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,
1082 if (comp->llc_pattern != LOV_PATTERN_MDT) {
1083 if (append_stripes) {
1084 comp->llc_stripe_count = append_stripes;
1085 } else if (!comp->llc_stripe_count) {
1086 comp->llc_stripe_count =
1087 desc->ld_default_stripe_count;
1090 if (comp->llc_stripe_size <= 0)
1091 comp->llc_stripe_size = desc->ld_default_stripe_size;
1094 int lod_obj_for_each_stripe(const struct lu_env *env, struct lod_object *lo,
1096 struct lod_obj_stripe_cb_data *data)
1098 struct lod_layout_component *lod_comp;
1102 LASSERT(lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL);
1103 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1104 lod_comp = &lo->ldo_comp_entries[i];
1106 if (lod_comp->llc_stripe == NULL)
1109 /* has stripe but not inited yet, this component has been
1110 * declared to be created, but hasn't created yet.
1112 if (!lod_comp_inited(lod_comp))
1115 if (data->locd_comp_skip_cb &&
1116 data->locd_comp_skip_cb(env, lo, i, data))
1119 if (data->locd_comp_cb) {
1120 rc = data->locd_comp_cb(env, lo, i, data);
1125 /* could used just to do sth about component, not each
1128 if (!data->locd_stripe_cb)
1131 LASSERT(lod_comp->llc_stripe_count > 0);
1132 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
1133 struct dt_object *dt = lod_comp->llc_stripe[j];
1137 rc = data->locd_stripe_cb(env, lo, dt, th, i, j, data);
1145 static bool lod_obj_attr_set_comp_skip_cb(const struct lu_env *env,
1146 struct lod_object *lo, int comp_idx,
1147 struct lod_obj_stripe_cb_data *data)
1149 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[comp_idx];
1150 bool skipped = false;
1152 if (!(data->locd_attr->la_valid & LA_LAYOUT_VERSION))
1155 switch (lo->ldo_flr_state) {
1156 case LCM_FL_WRITE_PENDING: {
1159 /* skip stale components */
1160 if (lod_comp->llc_flags & LCME_FL_STALE) {
1165 /* skip valid and overlapping components, therefore any
1166 * attempts to write overlapped components will never succeed
1167 * because client will get EINPROGRESS. */
1168 for (i = 0; i < lo->ldo_comp_cnt; i++) {
1172 if (lo->ldo_comp_entries[i].llc_flags & LCME_FL_STALE)
1175 if (lu_extent_is_overlapped(&lod_comp->llc_extent,
1176 &lo->ldo_comp_entries[i].llc_extent)) {
1184 LASSERTF(0, "impossible: %d\n", lo->ldo_flr_state);
1185 case LCM_FL_SYNC_PENDING:
1189 CDEBUG(D_LAYOUT, DFID": %s to set component %x to version: %u\n",
1190 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
1191 skipped ? "skipped" : "chose", lod_comp->llc_id,
1192 data->locd_attr->la_layout_version);
1198 lod_obj_stripe_attr_set_cb(const struct lu_env *env, struct lod_object *lo,
1199 struct dt_object *dt, struct thandle *th,
1200 int comp_idx, int stripe_idx,
1201 struct lod_obj_stripe_cb_data *data)
1203 if (data->locd_declare)
1204 return lod_sub_declare_attr_set(env, dt, data->locd_attr, th);
1206 if (data->locd_attr->la_valid & LA_LAYOUT_VERSION) {
1207 CDEBUG(D_LAYOUT, DFID": set layout version: %u, comp_idx: %d\n",
1208 PFID(lu_object_fid(&dt->do_lu)),
1209 data->locd_attr->la_layout_version, comp_idx);
1212 return lod_sub_attr_set(env, dt, data->locd_attr, th);
1216 * Implementation of dt_object_operations::do_declare_attr_set.
1218 * If the object is striped, then apply the changes to all the stripes.
1220 * \see dt_object_operations::do_declare_attr_set() in the API description
1223 static int lod_declare_attr_set(const struct lu_env *env,
1224 struct dt_object *dt,
1225 const struct lu_attr *attr,
1228 struct dt_object *next = dt_object_child(dt);
1229 struct lod_object *lo = lod_dt_obj(dt);
1234 * declare setattr on the local object
1236 rc = lod_sub_declare_attr_set(env, next, attr, th);
1240 /* osp_declare_attr_set() ignores all attributes other than
1241 * UID, GID, PROJID, and size, and osp_attr_set() ignores all
1242 * but UID, GID and PROJID. Declaration of size attr setting
1243 * happens through lod_declare_init_size(), and not through
1244 * this function. Therefore we need not load striping unless
1245 * ownership is changing. This should save memory and (we hope)
1246 * speed up rename().
1248 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1249 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1252 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1255 if (!(attr->la_valid & (LA_UID | LA_GID | LA_PROJID | LA_MODE |
1256 LA_ATIME | LA_MTIME | LA_CTIME |
1261 * load striping information, notice we don't do this when object
1262 * is being initialized as we don't need this information till
1263 * few specific cases like destroy, chown
1265 rc = lod_striping_load(env, lo);
1269 if (!lod_obj_is_striped(dt))
1273 * if object is striped declare changes on the stripes
1275 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1276 LASSERT(lo->ldo_stripe);
1277 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1278 if (lo->ldo_stripe[i] == NULL)
1280 if (!dt_object_exists(lo->ldo_stripe[i]))
1282 rc = lod_sub_declare_attr_set(env, lo->ldo_stripe[i],
1288 struct lod_obj_stripe_cb_data data = { { 0 } };
1290 data.locd_attr = attr;
1291 data.locd_declare = true;
1292 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1293 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1299 if (!dt_object_exists(next) || dt_object_remote(next) ||
1300 !S_ISREG(attr->la_mode))
1303 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1304 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
1308 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) ||
1309 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1310 struct lod_thread_info *info = lod_env_info(env);
1311 struct lu_buf *buf = &info->lti_buf;
1313 buf->lb_buf = info->lti_ea_store;
1314 buf->lb_len = info->lti_ea_store_size;
1315 rc = lod_sub_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1316 LU_XATTR_REPLACE, th);
1323 * Implementation of dt_object_operations::do_attr_set.
1325 * If the object is striped, then apply the changes to all or subset of
1326 * the stripes depending on the object type and specific attributes.
1328 * \see dt_object_operations::do_attr_set() in the API description for details.
1330 static int lod_attr_set(const struct lu_env *env,
1331 struct dt_object *dt,
1332 const struct lu_attr *attr,
1335 struct dt_object *next = dt_object_child(dt);
1336 struct lod_object *lo = lod_dt_obj(dt);
1341 * apply changes to the local object
1343 rc = lod_sub_attr_set(env, next, attr, th);
1347 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1348 if (!(attr->la_valid & LA_REMOTE_ATTR_SET))
1351 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1354 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE | LA_PROJID |
1355 LA_ATIME | LA_MTIME | LA_CTIME |
1360 /* FIXME: a tricky case in the code path of mdd_layout_change():
1361 * the in-memory striping information has been freed in lod_xattr_set()
1362 * due to layout change. It has to load stripe here again. It only
1363 * changes flags of layout so declare_attr_set() is still accurate */
1364 rc = lod_striping_load(env, lo);
1368 if (!lod_obj_is_striped(dt))
1372 * if object is striped, apply changes to all the stripes
1374 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1375 LASSERT(lo->ldo_stripe);
1376 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1377 if (unlikely(lo->ldo_stripe[i] == NULL))
1380 if ((dt_object_exists(lo->ldo_stripe[i]) == 0))
1383 rc = lod_sub_attr_set(env, lo->ldo_stripe[i], attr, th);
1388 struct lod_obj_stripe_cb_data data = { { 0 } };
1390 data.locd_attr = attr;
1391 data.locd_declare = false;
1392 data.locd_comp_skip_cb = lod_obj_attr_set_comp_skip_cb;
1393 data.locd_stripe_cb = lod_obj_stripe_attr_set_cb;
1394 rc = lod_obj_for_each_stripe(env, lo, th, &data);
1400 if (!dt_object_exists(next) || dt_object_remote(next) ||
1401 !S_ISREG(attr->la_mode))
1404 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE)) {
1405 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
1409 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE)) {
1410 struct lod_thread_info *info = lod_env_info(env);
1411 struct lu_buf *buf = &info->lti_buf;
1412 struct ost_id *oi = &info->lti_ostid;
1413 struct lu_fid *fid = &info->lti_fid;
1414 struct lov_mds_md_v1 *lmm;
1415 struct lov_ost_data_v1 *objs;
1418 rc = lod_get_lov_ea(env, lo);
1422 buf->lb_buf = info->lti_ea_store;
1423 buf->lb_len = info->lti_ea_store_size;
1424 lmm = info->lti_ea_store;
1425 magic = le32_to_cpu(lmm->lmm_magic);
1426 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
1427 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1428 struct lov_comp_md_entry_v1 *lcme =
1429 &lcm->lcm_entries[0];
1431 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
1432 magic = le32_to_cpu(lmm->lmm_magic);
1435 if (magic == LOV_MAGIC_V1)
1436 objs = &(lmm->lmm_objects[0]);
1438 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1439 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1440 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1442 fid_to_ostid(fid, oi);
1443 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1445 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1446 LU_XATTR_REPLACE, th);
1447 } else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PFL_RANGE)) {
1448 struct lod_thread_info *info = lod_env_info(env);
1449 struct lu_buf *buf = &info->lti_buf;
1450 struct lov_comp_md_v1 *lcm;
1451 struct lov_comp_md_entry_v1 *lcme;
1453 rc = lod_get_lov_ea(env, lo);
1457 buf->lb_buf = info->lti_ea_store;
1458 buf->lb_len = info->lti_ea_store_size;
1460 if (le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_COMP_V1 &&
1461 le32_to_cpu(lcm->lcm_magic) != LOV_MAGIC_SEL)
1464 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1465 lcme = &lcm->lcm_entries[0];
1466 le64_add_cpu(&lcme->lcme_extent.e_start, 1);
1467 le64_add_cpu(&lcme->lcme_extent.e_end, -1);
1469 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LOV,
1470 LU_XATTR_REPLACE, th);
1477 * Implementation of dt_object_operations::do_xattr_get.
1479 * If LOV EA is requested from the root object and it's not
1480 * found, then return default striping for the filesystem.
1482 * \see dt_object_operations::do_xattr_get() in the API description for details.
1484 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1485 struct lu_buf *buf, const char *name)
1487 struct lod_thread_info *info = lod_env_info(env);
1488 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1493 rc = dt_xattr_get(env, dt_object_child(dt), buf, name);
1494 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1495 struct lmv_mds_md_v1 *lmv1;
1496 struct lmv_foreign_md *lfm;
1499 if (rc > (typeof(rc))sizeof(*lmv1))
1502 /* short (<= sizeof(struct lmv_mds_md_v1)) foreign LMV case */
1503 /* XXX empty foreign LMV is not allowed */
1504 if (rc <= offsetof(typeof(*lfm), lfm_value))
1505 RETURN(rc = rc > 0 ? -EINVAL : rc);
1507 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1508 CLASSERT(sizeof(*lmv1) <= sizeof(info->lti_key));
1510 /* lti_buf is large enough for *lmv1 or a short
1511 * (<= sizeof(struct lmv_mds_md_v1)) foreign LMV
1513 info->lti_buf.lb_buf = info->lti_key;
1514 info->lti_buf.lb_len = sizeof(*lmv1);
1515 rc = dt_xattr_get(env, dt_object_child(dt),
1516 &info->lti_buf, name);
1517 if (unlikely(rc <= offsetof(typeof(*lfm),
1519 RETURN(rc = rc > 0 ? -EINVAL : rc);
1521 lfm = info->lti_buf.lb_buf;
1522 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN)
1525 if (unlikely(rc != sizeof(*lmv1)))
1526 RETURN(rc = rc > 0 ? -EINVAL : rc);
1528 lmv1 = info->lti_buf.lb_buf;
1529 /* The on-disk LMV EA only contains header, but the
1530 * returned LMV EA size should contain the space for
1531 * the FIDs of all shards of the striped directory. */
1532 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1533 rc = lmv_mds_md_size(
1534 le32_to_cpu(lmv1->lmv_stripe_count),
1538 if (le32_to_cpu(lfm->lfm_magic) == LMV_MAGIC_FOREIGN)
1541 if (rc != sizeof(*lmv1))
1542 RETURN(rc = rc > 0 ? -EINVAL : rc);
1544 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1548 RETURN(rc = rc1 != 0 ? rc1 : rc);
1551 if ((rc > 0) && buf->lb_buf && strcmp(name, XATTR_NAME_LOV) == 0) {
1552 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1554 if (lcm->lcm_magic == cpu_to_le32(LOV_MAGIC_SEL))
1555 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
1558 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1562 * XXX: Only used by lfsck
1564 * lod returns default striping on the real root of the device
1565 * this is like the root stores default striping for the whole
1566 * filesystem. historically we've been using a different approach
1567 * and store it in the config.
1569 dt_root_get(env, dev->lod_child, &info->lti_fid);
1570 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1572 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1573 struct lov_user_md *lum = buf->lb_buf;
1574 struct lov_desc *desc = &dev->lod_desc;
1576 if (buf->lb_buf == NULL) {
1578 } else if (buf->lb_len >= sizeof(*lum)) {
1579 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1580 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1581 lmm_oi_set_id(&lum->lmm_oi, 0);
1582 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1583 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1584 lum->lmm_stripe_size = cpu_to_le32(
1585 desc->ld_default_stripe_size);
1586 lum->lmm_stripe_count = cpu_to_le16(
1587 desc->ld_default_stripe_count);
1588 lum->lmm_stripe_offset = cpu_to_le16(
1589 desc->ld_default_stripe_offset);
1602 * Checks that the magic of the stripe is sane.
1604 * \param[in] lod lod device
1605 * \param[in] lum a buffer storing LMV EA to verify
1607 * \retval 0 if the EA is sane
1608 * \retval negative otherwise
1610 static int lod_verify_md_striping(struct lod_device *lod,
1611 const struct lmv_user_md_v1 *lum)
1613 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1614 CERROR("%s: invalid lmv_user_md: magic = %x, "
1615 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1616 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1617 (int)le32_to_cpu(lum->lum_stripe_offset),
1618 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1626 * Initialize LMV EA for a slave.
1628 * Initialize slave's LMV EA from the master's LMV EA.
1630 * \param[in] master_lmv a buffer containing master's EA
1631 * \param[out] slave_lmv a buffer where slave's EA will be stored
1634 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1635 const struct lmv_mds_md_v1 *master_lmv)
1637 *slave_lmv = *master_lmv;
1638 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1644 * Generate LMV EA from the object passed as \a dt. The object must have
1645 * the stripes created and initialized.
1647 * \param[in] env execution environment
1648 * \param[in] dt object
1649 * \param[out] lmv_buf buffer storing generated LMV EA
1651 * \retval 0 on success
1652 * \retval negative if failed
1654 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1655 struct lu_buf *lmv_buf)
1657 struct lod_thread_info *info = lod_env_info(env);
1658 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1659 struct lod_object *lo = lod_dt_obj(dt);
1660 struct lmv_mds_md_v1 *lmm1;
1662 int type = LU_SEQ_RANGE_ANY;
1667 LASSERT(lo->ldo_dir_striped != 0);
1668 LASSERT(lo->ldo_dir_stripe_count > 0);
1669 stripe_count = lo->ldo_dir_stripe_count;
1670 /* Only store the LMV EA heahder on the disk. */
1671 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1672 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1676 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1679 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1680 memset(lmm1, 0, sizeof(*lmm1));
1681 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1682 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1683 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1684 if (lo->ldo_dir_hash_type & LMV_HASH_FLAG_MIGRATION) {
1685 lmm1->lmv_migrate_hash = cpu_to_le32(lo->ldo_dir_migrate_hash);
1686 lmm1->lmv_migrate_offset =
1687 cpu_to_le32(lo->ldo_dir_migrate_offset);
1689 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1694 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1695 lmv_buf->lb_buf = info->lti_ea_store;
1696 lmv_buf->lb_len = sizeof(*lmm1);
1702 * Create in-core represenation for a striped directory.
1704 * Parse the buffer containing LMV EA and instantiate LU objects
1705 * representing the stripe objects. The pointers to the objects are
1706 * stored in ldo_stripe field of \a lo. This function is used when
1707 * we need to access an already created object (i.e. load from a disk).
1709 * \param[in] env execution environment
1710 * \param[in] lo lod object
1711 * \param[in] buf buffer containing LMV EA
1713 * \retval 0 on success
1714 * \retval negative if failed
1716 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1717 const struct lu_buf *buf)
1719 struct lod_thread_info *info = lod_env_info(env);
1720 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1721 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1722 struct dt_object **stripe;
1723 union lmv_mds_md *lmm = buf->lb_buf;
1724 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1725 struct lu_fid *fid = &info->lti_fid;
1730 LASSERT(mutex_is_locked(&lo->ldo_layout_mutex));
1732 /* XXX may be useless as not called for foreign LMV ?? */
1733 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_FOREIGN)
1736 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1737 lo->ldo_dir_slave_stripe = 1;
1741 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1744 if (le32_to_cpu(lmv1->lmv_stripe_count) < 1)
1747 LASSERT(lo->ldo_stripe == NULL);
1748 OBD_ALLOC(stripe, sizeof(stripe[0]) *
1749 (le32_to_cpu(lmv1->lmv_stripe_count)));
1753 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1754 struct dt_device *tgt_dt;
1755 struct dt_object *dto;
1756 int type = LU_SEQ_RANGE_ANY;
1759 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1760 if (!fid_is_sane(fid)) {
1765 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1769 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1770 tgt_dt = lod->lod_child;
1772 struct lod_tgt_desc *tgt;
1774 tgt = LTD_TGT(ltd, idx);
1776 GOTO(out, rc = -ESTALE);
1777 tgt_dt = tgt->ltd_tgt;
1780 dto = dt_locate_at(env, tgt_dt, fid,
1781 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1784 GOTO(out, rc = PTR_ERR(dto));
1789 lo->ldo_stripe = stripe;
1790 lo->ldo_dir_stripe_count = le32_to_cpu(lmv1->lmv_stripe_count);
1791 lo->ldo_dir_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1793 lod_striping_free_nolock(env, lo);
1799 * Declare create a striped directory.
1801 * Declare creating a striped directory with a given stripe pattern on the
1802 * specified MDTs. A striped directory is represented as a regular directory
1803 * - an index listing all the stripes. The stripes point back to the master
1804 * object with ".." and LinkEA. The master object gets LMV EA which
1805 * identifies it as a striped directory. The function allocates FIDs
1808 * \param[in] env execution environment
1809 * \param[in] dt object
1810 * \param[in] attr attributes to initialize the objects with
1811 * \param[in] dof type of objects to be created
1812 * \param[in] th transaction handle
1814 * \retval 0 on success
1815 * \retval negative if failed
1817 static int lod_dir_declare_create_stripes(const struct lu_env *env,
1818 struct dt_object *dt,
1819 struct lu_attr *attr,
1820 struct dt_object_format *dof,
1823 struct lod_thread_info *info = lod_env_info(env);
1824 struct lu_buf lmv_buf;
1825 struct lu_buf slave_lmv_buf;
1826 struct lmv_mds_md_v1 *lmm;
1827 struct lmv_mds_md_v1 *slave_lmm = NULL;
1828 struct dt_insert_rec *rec = &info->lti_dt_rec;
1829 struct lod_object *lo = lod_dt_obj(dt);
1834 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1837 lmm = lmv_buf.lb_buf;
1839 OBD_ALLOC_PTR(slave_lmm);
1840 if (slave_lmm == NULL)
1841 GOTO(out, rc = -ENOMEM);
1843 lod_prep_slave_lmv_md(slave_lmm, lmm);
1844 slave_lmv_buf.lb_buf = slave_lmm;
1845 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1847 if (!dt_try_as_dir(env, dt_object_child(dt)))
1848 GOTO(out, rc = -EINVAL);
1850 rec->rec_type = S_IFDIR;
1851 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
1852 struct dt_object *dto = lo->ldo_stripe[i];
1853 char *stripe_name = info->lti_key;
1854 struct lu_name *sname;
1855 struct linkea_data ldata = { NULL };
1856 struct lu_buf linkea_buf;
1858 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
1862 rc = lod_sub_declare_create(env, dto, attr, NULL, dof, th);
1866 if (!dt_try_as_dir(env, dto))
1867 GOTO(out, rc = -EINVAL);
1869 rc = lod_sub_declare_ref_add(env, dto, th);
1873 rec->rec_fid = lu_object_fid(&dto->do_lu);
1874 rc = lod_sub_declare_insert(env, dto,
1875 (const struct dt_rec *)rec,
1876 (const struct dt_key *)dot, th);
1880 /* master stripe FID will be put to .. */
1881 rec->rec_fid = lu_object_fid(&dt->do_lu);
1882 rc = lod_sub_declare_insert(env, dto,
1883 (const struct dt_rec *)rec,
1884 (const struct dt_key *)dotdot, th);
1888 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1889 cfs_fail_val != i) {
1890 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1892 slave_lmm->lmv_master_mdt_index =
1895 slave_lmm->lmv_master_mdt_index =
1897 rc = lod_sub_declare_xattr_set(env, dto, &slave_lmv_buf,
1898 XATTR_NAME_LMV, 0, th);
1903 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1905 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1906 PFID(lu_object_fid(&dto->do_lu)), i + 1);
1908 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1909 PFID(lu_object_fid(&dto->do_lu)), i);
1911 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
1912 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
1913 sname, lu_object_fid(&dt->do_lu));
1917 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1918 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1919 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
1920 XATTR_NAME_LINK, 0, th);
1924 rec->rec_fid = lu_object_fid(&dto->do_lu);
1925 rc = lod_sub_declare_insert(env, dt_object_child(dt),
1926 (const struct dt_rec *)rec,
1927 (const struct dt_key *)stripe_name,
1932 rc = lod_sub_declare_ref_add(env, dt_object_child(dt), th);
1937 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt),
1938 &lmv_buf, XATTR_NAME_LMV, 0, th);
1942 if (slave_lmm != NULL)
1943 OBD_FREE_PTR(slave_lmm);
1948 static int lod_prep_md_striped_create(const struct lu_env *env,
1949 struct dt_object *dt,
1950 struct lu_attr *attr,
1951 const struct lmv_user_md_v1 *lum,
1952 struct dt_object_format *dof,
1955 struct lod_thread_info *info = lod_env_info(env);
1956 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1957 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1958 struct lod_object *lo = lod_dt_obj(dt);
1959 struct dt_object **stripe;
1966 bool is_specific = false;
1969 /* The lum has been verifed in lod_verify_md_striping */
1970 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1971 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC);
1973 stripe_count = lo->ldo_dir_stripe_count;
1975 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
1976 if (idx_array == NULL)
1979 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_count);
1981 GOTO(out_free, rc = -ENOMEM);
1983 /* Start index must be the master MDT */
1984 master_index = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_node_id;
1985 idx_array[0] = master_index;
1986 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
1988 for (i = 1; i < stripe_count; i++)
1989 idx_array[i] = le32_to_cpu(lum->lum_objects[i].lum_mds);
1992 for (i = 0; i < stripe_count; i++) {
1993 struct lod_tgt_desc *tgt = NULL;
1994 struct dt_object *dto;
1995 struct lu_fid fid = { 0 };
1997 struct lu_object_conf conf = { 0 };
1998 struct dt_device *tgt_dt = NULL;
2000 /* Try to find next avaible target */
2002 for (j = 0; j < lod->lod_remote_mdt_count;
2003 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
2004 bool already_allocated = false;
2007 CDEBUG(D_INFO, "try idx %d, mdt cnt %u, allocated %u\n",
2008 idx, lod->lod_remote_mdt_count + 1, i);
2010 if (likely(!is_specific &&
2011 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE))) {
2012 /* check whether the idx already exists
2013 * in current allocated array */
2014 for (k = 0; k < i; k++) {
2015 if (idx_array[k] == idx) {
2016 already_allocated = true;
2021 if (already_allocated)
2025 /* Sigh, this index is not in the bitmap, let's check
2026 * next available target */
2027 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) &&
2028 idx != master_index)
2031 if (idx == master_index) {
2032 /* Allocate the FID locally */
2033 rc = obd_fid_alloc(env, lod->lod_child_exp,
2037 tgt_dt = lod->lod_child;
2041 /* check the status of the OSP */
2042 tgt = LTD_TGT(ltd, idx);
2046 tgt_dt = tgt->ltd_tgt;
2047 rc = dt_statfs(env, tgt_dt, &info->lti_osfs);
2049 /* this OSP doesn't feel well */
2054 rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
2063 /* Can not allocate more stripes */
2064 if (j == lod->lod_remote_mdt_count) {
2065 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
2066 lod2obd(lod)->obd_name, stripe_count, i);
2070 CDEBUG(D_INFO, "Get idx %d, for stripe %d "DFID"\n",
2071 idx, i, PFID(&fid));
2073 /* Set the start index for next stripe allocation */
2074 if (!is_specific && i < stripe_count - 1) {
2076 * for large dir test, put all other slaves on one
2077 * remote MDT, otherwise we may save too many local
2078 * slave locks which will exceed RS_MAX_LOCKS.
2080 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)))
2082 idx_array[i + 1] = (idx + 1) %
2083 (lod->lod_remote_mdt_count + 1);
2085 /* tgt_dt and fid must be ready after search avaible OSP
2086 * in the above loop */
2087 LASSERT(tgt_dt != NULL);
2088 LASSERT(fid_is_sane(&fid));
2090 /* fail a remote stripe FID allocation */
2091 if (i && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_FID))
2094 conf.loc_flags = LOC_F_NEW;
2095 dto = dt_locate_at(env, tgt_dt, &fid,
2096 dt->do_lu.lo_dev->ld_site->ls_top_dev,
2099 GOTO(out_put, rc = PTR_ERR(dto));
2103 lo->ldo_dir_striped = 1;
2104 lo->ldo_stripe = stripe;
2105 lo->ldo_dir_stripe_count = i;
2106 lo->ldo_dir_stripes_allocated = stripe_count;
2108 lo->ldo_dir_stripe_loaded = 1;
2110 if (lo->ldo_dir_stripe_count == 0)
2111 GOTO(out_put, rc = -ENOSPC);
2113 rc = lod_dir_declare_create_stripes(env, dt, attr, dof, th);
2119 for (i = 0; i < stripe_count; i++)
2120 if (stripe[i] != NULL)
2121 dt_object_put(env, stripe[i]);
2122 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_count);
2123 lo->ldo_dir_stripe_count = 0;
2124 lo->ldo_dir_stripes_allocated = 0;
2125 lo->ldo_stripe = NULL;
2129 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
2136 * Alloc cached foreign LMV
2138 * \param[in] lo object
2139 * \param[in] size size of foreign LMV
2141 * \retval 0 on success
2142 * \retval negative if failed
2144 int lod_alloc_foreign_lmv(struct lod_object *lo, size_t size)
2146 OBD_ALLOC_LARGE(lo->ldo_foreign_lmv, size);
2147 if (lo->ldo_foreign_lmv == NULL)
2149 lo->ldo_foreign_lmv_size = size;
2150 lo->ldo_dir_is_foreign = 1;
2156 * Declare create striped md object.
2158 * The function declares intention to create a striped directory. This is a
2159 * wrapper for lod_prep_md_striped_create(). The only additional functionality
2160 * is to verify pattern \a lum_buf is good. Check that function for the details.
2162 * \param[in] env execution environment
2163 * \param[in] dt object
2164 * \param[in] attr attributes to initialize the objects with
2165 * \param[in] lum_buf a pattern specifying the number of stripes and
2167 * \param[in] dof type of objects to be created
2168 * \param[in] th transaction handle
2170 * \retval 0 on success
2171 * \retval negative if failed
2174 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
2175 struct dt_object *dt,
2176 struct lu_attr *attr,
2177 const struct lu_buf *lum_buf,
2178 struct dt_object_format *dof,
2181 struct lod_object *lo = lod_dt_obj(dt);
2182 struct lmv_user_md_v1 *lum = lum_buf->lb_buf;
2186 LASSERT(lum != NULL);
2188 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
2189 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
2190 (int)le32_to_cpu(lum->lum_stripe_offset));
2192 if (lo->ldo_dir_stripe_count == 0) {
2193 if (lo->ldo_dir_is_foreign) {
2194 rc = lod_alloc_foreign_lmv(lo, lum_buf->lb_len);
2197 memcpy(lo->ldo_foreign_lmv, lum, lum_buf->lb_len);
2198 lo->ldo_dir_stripe_loaded = 1;
2203 /* prepare dir striped objects */
2204 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
2206 /* failed to create striping, let's reset
2207 * config so that others don't get confused */
2208 lod_striping_free(env, lo);
2216 * Append source stripes after target stripes for migrating directory. NB, we
2217 * only need to declare this, the append is done inside lod_xattr_set_lmv().
2219 * \param[in] env execution environment
2220 * \param[in] dt target object
2221 * \param[in] buf LMV buf which contains source stripe fids
2222 * \param[in] th transaction handle
2224 * \retval 0 on success
2225 * \retval negative if failed
2227 static int lod_dir_declare_layout_add(const struct lu_env *env,
2228 struct dt_object *dt,
2229 const struct lu_buf *buf,
2232 struct lod_thread_info *info = lod_env_info(env);
2233 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
2234 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
2235 struct lod_object *lo = lod_dt_obj(dt);
2236 struct dt_object *next = dt_object_child(dt);
2237 struct dt_object_format *dof = &info->lti_format;
2238 struct lmv_mds_md_v1 *lmv = buf->lb_buf;
2239 struct dt_object **stripe;
2240 __u32 stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
2241 struct lu_fid *fid = &info->lti_fid;
2242 struct lod_tgt_desc *tgt;
2243 struct dt_object *dto;
2244 struct dt_device *tgt_dt;
2245 int type = LU_SEQ_RANGE_ANY;
2246 struct dt_insert_rec *rec = &info->lti_dt_rec;
2247 char *stripe_name = info->lti_key;
2248 struct lu_name *sname;
2249 struct linkea_data ldata = { NULL };
2250 struct lu_buf linkea_buf;
2257 if (le32_to_cpu(lmv->lmv_magic) != LMV_MAGIC_V1)
2260 if (stripe_count == 0)
2263 dof->dof_type = DFT_DIR;
2266 sizeof(*stripe) * (lo->ldo_dir_stripe_count + stripe_count));
2270 for (i = 0; i < lo->ldo_dir_stripe_count; i++)
2271 stripe[i] = lo->ldo_stripe[i];
2273 for (i = 0; i < stripe_count; i++) {
2275 &lmv->lmv_stripe_fids[i]);
2276 if (!fid_is_sane(fid))
2279 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
2283 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
2284 tgt_dt = lod->lod_child;
2286 tgt = LTD_TGT(ltd, idx);
2288 GOTO(out, rc = -ESTALE);
2289 tgt_dt = tgt->ltd_tgt;
2292 dto = dt_locate_at(env, tgt_dt, fid,
2293 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
2296 GOTO(out, rc = PTR_ERR(dto));
2298 stripe[i + lo->ldo_dir_stripe_count] = dto;
2300 if (!dt_try_as_dir(env, dto))
2301 GOTO(out, rc = -ENOTDIR);
2303 rc = lod_sub_declare_ref_add(env, dto, th);
2307 rc = lod_sub_declare_insert(env, dto,
2308 (const struct dt_rec *)rec,
2309 (const struct dt_key *)dot, th);
2313 rc = lod_sub_declare_insert(env, dto,
2314 (const struct dt_rec *)rec,
2315 (const struct dt_key *)dotdot, th);
2319 rc = lod_sub_declare_xattr_set(env, dto, buf,
2320 XATTR_NAME_LMV, 0, th);
2324 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
2325 PFID(lu_object_fid(&dto->do_lu)),
2326 i + lo->ldo_dir_stripe_count);
2328 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
2329 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
2330 sname, lu_object_fid(&dt->do_lu));
2334 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
2335 linkea_buf.lb_len = ldata.ld_leh->leh_len;
2336 rc = lod_sub_declare_xattr_set(env, dto, &linkea_buf,
2337 XATTR_NAME_LINK, 0, th);
2341 rc = lod_sub_declare_insert(env, next,
2342 (const struct dt_rec *)rec,
2343 (const struct dt_key *)stripe_name,
2348 rc = lod_sub_declare_ref_add(env, next, th);
2354 OBD_FREE(lo->ldo_stripe,
2355 sizeof(*stripe) * lo->ldo_dir_stripes_allocated);
2356 lo->ldo_stripe = stripe;
2357 lo->ldo_dir_migrate_offset = lo->ldo_dir_stripe_count;
2358 lo->ldo_dir_migrate_hash = le32_to_cpu(lmv->lmv_hash_type);
2359 lo->ldo_dir_stripe_count += stripe_count;
2360 lo->ldo_dir_stripes_allocated += stripe_count;
2361 lo->ldo_dir_hash_type |= LMV_HASH_FLAG_MIGRATION;
2365 i = lo->ldo_dir_stripe_count;
2366 while (i < lo->ldo_dir_stripe_count + stripe_count && stripe[i])
2367 dt_object_put(env, stripe[i++]);
2370 sizeof(*stripe) * (stripe_count + lo->ldo_dir_stripe_count));
2374 static int lod_dir_declare_layout_delete(const struct lu_env *env,
2375 struct dt_object *dt,
2376 const struct lu_buf *buf,
2379 struct lod_thread_info *info = lod_env_info(env);
2380 struct lod_object *lo = lod_dt_obj(dt);
2381 struct dt_object *next = dt_object_child(dt);
2382 struct lmv_user_md *lmu = buf->lb_buf;
2383 __u32 final_stripe_count;
2384 char *stripe_name = info->lti_key;
2385 struct dt_object *dto;
2392 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
2393 if (final_stripe_count >= lo->ldo_dir_stripe_count)
2396 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
2397 dto = lo->ldo_stripe[i];
2401 if (!dt_try_as_dir(env, dto))
2404 rc = lod_sub_declare_delete(env, dto,
2405 (const struct dt_key *)dot, th);
2409 rc = lod_sub_declare_ref_del(env, dto, th);
2413 rc = lod_sub_declare_delete(env, dto,
2414 (const struct dt_key *)dotdot, th);
2418 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2419 PFID(lu_object_fid(&dto->do_lu)), i);
2421 rc = lod_sub_declare_delete(env, next,
2422 (const struct dt_key *)stripe_name, th);
2426 rc = lod_sub_declare_ref_del(env, next, th);
2435 * delete stripes from dir master object, the lum_stripe_count in argument is
2436 * the final stripe count, the stripes after that will be deleted, NB, they
2437 * are not destroyed, but deleted from it's parent namespace, this function
2438 * will be called in two places:
2439 * 1. mdd_migrate_create() delete stripes from source, and append them to
2441 * 2. mdd_dir_layout_shrink() delete stripes from source, and destroy them.
2443 static int lod_dir_layout_delete(const struct lu_env *env,
2444 struct dt_object *dt,
2445 const struct lu_buf *buf,
2448 struct lod_thread_info *info = lod_env_info(env);
2449 struct lod_object *lo = lod_dt_obj(dt);
2450 struct dt_object *next = dt_object_child(dt);
2451 struct lmv_user_md *lmu = buf->lb_buf;
2452 __u32 final_stripe_count;
2453 char *stripe_name = info->lti_key;
2454 struct dt_object *dto;
2463 final_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
2464 if (final_stripe_count >= lo->ldo_dir_stripe_count)
2467 for (i = final_stripe_count; i < lo->ldo_dir_stripe_count; i++) {
2468 dto = lo->ldo_stripe[i];
2472 rc = lod_sub_delete(env, dto,
2473 (const struct dt_key *)dotdot, th);
2477 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2478 PFID(lu_object_fid(&dto->do_lu)), i);
2480 rc = lod_sub_delete(env, next,
2481 (const struct dt_key *)stripe_name, th);
2485 rc = lod_sub_ref_del(env, next, th);
2490 lod_striping_free(env, lod_dt_obj(dt));
2496 * Implementation of dt_object_operations::do_declare_xattr_set.
2498 * Used with regular (non-striped) objects. Basically it
2499 * initializes the striping information and applies the
2500 * change to all the stripes.
2502 * \see dt_object_operations::do_declare_xattr_set() in the API description
2505 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2506 struct dt_object *dt,
2507 const struct lu_buf *buf,
2508 const char *name, int fl,
2511 struct dt_object *next = dt_object_child(dt);
2512 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2513 struct lod_object *lo = lod_dt_obj(dt);
2518 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2519 struct lmv_user_md_v1 *lum;
2521 LASSERT(buf != NULL && buf->lb_buf != NULL);
2523 rc = lod_verify_md_striping(d, lum);
2526 } else if (strcmp(name, XATTR_NAME_LOV) == 0) {
2527 rc = lod_verify_striping(d, lo, buf, false);
2532 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
2536 /* Note: Do not set LinkEA on sub-stripes, otherwise
2537 * it will confuse the fid2path process(see mdt_path_current()).
2538 * The linkEA between master and sub-stripes is set in
2539 * lod_xattr_set_lmv(). */
2540 if (strcmp(name, XATTR_NAME_LINK) == 0)
2543 /* set xattr to each stripes, if needed */
2544 rc = lod_striping_load(env, lo);
2548 if (lo->ldo_dir_stripe_count == 0)
2551 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
2552 if (!lo->ldo_stripe[i])
2555 if (!dt_object_exists(lo->ldo_stripe[i]))
2558 rc = lod_sub_declare_xattr_set(env, lo->ldo_stripe[i],
2568 lod_obj_stripe_replace_parent_fid_cb(const struct lu_env *env,
2569 struct lod_object *lo,
2570 struct dt_object *dt, struct thandle *th,
2571 int comp_idx, int stripe_idx,
2572 struct lod_obj_stripe_cb_data *data)
2574 struct lod_thread_info *info = lod_env_info(env);
2575 struct lod_layout_component *comp = &lo->ldo_comp_entries[comp_idx];
2576 struct filter_fid *ff = &info->lti_ff;
2577 struct lu_buf *buf = &info->lti_buf;
2581 buf->lb_len = sizeof(*ff);
2582 rc = dt_xattr_get(env, dt, buf, XATTR_NAME_FID);
2590 * locd_buf is set if it's called by dir migration, which doesn't check
2593 if (data->locd_buf) {
2594 memset(ff, 0, sizeof(*ff));
2595 ff->ff_parent = *(struct lu_fid *)data->locd_buf->lb_buf;
2597 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
2599 if (lu_fid_eq(lod_object_fid(lo), &ff->ff_parent) &&
2600 ff->ff_layout.ol_comp_id == comp->llc_id)
2603 memset(ff, 0, sizeof(*ff));
2604 ff->ff_parent = *lu_object_fid(&lo->ldo_obj.do_lu);
2607 /* rewrite filter_fid */
2608 ff->ff_parent.f_ver = stripe_idx;
2609 ff->ff_layout.ol_stripe_size = comp->llc_stripe_size;
2610 ff->ff_layout.ol_stripe_count = comp->llc_stripe_count;
2611 ff->ff_layout.ol_comp_id = comp->llc_id;
2612 ff->ff_layout.ol_comp_start = comp->llc_extent.e_start;
2613 ff->ff_layout.ol_comp_end = comp->llc_extent.e_end;
2614 filter_fid_cpu_to_le(ff, ff, sizeof(*ff));
2616 if (data->locd_declare)
2617 rc = lod_sub_declare_xattr_set(env, dt, buf, XATTR_NAME_FID,
2618 LU_XATTR_REPLACE, th);
2620 rc = lod_sub_xattr_set(env, dt, buf, XATTR_NAME_FID,
2621 LU_XATTR_REPLACE, th);
2627 * Reset parent FID on OST object
2629 * Replace parent FID with @dt object FID, which is only called during migration
2630 * to reset the parent FID after the MDT object is migrated to the new MDT, i.e.
2631 * the FID is changed.
2633 * \param[in] env execution environment
2634 * \param[in] dt dt_object whose stripes's parent FID will be reset
2635 * \parem[in] th thandle
2636 * \param[in] declare if it is declare
2638 * \retval 0 if reset succeeds
2639 * \retval negative errno if reset fails
2641 static int lod_replace_parent_fid(const struct lu_env *env,
2642 struct dt_object *dt,
2643 const struct lu_buf *buf,
2644 struct thandle *th, bool declare)
2646 struct lod_object *lo = lod_dt_obj(dt);
2647 struct lod_thread_info *info = lod_env_info(env);
2648 struct filter_fid *ff;
2649 struct lod_obj_stripe_cb_data data = { { 0 } };
2653 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
2655 /* set xattr to each stripes, if needed */
2656 rc = lod_striping_load(env, lo);
2660 if (!lod_obj_is_striped(dt))
2663 if (info->lti_ea_store_size < sizeof(*ff)) {
2664 rc = lod_ea_store_resize(info, sizeof(*ff));
2669 data.locd_declare = declare;
2670 data.locd_stripe_cb = lod_obj_stripe_replace_parent_fid_cb;
2671 data.locd_buf = buf;
2672 rc = lod_obj_for_each_stripe(env, lo, th, &data);
2677 inline __u16 lod_comp_entry_stripe_count(struct lod_object *lo,
2678 struct lod_layout_component *entry,
2681 struct lod_device *lod = lu2lod_dev(lod2lu_obj(lo)->lo_dev);
2685 else if (lod_comp_inited(entry))
2686 return entry->llc_stripe_count;
2687 else if ((__u16)-1 == entry->llc_stripe_count)
2688 return lod->lod_desc.ld_tgt_count;
2690 return lod_get_stripe_count(lod, lo,
2691 entry->llc_stripe_count, false);
2694 static int lod_comp_md_size(struct lod_object *lo, bool is_dir)
2696 int magic, size = 0, i;
2697 struct lod_layout_component *comp_entries;
2699 bool is_composite, is_foreign = false;
2702 comp_cnt = lo->ldo_def_striping->lds_def_comp_cnt;
2703 comp_entries = lo->ldo_def_striping->lds_def_comp_entries;
2705 lo->ldo_def_striping->lds_def_striping_is_composite;
2707 comp_cnt = lo->ldo_comp_cnt;
2708 comp_entries = lo->ldo_comp_entries;
2709 is_composite = lo->ldo_is_composite;
2710 is_foreign = lo->ldo_is_foreign;
2714 return lo->ldo_foreign_lov_size;
2716 LASSERT(comp_cnt != 0 && comp_entries != NULL);
2718 size = sizeof(struct lov_comp_md_v1) +
2719 sizeof(struct lov_comp_md_entry_v1) * comp_cnt;
2720 LASSERT(size % sizeof(__u64) == 0);
2723 for (i = 0; i < comp_cnt; i++) {
2726 magic = comp_entries[i].llc_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
2727 stripe_count = lod_comp_entry_stripe_count(lo, &comp_entries[i],
2729 if (!is_dir && is_composite)
2730 lod_comp_shrink_stripe_count(&comp_entries[i],
2733 size += lov_user_md_size(stripe_count, magic);
2734 LASSERT(size % sizeof(__u64) == 0);
2740 * Declare component add. The xattr name is XATTR_LUSTRE_LOV.add, and
2741 * the xattr value is binary lov_comp_md_v1 which contains component(s)
2744 * \param[in] env execution environment
2745 * \param[in] dt dt_object to add components on
2746 * \param[in] buf buffer contains components to be added
2747 * \parem[in] th thandle
2749 * \retval 0 on success
2750 * \retval negative errno on failure
2752 static int lod_declare_layout_add(const struct lu_env *env,
2753 struct dt_object *dt,
2754 const struct lu_buf *buf,
2757 struct lod_thread_info *info = lod_env_info(env);
2758 struct lod_layout_component *comp_array, *lod_comp, *old_array;
2759 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2760 struct dt_object *next = dt_object_child(dt);
2761 struct lov_desc *desc = &d->lod_desc;
2762 struct lod_object *lo = lod_dt_obj(dt);
2763 struct lov_user_md_v3 *v3;
2764 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2766 int i, rc, array_cnt, old_array_cnt;
2769 LASSERT(lo->ldo_is_composite);
2771 if (lo->ldo_flr_state != LCM_FL_NONE)
2774 rc = lod_verify_striping(d, lo, buf, false);
2778 magic = comp_v1->lcm_magic;
2779 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2780 lustre_swab_lov_comp_md_v1(comp_v1);
2781 magic = comp_v1->lcm_magic;
2784 if (magic != LOV_USER_MAGIC_COMP_V1)
2787 array_cnt = lo->ldo_comp_cnt + comp_v1->lcm_entry_count;
2788 OBD_ALLOC(comp_array, sizeof(*comp_array) * array_cnt);
2789 if (comp_array == NULL)
2792 memcpy(comp_array, lo->ldo_comp_entries,
2793 sizeof(*comp_array) * lo->ldo_comp_cnt);
2795 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2796 struct lov_user_md_v1 *v1;
2797 struct lu_extent *ext;
2799 v1 = (struct lov_user_md *)((char *)comp_v1 +
2800 comp_v1->lcm_entries[i].lcme_offset);
2801 ext = &comp_v1->lcm_entries[i].lcme_extent;
2803 lod_comp = &comp_array[lo->ldo_comp_cnt + i];
2804 lod_comp->llc_extent.e_start = ext->e_start;
2805 lod_comp->llc_extent.e_end = ext->e_end;
2806 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
2807 lod_comp->llc_flags = comp_v1->lcm_entries[i].lcme_flags;
2809 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
2810 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
2811 lod_adjust_stripe_info(lod_comp, desc, 0);
2813 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2814 v3 = (struct lov_user_md_v3 *) v1;
2815 if (v3->lmm_pool_name[0] != '\0') {
2816 rc = lod_set_pool(&lod_comp->llc_pool,
2824 old_array = lo->ldo_comp_entries;
2825 old_array_cnt = lo->ldo_comp_cnt;
2827 lo->ldo_comp_entries = comp_array;
2828 lo->ldo_comp_cnt = array_cnt;
2830 /* No need to increase layout generation here, it will be increased
2831 * later when generating component ID for the new components */
2833 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2834 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
2835 XATTR_NAME_LOV, 0, th);
2837 lo->ldo_comp_entries = old_array;
2838 lo->ldo_comp_cnt = old_array_cnt;
2842 OBD_FREE(old_array, sizeof(*lod_comp) * old_array_cnt);
2844 LASSERT(lo->ldo_mirror_count == 1);
2845 lo->ldo_mirrors[0].lme_end = array_cnt - 1;
2850 for (i = lo->ldo_comp_cnt; i < array_cnt; i++) {
2851 lod_comp = &comp_array[i];
2852 if (lod_comp->llc_pool != NULL) {
2853 OBD_FREE(lod_comp->llc_pool,
2854 strlen(lod_comp->llc_pool) + 1);
2855 lod_comp->llc_pool = NULL;
2858 OBD_FREE(comp_array, sizeof(*comp_array) * array_cnt);
2863 * Declare component set. The xattr is name XATTR_LUSTRE_LOV.set.$field,
2864 * the '$field' can only be 'flags' now. The xattr value is binary
2865 * lov_comp_md_v1 which contains the component ID(s) and the value of
2866 * the field to be modified.
2868 * \param[in] env execution environment
2869 * \param[in] dt dt_object to be modified
2870 * \param[in] op operation string, like "set.flags"
2871 * \param[in] buf buffer contains components to be set
2872 * \parem[in] th thandle
2874 * \retval 0 on success
2875 * \retval negative errno on failure
2877 static int lod_declare_layout_set(const struct lu_env *env,
2878 struct dt_object *dt,
2879 char *op, const struct lu_buf *buf,
2882 struct lod_layout_component *lod_comp;
2883 struct lod_thread_info *info = lod_env_info(env);
2884 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2885 struct lod_object *lo = lod_dt_obj(dt);
2886 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2889 bool changed = false;
2892 if (strcmp(op, "set.flags") != 0) {
2893 CDEBUG(D_LAYOUT, "%s: operation (%s) not supported.\n",
2894 lod2obd(d)->obd_name, op);
2898 magic = comp_v1->lcm_magic;
2899 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
2900 lustre_swab_lov_comp_md_v1(comp_v1);
2901 magic = comp_v1->lcm_magic;
2904 if (magic != LOV_USER_MAGIC_COMP_V1)
2907 if (comp_v1->lcm_entry_count == 0) {
2908 CDEBUG(D_LAYOUT, "%s: entry count is zero.\n",
2909 lod2obd(d)->obd_name);
2913 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
2914 __u32 id = comp_v1->lcm_entries[i].lcme_id;
2915 __u32 flags = comp_v1->lcm_entries[i].lcme_flags;
2916 __u32 mirror_flag = flags & LCME_MIRROR_FLAGS;
2917 bool neg = flags & LCME_FL_NEG;
2919 if (flags & LCME_FL_INIT) {
2921 lod_striping_free(env, lo);
2925 flags &= ~(LCME_MIRROR_FLAGS | LCME_FL_NEG);
2926 for (j = 0; j < lo->ldo_comp_cnt; j++) {
2927 lod_comp = &lo->ldo_comp_entries[j];
2929 /* lfs only put one flag in each entry */
2930 if ((flags && id != lod_comp->llc_id) ||
2931 (mirror_flag && mirror_id_of(id) !=
2932 mirror_id_of(lod_comp->llc_id)))
2937 lod_comp->llc_flags &= ~flags;
2939 lod_comp->llc_flags &= ~mirror_flag;
2942 lod_comp->llc_flags |= flags;
2944 lod_comp->llc_flags |= mirror_flag;
2945 if (mirror_flag & LCME_FL_NOSYNC)
2946 lod_comp->llc_timestamp =
2947 ktime_get_real_seconds();
2955 CDEBUG(D_LAYOUT, "%s: requested component(s) not found.\n",
2956 lod2obd(d)->obd_name);
2960 lod_obj_inc_layout_gen(lo);
2962 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
2963 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), &info->lti_buf,
2964 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
2969 * Declare component deletion. The xattr name is XATTR_LUSTRE_LOV.del,
2970 * and the xattr value is a unique component ID or a special lcme_id.
2972 * \param[in] env execution environment
2973 * \param[in] dt dt_object to be operated on
2974 * \param[in] buf buffer contains component ID or lcme_id
2975 * \parem[in] th thandle
2977 * \retval 0 on success
2978 * \retval negative errno on failure
2980 static int lod_declare_layout_del(const struct lu_env *env,
2981 struct dt_object *dt,
2982 const struct lu_buf *buf,
2985 struct lod_thread_info *info = lod_env_info(env);
2986 struct dt_object *next = dt_object_child(dt);
2987 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2988 struct lod_object *lo = lod_dt_obj(dt);
2989 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2990 struct lov_comp_md_v1 *comp_v1 = buf->lb_buf;
2991 __u32 magic, id, flags, neg_flags = 0;
2995 LASSERT(lo->ldo_is_composite);
2997 if (lo->ldo_flr_state != LCM_FL_NONE)
3000 magic = comp_v1->lcm_magic;
3001 if (magic == __swab32(LOV_USER_MAGIC_COMP_V1)) {
3002 lustre_swab_lov_comp_md_v1(comp_v1);
3003 magic = comp_v1->lcm_magic;
3006 if (magic != LOV_USER_MAGIC_COMP_V1)
3009 id = comp_v1->lcm_entries[0].lcme_id;
3010 flags = comp_v1->lcm_entries[0].lcme_flags;
3012 if (id > LCME_ID_MAX || (flags & ~LCME_KNOWN_FLAGS)) {
3013 CDEBUG(D_LAYOUT, "%s: invalid component id %#x, flags %#x\n",
3014 lod2obd(d)->obd_name, id, flags);
3018 if (id != LCME_ID_INVAL && flags != 0) {
3019 CDEBUG(D_LAYOUT, "%s: specified both id and flags.\n",
3020 lod2obd(d)->obd_name);
3024 if (id == LCME_ID_INVAL && !flags) {
3025 CDEBUG(D_LAYOUT, "%s: no id or flags specified.\n",
3026 lod2obd(d)->obd_name);
3030 if (flags & LCME_FL_NEG) {
3031 neg_flags = flags & ~LCME_FL_NEG;
3035 left = lo->ldo_comp_cnt;
3039 for (i = (lo->ldo_comp_cnt - 1); i >= 0; i--) {
3040 struct lod_layout_component *lod_comp;
3042 lod_comp = &lo->ldo_comp_entries[i];
3044 if (id != LCME_ID_INVAL && id != lod_comp->llc_id)
3046 else if (flags && !(flags & lod_comp->llc_flags))
3048 else if (neg_flags && (neg_flags & lod_comp->llc_flags))
3051 if (left != (i + 1)) {
3052 CDEBUG(D_LAYOUT, "%s: this deletion will create "
3053 "a hole.\n", lod2obd(d)->obd_name);
3058 /* Mark the component as deleted */
3059 lod_comp->llc_id = LCME_ID_INVAL;
3061 /* Not instantiated component */
3062 if (lod_comp->llc_stripe == NULL)
3065 LASSERT(lod_comp->llc_stripe_count > 0);
3066 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
3067 struct dt_object *obj = lod_comp->llc_stripe[j];
3071 rc = lod_sub_declare_destroy(env, obj, th);
3077 LASSERTF(left >= 0, "left = %d\n", left);
3078 if (left == lo->ldo_comp_cnt) {
3079 CDEBUG(D_LAYOUT, "%s: requested component id:%#x not found\n",
3080 lod2obd(d)->obd_name, id);
3084 memset(attr, 0, sizeof(*attr));
3085 attr->la_valid = LA_SIZE;
3086 rc = lod_sub_declare_attr_set(env, next, attr, th);
3091 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
3092 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
3093 XATTR_NAME_LOV, 0, th);
3095 rc = lod_sub_declare_xattr_del(env, next, XATTR_NAME_LOV, th);
3102 * Declare layout add/set/del operations issued by special xattr names:
3104 * XATTR_LUSTRE_LOV.add add component(s) to existing file
3105 * XATTR_LUSTRE_LOV.del delete component(s) from existing file
3106 * XATTR_LUSTRE_LOV.set.$field set specified field of certain component(s)
3108 * \param[in] env execution environment
3109 * \param[in] dt object
3110 * \param[in] name name of xattr
3111 * \param[in] buf lu_buf contains xattr value
3112 * \param[in] th transaction handle
3114 * \retval 0 on success
3115 * \retval negative if failed
3117 static int lod_declare_modify_layout(const struct lu_env *env,
3118 struct dt_object *dt,
3120 const struct lu_buf *buf,
3123 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
3124 struct lod_object *lo = lod_dt_obj(dt);
3126 int rc, len = strlen(XATTR_LUSTRE_LOV);
3129 LASSERT(dt_object_exists(dt));
3131 if (strlen(name) <= len || name[len] != '.') {
3132 CDEBUG(D_LAYOUT, "%s: invalid xattr name: %s\n",
3133 lod2obd(d)->obd_name, name);
3138 rc = lod_striping_load(env, lo);
3142 /* the layout to be modified must be a composite layout */
3143 if (!lo->ldo_is_composite) {
3144 CDEBUG(D_LAYOUT, "%s: object "DFID" isn't a composite file.\n",
3145 lod2obd(d)->obd_name, PFID(lu_object_fid(&dt->do_lu)));
3146 GOTO(unlock, rc = -EINVAL);
3149 op = (char *)name + len;
3150 if (strcmp(op, "add") == 0) {
3151 rc = lod_declare_layout_add(env, dt, buf, th);
3152 } else if (strcmp(op, "del") == 0) {
3153 rc = lod_declare_layout_del(env, dt, buf, th);
3154 } else if (strncmp(op, "set", strlen("set")) == 0) {
3155 rc = lod_declare_layout_set(env, dt, op, buf, th);
3157 CDEBUG(D_LAYOUT, "%s: unsupported xattr name:%s\n",
3158 lod2obd(d)->obd_name, name);
3159 GOTO(unlock, rc = -ENOTSUPP);
3163 lod_striping_free(env, lo);
3169 * Convert a plain file lov_mds_md to a composite layout.
3171 * \param[in,out] info the thread info::lti_ea_store buffer contains little
3172 * endian plain file layout
3174 * \retval 0 on success, <0 on failure
3176 static int lod_layout_convert(struct lod_thread_info *info)
3178 struct lov_mds_md *lmm = info->lti_ea_store;
3179 struct lov_mds_md *lmm_save;
3180 struct lov_comp_md_v1 *lcm;
3181 struct lov_comp_md_entry_v1 *lcme;
3187 /* realloc buffer to a composite layout which contains one component */
3188 blob_size = lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3189 le32_to_cpu(lmm->lmm_magic));
3190 size = sizeof(*lcm) + sizeof(*lcme) + blob_size;
3192 OBD_ALLOC_LARGE(lmm_save, blob_size);
3194 GOTO(out, rc = -ENOMEM);
3196 memcpy(lmm_save, lmm, blob_size);
3198 if (info->lti_ea_store_size < size) {
3199 rc = lod_ea_store_resize(info, size);
3204 lcm = info->lti_ea_store;
3205 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
3206 lcm->lcm_size = cpu_to_le32(size);
3207 lcm->lcm_layout_gen = cpu_to_le32(le16_to_cpu(
3208 lmm_save->lmm_layout_gen));
3209 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
3210 lcm->lcm_entry_count = cpu_to_le16(1);
3211 lcm->lcm_mirror_count = 0;
3213 lcme = &lcm->lcm_entries[0];
3214 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
3215 lcme->lcme_extent.e_start = 0;
3216 lcme->lcme_extent.e_end = cpu_to_le64(OBD_OBJECT_EOF);
3217 lcme->lcme_offset = cpu_to_le32(sizeof(*lcm) + sizeof(*lcme));
3218 lcme->lcme_size = cpu_to_le32(blob_size);
3220 memcpy((char *)lcm + lcme->lcme_offset, (char *)lmm_save, blob_size);
3225 OBD_FREE_LARGE(lmm_save, blob_size);
3230 * Merge layouts to form a mirrored file.
3232 static int lod_declare_layout_merge(const struct lu_env *env,
3233 struct dt_object *dt, const struct lu_buf *mbuf,
3236 struct lod_thread_info *info = lod_env_info(env);
3237 struct lu_buf *buf = &info->lti_buf;
3238 struct lod_object *lo = lod_dt_obj(dt);
3239 struct lov_comp_md_v1 *lcm;
3240 struct lov_comp_md_v1 *cur_lcm;
3241 struct lov_comp_md_v1 *merge_lcm;
3242 struct lov_comp_md_entry_v1 *lcme;
3243 struct lov_mds_md_v1 *lmm;
3246 __u16 cur_entry_count;
3247 __u16 merge_entry_count;
3249 __u16 mirror_id = 0;
3256 merge_lcm = mbuf->lb_buf;
3257 if (mbuf->lb_len < sizeof(*merge_lcm))
3260 /* must be an existing layout from disk */
3261 if (le32_to_cpu(merge_lcm->lcm_magic) != LOV_MAGIC_COMP_V1)
3264 merge_entry_count = le16_to_cpu(merge_lcm->lcm_entry_count);
3266 /* do not allow to merge two mirrored files */
3267 if (le16_to_cpu(merge_lcm->lcm_mirror_count))
3270 /* verify the target buffer */
3271 rc = lod_get_lov_ea(env, lo);
3273 RETURN(rc ? : -ENODATA);
3275 cur_lcm = info->lti_ea_store;
3276 switch (le32_to_cpu(cur_lcm->lcm_magic)) {
3279 rc = lod_layout_convert(info);
3281 case LOV_MAGIC_COMP_V1:
3291 /* info->lti_ea_store could be reallocated in lod_layout_convert() */
3292 cur_lcm = info->lti_ea_store;
3293 cur_entry_count = le16_to_cpu(cur_lcm->lcm_entry_count);
3295 /* 'lcm_mirror_count + 1' is the current # of mirrors the file has */
3296 mirror_count = le16_to_cpu(cur_lcm->lcm_mirror_count) + 1;
3297 if (mirror_count + 1 > LUSTRE_MIRROR_COUNT_MAX)
3300 /* size of new layout */
3301 size = le32_to_cpu(cur_lcm->lcm_size) +
3302 le32_to_cpu(merge_lcm->lcm_size) - sizeof(*cur_lcm);
3304 memset(buf, 0, sizeof(*buf));
3305 lu_buf_alloc(buf, size);
3306 if (buf->lb_buf == NULL)
3310 memcpy(lcm, cur_lcm, sizeof(*lcm) + cur_entry_count * sizeof(*lcme));
3312 offset = sizeof(*lcm) +
3313 sizeof(*lcme) * (cur_entry_count + merge_entry_count);
3314 for (i = 0; i < cur_entry_count; i++) {
3315 struct lov_comp_md_entry_v1 *cur_lcme;
3317 lcme = &lcm->lcm_entries[i];
3318 cur_lcme = &cur_lcm->lcm_entries[i];
3320 lcme->lcme_offset = cpu_to_le32(offset);
3321 memcpy((char *)lcm + offset,
3322 (char *)cur_lcm + le32_to_cpu(cur_lcme->lcme_offset),
3323 le32_to_cpu(lcme->lcme_size));
3325 offset += le32_to_cpu(lcme->lcme_size);
3327 if (mirror_count == 1 &&
3328 mirror_id_of(le32_to_cpu(lcme->lcme_id)) == 0) {
3329 /* Add mirror from a non-flr file, create new mirror ID.
3330 * Otherwise, keep existing mirror's component ID, used
3331 * for mirror extension.
3333 id = pflr_id(1, i + 1);
3334 lcme->lcme_id = cpu_to_le32(id);
3337 id = MAX(le32_to_cpu(lcme->lcme_id), id);
3340 mirror_id = mirror_id_of(id) + 1;
3342 /* check if first entry in new layout is DOM */
3343 lmm = (struct lov_mds_md_v1 *)((char *)merge_lcm +
3344 merge_lcm->lcm_entries[0].lcme_offset);
3345 merge_has_dom = lov_pattern(le32_to_cpu(lmm->lmm_pattern)) ==
3348 for (i = 0; i < merge_entry_count; i++) {
3349 struct lov_comp_md_entry_v1 *merge_lcme;
3351 merge_lcme = &merge_lcm->lcm_entries[i];
3352 lcme = &lcm->lcm_entries[cur_entry_count + i];
3354 *lcme = *merge_lcme;
3355 lcme->lcme_offset = cpu_to_le32(offset);
3356 if (merge_has_dom && i == 0)
3357 lcme->lcme_flags |= cpu_to_le32(LCME_FL_STALE);
3359 id = pflr_id(mirror_id, i + 1);
3360 lcme->lcme_id = cpu_to_le32(id);
3362 memcpy((char *)lcm + offset,
3363 (char *)merge_lcm + le32_to_cpu(merge_lcme->lcme_offset),
3364 le32_to_cpu(lcme->lcme_size));
3366 offset += le32_to_cpu(lcme->lcme_size);
3369 /* fixup layout information */
3370 lod_obj_inc_layout_gen(lo);
3371 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3372 lcm->lcm_size = cpu_to_le32(size);
3373 lcm->lcm_entry_count = cpu_to_le16(cur_entry_count + merge_entry_count);
3374 lcm->lcm_mirror_count = cpu_to_le16(mirror_count);
3375 if ((le16_to_cpu(lcm->lcm_flags) & LCM_FL_FLR_MASK) == LCM_FL_NONE)
3376 lcm->lcm_flags = cpu_to_le32(LCM_FL_RDONLY);
3378 rc = lod_striping_reload(env, lo, buf);
3382 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), buf,
3383 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3391 * Split layouts, just set the LOVEA with the layout from mbuf.
3393 static int lod_declare_layout_split(const struct lu_env *env,
3394 struct dt_object *dt, const struct lu_buf *mbuf,
3397 struct lod_object *lo = lod_dt_obj(dt);
3398 struct lov_comp_md_v1 *lcm = mbuf->lb_buf;
3402 lod_obj_inc_layout_gen(lo);
3403 lcm->lcm_layout_gen = cpu_to_le32(lo->ldo_layout_gen);
3405 rc = lod_striping_reload(env, lo, mbuf);
3409 rc = lod_sub_declare_xattr_set(env, dt_object_child(dt), mbuf,
3410 XATTR_NAME_LOV, LU_XATTR_REPLACE, th);
3415 * Implementation of dt_object_operations::do_declare_xattr_set.
3417 * \see dt_object_operations::do_declare_xattr_set() in the API description
3420 * the extension to the API:
3421 * - declaring LOVEA requests striping creation
3422 * - LU_XATTR_REPLACE means layout swap
3424 static int lod_declare_xattr_set(const struct lu_env *env,
3425 struct dt_object *dt,
3426 const struct lu_buf *buf,
3427 const char *name, int fl,
3430 struct dt_object *next = dt_object_child(dt);
3431 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3436 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
3437 if ((S_ISREG(mode) || mode == 0) &&
3438 !(fl & (LU_XATTR_REPLACE | LU_XATTR_MERGE | LU_XATTR_SPLIT)) &&
3439 (strcmp(name, XATTR_NAME_LOV) == 0 ||
3440 strcmp(name, XATTR_LUSTRE_LOV) == 0)) {
3442 * this is a request to create object's striping.
3444 * allow to declare predefined striping on a new (!mode) object
3445 * which is supposed to be replay of regular file creation
3446 * (when LOV setting is declared)
3448 * LU_XATTR_REPLACE is set to indicate a layout swap
3450 if (dt_object_exists(dt)) {
3451 rc = dt_attr_get(env, next, attr);
3455 memset(attr, 0, sizeof(*attr));
3456 attr->la_valid = LA_TYPE | LA_MODE;
3457 attr->la_mode = S_IFREG;
3459 rc = lod_declare_striped_create(env, dt, attr, buf, th);
3460 } else if (fl & LU_XATTR_MERGE) {
3461 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3462 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3463 rc = lod_declare_layout_merge(env, dt, buf, th);
3464 } else if (fl & LU_XATTR_SPLIT) {
3465 LASSERT(strcmp(name, XATTR_NAME_LOV) == 0 ||
3466 strcmp(name, XATTR_LUSTRE_LOV) == 0);
3467 rc = lod_declare_layout_split(env, dt, buf, th);
3468 } else if (S_ISREG(mode) &&
3469 strlen(name) > strlen(XATTR_LUSTRE_LOV) + 1 &&
3470 strncmp(name, XATTR_LUSTRE_LOV,
3471 strlen(XATTR_LUSTRE_LOV)) == 0) {
3473 * this is a request to modify object's striping.
3474 * add/set/del component(s).
3476 if (!dt_object_exists(dt))
3479 rc = lod_declare_modify_layout(env, dt, name, buf, th);
3480 } else if (strncmp(name, XATTR_NAME_LMV, strlen(XATTR_NAME_LMV)) == 0 &&
3481 strlen(name) > strlen(XATTR_NAME_LMV) + 1) {
3482 const char *op = name + strlen(XATTR_NAME_LMV) + 1;
3485 if (strcmp(op, "add") == 0)
3486 rc = lod_dir_declare_layout_add(env, dt, buf, th);
3487 else if (strcmp(op, "del") == 0)
3488 rc = lod_dir_declare_layout_delete(env, dt, buf, th);
3489 else if (strcmp(op, "set") == 0)
3490 rc = lod_sub_declare_xattr_set(env, next, buf,
3491 XATTR_NAME_LMV, fl, th);
3494 } else if (S_ISDIR(mode)) {
3495 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
3496 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3497 rc = lod_replace_parent_fid(env, dt, buf, th, true);
3499 rc = lod_sub_declare_xattr_set(env, next, buf, name, fl, th);
3506 * Apply xattr changes to the object.
3508 * Applies xattr changes to the object and the stripes if the latter exist.
3510 * \param[in] env execution environment
3511 * \param[in] dt object
3512 * \param[in] buf buffer pointing to the new value of xattr
3513 * \param[in] name name of xattr
3514 * \param[in] fl flags
3515 * \param[in] th transaction handle
3517 * \retval 0 on success
3518 * \retval negative if failed
3520 static int lod_xattr_set_internal(const struct lu_env *env,
3521 struct dt_object *dt,
3522 const struct lu_buf *buf,
3523 const char *name, int fl,
3526 struct dt_object *next = dt_object_child(dt);
3527 struct lod_object *lo = lod_dt_obj(dt);
3532 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
3533 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3536 /* Note: Do not set LinkEA on sub-stripes, otherwise
3537 * it will confuse the fid2path process(see mdt_path_current()).
3538 * The linkEA between master and sub-stripes is set in
3539 * lod_xattr_set_lmv(). */
3540 if (lo->ldo_dir_stripe_count == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
3543 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3544 if (!lo->ldo_stripe[i])
3547 if (!dt_object_exists(lo->ldo_stripe[i]))
3550 rc = lod_sub_xattr_set(env, lo->ldo_stripe[i], buf, name,
3560 * Delete an extended attribute.
3562 * Deletes specified xattr from the object and the stripes if the latter exist.
3564 * \param[in] env execution environment
3565 * \param[in] dt object
3566 * \param[in] name name of xattr
3567 * \param[in] th transaction handle
3569 * \retval 0 on success
3570 * \retval negative if failed
3572 static int lod_xattr_del_internal(const struct lu_env *env,
3573 struct dt_object *dt,
3574 const char *name, struct thandle *th)
3576 struct dt_object *next = dt_object_child(dt);
3577 struct lod_object *lo = lod_dt_obj(dt);
3582 rc = lod_sub_xattr_del(env, next, name, th);
3583 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
3586 if (lo->ldo_dir_stripe_count == 0)
3589 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3590 LASSERT(lo->ldo_stripe[i]);
3592 rc = lod_sub_xattr_del(env, lo->ldo_stripe[i], name, th);
3601 * Set default striping on a directory.
3603 * Sets specified striping on a directory object unless it matches the default
3604 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3605 * EA. This striping will be used when regular file is being created in this
3608 * \param[in] env execution environment
3609 * \param[in] dt the striped object
3610 * \param[in] buf buffer with the striping
3611 * \param[in] name name of EA
3612 * \param[in] fl xattr flag (see OSD API description)
3613 * \param[in] th transaction handle
3615 * \retval 0 on success
3616 * \retval negative if failed
3618 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
3619 struct dt_object *dt,
3620 const struct lu_buf *buf,
3621 const char *name, int fl,
3624 struct lov_user_md_v1 *lum;
3625 struct lov_user_md_v3 *v3 = NULL;
3626 const char *pool_name = NULL;
3631 LASSERT(buf != NULL && buf->lb_buf != NULL);
3634 switch (lum->lmm_magic) {
3635 case LOV_USER_MAGIC_SPECIFIC:
3636 case LOV_USER_MAGIC_V3:
3638 if (v3->lmm_pool_name[0] != '\0')
3639 pool_name = v3->lmm_pool_name;
3641 case LOV_USER_MAGIC_V1:
3642 /* if { size, offset, count } = { 0, -1, 0 } and no pool
3643 * (i.e. all default values specified) then delete default
3644 * striping from dir. */
3646 "set default striping: sz %u # %u offset %d %s %s\n",
3647 (unsigned)lum->lmm_stripe_size,
3648 (unsigned)lum->lmm_stripe_count,
3649 (int)lum->lmm_stripe_offset,
3650 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
3652 is_del = LOVEA_DELETE_VALUES(lum->lmm_stripe_size,
3653 lum->lmm_stripe_count,
3654 lum->lmm_stripe_offset,
3657 case LOV_USER_MAGIC_COMP_V1:
3659 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lum;
3660 struct lov_comp_md_entry_v1 *lcme;
3663 comp_cnt = le16_to_cpu(lcm->lcm_entry_count);
3664 for (i = 0; i < comp_cnt; i++) {
3665 lcme = &lcm->lcm_entries[i];
3666 if (lcme->lcme_flags & cpu_to_le32(LCME_FL_EXTENSION)) {
3667 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_SEL);
3676 CERROR("Invalid magic %x\n", lum->lmm_magic);
3681 rc = lod_xattr_del_internal(env, dt, name, th);
3685 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3692 * Set default striping on a directory object.
3694 * Sets specified striping on a directory object unless it matches the default
3695 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
3696 * EA. This striping will be used when a new directory is being created in the
3699 * \param[in] env execution environment
3700 * \param[in] dt the striped object
3701 * \param[in] buf buffer with the striping
3702 * \param[in] name name of EA
3703 * \param[in] fl xattr flag (see OSD API description)
3704 * \param[in] th transaction handle
3706 * \retval 0 on success
3707 * \retval negative if failed
3709 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
3710 struct dt_object *dt,
3711 const struct lu_buf *buf,
3712 const char *name, int fl,
3715 struct lmv_user_md_v1 *lum;
3720 LASSERT(buf != NULL && buf->lb_buf != NULL);
3724 "set default stripe_count # %u stripe_offset %d hash %u\n",
3725 le32_to_cpu(lum->lum_stripe_count),
3726 (int)le32_to_cpu(lum->lum_stripe_offset),
3727 le32_to_cpu(lum->lum_hash_type));
3729 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
3730 le32_to_cpu(lum->lum_stripe_offset)) &&
3731 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC &&
3732 !(le32_to_cpu(lum->lum_hash_type) & LMV_HASH_FLAG_SPACE)) {
3733 rc = lod_xattr_del_internal(env, dt, name, th);
3737 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
3746 * Turn directory into a striped directory.
3748 * During replay the client sends the striping created before MDT
3749 * failure, then the layer above LOD sends this defined striping
3750 * using ->do_xattr_set(), so LOD uses this method to replay creation
3751 * of the stripes. Notice the original information for the striping
3752 * (#stripes, FIDs, etc) was transferred in declare path.
3754 * \param[in] env execution environment
3755 * \param[in] dt the striped object
3756 * \param[in] buf not used currently
3757 * \param[in] name not used currently
3758 * \param[in] fl xattr flag (see OSD API description)
3759 * \param[in] th transaction handle
3761 * \retval 0 on success
3762 * \retval negative if failed
3764 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
3765 const struct lu_buf *buf, const char *name,
3766 int fl, struct thandle *th)
3768 struct lod_object *lo = lod_dt_obj(dt);
3769 struct lod_thread_info *info = lod_env_info(env);
3770 struct lu_attr *attr = &info->lti_attr;
3771 struct dt_object_format *dof = &info->lti_format;
3772 struct lu_buf lmv_buf;
3773 struct lu_buf slave_lmv_buf;
3774 struct lmv_mds_md_v1 *lmm;
3775 struct lmv_mds_md_v1 *slave_lmm = NULL;
3776 struct dt_insert_rec *rec = &info->lti_dt_rec;
3781 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3784 /* The stripes are supposed to be allocated in declare phase,
3785 * if there are no stripes being allocated, it will skip */
3786 if (lo->ldo_dir_stripe_count == 0) {
3787 if (lo->ldo_dir_is_foreign) {
3788 rc = lod_sub_xattr_set(env, dt_object_child(dt), buf,
3789 XATTR_NAME_LMV, fl, th);
3796 rc = dt_attr_get(env, dt_object_child(dt), attr);
3800 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME |
3801 LA_MODE | LA_UID | LA_GID | LA_TYPE | LA_PROJID;
3802 dof->dof_type = DFT_DIR;
3804 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
3807 lmm = lmv_buf.lb_buf;
3809 OBD_ALLOC_PTR(slave_lmm);
3810 if (slave_lmm == NULL)
3813 lod_prep_slave_lmv_md(slave_lmm, lmm);
3814 slave_lmv_buf.lb_buf = slave_lmm;
3815 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
3817 rec->rec_type = S_IFDIR;
3818 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
3819 struct dt_object *dto = lo->ldo_stripe[i];
3820 char *stripe_name = info->lti_key;
3821 struct lu_name *sname;
3822 struct linkea_data ldata = { NULL };
3823 struct lu_buf linkea_buf;
3825 /* OBD_FAIL_MDS_STRIPE_FID may leave stripe uninitialized */
3829 /* fail a remote stripe creation */
3830 if (i && OBD_FAIL_CHECK(OBD_FAIL_MDS_STRIPE_CREATE))
3833 /* if it's source stripe of migrating directory, don't create */
3834 if (!((lo->ldo_dir_hash_type & LMV_HASH_FLAG_MIGRATION) &&
3835 i >= lo->ldo_dir_migrate_offset)) {
3836 dt_write_lock(env, dto, DT_TGT_CHILD);
3837 rc = lod_sub_create(env, dto, attr, NULL, dof, th);
3839 dt_write_unlock(env, dto);
3843 rc = lod_sub_ref_add(env, dto, th);
3844 dt_write_unlock(env, dto);
3848 rec->rec_fid = lu_object_fid(&dto->do_lu);
3849 rc = lod_sub_insert(env, dto,
3850 (const struct dt_rec *)rec,
3851 (const struct dt_key *)dot, th);
3856 rec->rec_fid = lu_object_fid(&dt->do_lu);
3857 rc = lod_sub_insert(env, dto, (struct dt_rec *)rec,
3858 (const struct dt_key *)dotdot, th);
3862 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
3863 cfs_fail_val != i) {
3864 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
3866 slave_lmm->lmv_master_mdt_index =
3869 slave_lmm->lmv_master_mdt_index =
3872 rc = lod_sub_xattr_set(env, dto, &slave_lmv_buf,
3873 XATTR_NAME_LMV, fl, th);
3878 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
3880 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3881 PFID(lu_object_fid(&dto->do_lu)), i + 1);
3883 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3884 PFID(lu_object_fid(&dto->do_lu)), i);
3886 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
3887 rc = linkea_links_new(&ldata, &info->lti_linkea_buf,
3888 sname, lu_object_fid(&dt->do_lu));
3892 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
3893 linkea_buf.lb_len = ldata.ld_leh->leh_len;
3894 rc = lod_sub_xattr_set(env, dto, &linkea_buf,
3895 XATTR_NAME_LINK, 0, th);
3899 rec->rec_fid = lu_object_fid(&dto->do_lu);
3900 rc = lod_sub_insert(env, dt_object_child(dt),
3901 (const struct dt_rec *)rec,
3902 (const struct dt_key *)stripe_name, th);
3906 rc = lod_sub_ref_add(env, dt_object_child(dt), th);
3911 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
3912 rc = lod_sub_xattr_set(env, dt_object_child(dt),
3913 &lmv_buf, XATTR_NAME_LMV, fl, th);
3915 if (slave_lmm != NULL)
3916 OBD_FREE_PTR(slave_lmm);
3922 * Helper function to declare/execute creation of a striped directory
3924 * Called in declare/create object path, prepare striping for a directory
3925 * and prepare defaults data striping for the objects to be created in
3926 * that directory. Notice the function calls "declaration" or "execution"
3927 * methods depending on \a declare param. This is a consequence of the
3928 * current approach while we don't have natural distributed transactions:
3929 * we basically execute non-local updates in the declare phase. So, the
3930 * arguments for the both phases are the same and this is the reason for
3931 * this function to exist.
3933 * \param[in] env execution environment
3934 * \param[in] dt object
3935 * \param[in] attr attributes the stripes will be created with
3936 * \param[in] lmu lmv_user_md if MDT indices are specified
3937 * \param[in] dof format of stripes (see OSD API description)
3938 * \param[in] th transaction handle
3939 * \param[in] declare where to call "declare" or "execute" methods
3941 * \retval 0 on success
3942 * \retval negative if failed
3944 static int lod_dir_striping_create_internal(const struct lu_env *env,
3945 struct dt_object *dt,
3946 struct lu_attr *attr,
3947 const struct lu_buf *lmu,
3948 struct dt_object_format *dof,
3952 struct lod_thread_info *info = lod_env_info(env);
3953 struct lod_object *lo = lod_dt_obj(dt);
3954 const struct lod_default_striping *lds = lo->ldo_def_striping;
3958 LASSERT(ergo(lds != NULL,
3959 lds->lds_def_striping_set ||
3960 lds->lds_dir_def_striping_set));
3962 if (!LMVEA_DELETE_VALUES(lo->ldo_dir_stripe_count,
3963 lo->ldo_dir_stripe_offset)) {
3965 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
3966 int stripe_count = lo->ldo_dir_stripe_count;
3968 if (info->lti_ea_store_size < sizeof(*v1)) {
3969 rc = lod_ea_store_resize(info, sizeof(*v1));
3972 v1 = info->lti_ea_store;
3975 memset(v1, 0, sizeof(*v1));
3976 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
3977 v1->lum_stripe_count = cpu_to_le32(stripe_count);
3978 v1->lum_stripe_offset =
3979 cpu_to_le32(lo->ldo_dir_stripe_offset);
3981 info->lti_buf.lb_buf = v1;
3982 info->lti_buf.lb_len = sizeof(*v1);
3983 lmu = &info->lti_buf;
3987 rc = lod_declare_xattr_set_lmv(env, dt, attr, lmu, dof,
3990 rc = lod_xattr_set_lmv(env, dt, lmu, XATTR_NAME_LMV, 0,
3995 /* foreign LMV EA case */
3997 struct lmv_foreign_md *lfm = lmu->lb_buf;
3999 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
4000 rc = lod_declare_xattr_set_lmv(env, dt, attr,
4004 if (lo->ldo_dir_is_foreign) {
4005 LASSERT(lo->ldo_foreign_lmv != NULL &&
4006 lo->ldo_foreign_lmv_size > 0);
4007 info->lti_buf.lb_buf = lo->ldo_foreign_lmv;
4008 info->lti_buf.lb_len = lo->ldo_foreign_lmv_size;
4009 lmu = &info->lti_buf;
4010 rc = lod_xattr_set_lmv(env, dt, lmu,
4011 XATTR_NAME_LMV, 0, th);
4016 /* Transfer default LMV striping from the parent */
4017 if (lds != NULL && lds->lds_dir_def_striping_set &&
4018 !(LMVEA_DELETE_VALUES(lds->lds_dir_def_stripe_count,
4019 lds->lds_dir_def_stripe_offset) &&
4020 le32_to_cpu(lds->lds_dir_def_hash_type) !=
4021 LMV_HASH_TYPE_UNKNOWN)) {
4022 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
4024 if (info->lti_ea_store_size < sizeof(*v1)) {
4025 rc = lod_ea_store_resize(info, sizeof(*v1));
4028 v1 = info->lti_ea_store;
4031 memset(v1, 0, sizeof(*v1));
4032 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
4033 v1->lum_stripe_count =
4034 cpu_to_le32(lds->lds_dir_def_stripe_count);
4035 v1->lum_stripe_offset =
4036 cpu_to_le32(lds->lds_dir_def_stripe_offset);
4038 cpu_to_le32(lds->lds_dir_def_hash_type);
4040 info->lti_buf.lb_buf = v1;
4041 info->lti_buf.lb_len = sizeof(*v1);
4043 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4044 XATTR_NAME_DEFAULT_LMV,
4047 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
4049 XATTR_NAME_DEFAULT_LMV, 0,
4055 /* Transfer default LOV striping from the parent */
4056 if (lds != NULL && lds->lds_def_striping_set &&
4057 lds->lds_def_comp_cnt != 0) {
4058 struct lov_mds_md *lmm;
4059 int lmm_size = lod_comp_md_size(lo, true);
4061 if (info->lti_ea_store_size < lmm_size) {
4062 rc = lod_ea_store_resize(info, lmm_size);
4066 lmm = info->lti_ea_store;
4068 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, true);
4072 info->lti_buf.lb_buf = lmm;
4073 info->lti_buf.lb_len = lmm_size;
4076 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
4077 XATTR_NAME_LOV, 0, th);
4079 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4080 XATTR_NAME_LOV, 0, th);
4088 static int lod_declare_dir_striping_create(const struct lu_env *env,
4089 struct dt_object *dt,
4090 struct lu_attr *attr,
4092 struct dt_object_format *dof,
4095 return lod_dir_striping_create_internal(env, dt, attr, lmu, dof, th,
4099 static int lod_dir_striping_create(const struct lu_env *env,
4100 struct dt_object *dt,
4101 struct lu_attr *attr,
4102 struct dt_object_format *dof,
4105 return lod_dir_striping_create_internal(env, dt, attr, NULL, dof, th,
4110 * Make LOV EA for striped object.
4112 * Generate striping information and store it in the LOV EA of the given
4113 * object. The caller must ensure nobody else is calling the function
4114 * against the object concurrently. The transaction must be started.
4115 * FLDB service must be running as well; it's used to map FID to the target,
4116 * which is stored in LOV EA.
4118 * \param[in] env execution environment for this thread
4119 * \param[in] lo LOD object
4120 * \param[in] th transaction handle
4122 * \retval 0 if LOV EA is stored successfully
4123 * \retval negative error number on failure
4125 static int lod_generate_and_set_lovea(const struct lu_env *env,
4126 struct lod_object *lo,
4129 struct lod_thread_info *info = lod_env_info(env);
4130 struct dt_object *next = dt_object_child(&lo->ldo_obj);
4131 struct lov_mds_md_v1 *lmm;
4137 if (lo->ldo_comp_cnt == 0 && !lo->ldo_is_foreign) {
4138 lod_striping_free(env, lo);
4139 rc = lod_sub_xattr_del(env, next, XATTR_NAME_LOV, th);
4143 lmm_size = lod_comp_md_size(lo, false);
4144 if (info->lti_ea_store_size < lmm_size) {
4145 rc = lod_ea_store_resize(info, lmm_size);
4149 lmm = info->lti_ea_store;
4151 rc = lod_generate_lovea(env, lo, lmm, &lmm_size, false);
4155 info->lti_buf.lb_buf = lmm;
4156 info->lti_buf.lb_len = lmm_size;
4157 rc = lod_sub_xattr_set(env, next, &info->lti_buf,
4158 XATTR_NAME_LOV, 0, th);
4162 static __u32 lod_gen_component_id(struct lod_object *lo,
4163 int mirror_id, int comp_idx);
4166 * Repeat an existing component
4168 * Creates a new layout by replicating an existing component. Uses striping
4169 * policy from previous component as a template for the striping for the new
4172 * New component starts with zero length, will be extended (or removed) before
4173 * returning layout to client.
4175 * NB: Reallocates layout components array (lo->ldo_comp_entries), invalidating
4176 * any pre-existing pointers to components. Handle with care.
4178 * \param[in] env execution environment for this thread
4179 * \param[in,out] lo object to update the layout of
4180 * \param[in] index index of component to copy
4182 * \retval 0 on success
4183 * \retval negative errno on error
4185 static int lod_layout_repeat_comp(const struct lu_env *env,
4186 struct lod_object *lo, int index)
4188 struct lod_layout_component *lod_comp;
4189 struct lod_layout_component *new_comp = NULL;
4190 struct lod_layout_component *comp_array;
4191 int rc = 0, i, new_cnt = lo->ldo_comp_cnt + 1;
4196 lod_comp = &lo->ldo_comp_entries[index];
4197 LASSERT(lod_comp_inited(lod_comp) && lod_comp->llc_id != LCME_ID_INVAL);
4199 CDEBUG(D_LAYOUT, "repeating component %d\n", index);
4201 OBD_ALLOC(comp_array, sizeof(*comp_array) * new_cnt);
4202 if (comp_array == NULL)
4203 GOTO(out, rc = -ENOMEM);
4205 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4206 memcpy(&comp_array[i + offset], &lo->ldo_comp_entries[i],
4207 sizeof(*comp_array));
4209 /* Duplicate this component in to the next slot */
4211 new_comp = &comp_array[i + 1];
4212 memcpy(&comp_array[i + 1], &lo->ldo_comp_entries[i],
4213 sizeof(*comp_array));
4214 /* We must now skip this new component when copying */
4219 /* Set up copied component */
4220 new_comp->llc_flags &= ~LCME_FL_INIT;
4221 new_comp->llc_stripe = NULL;
4222 new_comp->llc_stripes_allocated = 0;
4223 new_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
4224 /* for uninstantiated components, layout gen stores default stripe
4226 new_comp->llc_layout_gen = lod_comp->llc_stripe_offset;
4227 /* This makes the repeated component zero-length, placed at the end of
4228 * the preceding component */
4229 new_comp->llc_extent.e_start = new_comp->llc_extent.e_end;
4230 new_comp->llc_timestamp = lod_comp->llc_timestamp;
4231 new_comp->llc_pool = NULL;
4233 rc = lod_set_pool(&new_comp->llc_pool, lod_comp->llc_pool);
4237 if (new_comp->llc_ostlist.op_array) {
4238 __u32 *op_array = NULL;
4240 OBD_ALLOC(op_array, new_comp->llc_ostlist.op_size);
4242 GOTO(out, rc = -ENOMEM);
4243 memcpy(op_array, &new_comp->llc_ostlist.op_array,
4244 new_comp->llc_ostlist.op_size);
4245 new_comp->llc_ostlist.op_array = op_array;
4248 OBD_FREE(lo->ldo_comp_entries,
4249 sizeof(*comp_array) * lo->ldo_comp_cnt);
4250 lo->ldo_comp_entries = comp_array;
4251 lo->ldo_comp_cnt = new_cnt;
4253 /* Generate an id for the new component */
4254 mirror_id = mirror_id_of(new_comp->llc_id);
4255 new_comp->llc_id = LCME_ID_INVAL;
4256 new_comp->llc_id = lod_gen_component_id(lo, mirror_id, index + 1);
4257 if (new_comp->llc_id == LCME_ID_INVAL)
4258 GOTO(out, rc = -ERANGE);
4263 OBD_FREE(comp_array, sizeof(*comp_array) * new_cnt);
4268 static int lod_layout_data_init(struct lod_thread_info *info, __u32 comp_cnt)
4272 /* clear memory region that will be used for layout change */
4273 memset(&info->lti_layout_attr, 0, sizeof(struct lu_attr));
4274 info->lti_count = 0;
4276 if (info->lti_comp_size >= comp_cnt)
4279 if (info->lti_comp_size > 0) {
4280 OBD_FREE(info->lti_comp_idx,
4281 info->lti_comp_size * sizeof(__u32));
4282 info->lti_comp_size = 0;
4285 OBD_ALLOC(info->lti_comp_idx, comp_cnt * sizeof(__u32));
4286 if (!info->lti_comp_idx)
4289 info->lti_comp_size = comp_cnt;
4294 * Prepare new layout minus deleted components
4296 * Removes components marked for deletion (LCME_ID_INVAL) by copying to a new
4297 * layout and skipping those components. Removes stripe objects if any exist.
4300 * Reallocates layout components array (lo->ldo_comp_entries), invalidating
4301 * any pre-existing pointers to components.
4303 * Caller is responsible for updating mirror end (ldo_mirror[].lme_end).
4305 * \param[in] env execution environment for this thread
4306 * \param[in,out] lo object to update the layout of
4307 * \param[in] th transaction handle for this operation
4309 * \retval # of components deleted
4310 * \retval negative errno on error
4312 static int lod_layout_del_prep_layout(const struct lu_env *env,
4313 struct lod_object *lo,
4316 struct lod_layout_component *lod_comp;
4317 struct lod_thread_info *info = lod_env_info(env);
4318 int rc = 0, i, j, deleted = 0;
4322 LASSERT(lo->ldo_is_composite);
4323 LASSERT(lo->ldo_comp_cnt > 0 && lo->ldo_comp_entries != NULL);
4325 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
4329 for (i = 0; i < lo->ldo_comp_cnt; i++) {
4330 lod_comp = &lo->ldo_comp_entries[i];
4332 if (lod_comp->llc_id != LCME_ID_INVAL) {
4333 /* Build array of things to keep */
4334 info->lti_comp_idx[info->lti_count++] = i;
4338 lod_obj_set_pool(lo, i, NULL);
4339 if (lod_comp->llc_ostlist.op_array) {
4340 OBD_FREE(lod_comp->llc_ostlist.op_array,
4341 lod_comp->llc_ostlist.op_size);
4342 lod_comp->llc_ostlist.op_array = NULL;
4343 lod_comp->llc_ostlist.op_size = 0;
4347 CDEBUG(D_LAYOUT, "deleting comp %d, left %d\n", i,
4348 lo->ldo_comp_cnt - deleted);
4350 /* No striping info for this component */
4351 if (lod_comp->llc_stripe == NULL)
4354 LASSERT(lod_comp->llc_stripe_count > 0);
4355 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
4356 struct dt_object *obj = lod_comp->llc_stripe[j];
4361 /* components which are not init have no sub objects
4363 if (lod_comp_inited(lod_comp)) {
4364 rc = lod_sub_destroy(env, obj, th);
4369 lu_object_put(env, &obj->do_lu);
4370 lod_comp->llc_stripe[j] = NULL;
4372 OBD_FREE(lod_comp->llc_stripe, sizeof(struct dt_object *) *
4373 lod_comp->llc_stripes_allocated);
4374 lod_comp->llc_stripe = NULL;
4375 OBD_FREE(lod_comp->llc_ost_indices,
4376 sizeof(__u32) * lod_comp->llc_stripes_allocated);
4377 lod_comp->llc_ost_indices = NULL;
4378 lod_comp->llc_stripes_allocated = 0;
4381 /* info->lti_count has the amount of left components */
4382 LASSERTF(info->lti_count >= 0 && info->lti_count < lo->ldo_comp_cnt,
4383 "left = %d, lo->ldo_comp_cnt %d\n", (int)info->lti_count,
4384 (int)lo->ldo_comp_cnt);
4386 if (info->lti_count > 0) {
4387 struct lod_layout_component *comp_array;
4389 OBD_ALLOC(comp_array, sizeof(*comp_array) * info->lti_count);
4390 if (comp_array == NULL)
4391 GOTO(out, rc = -ENOMEM);
4393 for (i = 0; i < info->lti_count; i++) {
4394 memcpy(&comp_array[i],
4395 &lo->ldo_comp_entries[info->lti_comp_idx[i]],
4396 sizeof(*comp_array));
4399 OBD_FREE(lo->ldo_comp_entries,
4400 sizeof(*comp_array) * lo->ldo_comp_cnt);
4401 lo->ldo_comp_entries = comp_array;
4402 lo->ldo_comp_cnt = info->lti_count;
4404 lod_free_comp_entries(lo);
4409 return rc ? rc : deleted;
4413 * Delete layout component(s)
4415 * This function sets up the layout data in the env and does the setattrs
4416 * required to write out the new layout. The layout itself is modified in
4417 * lod_layout_del_prep_layout.
4419 * \param[in] env execution environment for this thread
4420 * \param[in] dt object
4421 * \param[in] th transaction handle
4423 * \retval 0 on success
4424 * \retval negative error number on failure
4426 static int lod_layout_del(const struct lu_env *env, struct dt_object *dt,
4429 struct lod_object *lo = lod_dt_obj(dt);
4430 struct dt_object *next = dt_object_child(dt);
4431 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
4434 LASSERT(lo->ldo_mirror_count == 1);
4436 rc = lod_layout_del_prep_layout(env, lo, th);
4440 /* Only do this if we didn't delete all components */
4441 if (lo->ldo_comp_cnt > 0) {
4442 lo->ldo_mirrors[0].lme_end = lo->ldo_comp_cnt - 1;
4443 lod_obj_inc_layout_gen(lo);
4446 LASSERT(dt_object_exists(dt));
4447 rc = dt_attr_get(env, next, attr);
4451 if (attr->la_size > 0) {
4453 attr->la_valid = LA_SIZE;
4454 rc = lod_sub_attr_set(env, next, attr, th);
4459 rc = lod_generate_and_set_lovea(env, lo, th);
4463 lod_striping_free(env, lo);
4468 static int lod_get_default_lov_striping(const struct lu_env *env,
4469 struct lod_object *lo,
4470 struct lod_default_striping *lds,
4471 struct dt_allocation_hint *ah);
4473 * Implementation of dt_object_operations::do_xattr_set.
4475 * Sets specified extended attribute on the object. Three types of EAs are
4477 * LOV EA - stores striping for a regular file or default striping (when set
4479 * LMV EA - stores a marker for the striped directories
4480 * DMV EA - stores default directory striping
4482 * When striping is applied to a non-striped existing object (this is called
4483 * late striping), then LOD notices the caller wants to turn the object into a
4484 * striped one. The stripe objects are created and appropriate EA is set:
4485 * LOV EA storing all the stripes directly or LMV EA storing just a small header
4486 * with striping configuration.
4488 * \see dt_object_operations::do_xattr_set() in the API description for details.
4490 static int lod_xattr_set(const struct lu_env *env,
4491 struct dt_object *dt, const struct lu_buf *buf,
4492 const char *name, int fl, struct thandle *th)
4494 struct dt_object *next = dt_object_child(dt);
4498 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4499 strcmp(name, XATTR_NAME_LMV) == 0) {
4500 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
4502 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4503 strncmp(name, XATTR_NAME_LMV, strlen(XATTR_NAME_LMV)) == 0 &&
4504 strlen(name) > strlen(XATTR_NAME_LMV) + 1) {
4505 const char *op = name + strlen(XATTR_NAME_LMV) + 1;
4509 * XATTR_NAME_LMV".add" is never called, but only declared,
4510 * because lod_xattr_set_lmv() will do the addition.
4512 if (strcmp(op, "del") == 0)
4513 rc = lod_dir_layout_delete(env, dt, buf, th);
4514 else if (strcmp(op, "set") == 0)
4515 rc = lod_sub_xattr_set(env, next, buf, XATTR_NAME_LMV,
4519 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4520 strcmp(name, XATTR_NAME_LOV) == 0) {
4521 struct lod_default_striping *lds = lod_lds_buf_get(env);
4522 struct lov_user_md_v1 *v1 = buf->lb_buf;
4523 char pool[LOV_MAXPOOLNAME + 1];
4526 /* get existing striping config */
4527 rc = lod_get_default_lov_striping(env, lod_dt_obj(dt), lds,
4532 memset(pool, 0, sizeof(pool));
4533 if (lds->lds_def_striping_set == 1)
4534 lod_layout_get_pool(lds->lds_def_comp_entries,
4535 lds->lds_def_comp_cnt, pool,
4538 is_del = LOVEA_DELETE_VALUES(v1->lmm_stripe_size,
4539 v1->lmm_stripe_count,
4540 v1->lmm_stripe_offset,
4543 /* Retain the pool name if it is not given */
4544 if (v1->lmm_magic == LOV_USER_MAGIC_V1 && pool[0] != '\0' &&
4546 struct lod_thread_info *info = lod_env_info(env);
4547 struct lov_user_md_v3 *v3 = info->lti_ea_store;
4549 memset(v3, 0, sizeof(*v3));
4550 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
4551 v3->lmm_pattern = cpu_to_le32(v1->lmm_pattern);
4552 v3->lmm_stripe_count =
4553 cpu_to_le32(v1->lmm_stripe_count);
4554 v3->lmm_stripe_offset =
4555 cpu_to_le32(v1->lmm_stripe_offset);
4556 v3->lmm_stripe_size = cpu_to_le32(v1->lmm_stripe_size);
4558 strlcpy(v3->lmm_pool_name, pool,
4559 sizeof(v3->lmm_pool_name));
4561 info->lti_buf.lb_buf = v3;
4562 info->lti_buf.lb_len = sizeof(*v3);
4563 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
4566 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name,
4570 if (lds->lds_def_striping_set == 1 &&
4571 lds->lds_def_comp_entries != NULL)
4572 lod_free_def_comp_entries(lds);
4575 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
4576 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
4578 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
4581 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
4582 (!strcmp(name, XATTR_NAME_LOV) ||
4583 !strncmp(name, XATTR_LUSTRE_LOV,
4584 strlen(XATTR_LUSTRE_LOV)))) {
4585 /* in case of lov EA swap, just set it
4586 * if not, it is a replay so check striping match what we
4587 * already have during req replay, declare_xattr_set()
4588 * defines striping, then create() does the work */
4589 if (fl & LU_XATTR_REPLACE) {
4590 /* free stripes, then update disk */
4591 lod_striping_free(env, lod_dt_obj(dt));
4593 rc = lod_sub_xattr_set(env, next, buf, name, fl, th);
4594 } else if (dt_object_remote(dt)) {
4595 /* This only happens during migration, see
4596 * mdd_migrate_create(), in which Master MDT will
4597 * create a remote target object, and only set
4598 * (migrating) stripe EA on the remote object,
4599 * and does not need creating each stripes. */
4600 rc = lod_sub_xattr_set(env, next, buf, name,
4602 } else if (strcmp(name, XATTR_LUSTRE_LOV".del") == 0) {
4603 /* delete component(s) */
4604 LASSERT(lod_dt_obj(dt)->ldo_comp_cached);
4605 rc = lod_layout_del(env, dt, th);
4608 * When 'name' is XATTR_LUSTRE_LOV or XATTR_NAME_LOV,
4609 * it's going to create create file with specified
4610 * component(s), the striping must have not being
4611 * cached in this case;
4613 * Otherwise, it's going to add/change component(s) to
4614 * an existing file, the striping must have been cached
4617 LASSERT(equi(!strcmp(name, XATTR_LUSTRE_LOV) ||
4618 !strcmp(name, XATTR_NAME_LOV),
4619 !lod_dt_obj(dt)->ldo_comp_cached));
4621 rc = lod_striped_create(env, dt, NULL, NULL, th);
4624 } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4625 rc = lod_replace_parent_fid(env, dt, buf, th, false);
4630 /* then all other xattr */
4631 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th);
4637 * Implementation of dt_object_operations::do_declare_xattr_del.
4639 * \see dt_object_operations::do_declare_xattr_del() in the API description
4642 static int lod_declare_xattr_del(const struct lu_env *env,
4643 struct dt_object *dt, const char *name,
4646 struct lod_object *lo = lod_dt_obj(dt);
4647 struct dt_object *next = dt_object_child(dt);
4652 rc = lod_sub_declare_xattr_del(env, next, name, th);
4656 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
4659 /* NB: don't delete stripe LMV, because when we do this, normally we
4660 * will remove stripes, besides, if directory LMV is corrupt, this will
4661 * prevent deleting its LMV and fixing it (via LFSCK).
4663 if (!strcmp(name, XATTR_NAME_LMV))
4666 rc = lod_striping_load(env, lo);
4670 if (lo->ldo_dir_stripe_count == 0)
4673 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4674 struct dt_object *dto = lo->ldo_stripe[i];
4679 rc = lod_sub_declare_xattr_del(env, dto, name, th);
4688 * Implementation of dt_object_operations::do_xattr_del.
4690 * If EA storing a regular striping is being deleted, then release
4691 * all the references to the stripe objects in core.
4693 * \see dt_object_operations::do_xattr_del() in the API description for details.
4695 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
4696 const char *name, struct thandle *th)
4698 struct dt_object *next = dt_object_child(dt);
4699 struct lod_object *lo = lod_dt_obj(dt);
4704 if (!strcmp(name, XATTR_NAME_LOV) || !strcmp(name, XATTR_NAME_LMV))
4705 lod_striping_free(env, lod_dt_obj(dt));
4707 rc = lod_sub_xattr_del(env, next, name, th);
4708 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
4711 if (!strcmp(name, XATTR_NAME_LMV))
4714 if (lo->ldo_dir_stripe_count == 0)
4717 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
4718 struct dt_object *dto = lo->ldo_stripe[i];
4723 rc = lod_sub_xattr_del(env, dto, name, th);
4732 * Implementation of dt_object_operations::do_xattr_list.
4734 * \see dt_object_operations::do_xattr_list() in the API description
4737 static int lod_xattr_list(const struct lu_env *env,
4738 struct dt_object *dt, const struct lu_buf *buf)
4740 return dt_xattr_list(env, dt_object_child(dt), buf);
4743 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
4745 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
4749 * Copy OST list from layout provided by user.
4751 * \param[in] lod_comp layout_component to be filled
4752 * \param[in] v3 LOV EA V3 user data
4754 * \retval 0 on success
4755 * \retval negative if failed
4757 int lod_comp_copy_ost_lists(struct lod_layout_component *lod_comp,
4758 struct lov_user_md_v3 *v3)
4764 if (v3->lmm_stripe_offset == LOV_OFFSET_DEFAULT)
4765 v3->lmm_stripe_offset = v3->lmm_objects[0].l_ost_idx;
4767 if (lod_comp->llc_ostlist.op_array) {
4768 if (lod_comp->llc_ostlist.op_size >=
4769 v3->lmm_stripe_count * sizeof(__u32)) {
4770 lod_comp->llc_ostlist.op_count =
4771 v3->lmm_stripe_count;
4774 OBD_FREE(lod_comp->llc_ostlist.op_array,
4775 lod_comp->llc_ostlist.op_size);
4778 /* copy ost list from lmm */
4779 lod_comp->llc_ostlist.op_count = v3->lmm_stripe_count;
4780 lod_comp->llc_ostlist.op_size = v3->lmm_stripe_count * sizeof(__u32);
4781 OBD_ALLOC(lod_comp->llc_ostlist.op_array,
4782 lod_comp->llc_ostlist.op_size);
4783 if (!lod_comp->llc_ostlist.op_array)
4786 for (j = 0; j < v3->lmm_stripe_count; j++) {
4787 lod_comp->llc_ostlist.op_array[j] =
4788 v3->lmm_objects[j].l_ost_idx;
4796 * Get default striping.
4798 * \param[in] env execution environment
4799 * \param[in] lo object
4800 * \param[out] lds default striping
4802 * \retval 0 on success
4803 * \retval negative if failed
4805 static int lod_get_default_lov_striping(const struct lu_env *env,
4806 struct lod_object *lo,
4807 struct lod_default_striping *lds,
4808 struct dt_allocation_hint *ah)
4810 struct lod_thread_info *info = lod_env_info(env);
4811 struct lov_user_md_v1 *v1 = NULL;
4812 struct lov_user_md_v3 *v3 = NULL;
4813 struct lov_comp_md_v1 *comp_v1 = NULL;
4821 rc = lod_get_lov_ea(env, lo);
4825 if (rc < (typeof(rc))sizeof(struct lov_user_md))
4828 v1 = info->lti_ea_store;
4829 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
4830 lustre_swab_lov_user_md_v1(v1);
4831 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
4832 v3 = (struct lov_user_md_v3 *)v1;
4833 lustre_swab_lov_user_md_v3(v3);
4834 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_SPECIFIC)) {
4835 v3 = (struct lov_user_md_v3 *)v1;
4836 lustre_swab_lov_user_md_v3(v3);
4837 lustre_swab_lov_user_md_objects(v3->lmm_objects,
4838 v3->lmm_stripe_count);
4839 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_COMP_V1) ||
4840 v1->lmm_magic == __swab32(LOV_USER_MAGIC_SEL)) {
4841 comp_v1 = (struct lov_comp_md_v1 *)v1;
4842 lustre_swab_lov_comp_md_v1(comp_v1);
4845 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1 &&
4846 v1->lmm_magic != LOV_MAGIC_COMP_V1 &&
4847 v1->lmm_magic != LOV_MAGIC_SEL &&
4848 v1->lmm_magic != LOV_USER_MAGIC_SPECIFIC)
4851 if ((v1->lmm_magic == LOV_MAGIC_COMP_V1 ||
4852 v1->lmm_magic == LOV_MAGIC_SEL) &&
4853 !(ah && ah->dah_append_stripes)) {
4854 comp_v1 = (struct lov_comp_md_v1 *)v1;
4855 comp_cnt = comp_v1->lcm_entry_count;
4858 mirror_cnt = comp_v1->lcm_mirror_count + 1;
4866 /* realloc default comp entries if necessary */
4867 rc = lod_def_striping_comp_resize(lds, comp_cnt);
4871 lds->lds_def_comp_cnt = comp_cnt;
4872 lds->lds_def_striping_is_composite = composite;
4873 lds->lds_def_mirror_cnt = mirror_cnt;
4875 for (i = 0; i < comp_cnt; i++) {
4876 struct lod_layout_component *lod_comp;
4879 lod_comp = &lds->lds_def_comp_entries[i];
4881 * reset lod_comp values, llc_stripes is always NULL in
4882 * the default striping template, llc_pool will be reset
4885 memset(lod_comp, 0, offsetof(typeof(*lod_comp), llc_pool));
4888 v1 = (struct lov_user_md *)((char *)comp_v1 +
4889 comp_v1->lcm_entries[i].lcme_offset);
4890 lod_comp->llc_extent =
4891 comp_v1->lcm_entries[i].lcme_extent;
4892 /* We only inherit certain flags from the layout */
4893 lod_comp->llc_flags =
4894 comp_v1->lcm_entries[i].lcme_flags &
4895 LCME_TEMPLATE_FLAGS;
4898 if (!lov_pattern_supported(v1->lmm_pattern) &&
4899 !(v1->lmm_pattern & LOV_PATTERN_F_RELEASED)) {
4900 lod_free_def_comp_entries(lds);
4904 CDEBUG(D_LAYOUT, DFID" stripe_count=%d stripe_size=%d stripe_offset=%d append_stripes=%d\n",
4905 PFID(lu_object_fid(&lo->ldo_obj.do_lu)),
4906 (int)v1->lmm_stripe_count, (int)v1->lmm_stripe_size,
4907 (int)v1->lmm_stripe_offset,
4908 ah ? ah->dah_append_stripes : 0);
4910 if (ah && ah->dah_append_stripes)
4911 lod_comp->llc_stripe_count = ah->dah_append_stripes;
4913 lod_comp->llc_stripe_count = v1->lmm_stripe_count;
4914 lod_comp->llc_stripe_size = v1->lmm_stripe_size;
4915 lod_comp->llc_stripe_offset = v1->lmm_stripe_offset;
4916 lod_comp->llc_pattern = v1->lmm_pattern;
4919 if (ah && ah->dah_append_pool && ah->dah_append_pool[0]) {
4920 pool = ah->dah_append_pool;
4921 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
4922 /* XXX: sanity check here */
4923 v3 = (struct lov_user_md_v3 *) v1;
4924 if (v3->lmm_pool_name[0] != '\0')
4925 pool = v3->lmm_pool_name;
4927 lod_set_def_pool(lds, i, pool);
4928 if (v1->lmm_magic == LOV_USER_MAGIC_SPECIFIC) {
4929 v3 = (struct lov_user_md_v3 *)v1;
4930 rc = lod_comp_copy_ost_lists(lod_comp, v3);
4933 } else if (lod_comp->llc_ostlist.op_array &&
4934 lod_comp->llc_ostlist.op_count) {
4935 for (j = 0; j < lod_comp->llc_ostlist.op_count; j++)
4936 lod_comp->llc_ostlist.op_array[j] = -1;
4937 lod_comp->llc_ostlist.op_count = 0;
4941 lds->lds_def_striping_set = 1;
4946 * Get default directory striping.
4948 * \param[in] env execution environment
4949 * \param[in] lo object
4950 * \param[out] lds default striping
4952 * \retval 0 on success
4953 * \retval negative if failed
4955 static int lod_get_default_lmv_striping(const struct lu_env *env,
4956 struct lod_object *lo,
4957 struct lod_default_striping *lds)
4959 struct lmv_user_md *lmu;
4962 lds->lds_dir_def_striping_set = 0;
4964 rc = lod_get_default_lmv_ea(env, lo);
4968 if (rc >= (int)sizeof(*lmu)) {
4969 struct lod_thread_info *info = lod_env_info(env);
4971 lmu = info->lti_ea_store;
4973 lds->lds_dir_def_stripe_count =
4974 le32_to_cpu(lmu->lum_stripe_count);
4975 lds->lds_dir_def_stripe_offset =
4976 le32_to_cpu(lmu->lum_stripe_offset);
4977 lds->lds_dir_def_hash_type =
4978 le32_to_cpu(lmu->lum_hash_type);
4979 lds->lds_dir_def_striping_set = 1;
4986 * Get default striping in the object.
4988 * Get object default striping and default directory striping.
4990 * \param[in] env execution environment
4991 * \param[in] lo object
4992 * \param[out] lds default striping
4994 * \retval 0 on success
4995 * \retval negative if failed
4997 static int lod_get_default_striping(const struct lu_env *env,
4998 struct lod_object *lo,
4999 struct lod_default_striping *lds)
5003 rc = lod_get_default_lov_striping(env, lo, lds, NULL);
5004 rc1 = lod_get_default_lmv_striping(env, lo, lds);
5005 if (rc == 0 && rc1 < 0)
5012 * Apply default striping on object.
5014 * If object striping pattern is not set, set to the one in default striping.
5015 * The default striping is from parent or fs.
5017 * \param[in] lo new object
5018 * \param[in] lds default striping
5019 * \param[in] mode new object's mode
5021 static void lod_striping_from_default(struct lod_object *lo,
5022 const struct lod_default_striping *lds,
5025 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5026 struct lov_desc *desc = &d->lod_desc;
5029 if (lds->lds_def_striping_set && S_ISREG(mode)) {
5030 rc = lod_alloc_comp_entries(lo, lds->lds_def_mirror_cnt,
5031 lds->lds_def_comp_cnt);
5035 lo->ldo_is_composite = lds->lds_def_striping_is_composite;
5036 if (lds->lds_def_mirror_cnt > 1)
5037 lo->ldo_flr_state = LCM_FL_RDONLY;
5039 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5040 struct lod_layout_component *obj_comp =
5041 &lo->ldo_comp_entries[i];
5042 struct lod_layout_component *def_comp =
5043 &lds->lds_def_comp_entries[i];
5045 CDEBUG(D_LAYOUT, "Inherit from default: flags=%#x "
5046 "size=%hu nr=%u offset=%u pattern=%#x pool=%s\n",
5047 def_comp->llc_flags,
5048 def_comp->llc_stripe_size,
5049 def_comp->llc_stripe_count,
5050 def_comp->llc_stripe_offset,
5051 def_comp->llc_pattern,
5052 def_comp->llc_pool ?: "");
5054 *obj_comp = *def_comp;
5055 if (def_comp->llc_pool != NULL) {
5056 /* pointer was copied from def_comp */
5057 obj_comp->llc_pool = NULL;
5058 lod_obj_set_pool(lo, i, def_comp->llc_pool);
5062 if (def_comp->llc_ostlist.op_array &&
5063 def_comp->llc_ostlist.op_count) {
5064 OBD_ALLOC(obj_comp->llc_ostlist.op_array,
5065 obj_comp->llc_ostlist.op_size);
5066 if (!obj_comp->llc_ostlist.op_array)
5068 memcpy(obj_comp->llc_ostlist.op_array,
5069 def_comp->llc_ostlist.op_array,
5070 obj_comp->llc_ostlist.op_size);
5071 } else if (def_comp->llc_ostlist.op_array) {
5072 obj_comp->llc_ostlist.op_array = NULL;
5076 * Don't initialize these fields for plain layout
5077 * (v1/v3) here, they are inherited in the order of
5078 * 'parent' -> 'fs default (root)' -> 'global default
5079 * values for stripe_count & stripe_size'.
5081 * see lod_ah_init().
5083 if (!lo->ldo_is_composite)
5086 lod_adjust_stripe_info(obj_comp, desc, 0);
5088 } else if (lds->lds_dir_def_striping_set && S_ISDIR(mode)) {
5089 if (lo->ldo_dir_stripe_count == 0)
5090 lo->ldo_dir_stripe_count =
5091 lds->lds_dir_def_stripe_count;
5092 if (lo->ldo_dir_stripe_offset == -1)
5093 lo->ldo_dir_stripe_offset =
5094 lds->lds_dir_def_stripe_offset;
5095 if (lo->ldo_dir_hash_type == 0)
5096 lo->ldo_dir_hash_type = lds->lds_dir_def_hash_type &
5097 ~LMV_HASH_FLAG_SPACE;
5099 CDEBUG(D_LAYOUT, "striping from default dir: count:%hu, "
5100 "offset:%u, hash_type:%u\n",
5101 lo->ldo_dir_stripe_count, lo->ldo_dir_stripe_offset,
5102 lo->ldo_dir_hash_type);
5106 static inline bool lod_need_inherit_more(struct lod_object *lo, bool from_root,
5109 struct lod_layout_component *lod_comp;
5111 if (lo->ldo_comp_cnt == 0)
5114 if (lo->ldo_is_composite)
5117 lod_comp = &lo->ldo_comp_entries[0];
5119 if (lod_comp->llc_stripe_count <= 0 ||
5120 lod_comp->llc_stripe_size <= 0)
5123 if (from_root && (lod_comp->llc_pool == NULL ||
5124 lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT))
5127 if (append_pool && append_pool[0])
5134 * Implementation of dt_object_operations::do_ah_init.
5136 * This method is used to make a decision on the striping configuration for the
5137 * object being created. It can be taken from the \a parent object if it exists,
5138 * or filesystem's default. The resulting configuration (number of stripes,
5139 * stripe size/offset, pool name, etc) is stored in the object itself and will
5140 * be used by the methods like ->doo_declare_create().
5142 * \see dt_object_operations::do_ah_init() in the API description for details.
5144 static void lod_ah_init(const struct lu_env *env,
5145 struct dt_allocation_hint *ah,
5146 struct dt_object *parent,
5147 struct dt_object *child,
5150 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
5151 struct lod_thread_info *info = lod_env_info(env);
5152 struct lod_default_striping *lds = lod_lds_buf_get(env);
5153 struct dt_object *nextp = NULL;
5154 struct dt_object *nextc;
5155 struct lod_object *lp = NULL;
5156 struct lod_object *lc;
5157 struct lov_desc *desc;
5158 struct lod_layout_component *lod_comp;
5164 if (ah->dah_append_stripes == -1)
5165 ah->dah_append_stripes = d->lod_desc.ld_tgt_count;
5167 if (likely(parent)) {
5168 nextp = dt_object_child(parent);
5169 lp = lod_dt_obj(parent);
5172 nextc = dt_object_child(child);
5173 lc = lod_dt_obj(child);
5175 LASSERT(!lod_obj_is_striped(child));
5176 /* default layout template may have been set on the regular file
5177 * when this is called from mdd_create_data() */
5178 if (S_ISREG(child_mode))
5179 lod_free_comp_entries(lc);
5181 if (!dt_object_exists(nextc))
5182 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
5184 if (S_ISDIR(child_mode)) {
5185 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
5187 /* other default values are 0 */
5188 lc->ldo_dir_stripe_offset = -1;
5190 /* no default striping configuration is needed for
5193 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5194 le32_to_cpu(lum1->lum_magic) == LMV_MAGIC_FOREIGN) {
5195 lc->ldo_dir_is_foreign = true;
5196 /* keep stripe_count 0 and stripe_offset -1 */
5197 CDEBUG(D_INFO, "no default striping for foreign dir\n");
5202 * If parent object is not root directory,
5203 * then get default striping from parent object.
5205 if (likely(lp != NULL) && !fid_is_root(lod_object_fid(lp)))
5206 lod_get_default_striping(env, lp, lds);
5208 /* set child default striping info, default value is NULL */
5209 if (lds->lds_def_striping_set || lds->lds_dir_def_striping_set)
5210 lc->ldo_def_striping = lds;
5212 /* It should always honour the specified stripes */
5213 /* Note: old client (< 2.7)might also do lfs mkdir, whose EA
5214 * will have old magic. In this case, we should ignore the
5215 * stripe count and try to create dir by default stripe.
5217 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0 &&
5218 (le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC ||
5219 le32_to_cpu(lum1->lum_magic) == LMV_USER_MAGIC_SPECIFIC)) {
5220 lc->ldo_dir_stripe_count =
5221 le32_to_cpu(lum1->lum_stripe_count);
5222 lc->ldo_dir_stripe_offset =
5223 le32_to_cpu(lum1->lum_stripe_offset);
5224 lc->ldo_dir_hash_type =
5225 le32_to_cpu(lum1->lum_hash_type);
5227 "set dirstripe: count %hu, offset %d, hash %u\n",
5228 lc->ldo_dir_stripe_count,
5229 (int)lc->ldo_dir_stripe_offset,
5230 lc->ldo_dir_hash_type);
5232 /* transfer defaults LMV to new directory */
5233 lod_striping_from_default(lc, lds, child_mode);
5235 /* set count 0 to create normal directory */
5236 if (lc->ldo_dir_stripe_count == 1)
5237 lc->ldo_dir_stripe_count = 0;
5240 /* shrink the stripe_count to the avaible MDT count */
5241 if (lc->ldo_dir_stripe_count > d->lod_remote_mdt_count + 1 &&
5242 !OBD_FAIL_CHECK(OBD_FAIL_LARGE_STRIPE)) {
5243 lc->ldo_dir_stripe_count = d->lod_remote_mdt_count + 1;
5244 if (lc->ldo_dir_stripe_count == 1)
5245 lc->ldo_dir_stripe_count = 0;
5248 CDEBUG(D_INFO, "final dir stripe [%hu %d %u]\n",
5249 lc->ldo_dir_stripe_count,
5250 (int)lc->ldo_dir_stripe_offset, lc->ldo_dir_hash_type);
5255 /* child object regular file*/
5257 if (!lod_object_will_be_striped(S_ISREG(child_mode),
5258 lu_object_fid(&child->do_lu)))
5261 /* If object is going to be striped over OSTs, transfer default
5262 * striping information to the child, so that we can use it
5263 * during declaration and creation.
5265 * Try from the parent first.
5267 if (likely(lp != NULL)) {
5268 rc = lod_get_default_lov_striping(env, lp, lds, ah);
5270 lod_striping_from_default(lc, lds, child_mode);
5273 /* Initialize lod_device::lod_md_root object reference */
5274 if (d->lod_md_root == NULL) {
5275 struct dt_object *root;
5276 struct lod_object *lroot;
5278 lu_root_fid(&info->lti_fid);
5279 root = dt_locate(env, &d->lod_dt_dev, &info->lti_fid);
5280 if (!IS_ERR(root)) {
5281 lroot = lod_dt_obj(root);
5283 spin_lock(&d->lod_lock);
5284 if (d->lod_md_root != NULL)
5285 dt_object_put(env, &d->lod_md_root->ldo_obj);
5286 d->lod_md_root = lroot;
5287 spin_unlock(&d->lod_lock);
5291 /* try inherit layout from the root object (fs default) when:
5292 * - parent does not have default layout; or
5293 * - parent has plain(v1/v3) default layout, and some attributes
5294 * are not specified in the default layout;
5296 if (d->lod_md_root != NULL &&
5297 lod_need_inherit_more(lc, true, ah->dah_append_pool)) {
5298 rc = lod_get_default_lov_striping(env, d->lod_md_root, lds,
5302 if (lc->ldo_comp_cnt == 0) {
5303 lod_striping_from_default(lc, lds, child_mode);
5304 } else if (!lds->lds_def_striping_is_composite) {
5305 struct lod_layout_component *def_comp;
5307 LASSERT(!lc->ldo_is_composite);
5308 lod_comp = &lc->ldo_comp_entries[0];
5309 def_comp = &lds->lds_def_comp_entries[0];
5311 if (lod_comp->llc_stripe_count <= 0)
5312 lod_comp->llc_stripe_count =
5313 def_comp->llc_stripe_count;
5314 if (lod_comp->llc_stripe_size <= 0)
5315 lod_comp->llc_stripe_size =
5316 def_comp->llc_stripe_size;
5317 if (lod_comp->llc_stripe_offset == LOV_OFFSET_DEFAULT &&
5318 (!lod_comp->llc_pool || !lod_comp->llc_pool[0]))
5319 lod_comp->llc_stripe_offset =
5320 def_comp->llc_stripe_offset;
5321 if (lod_comp->llc_pool == NULL)
5322 lod_obj_set_pool(lc, 0, def_comp->llc_pool);
5327 * fs default striping may not be explicitly set, or historically set
5328 * in config log, use them.
5330 if (lod_need_inherit_more(lc, false, ah->dah_append_pool)) {
5331 if (lc->ldo_comp_cnt == 0) {
5332 rc = lod_alloc_comp_entries(lc, 0, 1);
5334 /* fail to allocate memory, will create a
5335 * non-striped file. */
5337 lc->ldo_is_composite = 0;
5338 lod_comp = &lc->ldo_comp_entries[0];
5339 lod_comp->llc_stripe_offset = LOV_OFFSET_DEFAULT;
5341 LASSERT(!lc->ldo_is_composite);
5342 lod_comp = &lc->ldo_comp_entries[0];
5343 desc = &d->lod_desc;
5344 lod_adjust_stripe_info(lod_comp, desc, ah->dah_append_stripes);
5345 if (ah->dah_append_pool && ah->dah_append_pool[0])
5346 lod_obj_set_pool(lc, 0, ah->dah_append_pool);
5352 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
5354 * Size initialization on late striping.
5356 * Propagate the size of a truncated object to a deferred striping.
5357 * This function handles a special case when truncate was done on a
5358 * non-striped object and now while the striping is being created
5359 * we can't lose that size, so we have to propagate it to the stripes
5362 * \param[in] env execution environment
5363 * \param[in] dt object
5364 * \param[in] th transaction handle
5366 * \retval 0 on success
5367 * \retval negative if failed
5369 static int lod_declare_init_size(const struct lu_env *env,
5370 struct dt_object *dt, struct thandle *th)
5372 struct dt_object *next = dt_object_child(dt);
5373 struct lod_object *lo = lod_dt_obj(dt);
5374 struct dt_object **objects = NULL;
5375 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
5376 uint64_t size, offs;
5377 int i, rc, stripe, stripe_count = 0, stripe_size = 0;
5378 struct lu_extent size_ext;
5381 if (!lod_obj_is_striped(dt))
5384 rc = dt_attr_get(env, next, attr);
5385 LASSERT(attr->la_valid & LA_SIZE);
5389 size = attr->la_size;
5393 size_ext = (typeof(size_ext)){ .e_start = size - 1, .e_end = size };
5394 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5395 struct lod_layout_component *lod_comp;
5396 struct lu_extent *extent;
5398 lod_comp = &lo->ldo_comp_entries[i];
5400 if (lod_comp->llc_stripe == NULL)
5403 extent = &lod_comp->llc_extent;
5404 CDEBUG(D_INFO, "%lld "DEXT"\n", size, PEXT(extent));
5405 if (!lo->ldo_is_composite ||
5406 lu_extent_is_overlapped(extent, &size_ext)) {
5407 objects = lod_comp->llc_stripe;
5408 stripe_count = lod_comp->llc_stripe_count;
5409 stripe_size = lod_comp->llc_stripe_size;
5412 if (stripe_count == 0)
5415 LASSERT(objects != NULL && stripe_size != 0);
5416 /* ll_do_div64(a, b) returns a % b, and a = a / b */
5417 ll_do_div64(size, (__u64)stripe_size);
5418 stripe = ll_do_div64(size, (__u64)stripe_count);
5419 LASSERT(objects[stripe] != NULL);
5421 size = size * stripe_size;
5422 offs = attr->la_size;
5423 size += ll_do_div64(offs, stripe_size);
5425 attr->la_valid = LA_SIZE;
5426 attr->la_size = size;
5428 rc = lod_sub_declare_attr_set(env, objects[stripe],
5437 * Declare creation of striped object.
5439 * The function declares creation stripes for a regular object. The function
5440 * also declares whether the stripes will be created with non-zero size if
5441 * previously size was set non-zero on the master object. If object \a dt is
5442 * not local, then only fully defined striping can be applied in \a lovea.
5443 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
5446 * \param[in] env execution environment
5447 * \param[in] dt object
5448 * \param[in] attr attributes the stripes will be created with
5449 * \param[in] lovea a buffer containing striping description
5450 * \param[in] th transaction handle
5452 * \retval 0 on success
5453 * \retval negative if failed
5455 int lod_declare_striped_create(const struct lu_env *env, struct dt_object *dt,
5456 struct lu_attr *attr,
5457 const struct lu_buf *lovea, struct thandle *th)
5459 struct lod_thread_info *info = lod_env_info(env);
5460 struct dt_object *next = dt_object_child(dt);
5461 struct lod_object *lo = lod_dt_obj(dt);
5465 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO))
5466 GOTO(out, rc = -ENOMEM);
5468 if (!dt_object_remote(next)) {
5469 /* choose OST and generate appropriate objects */
5470 rc = lod_prepare_create(env, lo, attr, lovea, th);
5475 * declare storage for striping data
5477 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
5479 /* LOD can not choose OST objects for remote objects, i.e.
5480 * stripes must be ready before that. Right now, it can only
5481 * happen during migrate, i.e. migrate process needs to create
5482 * remote regular file (mdd_migrate_create), then the migrate
5483 * process will provide stripeEA. */
5484 LASSERT(lovea != NULL);
5485 info->lti_buf = *lovea;
5488 rc = lod_sub_declare_xattr_set(env, next, &info->lti_buf,
5489 XATTR_NAME_LOV, 0, th);
5494 * if striping is created with local object's size > 0,
5495 * we have to propagate this size to specific object
5496 * the case is possible only when local object was created previously
5498 if (dt_object_exists(next))
5499 rc = lod_declare_init_size(env, dt, th);
5502 /* failed to create striping or to set initial size, let's reset
5503 * config so that others don't get confused */
5505 lod_striping_free(env, lo);
5511 * Whether subdirectories under \a dt should be created on MDTs by space QoS
5513 * If LMV_HASH_FLAG_SPACE is set on directory default layout, its subdirectories
5514 * should be created on MDT by space QoS.
5516 * \param[in] env execution environment
5517 * \param[in] dev lu device
5518 * \param[in] dt object
5520 * \retval 1 if directory should create subdir by space usage
5522 * \retval -ev if failed
5524 static inline int dt_object_qos_mkdir(const struct lu_env *env,
5525 struct lu_device *dev,
5526 struct dt_object *dt)
5528 struct lod_thread_info *info = lod_env_info(env);
5529 struct lu_object *obj;
5530 struct lod_object *lo;
5531 struct lmv_user_md *lmu;
5534 obj = lu_object_find_slice(env, dev, lu_object_fid(&dt->do_lu), NULL);
5536 return PTR_ERR(obj);
5538 lo = lu2lod_obj(obj);
5540 rc = lod_get_default_lmv_ea(env, lo);
5541 dt_object_put(env, dt);
5545 if (rc < (int)sizeof(*lmu))
5548 lmu = info->lti_ea_store;
5549 return !!(le32_to_cpu(lmu->lum_hash_type) & LMV_HASH_FLAG_SPACE);
5553 * Implementation of dt_object_operations::do_declare_create.
5555 * The method declares creation of a new object. If the object will be striped,
5556 * then helper functions are called to find FIDs for the stripes, declare
5557 * creation of the stripes and declare initialization of the striping
5558 * information to be stored in the master object.
5560 * \see dt_object_operations::do_declare_create() in the API description
5563 static int lod_declare_create(const struct lu_env *env, struct dt_object *dt,
5564 struct lu_attr *attr,
5565 struct dt_allocation_hint *hint,
5566 struct dt_object_format *dof, struct thandle *th)
5568 struct dt_object *next = dt_object_child(dt);
5569 struct lod_object *lo = lod_dt_obj(dt);
5578 * first of all, we declare creation of local object
5580 rc = lod_sub_declare_create(env, next, attr, hint, dof, th);
5585 * it's lod_ah_init() that has decided the object will be striped
5587 if (dof->dof_type == DFT_REGULAR) {
5588 /* callers don't want stripes */
5589 /* XXX: all tricky interactions with ->ah_make_hint() decided
5590 * to use striping, then ->declare_create() behaving differently
5591 * should be cleaned */
5592 if (dof->u.dof_reg.striped != 0)
5593 rc = lod_declare_striped_create(env, dt, attr,
5595 } else if (dof->dof_type == DFT_DIR) {
5596 struct seq_server_site *ss;
5597 struct lu_buf buf = { NULL };
5598 struct lu_buf *lmu = NULL;
5600 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
5602 /* If the parent has default stripeEA, and client
5603 * did not find it before sending create request,
5604 * then MDT will return -EREMOTE, and client will
5605 * retrieve the default stripeEA and re-create the
5608 * Note: if dah_eadata != NULL, it means creating the
5609 * striped directory with specified stripeEA, then it
5610 * should ignore the default stripeEA */
5611 if (hint != NULL && hint->dah_eadata == NULL) {
5612 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STALE_DIR_LAYOUT))
5613 GOTO(out, rc = -EREMOTE);
5615 if (lo->ldo_dir_stripe_offset == -1) {
5617 * child and parent should be in the same MDT,
5618 * but if parent has plain layout, it's allowed.
5620 if (hint->dah_parent &&
5621 dt_object_remote(hint->dah_parent)) {
5622 rc = dt_object_qos_mkdir(env,
5623 lo->ldo_obj.do_lu.lo_dev,
5626 GOTO(out, rc ? rc : -EREMOTE);
5628 } else if (lo->ldo_dir_stripe_offset !=
5630 struct lod_device *lod;
5631 struct lod_tgt_descs *ltd;
5632 struct lod_tgt_desc *tgt = NULL;
5633 bool found_mdt = false;
5635 lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
5636 ltd = &lod->lod_mdt_descs;
5637 ltd_foreach_tgt(ltd, tgt) {
5638 if (tgt->ltd_index ==
5639 lo->ldo_dir_stripe_offset) {
5645 /* If the MDT indicated by stripe_offset can be
5646 * found, then tell client to resend the create
5647 * request to the correct MDT, otherwise return
5648 * error to client */
5650 GOTO(out, rc = -EREMOTE);
5652 GOTO(out, rc = -EINVAL);
5654 } else if (hint && hint->dah_eadata) {
5656 lmu->lb_buf = (void *)hint->dah_eadata;
5657 lmu->lb_len = hint->dah_eadata_len;
5660 rc = lod_declare_dir_striping_create(env, dt, attr, lmu, dof,
5664 /* failed to create striping or to set initial size, let's reset
5665 * config so that others don't get confused */
5667 lod_striping_free(env, lo);
5672 * Generate component ID for new created component.
5674 * \param[in] lo LOD object
5675 * \param[in] comp_idx index of ldo_comp_entries
5677 * \retval component ID on success
5678 * \retval LCME_ID_INVAL on failure
5680 static __u32 lod_gen_component_id(struct lod_object *lo,
5681 int mirror_id, int comp_idx)
5683 struct lod_layout_component *lod_comp;
5684 __u32 id, start, end;
5687 LASSERT(lo->ldo_comp_entries[comp_idx].llc_id == LCME_ID_INVAL);
5689 lod_obj_inc_layout_gen(lo);
5690 id = lo->ldo_layout_gen;
5691 if (likely(id <= SEQ_ID_MAX))
5692 RETURN(pflr_id(mirror_id, id & SEQ_ID_MASK));
5694 /* Layout generation wraps, need to check collisions. */
5695 start = id & SEQ_ID_MASK;
5698 for (id = start; id <= end; id++) {
5699 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5700 lod_comp = &lo->ldo_comp_entries[i];
5701 if (pflr_id(mirror_id, id) == lod_comp->llc_id)
5704 /* Found the ununsed ID */
5705 if (i == lo->ldo_comp_cnt)
5706 RETURN(pflr_id(mirror_id, id));
5708 if (end == LCME_ID_MAX) {
5710 end = min(lo->ldo_layout_gen & LCME_ID_MASK,
5711 (__u32)(LCME_ID_MAX - 1));
5715 RETURN(LCME_ID_INVAL);
5719 * Creation of a striped regular object.
5721 * The function is called to create the stripe objects for a regular
5722 * striped file. This can happen at the initial object creation or
5723 * when the caller asks LOD to do so using ->do_xattr_set() method
5724 * (so called late striping). Notice all the information are already
5725 * prepared in the form of the list of objects (ldo_stripe field).
5726 * This is done during declare phase.
5728 * \param[in] env execution environment
5729 * \param[in] dt object
5730 * \param[in] attr attributes the stripes will be created with
5731 * \param[in] dof format of stripes (see OSD API description)
5732 * \param[in] th transaction handle
5734 * \retval 0 on success
5735 * \retval negative if failed
5737 int lod_striped_create(const struct lu_env *env, struct dt_object *dt,
5738 struct lu_attr *attr, struct dt_object_format *dof,
5741 struct lod_layout_component *lod_comp;
5742 struct lod_object *lo = lod_dt_obj(dt);
5747 LASSERT((lo->ldo_comp_cnt != 0 && lo->ldo_comp_entries != NULL) ||
5748 lo->ldo_is_foreign);
5750 mirror_id = 0; /* non-flr file's mirror_id is 0 */
5751 if (lo->ldo_mirror_count > 1) {
5752 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5753 lod_comp = &lo->ldo_comp_entries[i];
5754 if (lod_comp->llc_id != LCME_ID_INVAL &&
5755 mirror_id_of(lod_comp->llc_id) > mirror_id)
5756 mirror_id = mirror_id_of(lod_comp->llc_id);
5760 /* create all underlying objects */
5761 for (i = 0; i < lo->ldo_comp_cnt; i++) {
5762 lod_comp = &lo->ldo_comp_entries[i];
5764 if (lod_comp->llc_id == LCME_ID_INVAL) {
5765 /* only the component of FLR layout with more than 1
5766 * mirror has mirror ID in its component ID.
5768 if (lod_comp->llc_extent.e_start == 0 &&
5769 lo->ldo_mirror_count > 1)
5772 lod_comp->llc_id = lod_gen_component_id(lo,
5774 if (lod_comp->llc_id == LCME_ID_INVAL)
5775 GOTO(out, rc = -ERANGE);
5778 if (lod_comp_inited(lod_comp))
5781 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
5782 lod_comp_set_init(lod_comp);
5784 if (lov_pattern(lod_comp->llc_pattern) == LOV_PATTERN_MDT)
5785 lod_comp_set_init(lod_comp);
5787 if (lod_comp->llc_stripe == NULL)
5790 LASSERT(lod_comp->llc_stripe_count);
5791 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
5792 struct dt_object *object = lod_comp->llc_stripe[j];
5793 LASSERT(object != NULL);
5794 rc = lod_sub_create(env, object, attr, NULL, dof, th);
5798 lod_comp_set_init(lod_comp);
5801 rc = lod_fill_mirrors(lo);
5805 rc = lod_generate_and_set_lovea(env, lo, th);
5809 lo->ldo_comp_cached = 1;
5813 lod_striping_free(env, lo);
5817 static inline bool lod_obj_is_dom(struct dt_object *dt)
5819 struct lod_object *lo = lod_dt_obj(dt);
5821 if (!dt_object_exists(dt_object_child(dt)))
5824 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
5827 if (!lo->ldo_comp_cnt)
5830 return (lov_pattern(lo->ldo_comp_entries[0].llc_pattern) ==
5835 * Implementation of dt_object_operations::do_create.
5837 * If any of preceeding methods (like ->do_declare_create(),
5838 * ->do_ah_init(), etc) chose to create a striped object,
5839 * then this method will create the master and the stripes.
5841 * \see dt_object_operations::do_create() in the API description for details.
5843 static int lod_create(const struct lu_env *env, struct dt_object *dt,
5844 struct lu_attr *attr, struct dt_allocation_hint *hint,
5845 struct dt_object_format *dof, struct thandle *th)
5850 /* create local object */
5851 rc = lod_sub_create(env, dt_object_child(dt), attr, hint, dof, th);
5855 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
5856 (lod_obj_is_striped(dt) || lod_obj_is_dom(dt)) &&
5857 dof->u.dof_reg.striped != 0) {
5858 LASSERT(lod_dt_obj(dt)->ldo_comp_cached == 0);
5859 rc = lod_striped_create(env, dt, attr, dof, th);
5866 lod_obj_stripe_destroy_cb(const struct lu_env *env, struct lod_object *lo,
5867 struct dt_object *dt, struct thandle *th,
5868 int comp_idx, int stripe_idx,
5869 struct lod_obj_stripe_cb_data *data)
5871 if (data->locd_declare)
5872 return lod_sub_declare_destroy(env, dt, th);
5873 else if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
5874 stripe_idx == cfs_fail_val)
5875 return lod_sub_destroy(env, dt, th);
5881 * Implementation of dt_object_operations::do_declare_destroy.
5883 * If the object is a striped directory, then the function declares reference
5884 * removal from the master object (this is an index) to the stripes and declares
5885 * destroy of all the stripes. In all the cases, it declares an intention to
5886 * destroy the object itself.
5888 * \see dt_object_operations::do_declare_destroy() in the API description
5891 static int lod_declare_destroy(const struct lu_env *env, struct dt_object *dt,
5894 struct dt_object *next = dt_object_child(dt);
5895 struct lod_object *lo = lod_dt_obj(dt);
5896 struct lod_thread_info *info = lod_env_info(env);
5897 struct dt_object *stripe;
5898 char *stripe_name = info->lti_key;
5904 * load striping information, notice we don't do this when object
5905 * is being initialized as we don't need this information till
5906 * few specific cases like destroy, chown
5908 rc = lod_striping_load(env, lo);
5912 /* declare destroy for all underlying objects */
5913 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5914 rc = next->do_ops->do_index_try(env, next,
5915 &dt_directory_features);
5919 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5920 stripe = lo->ldo_stripe[i];
5924 rc = lod_sub_declare_ref_del(env, next, th);
5928 snprintf(stripe_name, sizeof(info->lti_key),
5930 PFID(lu_object_fid(&stripe->do_lu)), i);
5931 rc = lod_sub_declare_delete(env, next,
5932 (const struct dt_key *)stripe_name, th);
5939 * we declare destroy for the local object
5941 rc = lod_sub_declare_destroy(env, next, th);
5945 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
5946 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
5949 if (!lod_obj_is_striped(dt))
5952 /* declare destroy all striped objects */
5953 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
5954 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
5955 stripe = lo->ldo_stripe[i];
5959 if (!dt_object_exists(stripe))
5962 rc = lod_sub_declare_ref_del(env, stripe, th);
5966 rc = lod_sub_declare_destroy(env, stripe, th);
5971 struct lod_obj_stripe_cb_data data = { { 0 } };
5973 data.locd_declare = true;
5974 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
5975 rc = lod_obj_for_each_stripe(env, lo, th, &data);
5982 * Implementation of dt_object_operations::do_destroy.
5984 * If the object is a striped directory, then the function removes references
5985 * from the master object (this is an index) to the stripes and destroys all
5986 * the stripes. In all the cases, the function destroys the object itself.
5988 * \see dt_object_operations::do_destroy() in the API description for details.
5990 static int lod_destroy(const struct lu_env *env, struct dt_object *dt,
5993 struct dt_object *next = dt_object_child(dt);
5994 struct lod_object *lo = lod_dt_obj(dt);
5995 struct lod_thread_info *info = lod_env_info(env);
5996 char *stripe_name = info->lti_key;
5997 struct dt_object *stripe;
6003 /* destroy sub-stripe of master object */
6004 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6005 rc = next->do_ops->do_index_try(env, next,
6006 &dt_directory_features);
6010 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6011 stripe = lo->ldo_stripe[i];
6015 rc = lod_sub_ref_del(env, next, th);
6019 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
6020 PFID(lu_object_fid(&stripe->do_lu)), i);
6022 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
6023 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
6024 PFID(lu_object_fid(&stripe->do_lu)));
6026 rc = lod_sub_delete(env, next,
6027 (const struct dt_key *)stripe_name, th);
6033 rc = lod_sub_destroy(env, next, th);
6037 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ) ||
6038 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
6041 if (!lod_obj_is_striped(dt))
6044 /* destroy all striped objects */
6045 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
6046 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6047 stripe = lo->ldo_stripe[i];
6051 if (!dt_object_exists(stripe))
6054 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
6055 i == cfs_fail_val) {
6056 dt_write_lock(env, stripe, DT_TGT_CHILD);
6057 rc = lod_sub_ref_del(env, stripe, th);
6058 dt_write_unlock(env, stripe);
6062 rc = lod_sub_destroy(env, stripe, th);
6068 struct lod_obj_stripe_cb_data data = { { 0 } };
6070 data.locd_declare = false;
6071 data.locd_stripe_cb = lod_obj_stripe_destroy_cb;
6072 rc = lod_obj_for_each_stripe(env, lo, th, &data);
6079 * Implementation of dt_object_operations::do_declare_ref_add.
6081 * \see dt_object_operations::do_declare_ref_add() in the API description
6084 static int lod_declare_ref_add(const struct lu_env *env,
6085 struct dt_object *dt, struct thandle *th)
6087 return lod_sub_declare_ref_add(env, dt_object_child(dt), th);
6091 * Implementation of dt_object_operations::do_ref_add.
6093 * \see dt_object_operations::do_ref_add() in the API description for details.
6095 static int lod_ref_add(const struct lu_env *env,
6096 struct dt_object *dt, struct thandle *th)
6098 return lod_sub_ref_add(env, dt_object_child(dt), th);
6102 * Implementation of dt_object_operations::do_declare_ref_del.
6104 * \see dt_object_operations::do_declare_ref_del() in the API description
6107 static int lod_declare_ref_del(const struct lu_env *env,
6108 struct dt_object *dt, struct thandle *th)
6110 return lod_sub_declare_ref_del(env, dt_object_child(dt), th);
6114 * Implementation of dt_object_operations::do_ref_del
6116 * \see dt_object_operations::do_ref_del() in the API description for details.
6118 static int lod_ref_del(const struct lu_env *env,
6119 struct dt_object *dt, struct thandle *th)
6121 return lod_sub_ref_del(env, dt_object_child(dt), th);
6125 * Implementation of dt_object_operations::do_object_sync.
6127 * \see dt_object_operations::do_object_sync() in the API description
6130 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
6131 __u64 start, __u64 end)
6133 return dt_object_sync(env, dt_object_child(dt), start, end);
6137 * Implementation of dt_object_operations::do_object_unlock.
6139 * Used to release LDLM lock(s).
6141 * \see dt_object_operations::do_object_unlock() in the API description
6144 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
6145 struct ldlm_enqueue_info *einfo,
6146 union ldlm_policy_data *policy)
6148 struct lod_object *lo = lod_dt_obj(dt);
6149 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
6150 int slave_locks_size;
6154 if (slave_locks == NULL)
6157 LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
6158 /* Note: for remote lock for single stripe dir, MDT will cancel
6159 * the lock by lockh directly */
6160 LASSERT(!dt_object_remote(dt_object_child(dt)));
6162 /* locks were unlocked in MDT layer */
6163 for (i = 0; i < slave_locks->ha_count; i++)
6164 LASSERT(!lustre_handle_is_used(&slave_locks->ha_handles[i]));
6167 * NB, ha_count may not equal to ldo_dir_stripe_count, because dir
6168 * layout may change, e.g., shrink dir layout after migration.
6170 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6171 if (lo->ldo_stripe[i])
6172 dt_invalidate(env, lo->ldo_stripe[i]);
6175 slave_locks_size = offsetof(typeof(*slave_locks),
6176 ha_handles[slave_locks->ha_count]);
6177 OBD_FREE(slave_locks, slave_locks_size);
6178 einfo->ei_cbdata = NULL;
6184 * Implementation of dt_object_operations::do_object_lock.
6186 * Used to get LDLM lock on the non-striped and striped objects.
6188 * \see dt_object_operations::do_object_lock() in the API description
6191 static int lod_object_lock(const struct lu_env *env,
6192 struct dt_object *dt,
6193 struct lustre_handle *lh,
6194 struct ldlm_enqueue_info *einfo,
6195 union ldlm_policy_data *policy)
6197 struct lod_object *lo = lod_dt_obj(dt);
6198 int slave_locks_size;
6199 struct lustre_handle_array *slave_locks = NULL;
6204 /* remote object lock */
6205 if (!einfo->ei_enq_slave) {
6206 LASSERT(dt_object_remote(dt));
6207 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
6211 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
6214 rc = lod_striping_load(env, lo);
6219 if (lo->ldo_dir_stripe_count <= 1)
6222 slave_locks_size = offsetof(typeof(*slave_locks),
6223 ha_handles[lo->ldo_dir_stripe_count]);
6224 /* Freed in lod_object_unlock */
6225 OBD_ALLOC(slave_locks, slave_locks_size);
6228 slave_locks->ha_count = lo->ldo_dir_stripe_count;
6230 /* striped directory lock */
6231 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
6232 struct lustre_handle lockh;
6233 struct ldlm_res_id *res_id;
6234 struct dt_object *stripe;
6236 stripe = lo->ldo_stripe[i];
6240 res_id = &lod_env_info(env)->lti_res_id;
6241 fid_build_reg_res_name(lu_object_fid(&stripe->do_lu), res_id);
6242 einfo->ei_res_id = res_id;
6244 if (dt_object_remote(stripe)) {
6245 set_bit(i, (void *)slave_locks->ha_map);
6246 rc = dt_object_lock(env, stripe, &lockh, einfo, policy);
6248 struct ldlm_namespace *ns = einfo->ei_namespace;
6249 ldlm_blocking_callback blocking = einfo->ei_cb_local_bl;
6250 ldlm_completion_callback completion = einfo->ei_cb_cp;
6251 __u64 dlmflags = LDLM_FL_ATOMIC_CB;
6253 if (einfo->ei_mode == LCK_PW ||
6254 einfo->ei_mode == LCK_EX)
6255 dlmflags |= LDLM_FL_COS_INCOMPAT;
6257 LASSERT(ns != NULL);
6258 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS,
6259 policy, einfo->ei_mode,
6260 &dlmflags, blocking,
6262 NULL, 0, LVB_T_NONE,
6267 ldlm_lock_decref_and_cancel(
6268 &slave_locks->ha_handles[i],
6270 OBD_FREE(slave_locks, slave_locks_size);
6273 slave_locks->ha_handles[i] = lockh;
6275 einfo->ei_cbdata = slave_locks;
6281 * Implementation of dt_object_operations::do_invalidate.
6283 * \see dt_object_operations::do_invalidate() in the API description for details
6285 static int lod_invalidate(const struct lu_env *env, struct dt_object *dt)
6287 return dt_invalidate(env, dt_object_child(dt));
6290 static int lod_declare_instantiate_components(const struct lu_env *env,
6291 struct lod_object *lo, struct thandle *th)
6293 struct lod_thread_info *info = lod_env_info(env);
6298 LASSERT(info->lti_count < lo->ldo_comp_cnt);
6300 for (i = 0; i < info->lti_count; i++) {
6301 rc = lod_qos_prep_create(env, lo, NULL, th,
6302 info->lti_comp_idx[i]);
6308 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
6309 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
6310 &info->lti_buf, XATTR_NAME_LOV, 0, th);
6317 * Check OSTs for an existing component for further extension
6319 * Checks if OSTs are still healthy and not out of space. Gets free space
6320 * on OSTs (relative to allocation watermark rmb_low) and compares to
6321 * the proposed new_end for this component.
6323 * Decides whether or not to extend a component on its current OSTs.
6325 * \param[in] env execution environment for this thread
6326 * \param[in] lo object we're checking
6327 * \param[in] index index of this component
6328 * \param[in] extension_size extension size for this component
6329 * \param[in] extent layout extent for requested operation
6330 * \param[in] comp_extent extension component extent
6331 * \param[in] write if this is write operation
6333 * \retval true - OK to extend on current OSTs
6334 * \retval false - do not extend on current OSTs
6336 static bool lod_sel_osts_allowed(const struct lu_env *env,
6337 struct lod_object *lo,
6338 int index, __u64 extension_size,
6339 struct lu_extent *extent,
6340 struct lu_extent *comp_extent, int write)
6342 struct lod_layout_component *lod_comp = &lo->ldo_comp_entries[index];
6343 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6344 struct obd_statfs *sfs = &lod_env_info(env)->lti_osfs;
6345 __u64 available = 0;
6352 LASSERT(lod_comp->llc_stripe_count != 0);
6355 (extent->e_start == 0 && extent->e_end == OBD_OBJECT_EOF)) {
6356 /* truncate or append */
6357 size = extension_size;
6359 /* In case of write op, check the real write extent,
6360 * it may be larger than the extension_size */
6361 size = roundup(min(extent->e_end, comp_extent->e_end) -
6362 max(extent->e_start, comp_extent->e_start),
6365 /* extension_size is file level, so we must divide by stripe count to
6366 * compare it to available space on a single OST */
6367 size /= lod_comp->llc_stripe_count;
6369 lod_getref(&lod->lod_ost_descs);
6370 for (i = 0; i < lod_comp->llc_stripe_count; i++) {
6371 int index = lod_comp->llc_ost_indices[i];
6372 struct lod_tgt_desc *ost = OST_TGT(lod, index);
6373 struct obd_statfs_info info = { 0 };
6374 int j, repeated = 0;
6378 /* Get the number of times this OST repeats in this component.
6379 * Note: inter-component repeats are not counted as this is
6380 * considered as a rare case: we try to not repeat OST in other
6381 * components if possible. */
6382 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
6383 if (index != lod_comp->llc_ost_indices[j])
6386 /* already handled */
6392 if (j < lod_comp->llc_stripe_count)
6395 if (!cfs_bitmap_check(lod->lod_ost_bitmap, index)) {
6396 CDEBUG(D_LAYOUT, "ost %d no longer present\n", index);
6401 rc = dt_statfs_info(env, ost->ltd_ost, sfs, &info);
6403 CDEBUG(D_LAYOUT, "statfs failed for ost %d, error %d\n",
6409 if (sfs->os_state & OS_STATE_ENOSPC ||
6410 sfs->os_state & OS_STATE_READONLY ||
6411 sfs->os_state & OS_STATE_DEGRADED) {
6412 CDEBUG(D_LAYOUT, "ost %d is not availble for SEL "
6413 "extension, state %u\n", index, sfs->os_state);
6419 available = sfs->os_bavail * sfs->os_bsize;
6420 /* 'available' is relative to the allocation threshold */
6421 available -= (__u64) info.os_reserved_mb_low << 20;
6423 CDEBUG(D_LAYOUT, "ost %d lowwm: %d highwm: %d, "
6424 "%llu %% blocks available, %llu %% blocks free\n",
6425 index, info.os_reserved_mb_low, info.os_reserved_mb_high,
6426 (100ull * sfs->os_bavail) / sfs->os_blocks,
6427 (100ull * sfs->os_bfree) / sfs->os_blocks);
6429 if (size * repeated > available) {
6431 CDEBUG(D_LAYOUT, "low space on ost %d, available %llu "
6432 "< extension size %llu\n", index, available,
6437 lod_putref(lod, &lod->lod_ost_descs);
6443 * Adjust extents after component removal
6445 * When we remove an extension component, we move the start of the next
6446 * component to match the start of the extension component, so no space is left
6449 * \param[in] env execution environment for this thread
6450 * \param[in] lo object
6451 * \param[in] max_comp layout component
6452 * \param[in] index index of this component
6454 * \retval 0 on success
6455 * \retval negative errno on error
6457 static void lod_sel_adjust_extents(const struct lu_env *env,
6458 struct lod_object *lo,
6459 int max_comp, int index)
6461 struct lod_layout_component *lod_comp = NULL;
6462 struct lod_layout_component *next = NULL;
6463 struct lod_layout_component *prev = NULL;
6464 __u64 new_start = 0;
6468 /* Extension space component */
6469 lod_comp = &lo->ldo_comp_entries[index];
6470 next = &lo->ldo_comp_entries[index + 1];
6471 prev = &lo->ldo_comp_entries[index - 1];
6473 LASSERT(lod_comp != NULL && prev != NULL && next != NULL);
6474 LASSERT(lod_comp->llc_flags & LCME_FL_EXTENSION);
6476 /* Previous is being removed */
6477 if (prev && prev->llc_id == LCME_ID_INVAL)
6478 new_start = prev->llc_extent.e_start;
6480 new_start = lod_comp->llc_extent.e_start;
6482 for (i = index + 1; i < max_comp; i++) {
6483 lod_comp = &lo->ldo_comp_entries[i];
6485 start = lod_comp->llc_extent.e_start;
6486 lod_comp->llc_extent.e_start = new_start;
6488 /* We only move zero length extendable components */
6489 if (!(start == lod_comp->llc_extent.e_end))
6492 LASSERT(!(lod_comp->llc_flags & LCME_FL_INIT));
6494 lod_comp->llc_extent.e_end = new_start;
6498 /* Calculate the proposed 'new end' for a component we're extending */
6499 static __u64 lod_extension_new_end(__u64 extension_size, __u64 extent_end,
6500 __u32 stripe_size, __u64 component_end,
6501 __u64 extension_end)
6505 LASSERT(extension_size != 0 && stripe_size != 0);
6507 /* Round up to extension size */
6508 if (extent_end == OBD_OBJECT_EOF) {
6509 new_end = OBD_OBJECT_EOF;
6511 /* Add at least extension_size to the previous component_end,
6512 * covering the req layout extent */
6513 new_end = max(extent_end - component_end, extension_size);
6514 new_end = roundup(new_end, extension_size);
6515 new_end += component_end;
6517 /* Component end must be min stripe size aligned */
6518 if (new_end % stripe_size) {
6519 CDEBUG(D_LAYOUT, "new component end is not aligned "
6520 "by the stripe size %u: [%llu, %llu) ext size "
6521 "%llu new end %llu, aligning\n",
6522 stripe_size, component_end, extent_end,
6523 extension_size, new_end);
6524 new_end = roundup(new_end, stripe_size);
6528 if (new_end < extent_end)
6529 new_end = OBD_OBJECT_EOF;
6532 /* Don't extend past the end of the extension component */
6533 if (new_end > extension_end)
6534 new_end = extension_end;
6539 /* As lod_sel_handler() could be re-entered for the same component several
6540 * times, this is the data for the next call. Fields could be changed to
6541 * component indexes when needed, (e.g. if there is no need to instantiate
6542 * all the previous components up to the current position) to tell the caller
6543 * where to start over from. */
6550 * Process extent updates for a particular layout component
6552 * Handle layout updates for a particular extension space component touched by
6553 * a layout update operation. Core function of self-extending PFL feature.
6555 * In general, this function processes exactly *one* stage of an extension
6556 * operation, modifying the layout accordingly, then returns to the caller.
6557 * The caller is responsible for restarting processing with the new layout,
6558 * which may repeatedly return to this function until the extension updates
6561 * This function does one of a few things to the layout:
6562 * 1. Extends the component before the current extension space component to
6563 * allow it to accomodate the requested operation (if space/policy permit that
6564 * component to continue on its current OSTs)
6566 * 2. If extension of the existing component fails, we do one of two things:
6567 * a. If there is a component after the extension space, we remove the
6568 * extension space component, move the start of the next component down
6569 * accordingly, then notify the caller to restart processing w/the new
6571 * b. If there is no following component, we try repeating the current
6572 * component, creating a new component using the current one as a
6573 * template (keeping its stripe properties but not specific striping),
6574 * and try assigning striping for this component. If there is sufficient
6575 * free space on the OSTs chosen for this component, it is instantiated
6576 * and i/o continues there.
6578 * If there is not sufficient space on the new OSTs, we remove this new
6579 * component & extend the current component.
6581 * Note further that uninited components followed by extension space can be zero
6582 * length meaning that we will try to extend them before initializing them, and
6583 * if that fails, they will be removed without initialization.
6585 * 3. If we extend to/beyond the end of an extension space component, that
6586 * component is exhausted (all of its range has been given to real components),
6587 * so we remove it and restart processing.
6589 * \param[in] env execution environment for this thread
6590 * \param[in,out] lo object to update the layout of
6591 * \param[in] extent layout extent for requested operation, update
6592 * layout to fit this operation
6593 * \param[in] th transaction handle for this operation
6594 * \param[in,out] max_comp the highest comp for the portion of the layout
6595 * we are operating on (For FLR, the chosen
6596 * replica). Updated because we may remove
6598 * \param[in] index index of the extension space component we're
6600 * \param[in] write if this is write op
6601 * \param[in,out] force if the extension is to be forced; set here
6602 to force it on the 2nd call for the same
6605 * \retval 0 on success
6606 * \retval negative errno on error
6608 static int lod_sel_handler(const struct lu_env *env,
6609 struct lod_object *lo,
6610 struct lu_extent *extent,
6611 struct thandle *th, int *max_comp,
6612 int index, int write,
6613 struct sel_data *sd)
6615 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6616 struct lod_thread_info *info = lod_env_info(env);
6617 struct lod_layout_component *lod_comp;
6618 struct lod_layout_component *prev;
6619 struct lod_layout_component *next = NULL;
6620 __u64 extension_size;
6627 /* First component cannot be extension space */
6629 CERROR("%s: "DFID" first component cannot be extension space\n",
6630 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
6634 lod_comp = &lo->ldo_comp_entries[index];
6635 prev = &lo->ldo_comp_entries[index - 1];
6636 if ((index + 1) < *max_comp)
6637 next = &lo->ldo_comp_entries[index + 1];
6639 /* extension size uses the stripe size field as KiB */
6640 extension_size = lod_comp->llc_stripe_size * SEL_UNIT_SIZE;
6642 CDEBUG(D_LAYOUT, "prev start %llu, extension start %llu, extension end"
6643 " %llu, extension size %llu\n", prev->llc_extent.e_start,
6644 lod_comp->llc_extent.e_start, lod_comp->llc_extent.e_end,
6647 /* Two extension space components cannot be adjacent & extension space
6648 * components cannot be init */
6649 if ((prev->llc_flags & LCME_FL_EXTENSION) ||
6650 !(ergo(next, !(next->llc_flags & LCME_FL_EXTENSION))) ||
6651 lod_comp_inited(lod_comp)) {
6652 CERROR("%s: "DFID" invalid extension space components\n",
6653 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)));
6657 if (!prev->llc_stripe) {
6658 CDEBUG(D_LAYOUT, "Previous component not inited\n");
6659 info->lti_count = 1;
6660 info->lti_comp_idx[0] = index - 1;
6661 rc = lod_declare_instantiate_components(env, lo, th);
6662 /* ENOSPC tells us we can't use this component. If there is
6663 * a next or we are repeating, we either spill over (next) or
6664 * extend the original comp (repeat). Otherwise, return the
6665 * error to the user. */
6666 if (rc == -ENOSPC && (next || sd->sd_repeat))
6672 if (sd->sd_force == 0 && rc == 0)
6673 rc = !lod_sel_osts_allowed(env, lo, index - 1,
6674 extension_size, extent,
6675 &lod_comp->llc_extent, write);
6677 repeated = !!(sd->sd_repeat);
6681 /* Extend previous component */
6683 new_end = lod_extension_new_end(extension_size, extent->e_end,
6684 prev->llc_stripe_size,
6685 prev->llc_extent.e_end,
6686 lod_comp->llc_extent.e_end);
6688 CDEBUG(D_LAYOUT, "new end %llu\n", new_end);
6689 lod_comp->llc_extent.e_start = new_end;
6690 prev->llc_extent.e_end = new_end;
6692 if (prev->llc_extent.e_end == lod_comp->llc_extent.e_end) {
6693 CDEBUG(D_LAYOUT, "Extension component exhausted\n");
6694 lod_comp->llc_id = LCME_ID_INVAL;
6698 /* rc == 1, failed to extend current component */
6701 /* Normal 'spillover' case - Remove the extension
6702 * space component & bring down the start of the next
6704 lod_comp->llc_id = LCME_ID_INVAL;
6706 if (!(prev->llc_flags & LCME_FL_INIT)) {
6707 prev->llc_id = LCME_ID_INVAL;
6710 lod_sel_adjust_extents(env, lo, *max_comp, index);
6711 } else if (lod_comp_inited(prev)) {
6712 /* If there is no next, and the previous component is
6713 * INIT'ed, try repeating the previous component. */
6714 LASSERT(repeated == 0);
6715 rc = lod_layout_repeat_comp(env, lo, index - 1);
6719 /* The previous component is a repeated component.
6720 * Record this so we don't keep trying to repeat it. */
6723 /* If the previous component is not INIT'ed, this may
6724 * be a component we have just instantiated but failed
6725 * to extend. Or even a repeated component we failed
6726 * to prepare a striping for. Do not repeat but instead
6727 * remove the repeated component & force the extention
6728 * of the original one */
6731 prev->llc_id = LCME_ID_INVAL;
6738 rc = lod_layout_del_prep_layout(env, lo, NULL);
6741 LASSERTF(-rc == change,
6742 "number deleted %d != requested %d\n", -rc,
6745 *max_comp = *max_comp + change;
6747 /* lod_del_prep_layout reallocates ldo_comp_entries, so we must
6748 * refresh these pointers before using them */
6749 lod_comp = &lo->ldo_comp_entries[index];
6750 prev = &lo->ldo_comp_entries[index - 1];
6751 CDEBUG(D_LAYOUT, "After extent updates: prev start %llu, current start "
6752 "%llu, current end %llu max_comp %d ldo_comp_cnt %d\n",
6753 prev->llc_extent.e_start, lod_comp->llc_extent.e_start,
6754 lod_comp->llc_extent.e_end, *max_comp, lo->ldo_comp_cnt);
6756 /* Layout changed successfully */
6761 * Declare layout extent updates
6763 * Handles extensions. Identifies extension components touched by current
6764 * operation and passes them to processing function.
6766 * Restarts with updated layouts from the processing function until the current
6767 * operation no longer touches an extension space component.
6769 * \param[in] env execution environment for this thread
6770 * \param[in,out] lo object to update the layout of
6771 * \param[in] extent layout extent for requested operation, update layout to
6772 * fit this operation
6773 * \param[in] th transaction handle for this operation
6774 * \param[in] pick identifies chosen mirror for FLR layouts
6775 * \param[in] write if this is write op
6777 * \retval 1 on layout changed, 0 on no change
6778 * \retval negative errno on error
6780 static int lod_declare_update_extents(const struct lu_env *env,
6781 struct lod_object *lo, struct lu_extent *extent,
6782 struct thandle *th, int pick, int write)
6784 struct lod_thread_info *info = lod_env_info(env);
6785 struct lod_layout_component *lod_comp;
6786 bool layout_changed = false;
6787 struct sel_data sd = { 0 };
6795 /* This makes us work on the components of the chosen mirror */
6796 start_index = lo->ldo_mirrors[pick].lme_start;
6797 max_comp = lo->ldo_mirrors[pick].lme_end + 1;
6798 if (lo->ldo_flr_state == LCM_FL_NONE)
6799 LASSERT(start_index == 0 && max_comp == lo->ldo_comp_cnt);
6801 CDEBUG(D_LAYOUT, "extent->e_start %llu, extent->e_end %llu\n",
6802 extent->e_start, extent->e_end);
6803 for (i = start_index; i < max_comp; i++) {
6804 lod_comp = &lo->ldo_comp_entries[i];
6806 /* We've passed all components of interest */
6807 if (lod_comp->llc_extent.e_start >= extent->e_end)
6810 if (lod_comp->llc_flags & LCME_FL_EXTENSION) {
6811 layout_changed = true;
6812 rc = lod_sel_handler(env, lo, extent, th, &max_comp,
6817 /* Nothing has changed behind the prev one */
6823 /* We may have added or removed components. If so, we must update the
6824 * start & ends of all the mirrors after the current one, and the end
6825 * of the current mirror. */
6826 change = max_comp - 1 - lo->ldo_mirrors[pick].lme_end;
6828 lo->ldo_mirrors[pick].lme_end += change;
6829 for (i = pick + 1; i < lo->ldo_mirror_count; i++) {
6830 lo->ldo_mirrors[i].lme_start += change;
6831 lo->ldo_mirrors[i].lme_end += change;
6837 /* The amount of components has changed, adjust the lti_comp_idx */
6838 rc2 = lod_layout_data_init(info, lo->ldo_comp_cnt);
6840 return rc < 0 ? rc : rc2 < 0 ? rc2 : layout_changed;
6843 /* If striping is already instantiated or INIT'ed DOM? */
6844 static bool lod_is_instantiation_needed(struct lod_layout_component *comp)
6846 return !(((lov_pattern(comp->llc_pattern) == LOV_PATTERN_MDT) &&
6847 lod_comp_inited(comp)) || comp->llc_stripe);
6851 * Declare layout update for a non-FLR layout.
6853 * \param[in] env execution environment for this thread
6854 * \param[in,out] lo object to update the layout of
6855 * \param[in] layout layout intent for requested operation, "update" is
6856 * a process of reacting to this
6857 * \param[in] buf buffer containing lov ea (see comment on usage inline)
6858 * \param[in] th transaction handle for this operation
6860 * \retval 0 on success
6861 * \retval negative errno on error
6863 static int lod_declare_update_plain(const struct lu_env *env,
6864 struct lod_object *lo, struct layout_intent *layout,
6865 const struct lu_buf *buf, struct thandle *th)
6867 struct lod_thread_info *info = lod_env_info(env);
6868 struct lod_device *d = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
6869 struct lod_layout_component *lod_comp;
6870 struct lov_comp_md_v1 *comp_v1 = NULL;
6871 bool layout_changed = false;
6872 bool replay = false;
6876 LASSERT(lo->ldo_flr_state == LCM_FL_NONE);
6879 * In case the client is passing lovea, which only happens during
6880 * the replay of layout intent write RPC for now, we may need to
6881 * parse the lovea and apply new layout configuration.
6883 if (buf && buf->lb_len) {
6884 struct lov_user_md_v1 *v1 = buf->lb_buf;
6886 if (v1->lmm_magic != (LOV_MAGIC_DEFINED | LOV_MAGIC_COMP_V1) &&
6887 v1->lmm_magic != __swab32(LOV_MAGIC_DEFINED |
6888 LOV_MAGIC_COMP_V1)) {
6889 CERROR("%s: the replay buffer of layout extend "
6890 "(magic %#x) does not contain expected "
6891 "composite layout.\n",
6892 lod2obd(d)->obd_name, v1->lmm_magic);
6893 GOTO(out, rc = -EINVAL);
6896 rc = lod_use_defined_striping(env, lo, buf);
6899 lo->ldo_comp_cached = 1;
6901 rc = lod_get_lov_ea(env, lo);
6904 /* old on-disk EA is stored in info->lti_buf */
6905 comp_v1 = (struct lov_comp_md_v1 *)info->lti_buf.lb_buf;
6907 layout_changed = true;
6909 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
6913 /* non replay path */
6914 rc = lod_striping_load(env, lo);
6919 /* Make sure defined layout covers the requested write range. */
6920 lod_comp = &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1];
6921 if (lo->ldo_comp_cnt > 1 &&
6922 lod_comp->llc_extent.e_end != OBD_OBJECT_EOF &&
6923 lod_comp->llc_extent.e_end < layout->li_extent.e_end) {
6924 CDEBUG_LIMIT(replay ? D_ERROR : D_LAYOUT,
6925 "%s: the defined layout [0, %#llx) does not "
6926 "covers the write range "DEXT"\n",
6927 lod2obd(d)->obd_name, lod_comp->llc_extent.e_end,
6928 PEXT(&layout->li_extent));
6929 GOTO(out, rc = -EINVAL);
6932 CDEBUG(D_LAYOUT, "%s: "DFID": update components "DEXT"\n",
6933 lod2obd(d)->obd_name, PFID(lod_object_fid(lo)),
6934 PEXT(&layout->li_extent));
6937 rc = lod_declare_update_extents(env, lo, &layout->li_extent,
6938 th, 0, layout->li_opc == LAYOUT_INTENT_WRITE);
6942 layout_changed = true;
6946 * Iterate ld->ldo_comp_entries, find the component whose extent under
6947 * the write range and not instantianted.
6949 for (i = 0; i < lo->ldo_comp_cnt; i++) {
6950 lod_comp = &lo->ldo_comp_entries[i];
6952 if (lod_comp->llc_extent.e_start >= layout->li_extent.e_end)
6956 /* If striping is instantiated or INIT'ed DOM skip */
6957 if (!lod_is_instantiation_needed(lod_comp))
6961 * In replay path, lod_comp is the EA passed by
6962 * client replay buffer, comp_v1 is the pre-recovery
6963 * on-disk EA, we'd sift out those components which
6964 * were init-ed in the on-disk EA.
6966 if (le32_to_cpu(comp_v1->lcm_entries[i].lcme_flags) &
6971 * this component hasn't instantiated in normal path, or during
6972 * replay it needs replay the instantiation.
6975 /* A released component is being extended */
6976 if (lod_comp->llc_pattern & LOV_PATTERN_F_RELEASED)
6977 GOTO(out, rc = -EINVAL);
6979 LASSERT(info->lti_comp_idx != NULL);
6980 info->lti_comp_idx[info->lti_count++] = i;
6981 layout_changed = true;
6984 if (!layout_changed)
6987 lod_obj_inc_layout_gen(lo);
6988 rc = lod_declare_instantiate_components(env, lo, th);
6992 lod_striping_free(env, lo);
6996 static inline int lod_comp_index(struct lod_object *lo,
6997 struct lod_layout_component *lod_comp)
6999 LASSERT(lod_comp >= lo->ldo_comp_entries &&
7000 lod_comp <= &lo->ldo_comp_entries[lo->ldo_comp_cnt - 1]);
7002 return lod_comp - lo->ldo_comp_entries;
7006 * Stale other mirrors by writing extent.
7008 static int lod_stale_components(const struct lu_env *env, struct lod_object *lo,
7009 int primary, struct lu_extent *extent,
7012 struct lod_layout_component *pri_comp, *lod_comp;
7013 struct lu_extent pri_extent;
7018 /* The writing extent decides which components in the primary
7019 * are affected... */
7020 CDEBUG(D_LAYOUT, "primary mirror %d, "DEXT"\n", primary, PEXT(extent));
7023 lod_foreach_mirror_comp(pri_comp, lo, primary) {
7024 if (!lu_extent_is_overlapped(extent, &pri_comp->llc_extent))
7027 CDEBUG(D_LAYOUT, "primary comp %u "DEXT"\n",
7028 lod_comp_index(lo, pri_comp),
7029 PEXT(&pri_comp->llc_extent));
7031 pri_extent.e_start = pri_comp->llc_extent.e_start;
7032 pri_extent.e_end = pri_comp->llc_extent.e_end;
7034 for (i = 0; i < lo->ldo_mirror_count; i++) {
7037 rc = lod_declare_update_extents(env, lo, &pri_extent,
7039 /* if update_extents changed the layout, it may have
7040 * reallocated the component array, so start over to
7041 * avoid using stale pointers */
7047 /* ... and then stale other components that are
7048 * overlapping with primary components */
7049 lod_foreach_mirror_comp(lod_comp, lo, i) {
7050 if (!lu_extent_is_overlapped(
7052 &lod_comp->llc_extent))
7055 CDEBUG(D_LAYOUT, "stale: %u / %u\n",
7056 i, lod_comp_index(lo, lod_comp));
7058 lod_comp->llc_flags |= LCME_FL_STALE;
7059 lo->ldo_mirrors[i].lme_stale = 1;
7068 * check an OST's availability
7069 * \param[in] env execution environment
7070 * \param[in] lo lod object
7071 * \param[in] dt dt object
7072 * \param[in] index mirror index
7074 * \retval negative if failed
7075 * \retval 1 if \a dt is available
7076 * \retval 0 if \a dt is not available
7078 static inline int lod_check_ost_avail(const struct lu_env *env,
7079 struct lod_object *lo,
7080 struct dt_object *dt, int index)
7082 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7083 struct lod_tgt_desc *ost;
7085 int type = LU_SEQ_RANGE_OST;
7088 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu), &idx, &type);
7090 CERROR("%s: can't locate "DFID":rc = %d\n",
7091 lod2obd(lod)->obd_name, PFID(lu_object_fid(&dt->do_lu)),
7096 ost = OST_TGT(lod, idx);
7097 if (ost->ltd_statfs.os_state &
7098 (OS_STATE_READONLY | OS_STATE_ENOSPC | OS_STATE_ENOINO |
7099 OS_STATE_NOPRECREATE) ||
7100 ost->ltd_active == 0) {
7101 CDEBUG(D_LAYOUT, DFID ": mirror %d OST%d unavail, rc = %d\n",
7102 PFID(lod_object_fid(lo)), index, idx, rc);
7110 * Pick primary mirror for write
7111 * \param[in] env execution environment
7112 * \param[in] lo object
7113 * \param[in] extent write range
7115 static int lod_primary_pick(const struct lu_env *env, struct lod_object *lo,
7116 struct lu_extent *extent)
7118 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
7119 unsigned int seq = 0;
7120 struct lod_layout_component *lod_comp;
7122 int picked = -1, second_pick = -1, third_pick = -1;
7125 if (OBD_FAIL_CHECK(OBD_FAIL_FLR_RANDOM_PICK_MIRROR)) {
7126 get_random_bytes(&seq, sizeof(seq));
7127 seq %= lo->ldo_mirror_count;
7131 * Pick a mirror as the primary, and check the availability of OSTs.
7133 * This algo can be revised later after knowing the topology of
7136 lod_qos_statfs_update(env, lod);
7137 for (i = 0; i < lo->ldo_mirror_count; i++) {
7138 bool ost_avail = true;
7139 int index = (i + seq) % lo->ldo_mirror_count;
7141 if (lo->ldo_mirrors[index].lme_stale) {
7142 CDEBUG(D_LAYOUT, DFID": mirror %d stale\n",
7143 PFID(lod_object_fid(lo)), index);
7147 /* 2nd pick is for the primary mirror containing unavail OST */
7148 if (lo->ldo_mirrors[index].lme_primary && second_pick < 0)
7149 second_pick = index;
7151 /* 3rd pick is for non-primary mirror containing unavail OST */
7152 if (second_pick < 0 && third_pick < 0)
7156 * we found a non-primary 1st pick, we'd like to find a
7157 * potential pirmary mirror.
7159 if (picked >= 0 && !lo->ldo_mirrors[index].lme_primary)
7162 /* check the availability of OSTs */
7163 lod_foreach_mirror_comp(lod_comp, lo, index) {
7164 if (!lod_comp_inited(lod_comp) || !lod_comp->llc_stripe)
7167 for (j = 0; j < lod_comp->llc_stripe_count; j++) {
7168 struct dt_object *dt = lod_comp->llc_stripe[j];
7170 rc = lod_check_ost_avail(env, lo, dt, index);
7177 } /* for all dt object in one component */
7180 } /* for all components in a mirror */
7183 * the OSTs where allocated objects locates in the components
7184 * of the mirror are available.
7189 /* this mirror has all OSTs available */
7193 * primary with all OSTs are available, this is the perfect
7196 if (lo->ldo_mirrors[index].lme_primary)
7198 } /* for all mirrors */
7200 /* failed to pick a sound mirror, lower our expectation */
7202 picked = second_pick;
7204 picked = third_pick;
7211 static int lod_prepare_resync_mirror(const struct lu_env *env,
7212 struct lod_object *lo,
7215 struct lod_thread_info *info = lod_env_info(env);
7216 struct lod_layout_component *lod_comp;
7217 bool neg = !!(MIRROR_ID_NEG & mirror_id);
7220 mirror_id &= ~MIRROR_ID_NEG;
7222 for (i = 0; i < lo->ldo_mirror_count; i++) {
7223 if ((!neg && lo->ldo_mirrors[i].lme_id != mirror_id) ||
7224 (neg && lo->ldo_mirrors[i].lme_id == mirror_id))
7227 lod_foreach_mirror_comp(lod_comp, lo, i) {
7228 if (lod_comp_inited(lod_comp))
7231 info->lti_comp_idx[info->lti_count++] =
7232 lod_comp_index(lo, lod_comp);
7240 * figure out the components should be instantiated for resync.
7242 static int lod_prepare_resync(const struct lu_env *env, struct lod_object *lo,
7243 struct lu_extent *extent)
7245 struct lod_thread_info *info = lod_env_info(env);
7246 struct lod_layout_component *lod_comp;
7247 unsigned int need_sync = 0;
7251 DFID": instantiate all stale components in "DEXT"\n",
7252 PFID(lod_object_fid(lo)), PEXT(extent));
7255 * instantiate all components within this extent, even non-stale
7258 for (i = 0; i < lo->ldo_mirror_count; i++) {
7259 if (!lo->ldo_mirrors[i].lme_stale)
7262 lod_foreach_mirror_comp(lod_comp, lo, i) {
7263 if (!lu_extent_is_overlapped(extent,
7264 &lod_comp->llc_extent))
7269 if (lod_comp_inited(lod_comp))
7272 CDEBUG(D_LAYOUT, "resync instantiate %d / %d\n",
7273 i, lod_comp_index(lo, lod_comp));
7274 info->lti_comp_idx[info->lti_count++] =
7275 lod_comp_index(lo, lod_comp);
7279 return need_sync ? 0 : -EALREADY;
7282 static int lod_declare_update_rdonly(const struct lu_env *env,
7283 struct lod_object *lo, struct md_layout_change *mlc,
7286 struct lod_thread_info *info = lod_env_info(env);
7287 struct lu_attr *layout_attr = &info->lti_layout_attr;
7288 struct lod_layout_component *lod_comp;
7289 struct lu_extent extent = { 0 };
7293 LASSERT(lo->ldo_flr_state == LCM_FL_RDONLY);
7294 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
7295 mlc->mlc_opc == MD_LAYOUT_RESYNC);
7296 LASSERT(lo->ldo_mirror_count > 0);
7298 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7299 struct layout_intent *layout = mlc->mlc_intent;
7300 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
7303 extent = layout->li_extent;
7304 CDEBUG(D_LAYOUT, DFID": trying to write :"DEXT"\n",
7305 PFID(lod_object_fid(lo)), PEXT(&extent));
7307 picked = lod_primary_pick(env, lo, &extent);
7311 CDEBUG(D_LAYOUT, DFID": picked mirror id %u as primary\n",
7312 PFID(lod_object_fid(lo)),
7313 lo->ldo_mirrors[picked].lme_id);
7315 /* Update extents of primary before staling */
7316 rc = lod_declare_update_extents(env, lo, &extent, th, picked,
7321 if (layout->li_opc == LAYOUT_INTENT_TRUNC) {
7323 * trunc transfers [0, size) in the intent extent, we'd
7324 * stale components overlapping [size, eof).
7326 extent.e_start = extent.e_end;
7327 extent.e_end = OBD_OBJECT_EOF;
7330 /* stale overlapping components from other mirrors */
7331 rc = lod_stale_components(env, lo, picked, &extent, th);
7335 /* restore truncate intent extent */
7336 if (layout->li_opc == LAYOUT_INTENT_TRUNC)
7337 extent.e_end = extent.e_start;
7339 /* instantiate components for the picked mirror, start from 0 */
7342 lod_foreach_mirror_comp(lod_comp, lo, picked) {
7343 if (!lu_extent_is_overlapped(&extent,
7344 &lod_comp->llc_extent))
7347 if (!lod_is_instantiation_needed(lod_comp))
7350 info->lti_comp_idx[info->lti_count++] =
7351 lod_comp_index(lo, lod_comp);
7354 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
7355 } else { /* MD_LAYOUT_RESYNC */
7359 * could contain multiple non-stale mirrors, so we need to
7360 * prep uninited all components assuming any non-stale mirror
7361 * could be picked as the primary mirror.
7363 if (mlc->mlc_mirror_id == 0) {
7365 for (i = 0; i < lo->ldo_mirror_count; i++) {
7366 if (lo->ldo_mirrors[i].lme_stale)
7369 lod_foreach_mirror_comp(lod_comp, lo, i) {
7370 if (!lod_comp_inited(lod_comp))
7374 lod_comp->llc_extent.e_end)
7376 lod_comp->llc_extent.e_end;
7379 rc = lod_prepare_resync(env, lo, &extent);
7383 /* mirror write, try to init its all components */
7384 rc = lod_prepare_resync_mirror(env, lo,
7385 mlc->mlc_mirror_id);
7390 /* change the file state to SYNC_PENDING */
7391 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
7394 /* Reset the layout version once it's becoming too large.
7395 * This way it can make sure that the layout version is
7396 * monotonously increased in this writing era. */
7397 lod_obj_inc_layout_gen(lo);
7398 if (lo->ldo_layout_gen > (LCME_ID_MAX >> 1)) {
7399 __u32 layout_version;
7401 get_random_bytes(&layout_version, sizeof(layout_version));
7402 lo->ldo_layout_gen = layout_version & 0xffff;
7405 rc = lod_declare_instantiate_components(env, lo, th);
7409 layout_attr->la_valid = LA_LAYOUT_VERSION;
7410 layout_attr->la_layout_version = 0; /* set current version */
7411 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
7412 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
7413 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
7419 lod_striping_free(env, lo);
7423 static int lod_declare_update_write_pending(const struct lu_env *env,
7424 struct lod_object *lo, struct md_layout_change *mlc,
7427 struct lod_thread_info *info = lod_env_info(env);
7428 struct lu_attr *layout_attr = &info->lti_layout_attr;
7429 struct lod_layout_component *lod_comp;
7430 struct lu_extent extent = { 0 };
7436 LASSERT(lo->ldo_flr_state == LCM_FL_WRITE_PENDING);
7437 LASSERT(mlc->mlc_opc == MD_LAYOUT_WRITE ||
7438 mlc->mlc_opc == MD_LAYOUT_RESYNC);
7440 /* look for the primary mirror */
7441 for (i = 0; i < lo->ldo_mirror_count; i++) {
7442 if (lo->ldo_mirrors[i].lme_stale)
7445 LASSERTF(primary < 0, DFID " has multiple primary: %u / %u",
7446 PFID(lod_object_fid(lo)),
7447 lo->ldo_mirrors[i].lme_id,
7448 lo->ldo_mirrors[primary].lme_id);
7453 CERROR(DFID ": doesn't have a primary mirror\n",
7454 PFID(lod_object_fid(lo)));
7455 GOTO(out, rc = -ENODATA);
7458 CDEBUG(D_LAYOUT, DFID": found primary %u\n",
7459 PFID(lod_object_fid(lo)), lo->ldo_mirrors[primary].lme_id);
7461 LASSERT(!lo->ldo_mirrors[primary].lme_stale);
7463 /* for LAYOUT_WRITE opc, it has to do the following operations:
7464 * 1. stale overlapping componets from stale mirrors;
7465 * 2. instantiate components of the primary mirror;
7466 * 3. transfter layout version to all objects of the primary;
7468 * for LAYOUT_RESYNC opc, it will do:
7469 * 1. instantiate components of all stale mirrors;
7470 * 2. transfer layout version to all objects to close write era. */
7472 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7473 struct layout_intent *layout = mlc->mlc_intent;
7474 int write = layout->li_opc == LAYOUT_INTENT_WRITE;
7476 LASSERT(mlc->mlc_intent != NULL);
7478 extent = mlc->mlc_intent->li_extent;
7480 CDEBUG(D_LAYOUT, DFID": intent to write: "DEXT"\n",
7481 PFID(lod_object_fid(lo)), PEXT(&extent));
7483 /* 1. Update extents of primary before staling */
7484 rc = lod_declare_update_extents(env, lo, &extent, th, primary,
7489 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC) {
7491 * trunc transfers [0, size) in the intent extent, we'd
7492 * stale components overlapping [size, eof).
7494 extent.e_start = extent.e_end;
7495 extent.e_end = OBD_OBJECT_EOF;
7498 /* 2. stale overlapping components */
7499 rc = lod_stale_components(env, lo, primary, &extent, th);
7503 /* 3. find the components which need instantiating.
7504 * instantiate [0, mlc->mlc_intent->e_end) */
7506 /* restore truncate intent extent */
7507 if (mlc->mlc_intent->li_opc == LAYOUT_INTENT_TRUNC)
7508 extent.e_end = extent.e_start;
7511 lod_foreach_mirror_comp(lod_comp, lo, primary) {
7512 if (!lu_extent_is_overlapped(&extent,
7513 &lod_comp->llc_extent))
7516 if (!lod_is_instantiation_needed(lod_comp))
7519 CDEBUG(D_LAYOUT, "write instantiate %d / %d\n",
7520 primary, lod_comp_index(lo, lod_comp));
7521 info->lti_comp_idx[info->lti_count++] =
7522 lod_comp_index(lo, lod_comp);
7524 } else { /* MD_LAYOUT_RESYNC */
7525 if (mlc->mlc_mirror_id == 0) {
7527 lod_foreach_mirror_comp(lod_comp, lo, primary) {
7528 if (!lod_comp_inited(lod_comp))
7531 extent.e_end = lod_comp->llc_extent.e_end;
7534 rc = lod_prepare_resync(env, lo, &extent);
7538 /* mirror write, try to init its all components */
7539 rc = lod_prepare_resync_mirror(env, lo,
7540 mlc->mlc_mirror_id);
7545 /* change the file state to SYNC_PENDING */
7546 lo->ldo_flr_state = LCM_FL_SYNC_PENDING;
7549 rc = lod_declare_instantiate_components(env, lo, th);
7553 /* 3. transfer layout version to OST objects.
7554 * transfer new layout version to OST objects so that stale writes
7555 * can be denied. It also ends an era of writing by setting
7556 * LU_LAYOUT_RESYNC. Normal client can never use this bit to
7557 * send write RPC; only resync RPCs could do it. */
7558 layout_attr->la_valid = LA_LAYOUT_VERSION;
7559 layout_attr->la_layout_version = 0; /* set current version */
7560 if (mlc->mlc_opc == MD_LAYOUT_RESYNC)
7561 layout_attr->la_layout_version = LU_LAYOUT_RESYNC;
7562 rc = lod_declare_attr_set(env, &lo->ldo_obj, layout_attr, th);
7566 lod_obj_inc_layout_gen(lo);
7569 lod_striping_free(env, lo);
7573 static int lod_declare_update_sync_pending(const struct lu_env *env,
7574 struct lod_object *lo, struct md_layout_change *mlc,
7577 struct lod_thread_info *info = lod_env_info(env);
7578 unsigned sync_components = 0;
7579 unsigned resync_components = 0;
7584 LASSERT(lo->ldo_flr_state == LCM_FL_SYNC_PENDING);
7585 LASSERT(mlc->mlc_opc == MD_LAYOUT_RESYNC_DONE ||
7586 mlc->mlc_opc == MD_LAYOUT_WRITE);
7588 CDEBUG(D_LAYOUT, DFID ": received op %d in sync pending\n",
7589 PFID(lod_object_fid(lo)), mlc->mlc_opc);
7591 if (mlc->mlc_opc == MD_LAYOUT_WRITE) {
7592 CDEBUG(D_LAYOUT, DFID": cocurrent write to sync pending\n",
7593 PFID(lod_object_fid(lo)));
7595 lo->ldo_flr_state = LCM_FL_WRITE_PENDING;
7596 return lod_declare_update_write_pending(env, lo, mlc, th);
7599 /* MD_LAYOUT_RESYNC_DONE */
7601 for (i = 0; i < lo->ldo_comp_cnt; i++) {
7602 struct lod_layout_component *lod_comp;
7605 lod_comp = &lo->ldo_comp_entries[i];
7607 if (!(lod_comp->llc_flags & LCME_FL_STALE)) {
7612 for (j = 0; j < mlc->mlc_resync_count; j++) {
7613 if (lod_comp->llc_id != mlc->mlc_resync_ids[j])
7616 mlc->mlc_resync_ids[j] = LCME_ID_INVAL;
7617 lod_comp->llc_flags &= ~LCME_FL_STALE;
7618 resync_components++;
7624 for (i = 0; i < mlc->mlc_resync_count; i++) {
7625 if (mlc->mlc_resync_ids[i] == LCME_ID_INVAL)
7628 CDEBUG(D_LAYOUT, DFID": lcme id %u (%d / %zd) not exist "
7629 "or already synced\n", PFID(lod_object_fid(lo)),
7630 mlc->mlc_resync_ids[i], i, mlc->mlc_resync_count);
7631 GOTO(out, rc = -EINVAL);
7634 if (!sync_components || (mlc->mlc_resync_count && !resync_components)) {
7635 CDEBUG(D_LAYOUT, DFID": no mirror in sync\n",
7636 PFID(lod_object_fid(lo)));
7638 /* tend to return an error code here to prevent
7639 * the MDT from setting SoM attribute */
7640 GOTO(out, rc = -EINVAL);
7643 CDEBUG(D_LAYOUT, DFID": synced %u resynced %u/%zu components\n",
7644 PFID(lod_object_fid(lo)),
7645 sync_components, resync_components, mlc->mlc_resync_count);
7647 lo->ldo_flr_state = LCM_FL_RDONLY;
7648 lod_obj_inc_layout_gen(lo);
7650 info->lti_buf.lb_len = lod_comp_md_size(lo, false);
7651 rc = lod_sub_declare_xattr_set(env, lod_object_child(lo),
7652 &info->lti_buf, XATTR_NAME_LOV, 0, th);
7657 lod_striping_free(env, lo);
7661 static int lod_declare_layout_change(const struct lu_env *env,
7662 struct dt_object *dt, struct md_layout_change *mlc,
7665 struct lod_thread_info *info = lod_env_info(env);
7666 struct lod_object *lo = lod_dt_obj(dt);
7670 if (!S_ISREG(dt->do_lu.lo_header->loh_attr) || !dt_object_exists(dt) ||
7671 dt_object_remote(dt_object_child(dt)))
7674 rc = lod_striping_load(env, lo);
7678 LASSERT(lo->ldo_comp_cnt > 0);
7680 rc = lod_layout_data_init(info, lo->ldo_comp_cnt);
7684 switch (lo->ldo_flr_state) {
7686 rc = lod_declare_update_plain(env, lo, mlc->mlc_intent,
7690 rc = lod_declare_update_rdonly(env, lo, mlc, th);
7692 case LCM_FL_WRITE_PENDING:
7693 rc = lod_declare_update_write_pending(env, lo, mlc, th);
7695 case LCM_FL_SYNC_PENDING:
7696 rc = lod_declare_update_sync_pending(env, lo, mlc, th);
7707 * Instantiate layout component objects which covers the intent write offset.
7709 static int lod_layout_change(const struct lu_env *env, struct dt_object *dt,
7710 struct md_layout_change *mlc, struct thandle *th)
7712 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
7713 struct lu_attr *layout_attr = &lod_env_info(env)->lti_layout_attr;
7714 struct lod_object *lo = lod_dt_obj(dt);
7717 rc = lod_striped_create(env, dt, attr, NULL, th);
7718 if (!rc && layout_attr->la_valid & LA_LAYOUT_VERSION) {
7719 layout_attr->la_layout_version |= lo->ldo_layout_gen;
7720 rc = lod_attr_set(env, dt, layout_attr, th);
7726 struct dt_object_operations lod_obj_ops = {
7727 .do_read_lock = lod_read_lock,
7728 .do_write_lock = lod_write_lock,
7729 .do_read_unlock = lod_read_unlock,
7730 .do_write_unlock = lod_write_unlock,
7731 .do_write_locked = lod_write_locked,
7732 .do_attr_get = lod_attr_get,
7733 .do_declare_attr_set = lod_declare_attr_set,
7734 .do_attr_set = lod_attr_set,
7735 .do_xattr_get = lod_xattr_get,
7736 .do_declare_xattr_set = lod_declare_xattr_set,
7737 .do_xattr_set = lod_xattr_set,
7738 .do_declare_xattr_del = lod_declare_xattr_del,
7739 .do_xattr_del = lod_xattr_del,
7740 .do_xattr_list = lod_xattr_list,
7741 .do_ah_init = lod_ah_init,
7742 .do_declare_create = lod_declare_create,
7743 .do_create = lod_create,
7744 .do_declare_destroy = lod_declare_destroy,
7745 .do_destroy = lod_destroy,
7746 .do_index_try = lod_index_try,
7747 .do_declare_ref_add = lod_declare_ref_add,
7748 .do_ref_add = lod_ref_add,
7749 .do_declare_ref_del = lod_declare_ref_del,
7750 .do_ref_del = lod_ref_del,
7751 .do_object_sync = lod_object_sync,
7752 .do_object_lock = lod_object_lock,
7753 .do_object_unlock = lod_object_unlock,
7754 .do_invalidate = lod_invalidate,
7755 .do_declare_layout_change = lod_declare_layout_change,
7756 .do_layout_change = lod_layout_change,
7760 * Implementation of dt_body_operations::dbo_read.
7762 * \see dt_body_operations::dbo_read() in the API description for details.
7764 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
7765 struct lu_buf *buf, loff_t *pos)
7767 struct dt_object *next = dt_object_child(dt);
7769 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
7770 S_ISLNK(dt->do_lu.lo_header->loh_attr));
7771 return next->do_body_ops->dbo_read(env, next, buf, pos);
7775 * Implementation of dt_body_operations::dbo_declare_write.
7777 * \see dt_body_operations::dbo_declare_write() in the API description
7780 static ssize_t lod_declare_write(const struct lu_env *env,
7781 struct dt_object *dt,
7782 const struct lu_buf *buf, loff_t pos,
7785 return lod_sub_declare_write(env, dt_object_child(dt), buf, pos, th);
7789 * Implementation of dt_body_operations::dbo_write.
7791 * \see dt_body_operations::dbo_write() in the API description for details.
7793 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
7794 const struct lu_buf *buf, loff_t *pos,
7797 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr) ||
7798 S_ISLNK(dt->do_lu.lo_header->loh_attr));
7799 return lod_sub_write(env, dt_object_child(dt), buf, pos, th);
7802 static int lod_declare_punch(const struct lu_env *env, struct dt_object *dt,
7803 __u64 start, __u64 end, struct thandle *th)
7805 if (dt_object_remote(dt))
7808 return lod_sub_declare_punch(env, dt_object_child(dt), start, end, th);
7811 static int lod_punch(const struct lu_env *env, struct dt_object *dt,
7812 __u64 start, __u64 end, struct thandle *th)
7814 if (dt_object_remote(dt))
7817 LASSERT(S_ISREG(dt->do_lu.lo_header->loh_attr));
7818 return lod_sub_punch(env, dt_object_child(dt), start, end, th);
7822 * different type of files use the same body_ops because object may be created
7823 * in OUT, where there is no chance to set correct body_ops for each type, so
7824 * body_ops themselves will check file type inside, see lod_read/write/punch for
7827 const struct dt_body_operations lod_body_ops = {
7828 .dbo_read = lod_read,
7829 .dbo_declare_write = lod_declare_write,
7830 .dbo_write = lod_write,
7831 .dbo_declare_punch = lod_declare_punch,
7832 .dbo_punch = lod_punch,
7836 * Implementation of lu_object_operations::loo_object_init.
7838 * The function determines the type and the index of the target device using
7839 * sequence of the object's FID. Then passes control down to the
7840 * corresponding device:
7841 * OSD for the local objects, OSP for remote
7843 * \see lu_object_operations::loo_object_init() in the API description
7846 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
7847 const struct lu_object_conf *conf)
7849 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
7850 struct lu_device *cdev = NULL;
7851 struct lu_object *cobj;
7852 struct lod_tgt_descs *ltd = NULL;
7853 struct lod_tgt_desc *tgt;
7855 int type = LU_SEQ_RANGE_ANY;
7859 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
7863 if (type == LU_SEQ_RANGE_MDT &&
7864 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
7865 cdev = &lod->lod_child->dd_lu_dev;
7866 } else if (type == LU_SEQ_RANGE_MDT) {
7867 ltd = &lod->lod_mdt_descs;
7869 } else if (type == LU_SEQ_RANGE_OST) {
7870 ltd = &lod->lod_ost_descs;
7877 if (ltd->ltd_tgts_size > idx &&
7878 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
7879 tgt = LTD_TGT(ltd, idx);
7881 LASSERT(tgt != NULL);
7882 LASSERT(tgt->ltd_tgt != NULL);
7884 cdev = &(tgt->ltd_tgt->dd_lu_dev);
7886 lod_putref(lod, ltd);
7889 if (unlikely(cdev == NULL))
7892 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
7893 if (unlikely(cobj == NULL))
7896 lu2lod_obj(lo)->ldo_obj.do_body_ops = &lod_body_ops;
7898 lu_object_add(lo, cobj);
7905 * Alloc cached foreign LOV
7907 * \param[in] lo object
7908 * \param[in] size size of foreign LOV
7910 * \retval 0 on success
7911 * \retval negative if failed
7913 int lod_alloc_foreign_lov(struct lod_object *lo, size_t size)
7915 OBD_ALLOC_LARGE(lo->ldo_foreign_lov, size);
7916 if (lo->ldo_foreign_lov == NULL)
7918 lo->ldo_foreign_lov_size = size;
7919 lo->ldo_is_foreign = 1;
7925 * Free cached foreign LOV
7927 * \param[in] lo object
7929 void lod_free_foreign_lov(struct lod_object *lo)
7931 if (lo->ldo_foreign_lov != NULL)
7932 OBD_FREE_LARGE(lo->ldo_foreign_lov, lo->ldo_foreign_lov_size);
7933 lo->ldo_foreign_lov = NULL;
7934 lo->ldo_foreign_lov_size = 0;
7935 lo->ldo_is_foreign = 0;
7940 * Free cached foreign LMV
7942 * \param[in] lo object
7944 void lod_free_foreign_lmv(struct lod_object *lo)
7946 if (lo->ldo_foreign_lmv != NULL)
7947 OBD_FREE_LARGE(lo->ldo_foreign_lmv, lo->ldo_foreign_lmv_size);
7948 lo->ldo_foreign_lmv = NULL;
7949 lo->ldo_foreign_lmv_size = 0;
7950 lo->ldo_dir_is_foreign = 0;
7955 * Release resources associated with striping.
7957 * If the object is striped (regular or directory), then release
7958 * the stripe objects references and free the ldo_stripe array.
7960 * \param[in] env execution environment
7961 * \param[in] lo object
7963 void lod_striping_free_nolock(const struct lu_env *env, struct lod_object *lo)
7965 struct lod_layout_component *lod_comp;
7968 if (unlikely(lo->ldo_is_foreign)) {
7969 lod_free_foreign_lov(lo);
7970 lo->ldo_comp_cached = 0;
7971 } else if (unlikely(lo->ldo_dir_is_foreign)) {
7972 lod_free_foreign_lmv(lo);
7973 lo->ldo_dir_stripe_loaded = 0;
7974 } else if (lo->ldo_stripe != NULL) {
7975 LASSERT(lo->ldo_comp_entries == NULL);
7976 LASSERT(lo->ldo_dir_stripes_allocated > 0);
7978 for (i = 0; i < lo->ldo_dir_stripe_count; i++) {
7979 if (lo->ldo_stripe[i])
7980 dt_object_put(env, lo->ldo_stripe[i]);
7983 j = sizeof(struct dt_object *) * lo->ldo_dir_stripes_allocated;
7984 OBD_FREE(lo->ldo_stripe, j);
7985 lo->ldo_stripe = NULL;
7986 lo->ldo_dir_stripes_allocated = 0;
7987 lo->ldo_dir_stripe_loaded = 0;
7988 lo->ldo_dir_stripe_count = 0;
7989 } else if (lo->ldo_comp_entries != NULL) {
7990 for (i = 0; i < lo->ldo_comp_cnt; i++) {
7991 /* free lod_layout_component::llc_stripe array */
7992 lod_comp = &lo->ldo_comp_entries[i];
7994 if (lod_comp->llc_stripe == NULL)
7996 LASSERT(lod_comp->llc_stripes_allocated != 0);
7997 for (j = 0; j < lod_comp->llc_stripes_allocated; j++) {
7998 if (lod_comp->llc_stripe[j] != NULL)
8000 &lod_comp->llc_stripe[j]->do_lu);
8002 OBD_FREE(lod_comp->llc_stripe,
8003 sizeof(struct dt_object *) *
8004 lod_comp->llc_stripes_allocated);
8005 lod_comp->llc_stripe = NULL;
8006 OBD_FREE(lod_comp->llc_ost_indices,
8008 lod_comp->llc_stripes_allocated);
8009 lod_comp->llc_ost_indices = NULL;
8010 lod_comp->llc_stripes_allocated = 0;
8012 lod_free_comp_entries(lo);
8013 lo->ldo_comp_cached = 0;
8017 void lod_striping_free(const struct lu_env *env, struct lod_object *lo)
8019 mutex_lock(&lo->ldo_layout_mutex);
8020 lod_striping_free_nolock(env, lo);
8021 mutex_unlock(&lo->ldo_layout_mutex);
8025 * Implementation of lu_object_operations::loo_object_free.
8027 * \see lu_object_operations::loo_object_free() in the API description
8030 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
8032 struct lod_object *lo = lu2lod_obj(o);
8034 /* release all underlying object pinned */
8035 lod_striping_free(env, lo);
8037 OBD_SLAB_FREE_PTR(lo, lod_object_kmem);
8041 * Implementation of lu_object_operations::loo_object_release.
8043 * \see lu_object_operations::loo_object_release() in the API description
8046 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
8048 /* XXX: shouldn't we release everything here in case if object
8049 * creation failed before? */
8053 * Implementation of lu_object_operations::loo_object_print.
8055 * \see lu_object_operations::loo_object_print() in the API description
8058 static int lod_object_print(const struct lu_env *env, void *cookie,
8059 lu_printer_t p, const struct lu_object *l)
8061 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
8063 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
8066 struct lu_object_operations lod_lu_obj_ops = {
8067 .loo_object_init = lod_object_init,
8068 .loo_object_free = lod_object_free,
8069 .loo_object_release = lod_object_release,
8070 .loo_object_print = lod_object_print,