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, Intel Corporation.
29 * lustre/lod/lod_object.c
31 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
35 # define EXPORT_SYMTAB
37 #define DEBUG_SUBSYSTEM S_MDS
40 #include <obd_class.h>
41 #include <lustre_ver.h>
42 #include <obd_support.h>
43 #include <lprocfs_status.h>
45 #include <lustre_fid.h>
46 #include <lustre_param.h>
47 #include <lustre_fid.h>
50 #include "lod_internal.h"
52 extern cfs_mem_cache_t *lod_object_kmem;
53 static const struct dt_body_operations lod_body_lnk_ops;
55 static int lod_index_lookup(const struct lu_env *env, struct dt_object *dt,
56 struct dt_rec *rec, const struct dt_key *key,
57 struct lustre_capa *capa)
59 struct dt_object *next = dt_object_child(dt);
60 return next->do_index_ops->dio_lookup(env, next, rec, key, capa);
63 static int lod_declare_index_insert(const struct lu_env *env,
65 const struct dt_rec *rec,
66 const struct dt_key *key,
67 struct thandle *handle)
69 return dt_declare_insert(env, dt_object_child(dt), rec, key, handle);
72 static int lod_index_insert(const struct lu_env *env,
74 const struct dt_rec *rec,
75 const struct dt_key *key,
77 struct lustre_capa *capa,
80 return dt_insert(env, dt_object_child(dt), rec, key, th, capa, ign);
83 static int lod_declare_index_delete(const struct lu_env *env,
85 const struct dt_key *key,
88 return dt_declare_delete(env, dt_object_child(dt), key, th);
91 static int lod_index_delete(const struct lu_env *env,
93 const struct dt_key *key,
95 struct lustre_capa *capa)
97 return dt_delete(env, dt_object_child(dt), key, th, capa);
100 static struct dt_it *lod_it_init(const struct lu_env *env,
101 struct dt_object *dt, __u32 attr,
102 struct lustre_capa *capa)
104 struct dt_object *next = dt_object_child(dt);
105 struct lod_it *it = &lod_env_info(env)->lti_it;
106 struct dt_it *it_next;
109 it_next = next->do_index_ops->dio_it.init(env, next, attr, capa);
113 /* currently we do not use more than one iterator per thread
114 * so we store it in thread info. if at some point we need
115 * more active iterators in a single thread, we can allocate
117 LASSERT(it->lit_obj == NULL);
119 it->lit_it = it_next;
122 return (struct dt_it *)it;
125 #define LOD_CHECK_IT(env, it) \
127 /* IT is supposed to be in thread info always */ \
128 LASSERT((it) == &lod_env_info(env)->lti_it); \
129 LASSERT((it)->lit_obj != NULL); \
130 LASSERT((it)->lit_it != NULL); \
133 void lod_it_fini(const struct lu_env *env, struct dt_it *di)
135 struct lod_it *it = (struct lod_it *)di;
137 LOD_CHECK_IT(env, it);
138 it->lit_obj->do_index_ops->dio_it.fini(env, it->lit_it);
140 /* the iterator not in use any more */
145 int lod_it_get(const struct lu_env *env, struct dt_it *di,
146 const struct dt_key *key)
148 const struct lod_it *it = (const struct lod_it *)di;
150 LOD_CHECK_IT(env, it);
151 return it->lit_obj->do_index_ops->dio_it.get(env, it->lit_it, key);
154 void lod_it_put(const struct lu_env *env, struct dt_it *di)
156 struct lod_it *it = (struct lod_it *)di;
158 LOD_CHECK_IT(env, it);
159 return it->lit_obj->do_index_ops->dio_it.put(env, it->lit_it);
162 int lod_it_next(const struct lu_env *env, struct dt_it *di)
164 struct lod_it *it = (struct lod_it *)di;
166 LOD_CHECK_IT(env, it);
167 return it->lit_obj->do_index_ops->dio_it.next(env, it->lit_it);
170 struct dt_key *lod_it_key(const struct lu_env *env, const struct dt_it *di)
172 const struct lod_it *it = (const struct lod_it *)di;
174 LOD_CHECK_IT(env, it);
175 return it->lit_obj->do_index_ops->dio_it.key(env, it->lit_it);
178 int lod_it_key_size(const struct lu_env *env, const struct dt_it *di)
180 struct lod_it *it = (struct lod_it *)di;
182 LOD_CHECK_IT(env, it);
183 return it->lit_obj->do_index_ops->dio_it.key_size(env, it->lit_it);
186 int lod_it_rec(const struct lu_env *env, const struct dt_it *di,
187 struct dt_rec *rec, __u32 attr)
189 const struct lod_it *it = (const struct lod_it *)di;
191 LOD_CHECK_IT(env, it);
192 return it->lit_obj->do_index_ops->dio_it.rec(env, it->lit_it, rec, attr);
195 __u64 lod_it_store(const struct lu_env *env, const struct dt_it *di)
197 const struct lod_it *it = (const struct lod_it *)di;
199 LOD_CHECK_IT(env, it);
200 return it->lit_obj->do_index_ops->dio_it.store(env, it->lit_it);
203 int lod_it_load(const struct lu_env *env, const struct dt_it *di, __u64 hash)
205 const struct lod_it *it = (const struct lod_it *)di;
207 LOD_CHECK_IT(env, it);
208 return it->lit_obj->do_index_ops->dio_it.load(env, it->lit_it, hash);
211 int lod_it_key_rec(const struct lu_env *env, const struct dt_it *di,
214 const struct lod_it *it = (const struct lod_it *)di;
216 LOD_CHECK_IT(env, it);
217 return it->lit_obj->do_index_ops->dio_it.key_rec(env, it->lit_it, key_rec);
220 static struct dt_index_operations lod_index_ops = {
221 .dio_lookup = lod_index_lookup,
222 .dio_declare_insert = lod_declare_index_insert,
223 .dio_insert = lod_index_insert,
224 .dio_declare_delete = lod_declare_index_delete,
225 .dio_delete = lod_index_delete,
233 .key_size = lod_it_key_size,
235 .store = lod_it_store,
237 .key_rec = lod_it_key_rec,
241 static void lod_object_read_lock(const struct lu_env *env,
242 struct dt_object *dt, unsigned role)
244 dt_read_lock(env, dt_object_child(dt), role);
247 static void lod_object_write_lock(const struct lu_env *env,
248 struct dt_object *dt, unsigned role)
250 dt_write_lock(env, dt_object_child(dt), role);
253 static void lod_object_read_unlock(const struct lu_env *env,
254 struct dt_object *dt)
256 dt_read_unlock(env, dt_object_child(dt));
259 static void lod_object_write_unlock(const struct lu_env *env,
260 struct dt_object *dt)
262 dt_write_unlock(env, dt_object_child(dt));
265 static int lod_object_write_locked(const struct lu_env *env,
266 struct dt_object *dt)
268 return dt_write_locked(env, dt_object_child(dt));
271 static int lod_attr_get(const struct lu_env *env,
272 struct dt_object *dt,
273 struct lu_attr *attr,
274 struct lustre_capa *capa)
276 return dt_attr_get(env, dt_object_child(dt), attr, capa);
279 static int lod_declare_attr_set(const struct lu_env *env,
280 struct dt_object *dt,
281 const struct lu_attr *attr,
282 struct thandle *handle)
284 struct dt_object *next = dt_object_child(dt);
285 struct lod_object *lo = lod_dt_obj(dt);
290 * declare setattr on the local object
292 rc = dt_declare_attr_set(env, next, attr, handle);
297 * load striping information, notice we don't do this when object
298 * is being initialized as we don't need this information till
299 * few specific cases like destroy, chown
301 rc = lod_load_striping(env, lo);
306 * if object is striped declare changes on the stripes
308 LASSERT(lo->ldo_stripe || lo->ldo_stripenr == 0);
309 for (i = 0; i < lo->ldo_stripenr; i++) {
310 LASSERT(lo->ldo_stripe[i]);
311 rc = dt_declare_attr_set(env, lo->ldo_stripe[i], attr, handle);
313 CERROR("failed declaration: %d\n", rc);
321 static int lod_attr_set(const struct lu_env *env,
322 struct dt_object *dt,
323 const struct lu_attr *attr,
324 struct thandle *handle,
325 struct lustre_capa *capa)
327 struct dt_object *next = dt_object_child(dt);
328 struct lod_object *lo = lod_dt_obj(dt);
333 * apply changes to the local object
335 rc = dt_attr_set(env, next, attr, handle, capa);
340 * if object is striped, apply changes to all the stripes
342 LASSERT(lo->ldo_stripe || lo->ldo_stripenr == 0);
343 for (i = 0; i < lo->ldo_stripenr; i++) {
344 LASSERT(lo->ldo_stripe[i]);
345 rc = dt_attr_set(env, lo->ldo_stripe[i], attr, handle, capa);
347 CERROR("failed declaration: %d\n", rc);
355 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
356 struct lu_buf *buf, const char *name,
357 struct lustre_capa *capa)
359 struct lod_thread_info *info = lod_env_info(env);
360 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
364 rc = dt_xattr_get(env, dt_object_child(dt), buf, name, capa);
365 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
369 * lod returns default striping on the real root of the device
370 * this is like the root stores default striping for the whole
371 * filesystem. historically we've been using a different approach
372 * and store it in the config.
374 dt_root_get(env, dev->lod_child, &info->lti_fid);
375 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
377 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
378 struct lov_user_md *lum = buf->lb_buf;
379 struct lov_desc *desc = &dev->lod_desc;
381 if (buf->lb_buf == NULL) {
382 rc = sizeof(struct lov_user_md_v1);
383 } else if (buf->lb_len >= sizeof(struct lov_user_md_v1)) {
384 lum->lmm_magic = LOV_USER_MAGIC_V1;
385 lum->lmm_object_seq = FID_SEQ_LOV_DEFAULT;
386 lum->lmm_pattern = desc->ld_pattern;
387 lum->lmm_stripe_size = desc->ld_default_stripe_size;
388 lum->lmm_stripe_count = desc->ld_default_stripe_count;
389 lum->lmm_stripe_offset = desc->ld_default_stripe_offset;
390 rc = sizeof(struct lov_user_md_v1);
400 * LOV xattr is a storage for striping, and LOD owns this xattr.
401 * but LOD allows others to control striping to some extent
403 * - to set new defined striping
404 * - to set new semi-defined striping
405 * - number of stripes is defined
406 * - number of stripes + osts are defined
409 static int lod_declare_xattr_set(const struct lu_env *env,
410 struct dt_object *dt,
411 const struct lu_buf *buf,
412 const char *name, int fl,
415 struct dt_object *next = dt_object_child(dt);
416 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
422 * allow to declare predefined striping on a new (!mode) object
423 * which is supposed to be replay of regular file creation
424 * (when LOV setting is declared)
425 * LU_XATTR_REPLACE is set to indicate a layout swap
427 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
428 if ((S_ISREG(mode) || !mode) && !strcmp(name, XATTR_NAME_LOV) &&
429 !(fl & LU_XATTR_REPLACE)) {
431 * this is a request to manipulate object's striping
433 if (dt_object_exists(dt)) {
434 rc = dt_attr_get(env, next, attr, BYPASS_CAPA);
438 memset(attr, 0, sizeof(attr));
439 attr->la_valid = LA_TYPE | LA_MODE;
440 attr->la_mode = S_IFREG;
442 rc = lod_declare_striped_object(env, dt, attr, buf, th);
447 rc = dt_declare_xattr_set(env, next, buf, name, fl, th);
452 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
453 struct dt_object *dt,
454 const struct lu_buf *buf,
455 const char *name, int fl,
457 struct lustre_capa *capa)
459 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
460 struct dt_object *next = dt_object_child(dt);
461 struct lod_object *l = lod_dt_obj(dt);
462 struct lov_user_md_v1 *lum;
463 struct lov_user_md_v3 *v3 = NULL;
467 LASSERT(l->ldo_stripe == NULL);
468 l->ldo_striping_cached = 0;
469 l->ldo_def_striping_set = 0;
470 lod_object_set_pool(l, NULL);
471 l->ldo_def_stripe_size = 0;
472 l->ldo_def_stripenr = 0;
475 LASSERT(buf->lb_buf);
478 rc = lod_verify_striping(d, buf, 0);
482 if (lum->lmm_magic == LOV_USER_MAGIC_V3)
485 /* if { size, offset, count } = { 0, -1, 0 } and no pool
486 * (i.e. all default values specified) then delete default
487 * striping from dir. */
489 "set default striping: sz %u # %u offset %d %s %s\n",
490 (unsigned)lum->lmm_stripe_size,
491 (unsigned)lum->lmm_stripe_count,
492 (int)lum->lmm_stripe_offset,
493 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
495 if (LOVEA_DELETE_VALUES((lum->lmm_stripe_size),
496 (lum->lmm_stripe_count),
497 (lum->lmm_stripe_offset)) &&
498 lum->lmm_magic == LOV_USER_MAGIC_V1) {
499 rc = dt_xattr_del(env, next, name, th, capa);
503 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
509 static int lod_xattr_set(const struct lu_env *env,
510 struct dt_object *dt, const struct lu_buf *buf,
511 const char *name, int fl, struct thandle *th,
512 struct lustre_capa *capa)
514 struct dt_object *next = dt_object_child(dt);
519 attr = dt->do_lu.lo_header->loh_attr & S_IFMT;
521 if (strncmp(name, XATTR_NAME_LOV, strlen(XATTR_NAME_LOV)) == 0)
522 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name,
525 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
527 } else if (S_ISREG(attr) && !strcmp(name, XATTR_NAME_LOV)) {
528 /* in case of lov EA swap, just set it
529 * if not, it is a replay so check striping match what we
530 * already have during req replay, declare_xattr_set()
531 * defines striping, then create() does the work
533 if (fl & LU_XATTR_REPLACE)
534 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
536 rc = lod_striping_create(env, dt, NULL, NULL, th);
540 * behave transparantly for all other EAs
542 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
548 static int lod_declare_xattr_del(const struct lu_env *env,
549 struct dt_object *dt, const char *name,
552 return dt_declare_xattr_del(env, dt_object_child(dt), name, th);
555 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
556 const char *name, struct thandle *th,
557 struct lustre_capa *capa)
559 return dt_xattr_del(env, dt_object_child(dt), name, th, capa);
562 static int lod_xattr_list(const struct lu_env *env,
563 struct dt_object *dt, struct lu_buf *buf,
564 struct lustre_capa *capa)
566 return dt_xattr_list(env, dt_object_child(dt), buf, capa);
569 int lod_object_set_pool(struct lod_object *o, char *pool)
574 len = strlen(o->ldo_pool);
575 OBD_FREE(o->ldo_pool, len + 1);
580 OBD_ALLOC(o->ldo_pool, len + 1);
581 if (o->ldo_pool == NULL)
583 strcpy(o->ldo_pool, pool);
588 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
590 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
593 static int lod_cache_parent_striping(const struct lu_env *env,
594 struct lod_object *lp)
596 struct lov_user_md_v1 *v1 = NULL;
597 struct lov_user_md_v3 *v3 = NULL;
601 /* dt_ah_init() is called from MDD without parent being write locked
603 dt_write_lock(env, dt_object_child(&lp->ldo_obj), 0);
604 if (lp->ldo_striping_cached)
605 GOTO(unlock, rc = 0);
607 rc = lod_get_lov_ea(env, lp);
611 if (rc < sizeof(struct lov_user_md)) {
612 /* don't lookup for non-existing or invalid striping */
613 lp->ldo_def_striping_set = 0;
614 lp->ldo_striping_cached = 1;
615 lp->ldo_def_stripe_size = 0;
616 lp->ldo_def_stripenr = 0;
617 lp->ldo_def_stripe_offset = (typeof(v1->lmm_stripe_offset))(-1);
618 GOTO(unlock, rc = 0);
621 v1 = (struct lov_user_md_v1 *)lod_env_info(env)->lti_ea_store;
622 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1))
623 lustre_swab_lov_user_md_v1(v1);
624 else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3))
625 lustre_swab_lov_user_md_v3(v3);
627 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1)
628 GOTO(unlock, rc = 0);
630 if (v1->lmm_pattern != LOV_PATTERN_RAID0 && v1->lmm_pattern != 0)
631 GOTO(unlock, rc = 0);
633 lp->ldo_def_stripenr = v1->lmm_stripe_count;
634 lp->ldo_def_stripe_size = v1->lmm_stripe_size;
635 lp->ldo_def_stripe_offset = v1->lmm_stripe_offset;
636 lp->ldo_striping_cached = 1;
637 lp->ldo_def_striping_set = 1;
639 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
640 /* XXX: sanity check here */
641 v3 = (struct lov_user_md_v3 *) v1;
642 if (v3->lmm_pool_name[0])
643 lod_object_set_pool(lp, v3->lmm_pool_name);
646 CDEBUG(D_OTHER, "def. striping: # %d, sz %d, off %d %s%s on "DFID"\n",
647 lp->ldo_def_stripenr, lp->ldo_def_stripe_size,
648 lp->ldo_def_stripe_offset, v3 ? "from " : "",
649 v3 ? lp->ldo_pool : "", PFID(lu_object_fid(&lp->ldo_obj.do_lu)));
653 dt_write_unlock(env, dt_object_child(&lp->ldo_obj));
658 * used to transfer default striping data to the object being created
660 static void lod_ah_init(const struct lu_env *env,
661 struct dt_allocation_hint *ah,
662 struct dt_object *parent,
663 struct dt_object *child,
664 cfs_umode_t child_mode)
666 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
667 struct dt_object *nextp = NULL;
668 struct dt_object *nextc;
669 struct lod_object *lp = NULL;
670 struct lod_object *lc;
671 struct lov_desc *desc;
676 if (likely(parent)) {
677 nextp = dt_object_child(parent);
678 lp = lod_dt_obj(parent);
681 nextc = dt_object_child(child);
682 lc = lod_dt_obj(child);
684 LASSERT(lc->ldo_stripenr == 0);
685 LASSERT(lc->ldo_stripe == NULL);
688 * local object may want some hints
689 * in case of late striping creation, ->ah_init()
690 * can be called with local object existing
692 if (!dt_object_exists(nextc))
693 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
695 if (S_ISDIR(child_mode)) {
696 if (lp->ldo_striping_cached == 0) {
697 /* we haven't tried to get default striping for
698 * the directory yet, let's cache it in the object */
699 lod_cache_parent_striping(env, lp);
701 /* transfer defaults to new directory */
702 if (lp->ldo_striping_cached) {
704 lod_object_set_pool(lc, lp->ldo_pool);
705 lc->ldo_def_stripenr = lp->ldo_def_stripenr;
706 lc->ldo_def_stripe_size = lp->ldo_def_stripe_size;
707 lc->ldo_def_stripe_offset = lp->ldo_def_stripe_offset;
708 lc->ldo_striping_cached = 1;
709 lc->ldo_def_striping_set = 1;
710 CDEBUG(D_OTHER, "inherite striping defaults\n");
716 * if object is going to be striped over OSTs, transfer default
717 * striping information to the child, so that we can use it
718 * during declaration and creation
720 if (!lod_object_will_be_striped(S_ISREG(child_mode),
721 lu_object_fid(&child->do_lu)))
725 * try from the parent
727 if (likely(parent)) {
728 if (lp->ldo_striping_cached == 0) {
729 /* we haven't tried to get default striping for
730 * the directory yet, let's cache it in the object */
731 lod_cache_parent_striping(env, lp);
734 lc->ldo_def_stripe_offset = (__u16) -1;
736 if (lp->ldo_def_striping_set) {
738 lod_object_set_pool(lc, lp->ldo_pool);
739 lc->ldo_stripenr = lp->ldo_def_stripenr;
740 lc->ldo_stripe_size = lp->ldo_def_stripe_size;
741 lc->ldo_def_stripe_offset = lp->ldo_def_stripe_offset;
742 CDEBUG(D_OTHER, "striping from parent: #%d, sz %d %s\n",
743 lc->ldo_stripenr, lc->ldo_stripe_size,
744 lp->ldo_pool ? lp->ldo_pool : "");
749 * if the parent doesn't provide with specific pattern, grab fs-wide one
752 if (lc->ldo_stripenr == 0)
753 lc->ldo_stripenr = desc->ld_default_stripe_count;
754 if (lc->ldo_stripe_size == 0)
755 lc->ldo_stripe_size = desc->ld_default_stripe_size;
756 CDEBUG(D_OTHER, "final striping: # %d stripes, sz %d from %s\n",
757 lc->ldo_stripenr, lc->ldo_stripe_size,
758 lc->ldo_pool ? lc->ldo_pool : "");
763 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
765 * this function handles a special case when truncate was done
766 * on a stripeless object and now striping is being created
767 * we can't lose that size, so we have to propagate it to newly
770 static int lod_declare_init_size(const struct lu_env *env,
771 struct dt_object *dt, struct thandle *th)
773 struct dt_object *next = dt_object_child(dt);
774 struct lod_object *lo = lod_dt_obj(dt);
775 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
780 /* XXX: we support the simplest (RAID0) striping so far */
781 LASSERT(lo->ldo_stripe || lo->ldo_stripenr == 0);
782 LASSERT(lo->ldo_stripe_size > 0);
784 rc = dt_attr_get(env, next, attr, BYPASS_CAPA);
785 LASSERT(attr->la_valid & LA_SIZE);
789 size = attr->la_size;
793 /* ll_do_div64(a, b) returns a % b, and a = a / b */
794 ll_do_div64(size, (__u64) lo->ldo_stripe_size);
795 stripe = ll_do_div64(size, (__u64) lo->ldo_stripenr);
797 size = size * lo->ldo_stripe_size;
798 offs = attr->la_size;
799 size += ll_do_div64(offs, lo->ldo_stripe_size);
801 attr->la_valid = LA_SIZE;
802 attr->la_size = size;
804 rc = dt_declare_attr_set(env, lo->ldo_stripe[stripe], attr, th);
811 * Create declaration of striped object
813 int lod_declare_striped_object(const struct lu_env *env, struct dt_object *dt,
814 struct lu_attr *attr,
815 const struct lu_buf *lovea, struct thandle *th)
817 struct lod_thread_info *info = lod_env_info(env);
818 struct dt_object *next = dt_object_child(dt);
819 struct lod_object *lo = lod_dt_obj(dt);
823 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO)) {
824 /* failed to create striping, let's reset
825 * config so that others don't get confused */
826 lod_object_free_striping(env, lo);
827 GOTO(out, rc = -ENOMEM);
830 /* choose OST and generate appropriate objects */
831 rc = lod_qos_prep_create(env, lo, attr, lovea, th);
833 /* failed to create striping, let's reset
834 * config so that others don't get confused */
835 lod_object_free_striping(env, lo);
840 * declare storage for striping data
842 info->lti_buf.lb_len = lov_mds_md_size(lo->ldo_stripenr,
843 lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1);
844 rc = dt_declare_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV,
850 * if striping is created with local object's size > 0,
851 * we have to propagate this size to specific object
852 * the case is possible only when local object was created previously
854 if (dt_object_exists(next))
855 rc = lod_declare_init_size(env, dt, th);
861 static int lod_declare_object_create(const struct lu_env *env,
862 struct dt_object *dt,
863 struct lu_attr *attr,
864 struct dt_allocation_hint *hint,
865 struct dt_object_format *dof,
868 struct dt_object *next = dt_object_child(dt);
869 struct lod_object *lo = lod_dt_obj(dt);
876 LASSERT(!dt_object_exists(next));
879 * first of all, we declare creation of local object
881 rc = dt_declare_create(env, next, attr, hint, dof, th);
885 if (dof->dof_type == DFT_SYM)
886 dt->do_body_ops = &lod_body_lnk_ops;
889 * it's lod_ah_init() who has decided the object will striped
891 if (dof->dof_type == DFT_REGULAR) {
892 /* callers don't want stripes */
893 /* XXX: all tricky interactions with ->ah_make_hint() decided
894 * to use striping, then ->declare_create() behaving differently
895 * should be cleaned */
896 if (dof->u.dof_reg.striped == 0)
897 lo->ldo_stripenr = 0;
898 if (lo->ldo_stripenr > 0)
899 rc = lod_declare_striped_object(env, dt, attr,
901 } else if (dof->dof_type == DFT_DIR && lo->ldo_striping_cached) {
902 struct lod_thread_info *info = lod_env_info(env);
904 info->lti_buf.lb_buf = NULL;
905 info->lti_buf.lb_len = sizeof(struct lov_user_md_v3);
906 /* to transfer default striping from the parent */
907 rc = dt_declare_xattr_set(env, next, &info->lti_buf,
908 XATTR_NAME_LOV, 0, th);
915 int lod_striping_create(const struct lu_env *env, struct dt_object *dt,
916 struct lu_attr *attr, struct dt_object_format *dof,
919 struct lod_object *lo = lod_dt_obj(dt);
923 LASSERT(lo->ldo_stripe);
924 LASSERT(lo->ldo_stripe > 0);
925 LASSERT(lo->ldo_striping_cached == 0);
927 /* create all underlying objects */
928 for (i = 0; i < lo->ldo_stripenr; i++) {
929 LASSERT(lo->ldo_stripe[i]);
930 rc = dt_create(env, lo->ldo_stripe[i], attr, NULL, dof, th);
936 rc = lod_generate_and_set_lovea(env, lo, th);
941 static int lod_object_create(const struct lu_env *env, struct dt_object *dt,
942 struct lu_attr *attr,
943 struct dt_allocation_hint *hint,
944 struct dt_object_format *dof, struct thandle *th)
946 struct dt_object *next = dt_object_child(dt);
947 struct lod_object *lo = lod_dt_obj(dt);
951 /* create local object */
952 rc = dt_create(env, next, attr, hint, dof, th);
955 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
956 rc = lod_store_def_striping(env, dt, th);
957 else if (lo->ldo_stripe)
958 rc = lod_striping_create(env, dt, attr, dof, th);
964 static int lod_declare_object_destroy(const struct lu_env *env,
965 struct dt_object *dt,
968 struct dt_object *next = dt_object_child(dt);
969 struct lod_object *lo = lod_dt_obj(dt);
974 * we declare destroy for the local object
976 rc = dt_declare_destroy(env, next, th);
981 * load striping information, notice we don't do this when object
982 * is being initialized as we don't need this information till
983 * few specific cases like destroy, chown
985 rc = lod_load_striping(env, lo);
989 /* declare destroy for all underlying objects */
990 for (i = 0; i < lo->ldo_stripenr; i++) {
991 LASSERT(lo->ldo_stripe[i]);
992 rc = dt_declare_destroy(env, lo->ldo_stripe[i], th);
1001 static int lod_object_destroy(const struct lu_env *env,
1002 struct dt_object *dt, struct thandle *th)
1004 struct dt_object *next = dt_object_child(dt);
1005 struct lod_object *lo = lod_dt_obj(dt);
1009 /* destroy local object */
1010 rc = dt_destroy(env, next, th);
1014 /* destroy all underlying objects */
1015 for (i = 0; i < lo->ldo_stripenr; i++) {
1016 LASSERT(lo->ldo_stripe[i]);
1017 rc = dt_destroy(env, lo->ldo_stripe[i], th);
1025 static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
1026 const struct dt_index_features *feat)
1028 struct dt_object *next = dt_object_child(dt);
1032 LASSERT(next->do_ops);
1033 LASSERT(next->do_ops->do_index_try);
1035 rc = next->do_ops->do_index_try(env, next, feat);
1036 if (next->do_index_ops && dt->do_index_ops == NULL)
1037 dt->do_index_ops = &lod_index_ops;
1042 static int lod_declare_ref_add(const struct lu_env *env,
1043 struct dt_object *dt, struct thandle *th)
1045 return dt_declare_ref_add(env, dt_object_child(dt), th);
1048 static int lod_ref_add(const struct lu_env *env,
1049 struct dt_object *dt, struct thandle *th)
1051 return dt_ref_add(env, dt_object_child(dt), th);
1054 static int lod_declare_ref_del(const struct lu_env *env,
1055 struct dt_object *dt, struct thandle *th)
1057 return dt_declare_ref_del(env, dt_object_child(dt), th);
1060 static int lod_ref_del(const struct lu_env *env,
1061 struct dt_object *dt, struct thandle *th)
1063 return dt_ref_del(env, dt_object_child(dt), th);
1066 static struct obd_capa *lod_capa_get(const struct lu_env *env,
1067 struct dt_object *dt,
1068 struct lustre_capa *old, __u64 opc)
1070 return dt_capa_get(env, dt_object_child(dt), old, opc);
1073 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt)
1075 return dt_object_sync(env, dt_object_child(dt));
1078 struct dt_object_operations lod_obj_ops = {
1079 .do_read_lock = lod_object_read_lock,
1080 .do_write_lock = lod_object_write_lock,
1081 .do_read_unlock = lod_object_read_unlock,
1082 .do_write_unlock = lod_object_write_unlock,
1083 .do_write_locked = lod_object_write_locked,
1084 .do_attr_get = lod_attr_get,
1085 .do_declare_attr_set = lod_declare_attr_set,
1086 .do_attr_set = lod_attr_set,
1087 .do_xattr_get = lod_xattr_get,
1088 .do_declare_xattr_set = lod_declare_xattr_set,
1089 .do_xattr_set = lod_xattr_set,
1090 .do_declare_xattr_del = lod_declare_xattr_del,
1091 .do_xattr_del = lod_xattr_del,
1092 .do_xattr_list = lod_xattr_list,
1093 .do_ah_init = lod_ah_init,
1094 .do_declare_create = lod_declare_object_create,
1095 .do_create = lod_object_create,
1096 .do_declare_destroy = lod_declare_object_destroy,
1097 .do_destroy = lod_object_destroy,
1098 .do_index_try = lod_index_try,
1099 .do_declare_ref_add = lod_declare_ref_add,
1100 .do_ref_add = lod_ref_add,
1101 .do_declare_ref_del = lod_declare_ref_del,
1102 .do_ref_del = lod_ref_del,
1103 .do_capa_get = lod_capa_get,
1104 .do_object_sync = lod_object_sync,
1107 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
1108 struct lu_buf *buf, loff_t *pos,
1109 struct lustre_capa *capa)
1111 struct dt_object *next = dt_object_child(dt);
1112 return next->do_body_ops->dbo_read(env, next, buf, pos, capa);
1115 static ssize_t lod_declare_write(const struct lu_env *env,
1116 struct dt_object *dt,
1117 const loff_t size, loff_t pos,
1120 return dt_declare_record_write(env, dt_object_child(dt),
1124 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
1125 const struct lu_buf *buf, loff_t *pos,
1126 struct thandle *th, struct lustre_capa *capa, int iq)
1128 struct dt_object *next = dt_object_child(dt);
1130 return next->do_body_ops->dbo_write(env, next, buf, pos, th, capa, iq);
1133 static const struct dt_body_operations lod_body_lnk_ops = {
1134 .dbo_read = lod_read,
1135 .dbo_declare_write = lod_declare_write,
1136 .dbo_write = lod_write
1139 static int lod_object_init(const struct lu_env *env, struct lu_object *o,
1140 const struct lu_object_conf *conf)
1142 struct lod_device *d = lu2lod_dev(o->lo_dev);
1143 struct lu_object *below;
1144 struct lu_device *under;
1148 * create local object
1150 under = &d->lod_child->dd_lu_dev;
1151 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
1155 lu_object_add(o, below);
1160 void lod_object_free_striping(const struct lu_env *env, struct lod_object *lo)
1164 if (lo->ldo_stripe) {
1165 LASSERT(lo->ldo_stripes_allocated > 0);
1167 for (i = 0; i < lo->ldo_stripenr; i++) {
1168 if (lo->ldo_stripe[i])
1169 lu_object_put(env, &lo->ldo_stripe[i]->do_lu);
1172 i = sizeof(struct dt_object *) * lo->ldo_stripes_allocated;
1173 OBD_FREE(lo->ldo_stripe, i);
1174 lo->ldo_stripe = NULL;
1175 lo->ldo_stripes_allocated = 0;
1177 lo->ldo_stripenr = 0;
1181 * ->start is called once all slices are initialized, including header's
1182 * cache for mode (object type). using the type we can initialize ops
1184 static int lod_object_start(const struct lu_env *env, struct lu_object *o)
1186 if (S_ISLNK(o->lo_header->loh_attr & S_IFMT))
1187 lu2lod_obj(o)->ldo_obj.do_body_ops = &lod_body_lnk_ops;
1191 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
1193 struct lod_object *mo = lu2lod_obj(o);
1196 * release all underlying object pinned
1199 lod_object_free_striping(env, mo);
1201 lod_object_set_pool(mo, NULL);
1204 OBD_SLAB_FREE_PTR(mo, lod_object_kmem);
1207 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
1209 /* XXX: shouldn't we release everything here in case if object
1210 * creation failed before? */
1213 static int lod_object_print(const struct lu_env *env, void *cookie,
1214 lu_printer_t p, const struct lu_object *l)
1216 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
1218 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
1221 struct lu_object_operations lod_lu_obj_ops = {
1222 .loo_object_init = lod_object_init,
1223 .loo_object_start = lod_object_start,
1224 .loo_object_free = lod_object_free,
1225 .loo_object_release = lod_object_release,
1226 .loo_object_print = lod_object_print,
1230 * Init remote lod object
1232 static int lod_robject_init(const struct lu_env *env, struct lu_object *lo,
1233 const struct lu_object_conf *conf)
1235 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
1236 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1237 struct lu_device *c_dev = NULL;
1238 struct lu_object *c_obj;
1243 if (ltd->ltd_tgts_size > 0) {
1244 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
1245 struct lod_tgt_desc *tgt;
1246 tgt = LTD_TGT(ltd, i);
1247 LASSERT(tgt && tgt->ltd_tgt);
1248 if (tgt->ltd_index ==
1249 lu2lod_obj(lo)->ldo_mds_num) {
1250 c_dev = &(tgt->ltd_tgt->dd_lu_dev);
1255 lod_putref(lod, ltd);
1257 if (unlikely(c_dev == NULL))
1260 c_obj = c_dev->ld_ops->ldo_object_alloc(env, lo->lo_header, c_dev);
1261 if (unlikely(c_obj == NULL))
1264 lu_object_add(lo, c_obj);
1269 struct lu_object_operations lod_lu_robj_ops = {
1270 .loo_object_init = lod_robject_init,
1271 .loo_object_start = lod_object_start,
1272 .loo_object_free = lod_object_free,
1273 .loo_object_release = lod_object_release,
1274 .loo_object_print = lod_object_print,