component
6514 * Handle layout updates for a particular extension space component touched by
6515 * a layout update operation. Core function of self-extending PFL feature.
6517 * In general, this function processes exactly *one* stage of an extension
6518 * operation, modifying the layout accordingly, then returns to the caller.
6519 * The caller is responsible for restarting processing with the new layout,
6520 * which may repeatedly return to this function until the extension updates
6523 * This function does one of a few things to the layout:
6524 * 1. Extends the component before the current extension space component to
6525 * allow it to accomodate the requested operation (if space/policy permit that
6526 * component to continue on its current OSTs)
6528 * 2. If extension of the existing component fails, we do one of two things:
6529 * a. If there is a component after the extension space, we remove the
6530 * extension space component, move the start of the next component down
6531 * accordingly, then notify the caller to restart processing w/the new
6533 * b. If there is no following component, we try repeating the current
6534 * component, creating a new component using the current one as a
6535 * template (keeping its stripe properties but not specific striping),
6536 * and try assigning striping for this component. If there is sufficient
6537 * free space on the OSTs chosen for this component, it is instantiated
6538 * and i/o continues there.
6540 * If there is not sufficient space on the new OSTs, we remove this new
6541 * component & extend the current component.
6543 * Note further that uninited components followed by extension space can be zero
6544 * length meaning that we will try to extend them before initializing them, and
6545 * if that fails, they will be removed without initialization.
6547 * 3. If we extend to/beyond the end of an extension space component, that
6548 * component is exhausted (all of its range has been given to real components),
6549 * so we remove it and restart processing.
6551 * \param[in] env execution environment for this thread
6552 * \param[in,out] lo object to update the layout of
6553 * \param[in] extent layout extent for requested operation, update
6554 * layout to fit this operation
6555 * \param[in] th transaction handle for this operation
6556 * \param[in,out] max_comp the highest comp for the portion of the layout
6557 * we are operating on (For FLR, the chosen
6558 * replica). Updated because we may remove
6560 * \param[in] index index of the extension space component we're
6562 * \param[in] write if this is write op
6563 * \param[in,out] force if the extension is to be forced; set here
6564 to force it on the 2nd call for the same
6567 * \retval 0 on success
6568 * \retval negative errno on error
6570 static int lod_sel_handler(const struct lu_env *env,
6571 struct lod_object *lo,
6572 struct lu_extent *extent,
6573 struct thandle *th, int *max_comp,
6574 int index, int write,
6575 struct sel_data *sd)
6577 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6578 struct lod_thread_info *info = lod_env_info(env);
6579 struct lod_layout_component *lod_comp;
6580 struct lod_layout_component *prev;
6581 struct lod_layout_component *next = NULL;
6582 __u64 extension_size;
6589 /* First component cannot be extension space */
6591 CERROR("%s: "DFID" first component cannot be extension space\n",
6592 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
6596 lod_comp = &lo->ldo_comp_entries[index];
6597 prev = &lo->ldo_comp_entries[index - 1];
6598 if ((index + 1) < *max_comp)
6599 next = &lo->ldo_comp_entries[index + 1];
6601 /* extension size uses the stripe size field as KiB */
6602 extension_size = lod_comp->llc_stripe_size * SEL_UNIT_SIZE;
6604 CDEBUG(D_LAYOUT, "prev start %llu, extension start %llu, extension end"
6605 " %llu, extension size %llu\n", prev->llc_extent.e_start,
6606 lod_comp->llc_extent.e_start, lod_comp->llc_extent.e_end,
6609 /* Two extension space components cannot be adjacent & extension space
6610 * components cannot be init */
6611 if ((prev->llc_flags & LCME_FL_EXTENSION) ||
6612 !(ergo(next, !(next->llc_flags & LCME_FL_EXTENSION))) ||
6613 lod_comp_inited(lod_comp)) {
6614 CERROR("%s: "DFID" invalid extension space components\n",
6615 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
6619 if (!prev->llc_stripe) {
6620 CDEBUG(D_LAYOUT, "Previous component not inited\n");
6621 info->lti_count = 1;
6622 info->lti_comp_idx[0] = index - 1;
6623 rc = lod_declare_instantiate_components(env, lo, th);
6624 /* ENOSPC tells us we can't use this component. If there is
6625 * a next or we are repeating, we either spill over (next) or
6626 * extend the original comp (repeat). Otherwise, return the
6627 * error to the user. */
6628 if (rc == -ENOSPC && (next || sd->sd_repeat))
6634 if (sd->sd_force == 0 && rc == 0)
6635 rc = !lod_sel_osts_allowed(env, lo, index - 1,
6636 extension_size, extent,
6637 &lod_comp->llc_extent, write);
6639 repeated = !!(sd->sd_repeat);
6643 /* Extend previous component */
6645 new_end = lod_extension_new_end(extension_size, extent->e_end,
6646 prev->llc_stripe_size,
6647 prev->llc_extent.e_end,
6648 lod_comp->llc_extent.e_end);
6650 CDEBUG(D_LAYOUT, "new end %llu\n", new_end);
6651 lod_comp->llc_extent.e_start = new_end;
6652 prev->llc_extent.e_end = new_end;
6654 if (prev->llc_extent.e_end == lod_comp->llc_extent.e_end) {
6655 CDEBUG(D_LAYOUT, "Extension component exhausted\n");
6656 lod_comp->llc_id = LCME_ID_INVAL;
6660 /* rc == 1, failed to extend current component */
6663 /* Normal 'spillover' case - Remove the extension
6664 * space component & bring down the start of the next
6666 lod_comp->llc_id = LCME_ID_INVAL;
6668 if (!(prev->llc_flags & LCME_FL_INIT)) {
6669 prev->llc_id = LCME_ID_INVAL;
6672 lod_sel_adjust_extents(env, lo, *max_comp, index);
6673 } else if (lod_comp_inited(prev)) {
6674 /* If there is no next, and the previous component is
6675 * INIT'ed, try repeating the previous component. */
6676 LASSERT(repeated == 0);
6677 rc = lod_layout_repeat_comp(env, lo, index - 1);
6681 /* The previous component is a repeated component.
6682 * Record this so we don't keep trying to repeat it. */
6685 /* If the previous component is not INIT'ed, this may
6686 * be a component we have just instantiated but failed
6687 * to extend. Or even a repeated component we failed
6688 * to prepare a striping for. Do not repeat but instead
6689 * remove the repeated component & force the extention
6690 * of the original one */
6693 prev->llc_id = LCME_ID_INVAL;
6700 rc = lod_layout_del_prep_layout(env, lo, NULL);
6703 LASSERTF(-rc == change,
6704 "number deleted %d != requested %d\n", -rc,
6707 *max_comp = *max_comp + change;
6709 /* lod_del_prep_layout reallocates ldo_comp_entries, so we must
6710 * refresh these pointers before using them */
6711 lod_comp = &lo->ldo_comp_entries[index];
6712 prev = &lo->ldo_comp_entries[index - 1];
6713 CDEBUG(D_LAYOUT, "After extent updates: prev start %llu, current start "
6714 "%llu, current end %llu max_comp %d ldo_comp_cnt %d\n",
6715 prev->llc_extent.e_start, lod_comp->llc_extent.e_start,
6716 lod_comp->llc_extent.e_end, *max_comp, lo->ldo_comp_cnt);
6718 /* Layout changed successfully */
6723 * Declare layout extent updates
6725 * Handles extensions. Identifies extension components touched by current
6726 * operation and passes them to processing function.
6728 * Restarts with updated layouts from the processing function until the current
6729 * operation no longer touches an extension space component.
6731 * \param[in] env execution environment for this thread
6732 * \param[in,out] lo object to update the layout of
6733 * \param[in] extent layout extent for requested operation, update layout to
6734 * fit this operation
6735 * \param[in] th transaction handle for this operation
6736 * \param[in] pick identifies chosen mirror for FLR layouts
6737 * \param[in] write if this is write op
6739 * \retval 1 on layout changed, 0 on no change
6740 * \retval negative errno on error
6742 static int lod_declare_update_extents(const struct lu_env *env,
6743 struct lod_object *lo, struct lu_extent *extent,
6744 struct thandle *th, int pick, int write)
6746 struct lod_thread_info *info = lod_env_info(env);
6747 struct lod_layout_component *lod_comp;
6748 bool layout_changed = false;
6749 struct sel_data sd = { 0 };
6757 /* This makes us work on the components of the chosen mirror */
6758 start_index = lo->ldo_mirrors[pick].lme_start;
6759 max_comp = lo->ldo_mirrors[pick].lme_end + 1;
6760 if (lo->ldo_flr_state == LCM_FL_NONE)
6761 LASSERT(start_index == 0 && max_comp == lo->ldo_comp_cnt);
6763 CDEBUG(D_LAYOUT, "extent->e_start %llu, extent->e_end %llu\n",
6764 extent->e_start, extent->e_end);
6765 for (i = start_index; i < max_comp; i++) {
6766 lod_comp = &lo->ldo_comp_entries[i];
6768 /* We've passed all components of interest */
6769 if (lod_comp->llc_extent.e_start >= extent->e_end)
6772 if (lod_comp->llc_flags & LCME_FL_EXTENSION) {
6773 layout_changed = true;
6774 rc = lod_sel_handler(env, lo, extent, th, &max_comp,
6779 /* Nothing has changed behind the prev one */
6785 /* We may have added or removed components. If so, we must update the
6786 * start & ends of all the mirrors after the current one, and the end
6787 * of the current mirror. */
6788 change = max_comp - 1 - lo->ldo_mirrors[pick].lme_end;
6790 lo->ldo_mirrors[pick].lme_end += change;
6791 for (i = pick + 1; i < lo->ldo_mirror_count; i++) {
6792 lo->ldo_mirrors[i].lme_start += change;
6793 lo->ldo_mirrors[i].lme_end += change;
6799 /* The amount of components has changed, adjust the lti_comp_idx */
6800 rc2 = lod_layout_data_init(info, lo->ldo_comp_cnt);
6802 return rc < 0 ? rc : rc2 < 0 ? rc2 : layout_changed;
6805 /* If striping is already instantiated or INIT'ed DOM? */
6806 static bool lod_is_instantiation_needed(struct lod_layout_component *comp)
6808 return !(((lov_pattern(comp->llc_pattern) == LOV_PATTERN_MDT) &&
6809 lod_comp_inited(comp)) || comp->llc_stripe);
6813 * Declare layout update for a non-FLR layout.
6815 * \param[in] env execution environment for this thread
6816 * \param[in,out] lo object to update the layout of
6817 * \param[in] layout layout intent for requested operation, "update" is
6818 * a process of reacting to this
6819 * \param[in] buf buffer containing lov ea (see comment on usage inline)
6820 * \param[in] th transaction handle for this operation
6822 * \retval 0 on success
6823 * \retval negative errno on error
6825 static int lod_declare_update_plain(const struct lu_env *env,
6826 struct lod_object *lo, struct layout_intent *layout,
6827 const struct lu_buf *buf, struct thandle *th)
6829 struct lod_thread_info *info = lod_env_info(env);
6830 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6831 struct lod_layout_component *lod_comp;
6832 struct lov_comp_md_v1 *comp_v1 = NULL;
6833 bool layout_changed = false;
6834 bool replay = false;
6838 LASSERT(lo->ldo_flr_state == LCM_FL_NONE);
6841 * In case the client is passing lovea, which only happens during
6842 * the replay of layout intent write RPC for now, we may need to
6843 * parse the lovea and apply new layout configuration.
6845 if (buf && buf->lb_len) {
6846 struct lov_user_md_v1 *v1 = buf->lb_buf;
6848 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
6849 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
6850 LOV_MAGIC_COMP_V1)) {
6851 CERROR("%s: the replay buffer of layout extend "
6852 "(magic %#x) does not contain expected "
6853 "composite layout.\n",
6854 lod2obd(d)->obd_name, v1->lmm_magic);
6855 GOTO(out, rc = -EINVAL);
6858 rc = lod_use_defined_striping(env, lo, buf);
6861 lo->ldo_comp_cached = 1;
6863 rc = lod_get_lov_ea(env, lo);
6866 /* old on-disk EA is stored in info->lti_buf */
6867 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
6869 layout_changed = true;
6871 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
6875 /* non replay path */
6876 rc = lod_striping_load(env, lo);
6881 /* Make sure defined layout covers the requested write range. */
6882 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
6883 if (lo->ldo_comp_cnt > 1 &&
6884 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
6885 lod_comp->llc_extent.e_end < layout->li_extent.e_end) {
6886 CDEBUG_LIMIT(replay ? D_ERROR : D_LAYOUT,
6887 "%s: the defined layout [0, %#llx) does not "
6888 "covers the write range "DEXT"\n",
6889 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
6890 PEXT(&layout->li_extent));
6891 GOTO(out, rc = -EINVAL);
6894 CDEBUG(D_LAYOUT, "%s: "DFID": update components "DEXT"\n",
6895 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
6896 PEXT(&layout->li_extent));
6899 rc = lod_declare_update_extents(env, lo, &layout->li_extent,
6900 th, 0, layout->li_opc == LAYOUT_INTENT_WRITE);
6904 layout_changed = true;
6908 * Iterate ld->ldo_comp_entries, find the component whose extent under
6909 * the write range and not instantianted.
6911 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6912 lod_comp = &lo->ldo_comp_entries[i];
6914 if (lod_comp->llc_extent.e_start >= layout->li_extent.e_end)
6918 /* If striping is instantiated or INIT'ed DOM skip */
6919 if (!lod_is_instantiation_needed(lod_comp))
6923 * In replay path, lod_comp is the EA passed by
6924 * client replay buffer, comp_v1 is the pre-recovery
6925 * on-disk EA, we'd sift out those components which
6926 * were init-ed in the on-disk EA.
6928 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
6933 * this component hasn't instantiated in normal path, or during
6934 * replay it needs replay the instantiation.
6937 /* A released component is being extended */
6938 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
6939 GOTO(out, rc = -EINVAL);
6941 LASSERT(info->lti_comp_idx != NULL);
6942 info->lti_comp_idx[info->lti_count++] = i;
6943 layout_changed = true;
6946 if (!layout_changed)
6949 lod_obj_inc_layout_gen(lo);
6950 rc = lod_declare_instantiate_components(env, lo, th);
6954 lod_striping_free(env, lo);
6958 static inline int lod_comp_index(struct lod_object *lo,
6959 struct lod_layout_component *lod_comp)
6961 LASSERT(lod_comp >= lo->ldo_comp_entries &&
6962 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
6964 return lod_comp - lo->ldo_comp_entries;
6968 * Stale other mirrors by writing extent.
6970 static int lod_stale_components(const struct lu_env *env, struct lod_object *lo,
6971 int primary, struct lu_extent *extent,
6974 struct lod_layout_component *pri_comp, *lod_comp;
6975 struct lu_extent pri_extent;
6980 /* The writing extent decides which components in the primary
6981 * are affected... */
6982 CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
6985 lod_foreach_mirror_comp(pri_comp, lo, primary) {
6986 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
6989 CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
6990 lod_comp_index(lo, pri_comp),
6991 PEXT(&pri_comp->llc_extent));
6993 pri_extent.e_start = pri_comp->llc_extent.e_start;
6994 pri_extent.e_end = pri_comp->llc_extent.e_end;
6996 for (i = 0; i < lo->ldo_mirror_count; i++) {
6999 rc = lod_declare_update_extents(env, lo, &pri_extent,
7001 /* if update_extents changed the layout, it may have
7002 * reallocated the component array, so start over to
7003 * avoid using stale pointers */
7009 /* ... and then stale other components that are
7010 * overlapping with primary components */
7011 lod_foreach_mirror_comp(lod_comp, lo, i) {
7012 if (!lu_extent_is_overlapped(
7014 &lod_comp->llc_extent))
7017 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
7018 i, lod_comp_index(lo, lod_comp));
7020 lod_comp->llc_flags |= LCME_FL_STALE;
7021 lo->ldo_mirrors[i].lme_stale = 1;
7030 * check an OST's availability
7031 * \param[in] env execution environment
7032 * \param[in] lo lod object
7033 * \param[in] dt dt object
7034 * \param[in] index mirror index
7036 * \retval negative if failed
7037 * \retval 1 if \a dt is available
7038 * \retval 0 if \a dt is not available
7040 static inline int lod_check_ost_avail(const struct lu_env *env,
7041 struct lod_object *lo,
7042 struct dt_object *dt, int index)
7044 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7045 struct lod_tgt_desc *ost;
7047 int type = LU_SEQ_RANGE_OST;
7050 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &idx, &type);
7052 CERROR("%s: can't locate "DFID":rc = %d\n",
7053 lod2obd(lod)->obd_name, PFID(lu_object_fid(&dt->do_lu)),
7058 ost = OST_TGT(lod, idx);
7059 if (ost->ltd_statfs.os_state &
7060 (OS_STATE_READONLY | OS_STATE_ENOSPC | OS_STATE_ENOINO |
7061 OS_STATE_NOPRECREATE) ||
7062 ost->ltd_active == 0) {
7063 CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail, rc = %d\n",
7064 PFID(lod_object_fid(lo)), index, idx, rc);
7072 * Pick primary mirror for write
7073 * \param[in] env execution environment
7074 * \param[in] lo object
7075 * \param[in] extent write range
7077 static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo,
7078 struct lu_extent *extent)
7080 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7081 unsigned int seq = 0;
7082 struct lod_layout_component *lod_comp;
7084 int picked = -1, second_pick = -1, third_pick = -1;
7087 if (OBD_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
7088 get_random_bytes(&seq, sizeof(seq));
7089 seq %= lo->ldo_mirror_count;
7093 * Pick a mirror as the primary, and check the availability of OSTs.
7095 * This algo can be revised later after knowing the topology of
7098 lod_qos_statfs_update(env, lod, &lod->lod_ost_descs);
7099 for (i = 0; i < lo->ldo_mirror_count; i++) {
7100 bool ost_avail = true;
7101 int index = (i + seq) % lo->ldo_mirror_count;
7103 if (lo->ldo_mirrors[index].lme_stale) {
7104 CDEBUG(D_LAYOUT, DFID": mirror %d stale\n",
7105 PFID(lod_object_fid(lo)), index);
7109 /* 2nd pick is for the primary mirror containing unavail OST */
7110 if (lo->ldo_mirrors[index].lme_primary && second_pick < 0)
7111 second_pick = index;
7113 /* 3rd pick is for non-primary mirror containing unavail OST */
7114 if (second_pick < 0 && third_pick < 0)
7118 * we found a non-primary 1st pick, we'd like to find a
7119 * potential pirmary mirror.
7121 if (picked >= 0 && !lo->ldo_mirrors[index].lme_primary)
7124 /* check the availability of OSTs */
7125 lod_foreach_mirror_comp(lod_comp, lo, index) {
7126 if (!lod_comp_inited(lod_comp) || !lod_comp->llc_stripe)
7129 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
7130 struct dt_object *dt = lod_comp->llc_stripe[j];
7132 rc = lod_check_ost_avail(env, lo, dt, index);
7139 } /* for all dt object in one component */
7142 } /* for all components in a mirror */
7145 * the OSTs where allocated objects locates in the components
7146 * of the mirror are available.
7151 /* this mirror has all OSTs available */
7155 * primary with all OSTs are available, this is the perfect
7158 if (lo->ldo_mirrors[index].lme_primary)
7160 } /* for all mirrors */
7162 /* failed to pick a sound mirror, lower our expectation */
7164 picked = second_pick;
7166 picked = third_pick;
7173 static int lod_prepare_resync_mirror(const struct lu_env *env,
7174 struct lod_object *lo,
7177 struct lod_thread_info *info = lod_env_info(env);
7178 struct lod_layout_component *lod_comp;
7179 bool neg = !!(MIRROR_ID_NEG & mirror_id);
7182 mirror_id &= ~MIRROR_ID_NEG;
7184 for (i = 0; i < lo->ldo_mirror_count; i++) {
7185 if ((!neg && lo->ldo_mirrors[i].lme_id != mirror_id) ||
7186 (neg && lo->ldo_mirrors[i].lme_id == mirror_id))
7189 lod_foreach_mirror_comp(lod_comp, lo, i) {
7190 if (lod_comp_inited(lod_comp))
7193 info->lti_comp_idx[info->lti_count++] =
7194 lod_comp_index(lo, lod_comp);
7202 * figure out the components should be instantiated for resync.
7204 static int lod_prepare_resync(const struct lu_env *env, struct lod_object *lo,
7205 struct lu_extent *extent)
7207 struct lod_thread_info *info = lod_env_info(env);
7208 struct lod_layout_component *lod_comp;
7209 unsigned int need_sync = 0;
7213 DFID": instantiate all stale components in "DEXT"\n",
7214 PFID(lod_object_fid(lo)), PEXT(extent));
7217 * instantiate all components within this extent, even non-stale
7220 for (i = 0; i < lo->ldo_mirror_count; i++) {
7221 if (!lo->ldo_mirrors[i].lme_stale)
7224 lod_foreach_mirror_comp(lod_comp, lo, i) {
7225 if (!lu_extent_is_overlapped(extent,
7226 &lod_comp->llc_extent))
7231 if (lod_comp_inited(lod_comp))
7234 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
7235 i, lod_comp_index(lo, lod_comp));
7236 info->lti_comp_idx[info->lti_count++] =
7237 lod_comp_index(lo, lod_comp);
7241 return need_sync ? 0 : -EALREADY;
7244 static int lod_declare_update_rdonly(const struct lu_env *env,
7245 struct lod_object *lo, struct md_layout_change *mlc,
7248 struct lod_thread_info *info = lod_env_info(env);
7249 struct lu_attr *layout_attr = &info->lti_layout_attr;
7250 struct lod_layout_component *lod_comp;
7251 struct lu_extent extent = { 0 };
7255 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
7256 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
7257 mlc->mlc_opc == MD_LAYOUT_RESYNC);
7258 LASSERT(lo->ldo_mirror_count > 0);
7260 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7261 struct layout_intent *layout = mlc->mlc_intent;
7262 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
7265 extent = layout->li_extent;
7266 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
7267 PFID(lod_object_fid(lo)), PEXT(&extent));
7269 picked = lod_primary_pick(env, lo, &extent);
7273 CDEBUG(D_LAYOUT, DFID": picked mirror id %u as primary\n",
7274 PFID(lod_object_fid(lo)),
7275 lo->ldo_mirrors[picked].lme_id);
7277 /* Update extents of primary before staling */
7278 rc = lod_declare_update_extents(env, lo, &extent, th, picked,
7283 if (layout->li_opc == LAYOUT_INTENT_TRUNC) {
7285 * trunc transfers [0, size) in the intent extent, we'd
7286 * stale components overlapping [size, eof).
7288 extent.e_start = extent.e_end;
7289 extent.e_end = OBD_OBJECT_EOF;
7292 /* stale overlapping components from other mirrors */
7293 rc = lod_stale_components(env, lo, picked, &extent, th);
7297 /* restore truncate intent extent */
7298 if (layout->li_opc == LAYOUT_INTENT_TRUNC)
7299 extent.e_end = extent.e_start;
7301 /* instantiate components for the picked mirror, start from 0 */
7304 lod_foreach_mirror_comp(lod_comp, lo, picked) {
7305 if (!lu_extent_is_overlapped(&extent,
7306 &lod_comp->llc_extent))
7309 if (!lod_is_instantiation_needed(lod_comp))
7312 info->lti_comp_idx[info->lti_count++] =
7313 lod_comp_index(lo, lod_comp);
7316 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
7317 } else { /* MD_LAYOUT_RESYNC */
7321 * could contain multiple non-stale mirrors, so we need to
7322 * prep uninited all components assuming any non-stale mirror
7323 * could be picked as the primary mirror.
7325 if (mlc->mlc_mirror_id == 0) {
7327 for (i = 0; i < lo->ldo_mirror_count; i++) {
7328 if (lo->ldo_mirrors[i].lme_stale)
7331 lod_foreach_mirror_comp(lod_comp, lo, i) {
7332 if (!lod_comp_inited(lod_comp))
7336 lod_comp->llc_extent.e_end)
7338 lod_comp->llc_extent.e_end;
7341 rc = lod_prepare_resync(env, lo, &extent);
7345 /* mirror write, try to init its all components */
7346 rc = lod_prepare_resync_mirror(env, lo,
7347 mlc->mlc_mirror_id);
7352 /* change the file state to SYNC_PENDING */
7353 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
7356 /* Reset the layout version once it's becoming too large.
7357 * This way it can make sure that the layout version is
7358 * monotonously increased in this writing era. */
7359 lod_obj_inc_layout_gen(lo);
7360 if (lo->ldo_layout_gen > (LCME_ID_MAX >> 1)) {
7361 __u32 layout_version;
7363 get_random_bytes(&layout_version, sizeof(layout_version));
7364 lo->ldo_layout_gen = layout_version & 0xffff;
7367 rc = lod_declare_instantiate_components(env, lo, th);
7371 layout_attr->la_valid = LA_LAYOUT_VERSION;
7372 layout_attr->la_layout_version = 0; /* set current version */
7373 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
7374 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
7375 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
7381 lod_striping_free(env, lo);
7385 static int lod_declare_update_write_pending(const struct lu_env *env,
7386 struct lod_object *lo, struct md_layout_change *mlc,
7389 struct lod_thread_info *info = lod_env_info(env);
7390 struct lu_attr *layout_attr = &info->lti_layout_attr;
7391 struct lod_layout_component *lod_comp;
7392 struct lu_extent extent = { 0 };
7398 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
7399 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
7400 mlc->mlc_opc == MD_LAYOUT_RESYNC);
7402 /* look for the primary mirror */
7403 for (i = 0; i < lo->ldo_mirror_count; i++) {
7404 if (lo->ldo_mirrors[i].lme_stale)
7407 LASSERTF(primary < 0, DFID " has multiple primary: %u / %u",
7408 PFID(lod_object_fid(lo)),
7409 lo->ldo_mirrors[i].lme_id,
7410 lo->ldo_mirrors[primary].lme_id);
7415 CERROR(DFID ": doesn't have a primary mirror\n",
7416 PFID(lod_object_fid(lo)));
7417 GOTO(out, rc = -ENODATA);
7420 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
7421 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
7423 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
7425 /* for LAYOUT_WRITE opc, it has to do the following operations:
7426 * 1. stale overlapping componets from stale mirrors;
7427 * 2. instantiate components of the primary mirror;
7428 * 3. transfter layout version to all objects of the primary;
7430 * for LAYOUT_RESYNC opc, it will do:
7431 * 1. instantiate components of all stale mirrors;
7432 * 2. transfer layout version to all objects to close write era. */
7434 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7435 struct layout_intent *layout = mlc->mlc_intent;
7436 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
7438 LASSERT(mlc->mlc_intent != NULL);
7440 extent = mlc->mlc_intent->li_extent;
7442 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
7443 PFID(lod_object_fid(lo)), PEXT(&extent));
7445 /* 1. Update extents of primary before staling */
7446 rc = lod_declare_update_extents(env, lo, &extent, th, primary,
7451 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC) {
7453 * trunc transfers [0, size) in the intent extent, we'd
7454 * stale components overlapping [size, eof).
7456 extent.e_start = extent.e_end;
7457 extent.e_end = OBD_OBJECT_EOF;
7460 /* 2. stale overlapping components */
7461 rc = lod_stale_components(env, lo, primary, &extent, th);
7465 /* 3. find the components which need instantiating.
7466 * instantiate [0, mlc->mlc_intent->e_end) */
7468 /* restore truncate intent extent */
7469 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC)
7470 extent.e_end = extent.e_start;
7473 lod_foreach_mirror_comp(lod_comp, lo, primary) {
7474 if (!lu_extent_is_overlapped(&extent,
7475 &lod_comp->llc_extent))
7478 if (!lod_is_instantiation_needed(lod_comp))
7481 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
7482 primary, lod_comp_index(lo, lod_comp));
7483 info->lti_comp_idx[info->lti_count++] =
7484 lod_comp_index(lo, lod_comp);
7486 } else { /* MD_LAYOUT_RESYNC */
7487 if (mlc->mlc_mirror_id == 0) {
7489 lod_foreach_mirror_comp(lod_comp, lo, primary) {
7490 if (!lod_comp_inited(lod_comp))
7493 extent.e_end = lod_comp->llc_extent.e_end;
7496 rc = lod_prepare_resync(env, lo, &extent);
7500 /* mirror write, try to init its all components */
7501 rc = lod_prepare_resync_mirror(env, lo,
7502 mlc->mlc_mirror_id);
7507 /* change the file state to SYNC_PENDING */
7508 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
7511 rc = lod_declare_instantiate_components(env, lo, th);
7515 /* 3. transfer layout version to OST objects.
7516 * transfer new layout version to OST objects so that stale writes
7517 * can be denied. It also ends an era of writing by setting
7518 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
7519 * send write RPC; only resync RPCs could do it. */
7520 layout_attr->la_valid = LA_LAYOUT_VERSION;
7521 layout_attr->la_layout_version = 0; /* set current version */
7522 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
7523 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
7524 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
7528 lod_obj_inc_layout_gen(lo);
7531 lod_striping_free(env, lo);
7535 static int lod_declare_update_sync_pending(const struct lu_env *env,
7536 struct lod_object *lo, struct md_layout_change *mlc,
7539 struct lod_thread_info *info = lod_env_info(env);
7540 unsigned sync_components = 0;
7541 unsigned resync_components = 0;
7546 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
7547 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
7548 mlc->mlc_opc == MD_LAYOUT_WRITE);
7550 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
7551 PFID(lod_object_fid(lo)), mlc->mlc_opc);
7553 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7554 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
7555 PFID(lod_object_fid(lo)));
7557 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
7558 return lod_declare_update_write_pending(env, lo, mlc, th);
7561 /* MD_LAYOUT_RESYNC_DONE */
7563 for (i = 0; i < lo->ldo_comp_cnt; i++) {
7564 struct lod_layout_component *lod_comp;
7567 lod_comp = &lo->ldo_comp_entries[i];
7569 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
7574 for (j = 0; j < mlc->mlc_resync_count; j++) {
7575 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
7578 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
7579 lod_comp->llc_flags &= ~LCME_FL_STALE;
7580 resync_components++;
7586 for (i = 0; i < mlc->mlc_resync_count; i++) {
7587 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
7590 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
7591 "or already synced\n", PFID(lod_object_fid(lo)),
7592 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
7593 GOTO(out, rc = -EINVAL);
7596 if (!sync_components || (mlc->mlc_resync_count && !resync_components)) {
7597 CDEBUG(D_LAYOUT, DFID": no mirror in sync\n",
7598 PFID(lod_object_fid(lo)));
7600 /* tend to return an error code here to prevent
7601 * the MDT from setting SoM attribute */
7602 GOTO(out, rc = -EINVAL);
7605 CDEBUG(D_LAYOUT, DFID": synced %u resynced %u/%zu components\n",
7606 PFID(lod_object_fid(lo)),
7607 sync_components, resync_components, mlc->mlc_resync_count);
7609 lo->ldo_flr_state = LCM_FL_RDONLY;
7610 lod_obj_inc_layout_gen(lo);
7612 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
7613 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
7614 &info->lti_buf, XATTR_NAME_LOV, 0, th);
7619 lod_striping_free(env, lo);
7623 typedef int (*mlc_handler)(const struct lu_env *env, struct dt_object *dt,
7624 const struct md_layout_change *mlc,
7625 struct thandle *th);
7628 * Attach stripes after target's for migrating directory. NB, we
7629 * only need to declare this, the actual work is done inside
7630 * lod_xattr_set_lmv().
7632 * \param[in] env execution environment
7633 * \param[in] dt target object
7634 * \param[in] mlc layout change data
7635 * \param[in] th transaction handle
7637 * \retval 0 on success
7638 * \retval negative if failed
7640 static int lod_dir_declare_layout_attach(const struct lu_env *env,
7641 struct dt_object *dt,
7642 const struct md_layout_change *mlc,
7645 struct lod_thread_info *info = lod_env_info(env);
7646 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
7647 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
7648 struct lod_object *lo = lod_dt_obj(dt);
7649 struct dt_object *next = dt_object_child(dt);
7650 struct dt_object_format *dof = &info->lti_format;
7651 struct lmv_mds_md_v1 *lmv = mlc->mlc_buf.lb_buf;
7652 struct dt_object **stripes;
7653 __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
7654 struct lu_fid *fid = &info->lti_fid;
7655 struct lod_tgt_desc *tgt;
7656 struct dt_object *dto;
7657 struct dt_device *tgt_dt;
7658 int type = LU_SEQ_RANGE_ANY;
7659 struct dt_insert_rec *rec = &info->lti_dt_rec;
7660 char *stripe_name = info->lti_key;
7661 struct lu_name *sname;
7662 struct linkea_data ldata = { NULL };
7663 struct lu_buf linkea_buf;
7670 if (!lmv_is_sane(lmv))
7673 if (!dt_try_as_dir(env, dt))
7676 dof->dof_type = DFT_DIR;
7679 sizeof(*stripes) * (lo->ldo_dir_stripe_count + stripe_count));
7683 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
7684 stripes[i] = lo->ldo_stripe[i];
7686 rec->rec_type = S_IFDIR;
7688 for (i = 0; i < stripe_count; i++) {
7690 &lmv->lmv_stripe_fids[i]);
7691 if (!fid_is_sane(fid))
7694 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
7698 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
7699 tgt_dt = lod->lod_child;
7701 tgt = LTD_TGT(ltd, idx);
7703 GOTO(out, rc = -ESTALE);
7704 tgt_dt = tgt->ltd_tgt;
7707 dto = dt_locate_at(env, tgt_dt, fid,
7708 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
7711 GOTO(out, rc = PTR_ERR(dto));
7713 stripes[i + lo->ldo_dir_stripe_count] = dto;
7715 if (!dt_try_as_dir(env, dto))
7716 GOTO(out, rc = -ENOTDIR);
7718 rc = lod_sub_declare_ref_add(env, dto, th);
7722 rec->rec_fid = lu_object_fid(&dto->do_lu);
7723 rc = lod_sub_declare_insert(env, dto,
7724 (const struct dt_rec *)rec,
7725 (const struct dt_key *)dot, th);
7729 rc = lod_sub_declare_insert(env, dto,
7730 (const struct dt_rec *)rec,
7731 (const struct dt_key *)dotdot, th);
7735 rc = lod_sub_declare_xattr_set(env, dto, &mlc->mlc_buf,
7736 XATTR_NAME_LMV, 0, th);
7740 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
7741 PFID(lu_object_fid(&dto->do_lu)),
7742 i + lo->ldo_dir_stripe_count);
7744 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
7745 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
7746 sname, lu_object_fid(&dt->do_lu));
7750 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
7751 linkea_buf.lb_len = ldata.ld_leh->leh_len;
7752 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
7753 XATTR_NAME_LINK, 0, th);
7757 rc = lod_sub_declare_insert(env, next,
7758 (const struct dt_rec *)rec,
7759 (const struct dt_key *)stripe_name,
7764 rc = lod_sub_declare_ref_add(env, next, th);
7770 OBD_FREE(lo->ldo_stripe,
7771 sizeof(*stripes) * lo->ldo_dir_stripes_allocated);
7772 lo->ldo_stripe = stripes;
7773 lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count;
7774 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type);
7775 lo->ldo_dir_stripe_count += stripe_count;
7776 lo->ldo_dir_stripes_allocated += stripe_count;
7778 /* plain directory split creates target as a plain directory, while
7779 * after source attached as the first stripe, it becomes a striped
7780 * directory, set correct do_index_ops, otherwise it can't be unlinked.
7782 dt->do_index_ops = &lod_striped_index_ops;
7786 i = lo->ldo_dir_stripe_count;
7787 while (i < lo->ldo_dir_stripe_count + stripe_count && stripes[i])
7788 dt_object_put(env, stripes[i++]);
7791 sizeof(*stripes) * (stripe_count + lo->ldo_dir_stripe_count));
7795 static int lod_dir_declare_layout_detach(const struct lu_env *env,
7796 struct dt_object *dt,
7797 const struct md_layout_change *unused,
7800 struct lod_thread_info *info = lod_env_info(env);
7801 struct lod_object *lo = lod_dt_obj(dt);
7802 struct dt_object *next = dt_object_child(dt);
7803 char *stripe_name = info->lti_key;
7804 struct dt_object *dto;
7808 if (!dt_try_as_dir(env, dt))
7811 if (!lo->ldo_dir_stripe_count)
7812 return lod_sub_declare_delete(env, next,
7813 (const struct dt_key *)dotdot, th);
7815 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
7816 dto = lo->ldo_stripe[i];
7820 if (!dt_try_as_dir(env, dto))
7823 rc = lod_sub_declare_delete(env, dto,
7824 (const struct dt_key *)dotdot, th);
7828 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
7829 PFID(lu_object_fid(&dto->do_lu)), i);
7831 rc = lod_sub_declare_delete(env, next,
7832 (const struct dt_key *)stripe_name, th);
7836 rc = lod_sub_declare_ref_del(env, next, th);
7844 static int dt_dir_is_empty(const struct lu_env *env,
7845 struct dt_object *obj)
7848 const struct dt_it_ops *iops;
7853 if (!dt_try_as_dir(env, obj))
7856 iops = &obj->do_index_ops->dio_it;
7857 it = iops->init(env, obj, LUDA_64BITHASH);
7859 RETURN(PTR_ERR(it));
7861 rc = iops->get(env, it, (const struct dt_key *)"");
7865 for (rc = 0, i = 0; rc == 0 && i < 3; ++i)
7866 rc = iops->next(env, it);
7872 /* Huh? Index contains no zero key? */
7877 iops->fini(env, it);
7882 static int lod_dir_declare_layout_shrink(const struct lu_env *env,
7883 struct dt_object *dt,
7884 const struct md_layout_change *mlc,
7887 struct lod_thread_info *info = lod_env_info(env);
7888 struct lod_object *lo = lod_dt_obj(dt);
7889 struct dt_object *next = dt_object_child(dt);
7890 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
7891 __u32 final_stripe_count;
7892 char *stripe_name = info->lti_key;
7893 struct lu_buf *lmv_buf = &info->lti_buf;
7894 struct dt_object *dto;
7900 if (!dt_try_as_dir(env, dt))
7903 /* shouldn't be called on plain directory */
7904 LASSERT(lo->ldo_dir_stripe_count);
7906 lmv_buf->lb_buf = &info->lti_lmv.lmv_md_v1;
7907 lmv_buf->lb_len = sizeof(info->lti_lmv.lmv_md_v1);
7909 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
7910 LASSERT(final_stripe_count &&
7911 final_stripe_count < lo->ldo_dir_stripe_count);
7913 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
7914 dto = lo->ldo_stripe[i];
7918 if (i < final_stripe_count) {
7919 if (final_stripe_count == 1)
7922 rc = lod_sub_declare_xattr_set(env, dto, lmv_buf,
7924 LU_XATTR_REPLACE, th);
7931 rc = dt_dir_is_empty(env, dto);
7935 rc = lod_sub_declare_ref_del(env, dto, th);
7939 rc = lod_sub_declare_destroy(env, dto, th);
7943 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
7944 PFID(lu_object_fid(&dto->do_lu)), i);
7946 rc = lod_sub_declare_delete(env, next,
7947 (const struct dt_key *)stripe_name, th);
7951 rc = lod_sub_declare_ref_del(env, next, th);
7956 rc = lod_sub_declare_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
7957 LU_XATTR_REPLACE, th);
7962 * Allocate stripes for split directory.
7964 * \param[in] env execution environment
7965 * \param[in] dt target object
7966 * \param[in] mlc layout change data
7967 * \param[in] th transaction handle
7969 * \retval 0 on success
7970 * \retval negative if failed
7972 static int lod_dir_declare_layout_split(const struct lu_env *env,
7973 struct dt_object *dt,
7974 const struct md_layout_change *mlc,
7977 struct lod_thread_info *info = lod_env_info(env);
7978 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
7979 struct lod_object *lo = lod_dt_obj(dt);
7980 struct dt_object_format *dof = &info->lti_format;
7981 struct lmv_user_md_v1 *lum = mlc->mlc_spec->u.sp_ea.eadata;
7982 struct dt_object **stripes;
7990 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
7991 LASSERT(le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT);
7993 saved_count = lo->ldo_dir_stripes_allocated;
7994 stripe_count = le32_to_cpu(lum->lum_stripe_count);
7995 if (stripe_count <= saved_count)
7998 dof->dof_type = DFT_DIR;
8000 OBD_ALLOC(stripes, sizeof(*stripes) * stripe_count);
8004 for (i = 0; i < lo->ldo_dir_stripes_allocated; i++)
8005 stripes[i] = lo->ldo_stripe[i];
8007 lod_qos_statfs_update(env, lod, &lod->lod_mdt_descs);
8008 rc = lod_mdt_alloc_qos(env, lo, stripes, saved_count, stripe_count);
8010 rc = lod_mdt_alloc_rr(env, lo, stripes, saved_count,
8013 OBD_FREE(stripes, sizeof(*stripes) * stripe_count);
8017 LASSERT(rc > saved_count);
8018 OBD_FREE(lo->ldo_stripe,
8019 sizeof(*stripes) * lo->ldo_dir_stripes_allocated);
8020 lo->ldo_stripe = stripes;
8021 lo->ldo_dir_striped = 1;
8022 lo->ldo_dir_stripe_count = rc;
8023 lo->ldo_dir_stripes_allocated = stripe_count;
8024 lo->ldo_dir_split_hash = lo->ldo_dir_hash_type;
8025 lo->ldo_dir_hash_type = le32_to_cpu(lum->lum_hash_type);
8026 if (!lmv_is_known_hash_type(lo->ldo_dir_hash_type))
8027 lo->ldo_dir_hash_type =
8028 lod->lod_mdt_descs.ltd_lmv_desc.ld_pattern;
8029 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_SPLIT | LMV_HASH_FLAG_MIGRATION;
8030 lo->ldo_dir_split_offset = saved_count;
8031 lo->ldo_dir_layout_version++;
8032 lo->ldo_dir_stripe_loaded = 1;
8034 rc = lod_dir_declare_create_stripes(env, dt, mlc->mlc_attr, dof, th);
8036 lod_striping_free(env, lo);
8042 * detach all stripes from dir master object, NB, stripes are not destroyed, but
8043 * deleted from it's parent namespace, this function is called in two places:
8044 * 1. mdd_migrate_mdt() detach stripes from source, and attach them to
8046 * 2. mdd_dir_layout_update() detach stripe before turning 1-stripe directory to
8047 * a plain directory.
8049 * \param[in] env execution environment
8050 * \param[in] dt target object
8051 * \param[in] mlc layout change data
8052 * \param[in] th transaction handle
8054 * \retval 0 on success
8055 * \retval negative if failed
8057 static int lod_dir_layout_detach(const struct lu_env *env,
8058 struct dt_object *dt,
8059 const struct md_layout_change *mlc,
8062 struct lod_thread_info *info = lod_env_info(env);
8063 struct lod_object *lo = lod_dt_obj(dt);
8064 struct dt_object *next = dt_object_child(dt);
8065 char *stripe_name = info->lti_key;
8066 struct dt_object *dto;
8072 if (!lo->ldo_dir_stripe_count) {
8073 /* plain directory delete .. */
8074 rc = lod_sub_delete(env, next,
8075 (const struct dt_key *)dotdot, th);
8079 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8080 dto = lo->ldo_stripe[i];
8084 rc = lod_sub_delete(env, dto,
8085 (const struct dt_key *)dotdot, th);
8089 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8090 PFID(lu_object_fid(&dto->do_lu)), i);
8092 rc = lod_sub_delete(env, next,
8093 (const struct dt_key *)stripe_name, th);
8097 rc = lod_sub_ref_del(env, next, th);
8102 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8103 dto = lo->ldo_stripe[i];
8105 dt_object_put(env, dto);
8107 OBD_FREE(lo->ldo_stripe,
8108 sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated);
8109 lo->ldo_stripe = NULL;
8110 lo->ldo_dir_stripes_allocated = 0;
8111 lo->ldo_dir_stripe_count = 0;
8116 static int lod_dir_layout_shrink(const struct lu_env *env,
8117 struct dt_object *dt,
8118 const struct md_layout_change *mlc,
8121 struct lod_thread_info *info = lod_env_info(env);
8122 struct lod_object *lo = lod_dt_obj(dt);
8123 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
8124 struct dt_object *next = dt_object_child(dt);
8125 struct lmv_user_md *lmu = mlc->mlc_buf.lb_buf;
8126 __u32 final_stripe_count;
8127 char *stripe_name = info->lti_key;
8128 struct dt_object *dto;
8129 struct lu_buf *lmv_buf = &info->lti_buf;
8130 struct lmv_mds_md_v1 *lmv = &info->lti_lmv.lmv_md_v1;
8132 int type = LU_SEQ_RANGE_ANY;
8138 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
8140 lmv_buf->lb_buf = lmv;
8141 lmv_buf->lb_len = sizeof(*lmv);
8142 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
8143 lmv->lmv_stripe_count = cpu_to_le32(final_stripe_count);
8144 lmv->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type) &
8145 cpu_to_le32(LMV_HASH_TYPE_MASK);
8146 lmv->lmv_layout_version =
8147 cpu_to_le32(lo->ldo_dir_layout_version + 1);
8149 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8150 dto = lo->ldo_stripe[i];
8154 if (i < final_stripe_count) {
8155 /* if only one stripe left, no need to update
8156 * LMV because this stripe will replace master
8157 * object and act as a plain directory.
8159 if (final_stripe_count == 1)
8163 rc = lod_fld_lookup(env, lod,
8164 lu_object_fid(&dto->do_lu),
8169 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
8170 rc = lod_sub_xattr_set(env, dto, lmv_buf,
8172 LU_XATTR_REPLACE, th);
8179 dt_write_lock(env, dto, DT_TGT_CHILD);
8180 rc = lod_sub_ref_del(env, dto, th);
8181 dt_write_unlock(env, dto);
8185 rc = lod_sub_destroy(env, dto, th);
8189 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
8190 PFID(lu_object_fid(&dto->do_lu)), i);
8192 rc = lod_sub_delete(env, next,
8193 (const struct dt_key *)stripe_name, th);
8197 rc = lod_sub_ref_del(env, next, th);
8202 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &mdtidx,
8207 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_V1);
8208 lmv->lmv_master_mdt_index = cpu_to_le32(mdtidx);
8209 rc = lod_sub_xattr_set(env, next, lmv_buf, XATTR_NAME_LMV,
8210 LU_XATTR_REPLACE, th);
8214 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
8215 dto = lo->ldo_stripe[i];
8217 dt_object_put(env, dto);
8219 lo->ldo_dir_stripe_count = final_stripe_count;
8224 static mlc_handler dir_mlc_declare_ops[MD_LAYOUT_MAX] = {
8225 [MD_LAYOUT_ATTACH] = lod_dir_declare_layout_attach,
8226 [MD_LAYOUT_DETACH] = lod_dir_declare_layout_detach,
8227 [MD_LAYOUT_SHRINK] = lod_dir_declare_layout_shrink,
8228 [MD_LAYOUT_SPLIT] = lod_dir_declare_layout_split,
8231 static mlc_handler dir_mlc_ops[MD_LAYOUT_MAX] = {
8232 [MD_LAYOUT_DETACH] = lod_dir_layout_detach,
8233 [MD_LAYOUT_SHRINK] = lod_dir_layout_shrink,
8236 static int lod_declare_layout_change(const struct lu_env *env,
8237 struct dt_object *dt, struct md_layout_change *mlc,
8240 struct lod_thread_info *info = lod_env_info(env);
8241 struct lod_object *lo = lod_dt_obj(dt);
8246 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
8247 LASSERT(dir_mlc_declare_ops[mlc->mlc_opc]);
8248 rc = dir_mlc_declare_ops[mlc->mlc_opc](env, dt, mlc, th);
8252 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
8253 dt_object_remote(dt_object_child(dt)))
8256 rc = lod_striping_load(env, lo);
8260 LASSERT(lo->ldo_comp_cnt > 0);
8262 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
8266 switch (lo->ldo_flr_state) {
8268 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
8272 rc = lod_declare_update_rdonly(env, lo, mlc, th);
8274 case LCM_FL_WRITE_PENDING:
8275 rc = lod_declare_update_write_pending(env, lo, mlc, th);
8277 case LCM_FL_SYNC_PENDING:
8278 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
8289 * Instantiate layout component objects which covers the intent write offset.
8291 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
8292 struct md_layout_change *mlc, struct thandle *th)
8294 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
8295 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
8296 struct lod_object *lo = lod_dt_obj(dt);
8301 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
8302 LASSERT(dir_mlc_ops[mlc->mlc_opc]);
8303 rc = dir_mlc_ops[mlc->mlc_opc](env, dt, mlc, th);
8307 rc = lod_striped_create(env, dt, attr, NULL, th);
8308 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
8309 layout_attr->la_layout_version |= lo->ldo_layout_gen;
8310 rc = lod_attr_set(env, dt, layout_attr, th);
8316 struct dt_object_operations lod_obj_ops = {
8317 .do_read_lock = lod_read_lock,
8318 .do_write_lock = lod_write_lock,
8319 .do_read_unlock = lod_read_unlock,
8320 .do_write_unlock = lod_write_unlock,
8321 .do_write_locked = lod_write_locked,
8322 .do_attr_get = lod_attr_get,
8323 .do_declare_attr_set = lod_declare_attr_set,
8324 .do_attr_set = lod_attr_set,
8325 .do_xattr_get = lod_xattr_get,
8326 .do_declare_xattr_set = lod_declare_xattr_set,
8327 .do_xattr_set = lod_xattr_set,
8328 .do_declare_xattr_del = lod_declare_xattr_del,
8329 .do_xattr_del = lod_xattr_del,
8330 .do_xattr_list = lod_xattr_list,
8331 .do_ah_init = lod_ah_init,
8332 .do_declare_create = lod_declare_create,
8333 .do_create = lod_create,
8334 .do_declare_destroy = lod_declare_destroy,
8335 .do_destroy = lod_destroy,
8336 .do_index_try = lod_index_try,
8337 .do_declare_ref_add = lod_declare_ref_add,
8338 .do_ref_add = lod_ref_add,
8339 .do_declare_ref_del = lod_declare_ref_del,
8340 .do_ref_del = lod_ref_del,
8341 .do_object_sync = lod_object_sync,
8342 .do_object_lock = lod_object_lock,
8343 .do_object_unlock = lod_object_unlock,
8344 .do_invalidate = lod_invalidate,
8345 .do_declare_layout_change = lod_declare_layout_change,
8346 .do_layout_change = lod_layout_change,
8350 * Implementation of dt_body_operations::dbo_read.
8352 * \see dt_body_operations::dbo_read() in the API description for details.
8354 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
8355 struct lu_buf *buf, loff_t *pos)
8357 struct dt_object *next = dt_object_child(dt);
8359 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
8360 S_ISLNK(dt->do_lu.lo_header->loh_attr));
8361 return next->do_body_ops->dbo_read(env, next, buf, pos);
8365 * Implementation of dt_body_operations::dbo_declare_write.
8367 * \see dt_body_operations::dbo_declare_write() in the API description
8370 static ssize_t lod_declare_write(const struct lu_env *env,
8371 struct dt_object *dt,
8372 const struct lu_buf *buf, loff_t pos,
8375 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
8379 * Implementation of dt_body_operations::dbo_write.
8381 * \see dt_body_operations::dbo_write() in the API description for details.
8383 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
8384 const struct lu_buf *buf, loff_t *pos,
8387 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
8388 S_ISLNK(dt->do_lu.lo_header->loh_attr));
8389 return lod_sub_write(env, dt_object_child(dt), buf, pos, th);
8392 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
8393 __u64 start, __u64 end, struct thandle *th)
8395 if (dt_object_remote(dt))
8398 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
8401 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
8402 __u64 start, __u64 end, struct thandle *th)
8404 if (dt_object_remote(dt))
8407 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
8408 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
8412 * different type of files use the same body_ops because object may be created
8413 * in OUT, where there is no chance to set correct body_ops for each type, so
8414 * body_ops themselves will check file type inside, see lod_read/write/punch for
8417 const struct dt_body_operations lod_body_ops = {
8418 .dbo_read = lod_read,
8419 .dbo_declare_write = lod_declare_write,
8420 .dbo_write = lod_write,
8421 .dbo_declare_punch = lod_declare_punch,
8422 .dbo_punch = lod_punch,
8426 * Implementation of lu_object_operations::loo_object_init.
8428 * The function determines the type and the index of the target device using
8429 * sequence of the object's FID. Then passes control down to the
8430 * corresponding device:
8431 * OSD for the local objects, OSP for remote
8433 * \see lu_object_operations::loo_object_init() in the API description
8436 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
8437 const struct lu_object_conf *conf)
8439 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
8440 struct lu_device *cdev = NULL;
8441 struct lu_object *cobj;
8442 struct lod_tgt_descs *ltd = NULL;
8443 struct lod_tgt_desc *tgt;
8445 int type = LU_SEQ_RANGE_ANY;
8449 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
8453 if (type == LU_SEQ_RANGE_MDT &&
8454 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
8455 cdev = &lod->lod_child->dd_lu_dev;
8456 } else if (type == LU_SEQ_RANGE_MDT) {
8457 ltd = &lod->lod_mdt_descs;
8459 } else if (type == LU_SEQ_RANGE_OST) {
8460 ltd = &lod->lod_ost_descs;
8467 if (ltd->ltd_tgts_size > idx &&
8468 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
8469 tgt = LTD_TGT(ltd, idx);
8471 LASSERT(tgt != NULL);
8472 LASSERT(tgt->ltd_tgt != NULL);
8474 cdev = &(tgt->ltd_tgt->dd_lu_dev);
8476 lod_putref(lod, ltd);
8479 if (unlikely(cdev == NULL))
8482 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
8483 if (unlikely(cobj == NULL))
8486 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
8488 lu_object_add(lo, cobj);
8495 * Alloc cached foreign LOV
8497 * \param[in] lo object
8498 * \param[in] size size of foreign LOV
8500 * \retval 0 on success
8501 * \retval negative if failed
8503 int lod_alloc_foreign_lov(struct lod_object *lo, size_t size)
8505 OBD_ALLOC_LARGE(lo->ldo_foreign_lov, size);
8506 if (lo->ldo_foreign_lov == NULL)
8508 lo->ldo_foreign_lov_size = size;
8509 lo->ldo_is_foreign = 1;
8515 * Free cached foreign LOV
8517 * \param[in] lo object
8519 void lod_free_foreign_lov(struct lod_object *lo)
8521 if (lo->ldo_foreign_lov != NULL)
8522 OBD_FREE_LARGE(lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
8523 lo->ldo_foreign_lov = NULL;
8524 lo->ldo_foreign_lov_size = 0;
8525 lo->ldo_is_foreign = 0;
8530 * Free cached foreign LMV
8532 * \param[in] lo object
8534 void lod_free_foreign_lmv(struct lod_object *lo)
8536 if (lo->ldo_foreign_lmv != NULL)
8537 OBD_FREE_LARGE(lo->ldo_foreign_lmv, lo->ldo_foreign_lmv_size);
8538 lo->ldo_foreign_lmv = NULL;
8539 lo->ldo_foreign_lmv_size = 0;
8540 lo->ldo_dir_is_foreign = 0;
8545 * Release resources associated with striping.
8547 * If the object is striped (regular or directory), then release
8548 * the stripe objects references and free the ldo_stripe array.
8550 * \param[in] env execution environment
8551 * \param[in] lo object
8553 void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo)
8555 struct lod_layout_component *lod_comp;
8558 if (unlikely(lo->ldo_is_foreign)) {
8559 lod_free_foreign_lov(lo);
8560 lo->ldo_comp_cached = 0;
8561 } else if (unlikely(lo->ldo_dir_is_foreign)) {
8562 lod_free_foreign_lmv(lo);
8563 lo->ldo_dir_stripe_loaded = 0;
8564 } else if (lo->ldo_stripe != NULL) {
8565 LASSERT(lo->ldo_comp_entries == NULL);
8566 LASSERT(lo->ldo_dir_stripes_allocated > 0);
8568 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
8569 if (lo->ldo_stripe[i])
8570 dt_object_put(env, lo->ldo_stripe[i]);
8573 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
8574 OBD_FREE(lo->ldo_stripe, j);
8575 lo->ldo_stripe = NULL;
8576 lo->ldo_dir_stripes_allocated = 0;
8577 lo->ldo_dir_stripe_loaded = 0;
8578 lo->ldo_dir_stripe_count = 0;
8579 } else if (lo->ldo_comp_entries != NULL) {
8580 for (i = 0; i < lo->ldo_comp_cnt; i++) {
8581 /* free lod_layout_component::llc_stripe array */
8582 lod_comp = &lo->ldo_comp_entries[i];
8584 if (lod_comp->llc_stripe == NULL)
8586 LASSERT(lod_comp->llc_stripes_allocated != 0);
8587 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
8588 if (lod_comp->llc_stripe[j] != NULL)
8590 &lod_comp->llc_stripe[j]->do_lu);
8592 OBD_FREE(lod_comp->llc_stripe,
8593 sizeof(struct dt_object *) *
8594 lod_comp->llc_stripes_allocated);
8595 lod_comp->llc_stripe = NULL;
8596 OBD_FREE(lod_comp->llc_ost_indices,
8598 lod_comp->llc_stripes_allocated);
8599 lod_comp->llc_ost_indices = NULL;
8600 lod_comp->llc_stripes_allocated = 0;
8602 lod_free_comp_entries(lo);
8603 lo->ldo_comp_cached = 0;
8607 void lod_striping_free(const struct lu_env *env, struct lod_object *lo)
8609 mutex_lock(&lo->ldo_layout_mutex);
8610 lod_striping_free_nolock(env, lo);
8611 mutex_unlock(&lo->ldo_layout_mutex);
8615 * Implementation of lu_object_operations::loo_object_free.
8617 * \see lu_object_operations::loo_object_free() in the API description
8620 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
8622 struct lod_object *lo = lu2lod_obj(o);
8624 /* release all underlying object pinned */
8625 lod_striping_free(env, lo);
8627 /* lo doesn't contain a lu_object_header, so we don't need call_rcu */
8628 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
8632 * Implementation of lu_object_operations::loo_object_release.
8634 * \see lu_object_operations::loo_object_release() in the API description
8637 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
8639 /* XXX: shouldn't we release everything here in case if object
8640 * creation failed before? */
8644 * Implementation of lu_object_operations::loo_object_print.
8646 * \see lu_object_operations::loo_object_print() in the API description
8649 static int lod_object_print(const struct lu_env *env, void *cookie,
8650 lu_printer_t p, const struct lu_object *l)
8652 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
8654 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
8657 struct lu_object_operations lod_lu_obj_ops = {
8658 .loo_object_init = lod_object_init,
8659 .loo_object_free = lod_object_free,
8660 .loo_object_release = lod_object_release,
8661 .loo_object_print = lod_object_print,