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 LASSERT((it)->lit_obj != NULL); \
128 LASSERT((it)->lit_it != NULL); \
131 void lod_it_fini(const struct lu_env *env, struct dt_it *di)
133 struct lod_it *it = (struct lod_it *)di;
135 LOD_CHECK_IT(env, it);
136 it->lit_obj->do_index_ops->dio_it.fini(env, it->lit_it);
138 /* the iterator not in use any more */
143 int lod_it_get(const struct lu_env *env, struct dt_it *di,
144 const struct dt_key *key)
146 const struct lod_it *it = (const struct lod_it *)di;
148 LOD_CHECK_IT(env, it);
149 return it->lit_obj->do_index_ops->dio_it.get(env, it->lit_it, key);
152 void lod_it_put(const struct lu_env *env, struct dt_it *di)
154 struct lod_it *it = (struct lod_it *)di;
156 LOD_CHECK_IT(env, it);
157 return it->lit_obj->do_index_ops->dio_it.put(env, it->lit_it);
160 int lod_it_next(const struct lu_env *env, struct dt_it *di)
162 struct lod_it *it = (struct lod_it *)di;
164 LOD_CHECK_IT(env, it);
165 return it->lit_obj->do_index_ops->dio_it.next(env, it->lit_it);
168 struct dt_key *lod_it_key(const struct lu_env *env, const struct dt_it *di)
170 const struct lod_it *it = (const struct lod_it *)di;
172 LOD_CHECK_IT(env, it);
173 return it->lit_obj->do_index_ops->dio_it.key(env, it->lit_it);
176 int lod_it_key_size(const struct lu_env *env, const struct dt_it *di)
178 struct lod_it *it = (struct lod_it *)di;
180 LOD_CHECK_IT(env, it);
181 return it->lit_obj->do_index_ops->dio_it.key_size(env, it->lit_it);
184 int lod_it_rec(const struct lu_env *env, const struct dt_it *di,
185 struct dt_rec *rec, __u32 attr)
187 const struct lod_it *it = (const struct lod_it *)di;
189 LOD_CHECK_IT(env, it);
190 return it->lit_obj->do_index_ops->dio_it.rec(env, it->lit_it, rec, attr);
193 __u64 lod_it_store(const struct lu_env *env, const struct dt_it *di)
195 const struct lod_it *it = (const struct lod_it *)di;
197 LOD_CHECK_IT(env, it);
198 return it->lit_obj->do_index_ops->dio_it.store(env, it->lit_it);
201 int lod_it_load(const struct lu_env *env, const struct dt_it *di, __u64 hash)
203 const struct lod_it *it = (const struct lod_it *)di;
205 LOD_CHECK_IT(env, it);
206 return it->lit_obj->do_index_ops->dio_it.load(env, it->lit_it, hash);
209 int lod_it_key_rec(const struct lu_env *env, const struct dt_it *di,
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.key_rec(env, it->lit_it, key_rec);
218 static struct dt_index_operations lod_index_ops = {
219 .dio_lookup = lod_index_lookup,
220 .dio_declare_insert = lod_declare_index_insert,
221 .dio_insert = lod_index_insert,
222 .dio_declare_delete = lod_declare_index_delete,
223 .dio_delete = lod_index_delete,
231 .key_size = lod_it_key_size,
233 .store = lod_it_store,
235 .key_rec = lod_it_key_rec,
239 static void lod_object_read_lock(const struct lu_env *env,
240 struct dt_object *dt, unsigned role)
242 dt_read_lock(env, dt_object_child(dt), role);
245 static void lod_object_write_lock(const struct lu_env *env,
246 struct dt_object *dt, unsigned role)
248 dt_write_lock(env, dt_object_child(dt), role);
251 static void lod_object_read_unlock(const struct lu_env *env,
252 struct dt_object *dt)
254 dt_read_unlock(env, dt_object_child(dt));
257 static void lod_object_write_unlock(const struct lu_env *env,
258 struct dt_object *dt)
260 dt_write_unlock(env, dt_object_child(dt));
263 static int lod_object_write_locked(const struct lu_env *env,
264 struct dt_object *dt)
266 return dt_write_locked(env, dt_object_child(dt));
269 static int lod_attr_get(const struct lu_env *env,
270 struct dt_object *dt,
271 struct lu_attr *attr,
272 struct lustre_capa *capa)
274 return dt_attr_get(env, dt_object_child(dt), attr, capa);
277 static int lod_declare_attr_set(const struct lu_env *env,
278 struct dt_object *dt,
279 const struct lu_attr *attr,
280 struct thandle *handle)
282 struct dt_object *next = dt_object_child(dt);
283 struct lod_object *lo = lod_dt_obj(dt);
288 * declare setattr on the local object
290 rc = dt_declare_attr_set(env, next, attr, handle);
295 * load striping information, notice we don't do this when object
296 * is being initialized as we don't need this information till
297 * few specific cases like destroy, chown
299 rc = lod_load_striping(env, lo);
304 * if object is striped declare changes on the stripes
306 LASSERT(lo->ldo_stripe || lo->ldo_stripenr == 0);
307 for (i = 0; i < lo->ldo_stripenr; i++) {
308 LASSERT(lo->ldo_stripe[i]);
309 rc = dt_declare_attr_set(env, lo->ldo_stripe[i], attr, handle);
311 CERROR("failed declaration: %d\n", rc);
319 static int lod_attr_set(const struct lu_env *env,
320 struct dt_object *dt,
321 const struct lu_attr *attr,
322 struct thandle *handle,
323 struct lustre_capa *capa)
325 struct dt_object *next = dt_object_child(dt);
326 struct lod_object *lo = lod_dt_obj(dt);
331 * apply changes to the local object
333 rc = dt_attr_set(env, next, attr, handle, capa);
338 * if object is striped, apply changes to all the stripes
340 LASSERT(lo->ldo_stripe || lo->ldo_stripenr == 0);
341 for (i = 0; i < lo->ldo_stripenr; i++) {
342 LASSERT(lo->ldo_stripe[i]);
343 rc = dt_attr_set(env, lo->ldo_stripe[i], attr, handle, capa);
345 CERROR("failed declaration: %d\n", rc);
353 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
354 struct lu_buf *buf, const char *name,
355 struct lustre_capa *capa)
357 struct lod_thread_info *info = lod_env_info(env);
358 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
362 rc = dt_xattr_get(env, dt_object_child(dt), buf, name, capa);
363 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
367 * lod returns default striping on the real root of the device
368 * this is like the root stores default striping for the whole
369 * filesystem. historically we've been using a different approach
370 * and store it in the config.
372 dt_root_get(env, dev->lod_child, &info->lti_fid);
373 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
375 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
376 struct lov_user_md *lum = buf->lb_buf;
377 struct lov_desc *desc = &dev->lod_desc;
379 if (buf->lb_buf == NULL) {
380 rc = sizeof(struct lov_user_md_v1);
381 } else if (buf->lb_len >= sizeof(struct lov_user_md_v1)) {
382 lum->lmm_magic = LOV_USER_MAGIC_V1;
383 lum->lmm_oi.oi_seq = FID_SEQ_LOV_DEFAULT;
384 lum->lmm_pattern = desc->ld_pattern;
385 lum->lmm_stripe_size = desc->ld_default_stripe_size;
386 lum->lmm_stripe_count = desc->ld_default_stripe_count;
387 lum->lmm_stripe_offset = desc->ld_default_stripe_offset;
388 rc = sizeof(struct lov_user_md_v1);
398 * LOV xattr is a storage for striping, and LOD owns this xattr.
399 * but LOD allows others to control striping to some extent
401 * - to set new defined striping
402 * - to set new semi-defined striping
403 * - number of stripes is defined
404 * - number of stripes + osts are defined
407 static int lod_declare_xattr_set(const struct lu_env *env,
408 struct dt_object *dt,
409 const struct lu_buf *buf,
410 const char *name, int fl,
413 struct dt_object *next = dt_object_child(dt);
414 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
420 * allow to declare predefined striping on a new (!mode) object
421 * which is supposed to be replay of regular file creation
422 * (when LOV setting is declared)
423 * LU_XATTR_REPLACE is set to indicate a layout swap
425 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
426 if ((S_ISREG(mode) || !mode) && !strcmp(name, XATTR_NAME_LOV) &&
427 !(fl & LU_XATTR_REPLACE)) {
429 * this is a request to manipulate object's striping
431 if (dt_object_exists(dt)) {
432 rc = dt_attr_get(env, next, attr, BYPASS_CAPA);
436 memset(attr, 0, sizeof(*attr));
437 attr->la_valid = LA_TYPE | LA_MODE;
438 attr->la_mode = S_IFREG;
440 rc = lod_declare_striped_object(env, dt, attr, buf, th);
445 rc = dt_declare_xattr_set(env, next, buf, name, fl, th);
450 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
451 struct dt_object *dt,
452 const struct lu_buf *buf,
453 const char *name, int fl,
455 struct lustre_capa *capa)
457 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
458 struct dt_object *next = dt_object_child(dt);
459 struct lod_object *l = lod_dt_obj(dt);
460 struct lov_user_md_v1 *lum;
461 struct lov_user_md_v3 *v3 = NULL;
465 LASSERT(l->ldo_stripe == NULL);
466 l->ldo_striping_cached = 0;
467 l->ldo_def_striping_set = 0;
468 lod_object_set_pool(l, NULL);
469 l->ldo_def_stripe_size = 0;
470 l->ldo_def_stripenr = 0;
473 LASSERT(buf->lb_buf);
476 rc = lod_verify_striping(d, buf, 0);
480 if (lum->lmm_magic == LOV_USER_MAGIC_V3)
483 /* if { size, offset, count } = { 0, -1, 0 } and no pool
484 * (i.e. all default values specified) then delete default
485 * striping from dir. */
487 "set default striping: sz %u # %u offset %d %s %s\n",
488 (unsigned)lum->lmm_stripe_size,
489 (unsigned)lum->lmm_stripe_count,
490 (int)lum->lmm_stripe_offset,
491 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
493 if (LOVEA_DELETE_VALUES((lum->lmm_stripe_size),
494 (lum->lmm_stripe_count),
495 (lum->lmm_stripe_offset)) &&
496 lum->lmm_magic == LOV_USER_MAGIC_V1) {
497 rc = dt_xattr_del(env, next, name, th, capa);
501 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
507 static int lod_xattr_set(const struct lu_env *env,
508 struct dt_object *dt, const struct lu_buf *buf,
509 const char *name, int fl, struct thandle *th,
510 struct lustre_capa *capa)
512 struct dt_object *next = dt_object_child(dt);
517 attr = dt->do_lu.lo_header->loh_attr & S_IFMT;
519 if (strncmp(name, XATTR_NAME_LOV, strlen(XATTR_NAME_LOV)) == 0)
520 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name,
523 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
525 } else if (S_ISREG(attr) && !strcmp(name, XATTR_NAME_LOV)) {
526 /* in case of lov EA swap, just set it
527 * if not, it is a replay so check striping match what we
528 * already have during req replay, declare_xattr_set()
529 * defines striping, then create() does the work
531 if (fl & LU_XATTR_REPLACE) {
532 /* free stripes, then update disk */
533 lod_object_free_striping(env, lod_dt_obj(dt));
534 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
536 rc = lod_striping_create(env, dt, NULL, NULL, th);
541 * behave transparantly for all other EAs
543 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
549 static int lod_declare_xattr_del(const struct lu_env *env,
550 struct dt_object *dt, const char *name,
553 return dt_declare_xattr_del(env, dt_object_child(dt), name, th);
556 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
557 const char *name, struct thandle *th,
558 struct lustre_capa *capa)
560 return dt_xattr_del(env, dt_object_child(dt), name, th, capa);
563 static int lod_xattr_list(const struct lu_env *env,
564 struct dt_object *dt, struct lu_buf *buf,
565 struct lustre_capa *capa)
567 return dt_xattr_list(env, dt_object_child(dt), buf, capa);
570 int lod_object_set_pool(struct lod_object *o, char *pool)
575 len = strlen(o->ldo_pool);
576 OBD_FREE(o->ldo_pool, len + 1);
581 OBD_ALLOC(o->ldo_pool, len + 1);
582 if (o->ldo_pool == NULL)
584 strcpy(o->ldo_pool, pool);
589 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
591 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
594 static int lod_cache_parent_striping(const struct lu_env *env,
595 struct lod_object *lp)
597 struct lov_user_md_v1 *v1 = NULL;
598 struct lov_user_md_v3 *v3 = NULL;
602 /* dt_ah_init() is called from MDD without parent being write locked
604 dt_write_lock(env, dt_object_child(&lp->ldo_obj), 0);
605 if (lp->ldo_striping_cached)
606 GOTO(unlock, rc = 0);
608 rc = lod_get_lov_ea(env, lp);
612 if (rc < sizeof(struct lov_user_md)) {
613 /* don't lookup for non-existing or invalid striping */
614 lp->ldo_def_striping_set = 0;
615 lp->ldo_striping_cached = 1;
616 lp->ldo_def_stripe_size = 0;
617 lp->ldo_def_stripenr = 0;
618 lp->ldo_def_stripe_offset = (typeof(v1->lmm_stripe_offset))(-1);
619 GOTO(unlock, rc = 0);
622 v1 = (struct lov_user_md_v1 *)lod_env_info(env)->lti_ea_store;
623 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1))
624 lustre_swab_lov_user_md_v1(v1);
625 else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3))
626 lustre_swab_lov_user_md_v3(v3);
628 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1)
629 GOTO(unlock, rc = 0);
631 if (v1->lmm_pattern != LOV_PATTERN_RAID0 && v1->lmm_pattern != 0)
632 GOTO(unlock, rc = 0);
634 lp->ldo_def_stripenr = v1->lmm_stripe_count;
635 lp->ldo_def_stripe_size = v1->lmm_stripe_size;
636 lp->ldo_def_stripe_offset = v1->lmm_stripe_offset;
637 lp->ldo_striping_cached = 1;
638 lp->ldo_def_striping_set = 1;
640 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
641 /* XXX: sanity check here */
642 v3 = (struct lov_user_md_v3 *) v1;
643 if (v3->lmm_pool_name[0])
644 lod_object_set_pool(lp, v3->lmm_pool_name);
647 CDEBUG(D_OTHER, "def. striping: # %d, sz %d, off %d %s%s on "DFID"\n",
648 lp->ldo_def_stripenr, lp->ldo_def_stripe_size,
649 lp->ldo_def_stripe_offset, v3 ? "from " : "",
650 v3 ? lp->ldo_pool : "", PFID(lu_object_fid(&lp->ldo_obj.do_lu)));
654 dt_write_unlock(env, dt_object_child(&lp->ldo_obj));
659 * used to transfer default striping data to the object being created
661 static void lod_ah_init(const struct lu_env *env,
662 struct dt_allocation_hint *ah,
663 struct dt_object *parent,
664 struct dt_object *child,
665 cfs_umode_t child_mode)
667 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
668 struct dt_object *nextp = NULL;
669 struct dt_object *nextc;
670 struct lod_object *lp = NULL;
671 struct lod_object *lc;
672 struct lov_desc *desc;
677 if (likely(parent)) {
678 nextp = dt_object_child(parent);
679 lp = lod_dt_obj(parent);
682 nextc = dt_object_child(child);
683 lc = lod_dt_obj(child);
685 LASSERT(lc->ldo_stripenr == 0);
686 LASSERT(lc->ldo_stripe == NULL);
689 * local object may want some hints
690 * in case of late striping creation, ->ah_init()
691 * can be called with local object existing
693 if (!dt_object_exists(nextc) || dt_object_remote(nextc))
694 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
696 if (S_ISDIR(child_mode)) {
697 if (lp->ldo_striping_cached == 0) {
698 /* we haven't tried to get default striping for
699 * the directory yet, let's cache it in the object */
700 lod_cache_parent_striping(env, lp);
702 /* transfer defaults to new directory */
703 if (lp->ldo_striping_cached) {
705 lod_object_set_pool(lc, lp->ldo_pool);
706 lc->ldo_def_stripenr = lp->ldo_def_stripenr;
707 lc->ldo_def_stripe_size = lp->ldo_def_stripe_size;
708 lc->ldo_def_stripe_offset = lp->ldo_def_stripe_offset;
709 lc->ldo_striping_cached = 1;
710 lc->ldo_def_striping_set = 1;
711 CDEBUG(D_OTHER, "inherite EA sz:%d off:%d nr:%d\n",
712 (int)lc->ldo_def_stripenr,
713 (int)lc->ldo_def_stripe_size,
714 (int)lc->ldo_def_stripe_offset);
720 * if object is going to be striped over OSTs, transfer default
721 * striping information to the child, so that we can use it
722 * during declaration and creation
724 if (!lod_object_will_be_striped(S_ISREG(child_mode),
725 lu_object_fid(&child->do_lu)))
729 * try from the parent
731 if (likely(parent)) {
732 if (lp->ldo_striping_cached == 0) {
733 /* we haven't tried to get default striping for
734 * the directory yet, let's cache it in the object */
735 lod_cache_parent_striping(env, lp);
738 lc->ldo_def_stripe_offset = (__u16) -1;
740 if (lp->ldo_def_striping_set) {
742 lod_object_set_pool(lc, lp->ldo_pool);
743 lc->ldo_stripenr = lp->ldo_def_stripenr;
744 lc->ldo_stripe_size = lp->ldo_def_stripe_size;
745 lc->ldo_def_stripe_offset = lp->ldo_def_stripe_offset;
746 CDEBUG(D_OTHER, "striping from parent: #%d, sz %d %s\n",
747 lc->ldo_stripenr, lc->ldo_stripe_size,
748 lp->ldo_pool ? lp->ldo_pool : "");
753 * if the parent doesn't provide with specific pattern, grab fs-wide one
756 if (lc->ldo_stripenr == 0)
757 lc->ldo_stripenr = desc->ld_default_stripe_count;
758 if (lc->ldo_stripe_size == 0)
759 lc->ldo_stripe_size = desc->ld_default_stripe_size;
760 CDEBUG(D_OTHER, "final striping: # %d stripes, sz %d from %s\n",
761 lc->ldo_stripenr, lc->ldo_stripe_size,
762 lc->ldo_pool ? lc->ldo_pool : "");
767 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
769 * this function handles a special case when truncate was done
770 * on a stripeless object and now striping is being created
771 * we can't lose that size, so we have to propagate it to newly
774 static int lod_declare_init_size(const struct lu_env *env,
775 struct dt_object *dt, struct thandle *th)
777 struct dt_object *next = dt_object_child(dt);
778 struct lod_object *lo = lod_dt_obj(dt);
779 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
784 /* XXX: we support the simplest (RAID0) striping so far */
785 LASSERT(lo->ldo_stripe || lo->ldo_stripenr == 0);
786 LASSERT(lo->ldo_stripe_size > 0);
788 rc = dt_attr_get(env, next, attr, BYPASS_CAPA);
789 LASSERT(attr->la_valid & LA_SIZE);
793 size = attr->la_size;
797 /* ll_do_div64(a, b) returns a % b, and a = a / b */
798 ll_do_div64(size, (__u64) lo->ldo_stripe_size);
799 stripe = ll_do_div64(size, (__u64) lo->ldo_stripenr);
801 size = size * lo->ldo_stripe_size;
802 offs = attr->la_size;
803 size += ll_do_div64(offs, lo->ldo_stripe_size);
805 attr->la_valid = LA_SIZE;
806 attr->la_size = size;
808 rc = dt_declare_attr_set(env, lo->ldo_stripe[stripe], attr, th);
815 * Create declaration of striped object
817 int lod_declare_striped_object(const struct lu_env *env, struct dt_object *dt,
818 struct lu_attr *attr,
819 const struct lu_buf *lovea, struct thandle *th)
821 struct lod_thread_info *info = lod_env_info(env);
822 struct dt_object *next = dt_object_child(dt);
823 struct lod_object *lo = lod_dt_obj(dt);
827 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO)) {
828 /* failed to create striping, let's reset
829 * config so that others don't get confused */
830 lod_object_free_striping(env, lo);
831 GOTO(out, rc = -ENOMEM);
834 /* choose OST and generate appropriate objects */
835 rc = lod_qos_prep_create(env, lo, attr, lovea, th);
837 /* failed to create striping, let's reset
838 * config so that others don't get confused */
839 lod_object_free_striping(env, lo);
844 * declare storage for striping data
846 info->lti_buf.lb_len = lov_mds_md_size(lo->ldo_stripenr,
847 lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1);
848 rc = dt_declare_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV,
854 * if striping is created with local object's size > 0,
855 * we have to propagate this size to specific object
856 * the case is possible only when local object was created previously
858 if (dt_object_exists(next))
859 rc = lod_declare_init_size(env, dt, th);
865 static int lod_declare_object_create(const struct lu_env *env,
866 struct dt_object *dt,
867 struct lu_attr *attr,
868 struct dt_allocation_hint *hint,
869 struct dt_object_format *dof,
872 struct dt_object *next = dt_object_child(dt);
873 struct lod_object *lo = lod_dt_obj(dt);
882 * first of all, we declare creation of local object
884 rc = dt_declare_create(env, next, attr, hint, dof, th);
888 if (dof->dof_type == DFT_SYM)
889 dt->do_body_ops = &lod_body_lnk_ops;
892 * it's lod_ah_init() who has decided the object will striped
894 if (dof->dof_type == DFT_REGULAR) {
895 /* callers don't want stripes */
896 /* XXX: all tricky interactions with ->ah_make_hint() decided
897 * to use striping, then ->declare_create() behaving differently
898 * should be cleaned */
899 if (dof->u.dof_reg.striped == 0)
900 lo->ldo_stripenr = 0;
901 if (lo->ldo_stripenr > 0)
902 rc = lod_declare_striped_object(env, dt, attr,
904 } else if (dof->dof_type == DFT_DIR && lo->ldo_striping_cached) {
905 struct lod_thread_info *info = lod_env_info(env);
907 struct lov_user_md_v3 *v3;
909 if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size,
910 lo->ldo_def_stripenr,
911 lo->ldo_def_stripe_offset))
918 v3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
919 v3->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
920 fid_ostid_pack(lu_object_fid(&dt->do_lu), &v3->lmm_oi);
921 ostid_cpu_to_le(&v3->lmm_oi, &v3->lmm_oi);
922 v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
923 v3->lmm_stripe_count = cpu_to_le32(lo->ldo_def_stripenr);
924 v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
926 strncpy(v3->lmm_pool_name, lo->ldo_pool,
929 info->lti_buf.lb_buf = v3;
930 info->lti_buf.lb_len = sizeof(*v3);
932 /* to transfer default striping from the parent */
933 rc = dt_declare_xattr_set(env, next, &info->lti_buf,
934 XATTR_NAME_LOV, 0, th);
942 int lod_striping_create(const struct lu_env *env, struct dt_object *dt,
943 struct lu_attr *attr, struct dt_object_format *dof,
946 struct lod_object *lo = lod_dt_obj(dt);
950 LASSERT(lo->ldo_stripe);
951 LASSERT(lo->ldo_stripenr > 0);
952 LASSERT(lo->ldo_striping_cached == 0);
954 /* create all underlying objects */
955 for (i = 0; i < lo->ldo_stripenr; i++) {
956 LASSERT(lo->ldo_stripe[i]);
957 rc = dt_create(env, lo->ldo_stripe[i], attr, NULL, dof, th);
963 rc = lod_generate_and_set_lovea(env, lo, th);
968 static int lod_object_create(const struct lu_env *env, struct dt_object *dt,
969 struct lu_attr *attr,
970 struct dt_allocation_hint *hint,
971 struct dt_object_format *dof, struct thandle *th)
973 struct dt_object *next = dt_object_child(dt);
974 struct lod_object *lo = lod_dt_obj(dt);
978 /* create local object */
979 rc = dt_create(env, next, attr, hint, dof, th);
982 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
983 rc = lod_store_def_striping(env, dt, th);
984 else if (lo->ldo_stripe)
985 rc = lod_striping_create(env, dt, attr, dof, th);
991 static int lod_declare_object_destroy(const struct lu_env *env,
992 struct dt_object *dt,
995 struct dt_object *next = dt_object_child(dt);
996 struct lod_object *lo = lod_dt_obj(dt);
1001 * we declare destroy for the local object
1003 rc = dt_declare_destroy(env, next, th);
1008 * load striping information, notice we don't do this when object
1009 * is being initialized as we don't need this information till
1010 * few specific cases like destroy, chown
1012 rc = lod_load_striping(env, lo);
1016 /* declare destroy for all underlying objects */
1017 for (i = 0; i < lo->ldo_stripenr; i++) {
1018 LASSERT(lo->ldo_stripe[i]);
1019 rc = dt_declare_destroy(env, lo->ldo_stripe[i], th);
1028 static int lod_object_destroy(const struct lu_env *env,
1029 struct dt_object *dt, struct thandle *th)
1031 struct dt_object *next = dt_object_child(dt);
1032 struct lod_object *lo = lod_dt_obj(dt);
1036 /* destroy local object */
1037 rc = dt_destroy(env, next, th);
1041 /* destroy all underlying objects */
1042 for (i = 0; i < lo->ldo_stripenr; i++) {
1043 LASSERT(lo->ldo_stripe[i]);
1044 rc = dt_destroy(env, lo->ldo_stripe[i], th);
1052 static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
1053 const struct dt_index_features *feat)
1055 struct dt_object *next = dt_object_child(dt);
1059 LASSERT(next->do_ops);
1060 LASSERT(next->do_ops->do_index_try);
1062 rc = next->do_ops->do_index_try(env, next, feat);
1063 if (next->do_index_ops && dt->do_index_ops == NULL)
1064 dt->do_index_ops = &lod_index_ops;
1069 static int lod_declare_ref_add(const struct lu_env *env,
1070 struct dt_object *dt, struct thandle *th)
1072 return dt_declare_ref_add(env, dt_object_child(dt), th);
1075 static int lod_ref_add(const struct lu_env *env,
1076 struct dt_object *dt, struct thandle *th)
1078 return dt_ref_add(env, dt_object_child(dt), th);
1081 static int lod_declare_ref_del(const struct lu_env *env,
1082 struct dt_object *dt, struct thandle *th)
1084 return dt_declare_ref_del(env, dt_object_child(dt), th);
1087 static int lod_ref_del(const struct lu_env *env,
1088 struct dt_object *dt, struct thandle *th)
1090 return dt_ref_del(env, dt_object_child(dt), th);
1093 static struct obd_capa *lod_capa_get(const struct lu_env *env,
1094 struct dt_object *dt,
1095 struct lustre_capa *old, __u64 opc)
1097 return dt_capa_get(env, dt_object_child(dt), old, opc);
1100 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt)
1102 return dt_object_sync(env, dt_object_child(dt));
1105 struct dt_object_operations lod_obj_ops = {
1106 .do_read_lock = lod_object_read_lock,
1107 .do_write_lock = lod_object_write_lock,
1108 .do_read_unlock = lod_object_read_unlock,
1109 .do_write_unlock = lod_object_write_unlock,
1110 .do_write_locked = lod_object_write_locked,
1111 .do_attr_get = lod_attr_get,
1112 .do_declare_attr_set = lod_declare_attr_set,
1113 .do_attr_set = lod_attr_set,
1114 .do_xattr_get = lod_xattr_get,
1115 .do_declare_xattr_set = lod_declare_xattr_set,
1116 .do_xattr_set = lod_xattr_set,
1117 .do_declare_xattr_del = lod_declare_xattr_del,
1118 .do_xattr_del = lod_xattr_del,
1119 .do_xattr_list = lod_xattr_list,
1120 .do_ah_init = lod_ah_init,
1121 .do_declare_create = lod_declare_object_create,
1122 .do_create = lod_object_create,
1123 .do_declare_destroy = lod_declare_object_destroy,
1124 .do_destroy = lod_object_destroy,
1125 .do_index_try = lod_index_try,
1126 .do_declare_ref_add = lod_declare_ref_add,
1127 .do_ref_add = lod_ref_add,
1128 .do_declare_ref_del = lod_declare_ref_del,
1129 .do_ref_del = lod_ref_del,
1130 .do_capa_get = lod_capa_get,
1131 .do_object_sync = lod_object_sync,
1134 static int lod_object_lock(const struct lu_env *env,
1135 struct dt_object *dt, struct lustre_handle *lh,
1136 struct ldlm_enqueue_info *einfo,
1139 struct dt_object *next = dt_object_child(dt);
1144 * declare setattr on the local object
1146 rc = dt_object_lock(env, next, lh, einfo, policy);
1151 struct dt_lock_operations lod_lock_ops = {
1152 .do_object_lock = lod_object_lock,
1154 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
1155 struct lu_buf *buf, loff_t *pos,
1156 struct lustre_capa *capa)
1158 struct dt_object *next = dt_object_child(dt);
1159 return next->do_body_ops->dbo_read(env, next, buf, pos, capa);
1162 static ssize_t lod_declare_write(const struct lu_env *env,
1163 struct dt_object *dt,
1164 const loff_t size, loff_t pos,
1167 return dt_declare_record_write(env, dt_object_child(dt),
1171 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
1172 const struct lu_buf *buf, loff_t *pos,
1173 struct thandle *th, struct lustre_capa *capa, int iq)
1175 struct dt_object *next = dt_object_child(dt);
1177 return next->do_body_ops->dbo_write(env, next, buf, pos, th, capa, iq);
1180 static const struct dt_body_operations lod_body_lnk_ops = {
1181 .dbo_read = lod_read,
1182 .dbo_declare_write = lod_declare_write,
1183 .dbo_write = lod_write
1186 static int lod_object_init(const struct lu_env *env, struct lu_object *o,
1187 const struct lu_object_conf *conf)
1189 struct lod_device *d = lu2lod_dev(o->lo_dev);
1190 struct lu_object *below;
1191 struct lu_device *under;
1195 * create local object
1197 under = &d->lod_child->dd_lu_dev;
1198 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
1202 lu_object_add(o, below);
1207 void lod_object_free_striping(const struct lu_env *env, struct lod_object *lo)
1211 if (lo->ldo_stripe) {
1212 LASSERT(lo->ldo_stripes_allocated > 0);
1214 for (i = 0; i < lo->ldo_stripenr; i++) {
1215 if (lo->ldo_stripe[i])
1216 lu_object_put(env, &lo->ldo_stripe[i]->do_lu);
1219 i = sizeof(struct dt_object *) * lo->ldo_stripes_allocated;
1220 OBD_FREE(lo->ldo_stripe, i);
1221 lo->ldo_stripe = NULL;
1222 lo->ldo_stripes_allocated = 0;
1224 lo->ldo_stripenr = 0;
1228 * ->start is called once all slices are initialized, including header's
1229 * cache for mode (object type). using the type we can initialize ops
1231 static int lod_object_start(const struct lu_env *env, struct lu_object *o)
1233 if (S_ISLNK(o->lo_header->loh_attr & S_IFMT))
1234 lu2lod_obj(o)->ldo_obj.do_body_ops = &lod_body_lnk_ops;
1238 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
1240 struct lod_object *mo = lu2lod_obj(o);
1243 * release all underlying object pinned
1246 lod_object_free_striping(env, mo);
1248 lod_object_set_pool(mo, NULL);
1251 OBD_SLAB_FREE_PTR(mo, lod_object_kmem);
1254 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
1256 /* XXX: shouldn't we release everything here in case if object
1257 * creation failed before? */
1260 static int lod_object_print(const struct lu_env *env, void *cookie,
1261 lu_printer_t p, const struct lu_object *l)
1263 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
1265 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
1268 struct lu_object_operations lod_lu_obj_ops = {
1269 .loo_object_init = lod_object_init,
1270 .loo_object_start = lod_object_start,
1271 .loo_object_free = lod_object_free,
1272 .loo_object_release = lod_object_release,
1273 .loo_object_print = lod_object_print,
1277 * Init remote lod object
1279 static int lod_robject_init(const struct lu_env *env, struct lu_object *lo,
1280 const struct lu_object_conf *conf)
1282 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
1283 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1284 struct lu_device *c_dev = NULL;
1285 struct lu_object *c_obj;
1290 if (ltd->ltd_tgts_size > 0) {
1291 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
1292 struct lod_tgt_desc *tgt;
1293 tgt = LTD_TGT(ltd, i);
1294 LASSERT(tgt && tgt->ltd_tgt);
1295 if (tgt->ltd_index ==
1296 lu2lod_obj(lo)->ldo_mds_num) {
1297 c_dev = &(tgt->ltd_tgt->dd_lu_dev);
1302 lod_putref(lod, ltd);
1304 if (unlikely(c_dev == NULL))
1307 c_obj = c_dev->ld_ops->ldo_object_alloc(env, lo->lo_header, c_dev);
1308 if (unlikely(c_obj == NULL))
1311 lu_object_add(lo, c_obj);
1316 struct lu_object_operations lod_lu_robj_ops = {
1317 .loo_object_init = lod_robject_init,
1318 .loo_object_start = lod_object_start,
1319 .loo_object_free = lod_object_free,
1320 .loo_object_release = lod_object_release,
1321 .loo_object_print = lod_object_print,