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, 2014, 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 * lustre/doc/osd-api.txt.
38 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
41 #define DEBUG_SUBSYSTEM S_MDS
44 #include <obd_class.h>
45 #include <lustre_ver.h>
46 #include <obd_support.h>
47 #include <lprocfs_status.h>
49 #include <lustre_fid.h>
50 #include <lustre_param.h>
51 #include <lustre_fid.h>
52 #include <lustre_lmv.h>
53 #include <md_object.h>
54 #include <lustre_linkea.h>
56 #include "lod_internal.h"
58 static const char dot[] = ".";
59 static const char dotdot[] = "..";
61 static const struct dt_body_operations lod_body_lnk_ops;
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_index_lookup(const struct lu_env *env, struct dt_object *dt,
71 struct dt_rec *rec, const struct dt_key *key,
72 struct lustre_capa *capa)
74 struct dt_object *next = dt_object_child(dt);
75 return next->do_index_ops->dio_lookup(env, next, rec, key, capa);
79 * Implementation of dt_index_operations::dio_declare_insert.
81 * Used with regular (non-striped) objects.
83 * \see dt_index_operations::dio_declare_insert() in the API description
86 static int lod_declare_index_insert(const struct lu_env *env,
88 const struct dt_rec *rec,
89 const struct dt_key *key,
90 struct thandle *handle)
92 return dt_declare_insert(env, dt_object_child(dt), rec, key, handle);
96 * Implementation of dt_index_operations::dio_insert.
98 * Used with regular (non-striped) objects
100 * \see dt_index_operations::dio_insert() in the API description for details.
102 static int lod_index_insert(const struct lu_env *env,
103 struct dt_object *dt,
104 const struct dt_rec *rec,
105 const struct dt_key *key,
107 struct lustre_capa *capa,
110 return dt_insert(env, dt_object_child(dt), rec, key, th, capa, ign);
114 * Implementation of dt_index_operations::dio_declare_delete.
116 * Used with regular (non-striped) objects.
118 * \see dt_index_operations::dio_declare_delete() in the API description
121 static int lod_declare_index_delete(const struct lu_env *env,
122 struct dt_object *dt,
123 const struct dt_key *key,
126 return dt_declare_delete(env, dt_object_child(dt), key, th);
130 * Implementation of dt_index_operations::dio_delete.
132 * Used with regular (non-striped) objects.
134 * \see dt_index_operations::dio_delete() in the API description for details.
136 static int lod_index_delete(const struct lu_env *env,
137 struct dt_object *dt,
138 const struct dt_key *key,
140 struct lustre_capa *capa)
142 return dt_delete(env, dt_object_child(dt), key, th, capa);
146 * Implementation of dt_it_ops::init.
148 * Used with regular (non-striped) objects.
150 * \see dt_it_ops::init() in the API description for details.
152 static struct dt_it *lod_it_init(const struct lu_env *env,
153 struct dt_object *dt, __u32 attr,
154 struct lustre_capa *capa)
156 struct dt_object *next = dt_object_child(dt);
157 struct lod_it *it = &lod_env_info(env)->lti_it;
158 struct dt_it *it_next;
161 it_next = next->do_index_ops->dio_it.init(env, next, attr, capa);
165 /* currently we do not use more than one iterator per thread
166 * so we store it in thread info. if at some point we need
167 * more active iterators in a single thread, we can allocate
169 LASSERT(it->lit_obj == NULL);
171 it->lit_it = it_next;
174 return (struct dt_it *)it;
177 #define LOD_CHECK_IT(env, it) \
179 LASSERT((it)->lit_obj != NULL); \
180 LASSERT((it)->lit_it != NULL); \
184 * Implementation of dt_index_operations::dio_it.fini.
186 * Used with regular (non-striped) objects.
188 * \see dt_index_operations::dio_it.fini() in the API description for details.
190 static void lod_it_fini(const struct lu_env *env, struct dt_it *di)
192 struct lod_it *it = (struct lod_it *)di;
194 LOD_CHECK_IT(env, it);
195 it->lit_obj->do_index_ops->dio_it.fini(env, it->lit_it);
197 /* the iterator not in use any more */
203 * Implementation of dt_it_ops::get.
205 * Used with regular (non-striped) objects.
207 * \see dt_it_ops::get() in the API description for details.
209 static int lod_it_get(const struct lu_env *env, struct dt_it *di,
210 const struct dt_key *key)
212 const struct lod_it *it = (const struct lod_it *)di;
214 LOD_CHECK_IT(env, it);
215 return it->lit_obj->do_index_ops->dio_it.get(env, it->lit_it, key);
219 * Implementation of dt_it_ops::put.
221 * Used with regular (non-striped) objects.
223 * \see dt_it_ops::put() in the API description for details.
225 static void lod_it_put(const struct lu_env *env, struct dt_it *di)
227 struct lod_it *it = (struct lod_it *)di;
229 LOD_CHECK_IT(env, it);
230 return it->lit_obj->do_index_ops->dio_it.put(env, it->lit_it);
234 * Implementation of dt_it_ops::next.
236 * Used with regular (non-striped) objects
238 * \see dt_it_ops::next() in the API description for details.
240 static int lod_it_next(const struct lu_env *env, struct dt_it *di)
242 struct lod_it *it = (struct lod_it *)di;
244 LOD_CHECK_IT(env, it);
245 return it->lit_obj->do_index_ops->dio_it.next(env, it->lit_it);
249 * Implementation of dt_it_ops::key.
251 * Used with regular (non-striped) objects.
253 * \see dt_it_ops::key() in the API description for details.
255 static struct dt_key *lod_it_key(const struct lu_env *env,
256 const struct dt_it *di)
258 const struct lod_it *it = (const struct lod_it *)di;
260 LOD_CHECK_IT(env, it);
261 return it->lit_obj->do_index_ops->dio_it.key(env, it->lit_it);
265 * Implementation of dt_it_ops::key_size.
267 * Used with regular (non-striped) objects.
269 * \see dt_it_ops::key_size() in the API description for details.
271 static int lod_it_key_size(const struct lu_env *env, const struct dt_it *di)
273 struct lod_it *it = (struct lod_it *)di;
275 LOD_CHECK_IT(env, it);
276 return it->lit_obj->do_index_ops->dio_it.key_size(env, it->lit_it);
280 * Implementation of dt_it_ops::rec.
282 * Used with regular (non-striped) objects.
284 * \see dt_it_ops::rec() in the API description for details.
286 static int lod_it_rec(const struct lu_env *env, const struct dt_it *di,
287 struct dt_rec *rec, __u32 attr)
289 const struct lod_it *it = (const struct lod_it *)di;
291 LOD_CHECK_IT(env, it);
292 return it->lit_obj->do_index_ops->dio_it.rec(env, it->lit_it, rec,
297 * Implementation of dt_it_ops::rec_size.
299 * Used with regular (non-striped) objects.
301 * \see dt_it_ops::rec_size() in the API description for details.
303 static int lod_it_rec_size(const struct lu_env *env, const struct dt_it *di,
306 const struct lod_it *it = (const struct lod_it *)di;
308 LOD_CHECK_IT(env, it);
309 return it->lit_obj->do_index_ops->dio_it.rec_size(env, it->lit_it,
314 * Implementation of dt_it_ops::store.
316 * Used with regular (non-striped) objects.
318 * \see dt_it_ops::store() in the API description for details.
320 static __u64 lod_it_store(const struct lu_env *env, const struct dt_it *di)
322 const struct lod_it *it = (const struct lod_it *)di;
324 LOD_CHECK_IT(env, it);
325 return it->lit_obj->do_index_ops->dio_it.store(env, it->lit_it);
329 * Implementation of dt_it_ops::load.
331 * Used with regular (non-striped) objects.
333 * \see dt_it_ops::load() in the API description for details.
335 static int lod_it_load(const struct lu_env *env, const struct dt_it *di,
338 const struct lod_it *it = (const struct lod_it *)di;
340 LOD_CHECK_IT(env, it);
341 return it->lit_obj->do_index_ops->dio_it.load(env, it->lit_it, hash);
345 * Implementation of dt_it_ops::key_rec.
347 * Used with regular (non-striped) objects.
349 * \see dt_it_ops::rec() in the API description for details.
351 static int lod_it_key_rec(const struct lu_env *env, const struct dt_it *di,
354 const struct lod_it *it = (const struct lod_it *)di;
356 LOD_CHECK_IT(env, it);
357 return it->lit_obj->do_index_ops->dio_it.key_rec(env, it->lit_it,
361 static struct dt_index_operations lod_index_ops = {
362 .dio_lookup = lod_index_lookup,
363 .dio_declare_insert = lod_declare_index_insert,
364 .dio_insert = lod_index_insert,
365 .dio_declare_delete = lod_declare_index_delete,
366 .dio_delete = lod_index_delete,
374 .key_size = lod_it_key_size,
376 .rec_size = lod_it_rec_size,
377 .store = lod_it_store,
379 .key_rec = lod_it_key_rec,
384 * Implementation of dt_it_ops::init.
386 * Used with striped objects. Internally just initializes the iterator
387 * on the first stripe.
389 * \see dt_it_ops::init() in the API description for details.
391 static struct dt_it *lod_striped_it_init(const struct lu_env *env,
392 struct dt_object *dt, __u32 attr,
393 struct lustre_capa *capa)
395 struct lod_object *lo = lod_dt_obj(dt);
396 struct dt_object *next;
397 struct lod_it *it = &lod_env_info(env)->lti_it;
398 struct dt_it *it_next;
401 LASSERT(lo->ldo_stripenr > 0);
402 next = lo->ldo_stripe[0];
403 LASSERT(next != NULL);
404 LASSERT(next->do_index_ops != NULL);
406 it_next = next->do_index_ops->dio_it.init(env, next, attr, capa);
410 /* currently we do not use more than one iterator per thread
411 * so we store it in thread info. if at some point we need
412 * more active iterators in a single thread, we can allocate
414 LASSERT(it->lit_obj == NULL);
416 it->lit_stripe_index = 0;
418 it->lit_it = it_next;
421 return (struct dt_it *)it;
424 #define LOD_CHECK_STRIPED_IT(env, it, lo) \
426 LASSERT((it)->lit_obj != NULL); \
427 LASSERT((it)->lit_it != NULL); \
428 LASSERT((lo)->ldo_stripenr > 0); \
429 LASSERT((it)->lit_stripe_index < (lo)->ldo_stripenr); \
433 * Implementation of dt_it_ops::fini.
435 * Used with striped objects.
437 * \see dt_it_ops::fini() in the API description for details.
439 static void lod_striped_it_fini(const struct lu_env *env, struct dt_it *di)
441 struct lod_it *it = (struct lod_it *)di;
442 struct lod_object *lo = lod_dt_obj(it->lit_obj);
443 struct dt_object *next;
445 LOD_CHECK_STRIPED_IT(env, it, lo);
447 next = lo->ldo_stripe[it->lit_stripe_index];
448 LASSERT(next != NULL);
449 LASSERT(next->do_index_ops != NULL);
451 next->do_index_ops->dio_it.fini(env, it->lit_it);
453 /* the iterator not in use any more */
456 it->lit_stripe_index = 0;
460 * Implementation of dt_it_ops::get.
462 * Right now it's not used widely, only to reset the iterator to the
463 * initial position. It should be possible to implement a full version
464 * which chooses a correct stripe to be able to position with any key.
466 * \see dt_it_ops::get() in the API description for details.
468 static int lod_striped_it_get(const struct lu_env *env, struct dt_it *di,
469 const struct dt_key *key)
471 const struct lod_it *it = (const struct lod_it *)di;
472 struct lod_object *lo = lod_dt_obj(it->lit_obj);
473 struct dt_object *next;
476 LOD_CHECK_STRIPED_IT(env, it, lo);
478 next = lo->ldo_stripe[it->lit_stripe_index];
479 LASSERT(next != NULL);
480 LASSERT(next->do_index_ops != NULL);
482 return next->do_index_ops->dio_it.get(env, it->lit_it, key);
486 * Implementation of dt_it_ops::put.
488 * Used with striped objects.
490 * \see dt_it_ops::put() in the API description for details.
492 static void lod_striped_it_put(const struct lu_env *env, struct dt_it *di)
494 struct lod_it *it = (struct lod_it *)di;
495 struct lod_object *lo = lod_dt_obj(it->lit_obj);
496 struct dt_object *next;
498 LOD_CHECK_STRIPED_IT(env, it, lo);
500 next = lo->ldo_stripe[it->lit_stripe_index];
501 LASSERT(next != NULL);
502 LASSERT(next->do_index_ops != NULL);
504 return next->do_index_ops->dio_it.put(env, it->lit_it);
508 * Implementation of dt_it_ops::next.
510 * Used with striped objects. When the end of the current stripe is
511 * reached, the method takes the next stripe's iterator.
513 * \see dt_it_ops::next() in the API description for details.
515 static int lod_striped_it_next(const struct lu_env *env, struct dt_it *di)
517 struct lod_it *it = (struct lod_it *)di;
518 struct lod_object *lo = lod_dt_obj(it->lit_obj);
519 struct dt_object *next;
520 struct dt_it *it_next;
524 LOD_CHECK_STRIPED_IT(env, it, lo);
526 next = lo->ldo_stripe[it->lit_stripe_index];
527 LASSERT(next != NULL);
528 LASSERT(next->do_index_ops != NULL);
530 rc = next->do_index_ops->dio_it.next(env, it->lit_it);
534 if (rc == 0 && it->lit_stripe_index == 0)
537 if (rc == 0 && it->lit_stripe_index > 0) {
538 struct lu_dirent *ent;
540 ent = (struct lu_dirent *)lod_env_info(env)->lti_key;
542 rc = next->do_index_ops->dio_it.rec(env, it->lit_it,
543 (struct dt_rec *)ent,
548 /* skip . and .. for slave stripe */
549 if ((strncmp(ent->lde_name, ".",
550 le16_to_cpu(ent->lde_namelen)) == 0 &&
551 le16_to_cpu(ent->lde_namelen) == 1) ||
552 (strncmp(ent->lde_name, "..",
553 le16_to_cpu(ent->lde_namelen)) == 0 &&
554 le16_to_cpu(ent->lde_namelen) == 2))
560 /* go to next stripe */
561 if (it->lit_stripe_index + 1 >= lo->ldo_stripenr)
564 it->lit_stripe_index++;
566 next->do_index_ops->dio_it.put(env, it->lit_it);
567 next->do_index_ops->dio_it.fini(env, it->lit_it);
569 rc = next->do_ops->do_index_try(env, next, &dt_directory_features);
573 next = lo->ldo_stripe[it->lit_stripe_index];
574 LASSERT(next != NULL);
575 LASSERT(next->do_index_ops != NULL);
577 it_next = next->do_index_ops->dio_it.init(env, next, it->lit_attr,
579 if (!IS_ERR(it_next)) {
580 it->lit_it = it_next;
583 rc = PTR_ERR(it_next);
590 * Implementation of dt_it_ops::key.
592 * Used with striped objects.
594 * \see dt_it_ops::key() in the API description for details.
596 static struct dt_key *lod_striped_it_key(const struct lu_env *env,
597 const struct dt_it *di)
599 const struct lod_it *it = (const struct lod_it *)di;
600 struct lod_object *lo = lod_dt_obj(it->lit_obj);
601 struct dt_object *next;
603 LOD_CHECK_STRIPED_IT(env, it, lo);
605 next = lo->ldo_stripe[it->lit_stripe_index];
606 LASSERT(next != NULL);
607 LASSERT(next->do_index_ops != NULL);
609 return next->do_index_ops->dio_it.key(env, it->lit_it);
613 * Implementation of dt_it_ops::key_size.
615 * Used with striped objects.
617 * \see dt_it_ops::size() in the API description for details.
619 static int lod_striped_it_key_size(const struct lu_env *env,
620 const struct dt_it *di)
622 struct lod_it *it = (struct lod_it *)di;
623 struct lod_object *lo = lod_dt_obj(it->lit_obj);
624 struct dt_object *next;
626 LOD_CHECK_STRIPED_IT(env, it, lo);
628 next = lo->ldo_stripe[it->lit_stripe_index];
629 LASSERT(next != NULL);
630 LASSERT(next->do_index_ops != NULL);
632 return next->do_index_ops->dio_it.key_size(env, it->lit_it);
636 * Implementation of dt_it_ops::rec.
638 * Used with striped objects.
640 * \see dt_it_ops::rec() in the API description for details.
642 static int lod_striped_it_rec(const struct lu_env *env, const struct dt_it *di,
643 struct dt_rec *rec, __u32 attr)
645 const struct lod_it *it = (const struct lod_it *)di;
646 struct lod_object *lo = lod_dt_obj(it->lit_obj);
647 struct dt_object *next;
649 LOD_CHECK_STRIPED_IT(env, it, lo);
651 next = lo->ldo_stripe[it->lit_stripe_index];
652 LASSERT(next != NULL);
653 LASSERT(next->do_index_ops != NULL);
655 return next->do_index_ops->dio_it.rec(env, it->lit_it, rec, attr);
659 * Implementation of dt_it_ops::rec_size.
661 * Used with striped objects.
663 * \see dt_it_ops::rec_size() in the API description for details.
665 static int lod_striped_it_rec_size(const struct lu_env *env,
666 const struct dt_it *di, __u32 attr)
668 struct lod_it *it = (struct lod_it *)di;
669 struct lod_object *lo = lod_dt_obj(it->lit_obj);
670 struct dt_object *next;
672 LOD_CHECK_STRIPED_IT(env, it, lo);
674 next = lo->ldo_stripe[it->lit_stripe_index];
675 LASSERT(next != NULL);
676 LASSERT(next->do_index_ops != NULL);
678 return next->do_index_ops->dio_it.rec_size(env, it->lit_it, attr);
682 * Implementation of dt_it_ops::store.
684 * Used with striped objects.
686 * \see dt_it_ops::store() in the API description for details.
688 static __u64 lod_striped_it_store(const struct lu_env *env,
689 const struct dt_it *di)
691 const struct lod_it *it = (const struct lod_it *)di;
692 struct lod_object *lo = lod_dt_obj(it->lit_obj);
693 struct dt_object *next;
695 LOD_CHECK_STRIPED_IT(env, it, lo);
697 next = lo->ldo_stripe[it->lit_stripe_index];
698 LASSERT(next != NULL);
699 LASSERT(next->do_index_ops != NULL);
701 return next->do_index_ops->dio_it.store(env, it->lit_it);
705 * Implementation of dt_it_ops::load.
707 * Used with striped objects.
709 * \see dt_it_ops::load() in the API description for details.
711 static int lod_striped_it_load(const struct lu_env *env,
712 const struct dt_it *di, __u64 hash)
714 const struct lod_it *it = (const struct lod_it *)di;
715 struct lod_object *lo = lod_dt_obj(it->lit_obj);
716 struct dt_object *next;
718 LOD_CHECK_STRIPED_IT(env, it, lo);
720 next = lo->ldo_stripe[it->lit_stripe_index];
721 LASSERT(next != NULL);
722 LASSERT(next->do_index_ops != NULL);
724 return next->do_index_ops->dio_it.load(env, it->lit_it, hash);
727 static struct dt_index_operations lod_striped_index_ops = {
728 .dio_lookup = lod_index_lookup,
729 .dio_declare_insert = lod_declare_index_insert,
730 .dio_insert = lod_index_insert,
731 .dio_declare_delete = lod_declare_index_delete,
732 .dio_delete = lod_index_delete,
734 .init = lod_striped_it_init,
735 .fini = lod_striped_it_fini,
736 .get = lod_striped_it_get,
737 .put = lod_striped_it_put,
738 .next = lod_striped_it_next,
739 .key = lod_striped_it_key,
740 .key_size = lod_striped_it_key_size,
741 .rec = lod_striped_it_rec,
742 .rec_size = lod_striped_it_rec_size,
743 .store = lod_striped_it_store,
744 .load = lod_striped_it_load,
749 * Append the FID for each shard of the striped directory after the
750 * given LMV EA header.
752 * To simplify striped directory and the consistency verification,
753 * we only store the LMV EA header on disk, for both master object
754 * and slave objects. When someone wants to know the whole LMV EA,
755 * such as client readdir(), we can build the entrie LMV EA on the
756 * MDT side (in RAM) via iterating the sub-directory entries that
757 * are contained in the master object of the stripe directory.
759 * For the master object of the striped directroy, the valid name
760 * for each shard is composed of the ${shard_FID}:${shard_idx}.
762 * There may be holes in the LMV EA if some shards' name entries
763 * are corrupted or lost.
765 * \param[in] env pointer to the thread context
766 * \param[in] lo pointer to the master object of the striped directory
767 * \param[in] buf pointer to the lu_buf which will hold the LMV EA
768 * \param[in] resize whether re-allocate the buffer if it is not big enough
770 * \retval positive size of the LMV EA
771 * \retval 0 for nothing to be loaded
772 * \retval negative error number on failure
774 int lod_load_lmv_shards(const struct lu_env *env, struct lod_object *lo,
775 struct lu_buf *buf, bool resize)
777 struct lu_dirent *ent =
778 (struct lu_dirent *)lod_env_info(env)->lti_key;
779 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
780 struct dt_object *obj = dt_object_child(&lo->ldo_obj);
781 struct lmv_mds_md_v1 *lmv1 = buf->lb_buf;
783 const struct dt_it_ops *iops;
785 __u32 magic = le32_to_cpu(lmv1->lmv_magic);
790 /* If it is not a striped directory, then load nothing. */
791 if (magic != LMV_MAGIC_V1)
794 /* If it is in migration (or failure), then load nothing. */
795 if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
798 stripes = le32_to_cpu(lmv1->lmv_stripe_count);
802 rc = lmv_mds_md_size(stripes, magic);
806 if (buf->lb_len < lmv1_size) {
815 lu_buf_alloc(buf, lmv1_size);
820 memcpy(buf->lb_buf, tbuf.lb_buf, tbuf.lb_len);
823 if (unlikely(!dt_try_as_dir(env, obj)))
826 memset(&lmv1->lmv_stripe_fids[0], 0, stripes * sizeof(struct lu_fid));
827 iops = &obj->do_index_ops->dio_it;
828 it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA);
832 rc = iops->load(env, it, 0);
834 rc = iops->next(env, it);
839 char name[FID_LEN + 2] = "";
844 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
850 fid_le_to_cpu(&fid, &ent->lde_fid);
851 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
852 if (ent->lde_name[0] == '.') {
853 if (ent->lde_namelen == 1)
856 if (ent->lde_namelen == 2 && ent->lde_name[1] == '.')
860 len = snprintf(name, FID_LEN + 1, DFID":", PFID(&ent->lde_fid));
861 /* The ent->lde_name is composed of ${FID}:${index} */
862 if (ent->lde_namelen < len + 1 ||
863 memcmp(ent->lde_name, name, len) != 0) {
864 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
865 "%s: invalid shard name %.*s with the FID "DFID
866 " for the striped directory "DFID", %s\n",
867 lod2obd(lod)->obd_name, ent->lde_namelen,
868 ent->lde_name, PFID(&fid),
869 PFID(lu_object_fid(&obj->do_lu)),
870 lod->lod_lmv_failout ? "failout" : "skip");
872 if (lod->lod_lmv_failout)
880 if (ent->lde_name[len] < '0' ||
881 ent->lde_name[len] > '9') {
882 CDEBUG(lod->lod_lmv_failout ? D_ERROR : D_INFO,
883 "%s: invalid shard name %.*s with the "
884 "FID "DFID" for the striped directory "
886 lod2obd(lod)->obd_name, ent->lde_namelen,
887 ent->lde_name, PFID(&fid),
888 PFID(lu_object_fid(&obj->do_lu)),
889 lod->lod_lmv_failout ?
892 if (lod->lod_lmv_failout)
898 index = index * 10 + ent->lde_name[len++] - '0';
899 } while (len < ent->lde_namelen);
901 if (len == ent->lde_namelen) {
902 /* Out of LMV EA range. */
903 if (index >= stripes) {
904 CERROR("%s: the shard %.*s for the striped "
905 "directory "DFID" is out of the known "
906 "LMV EA range [0 - %u], failout\n",
907 lod2obd(lod)->obd_name, ent->lde_namelen,
909 PFID(lu_object_fid(&obj->do_lu)),
915 /* The slot has been occupied. */
916 if (!fid_is_zero(&lmv1->lmv_stripe_fids[index])) {
920 &lmv1->lmv_stripe_fids[index]);
921 CERROR("%s: both the shard "DFID" and "DFID
922 " for the striped directory "DFID
923 " claim the same LMV EA slot at the "
924 "index %d, failout\n",
925 lod2obd(lod)->obd_name,
926 PFID(&fid0), PFID(&fid),
927 PFID(lu_object_fid(&obj->do_lu)), index);
932 /* stored as LE mode */
933 lmv1->lmv_stripe_fids[index] = ent->lde_fid;
936 rc = iops->next(env, it);
943 RETURN(rc > 0 ? lmv_mds_md_size(stripes, magic) : rc);
947 * Implementation of dt_object_operations::do_index_try.
949 * \see dt_object_operations::do_index_try() in the API description for details.
951 static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
952 const struct dt_index_features *feat)
954 struct lod_object *lo = lod_dt_obj(dt);
955 struct dt_object *next = dt_object_child(dt);
959 LASSERT(next->do_ops);
960 LASSERT(next->do_ops->do_index_try);
962 rc = lod_load_striping_locked(env, lo);
966 rc = next->do_ops->do_index_try(env, next, feat);
970 if (lo->ldo_stripenr > 0) {
973 for (i = 0; i < lo->ldo_stripenr; i++) {
974 if (dt_object_exists(lo->ldo_stripe[i]) == 0)
976 rc = lo->ldo_stripe[i]->do_ops->do_index_try(env,
977 lo->ldo_stripe[i], feat);
981 dt->do_index_ops = &lod_striped_index_ops;
983 dt->do_index_ops = &lod_index_ops;
990 * Implementation of dt_object_operations::do_read_lock.
992 * \see dt_object_operations::do_read_lock() in the API description for details.
994 static void lod_object_read_lock(const struct lu_env *env,
995 struct dt_object *dt, unsigned role)
997 dt_read_lock(env, dt_object_child(dt), role);
1001 * Implementation of dt_object_operations::do_write_lock.
1003 * \see dt_object_operations::do_write_lock() in the API description for
1006 static void lod_object_write_lock(const struct lu_env *env,
1007 struct dt_object *dt, unsigned role)
1009 dt_write_lock(env, dt_object_child(dt), role);
1013 * Implementation of dt_object_operations::do_read_unlock.
1015 * \see dt_object_operations::do_read_unlock() in the API description for
1018 static void lod_object_read_unlock(const struct lu_env *env,
1019 struct dt_object *dt)
1021 dt_read_unlock(env, dt_object_child(dt));
1025 * Implementation of dt_object_operations::do_write_unlock.
1027 * \see dt_object_operations::do_write_unlock() in the API description for
1030 static void lod_object_write_unlock(const struct lu_env *env,
1031 struct dt_object *dt)
1033 dt_write_unlock(env, dt_object_child(dt));
1037 * Implementation of dt_object_operations::do_write_locked.
1039 * \see dt_object_operations::do_write_locked() in the API description for
1042 static int lod_object_write_locked(const struct lu_env *env,
1043 struct dt_object *dt)
1045 return dt_write_locked(env, dt_object_child(dt));
1049 * Implementation of dt_object_operations::do_attr_get.
1051 * \see dt_object_operations::do_attr_get() in the API description for details.
1053 static int lod_attr_get(const struct lu_env *env,
1054 struct dt_object *dt,
1055 struct lu_attr *attr,
1056 struct lustre_capa *capa)
1058 /* Note: for striped directory, client will merge attributes
1059 * from all of the sub-stripes see lmv_merge_attr(), and there
1060 * no MDD logic depend on directory nlink/size/time, so we can
1061 * always use master inode nlink and size for now. */
1062 return dt_attr_get(env, dt_object_child(dt), attr, capa);
1066 * Mark all of the striped directory sub-stripes dead.
1068 * When a striped object is a subject to removal, we have
1069 * to mark all the stripes to prevent further access to
1070 * them (e.g. create a new file in those). So we mark
1071 * all the stripes with LMV_HASH_FLAG_DEAD. The function
1072 * can be used to declare the changes and to apply them.
1073 * If the object isn't striped, then just return success.
1075 * \param[in] env execution environment
1076 * \param[in] dt the striped object
1077 * \param[in] handle transaction handle
1078 * \param[in] declare whether to declare the change or apply
1080 * \retval 0 on success
1081 * \retval negative if failed
1083 static int lod_mark_dead_object(const struct lu_env *env,
1084 struct dt_object *dt,
1085 struct thandle *handle,
1088 struct lod_object *lo = lod_dt_obj(dt);
1089 struct lmv_mds_md_v1 *lmv;
1090 __u32 dead_hash_type;
1096 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
1099 rc = lod_load_striping_locked(env, lo);
1103 if (lo->ldo_stripenr == 0)
1106 rc = lod_get_lmv_ea(env, lo);
1110 lmv = lod_env_info(env)->lti_ea_store;
1111 lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1112 dead_hash_type = le32_to_cpu(lmv->lmv_hash_type) | LMV_HASH_FLAG_DEAD;
1113 lmv->lmv_hash_type = cpu_to_le32(dead_hash_type);
1114 for (i = 0; i < lo->ldo_stripenr; i++) {
1117 lmv->lmv_master_mdt_index = i;
1119 buf.lb_len = sizeof(*lmv);
1121 rc = dt_declare_xattr_set(env, lo->ldo_stripe[i], &buf,
1123 LU_XATTR_REPLACE, handle);
1125 rc = dt_xattr_set(env, lo->ldo_stripe[i], &buf,
1126 XATTR_NAME_LMV, LU_XATTR_REPLACE,
1127 handle, BYPASS_CAPA);
1137 * Implementation of dt_object_operations::do_declare_attr_set.
1139 * If the object is striped, then apply the changes to all the stripes.
1141 * \see dt_object_operations::do_declare_attr_set() in the API description
1144 static int lod_declare_attr_set(const struct lu_env *env,
1145 struct dt_object *dt,
1146 const struct lu_attr *attr,
1147 struct thandle *handle)
1149 struct dt_object *next = dt_object_child(dt);
1150 struct lod_object *lo = lod_dt_obj(dt);
1154 /* Set dead object on all other stripes */
1155 if (attr->la_valid & LA_FLAGS && !(attr->la_valid & ~LA_FLAGS) &&
1156 attr->la_flags & LUSTRE_SLAVE_DEAD_FL) {
1157 rc = lod_mark_dead_object(env, dt, handle, true);
1162 * declare setattr on the local object
1164 rc = dt_declare_attr_set(env, next, attr, handle);
1168 /* osp_declare_attr_set() ignores all attributes other than
1169 * UID, GID, and size, and osp_attr_set() ignores all but UID
1170 * and GID. Declaration of size attr setting happens through
1171 * lod_declare_init_size(), and not through this function.
1172 * Therefore we need not load striping unless ownership is
1173 * changing. This should save memory and (we hope) speed up
1175 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1176 if (!(attr->la_valid & (LA_UID | LA_GID)))
1179 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1182 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE |
1183 LA_ATIME | LA_MTIME | LA_CTIME)))
1187 * load striping information, notice we don't do this when object
1188 * is being initialized as we don't need this information till
1189 * few specific cases like destroy, chown
1191 rc = lod_load_striping(env, lo);
1195 if (lo->ldo_stripenr == 0)
1199 * if object is striped declare changes on the stripes
1201 LASSERT(lo->ldo_stripe);
1202 for (i = 0; i < lo->ldo_stripenr; i++) {
1203 if (likely(lo->ldo_stripe[i] != NULL)) {
1204 rc = dt_declare_attr_set(env, lo->ldo_stripe[i], attr,
1207 CERROR("failed declaration: %d\n", rc);
1213 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE) &&
1214 dt_object_exists(next) != 0 &&
1215 dt_object_remote(next) == 0)
1216 dt_declare_xattr_del(env, next, XATTR_NAME_LOV, handle);
1218 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) &&
1219 dt_object_exists(next) &&
1220 dt_object_remote(next) == 0 && S_ISREG(attr->la_mode)) {
1221 struct lod_thread_info *info = lod_env_info(env);
1222 struct lu_buf *buf = &info->lti_buf;
1224 buf->lb_buf = info->lti_ea_store;
1225 buf->lb_len = info->lti_ea_store_size;
1226 dt_declare_xattr_set(env, next, buf, XATTR_NAME_LOV,
1227 LU_XATTR_REPLACE, handle);
1234 * Implementation of dt_object_operations::do_attr_set.
1236 * If the object is striped, then apply the changes to all or subset of
1237 * the stripes depending on the object type and specific attributes.
1239 * \see dt_object_operations::do_attr_set() in the API description for details.
1241 static int lod_attr_set(const struct lu_env *env,
1242 struct dt_object *dt,
1243 const struct lu_attr *attr,
1244 struct thandle *handle,
1245 struct lustre_capa *capa)
1247 struct dt_object *next = dt_object_child(dt);
1248 struct lod_object *lo = lod_dt_obj(dt);
1252 /* Set dead object on all other stripes */
1253 if (attr->la_valid & LA_FLAGS && !(attr->la_valid & ~LA_FLAGS) &&
1254 attr->la_flags & LUSTRE_SLAVE_DEAD_FL) {
1255 rc = lod_mark_dead_object(env, dt, handle, false);
1260 * apply changes to the local object
1262 rc = dt_attr_set(env, next, attr, handle, capa);
1266 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
1267 if (!(attr->la_valid & (LA_UID | LA_GID)))
1270 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
1273 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE |
1274 LA_ATIME | LA_MTIME | LA_CTIME)))
1278 if (lo->ldo_stripenr == 0)
1282 * if object is striped, apply changes to all the stripes
1284 LASSERT(lo->ldo_stripe);
1285 for (i = 0; i < lo->ldo_stripenr; i++) {
1286 if (unlikely(lo->ldo_stripe[i] == NULL))
1288 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
1289 (dt_object_exists(lo->ldo_stripe[i]) == 0))
1292 rc = dt_attr_set(env, lo->ldo_stripe[i], attr, handle, capa);
1294 CERROR("failed declaration: %d\n", rc);
1299 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_STRIPE) &&
1300 dt_object_exists(next) != 0 &&
1301 dt_object_remote(next) == 0)
1302 dt_xattr_del(env, next, XATTR_NAME_LOV, handle, BYPASS_CAPA);
1304 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CHANGE_STRIPE) &&
1305 dt_object_exists(next) &&
1306 dt_object_remote(next) == 0 && S_ISREG(attr->la_mode)) {
1307 struct lod_thread_info *info = lod_env_info(env);
1308 struct lu_buf *buf = &info->lti_buf;
1309 struct ost_id *oi = &info->lti_ostid;
1310 struct lu_fid *fid = &info->lti_fid;
1311 struct lov_mds_md_v1 *lmm;
1312 struct lov_ost_data_v1 *objs;
1316 rc1 = lod_get_lov_ea(env, lo);
1320 buf->lb_buf = info->lti_ea_store;
1321 buf->lb_len = info->lti_ea_store_size;
1322 lmm = info->lti_ea_store;
1323 magic = le32_to_cpu(lmm->lmm_magic);
1324 if (magic == LOV_MAGIC_V1)
1325 objs = &(lmm->lmm_objects[0]);
1327 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1328 ostid_le_to_cpu(&objs->l_ost_oi, oi);
1329 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
1331 fid_to_ostid(fid, oi);
1332 ostid_cpu_to_le(oi, &objs->l_ost_oi);
1333 dt_xattr_set(env, next, buf, XATTR_NAME_LOV,
1334 LU_XATTR_REPLACE, handle, BYPASS_CAPA);
1341 * Implementation of dt_object_operations::do_xattr_get.
1343 * If LOV EA is requested from the root object and it's not
1344 * found, then return default striping for the filesystem.
1346 * \see dt_object_operations::do_xattr_get() in the API description for details.
1348 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
1349 struct lu_buf *buf, const char *name,
1350 struct lustre_capa *capa)
1352 struct lod_thread_info *info = lod_env_info(env);
1353 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
1357 rc = dt_xattr_get(env, dt_object_child(dt), buf, name, capa);
1358 if (strcmp(name, XATTR_NAME_LMV) == 0) {
1359 struct lmv_mds_md_v1 *lmv1;
1362 if (rc > (typeof(rc))sizeof(*lmv1))
1365 if (rc < (typeof(rc))sizeof(*lmv1))
1366 RETURN(rc = rc > 0 ? -EINVAL : rc);
1368 if (buf->lb_buf == NULL || buf->lb_len == 0) {
1369 CLASSERT(sizeof(*lmv1) <= sizeof(info->lti_key));
1371 info->lti_buf.lb_buf = info->lti_key;
1372 info->lti_buf.lb_len = sizeof(*lmv1);
1373 rc = dt_xattr_get(env, dt_object_child(dt),
1374 &info->lti_buf, name, capa);
1375 if (unlikely(rc != sizeof(*lmv1)))
1376 RETURN(rc = rc > 0 ? -EINVAL : rc);
1378 lmv1 = info->lti_buf.lb_buf;
1379 /* The on-disk LMV EA only contains header, but the
1380 * returned LMV EA size should contain the space for
1381 * the FIDs of all shards of the striped directory. */
1382 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_V1)
1383 rc = lmv_mds_md_size(
1384 le32_to_cpu(lmv1->lmv_stripe_count),
1387 rc1 = lod_load_lmv_shards(env, lod_dt_obj(dt),
1391 RETURN(rc = rc1 != 0 ? rc1 : rc);
1394 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
1398 * lod returns default striping on the real root of the device
1399 * this is like the root stores default striping for the whole
1400 * filesystem. historically we've been using a different approach
1401 * and store it in the config.
1403 dt_root_get(env, dev->lod_child, &info->lti_fid);
1404 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
1406 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
1407 struct lov_user_md *lum = buf->lb_buf;
1408 struct lov_desc *desc = &dev->lod_desc;
1410 if (buf->lb_buf == NULL) {
1412 } else if (buf->lb_len >= sizeof(*lum)) {
1413 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
1414 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
1415 lmm_oi_set_id(&lum->lmm_oi, 0);
1416 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
1417 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
1418 lum->lmm_stripe_size = cpu_to_le32(
1419 desc->ld_default_stripe_size);
1420 lum->lmm_stripe_count = cpu_to_le16(
1421 desc->ld_default_stripe_count);
1422 lum->lmm_stripe_offset = cpu_to_le16(
1423 desc->ld_default_stripe_offset);
1436 * Checks that the magic of the stripe is sane.
1438 * \param[in] lod lod device
1439 * \param[in] lum a buffer storing LMV EA to verify
1441 * \retval 0 if the EA is sane
1442 * \retval negative otherwise
1444 static int lod_verify_md_striping(struct lod_device *lod,
1445 const struct lmv_user_md_v1 *lum)
1447 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC)) {
1448 CERROR("%s: invalid lmv_user_md: magic = %x, "
1449 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
1450 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
1451 (int)le32_to_cpu(lum->lum_stripe_offset),
1452 le32_to_cpu(lum->lum_stripe_count), -EINVAL);
1460 * Initialize LMV EA for a slave.
1462 * Initialize slave's LMV EA from the master's LMV EA.
1464 * \param[in] master_lmv a buffer containing master's EA
1465 * \param[out] slave_lmv a buffer where slave's EA will be stored
1468 static void lod_prep_slave_lmv_md(struct lmv_mds_md_v1 *slave_lmv,
1469 const struct lmv_mds_md_v1 *master_lmv)
1471 *slave_lmv = *master_lmv;
1472 slave_lmv->lmv_magic = cpu_to_le32(LMV_MAGIC_STRIPE);
1478 * Generate LMV EA from the object passed as \a dt. The object must have
1479 * the stripes created and initialized.
1481 * \param[in] env execution environment
1482 * \param[in] dt object
1483 * \param[out] lmv_buf buffer storing generated LMV EA
1485 * \retval 0 on success
1486 * \retval negative if failed
1488 static int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
1489 struct lu_buf *lmv_buf)
1491 struct lod_thread_info *info = lod_env_info(env);
1492 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1493 struct lod_object *lo = lod_dt_obj(dt);
1494 struct lmv_mds_md_v1 *lmm1;
1496 int type = LU_SEQ_RANGE_ANY;
1501 LASSERT(lo->ldo_dir_striped != 0);
1502 LASSERT(lo->ldo_stripenr > 0);
1503 stripe_count = lo->ldo_stripenr;
1504 /* Only store the LMV EA heahder on the disk. */
1505 if (info->lti_ea_store_size < sizeof(*lmm1)) {
1506 rc = lod_ea_store_resize(info, sizeof(*lmm1));
1510 memset(info->lti_ea_store, 0, sizeof(*lmm1));
1513 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
1514 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
1515 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
1516 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
1517 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
1522 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
1523 lmv_buf->lb_buf = info->lti_ea_store;
1524 lmv_buf->lb_len = sizeof(*lmm1);
1530 * Create in-core represenation for a striped directory.
1532 * Parse the buffer containing LMV EA and instantiate LU objects
1533 * representing the stripe objects. The pointers to the objects are
1534 * stored in ldo_stripe field of \a lo. This function is used when
1535 * we need to access an already created object (i.e. load from a disk).
1537 * \param[in] env execution environment
1538 * \param[in] lo lod object
1539 * \param[in] buf buffer containing LMV EA
1541 * \retval 0 on success
1542 * \retval negative if failed
1544 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
1545 const struct lu_buf *buf)
1547 struct lod_thread_info *info = lod_env_info(env);
1548 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
1549 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1550 struct dt_object **stripe;
1551 union lmv_mds_md *lmm = buf->lb_buf;
1552 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
1553 struct lu_fid *fid = &info->lti_fid;
1558 if (le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_MIGRATION)
1561 if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE) {
1562 lo->ldo_dir_slave_stripe = 1;
1566 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1569 if (le32_to_cpu(lmv1->lmv_stripe_count) < 1)
1572 LASSERT(lo->ldo_stripe == NULL);
1573 OBD_ALLOC(stripe, sizeof(stripe[0]) *
1574 (le32_to_cpu(lmv1->lmv_stripe_count)));
1578 for (i = 0; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
1579 struct dt_device *tgt_dt;
1580 struct dt_object *dto;
1581 int type = LU_SEQ_RANGE_ANY;
1584 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
1585 if (!fid_is_sane(fid))
1586 GOTO(out, rc = -ESTALE);
1588 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
1592 if (idx == lod2lu_dev(lod)->ld_site->ld_seq_site->ss_node_id) {
1593 tgt_dt = lod->lod_child;
1595 struct lod_tgt_desc *tgt;
1597 tgt = LTD_TGT(ltd, idx);
1599 GOTO(out, rc = -ESTALE);
1600 tgt_dt = tgt->ltd_tgt;
1603 dto = dt_locate_at(env, tgt_dt, fid,
1604 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
1607 GOTO(out, rc = PTR_ERR(dto));
1612 lo->ldo_stripe = stripe;
1613 lo->ldo_stripenr = le32_to_cpu(lmv1->lmv_stripe_count);
1614 lo->ldo_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count);
1616 lod_object_free_striping(env, lo);
1622 * Create a striped directory.
1624 * Create a striped directory with a given stripe pattern on the specified MDTs.
1625 * A striped directory is represented as a regular directory - an index listing
1626 * all the stripes. The stripes point back to the master object with ".." and
1627 * LinkEA. The master object gets LMV EA which identifies it as a striped
1628 * directory. The function allocates FIDs for all the stripes.
1630 * \param[in] env execution environment
1631 * \param[in] dt object
1632 * \param[in] attr attributes to initialize the objects with
1633 * \param[in] lum a pattern specifying the number of stripes and
1635 * \param[in] dof type of objects to be created
1636 * \param[in] th transaction handle
1638 * \retval 0 on success
1639 * \retval negative if failed
1641 static int lod_prep_md_striped_create(const struct lu_env *env,
1642 struct dt_object *dt,
1643 struct lu_attr *attr,
1644 const struct lmv_user_md_v1 *lum,
1645 struct dt_object_format *dof,
1648 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1649 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1650 struct lod_object *lo = lod_dt_obj(dt);
1651 struct lod_thread_info *info = lod_env_info(env);
1652 struct dt_object **stripe;
1653 struct lu_buf lmv_buf;
1654 struct lu_buf slave_lmv_buf;
1655 struct lmv_mds_md_v1 *lmm;
1656 struct lmv_mds_md_v1 *slave_lmm = NULL;
1657 struct dt_insert_rec *rec = &info->lti_dt_rec;
1665 /* The lum has been verifed in lod_verify_md_striping */
1666 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
1667 LASSERT(le32_to_cpu(lum->lum_stripe_count) > 0);
1669 stripe_count = le32_to_cpu(lum->lum_stripe_count);
1671 /* shrink the stripe_count to the avaible MDT count */
1672 if (stripe_count > lod->lod_remote_mdt_count + 1)
1673 stripe_count = lod->lod_remote_mdt_count + 1;
1675 OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_count);
1679 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
1680 if (idx_array == NULL)
1681 GOTO(out_free, rc = -ENOMEM);
1683 for (i = 0; i < stripe_count; i++) {
1684 struct lod_tgt_desc *tgt = NULL;
1685 struct dt_object *dto;
1686 struct lu_fid fid = { 0 };
1688 struct lu_object_conf conf = { 0 };
1689 struct dt_device *tgt_dt = NULL;
1692 /* Right now, master stripe and master object are
1693 * on the same MDT */
1694 idx = le32_to_cpu(lum->lum_stripe_offset);
1695 rc = obd_fid_alloc(env, lod->lod_child_exp, &fid,
1699 tgt_dt = lod->lod_child;
1703 idx = (idx_array[i - 1] + 1) % (lod->lod_remote_mdt_count + 1);
1705 for (j = 0; j < lod->lod_remote_mdt_count;
1706 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
1707 bool already_allocated = false;
1710 CDEBUG(D_INFO, "try idx %d, mdt cnt %u,"
1711 " allocated %u, last allocated %d\n", idx,
1712 lod->lod_remote_mdt_count, i, idx_array[i - 1]);
1714 /* Find next available target */
1715 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx))
1718 /* check whether the idx already exists
1719 * in current allocated array */
1720 for (k = 0; k < i; k++) {
1721 if (idx_array[k] == idx) {
1722 already_allocated = true;
1727 if (already_allocated)
1730 /* check the status of the OSP */
1731 tgt = LTD_TGT(ltd, idx);
1735 tgt_dt = tgt->ltd_tgt;
1736 rc = dt_statfs(env, tgt_dt, NULL);
1738 /* this OSP doesn't feel well */
1743 rc = obd_fid_alloc(env, tgt->ltd_exp, &fid, NULL);
1752 /* Can not allocate more stripes */
1753 if (j == lod->lod_remote_mdt_count) {
1754 CDEBUG(D_INFO, "%s: require stripes %u only get %d\n",
1755 lod2obd(lod)->obd_name, stripe_count, i - 1);
1759 CDEBUG(D_INFO, "idx %d, mdt cnt %u,"
1760 " allocated %u, last allocated %d\n", idx,
1761 lod->lod_remote_mdt_count, i, idx_array[i - 1]);
1764 /* tgt_dt and fid must be ready after search avaible OSP
1765 * in the above loop */
1766 LASSERT(tgt_dt != NULL);
1767 LASSERT(fid_is_sane(&fid));
1768 conf.loc_flags = LOC_F_NEW;
1769 dto = dt_locate_at(env, tgt_dt, &fid,
1770 dt->do_lu.lo_dev->ld_site->ls_top_dev,
1773 GOTO(out_put, rc = PTR_ERR(dto));
1778 lo->ldo_dir_striped = 1;
1779 lo->ldo_stripe = stripe;
1780 lo->ldo_stripenr = i;
1781 lo->ldo_stripes_allocated = stripe_count;
1783 if (lo->ldo_stripenr == 0)
1784 GOTO(out_put, rc = -ENOSPC);
1786 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1789 lmm = lmv_buf.lb_buf;
1791 OBD_ALLOC_PTR(slave_lmm);
1792 if (slave_lmm == NULL)
1793 GOTO(out_put, rc = -ENOMEM);
1795 lod_prep_slave_lmv_md(slave_lmm, lmm);
1796 slave_lmv_buf.lb_buf = slave_lmm;
1797 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
1799 if (!dt_try_as_dir(env, dt_object_child(dt)))
1800 GOTO(out_put, rc = -EINVAL);
1802 rec->rec_type = S_IFDIR;
1803 for (i = 0; i < lo->ldo_stripenr; i++) {
1804 struct dt_object *dto = stripe[i];
1805 char *stripe_name = info->lti_key;
1806 struct lu_name *sname;
1807 struct linkea_data ldata = { NULL };
1808 struct lu_buf linkea_buf;
1810 rc = dt_declare_create(env, dto, attr, NULL, dof, th);
1814 if (!dt_try_as_dir(env, dto))
1815 GOTO(out_put, rc = -EINVAL);
1817 rc = dt_declare_ref_add(env, dto, th);
1821 rec->rec_fid = lu_object_fid(&dto->do_lu);
1822 rc = dt_declare_insert(env, dto, (const struct dt_rec *)rec,
1823 (const struct dt_key *)dot, th);
1827 /* master stripe FID will be put to .. */
1828 rec->rec_fid = lu_object_fid(&dt->do_lu);
1829 rc = dt_declare_insert(env, dto, (const struct dt_rec *)rec,
1830 (const struct dt_key *)dotdot, th);
1834 /* probably nothing to inherite */
1835 if (lo->ldo_def_striping_set &&
1836 !LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size,
1837 lo->ldo_def_stripenr,
1838 lo->ldo_def_stripe_offset,
1840 struct lov_user_md_v3 *v3;
1842 /* sigh, lti_ea_store has been used for lmv_buf,
1843 * so we have to allocate buffer for default
1847 GOTO(out_put, rc = -ENOMEM);
1849 memset(v3, 0, sizeof(*v3));
1850 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
1851 v3->lmm_stripe_count =
1852 cpu_to_le16(lo->ldo_def_stripenr);
1853 v3->lmm_stripe_offset =
1854 cpu_to_le16(lo->ldo_def_stripe_offset);
1855 v3->lmm_stripe_size =
1856 cpu_to_le32(lo->ldo_def_stripe_size);
1857 if (lo->ldo_pool != NULL)
1858 strlcpy(v3->lmm_pool_name, lo->ldo_pool,
1859 sizeof(v3->lmm_pool_name));
1861 info->lti_buf.lb_buf = v3;
1862 info->lti_buf.lb_len = sizeof(*v3);
1863 rc = dt_declare_xattr_set(env, dto,
1872 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
1873 cfs_fail_val != i) {
1874 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
1876 slave_lmm->lmv_master_mdt_index =
1879 slave_lmm->lmv_master_mdt_index =
1881 rc = dt_declare_xattr_set(env, dto, &slave_lmv_buf,
1882 XATTR_NAME_LMV, 0, th);
1887 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
1889 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1890 PFID(lu_object_fid(&dto->do_lu)), i + 1);
1892 snprintf(stripe_name, sizeof(info->lti_key), DFID":%u",
1893 PFID(lu_object_fid(&dto->do_lu)), i);
1895 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
1896 rc = linkea_data_new(&ldata, &info->lti_linkea_buf);
1900 rc = linkea_add_buf(&ldata, sname, lu_object_fid(&dt->do_lu));
1904 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
1905 linkea_buf.lb_len = ldata.ld_leh->leh_len;
1906 rc = dt_declare_xattr_set(env, dto, &linkea_buf,
1907 XATTR_NAME_LINK, 0, th);
1911 rec->rec_fid = lu_object_fid(&dto->do_lu);
1912 rc = dt_declare_insert(env, dt_object_child(dt),
1913 (const struct dt_rec *)rec,
1914 (const struct dt_key *)stripe_name, th);
1918 rc = dt_declare_ref_add(env, dt_object_child(dt), th);
1923 rc = dt_declare_xattr_set(env, dt_object_child(dt), &lmv_buf,
1924 XATTR_NAME_LMV, 0, th);
1930 for (i = 0; i < stripe_count; i++)
1931 if (stripe[i] != NULL)
1932 lu_object_put(env, &stripe[i]->do_lu);
1933 OBD_FREE(stripe, sizeof(stripe[0]) * stripe_count);
1934 lo->ldo_stripenr = 0;
1935 lo->ldo_stripes_allocated = 0;
1936 lo->ldo_stripe = NULL;
1940 if (idx_array != NULL)
1941 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
1942 if (slave_lmm != NULL)
1943 OBD_FREE_PTR(slave_lmm);
1949 * Declare create striped md object.
1951 * The function declares intention to create a striped directory. This is a
1952 * wrapper for lod_prep_md_striped_create(). The only additional functionality
1953 * is to verify pattern \a lum_buf is good. Check that function for the details.
1955 * \param[in] env execution environment
1956 * \param[in] dt object
1957 * \param[in] attr attributes to initialize the objects with
1958 * \param[in] lum_buf a pattern specifying the number of stripes and
1960 * \param[in] dof type of objects to be created
1961 * \param[in] th transaction handle
1963 * \retval 0 on success
1964 * \retval negative if failed
1967 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
1968 struct dt_object *dt,
1969 struct lu_attr *attr,
1970 const struct lu_buf *lum_buf,
1971 struct dt_object_format *dof,
1974 struct lod_object *lo = lod_dt_obj(dt);
1975 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
1976 struct lmv_user_md_v1 *lum;
1980 lum = lum_buf->lb_buf;
1981 LASSERT(lum != NULL);
1983 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
1984 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
1985 (int)le32_to_cpu(lum->lum_stripe_offset));
1987 if (le32_to_cpu(lum->lum_stripe_count) == 0)
1990 rc = lod_verify_md_striping(lod, lum);
1994 /* prepare dir striped objects */
1995 rc = lod_prep_md_striped_create(env, dt, attr, lum, dof, th);
1997 /* failed to create striping, let's reset
1998 * config so that others don't get confused */
1999 lod_object_free_striping(env, lo);
2008 * Implementation of dt_object_operations::do_declare_xattr_set.
2010 * Used with regular (non-striped) objects. Basically it
2011 * initializes the striping information and applies the
2012 * change to all the stripes.
2014 * \see dt_object_operations::do_declare_xattr_set() in the API description
2017 static int lod_dir_declare_xattr_set(const struct lu_env *env,
2018 struct dt_object *dt,
2019 const struct lu_buf *buf,
2020 const char *name, int fl,
2023 struct dt_object *next = dt_object_child(dt);
2024 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2025 struct lod_object *lo = lod_dt_obj(dt);
2030 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2031 struct lmv_user_md_v1 *lum;
2033 LASSERT(buf != NULL && buf->lb_buf != NULL);
2035 rc = lod_verify_md_striping(d, lum);
2040 rc = dt_declare_xattr_set(env, next, buf, name, fl, th);
2044 /* set xattr to each stripes, if needed */
2045 rc = lod_load_striping(env, lo);
2049 /* Note: Do not set LinkEA on sub-stripes, otherwise
2050 * it will confuse the fid2path process(see mdt_path_current()).
2051 * The linkEA between master and sub-stripes is set in
2052 * lod_xattr_set_lmv(). */
2053 if (lo->ldo_stripenr == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
2056 for (i = 0; i < lo->ldo_stripenr; i++) {
2057 LASSERT(lo->ldo_stripe[i]);
2058 rc = dt_declare_xattr_set(env, lo->ldo_stripe[i], buf,
2068 * Implementation of dt_object_operations::do_declare_xattr_set.
2070 * \see dt_object_operations::do_declare_xattr_set() in the API description
2073 * the extension to the API:
2074 * - declaring LOVEA requests striping creation
2075 * - LU_XATTR_REPLACE means layout swap
2077 static int lod_declare_xattr_set(const struct lu_env *env,
2078 struct dt_object *dt,
2079 const struct lu_buf *buf,
2080 const char *name, int fl,
2083 struct dt_object *next = dt_object_child(dt);
2084 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
2090 * allow to declare predefined striping on a new (!mode) object
2091 * which is supposed to be replay of regular file creation
2092 * (when LOV setting is declared)
2093 * LU_XATTR_REPLACE is set to indicate a layout swap
2095 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
2096 if ((S_ISREG(mode) || mode == 0) && strcmp(name, XATTR_NAME_LOV) == 0 &&
2097 !(fl & LU_XATTR_REPLACE)) {
2099 * this is a request to manipulate object's striping
2101 if (dt_object_exists(dt)) {
2102 rc = dt_attr_get(env, next, attr, BYPASS_CAPA);
2106 memset(attr, 0, sizeof(*attr));
2107 attr->la_valid = LA_TYPE | LA_MODE;
2108 attr->la_mode = S_IFREG;
2110 rc = lod_declare_striped_object(env, dt, attr, buf, th);
2111 } else if (S_ISDIR(mode)) {
2112 rc = lod_dir_declare_xattr_set(env, dt, buf, name, fl, th);
2114 rc = dt_declare_xattr_set(env, next, buf, name, fl, th);
2121 * Resets cached default striping in the object.
2123 * \param[in] lo object
2125 static void lod_lov_stripe_cache_clear(struct lod_object *lo)
2127 lo->ldo_def_striping_set = 0;
2128 lo->ldo_def_striping_cached = 0;
2129 lod_object_set_pool(lo, NULL);
2130 lo->ldo_def_stripe_size = 0;
2131 lo->ldo_def_stripenr = 0;
2132 if (lo->ldo_dir_stripe != NULL)
2133 lo->ldo_dir_def_striping_cached = 0;
2137 * Apply xattr changes to the object.
2139 * Applies xattr changes to the object and the stripes if the latter exist.
2141 * \param[in] env execution environment
2142 * \param[in] dt object
2143 * \param[in] buf buffer pointing to the new value of xattr
2144 * \param[in] name name of xattr
2145 * \param[in] fl flags
2146 * \param[in] th transaction handle
2147 * \param[in] capa not used currently
2149 * \retval 0 on success
2150 * \retval negative if failed
2152 static int lod_xattr_set_internal(const struct lu_env *env,
2153 struct dt_object *dt,
2154 const struct lu_buf *buf,
2155 const char *name, int fl, struct thandle *th,
2156 struct lustre_capa *capa)
2158 struct dt_object *next = dt_object_child(dt);
2159 struct lod_object *lo = lod_dt_obj(dt);
2164 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
2165 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2168 /* Note: Do not set LinkEA on sub-stripes, otherwise
2169 * it will confuse the fid2path process(see mdt_path_current()).
2170 * The linkEA between master and sub-stripes is set in
2171 * lod_xattr_set_lmv(). */
2172 if (lo->ldo_stripenr == 0 || strcmp(name, XATTR_NAME_LINK) == 0)
2175 for (i = 0; i < lo->ldo_stripenr; i++) {
2176 LASSERT(lo->ldo_stripe[i]);
2177 rc = dt_xattr_set(env, lo->ldo_stripe[i], buf, name, fl, th,
2187 * Delete an extended attribute.
2189 * Deletes specified xattr from the object and the stripes if the latter exist.
2191 * \param[in] env execution environment
2192 * \param[in] dt object
2193 * \param[in] name name of xattr
2194 * \param[in] th transaction handle
2195 * \param[in] capa not used currently
2197 * \retval 0 on success
2198 * \retval negative if failed
2200 static int lod_xattr_del_internal(const struct lu_env *env,
2201 struct dt_object *dt,
2202 const char *name, struct thandle *th,
2203 struct lustre_capa *capa)
2205 struct dt_object *next = dt_object_child(dt);
2206 struct lod_object *lo = lod_dt_obj(dt);
2211 rc = dt_xattr_del(env, next, name, th, capa);
2212 if (rc != 0 || !S_ISDIR(dt->do_lu.lo_header->loh_attr))
2215 if (lo->ldo_stripenr == 0)
2218 for (i = 0; i < lo->ldo_stripenr; i++) {
2219 LASSERT(lo->ldo_stripe[i]);
2220 rc = dt_xattr_del(env, lo->ldo_stripe[i], name, th,
2230 * Set default striping on a directory.
2232 * Sets specified striping on a directory object unless it matches the default
2233 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
2234 * EA. This striping will be used when regular file is being created in this
2237 * \param[in] env execution environment
2238 * \param[in] dt the striped object
2239 * \param[in] buf buffer with the striping
2240 * \param[in] name name of EA
2241 * \param[in] fl xattr flag (see OSD API description)
2242 * \param[in] th transaction handle
2243 * \param[in] capa not used
2245 * \retval 0 on success
2246 * \retval negative if failed
2248 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
2249 struct dt_object *dt,
2250 const struct lu_buf *buf,
2251 const char *name, int fl,
2253 struct lustre_capa *capa)
2255 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
2256 struct lod_object *l = lod_dt_obj(dt);
2257 struct lov_user_md_v1 *lum;
2258 struct lov_user_md_v3 *v3 = NULL;
2259 const char *pool_name = NULL;
2263 /* If it is striped dir, we should clear the stripe cache for
2264 * slave stripe as well, but there are no effective way to
2265 * notify the LOD on the slave MDT, so we do not cache stripe
2266 * information for slave stripe for now. XXX*/
2267 lod_lov_stripe_cache_clear(l);
2268 LASSERT(buf != NULL && buf->lb_buf != NULL);
2271 rc = lod_verify_striping(d, buf, false);
2275 if (lum->lmm_magic == LOV_USER_MAGIC_V3) {
2277 if (v3->lmm_pool_name[0] != '\0')
2278 pool_name = v3->lmm_pool_name;
2281 /* if { size, offset, count } = { 0, -1, 0 } and no pool
2282 * (i.e. all default values specified) then delete default
2283 * striping from dir. */
2285 "set default striping: sz %u # %u offset %d %s %s\n",
2286 (unsigned)lum->lmm_stripe_size,
2287 (unsigned)lum->lmm_stripe_count,
2288 (int)lum->lmm_stripe_offset,
2289 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
2291 if (LOVEA_DELETE_VALUES(lum->lmm_stripe_size, lum->lmm_stripe_count,
2292 lum->lmm_stripe_offset, pool_name)) {
2293 rc = lod_xattr_del_internal(env, dt, name, th, capa);
2297 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th, capa);
2304 * Set default striping on a directory object.
2306 * Sets specified striping on a directory object unless it matches the default
2307 * striping (LOVEA_DELETE_VALUES() macro). In the latter case remove existing
2308 * EA. This striping will be used when a new directory is being created in the
2311 * \param[in] env execution environment
2312 * \param[in] dt the striped object
2313 * \param[in] buf buffer with the striping
2314 * \param[in] name name of EA
2315 * \param[in] fl xattr flag (see OSD API description)
2316 * \param[in] th transaction handle
2317 * \param[in] capa not used
2319 * \retval 0 on success
2320 * \retval negative if failed
2322 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
2323 struct dt_object *dt,
2324 const struct lu_buf *buf,
2325 const char *name, int fl,
2327 struct lustre_capa *capa)
2329 struct lod_object *l = lod_dt_obj(dt);
2330 struct lmv_user_md_v1 *lum;
2334 LASSERT(buf != NULL && buf->lb_buf != NULL);
2337 CDEBUG(D_OTHER, "set default stripe_count # %u stripe_offset %d\n",
2338 le32_to_cpu(lum->lum_stripe_count),
2339 (int)le32_to_cpu(lum->lum_stripe_offset));
2341 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
2342 le32_to_cpu(lum->lum_stripe_offset)) &&
2343 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
2344 rc = lod_xattr_del_internal(env, dt, name, th, capa);
2348 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th, capa);
2353 /* Update default stripe cache */
2354 if (l->ldo_dir_stripe == NULL) {
2355 OBD_ALLOC_PTR(l->ldo_dir_stripe);
2356 if (l->ldo_dir_stripe == NULL)
2360 l->ldo_dir_def_striping_cached = 0;
2365 * Turn directory into a striped directory.
2367 * During replay the client sends the striping created before MDT
2368 * failure, then the layer above LOD sends this defined striping
2369 * using ->do_xattr_set(), so LOD uses this method to replay creation
2370 * of the stripes. Notice the original information for the striping
2371 * (#stripes, FIDs, etc) was transfered in declare path.
2373 * \param[in] env execution environment
2374 * \param[in] dt the striped object
2375 * \param[in] buf not used currently
2376 * \param[in] name not used currently
2377 * \param[in] fl xattr flag (see OSD API description)
2378 * \param[in] th transaction handle
2379 * \param[in] capa not used
2381 * \retval 0 on success
2382 * \retval negative if failed
2384 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
2385 const struct lu_buf *buf, const char *name,
2386 int fl, struct thandle *th,
2387 struct lustre_capa *capa)
2389 struct lod_object *lo = lod_dt_obj(dt);
2390 struct lod_thread_info *info = lod_env_info(env);
2391 struct lu_attr *attr = &info->lti_attr;
2392 struct dt_object_format *dof = &info->lti_format;
2393 struct lu_buf lmv_buf;
2394 struct lu_buf slave_lmv_buf;
2395 struct lmv_mds_md_v1 *lmm;
2396 struct lmv_mds_md_v1 *slave_lmm = NULL;
2397 struct dt_insert_rec *rec = &info->lti_dt_rec;
2402 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
2405 /* The stripes are supposed to be allocated in declare phase,
2406 * if there are no stripes being allocated, it will skip */
2407 if (lo->ldo_stripenr == 0)
2410 rc = dt_attr_get(env, dt_object_child(dt), attr, BYPASS_CAPA);
2414 attr->la_valid = LA_ATIME | LA_MTIME | LA_CTIME |
2415 LA_MODE | LA_UID | LA_GID | LA_TYPE;
2416 dof->dof_type = DFT_DIR;
2418 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
2421 lmm = lmv_buf.lb_buf;
2423 OBD_ALLOC_PTR(slave_lmm);
2424 if (slave_lmm == NULL)
2427 lod_prep_slave_lmv_md(slave_lmm, lmm);
2428 slave_lmv_buf.lb_buf = slave_lmm;
2429 slave_lmv_buf.lb_len = sizeof(*slave_lmm);
2431 rec->rec_type = S_IFDIR;
2432 for (i = 0; i < lo->ldo_stripenr; i++) {
2433 struct dt_object *dto;
2434 char *stripe_name = info->lti_key;
2435 struct lu_name *sname;
2436 struct linkea_data ldata = { NULL };
2437 struct lu_buf linkea_buf;
2439 dto = lo->ldo_stripe[i];
2440 dt_write_lock(env, dto, MOR_TGT_CHILD);
2441 rc = dt_create(env, dto, attr, NULL, dof, th);
2443 dt_write_unlock(env, dto);
2447 rc = dt_ref_add(env, dto, th);
2448 dt_write_unlock(env, dto);
2452 rec->rec_fid = lu_object_fid(&dto->do_lu);
2453 rc = dt_insert(env, dto, (const struct dt_rec *)rec,
2454 (const struct dt_key *)dot, th, capa, 0);
2458 rec->rec_fid = lu_object_fid(&dt->do_lu);
2459 rc = dt_insert(env, dto, (struct dt_rec *)rec,
2460 (const struct dt_key *)dotdot, th, capa, 0);
2464 if (lo->ldo_def_striping_set &&
2465 !LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size,
2466 lo->ldo_def_stripenr,
2467 lo->ldo_def_stripe_offset,
2469 struct lov_user_md_v3 *v3;
2471 /* sigh, lti_ea_store has been used for lmv_buf,
2472 * so we have to allocate buffer for default
2478 memset(v3, 0, sizeof(*v3));
2479 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
2480 v3->lmm_stripe_count =
2481 cpu_to_le16(lo->ldo_def_stripenr);
2482 v3->lmm_stripe_offset =
2483 cpu_to_le16(lo->ldo_def_stripe_offset);
2484 v3->lmm_stripe_size =
2485 cpu_to_le32(lo->ldo_def_stripe_size);
2486 if (lo->ldo_pool != NULL)
2487 strlcpy(v3->lmm_pool_name, lo->ldo_pool,
2488 sizeof(v3->lmm_pool_name));
2490 info->lti_buf.lb_buf = v3;
2491 info->lti_buf.lb_len = sizeof(*v3);
2492 rc = dt_xattr_set(env, dto, &info->lti_buf,
2493 XATTR_NAME_LOV, 0, th, capa);
2499 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SLAVE_LMV) ||
2500 cfs_fail_val != i) {
2501 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_LMV) &&
2503 slave_lmm->lmv_master_mdt_index =
2506 slave_lmm->lmv_master_mdt_index =
2508 rc = dt_xattr_set(env, dto, &slave_lmv_buf,
2509 XATTR_NAME_LMV, fl, th, capa);
2514 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME) &&
2516 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2517 PFID(lu_object_fid(&dto->do_lu)), i + 1);
2519 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
2520 PFID(lu_object_fid(&dto->do_lu)), i);
2522 sname = lod_name_get(env, stripe_name, strlen(stripe_name));
2523 rc = linkea_data_new(&ldata, &info->lti_linkea_buf);
2527 rc = linkea_add_buf(&ldata, sname, lu_object_fid(&dt->do_lu));
2531 linkea_buf.lb_buf = ldata.ld_buf->lb_buf;
2532 linkea_buf.lb_len = ldata.ld_leh->leh_len;
2533 rc = dt_xattr_set(env, dto, &linkea_buf, XATTR_NAME_LINK,
2534 0, th, BYPASS_CAPA);
2538 rec->rec_fid = lu_object_fid(&dto->do_lu);
2539 rc = dt_insert(env, dt_object_child(dt),
2540 (const struct dt_rec *)rec,
2541 (const struct dt_key *)stripe_name, th, capa, 0);
2545 rc = dt_ref_add(env, dt_object_child(dt), th);
2550 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MASTER_LMV))
2551 rc = dt_xattr_set(env, dt_object_child(dt), &lmv_buf,
2552 XATTR_NAME_LMV, fl, th, capa);
2555 if (slave_lmm != NULL)
2556 OBD_FREE_PTR(slave_lmm);
2562 * Helper function to declare/execute creation of a striped directory
2564 * Called in declare/create object path, prepare striping for a directory
2565 * and prepare defaults data striping for the objects to be created in
2566 * that directory. Notice the function calls "declaration" or "execution"
2567 * methods depending on \a declare param. This is a consequence of the
2568 * current approach while we don't have natural distributed transactions:
2569 * we basically execute non-local updates in the declare phase. So, the
2570 * arguments for the both phases are the same and this is the reason for
2571 * this function to exist.
2573 * \param[in] env execution environment
2574 * \param[in] dt object
2575 * \param[in] attr attributes the stripes will be created with
2576 * \param[in] dof format of stripes (see OSD API description)
2577 * \param[in] th transaction handle
2578 * \param[in] declare where to call "declare" or "execute" methods
2580 * \retval 0 on success
2581 * \retval negative if failed
2583 static int lod_dir_striping_create_internal(const struct lu_env *env,
2584 struct dt_object *dt,
2585 struct lu_attr *attr,
2586 struct dt_object_format *dof,
2590 struct lod_thread_info *info = lod_env_info(env);
2591 struct lod_object *lo = lod_dt_obj(dt);
2595 if (!LMVEA_DELETE_VALUES(lo->ldo_stripenr,
2596 lo->ldo_dir_stripe_offset)) {
2597 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
2598 int stripe_count = lo->ldo_stripenr;
2600 if (info->lti_ea_store_size < sizeof(*v1)) {
2601 rc = lod_ea_store_resize(info, sizeof(*v1));
2604 v1 = info->lti_ea_store;
2607 memset(v1, 0, sizeof(*v1));
2608 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
2609 v1->lum_stripe_count = cpu_to_le32(stripe_count);
2610 v1->lum_stripe_offset =
2611 cpu_to_le32(lo->ldo_dir_stripe_offset);
2613 info->lti_buf.lb_buf = v1;
2614 info->lti_buf.lb_len = sizeof(*v1);
2617 rc = lod_declare_xattr_set_lmv(env, dt, attr,
2618 &info->lti_buf, dof, th);
2620 rc = lod_xattr_set_lmv(env, dt, &info->lti_buf,
2621 XATTR_NAME_LMV, 0, th,
2627 /* Transfer default LMV striping from the parent */
2628 if (lo->ldo_dir_def_striping_set &&
2629 !LMVEA_DELETE_VALUES(lo->ldo_dir_def_stripenr,
2630 lo->ldo_dir_def_stripe_offset)) {
2631 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
2632 int def_stripe_count = lo->ldo_dir_def_stripenr;
2634 if (info->lti_ea_store_size < sizeof(*v1)) {
2635 rc = lod_ea_store_resize(info, sizeof(*v1));
2638 v1 = info->lti_ea_store;
2641 memset(v1, 0, sizeof(*v1));
2642 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
2643 v1->lum_stripe_count = cpu_to_le32(def_stripe_count);
2644 v1->lum_stripe_offset =
2645 cpu_to_le32(lo->ldo_dir_def_stripe_offset);
2647 cpu_to_le32(lo->ldo_dir_def_hash_type);
2649 info->lti_buf.lb_buf = v1;
2650 info->lti_buf.lb_len = sizeof(*v1);
2652 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
2653 XATTR_NAME_DEFAULT_LMV,
2656 rc = lod_xattr_set_default_lmv_on_dir(env, dt,
2658 XATTR_NAME_DEFAULT_LMV, 0,
2664 /* Transfer default LOV striping from the parent */
2665 if (lo->ldo_def_striping_set &&
2666 !LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size,
2667 lo->ldo_def_stripenr,
2668 lo->ldo_def_stripe_offset,
2670 struct lov_user_md_v3 *v3 = info->lti_ea_store;
2672 if (info->lti_ea_store_size < sizeof(*v3)) {
2673 rc = lod_ea_store_resize(info, sizeof(*v3));
2676 v3 = info->lti_ea_store;
2679 memset(v3, 0, sizeof(*v3));
2680 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
2681 v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
2682 v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
2683 v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
2684 if (lo->ldo_pool != NULL)
2685 strlcpy(v3->lmm_pool_name, lo->ldo_pool,
2686 sizeof(v3->lmm_pool_name));
2688 info->lti_buf.lb_buf = v3;
2689 info->lti_buf.lb_len = sizeof(*v3);
2692 rc = lod_dir_declare_xattr_set(env, dt, &info->lti_buf,
2693 XATTR_NAME_LOV, 0, th);
2695 rc = lod_xattr_set_lov_on_dir(env, dt, &info->lti_buf,
2696 XATTR_NAME_LOV, 0, th,
2705 static int lod_declare_dir_striping_create(const struct lu_env *env,
2706 struct dt_object *dt,
2707 struct lu_attr *attr,
2708 struct dt_object_format *dof,
2711 return lod_dir_striping_create_internal(env, dt, attr, dof, th, true);
2714 static int lod_dir_striping_create(const struct lu_env *env,
2715 struct dt_object *dt,
2716 struct lu_attr *attr,
2717 struct dt_object_format *dof,
2720 struct lod_object *lo = lod_dt_obj(dt);
2723 rc = lod_dir_striping_create_internal(env, dt, attr, dof, th, false);
2725 lo->ldo_striping_cached = 1;
2731 * Implementation of dt_object_operations::do_xattr_set.
2733 * Sets specified extended attribute on the object. Three types of EAs are
2735 * LOV EA - stores striping for a regular file or default striping (when set
2737 * LMV EA - stores a marker for the striped directories
2738 * DMV EA - stores default directory striping
2740 * When striping is applied to a non-striped existing object (this is called
2741 * late striping), then LOD notices the caller wants to turn the object into a
2742 * striped one. The stripe objects are created and appropriate EA is set:
2743 * LOV EA storing all the stripes directly or LMV EA storing just a small header
2744 * with striping configuration.
2746 * \see dt_object_operations::do_xattr_set() in the API description for details.
2748 static int lod_xattr_set(const struct lu_env *env,
2749 struct dt_object *dt, const struct lu_buf *buf,
2750 const char *name, int fl, struct thandle *th,
2751 struct lustre_capa *capa)
2753 struct dt_object *next = dt_object_child(dt);
2757 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
2758 strcmp(name, XATTR_NAME_LMV) == 0) {
2759 struct lmv_mds_md_v1 *lmm = buf->lb_buf;
2761 if (lmm != NULL && le32_to_cpu(lmm->lmv_hash_type) &
2762 LMV_HASH_FLAG_MIGRATION)
2763 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
2765 rc = lod_dir_striping_create(env, dt, NULL, NULL, th);
2770 if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
2771 strcmp(name, XATTR_NAME_LOV) == 0) {
2773 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl, th, capa);
2775 } else if (S_ISDIR(dt->do_lu.lo_header->loh_attr) &&
2776 strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
2778 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
2781 } else if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
2782 !strcmp(name, XATTR_NAME_LOV)) {
2783 /* in case of lov EA swap, just set it
2784 * if not, it is a replay so check striping match what we
2785 * already have during req replay, declare_xattr_set()
2786 * defines striping, then create() does the work
2788 if (fl & LU_XATTR_REPLACE) {
2789 /* free stripes, then update disk */
2790 lod_object_free_striping(env, lod_dt_obj(dt));
2791 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
2793 rc = lod_striping_create(env, dt, NULL, NULL, th);
2798 /* then all other xattr */
2799 rc = lod_xattr_set_internal(env, dt, buf, name, fl, th, capa);
2805 * Implementation of dt_object_operations::do_declare_xattr_del.
2807 * \see dt_object_operations::do_declare_xattr_del() in the API description
2810 static int lod_declare_xattr_del(const struct lu_env *env,
2811 struct dt_object *dt, const char *name,
2814 return dt_declare_xattr_del(env, dt_object_child(dt), name, th);
2818 * Implementation of dt_object_operations::do_xattr_del.
2820 * If EA storing a regular striping is being deleted, then release
2821 * all the references to the stripe objects in core.
2823 * \see dt_object_operations::do_xattr_del() in the API description for details.
2825 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
2826 const char *name, struct thandle *th,
2827 struct lustre_capa *capa)
2829 if (!strcmp(name, XATTR_NAME_LOV))
2830 lod_object_free_striping(env, lod_dt_obj(dt));
2831 return dt_xattr_del(env, dt_object_child(dt), name, th, capa);
2835 * Implementation of dt_object_operations::do_xattr_list.
2837 * \see dt_object_operations::do_xattr_list() in the API description
2840 static int lod_xattr_list(const struct lu_env *env,
2841 struct dt_object *dt, const struct lu_buf *buf,
2842 struct lustre_capa *capa)
2844 return dt_xattr_list(env, dt_object_child(dt), buf, capa);
2848 * Initialize a pool the object belongs to.
2850 * When a striped object is being created, striping configuration
2851 * may demand the stripes are allocated on a limited set of the
2852 * targets. These limited sets are known as "pools". So we copy
2853 * a pool name into the object and later actual creation methods
2854 * (like lod_object_create()) will use this information to allocate
2855 * the stripes properly.
2857 * \param[in] o object
2858 * \param[in] pool pool name
2860 int lod_object_set_pool(struct lod_object *o, char *pool)
2865 len = strlen(o->ldo_pool);
2866 OBD_FREE(o->ldo_pool, len + 1);
2871 OBD_ALLOC(o->ldo_pool, len + 1);
2872 if (o->ldo_pool == NULL)
2874 strcpy(o->ldo_pool, pool);
2879 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
2881 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
2886 * Cache default regular striping in the object.
2888 * To improve performance of striped regular object creation we cache
2889 * default LOV striping (if it exists) in the parent directory object.
2891 * \param[in] env execution environment
2892 * \param[in] lp object
2894 * \retval 0 on success
2895 * \retval negative if failed
2897 static int lod_cache_parent_lov_striping(const struct lu_env *env,
2898 struct lod_object *lp)
2900 struct lod_thread_info *info = lod_env_info(env);
2901 struct lov_user_md_v1 *v1 = NULL;
2902 struct lov_user_md_v3 *v3 = NULL;
2906 /* called from MDD without parent being write locked,
2908 dt_write_lock(env, dt_object_child(&lp->ldo_obj), 0);
2909 rc = lod_get_lov_ea(env, lp);
2913 if (rc < (typeof(rc))sizeof(struct lov_user_md)) {
2914 /* don't lookup for non-existing or invalid striping */
2915 lp->ldo_def_striping_set = 0;
2916 lp->ldo_def_striping_cached = 1;
2917 lp->ldo_def_stripe_size = 0;
2918 lp->ldo_def_stripenr = 0;
2919 lp->ldo_def_stripe_offset = (typeof(v1->lmm_stripe_offset))(-1);
2920 GOTO(unlock, rc = 0);
2924 v1 = info->lti_ea_store;
2925 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
2926 lustre_swab_lov_user_md_v1(v1);
2927 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
2928 v3 = (struct lov_user_md_v3 *)v1;
2929 lustre_swab_lov_user_md_v3(v3);
2932 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1)
2933 GOTO(unlock, rc = 0);
2935 if (v1->lmm_pattern != LOV_PATTERN_RAID0 && v1->lmm_pattern != 0)
2936 GOTO(unlock, rc = 0);
2938 CDEBUG(D_INFO, DFID" stripe_count=%d stripe_size=%d stripe_offset=%d\n",
2939 PFID(lu_object_fid(&lp->ldo_obj.do_lu)),
2940 (int)v1->lmm_stripe_count,
2941 (int)v1->lmm_stripe_size, (int)v1->lmm_stripe_offset);
2943 lp->ldo_def_stripenr = v1->lmm_stripe_count;
2944 lp->ldo_def_stripe_size = v1->lmm_stripe_size;
2945 lp->ldo_def_stripe_offset = v1->lmm_stripe_offset;
2946 lp->ldo_def_striping_cached = 1;
2947 lp->ldo_def_striping_set = 1;
2948 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
2949 /* XXX: sanity check here */
2950 v3 = (struct lov_user_md_v3 *) v1;
2951 if (v3->lmm_pool_name[0])
2952 lod_object_set_pool(lp, v3->lmm_pool_name);
2956 dt_write_unlock(env, dt_object_child(&lp->ldo_obj));
2962 * Cache default directory striping in the object.
2964 * To improve performance of striped directory creation we cache default
2965 * directory striping (if it exists) in the parent directory object.
2967 * \param[in] env execution environment
2968 * \param[in] lp object
2970 * \retval 0 on success
2971 * \retval negative if failed
2973 static int lod_cache_parent_lmv_striping(const struct lu_env *env,
2974 struct lod_object *lp)
2976 struct lod_thread_info *info = lod_env_info(env);
2977 struct lmv_user_md_v1 *v1 = NULL;
2981 /* called from MDD without parent being write locked,
2983 dt_write_lock(env, dt_object_child(&lp->ldo_obj), 0);
2984 rc = lod_get_default_lmv_ea(env, lp);
2988 if (rc < (typeof(rc))sizeof(struct lmv_user_md)) {
2989 /* don't lookup for non-existing or invalid striping */
2990 lp->ldo_dir_def_striping_set = 0;
2991 lp->ldo_dir_def_striping_cached = 1;
2992 lp->ldo_dir_def_stripenr = 0;
2993 lp->ldo_dir_def_stripe_offset =
2994 (typeof(v1->lum_stripe_offset))(-1);
2995 lp->ldo_dir_def_hash_type = LMV_HASH_TYPE_FNV_1A_64;
2996 GOTO(unlock, rc = 0);
3000 v1 = info->lti_ea_store;
3002 lp->ldo_dir_def_stripenr = le32_to_cpu(v1->lum_stripe_count);
3003 lp->ldo_dir_def_stripe_offset = le32_to_cpu(v1->lum_stripe_offset);
3004 lp->ldo_dir_def_hash_type = le32_to_cpu(v1->lum_hash_type);
3005 lp->ldo_dir_def_striping_set = 1;
3006 lp->ldo_dir_def_striping_cached = 1;
3010 dt_write_unlock(env, dt_object_child(&lp->ldo_obj));
3015 * Cache default striping in the object.
3017 * To improve performance of striped object creation we cache default striping
3018 * (if it exists) in the parent directory object. We always cache default
3019 * striping for the regular files (stored in LOV EA) and we cache default
3020 * striping for the directories if requested by \a child_mode (when a new
3021 * directory is being created).
3023 * \param[in] env execution environment
3024 * \param[in] lp object
3025 * \param[in] child_mode new object's mode
3027 * \retval 0 on success
3028 * \retval negative if failed
3030 static int lod_cache_parent_striping(const struct lu_env *env,
3031 struct lod_object *lp,
3037 if (!lp->ldo_def_striping_cached) {
3038 /* we haven't tried to get default striping for
3039 * the directory yet, let's cache it in the object */
3040 rc = lod_cache_parent_lov_striping(env, lp);
3045 /* If the parent is on the remote MDT, we should always
3046 * try to refresh the default stripeEA cache, because we
3047 * do not cache default striping information for remote
3049 if (S_ISDIR(child_mode) && (!lp->ldo_dir_def_striping_cached ||
3050 dt_object_remote(&lp->ldo_obj)))
3051 rc = lod_cache_parent_lmv_striping(env, lp);
3057 * Implementation of dt_object_operations::do_ah_init.
3059 * This method is used to make a decision on the striping configuration for the
3060 * object being created. It can be taken from the \a parent object if it exists,
3061 * or filesystem's default. The resulting configuration (number of stripes,
3062 * stripe size/offset, pool name, etc) is stored in the object itself and will
3063 * be used by the methods like ->doo_declare_create().
3065 * \see dt_object_operations::do_ah_init() in the API description for details.
3067 static void lod_ah_init(const struct lu_env *env,
3068 struct dt_allocation_hint *ah,
3069 struct dt_object *parent,
3070 struct dt_object *child,
3073 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
3074 struct dt_object *nextp = NULL;
3075 struct dt_object *nextc;
3076 struct lod_object *lp = NULL;
3077 struct lod_object *lc;
3078 struct lov_desc *desc;
3084 if (likely(parent)) {
3085 nextp = dt_object_child(parent);
3086 lp = lod_dt_obj(parent);
3087 rc = lod_load_striping(env, lp);
3092 nextc = dt_object_child(child);
3093 lc = lod_dt_obj(child);
3095 LASSERT(lc->ldo_stripenr == 0);
3096 LASSERT(lc->ldo_stripe == NULL);
3099 * local object may want some hints
3100 * in case of late striping creation, ->ah_init()
3101 * can be called with local object existing
3103 if (!dt_object_exists(nextc) || dt_object_remote(nextc)) {
3104 struct dt_object *obj;
3106 obj = (nextp != NULL && dt_object_remote(nextp)) ? NULL : nextp;
3107 nextc->do_ops->do_ah_init(env, ah, obj, nextc, child_mode);
3110 if (S_ISDIR(child_mode)) {
3111 if (lc->ldo_dir_stripe == NULL) {
3112 OBD_ALLOC_PTR(lc->ldo_dir_stripe);
3113 if (lc->ldo_dir_stripe == NULL)
3117 LASSERT(lp != NULL);
3118 if (lp->ldo_dir_stripe == NULL) {
3119 OBD_ALLOC_PTR(lp->ldo_dir_stripe);
3120 if (lp->ldo_dir_stripe == NULL)
3124 rc = lod_cache_parent_striping(env, lp, child_mode);
3128 /* transfer defaults to new directory */
3129 if (lp->ldo_def_striping_set) {
3131 lod_object_set_pool(lc, lp->ldo_pool);
3132 lc->ldo_def_stripenr = lp->ldo_def_stripenr;
3133 lc->ldo_def_stripe_size = lp->ldo_def_stripe_size;
3134 lc->ldo_def_stripe_offset = lp->ldo_def_stripe_offset;
3135 lc->ldo_def_striping_set = 1;
3136 lc->ldo_def_striping_cached = 1;
3137 CDEBUG(D_OTHER, "inherite EA sz:%d off:%d nr:%d\n",
3138 (int)lc->ldo_def_stripe_size,
3139 (int)lc->ldo_def_stripe_offset,
3140 (int)lc->ldo_def_stripenr);
3143 /* transfer dir defaults to new directory */
3144 if (lp->ldo_dir_def_striping_set) {
3145 lc->ldo_dir_def_stripenr = lp->ldo_dir_def_stripenr;
3146 lc->ldo_dir_def_stripe_offset =
3147 lp->ldo_dir_def_stripe_offset;
3148 lc->ldo_dir_def_hash_type =
3149 lp->ldo_dir_def_hash_type;
3150 lc->ldo_dir_def_striping_set = 1;
3151 lc->ldo_dir_def_striping_cached = 1;
3152 CDEBUG(D_INFO, "inherit default EA nr:%d off:%d t%u\n",
3153 (int)lc->ldo_dir_def_stripenr,
3154 (int)lc->ldo_dir_def_stripe_offset,
3155 lc->ldo_dir_def_hash_type);
3158 /* It should always honour the specified stripes */
3159 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0) {
3160 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
3162 rc = lod_verify_md_striping(d, lum1);
3164 le32_to_cpu(lum1->lum_stripe_count) > 1) {
3165 /* Directory will be striped only if
3166 * stripe_count > 1 */
3168 le32_to_cpu(lum1->lum_stripe_count);
3169 lc->ldo_dir_stripe_offset =
3170 le32_to_cpu(lum1->lum_stripe_offset);
3171 lc->ldo_dir_hash_type =
3172 le32_to_cpu(lum1->lum_hash_type);
3173 CDEBUG(D_INFO, "set stripe EA nr:%hu off:%d\n",
3175 (int)lc->ldo_dir_stripe_offset);
3177 /* then check whether there is default stripes from parent */
3178 } else if (lp->ldo_dir_def_striping_set) {
3179 /* If there are default dir stripe from parent */
3180 lc->ldo_stripenr = lp->ldo_dir_def_stripenr;
3181 lc->ldo_dir_stripe_offset =
3182 lp->ldo_dir_def_stripe_offset;
3183 lc->ldo_dir_hash_type =
3184 lp->ldo_dir_def_hash_type;
3185 CDEBUG(D_INFO, "inherit EA nr:%hu off:%d\n",
3187 (int)lc->ldo_dir_stripe_offset);
3189 /* set default stripe for this directory */
3190 lc->ldo_stripenr = 0;
3191 lc->ldo_dir_stripe_offset = -1;
3194 CDEBUG(D_INFO, "final striping count:%hu, offset:%d\n",
3195 lc->ldo_stripenr, (int)lc->ldo_dir_stripe_offset);
3201 * if object is going to be striped over OSTs, transfer default
3202 * striping information to the child, so that we can use it
3203 * during declaration and creation
3205 if (!lod_object_will_be_striped(S_ISREG(child_mode),
3206 lu_object_fid(&child->do_lu)))
3209 * try from the parent
3211 if (likely(parent)) {
3212 lod_cache_parent_striping(env, lp, child_mode);
3214 lc->ldo_def_stripe_offset = LOV_OFFSET_DEFAULT;
3216 if (lp->ldo_def_striping_set) {
3218 lod_object_set_pool(lc, lp->ldo_pool);
3219 lc->ldo_stripenr = lp->ldo_def_stripenr;
3220 lc->ldo_stripe_size = lp->ldo_def_stripe_size;
3221 lc->ldo_def_stripe_offset = lp->ldo_def_stripe_offset;
3222 CDEBUG(D_OTHER, "striping from parent: #%d, sz %d %s\n",
3223 lc->ldo_stripenr, lc->ldo_stripe_size,
3224 lp->ldo_pool ? lp->ldo_pool : "");
3229 * if the parent doesn't provide with specific pattern, grab fs-wide one
3231 desc = &d->lod_desc;
3232 if (lc->ldo_stripenr == 0)
3233 lc->ldo_stripenr = desc->ld_default_stripe_count;
3234 if (lc->ldo_stripe_size == 0)
3235 lc->ldo_stripe_size = desc->ld_default_stripe_size;
3236 CDEBUG(D_OTHER, "final striping: # %d stripes, sz %d from %s\n",
3237 lc->ldo_stripenr, lc->ldo_stripe_size,
3238 lc->ldo_pool ? lc->ldo_pool : "");
3241 /* we do not cache stripe information for slave stripe, see
3242 * lod_xattr_set_lov_on_dir */
3243 if (lp != NULL && lp->ldo_dir_slave_stripe)
3244 lod_lov_stripe_cache_clear(lp);
3249 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
3251 * Size initialization on late striping.
3253 * Propagate the size of a truncated object to a deferred striping.
3254 * This function handles a special case when truncate was done on a
3255 * non-striped object and now while the striping is being created
3256 * we can't lose that size, so we have to propagate it to the stripes
3259 * \param[in] env execution environment
3260 * \param[in] dt object
3261 * \param[in] th transaction handle
3263 * \retval 0 on success
3264 * \retval negative if failed
3266 static int lod_declare_init_size(const struct lu_env *env,
3267 struct dt_object *dt, struct thandle *th)
3269 struct dt_object *next = dt_object_child(dt);
3270 struct lod_object *lo = lod_dt_obj(dt);
3271 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
3272 uint64_t size, offs;
3276 /* XXX: we support the simplest (RAID0) striping so far */
3277 LASSERT(lo->ldo_stripe || lo->ldo_stripenr == 0);
3278 LASSERT(lo->ldo_stripe_size > 0);
3280 rc = dt_attr_get(env, next, attr, BYPASS_CAPA);
3281 LASSERT(attr->la_valid & LA_SIZE);
3285 size = attr->la_size;
3289 /* ll_do_div64(a, b) returns a % b, and a = a / b */
3290 ll_do_div64(size, (__u64) lo->ldo_stripe_size);
3291 stripe = ll_do_div64(size, (__u64) lo->ldo_stripenr);
3293 size = size * lo->ldo_stripe_size;
3294 offs = attr->la_size;
3295 size += ll_do_div64(offs, lo->ldo_stripe_size);
3297 attr->la_valid = LA_SIZE;
3298 attr->la_size = size;
3300 rc = dt_declare_attr_set(env, lo->ldo_stripe[stripe], attr, th);
3306 * Declare creation of striped object.
3308 * The function declares creation stripes for a regular object. The function
3309 * also declares whether the stripes will be created with non-zero size if
3310 * previously size was set non-zero on the master object. If object \a dt is
3311 * not local, then only fully defined striping can be applied in \a lovea.
3312 * Otherwise \a lovea can be in the form of pattern, see lod_qos_parse_config()
3315 * \param[in] env execution environment
3316 * \param[in] dt object
3317 * \param[in] attr attributes the stripes will be created with
3318 * \param[in] lovea a buffer containing striping description
3319 * \param[in] th transaction handle
3321 * \retval 0 on success
3322 * \retval negative if failed
3324 int lod_declare_striped_object(const struct lu_env *env, struct dt_object *dt,
3325 struct lu_attr *attr,
3326 const struct lu_buf *lovea, struct thandle *th)
3328 struct lod_thread_info *info = lod_env_info(env);
3329 struct dt_object *next = dt_object_child(dt);
3330 struct lod_object *lo = lod_dt_obj(dt);
3334 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO)) {
3335 /* failed to create striping, let's reset
3336 * config so that others don't get confused */
3337 lod_object_free_striping(env, lo);
3338 GOTO(out, rc = -ENOMEM);
3341 if (!dt_object_remote(next)) {
3342 /* choose OST and generate appropriate objects */
3343 rc = lod_qos_prep_create(env, lo, attr, lovea, th);
3345 /* failed to create striping, let's reset
3346 * config so that others don't get confused */
3347 lod_object_free_striping(env, lo);
3352 * declare storage for striping data
3354 info->lti_buf.lb_len = lov_mds_md_size(lo->ldo_stripenr,
3355 lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1);
3357 /* LOD can not choose OST objects for remote objects, i.e.
3358 * stripes must be ready before that. Right now, it can only
3359 * happen during migrate, i.e. migrate process needs to create
3360 * remote regular file (mdd_migrate_create), then the migrate
3361 * process will provide stripeEA. */
3362 LASSERT(lovea != NULL);
3363 info->lti_buf = *lovea;
3366 rc = dt_declare_xattr_set(env, next, &info->lti_buf,
3367 XATTR_NAME_LOV, 0, th);
3372 * if striping is created with local object's size > 0,
3373 * we have to propagate this size to specific object
3374 * the case is possible only when local object was created previously
3376 if (dt_object_exists(next))
3377 rc = lod_declare_init_size(env, dt, th);
3384 * Implementation of dt_object_operations::do_declare_create.
3386 * The method declares creation of a new object. If the object will be striped,
3387 * then helper functions are called to find FIDs for the stripes, declare
3388 * creation of the stripes and declare initialization of the striping
3389 * information to be stored in the master object.
3391 * \see dt_object_operations::do_declare_create() in the API description
3394 static int lod_declare_object_create(const struct lu_env *env,
3395 struct dt_object *dt,
3396 struct lu_attr *attr,
3397 struct dt_allocation_hint *hint,
3398 struct dt_object_format *dof,
3401 struct dt_object *next = dt_object_child(dt);
3402 struct lod_object *lo = lod_dt_obj(dt);
3411 * first of all, we declare creation of local object
3413 rc = dt_declare_create(env, next, attr, hint, dof, th);
3417 if (dof->dof_type == DFT_SYM)
3418 dt->do_body_ops = &lod_body_lnk_ops;
3421 * it's lod_ah_init() who has decided the object will striped
3423 if (dof->dof_type == DFT_REGULAR) {
3424 /* callers don't want stripes */
3425 /* XXX: all tricky interactions with ->ah_make_hint() decided
3426 * to use striping, then ->declare_create() behaving differently
3427 * should be cleaned */
3428 if (dof->u.dof_reg.striped == 0)
3429 lo->ldo_stripenr = 0;
3430 if (lo->ldo_stripenr > 0)
3431 rc = lod_declare_striped_object(env, dt, attr,
3433 } else if (dof->dof_type == DFT_DIR) {
3434 struct seq_server_site *ss;
3436 ss = lu_site2seq(dt->do_lu.lo_dev->ld_site);
3438 /* If the parent has default stripeEA, and client
3439 * did not find it before sending create request,
3440 * then MDT will return -EREMOTE, and client will
3441 * retrieve the default stripeEA and re-create the
3444 * Note: if dah_eadata != NULL, it means creating the
3445 * striped directory with specified stripeEA, then it
3446 * should ignore the default stripeEA */
3447 if ((hint == NULL || hint->dah_eadata == NULL) &&
3448 lo->ldo_dir_stripe_offset != -1 &&
3449 lo->ldo_dir_stripe_offset != ss->ss_node_id)
3450 GOTO(out, rc = -EREMOTE);
3452 /* Orphan object (like migrating object) does not have
3453 * lod_dir_stripe, see lod_ah_init */
3454 if (lo->ldo_dir_stripe != NULL)
3455 rc = lod_declare_dir_striping_create(env, dt, attr,
3463 * Creation of a striped regular object.
3465 * The function is called to create the stripe objects for a regular
3466 * striped file. This can happen at the initial object creation or
3467 * when the caller asks LOD to do so using ->do_xattr_set() method
3468 * (so called late striping). Notice all the information are already
3469 * prepared in the form of the list of objects (ldo_stripe field).
3470 * This is done during declare phase.
3472 * \param[in] env execution environment
3473 * \param[in] dt object
3474 * \param[in] attr attributes the stripes will be created with
3475 * \param[in] dof format of stripes (see OSD API description)
3476 * \param[in] th transaction handle
3478 * \retval 0 on success
3479 * \retval negative if failed
3481 int lod_striping_create(const struct lu_env *env, struct dt_object *dt,
3482 struct lu_attr *attr, struct dt_object_format *dof,
3485 struct lod_object *lo = lod_dt_obj(dt);
3489 LASSERT(lo->ldo_striping_cached == 0);
3491 /* create all underlying objects */
3492 for (i = 0; i < lo->ldo_stripenr; i++) {
3493 LASSERT(lo->ldo_stripe[i]);
3494 rc = dt_create(env, lo->ldo_stripe[i], attr, NULL, dof, th);
3501 rc = lod_generate_and_set_lovea(env, lo, th);
3503 lo->ldo_striping_cached = 1;
3510 * Implementation of dt_object_operations::do_create.
3512 * If any of preceeding methods (like ->do_declare_create(),
3513 * ->do_ah_init(), etc) chose to create a striped object,
3514 * then this method will create the master and the stripes.
3516 * \see dt_object_operations::do_create() in the API description for details.
3518 static int lod_object_create(const struct lu_env *env, struct dt_object *dt,
3519 struct lu_attr *attr,
3520 struct dt_allocation_hint *hint,
3521 struct dt_object_format *dof, struct thandle *th)
3523 struct dt_object *next = dt_object_child(dt);
3524 struct lod_object *lo = lod_dt_obj(dt);
3528 /* create local object */
3529 rc = dt_create(env, next, attr, hint, dof, th);
3533 if (S_ISREG(dt->do_lu.lo_header->loh_attr) &&
3534 lo->ldo_stripe && dof->u.dof_reg.striped != 0)
3535 rc = lod_striping_create(env, dt, attr, dof, th);
3541 * Implementation of dt_object_operations::do_declare_destroy.
3543 * If the object is a striped directory, then the function declares reference
3544 * removal from the master object (this is an index) to the stripes and declares
3545 * destroy of all the stripes. In all the cases, it declares an intention to
3546 * destroy the object itself.
3548 * \see dt_object_operations::do_declare_destroy() in the API description
3551 static int lod_declare_object_destroy(const struct lu_env *env,
3552 struct dt_object *dt,
3555 struct dt_object *next = dt_object_child(dt);
3556 struct lod_object *lo = lod_dt_obj(dt);
3557 struct lod_thread_info *info = lod_env_info(env);
3558 char *stripe_name = info->lti_key;
3563 * load striping information, notice we don't do this when object
3564 * is being initialized as we don't need this information till
3565 * few specific cases like destroy, chown
3567 rc = lod_load_striping(env, lo);
3571 /* declare destroy for all underlying objects */
3572 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
3573 rc = next->do_ops->do_index_try(env, next,
3574 &dt_directory_features);
3578 for (i = 0; i < lo->ldo_stripenr; i++) {
3579 rc = dt_declare_ref_del(env, next, th);
3582 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3583 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
3585 rc = dt_declare_delete(env, next,
3586 (const struct dt_key *)stripe_name, th);
3592 * we declare destroy for the local object
3594 rc = dt_declare_destroy(env, next, th);
3598 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ))
3601 /* declare destroy all striped objects */
3602 for (i = 0; i < lo->ldo_stripenr; i++) {
3603 if (likely(lo->ldo_stripe[i] != NULL)) {
3604 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
3605 rc = dt_declare_ref_del(env, lo->ldo_stripe[i],
3611 rc = dt_declare_destroy(env, lo->ldo_stripe[i], th);
3621 * Implementation of dt_object_operations::do_destroy.
3623 * If the object is a striped directory, then the function removes references
3624 * from the master object (this is an index) to the stripes and destroys all
3625 * the stripes. In all the cases, the function destroys the object itself.
3627 * \see dt_object_operations::do_destroy() in the API description for details.
3629 static int lod_object_destroy(const struct lu_env *env,
3630 struct dt_object *dt, struct thandle *th)
3632 struct dt_object *next = dt_object_child(dt);
3633 struct lod_object *lo = lod_dt_obj(dt);
3634 struct lod_thread_info *info = lod_env_info(env);
3635 char *stripe_name = info->lti_key;
3640 /* destroy sub-stripe of master object */
3641 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
3642 rc = next->do_ops->do_index_try(env, next,
3643 &dt_directory_features);
3647 for (i = 0; i < lo->ldo_stripenr; i++) {
3648 rc = dt_ref_del(env, next, th);
3652 snprintf(stripe_name, sizeof(info->lti_key), DFID":%d",
3653 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)),
3656 CDEBUG(D_INFO, DFID" delete stripe %s "DFID"\n",
3657 PFID(lu_object_fid(&dt->do_lu)), stripe_name,
3658 PFID(lu_object_fid(&lo->ldo_stripe[i]->do_lu)));
3660 rc = dt_delete(env, next,
3661 (const struct dt_key *)stripe_name,
3667 rc = dt_destroy(env, next, th);
3671 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ))
3674 /* destroy all striped objects */
3675 for (i = 0; i < lo->ldo_stripenr; i++) {
3676 if (likely(lo->ldo_stripe[i] != NULL) &&
3677 (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_SPEOBJ) ||
3678 i == cfs_fail_val)) {
3679 if (S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
3680 dt_write_lock(env, lo->ldo_stripe[i],
3682 rc = dt_ref_del(env, lo->ldo_stripe[i], th);
3683 dt_write_unlock(env, lo->ldo_stripe[i]);
3688 rc = dt_destroy(env, lo->ldo_stripe[i], th);
3698 * Implementation of dt_object_operations::do_declare_ref_add.
3700 * \see dt_object_operations::do_declare_ref_add() in the API description
3703 static int lod_declare_ref_add(const struct lu_env *env,
3704 struct dt_object *dt, struct thandle *th)
3706 return dt_declare_ref_add(env, dt_object_child(dt), th);
3710 * Implementation of dt_object_operations::do_ref_add.
3712 * \see dt_object_operations::do_ref_add() in the API description for details.
3714 static int lod_ref_add(const struct lu_env *env,
3715 struct dt_object *dt, struct thandle *th)
3717 return dt_ref_add(env, dt_object_child(dt), th);
3721 * Implementation of dt_object_operations::do_declare_ref_del.
3723 * \see dt_object_operations::do_declare_ref_del() in the API description
3726 static int lod_declare_ref_del(const struct lu_env *env,
3727 struct dt_object *dt, struct thandle *th)
3729 return dt_declare_ref_del(env, dt_object_child(dt), th);
3733 * Implementation of dt_object_operations::do_ref_del
3735 * \see dt_object_operations::do_ref_del() in the API description for details.
3737 static int lod_ref_del(const struct lu_env *env,
3738 struct dt_object *dt, struct thandle *th)
3740 return dt_ref_del(env, dt_object_child(dt), th);
3744 * Implementation of dt_object_operations::do_capa_get.
3746 * \see dt_object_operations::do_capa_get() in the API description for details.
3748 static struct obd_capa *lod_capa_get(const struct lu_env *env,
3749 struct dt_object *dt,
3750 struct lustre_capa *old, __u64 opc)
3752 return dt_capa_get(env, dt_object_child(dt), old, opc);
3756 * Implementation of dt_object_operations::do_object_sync.
3758 * \see dt_object_operations::do_object_sync() in the API description
3761 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt,
3762 __u64 start, __u64 end)
3764 return dt_object_sync(env, dt_object_child(dt), start, end);
3767 struct lod_slave_locks {
3769 struct lustre_handle lsl_handle[0];
3773 * Release LDLM locks on the stripes of a striped directory.
3775 * Iterates over all the locks taken on the stripe objects and
3776 * release them using ->do_object_unlock() method.
3778 * \param[in] env execution environment
3779 * \param[in] dt striped object
3780 * \param[in] einfo lock description
3781 * \param[in] policy data describing requested lock
3783 * \retval 0 on success
3784 * \retval negative if failed
3786 static int lod_object_unlock_internal(const struct lu_env *env,
3787 struct dt_object *dt,
3788 struct ldlm_enqueue_info *einfo,
3789 ldlm_policy_data_t *policy)
3791 struct lod_object *lo = lod_dt_obj(dt);
3792 struct lod_slave_locks *slave_locks = einfo->ei_cbdata;
3797 if (slave_locks == NULL)
3800 for (i = 1; i < slave_locks->lsl_lock_count; i++) {
3801 if (lustre_handle_is_used(&slave_locks->lsl_handle[i])) {
3804 einfo->ei_cbdata = &slave_locks->lsl_handle[i];
3805 rc1 = dt_object_unlock(env, lo->ldo_stripe[i], einfo,
3808 rc = rc == 0 ? rc1 : rc;
3816 * Implementation of dt_object_operations::do_object_unlock.
3818 * Used to release LDLM lock(s).
3820 * \see dt_object_operations::do_object_unlock() in the API description
3823 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
3824 struct ldlm_enqueue_info *einfo,
3825 union ldlm_policy_data *policy)
3827 struct lod_object *lo = lod_dt_obj(dt);
3828 struct lod_slave_locks *slave_locks = einfo->ei_cbdata;
3829 int slave_locks_size;
3833 if (slave_locks == NULL)
3836 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3839 rc = lod_load_striping(env, lo);
3843 /* Note: for remote lock for single stripe dir, MDT will cancel
3844 * the lock by lockh directly */
3845 if (lo->ldo_stripenr <= 1 && dt_object_remote(dt_object_child(dt)))
3848 /* Only cancel slave lock for striped dir */
3849 rc = lod_object_unlock_internal(env, dt, einfo, policy);
3851 slave_locks_size = sizeof(*slave_locks) + slave_locks->lsl_lock_count *
3852 sizeof(slave_locks->lsl_handle[0]);
3853 OBD_FREE(slave_locks, slave_locks_size);
3854 einfo->ei_cbdata = NULL;
3860 * Implementation of dt_object_operations::do_object_lock.
3862 * Used to get LDLM lock on the non-striped and striped objects.
3864 * \see dt_object_operations::do_object_lock() in the API description
3867 static int lod_object_lock(const struct lu_env *env,
3868 struct dt_object *dt,
3869 struct lustre_handle *lh,
3870 struct ldlm_enqueue_info *einfo,
3871 union ldlm_policy_data *policy)
3873 struct lod_object *lo = lod_dt_obj(dt);
3876 int slave_locks_size;
3877 struct lod_slave_locks *slave_locks = NULL;
3880 /* remote object lock */
3881 if (!einfo->ei_enq_slave) {
3882 LASSERT(dt_object_remote(dt));
3883 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
3887 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
3890 rc = lod_load_striping(env, lo);
3895 if (lo->ldo_stripenr <= 1)
3898 slave_locks_size = sizeof(*slave_locks) + lo->ldo_stripenr *
3899 sizeof(slave_locks->lsl_handle[0]);
3900 /* Freed in lod_object_unlock */
3901 OBD_ALLOC(slave_locks, slave_locks_size);
3902 if (slave_locks == NULL)
3904 slave_locks->lsl_lock_count = lo->ldo_stripenr;
3906 /* striped directory lock */
3907 for (i = 1; i < lo->ldo_stripenr; i++) {
3908 struct lustre_handle lockh;
3909 struct ldlm_res_id *res_id;
3911 res_id = &lod_env_info(env)->lti_res_id;
3912 fid_build_reg_res_name(lu_object_fid(&lo->ldo_stripe[i]->do_lu),
3914 einfo->ei_res_id = res_id;
3916 LASSERT(lo->ldo_stripe[i]);
3917 rc = dt_object_lock(env, lo->ldo_stripe[i], &lockh, einfo,
3921 slave_locks->lsl_handle[i] = lockh;
3924 einfo->ei_cbdata = slave_locks;
3927 if (rc != 0 && slave_locks != NULL) {
3928 einfo->ei_cbdata = slave_locks;
3929 lod_object_unlock_internal(env, dt, einfo, policy);
3930 OBD_FREE(slave_locks, slave_locks_size);
3931 einfo->ei_cbdata = NULL;
3937 struct dt_object_operations lod_obj_ops = {
3938 .do_read_lock = lod_object_read_lock,
3939 .do_write_lock = lod_object_write_lock,
3940 .do_read_unlock = lod_object_read_unlock,
3941 .do_write_unlock = lod_object_write_unlock,
3942 .do_write_locked = lod_object_write_locked,
3943 .do_attr_get = lod_attr_get,
3944 .do_declare_attr_set = lod_declare_attr_set,
3945 .do_attr_set = lod_attr_set,
3946 .do_xattr_get = lod_xattr_get,
3947 .do_declare_xattr_set = lod_declare_xattr_set,
3948 .do_xattr_set = lod_xattr_set,
3949 .do_declare_xattr_del = lod_declare_xattr_del,
3950 .do_xattr_del = lod_xattr_del,
3951 .do_xattr_list = lod_xattr_list,
3952 .do_ah_init = lod_ah_init,
3953 .do_declare_create = lod_declare_object_create,
3954 .do_create = lod_object_create,
3955 .do_declare_destroy = lod_declare_object_destroy,
3956 .do_destroy = lod_object_destroy,
3957 .do_index_try = lod_index_try,
3958 .do_declare_ref_add = lod_declare_ref_add,
3959 .do_ref_add = lod_ref_add,
3960 .do_declare_ref_del = lod_declare_ref_del,
3961 .do_ref_del = lod_ref_del,
3962 .do_capa_get = lod_capa_get,
3963 .do_object_sync = lod_object_sync,
3964 .do_object_lock = lod_object_lock,
3965 .do_object_unlock = lod_object_unlock,
3969 * Implementation of dt_body_operations::dbo_read.
3971 * \see dt_body_operations::dbo_read() in the API description for details.
3973 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
3974 struct lu_buf *buf, loff_t *pos,
3975 struct lustre_capa *capa)
3977 struct dt_object *next = dt_object_child(dt);
3978 return next->do_body_ops->dbo_read(env, next, buf, pos, capa);
3982 * Implementation of dt_body_operations::dbo_declare_write.
3984 * \see dt_body_operations::dbo_declare_write() in the API description
3987 static ssize_t lod_declare_write(const struct lu_env *env,
3988 struct dt_object *dt,
3989 const struct lu_buf *buf, loff_t pos,
3992 return dt_declare_record_write(env, dt_object_child(dt),
3997 * Implementation of dt_body_operations::dbo_write.
3999 * \see dt_body_operations::dbo_write() in the API description for details.
4001 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
4002 const struct lu_buf *buf, loff_t *pos,
4003 struct thandle *th, struct lustre_capa *capa, int iq)
4005 struct dt_object *next = dt_object_child(dt);
4007 return next->do_body_ops->dbo_write(env, next, buf, pos, th, capa, iq);
4010 static const struct dt_body_operations lod_body_lnk_ops = {
4011 .dbo_read = lod_read,
4012 .dbo_declare_write = lod_declare_write,
4013 .dbo_write = lod_write
4017 * Implementation of lu_object_operations::loo_object_init.
4019 * The function determines the type and the index of the target device using
4020 * sequence of the object's FID. Then passes control down to the
4021 * corresponding device:
4022 * OSD for the local objects, OSP for remote
4024 * \see lu_object_operations::loo_object_init() in the API description
4027 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
4028 const struct lu_object_conf *conf)
4030 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
4031 struct lu_device *cdev = NULL;
4032 struct lu_object *cobj;
4033 struct lod_tgt_descs *ltd = NULL;
4034 struct lod_tgt_desc *tgt;
4036 int type = LU_SEQ_RANGE_ANY;
4040 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
4042 /* Note: Sometimes, it will Return EAGAIN here, see
4043 * ptrlpc_import_delay_req(), which might confuse
4044 * lu_object_find_at() and make it wait there incorrectly.
4045 * so we convert it to EIO here.*/
4052 if (type == LU_SEQ_RANGE_MDT &&
4053 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
4054 cdev = &lod->lod_child->dd_lu_dev;
4055 } else if (type == LU_SEQ_RANGE_MDT) {
4056 ltd = &lod->lod_mdt_descs;
4058 } else if (type == LU_SEQ_RANGE_OST) {
4059 ltd = &lod->lod_ost_descs;
4066 if (ltd->ltd_tgts_size > idx &&
4067 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
4068 tgt = LTD_TGT(ltd, idx);
4070 LASSERT(tgt != NULL);
4071 LASSERT(tgt->ltd_tgt != NULL);
4073 cdev = &(tgt->ltd_tgt->dd_lu_dev);
4075 lod_putref(lod, ltd);
4078 if (unlikely(cdev == NULL))
4081 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
4082 if (unlikely(cobj == NULL))
4085 lu_object_add(lo, cobj);
4092 * Release resources associated with striping.
4094 * If the object is striped (regular or directory), then release
4095 * the stripe objects references and free the ldo_stripe array.
4097 * \param[in] env execution environment
4098 * \param[in] lo object
4100 void lod_object_free_striping(const struct lu_env *env, struct lod_object *lo)
4104 if (lo->ldo_dir_stripe != NULL) {
4105 OBD_FREE_PTR(lo->ldo_dir_stripe);
4106 lo->ldo_dir_stripe = NULL;
4109 if (lo->ldo_stripe) {
4110 LASSERT(lo->ldo_stripes_allocated > 0);
4112 for (i = 0; i < lo->ldo_stripenr; i++) {
4113 if (lo->ldo_stripe[i])
4114 lu_object_put(env, &lo->ldo_stripe[i]->do_lu);
4117 i = sizeof(struct dt_object *) * lo->ldo_stripes_allocated;
4118 OBD_FREE(lo->ldo_stripe, i);
4119 lo->ldo_stripe = NULL;
4120 lo->ldo_stripes_allocated = 0;
4122 lo->ldo_striping_cached = 0;
4123 lo->ldo_stripenr = 0;
4124 lo->ldo_pattern = 0;
4128 * Implementation of lu_object_operations::loo_object_start.
4130 * \see lu_object_operations::loo_object_start() in the API description
4133 static int lod_object_start(const struct lu_env *env, struct lu_object *o)
4135 if (S_ISLNK(o->lo_header->loh_attr & S_IFMT))
4136 lu2lod_obj(o)->ldo_obj.do_body_ops = &lod_body_lnk_ops;
4141 * Implementation of lu_object_operations::loo_object_free.
4143 * \see lu_object_operations::loo_object_free() in the API description
4146 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
4148 struct lod_object *mo = lu2lod_obj(o);
4151 * release all underlying object pinned
4154 lod_object_free_striping(env, mo);
4156 lod_object_set_pool(mo, NULL);
4159 OBD_SLAB_FREE_PTR(mo, lod_object_kmem);
4163 * Implementation of lu_object_operations::loo_object_release.
4165 * \see lu_object_operations::loo_object_release() in the API description
4168 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
4170 /* XXX: shouldn't we release everything here in case if object
4171 * creation failed before? */
4175 * Implementation of lu_object_operations::loo_object_print.
4177 * \see lu_object_operations::loo_object_print() in the API description
4180 static int lod_object_print(const struct lu_env *env, void *cookie,
4181 lu_printer_t p, const struct lu_object *l)
4183 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
4185 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
4188 struct lu_object_operations lod_lu_obj_ops = {
4189 .loo_object_init = lod_object_init,
4190 .loo_object_start = lod_object_start,
4191 .loo_object_free = lod_object_free,
4192 .loo_object_release = lod_object_release,
4193 .loo_object_print = lod_object_print,