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 = snprintf(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 static inline void lod_adjust_stripe_info(struct lod_layout_component *comp,
1079 struct lov_desc *desc)
1081 if (comp->llc_pattern != LOV_PATTERN_MDT) {
1082 if (!comp->llc_stripe_count)
1083 comp->llc_stripe_count =
1084 desc->ld_default_stripe_count;
1086 if (comp->llc_stripe_size <= 0)
1087 comp->llc_stripe_size = desc->ld_default_stripe_size;
1090 int lod_obj_for_each_stripe(const struct lu_env *env, struct lod_object *lo,
1092 struct lod_obj_stripe_cb_data *data)
1094 struct lod_layout_component *lod_comp;
1098 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1099 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1100 lod_comp = &lo->ldo_comp_entries[i];
1102 if (lod_comp->llc_stripe == NULL)
1105 /* has stripe but not inited yet, this component has been
1106 * declared to be created, but hasn't created yet.
1108 if (!lod_comp_inited(lod_comp))
1111 if (data->locd_comp_skip_cb &&
1112 data->locd_comp_skip_cb(env, lo, i, data))
1115 if (data->locd_comp_cb) {
1116 rc = data->locd_comp_cb(env, lo, i, data);
1121 /* could used just to do sth about component, not each
1124 if (!data->locd_stripe_cb)
1127 LASSERT(lod_comp->llc_stripe_count > 0);
1128 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
1129 struct dt_object *dt = lod_comp->llc_stripe[j];
1133 rc = data->locd_stripe_cb(env, lo, dt, th, i, j, data);
1141 static bool lod_obj_attr_set_comp_skip_cb(const struct lu_env *env,
1142 struct lod_object *lo, int comp_idx,
1143 struct lod_obj_stripe_cb_data *data)
1145 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[comp_idx];
1146 bool skipped = false;
1148 if (!(data->locd_attr->la_valid & LA_LAYOUT_VERSION))
1151 switch (lo->ldo_flr_state) {
1152 case LCM_FL_WRITE_PENDING: {
1155 /* skip stale components */
1156 if (lod_comp->llc_flags & LCME_FL_STALE) {
1161 /* skip valid and overlapping components, therefore any
1162 * attempts to write overlapped components will never succeed
1163 * because client will get EINPROGRESS. */
1164 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1168 if (lo->ldo_comp_entries[i].llc_flags & LCME_FL_STALE)
1171 if (lu_extent_is_overlapped(&lod_comp->llc_extent,
1172 &lo->ldo_comp_entries[i].llc_extent)) {
1180 LASSERTF(0, "impossible: %d\n", lo->ldo_flr_state);
1181 case LCM_FL_SYNC_PENDING:
1185 CDEBUG(D_LAYOUT, DFID": %s to set component %x to version: %u\n",
1186 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
1187 skipped ? "skipped" : "chose", lod_comp->llc_id,
1188 data->locd_attr->la_layout_version);
1194 lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
1195 struct dt_object *dt, struct thandle *th,
1196 int comp_idx, int stripe_idx,
1197 struct lod_obj_stripe_cb_data *data)
1199 if (data->locd_declare)
1200 return lod_sub_declare_attr_set(env, dt, data->locd_attr, th);
1202 if (data->locd_attr->la_valid & LA_LAYOUT_VERSION) {
1203 CDEBUG(D_LAYOUT, DFID": set layout version: %u, comp_idx: %d\n",
1204 PFID(lu_object_fid(&dt->do_lu)),
1205 data->locd_attr->la_layout_version, comp_idx);
1208 return lod_sub_attr_set(env, dt, data->locd_attr, th);
1212 * Implementation of dt_object_operations::do_declare_attr_set.
1214 * If the object is striped, then apply the changes to all the stripes.
1216 * \see dt_object_operations::do_declare_attr_set() in the API description
1219 static int lod_declare_attr_set(const struct lu_env *env,
1220 struct dt_object *dt,
1221 const struct lu_attr *attr,
1224 struct dt_object *next = dt_object_child(dt);
1225 struct lod_object *lo = lod_dt_obj(dt);
1230 * declare setattr on the local object
1232 rc = lod_sub_declare_attr_set(env, next, attr, th);
1236 /* osp_declare_attr_set() ignores all attributes other than
1237 * UID, GID, PROJID, and size, and osp_attr_set() ignores all
1238 * but UID, GID and PROJID. Declaration of size attr setting
1239 * happens through lod_declare_init_size(), and not through
1240 * this function. Therefore we need not load striping unless
1241 * ownership is changing. This should save memory and (we hope)
1242 * speed up rename().
1244 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1245 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1248 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1251 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID | LA_MODE |
1252 LA_ATIME | LA_MTIME | LA_CTIME |
1257 * load striping information, notice we don't do this when object
1258 * is being initialized as we don't need this information till
1259 * few specific cases like destroy, chown
1261 rc = lod_striping_load(env, lo);
1265 if (!lod_obj_is_striped(dt))
1269 * if object is striped declare changes on the stripes
1271 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1272 LASSERT(lo->ldo_stripe);
1273 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1274 if (lo->ldo_stripe[i] == NULL)
1276 if (!dt_object_exists(lo->ldo_stripe[i]))
1278 rc = lod_sub_declare_attr_set(env, lo->ldo_stripe[i],
1284 struct lod_obj_stripe_cb_data data = { { 0 } };
1286 data.locd_attr = attr;
1287 data.locd_declare = true;
1288 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1289 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1295 if (!dt_object_exists(next) || dt_object_remote(next) ||
1296 !S_ISREG(attr->la_mode))
1299 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1300 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
1304 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) ||
1305 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1306 struct lod_thread_info *info = lod_env_info(env);
1307 struct lu_buf *buf = &info->lti_buf;
1309 buf->lb_buf = info->lti_ea_store;
1310 buf->lb_len = info->lti_ea_store_size;
1311 rc = lod_sub_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1312 LU_XATTR_REPLACE, th);
1319 * Implementation of dt_object_operations::do_attr_set.
1321 * If the object is striped, then apply the changes to all or subset of
1322 * the stripes depending on the object type and specific attributes.
1324 * \see dt_object_operations::do_attr_set() in the API description for details.
1326 static int lod_attr_set(const struct lu_env *env,
1327 struct dt_object *dt,
1328 const struct lu_attr *attr,
1331 struct dt_object *next = dt_object_child(dt);
1332 struct lod_object *lo = lod_dt_obj(dt);
1337 * apply changes to the local object
1339 rc = lod_sub_attr_set(env, next, attr, th);
1343 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1344 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1347 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1350 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE | LA_PROJID |
1351 LA_ATIME | LA_MTIME | LA_CTIME |
1356 /* FIXME: a tricky case in the code path of mdd_layout_change():
1357 * the in-memory striping information has been freed in lod_xattr_set()
1358 * due to layout change. It has to load stripe here again. It only
1359 * changes flags of layout so declare_attr_set() is still accurate */
1360 rc = lod_striping_load(env, lo);
1364 if (!lod_obj_is_striped(dt))
1368 * if object is striped, apply changes to all the stripes
1370 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1371 LASSERT(lo->ldo_stripe);
1372 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1373 if (unlikely(lo->ldo_stripe[i] == NULL))
1376 if ((dt_object_exists(lo->ldo_stripe[i]) == 0))
1379 rc = lod_sub_attr_set(env, lo->ldo_stripe[i], attr, th);
1384 struct lod_obj_stripe_cb_data data = { { 0 } };
1386 data.locd_attr = attr;
1387 data.locd_declare = false;
1388 data.locd_comp_skip_cb = lod_obj_attr_set_comp_skip_cb;
1389 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1390 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1396 if (!dt_object_exists(next) || dt_object_remote(next) ||
1397 !S_ISREG(attr->la_mode))
1400 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1401 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
1405 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE)) {
1406 struct lod_thread_info *info = lod_env_info(env);
1407 struct lu_buf *buf = &info->lti_buf;
1408 struct ost_id *oi = &info->lti_ostid;
1409 struct lu_fid *fid = &info->lti_fid;
1410 struct lov_mds_md_v1 *lmm;
1411 struct lov_ost_data_v1 *objs;
1414 rc = lod_get_lov_ea(env, lo);
1418 buf->lb_buf = info->lti_ea_store;
1419 buf->lb_len = info->lti_ea_store_size;
1420 lmm = info->lti_ea_store;
1421 magic = le32_to_cpu(lmm->lmm_magic);
1422 if (magic == LOV_MAGIC_COMP_V1) {
1423 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1424 struct lov_comp_md_entry_v1 *lcme =
1425 &lcm->lcm_entries[0];
1427 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
1428 magic = le32_to_cpu(lmm->lmm_magic);
1431 if (magic == LOV_MAGIC_V1)
1432 objs = &(lmm->lmm_objects[0]);
1434 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1435 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1436 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1438 fid_to_ostid(fid, oi);
1439 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1441 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1442 LU_XATTR_REPLACE, th);
1443 } else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1444 struct lod_thread_info *info = lod_env_info(env);
1445 struct lu_buf *buf = &info->lti_buf;
1446 struct lov_comp_md_v1 *lcm;
1447 struct lov_comp_md_entry_v1 *lcme;
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;
1456 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
1459 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1460 lcme = &lcm->lcm_entries[0];
1461 le64_add_cpu(&lcme->lcme_extent.e_start, 1);
1462 le64_add_cpu(&lcme->lcme_extent.e_end, -1);
1464 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1465 LU_XATTR_REPLACE, th);
1472 * Implementation of dt_object_operations::do_xattr_get.
1474 * If LOV EA is requested from the root object and it's not
1475 * found, then return default striping for the filesystem.
1477 * \see dt_object_operations::do_xattr_get() in the API description for details.
1479 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1480 struct lu_buf *buf, const char *name)
1482 struct lod_thread_info *info = lod_env_info(env);
1483 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1488 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1489 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1490 struct lmv_mds_md_v1 *lmv1;
1491 struct lmv_foreign_md *lfm;
1494 if (rc > (typeof(rc))sizeof(*lmv1))
1497 /* short (<= sizeof(struct lmv_mds_md_v1)) foreign LMV case */
1498 /* XXX empty foreign LMV is not allowed */
1499 if (rc <= offsetof(typeof(*lfm), lfm_value))
1500 RETURN(rc = rc > 0 ? -EINVAL : rc);
1502 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1503 CLASSERT(sizeof(*lmv1) <= sizeof(info->lti_key));
1505 /* lti_buf is large enough for *lmv1 or a short
1506 * (<= sizeof(struct lmv_mds_md_v1)) foreign LMV
1508 info->lti_buf.lb_buf = info->lti_key;
1509 info->lti_buf.lb_len = sizeof(*lmv1);
1510 rc = dt_xattr_get(env, dt_object_child(dt),
1511 &info->lti_buf, name);
1512 if (unlikely(rc <= offsetof(typeof(*lfm),
1514 RETURN(rc = rc > 0 ? -EINVAL : rc);
1516 lfm = info->lti_buf.lb_buf;
1517 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN)
1520 if (unlikely(rc != sizeof(*lmv1)))
1521 RETURN(rc = rc > 0 ? -EINVAL : rc);
1523 lmv1 = info->lti_buf.lb_buf;
1524 /* The on-disk LMV EA only contains header, but the
1525 * returned LMV EA size should contain the space for
1526 * the FIDs of all shards of the striped directory. */
1527 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1528 rc = lmv_mds_md_size(
1529 le32_to_cpu(lmv1->lmv_stripe_count),
1533 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN)
1536 if (rc != sizeof(*lmv1))
1537 RETURN(rc = rc > 0 ? -EINVAL : rc);
1539 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1543 RETURN(rc = rc1 != 0 ? rc1 : rc);
1546 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1550 * XXX: Only used by lfsck
1552 * lod returns default striping on the real root of the device
1553 * this is like the root stores default striping for the whole
1554 * filesystem. historically we've been using a different approach
1555 * and store it in the config.
1557 dt_root_get(env, dev->lod_child, &info->lti_fid);
1558 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1560 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1561 struct lov_user_md *lum = buf->lb_buf;
1562 struct lov_desc *desc = &dev->lod_desc;
1564 if (buf->lb_buf == NULL) {
1566 } else if (buf->lb_len >= sizeof(*lum)) {
1567 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1568 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1569 lmm_oi_set_id(&lum->lmm_oi, 0);
1570 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1571 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1572 lum->lmm_stripe_size = cpu_to_le32(
1573 desc->ld_default_stripe_size);
1574 lum->lmm_stripe_count = cpu_to_le16(
1575 desc->ld_default_stripe_count);
1576 lum->lmm_stripe_offset = cpu_to_le16(
1577 desc->ld_default_stripe_offset);
1590 * Checks that the magic of the stripe is sane.
1592 * \param[in] lod lod device
1593 * \param[in] lum a buffer storing LMV EA to verify
1595 * \retval 0 if the EA is sane
1596 * \retval negative otherwise
1598 static int lod_verify_md_striping(struct lod_device *lod,
1599 const struct lmv_user_md_v1 *lum)
1601 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1602 CERROR("%s: invalid lmv_user_md: magic = %x, "
1603 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1604 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1605 (int)le32_to_cpu(lum->lum_stripe_offset),
1606 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1614 * Initialize LMV EA for a slave.
1616 * Initialize slave's LMV EA from the master's LMV EA.
1618 * \param[in] master_lmv a buffer containing master's EA
1619 * \param[out] slave_lmv a buffer where slave's EA will be stored
1622 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1623 const struct lmv_mds_md_v1 *master_lmv)
1625 *slave_lmv = *master_lmv;
1626 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1632 * Generate LMV EA from the object passed as \a dt. The object must have
1633 * the stripes created and initialized.
1635 * \param[in] env execution environment
1636 * \param[in] dt object
1637 * \param[out] lmv_buf buffer storing generated LMV EA
1639 * \retval 0 on success
1640 * \retval negative if failed
1642 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1643 struct lu_buf *lmv_buf)
1645 struct lod_thread_info *info = lod_env_info(env);
1646 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1647 struct lod_object *lo = lod_dt_obj(dt);
1648 struct lmv_mds_md_v1 *lmm1;
1650 int type = LU_SEQ_RANGE_ANY;
1655 LASSERT(lo->ldo_dir_striped != 0);
1656 LASSERT(lo->ldo_dir_stripe_count > 0);
1657 stripe_count = lo->ldo_dir_stripe_count;
1658 /* Only store the LMV EA heahder on the disk. */
1659 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1660 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1664 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1667 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1668 memset(lmm1, 0, sizeof(*lmm1));
1669 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1670 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1671 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1672 if (lo->ldo_dir_hash_type & LMV_HASH_FLAG_MIGRATION) {
1673 lmm1->lmv_migrate_hash = cpu_to_le32(lo->ldo_dir_migrate_hash);
1674 lmm1->lmv_migrate_offset =
1675 cpu_to_le32(lo->ldo_dir_migrate_offset);
1677 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1682 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1683 lmv_buf->lb_buf = info->lti_ea_store;
1684 lmv_buf->lb_len = sizeof(*lmm1);
1690 * Create in-core represenation for a striped directory.
1692 * Parse the buffer containing LMV EA and instantiate LU objects
1693 * representing the stripe objects. The pointers to the objects are
1694 * stored in ldo_stripe field of \a lo. This function is used when
1695 * we need to access an already created object (i.e. load from a disk).
1697 * \param[in] env execution environment
1698 * \param[in] lo lod object
1699 * \param[in] buf buffer containing LMV EA
1701 * \retval 0 on success
1702 * \retval negative if failed
1704 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1705 const struct lu_buf *buf)
1707 struct lod_thread_info *info = lod_env_info(env);
1708 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1709 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1710 struct dt_object **stripe;
1711 union lmv_mds_md *lmm = buf->lb_buf;
1712 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1713 struct lu_fid *fid = &info->lti_fid;
1718 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1720 /* XXX may be useless as not called for foreign LMV ?? */
1721 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_FOREIGN)
1724 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1725 lo->ldo_dir_slave_stripe = 1;
1729 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1732 if (le32_to_cpu(lmv1->lmv_stripe_count) < 1)
1735 LASSERT(lo->ldo_stripe == NULL);
1736 OBD_ALLOC(stripe, sizeof(stripe[0]) *
1737 (le32_to_cpu(lmv1->lmv_stripe_count)));
1741 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1742 struct dt_device *tgt_dt;
1743 struct dt_object *dto;
1744 int type = LU_SEQ_RANGE_ANY;
1747 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1748 if (!fid_is_sane(fid)) {
1753 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1757 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1758 tgt_dt = lod->lod_child;
1760 struct lod_tgt_desc *tgt;
1762 tgt = LTD_TGT(ltd, idx);
1764 GOTO(out, rc = -ESTALE);
1765 tgt_dt = tgt->ltd_tgt;
1768 dto = dt_locate_at(env, tgt_dt, fid,
1769 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1772 GOTO(out, rc = PTR_ERR(dto));
1777 lo->ldo_stripe = stripe;
1778 lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1779 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1781 lod_striping_free_nolock(env, lo);
1787 * Declare create a striped directory.
1789 * Declare creating a striped directory with a given stripe pattern on the
1790 * specified MDTs. A striped directory is represented as a regular directory
1791 * - an index listing all the stripes. The stripes point back to the master
1792 * object with ".." and LinkEA. The master object gets LMV EA which
1793 * identifies it as a striped directory. The function allocates FIDs
1796 * \param[in] env execution environment
1797 * \param[in] dt object
1798 * \param[in] attr attributes to initialize the objects with
1799 * \param[in] dof type of objects to be created
1800 * \param[in] th transaction handle
1802 * \retval 0 on success
1803 * \retval negative if failed
1805 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1806 struct dt_object *dt,
1807 struct lu_attr *attr,
1808 struct dt_object_format *dof,
1811 struct lod_thread_info *info = lod_env_info(env);
1812 struct lu_buf lmv_buf;
1813 struct lu_buf slave_lmv_buf;
1814 struct lmv_mds_md_v1 *lmm;
1815 struct lmv_mds_md_v1 *slave_lmm = NULL;
1816 struct dt_insert_rec *rec = &info->lti_dt_rec;
1817 struct lod_object *lo = lod_dt_obj(dt);
1822 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1825 lmm = lmv_buf.lb_buf;
1827 OBD_ALLOC_PTR(slave_lmm);
1828 if (slave_lmm == NULL)
1829 GOTO(out, rc = -ENOMEM);
1831 lod_prep_slave_lmv_md(slave_lmm, lmm);
1832 slave_lmv_buf.lb_buf = slave_lmm;
1833 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1835 if (!dt_try_as_dir(env, dt_object_child(dt)))
1836 GOTO(out, rc = -EINVAL);
1838 rec->rec_type = S_IFDIR;
1839 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1840 struct dt_object *dto = lo->ldo_stripe[i];
1841 char *stripe_name = info->lti_key;
1842 struct lu_name *sname;
1843 struct linkea_data ldata = { NULL };
1844 struct lu_buf linkea_buf;
1846 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
1850 rc = lod_sub_declare_create(env, dto, attr, NULL, dof, th);
1854 if (!dt_try_as_dir(env, dto))
1855 GOTO(out, rc = -EINVAL);
1857 rc = lod_sub_declare_ref_add(env, dto, th);
1861 rec->rec_fid = lu_object_fid(&dto->do_lu);
1862 rc = lod_sub_declare_insert(env, dto,
1863 (const struct dt_rec *)rec,
1864 (const struct dt_key *)dot, th);
1868 /* master stripe FID will be put to .. */
1869 rec->rec_fid = lu_object_fid(&dt->do_lu);
1870 rc = lod_sub_declare_insert(env, dto,
1871 (const struct dt_rec *)rec,
1872 (const struct dt_key *)dotdot, th);
1876 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1877 cfs_fail_val != i) {
1878 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1880 slave_lmm->lmv_master_mdt_index =
1883 slave_lmm->lmv_master_mdt_index =
1885 rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
1886 XATTR_NAME_LMV, 0, th);
1891 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1893 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1894 PFID(lu_object_fid(&dto->do_lu)), i + 1);
1896 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1897 PFID(lu_object_fid(&dto->do_lu)), i);
1899 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
1900 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1901 sname, lu_object_fid(&dt->do_lu));
1905 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1906 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1907 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
1908 XATTR_NAME_LINK, 0, th);
1912 rec->rec_fid = lu_object_fid(&dto->do_lu);
1913 rc = lod_sub_declare_insert(env, dt_object_child(dt),
1914 (const struct dt_rec *)rec,
1915 (const struct dt_key *)stripe_name,
1920 rc = lod_sub_declare_ref_add(env, dt_object_child(dt), th);
1925 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
1926 &lmv_buf, XATTR_NAME_LMV, 0, th);
1930 if (slave_lmm != NULL)
1931 OBD_FREE_PTR(slave_lmm);
1936 static int lod_prep_md_striped_create(const struct lu_env *env,
1937 struct dt_object *dt,
1938 struct lu_attr *attr,
1939 const struct lmv_user_md_v1 *lum,
1940 struct dt_object_format *dof,
1943 struct lod_thread_info *info = lod_env_info(env);
1944 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1945 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1946 struct lod_object *lo = lod_dt_obj(dt);
1947 struct dt_object **stripe;
1954 bool is_specific = false;
1957 /* The lum has been verifed in lod_verify_md_striping */
1958 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1959 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
1961 stripe_count = lo->ldo_dir_stripe_count;
1963 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
1964 if (idx_array == NULL)
1967 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_count);
1969 GOTO(out_free, rc = -ENOMEM);
1971 /* Start index must be the master MDT */
1972 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1973 idx_array[0] = master_index;
1974 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
1976 for (i = 1; i < stripe_count; i++)
1977 idx_array[i] = le32_to_cpu(lum->lum_objects[i].lum_mds);
1980 for (i = 0; i < stripe_count; i++) {
1981 struct lod_tgt_desc *tgt = NULL;
1982 struct dt_object *dto;
1983 struct lu_fid fid = { 0 };
1985 struct lu_object_conf conf = { 0 };
1986 struct dt_device *tgt_dt = NULL;
1988 /* Try to find next avaible target */
1990 for (j = 0; j < lod->lod_remote_mdt_count;
1991 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
1992 bool already_allocated = false;
1995 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
1996 idx, lod->lod_remote_mdt_count + 1, i);
1998 if (likely(!is_specific &&
1999 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
2000 /* check whether the idx already exists
2001 * in current allocated array */
2002 for (k = 0; k < i; k++) {
2003 if (idx_array[k] == idx) {
2004 already_allocated = true;
2009 if (already_allocated)
2013 /* Sigh, this index is not in the bitmap, let's check
2014 * next available target */
2015 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) &&
2016 idx != master_index)
2019 if (idx == master_index) {
2020 /* Allocate the FID locally */
2021 rc = obd_fid_alloc(env, lod->lod_child_exp,
2025 tgt_dt = lod->lod_child;
2029 /* check the status of the OSP */
2030 tgt = LTD_TGT(ltd, idx);
2034 tgt_dt = tgt->ltd_tgt;
2035 rc = dt_statfs(env, tgt_dt, &info->lti_osfs);
2037 /* this OSP doesn't feel well */
2042 rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
2051 /* Can not allocate more stripes */
2052 if (j == lod->lod_remote_mdt_count) {
2053 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
2054 lod2obd(lod)->obd_name, stripe_count, i);
2058 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
2059 idx, i, PFID(&fid));
2061 /* Set the start index for next stripe allocation */
2062 if (!is_specific && i < stripe_count - 1) {
2064 * for large dir test, put all other slaves on one
2065 * remote MDT, otherwise we may save too many local
2066 * slave locks which will exceed RS_MAX_LOCKS.
2068 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)))
2070 idx_array[i + 1] = (idx + 1) %
2071 (lod->lod_remote_mdt_count + 1);
2073 /* tgt_dt and fid must be ready after search avaible OSP
2074 * in the above loop */
2075 LASSERT(tgt_dt != NULL);
2076 LASSERT(fid_is_sane(&fid));
2078 /* fail a remote stripe FID allocation */
2079 if (i && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_FID))
2082 conf.loc_flags = LOC_F_NEW;
2083 dto = dt_locate_at(env, tgt_dt, &fid,
2084 dt->do_lu.lo_dev->ld_site->ls_top_dev,
2087 GOTO(out_put, rc = PTR_ERR(dto));
2091 lo->ldo_dir_striped = 1;
2092 lo->ldo_stripe = stripe;
2093 lo->ldo_dir_stripe_count = i;
2094 lo->ldo_dir_stripes_allocated = stripe_count;
2096 lo->ldo_dir_stripe_loaded = 1;
2098 if (lo->ldo_dir_stripe_count == 0)
2099 GOTO(out_put, rc = -ENOSPC);
2101 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2107 for (i = 0; i < stripe_count; i++)
2108 if (stripe[i] != NULL)
2109 dt_object_put(env, stripe[i]);
2110 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_count);
2111 lo->ldo_dir_stripe_count = 0;
2112 lo->ldo_dir_stripes_allocated = 0;
2113 lo->ldo_stripe = NULL;
2117 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
2124 * Alloc cached foreign LMV
2126 * \param[in] lo object
2127 * \param[in] size size of foreign LMV
2129 * \retval 0 on success
2130 * \retval negative if failed
2132 int lod_alloc_foreign_lmv(struct lod_object *lo, size_t size)
2134 OBD_ALLOC_LARGE(lo->ldo_foreign_lmv, size);
2135 if (lo->ldo_foreign_lmv == NULL)
2137 lo->ldo_foreign_lmv_size = size;
2138 lo->ldo_dir_is_foreign = 1;
2144 * Declare create striped md object.
2146 * The function declares intention to create a striped directory. This is a
2147 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2148 * is to verify pattern \a lum_buf is good. Check that function for the details.
2150 * \param[in] env execution environment
2151 * \param[in] dt object
2152 * \param[in] attr attributes to initialize the objects with
2153 * \param[in] lum_buf a pattern specifying the number of stripes and
2155 * \param[in] dof type of objects to be created
2156 * \param[in] th transaction handle
2158 * \retval 0 on success
2159 * \retval negative if failed
2162 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2163 struct dt_object *dt,
2164 struct lu_attr *attr,
2165 const struct lu_buf *lum_buf,
2166 struct dt_object_format *dof,
2169 struct lod_object *lo = lod_dt_obj(dt);
2170 struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
2174 LASSERT(lum != NULL);
2176 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
2177 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
2178 (int)le32_to_cpu(lum->lum_stripe_offset));
2180 if (lo->ldo_dir_stripe_count == 0) {
2181 if (lo->ldo_dir_is_foreign) {
2182 rc = lod_alloc_foreign_lmv(lo, lum_buf->lb_len);
2185 memcpy(lo->ldo_foreign_lmv, lum, lum_buf->lb_len);
2186 lo->ldo_dir_stripe_loaded = 1;
2191 /* prepare dir striped objects */
2192 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2194 /* failed to create striping, let's reset
2195 * config so that others don't get confused */
2196 lod_striping_free(env, lo);
2204 * Append source stripes after target stripes for migrating directory. NB, we
2205 * only need to declare this, the append is done inside lod_xattr_set_lmv().
2207 * \param[in] env execution environment
2208 * \param[in] dt target object
2209 * \param[in] buf LMV buf which contains source stripe fids
2210 * \param[in] th transaction handle
2212 * \retval 0 on success
2213 * \retval negative if failed
2215 static int lod_dir_declare_layout_add(const struct lu_env *env,
2216 struct dt_object *dt,
2217 const struct lu_buf *buf,
2220 struct lod_thread_info *info = lod_env_info(env);
2221 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
2222 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
2223 struct lod_object *lo = lod_dt_obj(dt);
2224 struct dt_object *next = dt_object_child(dt);
2225 struct dt_object_format *dof = &info->lti_format;
2226 struct lmv_mds_md_v1 *lmv = buf->lb_buf;
2227 struct dt_object **stripe;
2228 __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
2229 struct lu_fid *fid = &info->lti_fid;
2230 struct lod_tgt_desc *tgt;
2231 struct dt_object *dto;
2232 struct dt_device *tgt_dt;
2233 int type = LU_SEQ_RANGE_ANY;
2234 struct dt_insert_rec *rec = &info->lti_dt_rec;
2235 char *stripe_name = info->lti_key;
2236 struct lu_name *sname;
2237 struct linkea_data ldata = { NULL };
2238 struct lu_buf linkea_buf;
2245 if (le32_to_cpu(lmv->lmv_magic) != LMV_MAGIC_V1)
2248 if (stripe_count == 0)
2251 dof->dof_type = DFT_DIR;
2254 sizeof(*stripe) * (lo->ldo_dir_stripe_count + stripe_count));
2258 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
2259 stripe[i] = lo->ldo_stripe[i];
2261 for (i = 0; i < stripe_count; i++) {
2263 &lmv->lmv_stripe_fids[i]);
2264 if (!fid_is_sane(fid))
2267 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
2271 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
2272 tgt_dt = lod->lod_child;
2274 tgt = LTD_TGT(ltd, idx);
2276 GOTO(out, rc = -ESTALE);
2277 tgt_dt = tgt->ltd_tgt;
2280 dto = dt_locate_at(env, tgt_dt, fid,
2281 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
2284 GOTO(out, rc = PTR_ERR(dto));
2286 stripe[i + lo->ldo_dir_stripe_count] = dto;
2288 if (!dt_try_as_dir(env, dto))
2289 GOTO(out, rc = -ENOTDIR);
2291 rc = lod_sub_declare_ref_add(env, dto, th);
2295 rc = lod_sub_declare_insert(env, dto,
2296 (const struct dt_rec *)rec,
2297 (const struct dt_key *)dot, th);
2301 rc = lod_sub_declare_insert(env, dto,
2302 (const struct dt_rec *)rec,
2303 (const struct dt_key *)dotdot, th);
2307 rc = lod_sub_declare_xattr_set(env, dto, buf,
2308 XATTR_NAME_LMV, 0, th);
2312 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
2313 PFID(lu_object_fid(&dto->do_lu)),
2314 i + lo->ldo_dir_stripe_count);
2316 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
2317 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
2318 sname, lu_object_fid(&dt->do_lu));
2322 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
2323 linkea_buf.lb_len = ldata.ld_leh->leh_len;
2324 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
2325 XATTR_NAME_LINK, 0, th);
2329 rc = lod_sub_declare_insert(env, next,
2330 (const struct dt_rec *)rec,
2331 (const struct dt_key *)stripe_name,
2336 rc = lod_sub_declare_ref_add(env, next, th);
2342 OBD_FREE(lo->ldo_stripe,
2343 sizeof(*stripe) * lo->ldo_dir_stripes_allocated);
2344 lo->ldo_stripe = stripe;
2345 lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count;
2346 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type);
2347 lo->ldo_dir_stripe_count += stripe_count;
2348 lo->ldo_dir_stripes_allocated += stripe_count;
2349 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_MIGRATION;
2353 i = lo->ldo_dir_stripe_count;
2354 while (i < lo->ldo_dir_stripe_count + stripe_count && stripe[i])
2355 dt_object_put(env, stripe[i++]);
2358 sizeof(*stripe) * (stripe_count + lo->ldo_dir_stripe_count));
2362 static int lod_dir_declare_layout_delete(const struct lu_env *env,
2363 struct dt_object *dt,
2364 const struct lu_buf *buf,
2367 struct lod_thread_info *info = lod_env_info(env);
2368 struct lod_object *lo = lod_dt_obj(dt);
2369 struct dt_object *next = dt_object_child(dt);
2370 struct lmv_user_md *lmu = buf->lb_buf;
2371 __u32 final_stripe_count;
2372 char *stripe_name = info->lti_key;
2373 struct dt_object *dto;
2380 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
2381 if (final_stripe_count >= lo->ldo_dir_stripe_count)
2384 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
2385 dto = lo->ldo_stripe[i];
2389 if (!dt_try_as_dir(env, dto))
2392 rc = lod_sub_declare_delete(env, dto,
2393 (const struct dt_key *)dot, th);
2397 rc = lod_sub_declare_ref_del(env, dto, th);
2401 rc = lod_sub_declare_delete(env, dto,
2402 (const struct dt_key *)dotdot, th);
2406 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2407 PFID(lu_object_fid(&dto->do_lu)), i);
2409 rc = lod_sub_declare_delete(env, next,
2410 (const struct dt_key *)stripe_name, th);
2414 rc = lod_sub_declare_ref_del(env, next, th);
2423 * delete stripes from dir master object, the lum_stripe_count in argument is
2424 * the final stripe count, the stripes after that will be deleted, NB, they
2425 * are not destroyed, but deleted from it's parent namespace, this function
2426 * will be called in two places:
2427 * 1. mdd_migrate_create() delete stripes from source, and append them to
2429 * 2. mdd_dir_layout_shrink() delete stripes from source, and destroy them.
2431 static int lod_dir_layout_delete(const struct lu_env *env,
2432 struct dt_object *dt,
2433 const struct lu_buf *buf,
2436 struct lod_thread_info *info = lod_env_info(env);
2437 struct lod_object *lo = lod_dt_obj(dt);
2438 struct dt_object *next = dt_object_child(dt);
2439 struct lmv_user_md *lmu = buf->lb_buf;
2440 __u32 final_stripe_count;
2441 char *stripe_name = info->lti_key;
2442 struct dt_object *dto;
2451 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
2452 if (final_stripe_count >= lo->ldo_dir_stripe_count)
2455 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
2456 dto = lo->ldo_stripe[i];
2460 rc = lod_sub_delete(env, dto,
2461 (const struct dt_key *)dotdot, th);
2465 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2466 PFID(lu_object_fid(&dto->do_lu)), i);
2468 rc = lod_sub_delete(env, next,
2469 (const struct dt_key *)stripe_name, th);
2473 rc = lod_sub_ref_del(env, next, th);
2478 lod_striping_free(env, lod_dt_obj(dt));
2484 * Implementation of dt_object_operations::do_declare_xattr_set.
2486 * Used with regular (non-striped) objects. Basically it
2487 * initializes the striping information and applies the
2488 * change to all the stripes.
2490 * \see dt_object_operations::do_declare_xattr_set() in the API description
2493 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2494 struct dt_object *dt,
2495 const struct lu_buf *buf,
2496 const char *name, int fl,
2499 struct dt_object *next = dt_object_child(dt);
2500 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2501 struct lod_object *lo = lod_dt_obj(dt);
2506 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2507 struct lmv_user_md_v1 *lum;
2509 LASSERT(buf != NULL && buf->lb_buf != NULL);
2511 rc = lod_verify_md_striping(d, lum);
2514 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2515 rc = lod_verify_striping(d, lo, buf, false);
2520 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2524 /* Note: Do not set LinkEA on sub-stripes, otherwise
2525 * it will confuse the fid2path process(see mdt_path_current()).
2526 * The linkEA between master and sub-stripes is set in
2527 * lod_xattr_set_lmv(). */
2528 if (strcmp(name, XATTR_NAME_LINK) == 0)
2531 /* set xattr to each stripes, if needed */
2532 rc = lod_striping_load(env, lo);
2536 if (lo->ldo_dir_stripe_count == 0)
2539 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2540 if (!lo->ldo_stripe[i])
2543 if (!dt_object_exists(lo->ldo_stripe[i]))
2546 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2556 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2557 struct lod_object *lo,
2558 struct dt_object *dt, struct thandle *th,
2559 int comp_idx, int stripe_idx,
2560 struct lod_obj_stripe_cb_data *data)
2562 struct lod_thread_info *info = lod_env_info(env);
2563 struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
2564 struct filter_fid *ff = &info->lti_ff;
2565 struct lu_buf *buf = &info->lti_buf;
2569 buf->lb_len = sizeof(*ff);
2570 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2578 * locd_buf is set if it's called by dir migration, which doesn't check
2581 if (data->locd_buf) {
2582 memset(ff, 0, sizeof(*ff));
2583 ff->ff_parent = *(struct lu_fid *)data->locd_buf->lb_buf;
2585 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
2587 if (lu_fid_eq(lod_object_fid(lo), &ff->ff_parent) &&
2588 ff->ff_layout.ol_comp_id == comp->llc_id)
2591 memset(ff, 0, sizeof(*ff));
2592 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2595 /* rewrite filter_fid */
2596 ff->ff_parent.f_ver = stripe_idx;
2597 ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
2598 ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
2599 ff->ff_layout.ol_comp_id = comp->llc_id;
2600 ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
2601 ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
2602 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
2604 if (data->locd_declare)
2605 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2606 LU_XATTR_REPLACE, th);
2608 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2609 LU_XATTR_REPLACE, th);
2615 * Reset parent FID on OST object
2617 * Replace parent FID with @dt object FID, which is only called during migration
2618 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2619 * the FID is changed.
2621 * \param[in] env execution environment
2622 * \param[in] dt dt_object whose stripes's parent FID will be reset
2623 * \parem[in] th thandle
2624 * \param[in] declare if it is declare
2626 * \retval 0 if reset succeeds
2627 * \retval negative errno if reset fails
2629 static int lod_replace_parent_fid(const struct lu_env *env,
2630 struct dt_object *dt,
2631 const struct lu_buf *buf,
2632 struct thandle *th, bool declare)
2634 struct lod_object *lo = lod_dt_obj(dt);
2635 struct lod_thread_info *info = lod_env_info(env);
2636 struct filter_fid *ff;
2637 struct lod_obj_stripe_cb_data data = { { 0 } };
2641 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2643 /* set xattr to each stripes, if needed */
2644 rc = lod_striping_load(env, lo);
2648 if (!lod_obj_is_striped(dt))
2651 if (info->lti_ea_store_size < sizeof(*ff)) {
2652 rc = lod_ea_store_resize(info, sizeof(*ff));
2657 data.locd_declare = declare;
2658 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2659 data.locd_buf = buf;
2660 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2665 inline __u16 lod_comp_entry_stripe_count(struct lod_object *lo,
2666 struct lod_layout_component *entry,
2669 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2673 else if (lod_comp_inited(entry))
2674 return entry->llc_stripe_count;
2675 else if ((__u16)-1 == entry->llc_stripe_count)
2676 return lod->lod_desc.ld_tgt_count;
2678 return lod_get_stripe_count(lod, lo,
2679 entry->llc_stripe_count, false);
2682 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2684 int magic, size = 0, i;
2685 struct lod_layout_component *comp_entries;
2687 bool is_composite, is_foreign = false;
2690 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2691 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2693 lo->ldo_def_striping->lds_def_striping_is_composite;
2695 comp_cnt = lo->ldo_comp_cnt;
2696 comp_entries = lo->ldo_comp_entries;
2697 is_composite = lo->ldo_is_composite;
2698 is_foreign = lo->ldo_is_foreign;
2702 return lo->ldo_foreign_lov_size;
2704 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2706 size = sizeof(struct lov_comp_md_v1) +
2707 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2708 LASSERT(size % sizeof(__u64) == 0);
2711 for (i = 0; i < comp_cnt; i++) {
2714 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2715 stripe_count = lod_comp_entry_stripe_count(lo, &comp_entries[i],
2717 if (!is_dir && is_composite)
2718 lod_comp_shrink_stripe_count(&comp_entries[i],
2721 size += lov_user_md_size(stripe_count, magic);
2722 LASSERT(size % sizeof(__u64) == 0);
2728 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2729 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2732 * \param[in] env execution environment
2733 * \param[in] dt dt_object to add components on
2734 * \param[in] buf buffer contains components to be added
2735 * \parem[in] th thandle
2737 * \retval 0 on success
2738 * \retval negative errno on failure
2740 static int lod_declare_layout_add(const struct lu_env *env,
2741 struct dt_object *dt,
2742 const struct lu_buf *buf,
2745 struct lod_thread_info *info = lod_env_info(env);
2746 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2747 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2748 struct dt_object *next = dt_object_child(dt);
2749 struct lov_desc *desc = &d->lod_desc;
2750 struct lod_object *lo = lod_dt_obj(dt);
2751 struct lov_user_md_v3 *v3;
2752 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2754 int i, rc, array_cnt, old_array_cnt;
2757 LASSERT(lo->ldo_is_composite);
2759 if (lo->ldo_flr_state != LCM_FL_NONE)
2762 rc = lod_verify_striping(d, lo, buf, false);
2766 magic = comp_v1->lcm_magic;
2767 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2768 lustre_swab_lov_comp_md_v1(comp_v1);
2769 magic = comp_v1->lcm_magic;
2772 if (magic != LOV_USER_MAGIC_COMP_V1)
2775 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2776 OBD_ALLOC(comp_array, sizeof(*comp_array) * array_cnt);
2777 if (comp_array == NULL)
2780 memcpy(comp_array, lo->ldo_comp_entries,
2781 sizeof(*comp_array) * lo->ldo_comp_cnt);
2783 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2784 struct lov_user_md_v1 *v1;
2785 struct lu_extent *ext;
2787 v1 = (struct lov_user_md *)((char *)comp_v1 +
2788 comp_v1->lcm_entries[i].lcme_offset);
2789 ext = &comp_v1->lcm_entries[i].lcme_extent;
2791 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2792 lod_comp->llc_extent.e_start = ext->e_start;
2793 lod_comp->llc_extent.e_end = ext->e_end;
2794 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2795 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2797 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2798 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2799 lod_adjust_stripe_info(lod_comp, desc);
2801 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2802 v3 = (struct lov_user_md_v3 *) v1;
2803 if (v3->lmm_pool_name[0] != '\0') {
2804 rc = lod_set_pool(&lod_comp->llc_pool,
2812 old_array = lo->ldo_comp_entries;
2813 old_array_cnt = lo->ldo_comp_cnt;
2815 lo->ldo_comp_entries = comp_array;
2816 lo->ldo_comp_cnt = array_cnt;
2818 /* No need to increase layout generation here, it will be increased
2819 * later when generating component ID for the new components */
2821 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2822 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2823 XATTR_NAME_LOV, 0, th);
2825 lo->ldo_comp_entries = old_array;
2826 lo->ldo_comp_cnt = old_array_cnt;
2830 OBD_FREE(old_array, sizeof(*lod_comp) * old_array_cnt);
2832 LASSERT(lo->ldo_mirror_count == 1);
2833 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2838 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2839 lod_comp = &comp_array[i];
2840 if (lod_comp->llc_pool != NULL) {
2841 OBD_FREE(lod_comp->llc_pool,
2842 strlen(lod_comp->llc_pool) + 1);
2843 lod_comp->llc_pool = NULL;
2846 OBD_FREE(comp_array, sizeof(*comp_array) * array_cnt);
2851 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2852 * the '$field' can only be 'flags' now. The xattr value is binary
2853 * lov_comp_md_v1 which contains the component ID(s) and the value of
2854 * the field to be modified.
2856 * \param[in] env execution environment
2857 * \param[in] dt dt_object to be modified
2858 * \param[in] op operation string, like "set.flags"
2859 * \param[in] buf buffer contains components to be set
2860 * \parem[in] th thandle
2862 * \retval 0 on success
2863 * \retval negative errno on failure
2865 static int lod_declare_layout_set(const struct lu_env *env,
2866 struct dt_object *dt,
2867 char *op, const struct lu_buf *buf,
2870 struct lod_layout_component *lod_comp;
2871 struct lod_thread_info *info = lod_env_info(env);
2872 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2873 struct lod_object *lo = lod_dt_obj(dt);
2874 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2877 bool changed = false;
2880 if (strcmp(op, "set.flags") != 0) {
2881 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2882 lod2obd(d)->obd_name, op);
2886 magic = comp_v1->lcm_magic;
2887 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2888 lustre_swab_lov_comp_md_v1(comp_v1);
2889 magic = comp_v1->lcm_magic;
2892 if (magic != LOV_USER_MAGIC_COMP_V1)
2895 if (comp_v1->lcm_entry_count == 0) {
2896 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2897 lod2obd(d)->obd_name);
2901 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2902 __u32 id = comp_v1->lcm_entries[i].lcme_id;
2903 __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
2904 __u32 mirror_flag = flags & LCME_MIRROR_FLAGS;
2905 bool neg = flags & LCME_FL_NEG;
2907 if (flags & LCME_FL_INIT) {
2909 lod_striping_free(env, lo);
2913 flags &= ~(LCME_MIRROR_FLAGS | LCME_FL_NEG);
2914 for (j = 0; j < lo->ldo_comp_cnt; j++) {
2915 lod_comp = &lo->ldo_comp_entries[j];
2917 /* lfs only put one flag in each entry */
2918 if ((flags && id != lod_comp->llc_id) ||
2919 (mirror_flag && mirror_id_of(id) !=
2920 mirror_id_of(lod_comp->llc_id)))
2925 lod_comp->llc_flags &= ~flags;
2927 lod_comp->llc_flags &= ~mirror_flag;
2930 lod_comp->llc_flags |= flags;
2932 lod_comp->llc_flags |= mirror_flag;
2933 if (mirror_flag & LCME_FL_NOSYNC)
2934 lod_comp->llc_timestamp =
2935 ktime_get_real_seconds();
2943 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
2944 lod2obd(d)->obd_name);
2948 lod_obj_inc_layout_gen(lo);
2950 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2951 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
2952 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
2957 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
2958 * and the xattr value is a unique component ID or a special lcme_id.
2960 * \param[in] env execution environment
2961 * \param[in] dt dt_object to be operated on
2962 * \param[in] buf buffer contains component ID or lcme_id
2963 * \parem[in] th thandle
2965 * \retval 0 on success
2966 * \retval negative errno on failure
2968 static int lod_declare_layout_del(const struct lu_env *env,
2969 struct dt_object *dt,
2970 const struct lu_buf *buf,
2973 struct lod_thread_info *info = lod_env_info(env);
2974 struct dt_object *next = dt_object_child(dt);
2975 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2976 struct lod_object *lo = lod_dt_obj(dt);
2977 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2978 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2979 __u32 magic, id, flags, neg_flags = 0;
2983 LASSERT(lo->ldo_is_composite);
2985 if (lo->ldo_flr_state != LCM_FL_NONE)
2988 magic = comp_v1->lcm_magic;
2989 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2990 lustre_swab_lov_comp_md_v1(comp_v1);
2991 magic = comp_v1->lcm_magic;
2994 if (magic != LOV_USER_MAGIC_COMP_V1)
2997 id = comp_v1->lcm_entries[0].lcme_id;
2998 flags = comp_v1->lcm_entries[0].lcme_flags;
3000 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
3001 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
3002 lod2obd(d)->obd_name, id, flags);
3006 if (id != LCME_ID_INVAL && flags != 0) {
3007 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
3008 lod2obd(d)->obd_name);
3012 if (id == LCME_ID_INVAL && !flags) {
3013 CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
3014 lod2obd(d)->obd_name);
3018 if (flags & LCME_FL_NEG) {
3019 neg_flags = flags & ~LCME_FL_NEG;
3023 left = lo->ldo_comp_cnt;
3027 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3028 struct lod_layout_component *lod_comp;
3030 lod_comp = &lo->ldo_comp_entries[i];
3032 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
3034 else if (flags && !(flags & lod_comp->llc_flags))
3036 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
3039 if (left != (i + 1)) {
3040 CDEBUG(D_LAYOUT, "%s: this deletion will create "
3041 "a hole.\n", lod2obd(d)->obd_name);
3046 /* Mark the component as deleted */
3047 lod_comp->llc_id = LCME_ID_INVAL;
3049 /* Not instantiated component */
3050 if (lod_comp->llc_stripe == NULL)
3053 LASSERT(lod_comp->llc_stripe_count > 0);
3054 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
3055 struct dt_object *obj = lod_comp->llc_stripe[j];
3059 rc = lod_sub_declare_destroy(env, obj, th);
3065 LASSERTF(left >= 0, "left = %d\n", left);
3066 if (left == lo->ldo_comp_cnt) {
3067 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
3068 lod2obd(d)->obd_name, id);
3072 memset(attr, 0, sizeof(*attr));
3073 attr->la_valid = LA_SIZE;
3074 rc = lod_sub_declare_attr_set(env, next, attr, th);
3079 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3080 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
3081 XATTR_NAME_LOV, 0, th);
3083 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
3090 * Declare layout add/set/del operations issued by special xattr names:
3092 * XATTR_LUSTRE_LOV.add add component(s) to existing file
3093 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
3094 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
3096 * \param[in] env execution environment
3097 * \param[in] dt object
3098 * \param[in] name name of xattr
3099 * \param[in] buf lu_buf contains xattr value
3100 * \param[in] th transaction handle
3102 * \retval 0 on success
3103 * \retval negative if failed
3105 static int lod_declare_modify_layout(const struct lu_env *env,
3106 struct dt_object *dt,
3108 const struct lu_buf *buf,
3111 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3112 struct lod_object *lo = lod_dt_obj(dt);
3114 int rc, len = strlen(XATTR_LUSTRE_LOV);
3117 LASSERT(dt_object_exists(dt));
3119 if (strlen(name) <= len || name[len] != '.') {
3120 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
3121 lod2obd(d)->obd_name, name);
3126 rc = lod_striping_load(env, lo);
3130 /* the layout to be modified must be a composite layout */
3131 if (!lo->ldo_is_composite) {
3132 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
3133 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3134 GOTO(unlock, rc = -EINVAL);
3137 op = (char *)name + len;
3138 if (strcmp(op, "add") == 0) {
3139 rc = lod_declare_layout_add(env, dt, buf, th);
3140 } else if (strcmp(op, "del") == 0) {
3141 rc = lod_declare_layout_del(env, dt, buf, th);
3142 } else if (strncmp(op, "set", strlen("set")) == 0) {
3143 rc = lod_declare_layout_set(env, dt, op, buf, th);
3145 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
3146 lod2obd(d)->obd_name, name);
3147 GOTO(unlock, rc = -ENOTSUPP);
3151 lod_striping_free(env, lo);
3157 * Convert a plain file lov_mds_md to a composite layout.
3159 * \param[in,out] info the thread info::lti_ea_store buffer contains little
3160 * endian plain file layout
3162 * \retval 0 on success, <0 on failure
3164 static int lod_layout_convert(struct lod_thread_info *info)
3166 struct lov_mds_md *lmm = info->lti_ea_store;
3167 struct lov_mds_md *lmm_save;
3168 struct lov_comp_md_v1 *lcm;
3169 struct lov_comp_md_entry_v1 *lcme;
3175 /* realloc buffer to a composite layout which contains one component */
3176 blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3177 le32_to_cpu(lmm->lmm_magic));
3178 size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
3180 OBD_ALLOC_LARGE(lmm_save, blob_size);
3182 GOTO(out, rc = -ENOMEM);
3184 memcpy(lmm_save, lmm, blob_size);
3186 if (info->lti_ea_store_size < size) {
3187 rc = lod_ea_store_resize(info, size);
3192 lcm = info->lti_ea_store;
3193 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
3194 lcm->lcm_size = cpu_to_le32(size);
3195 lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
3196 lmm_save->lmm_layout_gen));
3197 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
3198 lcm->lcm_entry_count = cpu_to_le16(1);
3199 lcm->lcm_mirror_count = 0;
3201 lcme = &lcm->lcm_entries[0];
3202 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
3203 lcme->lcme_extent.e_start = 0;
3204 lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
3205 lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
3206 lcme->lcme_size = cpu_to_le32(blob_size);
3208 memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
3213 OBD_FREE_LARGE(lmm_save, blob_size);
3218 * Merge layouts to form a mirrored file.
3220 static int lod_declare_layout_merge(const struct lu_env *env,
3221 struct dt_object *dt, const struct lu_buf *mbuf,
3224 struct lod_thread_info *info = lod_env_info(env);
3225 struct lu_buf *buf = &info->lti_buf;
3226 struct lod_object *lo = lod_dt_obj(dt);
3227 struct lov_comp_md_v1 *lcm;
3228 struct lov_comp_md_v1 *cur_lcm;
3229 struct lov_comp_md_v1 *merge_lcm;
3230 struct lov_comp_md_entry_v1 *lcme;
3233 __u16 cur_entry_count;
3234 __u16 merge_entry_count;
3236 __u16 mirror_id = 0;
3241 merge_lcm = mbuf->lb_buf;
3242 if (mbuf->lb_len < sizeof(*merge_lcm))
3245 /* must be an existing layout from disk */
3246 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
3249 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
3251 /* do not allow to merge two mirrored files */
3252 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
3255 /* verify the target buffer */
3256 rc = lod_get_lov_ea(env, lo);
3258 RETURN(rc ? : -ENODATA);
3260 cur_lcm = info->lti_ea_store;
3261 switch (le32_to_cpu(cur_lcm->lcm_magic)) {
3264 rc = lod_layout_convert(info);
3266 case LOV_MAGIC_COMP_V1:
3275 /* info->lti_ea_store could be reallocated in lod_layout_convert() */
3276 cur_lcm = info->lti_ea_store;
3277 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
3279 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
3280 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
3281 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
3284 /* size of new layout */
3285 size = le32_to_cpu(cur_lcm->lcm_size) +
3286 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
3288 memset(buf, 0, sizeof(*buf));
3289 lu_buf_alloc(buf, size);
3290 if (buf->lb_buf == NULL)
3294 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
3296 offset = sizeof(*lcm) +
3297 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
3298 for (i = 0; i < cur_entry_count; i++) {
3299 struct lov_comp_md_entry_v1 *cur_lcme;
3301 lcme = &lcm->lcm_entries[i];
3302 cur_lcme = &cur_lcm->lcm_entries[i];
3304 lcme->lcme_offset = cpu_to_le32(offset);
3305 memcpy((char *)lcm + offset,
3306 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
3307 le32_to_cpu(lcme->lcme_size));
3309 offset += le32_to_cpu(lcme->lcme_size);
3311 if (mirror_count == 1 &&
3312 mirror_id_of(le32_to_cpu(lcme->lcme_id)) == 0) {
3313 /* Add mirror from a non-flr file, create new mirror ID.
3314 * Otherwise, keep existing mirror's component ID, used
3315 * for mirror extension.
3317 id = pflr_id(1, i + 1);
3318 lcme->lcme_id = cpu_to_le32(id);
3321 id = MAX(le32_to_cpu(lcme->lcme_id), id);
3324 mirror_id = mirror_id_of(id) + 1;
3325 for (i = 0; i < merge_entry_count; i++) {
3326 struct lov_comp_md_entry_v1 *merge_lcme;
3328 merge_lcme = &merge_lcm->lcm_entries[i];
3329 lcme = &lcm->lcm_entries[cur_entry_count + i];
3331 *lcme = *merge_lcme;
3332 lcme->lcme_offset = cpu_to_le32(offset);
3334 id = pflr_id(mirror_id, i + 1);
3335 lcme->lcme_id = cpu_to_le32(id);
3337 memcpy((char *)lcm + offset,
3338 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
3339 le32_to_cpu(lcme->lcme_size));
3341 offset += le32_to_cpu(lcme->lcme_size);
3344 /* fixup layout information */
3345 lod_obj_inc_layout_gen(lo);
3346 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3347 lcm->lcm_size = cpu_to_le32(size);
3348 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
3349 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
3350 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NONE)
3351 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
3353 rc = lod_striping_reload(env, lo, buf);
3357 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
3358 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3366 * Split layouts, just set the LOVEA with the layout from mbuf.
3368 static int lod_declare_layout_split(const struct lu_env *env,
3369 struct dt_object *dt, const struct lu_buf *mbuf,
3372 struct lod_object *lo = lod_dt_obj(dt);
3373 struct lov_comp_md_v1 *lcm = mbuf->lb_buf;
3377 lod_obj_inc_layout_gen(lo);
3378 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3380 rc = lod_striping_reload(env, lo, mbuf);
3384 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), mbuf,
3385 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3390 * Implementation of dt_object_operations::do_declare_xattr_set.
3392 * \see dt_object_operations::do_declare_xattr_set() in the API description
3395 * the extension to the API:
3396 * - declaring LOVEA requests striping creation
3397 * - LU_XATTR_REPLACE means layout swap
3399 static int lod_declare_xattr_set(const struct lu_env *env,
3400 struct dt_object *dt,
3401 const struct lu_buf *buf,
3402 const char *name, int fl,
3405 struct dt_object *next = dt_object_child(dt);
3406 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3411 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
3412 if ((S_ISREG(mode) || mode == 0) &&
3413 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE | LU_XATTR_SPLIT)) &&
3414 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3415 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
3417 * this is a request to create object's striping.
3419 * allow to declare predefined striping on a new (!mode) object
3420 * which is supposed to be replay of regular file creation
3421 * (when LOV setting is declared)
3423 * LU_XATTR_REPLACE is set to indicate a layout swap
3425 if (dt_object_exists(dt)) {
3426 rc = dt_attr_get(env, next, attr);
3430 memset(attr, 0, sizeof(*attr));
3431 attr->la_valid = LA_TYPE | LA_MODE;
3432 attr->la_mode = S_IFREG;
3434 rc = lod_declare_striped_create(env, dt, attr, buf, th);
3435 } else if (fl & LU_XATTR_MERGE) {
3436 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3437 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3438 rc = lod_declare_layout_merge(env, dt, buf, th);
3439 } else if (fl & LU_XATTR_SPLIT) {
3440 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3441 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3442 rc = lod_declare_layout_split(env, dt, buf, th);
3443 } else if (S_ISREG(mode) &&
3444 strlen(name) > strlen(XATTR_LUSTRE_LOV) + 1 &&
3445 strncmp(name, XATTR_LUSTRE_LOV,
3446 strlen(XATTR_LUSTRE_LOV)) == 0) {
3448 * this is a request to modify object's striping.
3449 * add/set/del component(s).
3451 if (!dt_object_exists(dt))
3454 rc = lod_declare_modify_layout(env, dt, name, buf, th);
3455 } else if (strncmp(name, XATTR_NAME_LMV, strlen(XATTR_NAME_LMV)) == 0 &&
3456 strlen(name) > strlen(XATTR_NAME_LMV) + 1) {
3457 const char *op = name + strlen(XATTR_NAME_LMV) + 1;
3460 if (strcmp(op, "add") == 0)
3461 rc = lod_dir_declare_layout_add(env, dt, buf, th);
3462 else if (strcmp(op, "del") == 0)
3463 rc = lod_dir_declare_layout_delete(env, dt, buf, th);
3464 else if (strcmp(op, "set") == 0)
3465 rc = lod_sub_declare_xattr_set(env, next, buf,
3466 XATTR_NAME_LMV, fl, th);
3469 } else if (S_ISDIR(mode)) {
3470 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
3471 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3472 rc = lod_replace_parent_fid(env, dt, buf, th, true);
3474 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
3481 * Apply xattr changes to the object.
3483 * Applies xattr changes to the object and the stripes if the latter exist.
3485 * \param[in] env execution environment
3486 * \param[in] dt object
3487 * \param[in] buf buffer pointing to the new value of xattr
3488 * \param[in] name name of xattr
3489 * \param[in] fl flags
3490 * \param[in] th transaction handle
3492 * \retval 0 on success
3493 * \retval negative if failed
3495 static int lod_xattr_set_internal(const struct lu_env *env,
3496 struct dt_object *dt,
3497 const struct lu_buf *buf,
3498 const char *name, int fl,
3501 struct dt_object *next = dt_object_child(dt);
3502 struct lod_object *lo = lod_dt_obj(dt);
3507 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3508 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3511 /* Note: Do not set LinkEA on sub-stripes, otherwise
3512 * it will confuse the fid2path process(see mdt_path_current()).
3513 * The linkEA between master and sub-stripes is set in
3514 * lod_xattr_set_lmv(). */
3515 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
3518 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3519 if (!lo->ldo_stripe[i])
3522 if (!dt_object_exists(lo->ldo_stripe[i]))
3525 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
3535 * Delete an extended attribute.
3537 * Deletes specified xattr from the object and the stripes if the latter exist.
3539 * \param[in] env execution environment
3540 * \param[in] dt object
3541 * \param[in] name name of xattr
3542 * \param[in] th transaction handle
3544 * \retval 0 on success
3545 * \retval negative if failed
3547 static int lod_xattr_del_internal(const struct lu_env *env,
3548 struct dt_object *dt,
3549 const char *name, struct thandle *th)
3551 struct dt_object *next = dt_object_child(dt);
3552 struct lod_object *lo = lod_dt_obj(dt);
3557 rc = lod_sub_xattr_del(env, next, name, th);
3558 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3561 if (lo->ldo_dir_stripe_count == 0)
3564 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3565 LASSERT(lo->ldo_stripe[i]);
3567 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3576 * Set default striping on a directory.
3578 * Sets specified striping on a directory object unless it matches the default
3579 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3580 * EA. This striping will be used when regular file is being created in this
3583 * \param[in] env execution environment
3584 * \param[in] dt the striped object
3585 * \param[in] buf buffer with the striping
3586 * \param[in] name name of EA
3587 * \param[in] fl xattr flag (see OSD API description)
3588 * \param[in] th transaction handle
3590 * \retval 0 on success
3591 * \retval negative if failed
3593 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
3594 struct dt_object *dt,
3595 const struct lu_buf *buf,
3596 const char *name, int fl,
3599 struct lov_user_md_v1 *lum;
3600 struct lov_user_md_v3 *v3 = NULL;
3601 const char *pool_name = NULL;
3606 LASSERT(buf != NULL && buf->lb_buf != NULL);
3609 switch (lum->lmm_magic) {
3610 case LOV_USER_MAGIC_SPECIFIC:
3611 case LOV_USER_MAGIC_V3:
3613 if (v3->lmm_pool_name[0] != '\0')
3614 pool_name = v3->lmm_pool_name;
3616 case LOV_USER_MAGIC_V1:
3617 /* if { size, offset, count } = { 0, -1, 0 } and no pool
3618 * (i.e. all default values specified) then delete default
3619 * striping from dir. */
3621 "set default striping: sz %u # %u offset %d %s %s\n",
3622 (unsigned)lum->lmm_stripe_size,
3623 (unsigned)lum->lmm_stripe_count,
3624 (int)lum->lmm_stripe_offset,
3625 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
3627 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
3628 lum->lmm_stripe_count,
3629 lum->lmm_stripe_offset,
3632 case LOV_USER_MAGIC_COMP_V1:
3636 CERROR("Invalid magic %x\n", lum->lmm_magic);
3641 rc = lod_xattr_del_internal(env, dt, name, th);
3645 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3652 * Set default striping on a directory object.
3654 * Sets specified striping on a directory object unless it matches the default
3655 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3656 * EA. This striping will be used when a new directory is being created in the
3659 * \param[in] env execution environment
3660 * \param[in] dt the striped object
3661 * \param[in] buf buffer with the striping
3662 * \param[in] name name of EA
3663 * \param[in] fl xattr flag (see OSD API description)
3664 * \param[in] th transaction handle
3666 * \retval 0 on success
3667 * \retval negative if failed
3669 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
3670 struct dt_object *dt,
3671 const struct lu_buf *buf,
3672 const char *name, int fl,
3675 struct lmv_user_md_v1 *lum;
3680 LASSERT(buf != NULL && buf->lb_buf != NULL);
3684 "set default stripe_count # %u stripe_offset %d hash %u\n",
3685 le32_to_cpu(lum->lum_stripe_count),
3686 (int)le32_to_cpu(lum->lum_stripe_offset),
3687 le32_to_cpu(lum->lum_hash_type));
3689 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
3690 le32_to_cpu(lum->lum_stripe_offset)) &&
3691 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC &&
3692 !(le32_to_cpu(lum->lum_hash_type) & LMV_HASH_FLAG_SPACE)) {
3693 rc = lod_xattr_del_internal(env, dt, name, th);
3697 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3706 * Turn directory into a striped directory.
3708 * During replay the client sends the striping created before MDT
3709 * failure, then the layer above LOD sends this defined striping
3710 * using ->do_xattr_set(), so LOD uses this method to replay creation
3711 * of the stripes. Notice the original information for the striping
3712 * (#stripes, FIDs, etc) was transferred in declare path.
3714 * \param[in] env execution environment
3715 * \param[in] dt the striped object
3716 * \param[in] buf not used currently
3717 * \param[in] name not used currently
3718 * \param[in] fl xattr flag (see OSD API description)
3719 * \param[in] th transaction handle
3721 * \retval 0 on success
3722 * \retval negative if failed
3724 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
3725 const struct lu_buf *buf, const char *name,
3726 int fl, struct thandle *th)
3728 struct lod_object *lo = lod_dt_obj(dt);
3729 struct lod_thread_info *info = lod_env_info(env);
3730 struct lu_attr *attr = &info->lti_attr;
3731 struct dt_object_format *dof = &info->lti_format;
3732 struct lu_buf lmv_buf;
3733 struct lu_buf slave_lmv_buf;
3734 struct lmv_mds_md_v1 *lmm;
3735 struct lmv_mds_md_v1 *slave_lmm = NULL;
3736 struct dt_insert_rec *rec = &info->lti_dt_rec;
3741 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3744 /* The stripes are supposed to be allocated in declare phase,
3745 * if there are no stripes being allocated, it will skip */
3746 if (lo->ldo_dir_stripe_count == 0) {
3747 if (lo->ldo_dir_is_foreign) {
3748 rc = lod_sub_xattr_set(env, dt_object_child(dt), buf,
3749 XATTR_NAME_LMV, fl, th);
3756 rc = dt_attr_get(env, dt_object_child(dt), attr);
3760 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME |
3761 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
3762 dof->dof_type = DFT_DIR;
3764 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
3767 lmm = lmv_buf.lb_buf;
3769 OBD_ALLOC_PTR(slave_lmm);
3770 if (slave_lmm == NULL)
3773 lod_prep_slave_lmv_md(slave_lmm, lmm);
3774 slave_lmv_buf.lb_buf = slave_lmm;
3775 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
3777 rec->rec_type = S_IFDIR;
3778 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3779 struct dt_object *dto = lo->ldo_stripe[i];
3780 char *stripe_name = info->lti_key;
3781 struct lu_name *sname;
3782 struct linkea_data ldata = { NULL };
3783 struct lu_buf linkea_buf;
3785 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
3789 /* fail a remote stripe creation */
3790 if (i && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_CREATE))
3793 /* if it's source stripe of migrating directory, don't create */
3794 if (!((lo->ldo_dir_hash_type & LMV_HASH_FLAG_MIGRATION) &&
3795 i >= lo->ldo_dir_migrate_offset)) {
3796 dt_write_lock(env, dto, MOR_TGT_CHILD);
3797 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
3799 dt_write_unlock(env, dto);
3803 rc = lod_sub_ref_add(env, dto, th);
3804 dt_write_unlock(env, dto);
3808 rec->rec_fid = lu_object_fid(&dto->do_lu);
3809 rc = lod_sub_insert(env, dto,
3810 (const struct dt_rec *)rec,
3811 (const struct dt_key *)dot, th);
3816 rec->rec_fid = lu_object_fid(&dt->do_lu);
3817 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
3818 (const struct dt_key *)dotdot, th);
3822 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
3823 cfs_fail_val != i) {
3824 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
3826 slave_lmm->lmv_master_mdt_index =
3829 slave_lmm->lmv_master_mdt_index =
3832 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
3833 XATTR_NAME_LMV, fl, 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 int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
4126 /* clear memory region that will be used for layout change */
4127 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
4128 info->lti_count = 0;
4130 if (info->lti_comp_size >= comp_cnt)
4133 if (info->lti_comp_size > 0) {
4134 OBD_FREE(info->lti_comp_idx,
4135 info->lti_comp_size * sizeof(__u32));
4136 info->lti_comp_size = 0;
4139 OBD_ALLOC(info->lti_comp_idx, comp_cnt * sizeof(__u32));
4140 if (!info->lti_comp_idx)
4143 info->lti_comp_size = comp_cnt;
4148 * Prepare new layout minus deleted components
4150 * Removes components marked for deletion (LCME_ID_INVAL) by copying to a new
4151 * layout and skipping those components. Removes stripe objects if any exist.
4154 * Reallocates layout components array (lo->ldo_comp_entries), invalidating
4155 * any pre-existing pointers to components.
4157 * Caller is responsible for updating mirror end (ldo_mirror[].lme_end).
4159 * \param[in] env execution environment for this thread
4160 * \param[in,out] lo object to update the layout of
4161 * \param[in] th transaction handle for this operation
4163 * \retval # of components deleted
4164 * \retval negative errno on error
4166 static int lod_layout_del_prep_layout(const struct lu_env *env,
4167 struct lod_object *lo,
4170 struct lod_layout_component *lod_comp;
4171 struct lod_thread_info *info = lod_env_info(env);
4172 int rc = 0, i, j, deleted = 0;
4176 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
4180 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4181 lod_comp = &lo->ldo_comp_entries[i];
4183 if (lod_comp->llc_id != LCME_ID_INVAL) {
4184 /* Build array of things to keep */
4185 info->lti_comp_idx[info->lti_count++] = i;
4189 lod_obj_set_pool(lo, i, NULL);
4190 if (lod_comp->llc_ostlist.op_array) {
4191 OBD_FREE(lod_comp->llc_ostlist.op_array,
4192 lod_comp->llc_ostlist.op_size);
4193 lod_comp->llc_ostlist.op_array = NULL;
4194 lod_comp->llc_ostlist.op_size = 0;
4198 CDEBUG(D_LAYOUT, "deleting comp %d, left %d\n", i,
4199 lo->ldo_comp_cnt - deleted);
4201 /* No striping info for this component */
4202 if (lod_comp->llc_stripe == NULL)
4205 LASSERT(lod_comp->llc_stripe_count > 0);
4206 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4207 struct dt_object *obj = lod_comp->llc_stripe[j];
4212 /* components which are not init have no sub objects
4214 if (lod_comp_inited(lod_comp)) {
4215 rc = lod_sub_destroy(env, obj, th);
4220 lu_object_put(env, &obj->do_lu);
4221 lod_comp->llc_stripe[j] = NULL;
4223 OBD_FREE(lod_comp->llc_stripe, sizeof(struct dt_object *) *
4224 lod_comp->llc_stripes_allocated);
4225 lod_comp->llc_stripe = NULL;
4226 OBD_FREE(lod_comp->llc_ost_indices,
4227 sizeof(__u32) * lod_comp->llc_stripes_allocated);
4228 lod_comp->llc_ost_indices = NULL;
4229 lod_comp->llc_stripes_allocated = 0;
4232 /* info->lti_count has the amount of left components */
4233 LASSERTF(info->lti_count >= 0 && info->lti_count < lo->ldo_comp_cnt,
4234 "left = %d, lo->ldo_comp_cnt %d\n", (int)info->lti_count,
4235 (int)lo->ldo_comp_cnt);
4237 if (info->lti_count > 0) {
4238 struct lod_layout_component *comp_array;
4240 OBD_ALLOC(comp_array, sizeof(*comp_array) * info->lti_count);
4241 if (comp_array == NULL)
4242 GOTO(out, rc = -ENOMEM);
4244 for (i = 0; i < info->lti_count; i++) {
4245 memcpy(&comp_array[i],
4246 &lo->ldo_comp_entries[info->lti_comp_idx[i]],
4247 sizeof(*comp_array));
4250 OBD_FREE(lo->ldo_comp_entries,
4251 sizeof(*comp_array) * lo->ldo_comp_cnt);
4252 lo->ldo_comp_entries = comp_array;
4253 lo->ldo_comp_cnt = info->lti_count;
4255 lod_free_comp_entries(lo);
4260 return rc ? rc : deleted;
4264 * Delete layout component(s)
4266 * This function sets up the layout data in the env and does the setattrs
4267 * required to write out the new layout. The layout itself is modified in
4268 * lod_layout_del_prep_layout.
4270 * \param[in] env execution environment for this thread
4271 * \param[in] dt object
4272 * \param[in] th transaction handle
4274 * \retval 0 on success
4275 * \retval negative error number on failure
4277 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
4280 struct lod_object *lo = lod_dt_obj(dt);
4281 struct dt_object *next = dt_object_child(dt);
4282 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4285 LASSERT(lo->ldo_is_composite);
4286 LASSERT(lo->ldo_mirror_count == 1);
4287 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
4289 rc = lod_layout_del_prep_layout(env, lo, th);
4293 /* Only do this if we didn't delete all components */
4294 if (lo->ldo_comp_cnt > 0) {
4295 lo->ldo_mirrors[0].lme_end = lo->ldo_comp_cnt - 1;
4296 lod_obj_inc_layout_gen(lo);
4299 LASSERT(dt_object_exists(dt));
4300 rc = dt_attr_get(env, next, attr);
4304 if (attr->la_size > 0) {
4306 attr->la_valid = LA_SIZE;
4307 rc = lod_sub_attr_set(env, next, attr, th);
4312 rc = lod_generate_and_set_lovea(env, lo, th);
4316 lod_striping_free(env, lo);
4321 static int lod_get_default_lov_striping(const struct lu_env *env,
4322 struct lod_object *lo,
4323 struct lod_default_striping *lds);
4325 * Implementation of dt_object_operations::do_xattr_set.
4327 * Sets specified extended attribute on the object. Three types of EAs are
4329 * LOV EA - stores striping for a regular file or default striping (when set
4331 * LMV EA - stores a marker for the striped directories
4332 * DMV EA - stores default directory striping
4334 * When striping is applied to a non-striped existing object (this is called
4335 * late striping), then LOD notices the caller wants to turn the object into a
4336 * striped one. The stripe objects are created and appropriate EA is set:
4337 * LOV EA storing all the stripes directly or LMV EA storing just a small header
4338 * with striping configuration.
4340 * \see dt_object_operations::do_xattr_set() in the API description for details.
4342 static int lod_xattr_set(const struct lu_env *env,
4343 struct dt_object *dt, const struct lu_buf *buf,
4344 const char *name, int fl, struct thandle *th)
4346 struct dt_object *next = dt_object_child(dt);
4350 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4351 strcmp(name, XATTR_NAME_LMV) == 0) {
4352 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
4354 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4355 strncmp(name, XATTR_NAME_LMV, strlen(XATTR_NAME_LMV)) == 0 &&
4356 strlen(name) > strlen(XATTR_NAME_LMV) + 1) {
4357 const char *op = name + strlen(XATTR_NAME_LMV) + 1;
4361 * XATTR_NAME_LMV".add" is never called, but only declared,
4362 * because lod_xattr_set_lmv() will do the addition.
4364 if (strcmp(op, "del") == 0)
4365 rc = lod_dir_layout_delete(env, dt, buf, th);
4366 else if (strcmp(op, "set") == 0)
4367 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LMV,
4371 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4372 strcmp(name, XATTR_NAME_LOV) == 0) {
4373 struct lod_default_striping *lds = lod_lds_buf_get(env);
4374 struct lov_user_md_v1 *v1 = buf->lb_buf;
4375 char pool[LOV_MAXPOOLNAME + 1];
4378 /* get existing striping config */
4379 rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds);
4383 memset(pool, 0, sizeof(pool));
4384 if (lds->lds_def_striping_set == 1)
4385 lod_layout_get_pool(lds->lds_def_comp_entries,
4386 lds->lds_def_comp_cnt, pool,
4389 is_del = LOVEA_DELETE_VALUES(v1->lmm_stripe_size,
4390 v1->lmm_stripe_count,
4391 v1->lmm_stripe_offset,
4394 /* Retain the pool name if it is not given */
4395 if (v1->lmm_magic == LOV_USER_MAGIC_V1 && pool[0] != '\0' &&
4397 struct lod_thread_info *info = lod_env_info(env);
4398 struct lov_user_md_v3 *v3 = info->lti_ea_store;
4400 memset(v3, 0, sizeof(*v3));
4401 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4402 v3->lmm_pattern = cpu_to_le32(v1->lmm_pattern);
4403 v3->lmm_stripe_count =
4404 cpu_to_le32(v1->lmm_stripe_count);
4405 v3->lmm_stripe_offset =
4406 cpu_to_le32(v1->lmm_stripe_offset);
4407 v3->lmm_stripe_size = cpu_to_le32(v1->lmm_stripe_size);
4409 strlcpy(v3->lmm_pool_name, pool,
4410 sizeof(v3->lmm_pool_name));
4412 info->lti_buf.lb_buf = v3;
4413 info->lti_buf.lb_len = sizeof(*v3);
4414 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4417 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name,
4421 if (lds->lds_def_striping_set == 1 &&
4422 lds->lds_def_comp_entries != NULL)
4423 lod_free_def_comp_entries(lds);
4426 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4427 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
4429 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
4432 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
4433 (!strcmp(name, XATTR_NAME_LOV) ||
4434 !strncmp(name, XATTR_LUSTRE_LOV,
4435 strlen(XATTR_LUSTRE_LOV)))) {
4436 /* in case of lov EA swap, just set it
4437 * if not, it is a replay so check striping match what we
4438 * already have during req replay, declare_xattr_set()
4439 * defines striping, then create() does the work */
4440 if (fl & LU_XATTR_REPLACE) {
4441 /* free stripes, then update disk */
4442 lod_striping_free(env, lod_dt_obj(dt));
4444 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
4445 } else if (dt_object_remote(dt)) {
4446 /* This only happens during migration, see
4447 * mdd_migrate_create(), in which Master MDT will
4448 * create a remote target object, and only set
4449 * (migrating) stripe EA on the remote object,
4450 * and does not need creating each stripes. */
4451 rc = lod_sub_xattr_set(env, next, buf, name,
4453 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
4454 /* delete component(s) */
4455 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
4456 rc = lod_layout_del(env, dt, th);
4459 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
4460 * it's going to create create file with specified
4461 * component(s), the striping must have not being
4462 * cached in this case;
4464 * Otherwise, it's going to add/change component(s) to
4465 * an existing file, the striping must have been cached
4468 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
4469 !strcmp(name, XATTR_NAME_LOV),
4470 !lod_dt_obj(dt)->ldo_comp_cached));
4472 rc = lod_striped_create(env, dt, NULL, NULL, th);
4475 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4476 rc = lod_replace_parent_fid(env, dt, buf, th, false);
4481 /* then all other xattr */
4482 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4488 * Implementation of dt_object_operations::do_declare_xattr_del.
4490 * \see dt_object_operations::do_declare_xattr_del() in the API description
4493 static int lod_declare_xattr_del(const struct lu_env *env,
4494 struct dt_object *dt, const char *name,
4497 struct lod_object *lo = lod_dt_obj(dt);
4498 struct dt_object *next = dt_object_child(dt);
4503 rc = lod_sub_declare_xattr_del(env, next, name, th);
4507 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4510 /* set xattr to each stripes, if needed */
4511 rc = lod_striping_load(env, lo);
4515 if (lo->ldo_dir_stripe_count == 0)
4518 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4519 struct dt_object *dto = lo->ldo_stripe[i];
4524 rc = lod_sub_declare_xattr_del(env, dto, name, th);
4533 * Implementation of dt_object_operations::do_xattr_del.
4535 * If EA storing a regular striping is being deleted, then release
4536 * all the references to the stripe objects in core.
4538 * \see dt_object_operations::do_xattr_del() in the API description for details.
4540 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
4541 const char *name, struct thandle *th)
4543 struct dt_object *next = dt_object_child(dt);
4544 struct lod_object *lo = lod_dt_obj(dt);
4549 if (!strcmp(name, XATTR_NAME_LOV) || !strcmp(name, XATTR_NAME_LMV))
4550 lod_striping_free(env, lod_dt_obj(dt));
4552 rc = lod_sub_xattr_del(env, next, name, th);
4553 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
4556 if (lo->ldo_dir_stripe_count == 0)
4559 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4560 struct dt_object *dto = lo->ldo_stripe[i];
4565 rc = lod_sub_xattr_del(env, dto, name, th);
4574 * Implementation of dt_object_operations::do_xattr_list.
4576 * \see dt_object_operations::do_xattr_list() in the API description
4579 static int lod_xattr_list(const struct lu_env *env,
4580 struct dt_object *dt, const struct lu_buf *buf)
4582 return dt_xattr_list(env, dt_object_child(dt), buf);
4585 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
4587 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
4591 * Copy OST list from layout provided by user.
4593 * \param[in] lod_comp layout_component to be filled
4594 * \param[in] v3 LOV EA V3 user data
4596 * \retval 0 on success
4597 * \retval negative if failed
4599 int lod_comp_copy_ost_lists(struct lod_layout_component *lod_comp,
4600 struct lov_user_md_v3 *v3)
4606 if (v3->lmm_stripe_offset == LOV_OFFSET_DEFAULT)
4607 v3->lmm_stripe_offset = v3->lmm_objects[0].l_ost_idx;
4609 if (lod_comp->llc_ostlist.op_array) {
4610 if (lod_comp->llc_ostlist.op_size >=
4611 v3->lmm_stripe_count * sizeof(__u32)) {
4612 lod_comp->llc_ostlist.op_count =
4613 v3->lmm_stripe_count;
4616 OBD_FREE(lod_comp->llc_ostlist.op_array,
4617 lod_comp->llc_ostlist.op_size);
4620 /* copy ost list from lmm */
4621 lod_comp->llc_ostlist.op_count = v3->lmm_stripe_count;
4622 lod_comp->llc_ostlist.op_size = v3->lmm_stripe_count * sizeof(__u32);
4623 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
4624 lod_comp->llc_ostlist.op_size);
4625 if (!lod_comp->llc_ostlist.op_array)
4628 for (j = 0; j < v3->lmm_stripe_count; j++) {
4629 lod_comp->llc_ostlist.op_array[j] =
4630 v3->lmm_objects[j].l_ost_idx;
4638 * Get default striping.
4640 * \param[in] env execution environment
4641 * \param[in] lo object
4642 * \param[out] lds default striping
4644 * \retval 0 on success
4645 * \retval negative if failed
4647 static int lod_get_default_lov_striping(const struct lu_env *env,
4648 struct lod_object *lo,
4649 struct lod_default_striping *lds)
4651 struct lod_thread_info *info = lod_env_info(env);
4652 struct lov_user_md_v1 *v1 = NULL;
4653 struct lov_user_md_v3 *v3 = NULL;
4654 struct lov_comp_md_v1 *comp_v1 = NULL;
4662 rc = lod_get_lov_ea(env, lo);
4666 if (rc < (typeof(rc))sizeof(struct lov_user_md))
4669 v1 = info->lti_ea_store;
4670 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
4671 lustre_swab_lov_user_md_v1(v1);
4672 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
4673 v3 = (struct lov_user_md_v3 *)v1;
4674 lustre_swab_lov_user_md_v3(v3);
4675 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_SPECIFIC)) {
4676 v3 = (struct lov_user_md_v3 *)v1;
4677 lustre_swab_lov_user_md_v3(v3);
4678 lustre_swab_lov_user_md_objects(v3->lmm_objects,
4679 v3->lmm_stripe_count);
4680 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
4681 comp_v1 = (struct lov_comp_md_v1 *)v1;
4682 lustre_swab_lov_comp_md_v1(comp_v1);
4685 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1 &&
4686 v1->lmm_magic != LOV_MAGIC_COMP_V1 &&
4687 v1->lmm_magic != LOV_USER_MAGIC_SPECIFIC)
4690 if (v1->lmm_magic == LOV_MAGIC_COMP_V1) {
4691 comp_v1 = (struct lov_comp_md_v1 *)v1;
4692 comp_cnt = comp_v1->lcm_entry_count;
4695 mirror_cnt = comp_v1->lcm_mirror_count + 1;
4703 /* realloc default comp entries if necessary */
4704 rc = lod_def_striping_comp_resize(lds, comp_cnt);
4708 lds->lds_def_comp_cnt = comp_cnt;
4709 lds->lds_def_striping_is_composite = composite;
4710 lds->lds_def_mirror_cnt = mirror_cnt;
4712 for (i = 0; i < comp_cnt; i++) {
4713 struct lod_layout_component *lod_comp;
4716 lod_comp = &lds->lds_def_comp_entries[i];
4718 * reset lod_comp values, llc_stripes is always NULL in
4719 * the default striping template, llc_pool will be reset
4722 memset(lod_comp, 0, offsetof(typeof(*lod_comp), llc_pool));
4725 v1 = (struct lov_user_md *)((char *)comp_v1 +
4726 comp_v1->lcm_entries[i].lcme_offset);
4727 lod_comp->llc_extent =
4728 comp_v1->lcm_entries[i].lcme_extent;
4729 /* We only inherit certain flags from the layout */
4730 lod_comp->llc_flags =
4731 comp_v1->lcm_entries[i].lcme_flags &
4732 LCME_TEMPLATE_FLAGS;
4735 if (!lov_pattern_supported(v1->lmm_pattern) &&
4736 !(v1->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
4737 lod_free_def_comp_entries(lds);
4741 CDEBUG(D_LAYOUT, DFID" stripe_count=%d stripe_size=%d "
4742 "stripe_offset=%d\n",
4743 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
4744 (int)v1->lmm_stripe_count, (int)v1->lmm_stripe_size,
4745 (int)v1->lmm_stripe_offset);
4747 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
4748 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
4749 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
4750 lod_comp->llc_pattern = v1->lmm_pattern;
4753 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
4754 /* XXX: sanity check here */
4755 v3 = (struct lov_user_md_v3 *) v1;
4756 if (v3->lmm_pool_name[0] != '\0')
4757 pool = v3->lmm_pool_name;
4759 lod_set_def_pool(lds, i, pool);
4760 if (v1->lmm_magic == LOV_USER_MAGIC_SPECIFIC) {
4761 v3 = (struct lov_user_md_v3 *)v1;
4762 rc = lod_comp_copy_ost_lists(lod_comp, v3);
4765 } else if (lod_comp->llc_ostlist.op_array &&
4766 lod_comp->llc_ostlist.op_count) {
4767 for (j = 0; j < lod_comp->llc_ostlist.op_count; j++)
4768 lod_comp->llc_ostlist.op_array[j] = -1;
4769 lod_comp->llc_ostlist.op_count = 0;
4773 lds->lds_def_striping_set = 1;
4778 * Get default directory striping.
4780 * \param[in] env execution environment
4781 * \param[in] lo object
4782 * \param[out] lds default striping
4784 * \retval 0 on success
4785 * \retval negative if failed
4787 static int lod_get_default_lmv_striping(const struct lu_env *env,
4788 struct lod_object *lo,
4789 struct lod_default_striping *lds)
4791 struct lmv_user_md *lmu;
4794 lds->lds_dir_def_striping_set = 0;
4796 rc = lod_get_default_lmv_ea(env, lo);
4800 if (rc >= (int)sizeof(*lmu)) {
4801 struct lod_thread_info *info = lod_env_info(env);
4803 lmu = info->lti_ea_store;
4805 lds->lds_dir_def_stripe_count =
4806 le32_to_cpu(lmu->lum_stripe_count);
4807 lds->lds_dir_def_stripe_offset =
4808 le32_to_cpu(lmu->lum_stripe_offset);
4809 lds->lds_dir_def_hash_type =
4810 le32_to_cpu(lmu->lum_hash_type);
4811 lds->lds_dir_def_striping_set = 1;
4818 * Get default striping in the object.
4820 * Get object default striping and default directory striping.
4822 * \param[in] env execution environment
4823 * \param[in] lo object
4824 * \param[out] lds default striping
4826 * \retval 0 on success
4827 * \retval negative if failed
4829 static int lod_get_default_striping(const struct lu_env *env,
4830 struct lod_object *lo,
4831 struct lod_default_striping *lds)
4835 rc = lod_get_default_lov_striping(env, lo, lds);
4836 rc1 = lod_get_default_lmv_striping(env, lo, lds);
4837 if (rc == 0 && rc1 < 0)
4844 * Apply default striping on object.
4846 * If object striping pattern is not set, set to the one in default striping.
4847 * The default striping is from parent or fs.
4849 * \param[in] lo new object
4850 * \param[in] lds default striping
4851 * \param[in] mode new object's mode
4853 static void lod_striping_from_default(struct lod_object *lo,
4854 const struct lod_default_striping *lds,
4857 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
4858 struct lov_desc *desc = &d->lod_desc;
4861 if (lds->lds_def_striping_set && S_ISREG(mode)) {
4862 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
4863 lds->lds_def_comp_cnt);
4867 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
4868 if (lds->lds_def_mirror_cnt > 1)
4869 lo->ldo_flr_state = LCM_FL_RDONLY;
4871 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4872 struct lod_layout_component *obj_comp =
4873 &lo->ldo_comp_entries[i];
4874 struct lod_layout_component *def_comp =
4875 &lds->lds_def_comp_entries[i];
4877 CDEBUG(D_LAYOUT, "Inherit from default: flags=%#x "
4878 "size=%hu nr=%u offset=%u pattern=%#x pool=%s\n",
4879 def_comp->llc_flags,
4880 def_comp->llc_stripe_size,
4881 def_comp->llc_stripe_count,
4882 def_comp->llc_stripe_offset,
4883 def_comp->llc_pattern,
4884 def_comp->llc_pool ?: "");
4886 *obj_comp = *def_comp;
4887 if (def_comp->llc_pool != NULL) {
4888 /* pointer was copied from def_comp */
4889 obj_comp->llc_pool = NULL;
4890 lod_obj_set_pool(lo, i, def_comp->llc_pool);
4894 if (def_comp->llc_ostlist.op_array &&
4895 def_comp->llc_ostlist.op_count) {
4896 OBD_ALLOC(obj_comp->llc_ostlist.op_array,
4897 obj_comp->llc_ostlist.op_size);
4898 if (!obj_comp->llc_ostlist.op_array)
4900 memcpy(obj_comp->llc_ostlist.op_array,
4901 def_comp->llc_ostlist.op_array,
4902 obj_comp->llc_ostlist.op_size);
4903 } else if (def_comp->llc_ostlist.op_array) {
4904 obj_comp->llc_ostlist.op_array = NULL;
4908 * Don't initialize these fields for plain layout
4909 * (v1/v3) here, they are inherited in the order of
4910 * 'parent' -> 'fs default (root)' -> 'global default
4911 * values for stripe_count & stripe_size'.
4913 * see lod_ah_init().
4915 if (!lo->ldo_is_composite)
4918 lod_adjust_stripe_info(obj_comp, desc);
4920 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
4921 if (lo->ldo_dir_stripe_count == 0)
4922 lo->ldo_dir_stripe_count =
4923 lds->lds_dir_def_stripe_count;
4924 if (lo->ldo_dir_stripe_offset == -1)
4925 lo->ldo_dir_stripe_offset =
4926 lds->lds_dir_def_stripe_offset;
4927 if (lo->ldo_dir_hash_type == 0)
4928 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type &
4929 ~LMV_HASH_FLAG_SPACE;
4931 CDEBUG(D_LAYOUT, "striping from default dir: count:%hu, "
4932 "offset:%u, hash_type:%u\n",
4933 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
4934 lo->ldo_dir_hash_type);
4938 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root)
4940 struct lod_layout_component *lod_comp;
4942 if (lo->ldo_comp_cnt == 0)
4945 if (lo->ldo_is_composite)
4948 lod_comp = &lo->ldo_comp_entries[0];
4950 if (lod_comp->llc_stripe_count <= 0 ||
4951 lod_comp->llc_stripe_size <= 0)
4954 if (from_root && (lod_comp->llc_pool == NULL ||
4955 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
4962 * Implementation of dt_object_operations::do_ah_init.
4964 * This method is used to make a decision on the striping configuration for the
4965 * object being created. It can be taken from the \a parent object if it exists,
4966 * or filesystem's default. The resulting configuration (number of stripes,
4967 * stripe size/offset, pool name, etc) is stored in the object itself and will
4968 * be used by the methods like ->doo_declare_create().
4970 * \see dt_object_operations::do_ah_init() in the API description for details.
4972 static void lod_ah_init(const struct lu_env *env,
4973 struct dt_allocation_hint *ah,
4974 struct dt_object *parent,
4975 struct dt_object *child,
4978 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
4979 struct lod_thread_info *info = lod_env_info(env);
4980 struct lod_default_striping *lds = lod_lds_buf_get(env);
4981 struct dt_object *nextp = NULL;
4982 struct dt_object *nextc;
4983 struct lod_object *lp = NULL;
4984 struct lod_object *lc;
4985 struct lov_desc *desc;
4986 struct lod_layout_component *lod_comp;
4992 if (likely(parent)) {
4993 nextp = dt_object_child(parent);
4994 lp = lod_dt_obj(parent);
4997 nextc = dt_object_child(child);
4998 lc = lod_dt_obj(child);
5000 LASSERT(!lod_obj_is_striped(child));
5001 /* default layout template may have been set on the regular file
5002 * when this is called from mdd_create_data() */
5003 if (S_ISREG(child_mode))
5004 lod_free_comp_entries(lc);
5006 if (!dt_object_exists(nextc))
5007 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
5009 if (S_ISDIR(child_mode)) {
5010 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
5012 /* other default values are 0 */
5013 lc->ldo_dir_stripe_offset = -1;
5015 /* no default striping configuration is needed for
5018 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5019 le32_to_cpu(lum1->lum_magic) == LMV_MAGIC_FOREIGN) {
5020 lc->ldo_dir_is_foreign = true;
5021 /* keep stripe_count 0 and stripe_offset -1 */
5022 CDEBUG(D_INFO, "no default striping for foreign dir\n");
5027 * If parent object is not root directory,
5028 * then get default striping from parent object.
5030 if (likely(lp != NULL) && !fid_is_root(lod_object_fid(lp)))
5031 lod_get_default_striping(env, lp, lds);
5033 /* set child default striping info, default value is NULL */
5034 if (lds->lds_def_striping_set || lds->lds_dir_def_striping_set)
5035 lc->ldo_def_striping = lds;
5037 /* It should always honour the specified stripes */
5038 /* Note: old client (< 2.7)might also do lfs mkdir, whose EA
5039 * will have old magic. In this case, we should ignore the
5040 * stripe count and try to create dir by default stripe.
5042 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5043 (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
5044 le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC)) {
5045 lc->ldo_dir_stripe_count =
5046 le32_to_cpu(lum1->lum_stripe_count);
5047 lc->ldo_dir_stripe_offset =
5048 le32_to_cpu(lum1->lum_stripe_offset);
5049 lc->ldo_dir_hash_type =
5050 le32_to_cpu(lum1->lum_hash_type);
5052 "set dirstripe: count %hu, offset %d, hash %u\n",
5053 lc->ldo_dir_stripe_count,
5054 (int)lc->ldo_dir_stripe_offset,
5055 lc->ldo_dir_hash_type);
5057 /* transfer defaults LMV to new directory */
5058 lod_striping_from_default(lc, lds, child_mode);
5060 /* set count 0 to create normal directory */
5061 if (lc->ldo_dir_stripe_count == 1)
5062 lc->ldo_dir_stripe_count = 0;
5065 /* shrink the stripe_count to the avaible MDT count */
5066 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
5067 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)) {
5068 lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
5069 if (lc->ldo_dir_stripe_count == 1)
5070 lc->ldo_dir_stripe_count = 0;
5073 CDEBUG(D_INFO, "final dir stripe [%hu %d %u]\n",
5074 lc->ldo_dir_stripe_count,
5075 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
5080 /* child object regular file*/
5082 if (!lod_object_will_be_striped(S_ISREG(child_mode),
5083 lu_object_fid(&child->do_lu)))
5086 /* If object is going to be striped over OSTs, transfer default
5087 * striping information to the child, so that we can use it
5088 * during declaration and creation.
5090 * Try from the parent first.
5092 if (likely(lp != NULL)) {
5093 rc = lod_get_default_lov_striping(env, lp, lds);
5095 lod_striping_from_default(lc, lds, child_mode);
5098 /* Initialize lod_device::lod_md_root object reference */
5099 if (d->lod_md_root == NULL) {
5100 struct dt_object *root;
5101 struct lod_object *lroot;
5103 lu_root_fid(&info->lti_fid);
5104 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
5105 if (!IS_ERR(root)) {
5106 lroot = lod_dt_obj(root);
5108 spin_lock(&d->lod_lock);
5109 if (d->lod_md_root != NULL)
5110 dt_object_put(env, &d->lod_md_root->ldo_obj);
5111 d->lod_md_root = lroot;
5112 spin_unlock(&d->lod_lock);
5116 /* try inherit layout from the root object (fs default) when:
5117 * - parent does not have default layout; or
5118 * - parent has plain(v1/v3) default layout, and some attributes
5119 * are not specified in the default layout;
5121 if (d->lod_md_root != NULL && lod_need_inherit_more(lc, true)) {
5122 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds);
5125 if (lc->ldo_comp_cnt == 0) {
5126 lod_striping_from_default(lc, lds, child_mode);
5127 } else if (!lds->lds_def_striping_is_composite) {
5128 struct lod_layout_component *def_comp;
5130 LASSERT(!lc->ldo_is_composite);
5131 lod_comp = &lc->ldo_comp_entries[0];
5132 def_comp = &lds->lds_def_comp_entries[0];
5134 if (lod_comp->llc_stripe_count <= 0)
5135 lod_comp->llc_stripe_count =
5136 def_comp->llc_stripe_count;
5137 if (lod_comp->llc_stripe_size <= 0)
5138 lod_comp->llc_stripe_size =
5139 def_comp->llc_stripe_size;
5140 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT &&
5141 (!lod_comp->llc_pool || !lod_comp->llc_pool[0]))
5142 lod_comp->llc_stripe_offset =
5143 def_comp->llc_stripe_offset;
5144 if (lod_comp->llc_pool == NULL)
5145 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
5150 * fs default striping may not be explicitly set, or historically set
5151 * in config log, use them.
5153 if (lod_need_inherit_more(lc, false)) {
5154 if (lc->ldo_comp_cnt == 0) {
5155 rc = lod_alloc_comp_entries(lc, 0, 1);
5157 /* fail to allocate memory, will create a
5158 * non-striped file. */
5160 lc->ldo_is_composite = 0;
5161 lod_comp = &lc->ldo_comp_entries[0];
5162 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
5164 LASSERT(!lc->ldo_is_composite);
5165 lod_comp = &lc->ldo_comp_entries[0];
5166 desc = &d->lod_desc;
5167 lod_adjust_stripe_info(lod_comp, desc);
5173 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
5175 * Size initialization on late striping.
5177 * Propagate the size of a truncated object to a deferred striping.
5178 * This function handles a special case when truncate was done on a
5179 * non-striped object and now while the striping is being created
5180 * we can't lose that size, so we have to propagate it to the stripes
5183 * \param[in] env execution environment
5184 * \param[in] dt object
5185 * \param[in] th transaction handle
5187 * \retval 0 on success
5188 * \retval negative if failed
5190 static int lod_declare_init_size(const struct lu_env *env,
5191 struct dt_object *dt, struct thandle *th)
5193 struct dt_object *next = dt_object_child(dt);
5194 struct lod_object *lo = lod_dt_obj(dt);
5195 struct dt_object **objects = NULL;
5196 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
5197 uint64_t size, offs;
5198 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
5199 struct lu_extent size_ext;
5202 if (!lod_obj_is_striped(dt))
5205 rc = dt_attr_get(env, next, attr);
5206 LASSERT(attr->la_valid & LA_SIZE);
5210 size = attr->la_size;
5214 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
5215 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5216 struct lod_layout_component *lod_comp;
5217 struct lu_extent *extent;
5219 lod_comp = &lo->ldo_comp_entries[i];
5221 if (lod_comp->llc_stripe == NULL)
5224 extent = &lod_comp->llc_extent;
5225 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
5226 if (!lo->ldo_is_composite ||
5227 lu_extent_is_overlapped(extent, &size_ext)) {
5228 objects = lod_comp->llc_stripe;
5229 stripe_count = lod_comp->llc_stripe_count;
5230 stripe_size = lod_comp->llc_stripe_size;
5233 if (stripe_count == 0)
5236 LASSERT(objects != NULL && stripe_size != 0);
5237 /* ll_do_div64(a, b) returns a % b, and a = a / b */
5238 ll_do_div64(size, (__u64)stripe_size);
5239 stripe = ll_do_div64(size, (__u64)stripe_count);
5240 LASSERT(objects[stripe] != NULL);
5242 size = size * stripe_size;
5243 offs = attr->la_size;
5244 size += ll_do_div64(offs, stripe_size);
5246 attr->la_valid = LA_SIZE;
5247 attr->la_size = size;
5249 rc = lod_sub_declare_attr_set(env, objects[stripe],
5258 * Declare creation of striped object.
5260 * The function declares creation stripes for a regular object. The function
5261 * also declares whether the stripes will be created with non-zero size if
5262 * previously size was set non-zero on the master object. If object \a dt is
5263 * not local, then only fully defined striping can be applied in \a lovea.
5264 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
5267 * \param[in] env execution environment
5268 * \param[in] dt object
5269 * \param[in] attr attributes the stripes will be created with
5270 * \param[in] lovea a buffer containing striping description
5271 * \param[in] th transaction handle
5273 * \retval 0 on success
5274 * \retval negative if failed
5276 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
5277 struct lu_attr *attr,
5278 const struct lu_buf *lovea, struct thandle *th)
5280 struct lod_thread_info *info = lod_env_info(env);
5281 struct dt_object *next = dt_object_child(dt);
5282 struct lod_object *lo = lod_dt_obj(dt);
5286 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
5287 GOTO(out, rc = -ENOMEM);
5289 if (!dt_object_remote(next)) {
5290 /* choose OST and generate appropriate objects */
5291 rc = lod_prepare_create(env, lo, attr, lovea, th);
5296 * declare storage for striping data
5298 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
5300 /* LOD can not choose OST objects for remote objects, i.e.
5301 * stripes must be ready before that. Right now, it can only
5302 * happen during migrate, i.e. migrate process needs to create
5303 * remote regular file (mdd_migrate_create), then the migrate
5304 * process will provide stripeEA. */
5305 LASSERT(lovea != NULL);
5306 info->lti_buf = *lovea;
5309 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
5310 XATTR_NAME_LOV, 0, th);
5315 * if striping is created with local object's size > 0,
5316 * we have to propagate this size to specific object
5317 * the case is possible only when local object was created previously
5319 if (dt_object_exists(next))
5320 rc = lod_declare_init_size(env, dt, th);
5323 /* failed to create striping or to set initial size, let's reset
5324 * config so that others don't get confused */
5326 lod_striping_free(env, lo);
5332 * Whether subdirectories under \a dt should be created on MDTs by space QoS
5334 * If LMV_HASH_FLAG_SPACE is set on directory default layout, its subdirectories
5335 * should be created on MDT by space QoS.
5337 * \param[in] env execution environment
5338 * \param[in] dev lu device
5339 * \param[in] dt object
5341 * \retval 1 if directory should create subdir by space usage
5343 * \retval -ev if failed
5345 static inline int dt_object_qos_mkdir(const struct lu_env *env,
5346 struct lu_device *dev,
5347 struct dt_object *dt)
5349 struct lod_thread_info *info = lod_env_info(env);
5350 struct lu_object *obj;
5351 struct lod_object *lo;
5352 struct lmv_user_md *lmu;
5355 obj = lu_object_find_slice(env, dev, lu_object_fid(&dt->do_lu), NULL);
5357 return PTR_ERR(obj);
5359 lo = lu2lod_obj(obj);
5361 rc = lod_get_default_lmv_ea(env, lo);
5362 dt_object_put(env, dt);
5366 if (rc < (int)sizeof(*lmu))
5369 lmu = info->lti_ea_store;
5370 return !!(le32_to_cpu(lmu->lum_hash_type) & LMV_HASH_FLAG_SPACE);
5374 * Implementation of dt_object_operations::do_declare_create.
5376 * The method declares creation of a new object. If the object will be striped,
5377 * then helper functions are called to find FIDs for the stripes, declare
5378 * creation of the stripes and declare initialization of the striping
5379 * information to be stored in the master object.
5381 * \see dt_object_operations::do_declare_create() in the API description
5384 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
5385 struct lu_attr *attr,
5386 struct dt_allocation_hint *hint,
5387 struct dt_object_format *dof, struct thandle *th)
5389 struct dt_object *next = dt_object_child(dt);
5390 struct lod_object *lo = lod_dt_obj(dt);
5399 * first of all, we declare creation of local object
5401 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
5406 * it's lod_ah_init() that has decided the object will be striped
5408 if (dof->dof_type == DFT_REGULAR) {
5409 /* callers don't want stripes */
5410 /* XXX: all tricky interactions with ->ah_make_hint() decided
5411 * to use striping, then ->declare_create() behaving differently
5412 * should be cleaned */
5413 if (dof->u.dof_reg.striped != 0)
5414 rc = lod_declare_striped_create(env, dt, attr,
5416 } else if (dof->dof_type == DFT_DIR) {
5417 struct seq_server_site *ss;
5418 struct lu_buf buf = { NULL };
5419 struct lu_buf *lmu = NULL;
5421 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
5423 /* If the parent has default stripeEA, and client
5424 * did not find it before sending create request,
5425 * then MDT will return -EREMOTE, and client will
5426 * retrieve the default stripeEA and re-create the
5429 * Note: if dah_eadata != NULL, it means creating the
5430 * striped directory with specified stripeEA, then it
5431 * should ignore the default stripeEA */
5432 if (hint != NULL && hint->dah_eadata == NULL) {
5433 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
5434 GOTO(out, rc = -EREMOTE);
5436 if (lo->ldo_dir_stripe_offset == -1) {
5438 * child and parent should be in the same MDT,
5439 * but if parent has plain layout, it's allowed.
5441 if (hint->dah_parent &&
5442 dt_object_remote(hint->dah_parent)) {
5443 rc = dt_object_qos_mkdir(env,
5444 lo->ldo_obj.do_lu.lo_dev,
5447 GOTO(out, rc ? rc : -EREMOTE);
5449 } else if (lo->ldo_dir_stripe_offset !=
5451 struct lod_device *lod;
5452 struct lod_tgt_descs *ltd;
5453 struct lod_tgt_desc *tgt = NULL;
5454 bool found_mdt = false;
5457 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5458 ltd = &lod->lod_mdt_descs;
5459 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
5460 tgt = LTD_TGT(ltd, i);
5461 if (tgt->ltd_index ==
5462 lo->ldo_dir_stripe_offset) {
5468 /* If the MDT indicated by stripe_offset can be
5469 * found, then tell client to resend the create
5470 * request to the correct MDT, otherwise return
5471 * error to client */
5473 GOTO(out, rc = -EREMOTE);
5475 GOTO(out, rc = -EINVAL);
5477 } else if (hint && hint->dah_eadata) {
5479 lmu->lb_buf = (void *)hint->dah_eadata;
5480 lmu->lb_len = hint->dah_eadata_len;
5483 rc = lod_declare_dir_striping_create(env, dt, attr, lmu, dof,
5487 /* failed to create striping or to set initial size, let's reset
5488 * config so that others don't get confused */
5490 lod_striping_free(env, lo);
5495 * Generate component ID for new created component.
5497 * \param[in] lo LOD object
5498 * \param[in] comp_idx index of ldo_comp_entries
5500 * \retval component ID on success
5501 * \retval LCME_ID_INVAL on failure
5503 static __u32 lod_gen_component_id(struct lod_object *lo,
5504 int mirror_id, int comp_idx)
5506 struct lod_layout_component *lod_comp;
5507 __u32 id, start, end;
5510 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
5512 lod_obj_inc_layout_gen(lo);
5513 id = lo->ldo_layout_gen;
5514 if (likely(id <= SEQ_ID_MAX))
5515 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
5517 /* Layout generation wraps, need to check collisions. */
5518 start = id & SEQ_ID_MASK;
5521 for (id = start; id <= end; id++) {
5522 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5523 lod_comp = &lo->ldo_comp_entries[i];
5524 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
5527 /* Found the ununsed ID */
5528 if (i == lo->ldo_comp_cnt)
5529 RETURN(pflr_id(mirror_id, id));
5531 if (end == LCME_ID_MAX) {
5533 end = min(lo->ldo_layout_gen & LCME_ID_MASK,
5534 (__u32)(LCME_ID_MAX - 1));
5538 RETURN(LCME_ID_INVAL);
5542 * Creation of a striped regular object.
5544 * The function is called to create the stripe objects for a regular
5545 * striped file. This can happen at the initial object creation or
5546 * when the caller asks LOD to do so using ->do_xattr_set() method
5547 * (so called late striping). Notice all the information are already
5548 * prepared in the form of the list of objects (ldo_stripe field).
5549 * This is done during declare phase.
5551 * \param[in] env execution environment
5552 * \param[in] dt object
5553 * \param[in] attr attributes the stripes will be created with
5554 * \param[in] dof format of stripes (see OSD API description)
5555 * \param[in] th transaction handle
5557 * \retval 0 on success
5558 * \retval negative if failed
5560 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
5561 struct lu_attr *attr, struct dt_object_format *dof,
5564 struct lod_layout_component *lod_comp;
5565 struct lod_object *lo = lod_dt_obj(dt);
5570 LASSERT((lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL) ||
5571 lo->ldo_is_foreign);
5573 mirror_id = 0; /* non-flr file's mirror_id is 0 */
5574 if (lo->ldo_mirror_count > 1) {
5575 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5576 lod_comp = &lo->ldo_comp_entries[i];
5577 if (lod_comp->llc_id != LCME_ID_INVAL &&
5578 mirror_id_of(lod_comp->llc_id) > mirror_id)
5579 mirror_id = mirror_id_of(lod_comp->llc_id);
5583 /* create all underlying objects */
5584 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5585 lod_comp = &lo->ldo_comp_entries[i];
5587 if (lod_comp->llc_id == LCME_ID_INVAL) {
5588 /* only the component of FLR layout with more than 1
5589 * mirror has mirror ID in its component ID.
5591 if (lod_comp->llc_extent.e_start == 0 &&
5592 lo->ldo_mirror_count > 1)
5595 lod_comp->llc_id = lod_gen_component_id(lo,
5597 if (lod_comp->llc_id == LCME_ID_INVAL)
5598 GOTO(out, rc = -ERANGE);
5601 if (lod_comp_inited(lod_comp))
5604 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
5605 lod_comp_set_init(lod_comp);
5607 if (lov_pattern(lod_comp->llc_pattern) == LOV_PATTERN_MDT)
5608 lod_comp_set_init(lod_comp);
5610 if (lod_comp->llc_stripe == NULL)
5613 LASSERT(lod_comp->llc_stripe_count);
5614 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
5615 struct dt_object *object = lod_comp->llc_stripe[j];
5616 LASSERT(object != NULL);
5617 rc = lod_sub_create(env, object, attr, NULL, dof, th);
5621 lod_comp_set_init(lod_comp);
5624 rc = lod_fill_mirrors(lo);
5628 rc = lod_generate_and_set_lovea(env, lo, th);
5632 lo->ldo_comp_cached = 1;
5636 lod_striping_free(env, lo);
5640 static inline bool lod_obj_is_dom(struct dt_object *dt)
5642 struct lod_object *lo = lod_dt_obj(dt);
5644 if (!dt_object_exists(dt_object_child(dt)))
5647 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
5650 if (!lo->ldo_comp_cnt)
5653 return (lov_pattern(lo->ldo_comp_entries[0].llc_pattern) ==
5658 * Implementation of dt_object_operations::do_create.
5660 * If any of preceeding methods (like ->do_declare_create(),
5661 * ->do_ah_init(), etc) chose to create a striped object,
5662 * then this method will create the master and the stripes.
5664 * \see dt_object_operations::do_create() in the API description for details.
5666 static int lod_create(const struct lu_env *env, struct dt_object *dt,
5667 struct lu_attr *attr, struct dt_allocation_hint *hint,
5668 struct dt_object_format *dof, struct thandle *th)
5673 /* create local object */
5674 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
5678 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
5679 (lod_obj_is_striped(dt) || lod_obj_is_dom(dt)) &&
5680 dof->u.dof_reg.striped != 0) {
5681 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
5682 rc = lod_striped_create(env, dt, attr, dof, th);
5689 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
5690 struct dt_object *dt, struct thandle *th,
5691 int comp_idx, int stripe_idx,
5692 struct lod_obj_stripe_cb_data *data)
5694 if (data->locd_declare)
5695 return lod_sub_declare_destroy(env, dt, th);
5696 else if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
5697 stripe_idx == cfs_fail_val)
5698 return lod_sub_destroy(env, dt, th);
5704 * Implementation of dt_object_operations::do_declare_destroy.
5706 * If the object is a striped directory, then the function declares reference
5707 * removal from the master object (this is an index) to the stripes and declares
5708 * destroy of all the stripes. In all the cases, it declares an intention to
5709 * destroy the object itself.
5711 * \see dt_object_operations::do_declare_destroy() in the API description
5714 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
5717 struct dt_object *next = dt_object_child(dt);
5718 struct lod_object *lo = lod_dt_obj(dt);
5719 struct lod_thread_info *info = lod_env_info(env);
5720 struct dt_object *stripe;
5721 char *stripe_name = info->lti_key;
5727 * load striping information, notice we don't do this when object
5728 * is being initialized as we don't need this information till
5729 * few specific cases like destroy, chown
5731 rc = lod_striping_load(env, lo);
5735 /* declare destroy for all underlying objects */
5736 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5737 rc = next->do_ops->do_index_try(env, next,
5738 &dt_directory_features);
5742 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5743 stripe = lo->ldo_stripe[i];
5747 rc = lod_sub_declare_ref_del(env, next, th);
5751 snprintf(stripe_name, sizeof(info->lti_key),
5753 PFID(lu_object_fid(&stripe->do_lu)), i);
5754 rc = lod_sub_declare_delete(env, next,
5755 (const struct dt_key *)stripe_name, th);
5762 * we declare destroy for the local object
5764 rc = lod_sub_declare_destroy(env, next, th);
5768 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
5769 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
5772 if (!lod_obj_is_striped(dt))
5775 /* declare destroy all striped objects */
5776 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5777 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5778 stripe = lo->ldo_stripe[i];
5782 if (!dt_object_exists(stripe))
5785 rc = lod_sub_declare_ref_del(env, stripe, th);
5789 rc = lod_sub_declare_destroy(env, stripe, th);
5794 struct lod_obj_stripe_cb_data data = { { 0 } };
5796 data.locd_declare = true;
5797 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
5798 rc = lod_obj_for_each_stripe(env, lo, th, &data);
5805 * Implementation of dt_object_operations::do_destroy.
5807 * If the object is a striped directory, then the function removes references
5808 * from the master object (this is an index) to the stripes and destroys all
5809 * the stripes. In all the cases, the function destroys the object itself.
5811 * \see dt_object_operations::do_destroy() in the API description for details.
5813 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
5816 struct dt_object *next = dt_object_child(dt);
5817 struct lod_object *lo = lod_dt_obj(dt);
5818 struct lod_thread_info *info = lod_env_info(env);
5819 char *stripe_name = info->lti_key;
5820 struct dt_object *stripe;
5826 /* destroy sub-stripe of master object */
5827 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5828 rc = next->do_ops->do_index_try(env, next,
5829 &dt_directory_features);
5833 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5834 stripe = lo->ldo_stripe[i];
5838 rc = lod_sub_ref_del(env, next, th);
5842 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
5843 PFID(lu_object_fid(&stripe->do_lu)), i);
5845 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
5846 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
5847 PFID(lu_object_fid(&stripe->do_lu)));
5849 rc = lod_sub_delete(env, next,
5850 (const struct dt_key *)stripe_name, th);
5856 rc = lod_sub_destroy(env, next, th);
5860 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
5861 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
5864 if (!lod_obj_is_striped(dt))
5867 /* destroy all striped objects */
5868 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5869 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5870 stripe = lo->ldo_stripe[i];
5874 if (!dt_object_exists(stripe))
5877 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
5878 i == cfs_fail_val) {
5879 dt_write_lock(env, stripe, MOR_TGT_CHILD);
5880 rc = lod_sub_ref_del(env, stripe, th);
5881 dt_write_unlock(env, stripe);
5885 rc = lod_sub_destroy(env, stripe, th);
5891 struct lod_obj_stripe_cb_data data = { { 0 } };
5893 data.locd_declare = false;
5894 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
5895 rc = lod_obj_for_each_stripe(env, lo, th, &data);
5902 * Implementation of dt_object_operations::do_declare_ref_add.
5904 * \see dt_object_operations::do_declare_ref_add() in the API description
5907 static int lod_declare_ref_add(const struct lu_env *env,
5908 struct dt_object *dt, struct thandle *th)
5910 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
5914 * Implementation of dt_object_operations::do_ref_add.
5916 * \see dt_object_operations::do_ref_add() in the API description for details.
5918 static int lod_ref_add(const struct lu_env *env,
5919 struct dt_object *dt, struct thandle *th)
5921 return lod_sub_ref_add(env, dt_object_child(dt), th);
5925 * Implementation of dt_object_operations::do_declare_ref_del.
5927 * \see dt_object_operations::do_declare_ref_del() in the API description
5930 static int lod_declare_ref_del(const struct lu_env *env,
5931 struct dt_object *dt, struct thandle *th)
5933 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
5937 * Implementation of dt_object_operations::do_ref_del
5939 * \see dt_object_operations::do_ref_del() in the API description for details.
5941 static int lod_ref_del(const struct lu_env *env,
5942 struct dt_object *dt, struct thandle *th)
5944 return lod_sub_ref_del(env, dt_object_child(dt), th);
5948 * Implementation of dt_object_operations::do_object_sync.
5950 * \see dt_object_operations::do_object_sync() in the API description
5953 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
5954 __u64 start, __u64 end)
5956 return dt_object_sync(env, dt_object_child(dt), start, end);
5960 * Implementation of dt_object_operations::do_object_unlock.
5962 * Used to release LDLM lock(s).
5964 * \see dt_object_operations::do_object_unlock() in the API description
5967 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
5968 struct ldlm_enqueue_info *einfo,
5969 union ldlm_policy_data *policy)
5971 struct lod_object *lo = lod_dt_obj(dt);
5972 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
5973 int slave_locks_size;
5977 if (slave_locks == NULL)
5980 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
5981 /* Note: for remote lock for single stripe dir, MDT will cancel
5982 * the lock by lockh directly */
5983 LASSERT(!dt_object_remote(dt_object_child(dt)));
5985 /* locks were unlocked in MDT layer */
5986 for (i = 0; i < slave_locks->ha_count; i++)
5987 LASSERT(!lustre_handle_is_used(&slave_locks->ha_handles[i]));
5990 * NB, ha_count may not equal to ldo_dir_stripe_count, because dir
5991 * layout may change, e.g., shrink dir layout after migration.
5993 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5994 if (lo->ldo_stripe[i])
5995 dt_invalidate(env, lo->ldo_stripe[i]);
5998 slave_locks_size = offsetof(typeof(*slave_locks),
5999 ha_handles[slave_locks->ha_count]);
6000 OBD_FREE(slave_locks, slave_locks_size);
6001 einfo->ei_cbdata = NULL;
6007 * Implementation of dt_object_operations::do_object_lock.
6009 * Used to get LDLM lock on the non-striped and striped objects.
6011 * \see dt_object_operations::do_object_lock() in the API description
6014 static int lod_object_lock(const struct lu_env *env,
6015 struct dt_object *dt,
6016 struct lustre_handle *lh,
6017 struct ldlm_enqueue_info *einfo,
6018 union ldlm_policy_data *policy)
6020 struct lod_object *lo = lod_dt_obj(dt);
6021 int slave_locks_size;
6022 struct lustre_handle_array *slave_locks = NULL;
6027 /* remote object lock */
6028 if (!einfo->ei_enq_slave) {
6029 LASSERT(dt_object_remote(dt));
6030 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
6034 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
6037 rc = lod_striping_load(env, lo);
6042 if (lo->ldo_dir_stripe_count <= 1)
6045 slave_locks_size = offsetof(typeof(*slave_locks),
6046 ha_handles[lo->ldo_dir_stripe_count]);
6047 /* Freed in lod_object_unlock */
6048 OBD_ALLOC(slave_locks, slave_locks_size);
6051 slave_locks->ha_count = lo->ldo_dir_stripe_count;
6053 /* striped directory lock */
6054 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6055 struct lustre_handle lockh;
6056 struct ldlm_res_id *res_id;
6057 struct dt_object *stripe;
6059 stripe = lo->ldo_stripe[i];
6063 res_id = &lod_env_info(env)->lti_res_id;
6064 fid_build_reg_res_name(lu_object_fid(&stripe->do_lu), res_id);
6065 einfo->ei_res_id = res_id;
6067 if (dt_object_remote(stripe)) {
6068 set_bit(i, (void *)slave_locks->ha_map);
6069 rc = dt_object_lock(env, stripe, &lockh, einfo, policy);
6071 struct ldlm_namespace *ns = einfo->ei_namespace;
6072 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
6073 ldlm_completion_callback completion = einfo->ei_cb_cp;
6074 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
6076 if (einfo->ei_mode == LCK_PW ||
6077 einfo->ei_mode == LCK_EX)
6078 dlmflags |= LDLM_FL_COS_INCOMPAT;
6080 LASSERT(ns != NULL);
6081 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS,
6082 policy, einfo->ei_mode,
6083 &dlmflags, blocking,
6085 NULL, 0, LVB_T_NONE,
6090 ldlm_lock_decref_and_cancel(
6091 &slave_locks->ha_handles[i],
6093 OBD_FREE(slave_locks, slave_locks_size);
6096 slave_locks->ha_handles[i] = lockh;
6098 einfo->ei_cbdata = slave_locks;
6104 * Implementation of dt_object_operations::do_invalidate.
6106 * \see dt_object_operations::do_invalidate() in the API description for details
6108 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
6110 return dt_invalidate(env, dt_object_child(dt));
6113 static int lod_declare_instantiate_components(const struct lu_env *env,
6114 struct lod_object *lo, struct thandle *th)
6116 struct lod_thread_info *info = lod_env_info(env);
6121 LASSERT(info->lti_count < lo->ldo_comp_cnt);
6123 for (i = 0; i < info->lti_count; i++) {
6124 rc = lod_qos_prep_create(env, lo, NULL, th,
6125 info->lti_comp_idx[i]);
6131 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6132 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
6133 &info->lti_buf, XATTR_NAME_LOV, 0, th);
6139 static int lod_declare_update_plain(const struct lu_env *env,
6140 struct lod_object *lo, struct layout_intent *layout,
6141 const struct lu_buf *buf, struct thandle *th)
6143 struct lod_thread_info *info = lod_env_info(env);
6144 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6145 struct lod_layout_component *lod_comp;
6146 struct lov_comp_md_v1 *comp_v1 = NULL;
6147 bool replay = false;
6151 LASSERT(lo->ldo_flr_state == LCM_FL_NONE);
6154 * In case the client is passing lovea, which only happens during
6155 * the replay of layout intent write RPC for now, we may need to
6156 * parse the lovea and apply new layout configuration.
6158 if (buf && buf->lb_len) {
6159 struct lov_user_md_v1 *v1 = buf->lb_buf;
6161 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
6162 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
6163 LOV_MAGIC_COMP_V1)) {
6164 CERROR("%s: the replay buffer of layout extend "
6165 "(magic %#x) does not contain expected "
6166 "composite layout.\n",
6167 lod2obd(d)->obd_name, v1->lmm_magic);
6168 GOTO(out, rc = -EINVAL);
6171 rc = lod_use_defined_striping(env, lo, buf);
6174 lo->ldo_comp_cached = 1;
6176 rc = lod_get_lov_ea(env, lo);
6179 /* old on-disk EA is stored in info->lti_buf */
6180 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
6183 /* non replay path */
6184 rc = lod_striping_load(env, lo);
6189 /* Make sure defined layout covers the requested write range. */
6190 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
6191 if (lo->ldo_comp_cnt > 1 &&
6192 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
6193 lod_comp->llc_extent.e_end < layout->li_extent.e_end) {
6194 CDEBUG(replay ? D_ERROR : D_LAYOUT,
6195 "%s: the defined layout [0, %#llx) does not covers "
6196 "the write range "DEXT"\n",
6197 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
6198 PEXT(&layout->li_extent));
6199 GOTO(out, rc = -EINVAL);
6202 CDEBUG(D_LAYOUT, "%s: "DFID": instantiate components "DEXT"\n",
6203 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
6204 PEXT(&layout->li_extent));
6207 * Iterate ld->ldo_comp_entries, find the component whose extent under
6208 * the write range and not instantianted.
6210 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6211 lod_comp = &lo->ldo_comp_entries[i];
6213 if (lod_comp->llc_extent.e_start >= layout->li_extent.e_end)
6217 if (lod_comp_inited(lod_comp))
6221 * In replay path, lod_comp is the EA passed by
6222 * client replay buffer, comp_v1 is the pre-recovery
6223 * on-disk EA, we'd sift out those components which
6224 * were init-ed in the on-disk EA.
6226 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
6231 * this component hasn't instantiated in normal path, or during
6232 * replay it needs replay the instantiation.
6235 /* A released component is being extended */
6236 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
6237 GOTO(out, rc = -EINVAL);
6239 LASSERT(info->lti_comp_idx != NULL);
6240 info->lti_comp_idx[info->lti_count++] = i;
6243 if (info->lti_count == 0)
6246 lod_obj_inc_layout_gen(lo);
6247 rc = lod_declare_instantiate_components(env, lo, th);
6250 lod_striping_free(env, lo);
6254 static inline int lod_comp_index(struct lod_object *lo,
6255 struct lod_layout_component *lod_comp)
6257 LASSERT(lod_comp >= lo->ldo_comp_entries &&
6258 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
6260 return lod_comp - lo->ldo_comp_entries;
6264 * Stale other mirrors by writing extent.
6266 static void lod_stale_components(struct lod_object *lo, int primary,
6267 struct lu_extent *extent)
6269 struct lod_layout_component *pri_comp, *lod_comp;
6272 /* The writing extent decides which components in the primary
6273 * are affected... */
6274 CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
6275 lod_foreach_mirror_comp(pri_comp, lo, primary) {
6276 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
6279 CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
6280 lod_comp_index(lo, pri_comp),
6281 PEXT(&pri_comp->llc_extent));
6283 for (i = 0; i < lo->ldo_mirror_count; i++) {
6287 /* ... and then stale other components that are
6288 * overlapping with primary components */
6289 lod_foreach_mirror_comp(lod_comp, lo, i) {
6290 if (!lu_extent_is_overlapped(
6291 &pri_comp->llc_extent,
6292 &lod_comp->llc_extent))
6295 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
6296 i, lod_comp_index(lo, lod_comp));
6298 lod_comp->llc_flags |= LCME_FL_STALE;
6299 lo->ldo_mirrors[i].lme_stale = 1;
6306 * check an OST's availability
6307 * \param[in] env execution environment
6308 * \param[in] lo lod object
6309 * \param[in] dt dt object
6310 * \param[in] index mirror index
6312 * \retval negative if failed
6313 * \retval 1 if \a dt is available
6314 * \retval 0 if \a dt is not available
6316 static inline int lod_check_ost_avail(const struct lu_env *env,
6317 struct lod_object *lo,
6318 struct dt_object *dt, int index)
6320 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6321 struct lod_tgt_desc *ost;
6323 int type = LU_SEQ_RANGE_OST;
6326 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &idx, &type);
6328 CERROR("%s: can't locate "DFID":rc = %d\n",
6329 lod2obd(lod)->obd_name, PFID(lu_object_fid(&dt->do_lu)),
6334 ost = OST_TGT(lod, idx);
6335 if (ost->ltd_statfs.os_state &
6336 (OS_STATE_READONLY | OS_STATE_ENOSPC | OS_STATE_ENOINO |
6337 OS_STATE_NOPRECREATE) ||
6338 ost->ltd_active == 0) {
6339 CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail, rc = %d\n",
6340 PFID(lod_object_fid(lo)), index, idx, rc);
6348 * Pick primary mirror for write
6349 * \param[in] env execution environment
6350 * \param[in] lo object
6351 * \param[in] extent write range
6353 static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo,
6354 struct lu_extent *extent)
6356 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6357 unsigned int seq = 0;
6358 struct lod_layout_component *lod_comp;
6360 int picked = -1, second_pick = -1, third_pick = -1;
6363 if (OBD_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
6364 get_random_bytes(&seq, sizeof(seq));
6365 seq %= lo->ldo_mirror_count;
6369 * Pick a mirror as the primary, and check the availability of OSTs.
6371 * This algo can be revised later after knowing the topology of
6374 lod_qos_statfs_update(env, lod);
6375 for (i = 0; i < lo->ldo_mirror_count; i++) {
6376 bool ost_avail = true;
6377 int index = (i + seq) % lo->ldo_mirror_count;
6379 if (lo->ldo_mirrors[index].lme_stale) {
6380 CDEBUG(D_LAYOUT, DFID": mirror %d stale\n",
6381 PFID(lod_object_fid(lo)), index);
6385 /* 2nd pick is for the primary mirror containing unavail OST */
6386 if (lo->ldo_mirrors[index].lme_primary && second_pick < 0)
6387 second_pick = index;
6389 /* 3rd pick is for non-primary mirror containing unavail OST */
6390 if (second_pick < 0 && third_pick < 0)
6394 * we found a non-primary 1st pick, we'd like to find a
6395 * potential pirmary mirror.
6397 if (picked >= 0 && !lo->ldo_mirrors[index].lme_primary)
6400 /* check the availability of OSTs */
6401 lod_foreach_mirror_comp(lod_comp, lo, index) {
6402 if (!lod_comp_inited(lod_comp) || !lod_comp->llc_stripe)
6405 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6406 struct dt_object *dt = lod_comp->llc_stripe[j];
6408 rc = lod_check_ost_avail(env, lo, dt, index);
6415 } /* for all dt object in one component */
6418 } /* for all components in a mirror */
6421 * the OSTs where allocated objects locates in the components
6422 * of the mirror are available.
6427 /* this mirror has all OSTs available */
6431 * primary with all OSTs are available, this is the perfect
6434 if (lo->ldo_mirrors[index].lme_primary)
6436 } /* for all mirrors */
6438 /* failed to pick a sound mirror, lower our expectation */
6440 picked = second_pick;
6442 picked = third_pick;
6449 static int lod_prepare_resync_mirror(const struct lu_env *env,
6450 struct lod_object *lo,
6453 struct lod_thread_info *info = lod_env_info(env);
6454 struct lod_layout_component *lod_comp;
6455 bool neg = !!(MIRROR_ID_NEG & mirror_id);
6458 mirror_id &= ~MIRROR_ID_NEG;
6460 for (i = 0; i < lo->ldo_mirror_count; i++) {
6461 if ((!neg && lo->ldo_mirrors[i].lme_id != mirror_id) ||
6462 (neg && lo->ldo_mirrors[i].lme_id == mirror_id))
6465 lod_foreach_mirror_comp(lod_comp, lo, i) {
6466 if (lod_comp_inited(lod_comp))
6469 info->lti_comp_idx[info->lti_count++] =
6470 lod_comp_index(lo, lod_comp);
6478 * figure out the components should be instantiated for resync.
6480 static int lod_prepare_resync(const struct lu_env *env, struct lod_object *lo,
6481 struct lu_extent *extent)
6483 struct lod_thread_info *info = lod_env_info(env);
6484 struct lod_layout_component *lod_comp;
6485 unsigned int need_sync = 0;
6489 DFID": instantiate all stale components in "DEXT"\n",
6490 PFID(lod_object_fid(lo)), PEXT(extent));
6493 * instantiate all components within this extent, even non-stale
6496 for (i = 0; i < lo->ldo_mirror_count; i++) {
6497 if (!lo->ldo_mirrors[i].lme_stale)
6500 lod_foreach_mirror_comp(lod_comp, lo, i) {
6501 if (!lu_extent_is_overlapped(extent,
6502 &lod_comp->llc_extent))
6507 if (lod_comp_inited(lod_comp))
6510 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
6511 i, lod_comp_index(lo, lod_comp));
6512 info->lti_comp_idx[info->lti_count++] =
6513 lod_comp_index(lo, lod_comp);
6517 return need_sync ? 0 : -EALREADY;
6520 static int lod_declare_update_rdonly(const struct lu_env *env,
6521 struct lod_object *lo, struct md_layout_change *mlc,
6524 struct lod_thread_info *info = lod_env_info(env);
6525 struct lu_attr *layout_attr = &info->lti_layout_attr;
6526 struct lod_layout_component *lod_comp;
6527 struct lu_extent extent = { 0 };
6531 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
6532 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
6533 mlc->mlc_opc == MD_LAYOUT_RESYNC);
6534 LASSERT(lo->ldo_mirror_count > 0);
6536 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
6537 struct layout_intent *layout = mlc->mlc_intent;
6540 extent = layout->li_extent;
6541 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
6542 PFID(lod_object_fid(lo)), PEXT(&extent));
6544 picked = lod_primary_pick(env, lo, &extent);
6548 CDEBUG(D_LAYOUT, DFID": picked mirror id %u as primary\n",
6549 PFID(lod_object_fid(lo)),
6550 lo->ldo_mirrors[picked].lme_id);
6552 if (layout->li_opc == LAYOUT_INTENT_TRUNC) {
6554 * trunc transfers [0, size) in the intent extent, we'd
6555 * stale components overlapping [size, eof).
6557 extent.e_start = extent.e_end;
6558 extent.e_end = OBD_OBJECT_EOF;
6561 /* stale overlapping components from other mirrors */
6562 lod_stale_components(lo, picked, &extent);
6564 /* restore truncate intent extent */
6565 if (layout->li_opc == LAYOUT_INTENT_TRUNC)
6566 extent.e_end = extent.e_start;
6568 /* instantiate components for the picked mirror, start from 0 */
6571 lod_foreach_mirror_comp(lod_comp, lo, picked) {
6572 if (!lu_extent_is_overlapped(&extent,
6573 &lod_comp->llc_extent))
6576 if (lod_comp_inited(lod_comp))
6579 info->lti_comp_idx[info->lti_count++] =
6580 lod_comp_index(lo, lod_comp);
6583 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
6584 } else { /* MD_LAYOUT_RESYNC */
6588 * could contain multiple non-stale mirrors, so we need to
6589 * prep uninited all components assuming any non-stale mirror
6590 * could be picked as the primary mirror.
6592 if (mlc->mlc_mirror_id == 0) {
6594 for (i = 0; i < lo->ldo_mirror_count; i++) {
6595 if (lo->ldo_mirrors[i].lme_stale)
6598 lod_foreach_mirror_comp(lod_comp, lo, i) {
6599 if (!lod_comp_inited(lod_comp))
6603 lod_comp->llc_extent.e_end)
6605 lod_comp->llc_extent.e_end;
6608 rc = lod_prepare_resync(env, lo, &extent);
6612 /* mirror write, try to init its all components */
6613 rc = lod_prepare_resync_mirror(env, lo,
6614 mlc->mlc_mirror_id);
6619 /* change the file state to SYNC_PENDING */
6620 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
6623 /* Reset the layout version once it's becoming too large.
6624 * This way it can make sure that the layout version is
6625 * monotonously increased in this writing era. */
6626 lod_obj_inc_layout_gen(lo);
6627 if (lo->ldo_layout_gen > (LCME_ID_MAX >> 1)) {
6628 __u32 layout_version;
6630 get_random_bytes(&layout_version, sizeof(layout_version));
6631 lo->ldo_layout_gen = layout_version & 0xffff;
6634 rc = lod_declare_instantiate_components(env, lo, th);
6638 layout_attr->la_valid = LA_LAYOUT_VERSION;
6639 layout_attr->la_layout_version = 0; /* set current version */
6640 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
6641 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
6642 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
6648 lod_striping_free(env, lo);
6652 static int lod_declare_update_write_pending(const struct lu_env *env,
6653 struct lod_object *lo, struct md_layout_change *mlc,
6656 struct lod_thread_info *info = lod_env_info(env);
6657 struct lu_attr *layout_attr = &info->lti_layout_attr;
6658 struct lod_layout_component *lod_comp;
6659 struct lu_extent extent = { 0 };
6665 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
6666 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
6667 mlc->mlc_opc == MD_LAYOUT_RESYNC);
6669 /* look for the primary mirror */
6670 for (i = 0; i < lo->ldo_mirror_count; i++) {
6671 if (lo->ldo_mirrors[i].lme_stale)
6674 LASSERTF(primary < 0, DFID " has multiple primary: %u / %u",
6675 PFID(lod_object_fid(lo)),
6676 lo->ldo_mirrors[i].lme_id,
6677 lo->ldo_mirrors[primary].lme_id);
6682 CERROR(DFID ": doesn't have a primary mirror\n",
6683 PFID(lod_object_fid(lo)));
6684 GOTO(out, rc = -ENODATA);
6687 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
6688 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
6690 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
6692 /* for LAYOUT_WRITE opc, it has to do the following operations:
6693 * 1. stale overlapping componets from stale mirrors;
6694 * 2. instantiate components of the primary mirror;
6695 * 3. transfter layout version to all objects of the primary;
6697 * for LAYOUT_RESYNC opc, it will do:
6698 * 1. instantiate components of all stale mirrors;
6699 * 2. transfer layout version to all objects to close write era. */
6701 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
6702 LASSERT(mlc->mlc_intent != NULL);
6704 extent = mlc->mlc_intent->li_extent;
6706 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
6707 PFID(lod_object_fid(lo)), PEXT(&extent));
6709 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC) {
6711 * trunc transfers [0, size) in the intent extent, we'd
6712 * stale components overlapping [size, eof).
6714 extent.e_start = extent.e_end;
6715 extent.e_end = OBD_OBJECT_EOF;
6717 /* 1. stale overlapping components */
6718 lod_stale_components(lo, primary, &extent);
6720 /* 2. find out the components need instantiating.
6721 * instantiate [0, mlc->mlc_intent->e_end) */
6723 /* restore truncate intent extent */
6724 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC)
6725 extent.e_end = extent.e_start;
6728 lod_foreach_mirror_comp(lod_comp, lo, primary) {
6729 if (!lu_extent_is_overlapped(&extent,
6730 &lod_comp->llc_extent))
6733 if (lod_comp_inited(lod_comp))
6736 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
6737 primary, lod_comp_index(lo, lod_comp));
6738 info->lti_comp_idx[info->lti_count++] =
6739 lod_comp_index(lo, lod_comp);
6741 } else { /* MD_LAYOUT_RESYNC */
6742 if (mlc->mlc_mirror_id == 0) {
6744 lod_foreach_mirror_comp(lod_comp, lo, primary) {
6745 if (!lod_comp_inited(lod_comp))
6748 extent.e_end = lod_comp->llc_extent.e_end;
6751 rc = lod_prepare_resync(env, lo, &extent);
6755 /* mirror write, try to init its all components */
6756 rc = lod_prepare_resync_mirror(env, lo,
6757 mlc->mlc_mirror_id);
6762 /* change the file state to SYNC_PENDING */
6763 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
6766 rc = lod_declare_instantiate_components(env, lo, th);
6770 /* 3. transfer layout version to OST objects.
6771 * transfer new layout version to OST objects so that stale writes
6772 * can be denied. It also ends an era of writing by setting
6773 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
6774 * send write RPC; only resync RPCs could do it. */
6775 layout_attr->la_valid = LA_LAYOUT_VERSION;
6776 layout_attr->la_layout_version = 0; /* set current version */
6777 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
6778 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
6779 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
6783 lod_obj_inc_layout_gen(lo);
6786 lod_striping_free(env, lo);
6790 static int lod_declare_update_sync_pending(const struct lu_env *env,
6791 struct lod_object *lo, struct md_layout_change *mlc,
6794 struct lod_thread_info *info = lod_env_info(env);
6795 unsigned sync_components = 0;
6796 unsigned resync_components = 0;
6801 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
6802 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
6803 mlc->mlc_opc == MD_LAYOUT_WRITE);
6805 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
6806 PFID(lod_object_fid(lo)), mlc->mlc_opc);
6808 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
6809 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
6810 PFID(lod_object_fid(lo)));
6812 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
6813 return lod_declare_update_write_pending(env, lo, mlc, th);
6816 /* MD_LAYOUT_RESYNC_DONE */
6818 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6819 struct lod_layout_component *lod_comp;
6822 lod_comp = &lo->ldo_comp_entries[i];
6824 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
6829 for (j = 0; j < mlc->mlc_resync_count; j++) {
6830 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
6833 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
6834 lod_comp->llc_flags &= ~LCME_FL_STALE;
6835 resync_components++;
6841 for (i = 0; i < mlc->mlc_resync_count; i++) {
6842 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
6845 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
6846 "or already synced\n", PFID(lod_object_fid(lo)),
6847 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
6848 GOTO(out, rc = -EINVAL);
6851 if (!sync_components || (mlc->mlc_resync_count && !resync_components)) {
6852 CDEBUG(D_LAYOUT, DFID": no mirror in sync\n",
6853 PFID(lod_object_fid(lo)));
6855 /* tend to return an error code here to prevent
6856 * the MDT from setting SoM attribute */
6857 GOTO(out, rc = -EINVAL);
6860 CDEBUG(D_LAYOUT, DFID": resynced %u/%zu components\n",
6861 PFID(lod_object_fid(lo)),
6862 resync_components, mlc->mlc_resync_count);
6864 lo->ldo_flr_state = LCM_FL_RDONLY;
6865 lod_obj_inc_layout_gen(lo);
6867 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6868 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
6869 &info->lti_buf, XATTR_NAME_LOV, 0, th);
6874 lod_striping_free(env, lo);
6878 static int lod_declare_layout_change(const struct lu_env *env,
6879 struct dt_object *dt, struct md_layout_change *mlc,
6882 struct lod_thread_info *info = lod_env_info(env);
6883 struct lod_object *lo = lod_dt_obj(dt);
6887 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
6888 dt_object_remote(dt_object_child(dt)))
6891 rc = lod_striping_load(env, lo);
6895 LASSERT(lo->ldo_comp_cnt > 0);
6897 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
6901 switch (lo->ldo_flr_state) {
6903 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
6907 rc = lod_declare_update_rdonly(env, lo, mlc, th);
6909 case LCM_FL_WRITE_PENDING:
6910 rc = lod_declare_update_write_pending(env, lo, mlc, th);
6912 case LCM_FL_SYNC_PENDING:
6913 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
6924 * Instantiate layout component objects which covers the intent write offset.
6926 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
6927 struct md_layout_change *mlc, struct thandle *th)
6929 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
6930 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
6931 struct lod_object *lo = lod_dt_obj(dt);
6934 rc = lod_striped_create(env, dt, attr, NULL, th);
6935 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
6936 layout_attr->la_layout_version |= lo->ldo_layout_gen;
6937 rc = lod_attr_set(env, dt, layout_attr, th);
6943 struct dt_object_operations lod_obj_ops = {
6944 .do_read_lock = lod_read_lock,
6945 .do_write_lock = lod_write_lock,
6946 .do_read_unlock = lod_read_unlock,
6947 .do_write_unlock = lod_write_unlock,
6948 .do_write_locked = lod_write_locked,
6949 .do_attr_get = lod_attr_get,
6950 .do_declare_attr_set = lod_declare_attr_set,
6951 .do_attr_set = lod_attr_set,
6952 .do_xattr_get = lod_xattr_get,
6953 .do_declare_xattr_set = lod_declare_xattr_set,
6954 .do_xattr_set = lod_xattr_set,
6955 .do_declare_xattr_del = lod_declare_xattr_del,
6956 .do_xattr_del = lod_xattr_del,
6957 .do_xattr_list = lod_xattr_list,
6958 .do_ah_init = lod_ah_init,
6959 .do_declare_create = lod_declare_create,
6960 .do_create = lod_create,
6961 .do_declare_destroy = lod_declare_destroy,
6962 .do_destroy = lod_destroy,
6963 .do_index_try = lod_index_try,
6964 .do_declare_ref_add = lod_declare_ref_add,
6965 .do_ref_add = lod_ref_add,
6966 .do_declare_ref_del = lod_declare_ref_del,
6967 .do_ref_del = lod_ref_del,
6968 .do_object_sync = lod_object_sync,
6969 .do_object_lock = lod_object_lock,
6970 .do_object_unlock = lod_object_unlock,
6971 .do_invalidate = lod_invalidate,
6972 .do_declare_layout_change = lod_declare_layout_change,
6973 .do_layout_change = lod_layout_change,
6977 * Implementation of dt_body_operations::dbo_read.
6979 * \see dt_body_operations::dbo_read() in the API description for details.
6981 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
6982 struct lu_buf *buf, loff_t *pos)
6984 struct dt_object *next = dt_object_child(dt);
6986 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
6987 S_ISLNK(dt->do_lu.lo_header->loh_attr));
6988 return next->do_body_ops->dbo_read(env, next, buf, pos);
6992 * Implementation of dt_body_operations::dbo_declare_write.
6994 * \see dt_body_operations::dbo_declare_write() in the API description
6997 static ssize_t lod_declare_write(const struct lu_env *env,
6998 struct dt_object *dt,
6999 const struct lu_buf *buf, loff_t pos,
7002 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
7006 * Implementation of dt_body_operations::dbo_write.
7008 * \see dt_body_operations::dbo_write() in the API description for details.
7010 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
7011 const struct lu_buf *buf, loff_t *pos,
7014 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
7015 S_ISLNK(dt->do_lu.lo_header->loh_attr));
7016 return lod_sub_write(env, dt_object_child(dt), buf, pos, th);
7019 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
7020 __u64 start, __u64 end, struct thandle *th)
7022 if (dt_object_remote(dt))
7025 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
7028 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
7029 __u64 start, __u64 end, struct thandle *th)
7031 if (dt_object_remote(dt))
7034 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
7035 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
7039 * different type of files use the same body_ops because object may be created
7040 * in OUT, where there is no chance to set correct body_ops for each type, so
7041 * body_ops themselves will check file type inside, see lod_read/write/punch for
7044 const struct dt_body_operations lod_body_ops = {
7045 .dbo_read = lod_read,
7046 .dbo_declare_write = lod_declare_write,
7047 .dbo_write = lod_write,
7048 .dbo_declare_punch = lod_declare_punch,
7049 .dbo_punch = lod_punch,
7053 * Implementation of lu_object_operations::loo_object_init.
7055 * The function determines the type and the index of the target device using
7056 * sequence of the object's FID. Then passes control down to the
7057 * corresponding device:
7058 * OSD for the local objects, OSP for remote
7060 * \see lu_object_operations::loo_object_init() in the API description
7063 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
7064 const struct lu_object_conf *conf)
7066 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
7067 struct lu_device *cdev = NULL;
7068 struct lu_object *cobj;
7069 struct lod_tgt_descs *ltd = NULL;
7070 struct lod_tgt_desc *tgt;
7072 int type = LU_SEQ_RANGE_ANY;
7076 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
7078 /* Note: Sometimes, it will Return EAGAIN here, see
7079 * ptrlpc_import_delay_req(), which might confuse
7080 * lu_object_find_at() and make it wait there incorrectly.
7081 * so we convert it to EIO here.*/
7088 if (type == LU_SEQ_RANGE_MDT &&
7089 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
7090 cdev = &lod->lod_child->dd_lu_dev;
7091 } else if (type == LU_SEQ_RANGE_MDT) {
7092 ltd = &lod->lod_mdt_descs;
7094 } else if (type == LU_SEQ_RANGE_OST) {
7095 ltd = &lod->lod_ost_descs;
7102 if (ltd->ltd_tgts_size > idx &&
7103 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
7104 tgt = LTD_TGT(ltd, idx);
7106 LASSERT(tgt != NULL);
7107 LASSERT(tgt->ltd_tgt != NULL);
7109 cdev = &(tgt->ltd_tgt->dd_lu_dev);
7111 lod_putref(lod, ltd);
7114 if (unlikely(cdev == NULL))
7117 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
7118 if (unlikely(cobj == NULL))
7121 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
7123 lu_object_add(lo, cobj);
7130 * Alloc cached foreign LOV
7132 * \param[in] lo object
7133 * \param[in] size size of foreign LOV
7135 * \retval 0 on success
7136 * \retval negative if failed
7138 int lod_alloc_foreign_lov(struct lod_object *lo, size_t size)
7140 OBD_ALLOC_LARGE(lo->ldo_foreign_lov, size);
7141 if (lo->ldo_foreign_lov == NULL)
7143 lo->ldo_foreign_lov_size = size;
7144 lo->ldo_is_foreign = 1;
7150 * Free cached foreign LOV
7152 * \param[in] lo object
7154 void lod_free_foreign_lov(struct lod_object *lo)
7156 if (lo->ldo_foreign_lov != NULL)
7157 OBD_FREE_LARGE(lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
7158 lo->ldo_foreign_lov = NULL;
7159 lo->ldo_foreign_lov_size = 0;
7160 lo->ldo_is_foreign = 0;
7165 * Free cached foreign LMV
7167 * \param[in] lo object
7169 void lod_free_foreign_lmv(struct lod_object *lo)
7171 if (lo->ldo_foreign_lmv != NULL)
7172 OBD_FREE_LARGE(lo->ldo_foreign_lmv, lo->ldo_foreign_lmv_size);
7173 lo->ldo_foreign_lmv = NULL;
7174 lo->ldo_foreign_lmv_size = 0;
7175 lo->ldo_dir_is_foreign = 0;
7180 * Release resources associated with striping.
7182 * If the object is striped (regular or directory), then release
7183 * the stripe objects references and free the ldo_stripe array.
7185 * \param[in] env execution environment
7186 * \param[in] lo object
7188 void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo)
7190 struct lod_layout_component *lod_comp;
7193 if (unlikely(lo->ldo_is_foreign)) {
7194 lod_free_foreign_lov(lo);
7195 lo->ldo_comp_cached = 0;
7196 } else if (unlikely(lo->ldo_dir_is_foreign)) {
7197 lod_free_foreign_lmv(lo);
7198 lo->ldo_dir_stripe_loaded = 0;
7199 } else if (lo->ldo_stripe != NULL) {
7200 LASSERT(lo->ldo_comp_entries == NULL);
7201 LASSERT(lo->ldo_dir_stripes_allocated > 0);
7203 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
7204 if (lo->ldo_stripe[i])
7205 dt_object_put(env, lo->ldo_stripe[i]);
7208 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
7209 OBD_FREE(lo->ldo_stripe, j);
7210 lo->ldo_stripe = NULL;
7211 lo->ldo_dir_stripes_allocated = 0;
7212 lo->ldo_dir_stripe_loaded = 0;
7213 lo->ldo_dir_stripe_count = 0;
7214 } else if (lo->ldo_comp_entries != NULL) {
7215 for (i = 0; i < lo->ldo_comp_cnt; i++) {
7216 /* free lod_layout_component::llc_stripe array */
7217 lod_comp = &lo->ldo_comp_entries[i];
7219 if (lod_comp->llc_stripe == NULL)
7221 LASSERT(lod_comp->llc_stripes_allocated != 0);
7222 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
7223 if (lod_comp->llc_stripe[j] != NULL)
7225 &lod_comp->llc_stripe[j]->do_lu);
7227 OBD_FREE(lod_comp->llc_stripe,
7228 sizeof(struct dt_object *) *
7229 lod_comp->llc_stripes_allocated);
7230 lod_comp->llc_stripe = NULL;
7231 OBD_FREE(lod_comp->llc_ost_indices,
7233 lod_comp->llc_stripes_allocated);
7234 lod_comp->llc_ost_indices = NULL;
7235 lod_comp->llc_stripes_allocated = 0;
7237 lod_free_comp_entries(lo);
7238 lo->ldo_comp_cached = 0;
7242 void lod_striping_free(const struct lu_env *env, struct lod_object *lo)
7244 mutex_lock(&lo->ldo_layout_mutex);
7245 lod_striping_free_nolock(env, lo);
7246 mutex_unlock(&lo->ldo_layout_mutex);
7250 * Implementation of lu_object_operations::loo_object_free.
7252 * \see lu_object_operations::loo_object_free() in the API description
7255 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
7257 struct lod_object *lo = lu2lod_obj(o);
7259 /* release all underlying object pinned */
7260 lod_striping_free(env, lo);
7262 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
7266 * Implementation of lu_object_operations::loo_object_release.
7268 * \see lu_object_operations::loo_object_release() in the API description
7271 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
7273 /* XXX: shouldn't we release everything here in case if object
7274 * creation failed before? */
7278 * Implementation of lu_object_operations::loo_object_print.
7280 * \see lu_object_operations::loo_object_print() in the API description
7283 static int lod_object_print(const struct lu_env *env, void *cookie,
7284 lu_printer_t p, const struct lu_object *l)
7286 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
7288 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
7291 struct lu_object_operations lod_lu_obj_ops = {
7292 .loo_object_init = lod_object_init,
7293 .loo_object_free = lod_object_free,
7294 .loo_object_release = lod_object_release,
7295 .loo_object_print = lod_object_print,