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, 2013, Intel Corporation.
29 * lustre/lod/lod_object.c
31 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
34 #define DEBUG_SUBSYSTEM S_MDS
37 #include <obd_class.h>
38 #include <lustre_ver.h>
39 #include <obd_support.h>
40 #include <lprocfs_status.h>
42 #include <lustre_fid.h>
43 #include <lustre_param.h>
44 #include <lustre_fid.h>
47 #include "lod_internal.h"
49 extern struct kmem_cache *lod_object_kmem;
50 static const struct dt_body_operations lod_body_lnk_ops;
52 static int lod_index_lookup(const struct lu_env *env, struct dt_object *dt,
53 struct dt_rec *rec, const struct dt_key *key,
54 struct lustre_capa *capa)
56 struct dt_object *next = dt_object_child(dt);
57 return next->do_index_ops->dio_lookup(env, next, rec, key, capa);
60 static int lod_declare_index_insert(const struct lu_env *env,
62 const struct dt_rec *rec,
63 const struct dt_key *key,
64 struct thandle *handle)
66 return dt_declare_insert(env, dt_object_child(dt), rec, key, handle);
69 static int lod_index_insert(const struct lu_env *env,
71 const struct dt_rec *rec,
72 const struct dt_key *key,
74 struct lustre_capa *capa,
77 return dt_insert(env, dt_object_child(dt), rec, key, th, capa, ign);
80 static int lod_declare_index_delete(const struct lu_env *env,
82 const struct dt_key *key,
85 return dt_declare_delete(env, dt_object_child(dt), key, th);
88 static int lod_index_delete(const struct lu_env *env,
90 const struct dt_key *key,
92 struct lustre_capa *capa)
94 return dt_delete(env, dt_object_child(dt), key, th, capa);
97 static struct dt_it *lod_it_init(const struct lu_env *env,
98 struct dt_object *dt, __u32 attr,
99 struct lustre_capa *capa)
101 struct dt_object *next = dt_object_child(dt);
102 struct lod_it *it = &lod_env_info(env)->lti_it;
103 struct dt_it *it_next;
106 it_next = next->do_index_ops->dio_it.init(env, next, attr, capa);
110 /* currently we do not use more than one iterator per thread
111 * so we store it in thread info. if at some point we need
112 * more active iterators in a single thread, we can allocate
114 LASSERT(it->lit_obj == NULL);
116 it->lit_it = it_next;
119 return (struct dt_it *)it;
122 #define LOD_CHECK_IT(env, it) \
124 LASSERT((it)->lit_obj != NULL); \
125 LASSERT((it)->lit_it != NULL); \
128 void lod_it_fini(const struct lu_env *env, struct dt_it *di)
130 struct lod_it *it = (struct lod_it *)di;
132 LOD_CHECK_IT(env, it);
133 it->lit_obj->do_index_ops->dio_it.fini(env, it->lit_it);
135 /* the iterator not in use any more */
140 int lod_it_get(const struct lu_env *env, struct dt_it *di,
141 const struct dt_key *key)
143 const struct lod_it *it = (const struct lod_it *)di;
145 LOD_CHECK_IT(env, it);
146 return it->lit_obj->do_index_ops->dio_it.get(env, it->lit_it, key);
149 void lod_it_put(const struct lu_env *env, struct dt_it *di)
151 struct lod_it *it = (struct lod_it *)di;
153 LOD_CHECK_IT(env, it);
154 return it->lit_obj->do_index_ops->dio_it.put(env, it->lit_it);
157 int lod_it_next(const struct lu_env *env, struct dt_it *di)
159 struct lod_it *it = (struct lod_it *)di;
161 LOD_CHECK_IT(env, it);
162 return it->lit_obj->do_index_ops->dio_it.next(env, it->lit_it);
165 struct dt_key *lod_it_key(const struct lu_env *env, const struct dt_it *di)
167 const struct lod_it *it = (const struct lod_it *)di;
169 LOD_CHECK_IT(env, it);
170 return it->lit_obj->do_index_ops->dio_it.key(env, it->lit_it);
173 int lod_it_key_size(const struct lu_env *env, const struct dt_it *di)
175 struct lod_it *it = (struct lod_it *)di;
177 LOD_CHECK_IT(env, it);
178 return it->lit_obj->do_index_ops->dio_it.key_size(env, it->lit_it);
181 int lod_it_rec(const struct lu_env *env, const struct dt_it *di,
182 struct dt_rec *rec, __u32 attr)
184 const struct lod_it *it = (const struct lod_it *)di;
186 LOD_CHECK_IT(env, it);
187 return it->lit_obj->do_index_ops->dio_it.rec(env, it->lit_it, rec, attr);
190 __u64 lod_it_store(const struct lu_env *env, const struct dt_it *di)
192 const struct lod_it *it = (const struct lod_it *)di;
194 LOD_CHECK_IT(env, it);
195 return it->lit_obj->do_index_ops->dio_it.store(env, it->lit_it);
198 int lod_it_load(const struct lu_env *env, const struct dt_it *di, __u64 hash)
200 const struct lod_it *it = (const struct lod_it *)di;
202 LOD_CHECK_IT(env, it);
203 return it->lit_obj->do_index_ops->dio_it.load(env, it->lit_it, hash);
206 int lod_it_key_rec(const struct lu_env *env, const struct dt_it *di,
209 const struct lod_it *it = (const struct lod_it *)di;
211 LOD_CHECK_IT(env, it);
212 return it->lit_obj->do_index_ops->dio_it.key_rec(env, it->lit_it, key_rec);
215 static struct dt_index_operations lod_index_ops = {
216 .dio_lookup = lod_index_lookup,
217 .dio_declare_insert = lod_declare_index_insert,
218 .dio_insert = lod_index_insert,
219 .dio_declare_delete = lod_declare_index_delete,
220 .dio_delete = lod_index_delete,
228 .key_size = lod_it_key_size,
230 .store = lod_it_store,
232 .key_rec = lod_it_key_rec,
236 static void lod_object_read_lock(const struct lu_env *env,
237 struct dt_object *dt, unsigned role)
239 dt_read_lock(env, dt_object_child(dt), role);
242 static void lod_object_write_lock(const struct lu_env *env,
243 struct dt_object *dt, unsigned role)
245 dt_write_lock(env, dt_object_child(dt), role);
248 static void lod_object_read_unlock(const struct lu_env *env,
249 struct dt_object *dt)
251 dt_read_unlock(env, dt_object_child(dt));
254 static void lod_object_write_unlock(const struct lu_env *env,
255 struct dt_object *dt)
257 dt_write_unlock(env, dt_object_child(dt));
260 static int lod_object_write_locked(const struct lu_env *env,
261 struct dt_object *dt)
263 return dt_write_locked(env, dt_object_child(dt));
266 static int lod_attr_get(const struct lu_env *env,
267 struct dt_object *dt,
268 struct lu_attr *attr,
269 struct lustre_capa *capa)
271 return dt_attr_get(env, dt_object_child(dt), attr, capa);
274 static int lod_declare_attr_set(const struct lu_env *env,
275 struct dt_object *dt,
276 const struct lu_attr *attr,
277 struct thandle *handle)
279 struct dt_object *next = dt_object_child(dt);
280 struct lod_object *lo = lod_dt_obj(dt);
285 * declare setattr on the local object
287 rc = dt_declare_attr_set(env, next, attr, handle);
291 /* osp_declare_attr_set() ignores all attributes other than
292 * UID, GID, and size, and osp_attr_set() ignores all but UID
293 * and GID. Declaration of size attr setting happens through
294 * lod_declare_init_size(), and not through this function.
295 * Therefore we need not load striping unless ownership is
296 * changing. This should save memory and (we hope) speed up
298 if (!(attr->la_valid & (LA_UID | LA_GID)))
302 * load striping information, notice we don't do this when object
303 * is being initialized as we don't need this information till
304 * few specific cases like destroy, chown
306 rc = lod_load_striping(env, lo);
311 * if object is striped declare changes on the stripes
313 LASSERT(lo->ldo_stripe || lo->ldo_stripenr == 0);
314 for (i = 0; i < lo->ldo_stripenr; i++) {
315 LASSERT(lo->ldo_stripe[i]);
316 rc = dt_declare_attr_set(env, lo->ldo_stripe[i], attr, handle);
318 CERROR("failed declaration: %d\n", rc);
326 static int lod_attr_set(const struct lu_env *env,
327 struct dt_object *dt,
328 const struct lu_attr *attr,
329 struct thandle *handle,
330 struct lustre_capa *capa)
332 struct dt_object *next = dt_object_child(dt);
333 struct lod_object *lo = lod_dt_obj(dt);
338 * apply changes to the local object
340 rc = dt_attr_set(env, next, attr, handle, capa);
344 if (!(attr->la_valid & (LA_UID | LA_GID)))
348 * if object is striped, apply changes to all the stripes
350 LASSERT(lo->ldo_stripe || lo->ldo_stripenr == 0);
351 for (i = 0; i < lo->ldo_stripenr; i++) {
352 LASSERT(lo->ldo_stripe[i]);
353 rc = dt_attr_set(env, lo->ldo_stripe[i], attr, handle, capa);
355 CERROR("failed declaration: %d\n", rc);
363 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
364 struct lu_buf *buf, const char *name,
365 struct lustre_capa *capa)
367 struct lod_thread_info *info = lod_env_info(env);
368 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
372 rc = dt_xattr_get(env, dt_object_child(dt), buf, name, capa);
373 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
377 * lod returns default striping on the real root of the device
378 * this is like the root stores default striping for the whole
379 * filesystem. historically we've been using a different approach
380 * and store it in the config.
382 dt_root_get(env, dev->lod_child, &info->lti_fid);
383 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
385 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
386 struct lov_user_md *lum = buf->lb_buf;
387 struct lov_desc *desc = &dev->lod_desc;
389 if (buf->lb_buf == NULL) {
391 } else if (buf->lb_len >= sizeof(*lum)) {
392 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
393 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
394 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
395 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
396 lum->lmm_stripe_size = cpu_to_le32(
397 desc->ld_default_stripe_size);
398 lum->lmm_stripe_count = cpu_to_le16(
399 desc->ld_default_stripe_count);
400 lum->lmm_stripe_offset = cpu_to_le16(
401 desc->ld_default_stripe_offset);
412 * LOV xattr is a storage for striping, and LOD owns this xattr.
413 * but LOD allows others to control striping to some extent
415 * - to set new defined striping
416 * - to set new semi-defined striping
417 * - number of stripes is defined
418 * - number of stripes + osts are defined
421 static int lod_declare_xattr_set(const struct lu_env *env,
422 struct dt_object *dt,
423 const struct lu_buf *buf,
424 const char *name, int fl,
427 struct dt_object *next = dt_object_child(dt);
428 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
434 * allow to declare predefined striping on a new (!mode) object
435 * which is supposed to be replay of regular file creation
436 * (when LOV setting is declared)
437 * LU_XATTR_REPLACE is set to indicate a layout swap
439 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
440 if ((S_ISREG(mode) || !mode) && !strcmp(name, XATTR_NAME_LOV) &&
441 !(fl & LU_XATTR_REPLACE)) {
443 * this is a request to manipulate object's striping
445 if (dt_object_exists(dt)) {
446 rc = dt_attr_get(env, next, attr, BYPASS_CAPA);
450 memset(attr, 0, sizeof(*attr));
451 attr->la_valid = LA_TYPE | LA_MODE;
452 attr->la_mode = S_IFREG;
454 rc = lod_declare_striped_object(env, dt, attr, buf, th);
459 rc = dt_declare_xattr_set(env, next, buf, name, fl, th);
464 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
465 struct dt_object *dt,
466 const struct lu_buf *buf,
467 const char *name, int fl,
469 struct lustre_capa *capa)
471 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
472 struct dt_object *next = dt_object_child(dt);
473 struct lod_object *l = lod_dt_obj(dt);
474 struct lov_user_md_v1 *lum;
475 struct lov_user_md_v3 *v3 = NULL;
479 LASSERT(l->ldo_stripe == NULL);
480 l->ldo_striping_cached = 0;
481 l->ldo_def_striping_set = 0;
482 lod_object_set_pool(l, NULL);
483 l->ldo_def_stripe_size = 0;
484 l->ldo_def_stripenr = 0;
487 LASSERT(buf->lb_buf);
490 rc = lod_verify_striping(d, buf, 0);
494 if (lum->lmm_magic == LOV_USER_MAGIC_V3)
497 /* if { size, offset, count } = { 0, -1, 0 } and no pool
498 * (i.e. all default values specified) then delete default
499 * striping from dir. */
501 "set default striping: sz %u # %u offset %d %s %s\n",
502 (unsigned)lum->lmm_stripe_size,
503 (unsigned)lum->lmm_stripe_count,
504 (int)lum->lmm_stripe_offset,
505 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
507 if (LOVEA_DELETE_VALUES((lum->lmm_stripe_size),
508 (lum->lmm_stripe_count),
509 (lum->lmm_stripe_offset)) &&
510 lum->lmm_magic == LOV_USER_MAGIC_V1) {
511 rc = dt_xattr_del(env, next, name, th, capa);
515 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
521 static int lod_xattr_set(const struct lu_env *env,
522 struct dt_object *dt, const struct lu_buf *buf,
523 const char *name, int fl, struct thandle *th,
524 struct lustre_capa *capa)
526 struct dt_object *next = dt_object_child(dt);
531 attr = dt->do_lu.lo_header->loh_attr & S_IFMT;
533 if (strncmp(name, XATTR_NAME_LOV, strlen(XATTR_NAME_LOV)) == 0)
534 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name,
537 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
539 } else if (S_ISREG(attr) && !strcmp(name, XATTR_NAME_LOV)) {
540 /* in case of lov EA swap, just set it
541 * if not, it is a replay so check striping match what we
542 * already have during req replay, declare_xattr_set()
543 * defines striping, then create() does the work
545 if (fl & LU_XATTR_REPLACE) {
546 /* free stripes, then update disk */
547 lod_object_free_striping(env, lod_dt_obj(dt));
548 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
550 rc = lod_striping_create(env, dt, NULL, NULL, th);
555 * behave transparantly for all other EAs
557 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
563 static int lod_declare_xattr_del(const struct lu_env *env,
564 struct dt_object *dt, const char *name,
567 return dt_declare_xattr_del(env, dt_object_child(dt), name, th);
570 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
571 const char *name, struct thandle *th,
572 struct lustre_capa *capa)
574 if (!strcmp(name, XATTR_NAME_LOV))
575 lod_object_free_striping(env, lod_dt_obj(dt));
576 return dt_xattr_del(env, dt_object_child(dt), name, th, capa);
579 static int lod_xattr_list(const struct lu_env *env,
580 struct dt_object *dt, struct lu_buf *buf,
581 struct lustre_capa *capa)
583 return dt_xattr_list(env, dt_object_child(dt), buf, capa);
586 int lod_object_set_pool(struct lod_object *o, char *pool)
591 len = strlen(o->ldo_pool);
592 OBD_FREE(o->ldo_pool, len + 1);
597 OBD_ALLOC(o->ldo_pool, len + 1);
598 if (o->ldo_pool == NULL)
600 strcpy(o->ldo_pool, pool);
605 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
607 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
610 static int lod_cache_parent_striping(const struct lu_env *env,
611 struct lod_object *lp)
613 struct lov_user_md_v1 *v1 = NULL;
614 struct lov_user_md_v3 *v3 = NULL;
618 /* dt_ah_init() is called from MDD without parent being write locked
620 dt_write_lock(env, dt_object_child(&lp->ldo_obj), 0);
621 if (lp->ldo_striping_cached)
622 GOTO(unlock, rc = 0);
624 rc = lod_get_lov_ea(env, lp);
628 if (rc < sizeof(struct lov_user_md)) {
629 /* don't lookup for non-existing or invalid striping */
630 lp->ldo_def_striping_set = 0;
631 lp->ldo_striping_cached = 1;
632 lp->ldo_def_stripe_size = 0;
633 lp->ldo_def_stripenr = 0;
634 lp->ldo_def_stripe_offset = (typeof(v1->lmm_stripe_offset))(-1);
635 GOTO(unlock, rc = 0);
638 v1 = (struct lov_user_md_v1 *)lod_env_info(env)->lti_ea_store;
639 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1))
640 lustre_swab_lov_user_md_v1(v1);
641 else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3))
642 lustre_swab_lov_user_md_v3(v3);
644 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1)
645 GOTO(unlock, rc = 0);
647 if (v1->lmm_pattern != LOV_PATTERN_RAID0 && v1->lmm_pattern != 0)
648 GOTO(unlock, rc = 0);
650 lp->ldo_def_stripenr = v1->lmm_stripe_count;
651 lp->ldo_def_stripe_size = v1->lmm_stripe_size;
652 lp->ldo_def_stripe_offset = v1->lmm_stripe_offset;
653 lp->ldo_striping_cached = 1;
654 lp->ldo_def_striping_set = 1;
656 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
657 /* XXX: sanity check here */
658 v3 = (struct lov_user_md_v3 *) v1;
659 if (v3->lmm_pool_name[0])
660 lod_object_set_pool(lp, v3->lmm_pool_name);
663 CDEBUG(D_OTHER, "def. striping: # %d, sz %d, off %d %s%s on "DFID"\n",
664 lp->ldo_def_stripenr, lp->ldo_def_stripe_size,
665 lp->ldo_def_stripe_offset, v3 ? "from " : "",
666 v3 ? lp->ldo_pool : "", PFID(lu_object_fid(&lp->ldo_obj.do_lu)));
670 dt_write_unlock(env, dt_object_child(&lp->ldo_obj));
675 * used to transfer default striping data to the object being created
677 static void lod_ah_init(const struct lu_env *env,
678 struct dt_allocation_hint *ah,
679 struct dt_object *parent,
680 struct dt_object *child,
681 cfs_umode_t child_mode)
683 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
684 struct dt_object *nextp = NULL;
685 struct dt_object *nextc;
686 struct lod_object *lp = NULL;
687 struct lod_object *lc;
688 struct lov_desc *desc;
693 if (likely(parent)) {
694 nextp = dt_object_child(parent);
695 lp = lod_dt_obj(parent);
698 nextc = dt_object_child(child);
699 lc = lod_dt_obj(child);
701 LASSERT(lc->ldo_stripenr == 0);
702 LASSERT(lc->ldo_stripe == NULL);
705 * local object may want some hints
706 * in case of late striping creation, ->ah_init()
707 * can be called with local object existing
709 if (!dt_object_exists(nextc) || dt_object_remote(nextc))
710 nextc->do_ops->do_ah_init(env, ah, nextp, nextc, child_mode);
712 if (S_ISDIR(child_mode)) {
713 if (lp->ldo_striping_cached == 0) {
714 /* we haven't tried to get default striping for
715 * the directory yet, let's cache it in the object */
716 lod_cache_parent_striping(env, lp);
718 /* transfer defaults to new directory */
719 if (lp->ldo_striping_cached) {
721 lod_object_set_pool(lc, lp->ldo_pool);
722 lc->ldo_def_stripenr = lp->ldo_def_stripenr;
723 lc->ldo_def_stripe_size = lp->ldo_def_stripe_size;
724 lc->ldo_def_stripe_offset = lp->ldo_def_stripe_offset;
725 lc->ldo_striping_cached = 1;
726 lc->ldo_def_striping_set = 1;
727 CDEBUG(D_OTHER, "inherite EA sz:%d off:%d nr:%d\n",
728 (int)lc->ldo_def_stripenr,
729 (int)lc->ldo_def_stripe_size,
730 (int)lc->ldo_def_stripe_offset);
736 * if object is going to be striped over OSTs, transfer default
737 * striping information to the child, so that we can use it
738 * during declaration and creation
740 if (!lod_object_will_be_striped(S_ISREG(child_mode),
741 lu_object_fid(&child->do_lu)))
745 * try from the parent
747 if (likely(parent)) {
748 if (lp->ldo_striping_cached == 0) {
749 /* we haven't tried to get default striping for
750 * the directory yet, let's cache it in the object */
751 lod_cache_parent_striping(env, lp);
754 lc->ldo_def_stripe_offset = (__u16) -1;
756 if (lp->ldo_def_striping_set) {
758 lod_object_set_pool(lc, lp->ldo_pool);
759 lc->ldo_stripenr = lp->ldo_def_stripenr;
760 lc->ldo_stripe_size = lp->ldo_def_stripe_size;
761 lc->ldo_def_stripe_offset = lp->ldo_def_stripe_offset;
762 CDEBUG(D_OTHER, "striping from parent: #%d, sz %d %s\n",
763 lc->ldo_stripenr, lc->ldo_stripe_size,
764 lp->ldo_pool ? lp->ldo_pool : "");
769 * if the parent doesn't provide with specific pattern, grab fs-wide one
772 if (lc->ldo_stripenr == 0)
773 lc->ldo_stripenr = desc->ld_default_stripe_count;
774 if (lc->ldo_stripe_size == 0)
775 lc->ldo_stripe_size = desc->ld_default_stripe_size;
776 CDEBUG(D_OTHER, "final striping: # %d stripes, sz %d from %s\n",
777 lc->ldo_stripenr, lc->ldo_stripe_size,
778 lc->ldo_pool ? lc->ldo_pool : "");
783 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
785 * this function handles a special case when truncate was done
786 * on a stripeless object and now striping is being created
787 * we can't lose that size, so we have to propagate it to newly
790 static int lod_declare_init_size(const struct lu_env *env,
791 struct dt_object *dt, struct thandle *th)
793 struct dt_object *next = dt_object_child(dt);
794 struct lod_object *lo = lod_dt_obj(dt);
795 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
800 /* XXX: we support the simplest (RAID0) striping so far */
801 LASSERT(lo->ldo_stripe || lo->ldo_stripenr == 0);
802 LASSERT(lo->ldo_stripe_size > 0);
804 rc = dt_attr_get(env, next, attr, BYPASS_CAPA);
805 LASSERT(attr->la_valid & LA_SIZE);
809 size = attr->la_size;
813 /* ll_do_div64(a, b) returns a % b, and a = a / b */
814 ll_do_div64(size, (__u64) lo->ldo_stripe_size);
815 stripe = ll_do_div64(size, (__u64) lo->ldo_stripenr);
817 size = size * lo->ldo_stripe_size;
818 offs = attr->la_size;
819 size += ll_do_div64(offs, lo->ldo_stripe_size);
821 attr->la_valid = LA_SIZE;
822 attr->la_size = size;
824 rc = dt_declare_attr_set(env, lo->ldo_stripe[stripe], attr, th);
831 * Create declaration of striped object
833 int lod_declare_striped_object(const struct lu_env *env, struct dt_object *dt,
834 struct lu_attr *attr,
835 const struct lu_buf *lovea, struct thandle *th)
837 struct lod_thread_info *info = lod_env_info(env);
838 struct dt_object *next = dt_object_child(dt);
839 struct lod_object *lo = lod_dt_obj(dt);
843 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO)) {
844 /* failed to create striping, let's reset
845 * config so that others don't get confused */
846 lod_object_free_striping(env, lo);
847 GOTO(out, rc = -ENOMEM);
850 /* choose OST and generate appropriate objects */
851 rc = lod_qos_prep_create(env, lo, attr, lovea, th);
853 /* failed to create striping, let's reset
854 * config so that others don't get confused */
855 lod_object_free_striping(env, lo);
860 * declare storage for striping data
862 info->lti_buf.lb_len = lov_mds_md_size(lo->ldo_stripenr,
863 lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1);
864 rc = dt_declare_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV,
870 * if striping is created with local object's size > 0,
871 * we have to propagate this size to specific object
872 * the case is possible only when local object was created previously
874 if (dt_object_exists(next))
875 rc = lod_declare_init_size(env, dt, th);
881 static int lod_declare_object_create(const struct lu_env *env,
882 struct dt_object *dt,
883 struct lu_attr *attr,
884 struct dt_allocation_hint *hint,
885 struct dt_object_format *dof,
888 struct dt_object *next = dt_object_child(dt);
889 struct lod_object *lo = lod_dt_obj(dt);
898 * first of all, we declare creation of local object
900 rc = dt_declare_create(env, next, attr, hint, dof, th);
904 if (dof->dof_type == DFT_SYM)
905 dt->do_body_ops = &lod_body_lnk_ops;
908 * it's lod_ah_init() who has decided the object will striped
910 if (dof->dof_type == DFT_REGULAR) {
911 /* callers don't want stripes */
912 /* XXX: all tricky interactions with ->ah_make_hint() decided
913 * to use striping, then ->declare_create() behaving differently
914 * should be cleaned */
915 if (dof->u.dof_reg.striped == 0)
916 lo->ldo_stripenr = 0;
917 if (lo->ldo_stripenr > 0)
918 rc = lod_declare_striped_object(env, dt, attr,
920 } else if (dof->dof_type == DFT_DIR && lo->ldo_striping_cached) {
921 struct lod_thread_info *info = lod_env_info(env);
923 struct lov_user_md_v3 *v3;
925 if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size,
926 lo->ldo_def_stripenr,
927 lo->ldo_def_stripe_offset))
934 v3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
935 v3->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
936 fid_to_lmm_oi(lu_object_fid(&dt->do_lu), &v3->lmm_oi);
937 lmm_oi_cpu_to_le(&v3->lmm_oi, &v3->lmm_oi);
938 v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
939 v3->lmm_stripe_count = cpu_to_le32(lo->ldo_def_stripenr);
940 v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
942 strncpy(v3->lmm_pool_name, lo->ldo_pool,
945 info->lti_buf.lb_buf = v3;
946 info->lti_buf.lb_len = sizeof(*v3);
948 /* to transfer default striping from the parent */
949 rc = dt_declare_xattr_set(env, next, &info->lti_buf,
950 XATTR_NAME_LOV, 0, th);
958 int lod_striping_create(const struct lu_env *env, struct dt_object *dt,
959 struct lu_attr *attr, struct dt_object_format *dof,
962 struct lod_object *lo = lod_dt_obj(dt);
966 LASSERT(lo->ldo_striping_cached == 0);
968 /* create all underlying objects */
969 for (i = 0; i < lo->ldo_stripenr; i++) {
970 LASSERT(lo->ldo_stripe[i]);
971 rc = dt_create(env, lo->ldo_stripe[i], attr, NULL, dof, th);
977 rc = lod_generate_and_set_lovea(env, lo, th);
982 static int lod_object_create(const struct lu_env *env, struct dt_object *dt,
983 struct lu_attr *attr,
984 struct dt_allocation_hint *hint,
985 struct dt_object_format *dof, struct thandle *th)
987 struct dt_object *next = dt_object_child(dt);
988 struct lod_object *lo = lod_dt_obj(dt);
992 /* create local object */
993 rc = dt_create(env, next, attr, hint, dof, th);
996 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
997 rc = lod_store_def_striping(env, dt, th);
998 else if (lo->ldo_stripe)
999 rc = lod_striping_create(env, dt, attr, dof, th);
1005 static int lod_declare_object_destroy(const struct lu_env *env,
1006 struct dt_object *dt,
1009 struct dt_object *next = dt_object_child(dt);
1010 struct lod_object *lo = lod_dt_obj(dt);
1015 * we declare destroy for the local object
1017 rc = dt_declare_destroy(env, next, th);
1022 * load striping information, notice we don't do this when object
1023 * is being initialized as we don't need this information till
1024 * few specific cases like destroy, chown
1026 rc = lod_load_striping(env, lo);
1030 /* declare destroy for all underlying objects */
1031 for (i = 0; i < lo->ldo_stripenr; i++) {
1032 LASSERT(lo->ldo_stripe[i]);
1033 rc = dt_declare_destroy(env, lo->ldo_stripe[i], th);
1042 static int lod_object_destroy(const struct lu_env *env,
1043 struct dt_object *dt, struct thandle *th)
1045 struct dt_object *next = dt_object_child(dt);
1046 struct lod_object *lo = lod_dt_obj(dt);
1050 /* destroy local object */
1051 rc = dt_destroy(env, next, th);
1055 /* destroy all underlying objects */
1056 for (i = 0; i < lo->ldo_stripenr; i++) {
1057 LASSERT(lo->ldo_stripe[i]);
1058 rc = dt_destroy(env, lo->ldo_stripe[i], th);
1066 static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
1067 const struct dt_index_features *feat)
1069 struct dt_object *next = dt_object_child(dt);
1073 LASSERT(next->do_ops);
1074 LASSERT(next->do_ops->do_index_try);
1076 rc = next->do_ops->do_index_try(env, next, feat);
1077 if (next->do_index_ops && dt->do_index_ops == NULL)
1078 dt->do_index_ops = &lod_index_ops;
1083 static int lod_declare_ref_add(const struct lu_env *env,
1084 struct dt_object *dt, struct thandle *th)
1086 return dt_declare_ref_add(env, dt_object_child(dt), th);
1089 static int lod_ref_add(const struct lu_env *env,
1090 struct dt_object *dt, struct thandle *th)
1092 return dt_ref_add(env, dt_object_child(dt), th);
1095 static int lod_declare_ref_del(const struct lu_env *env,
1096 struct dt_object *dt, struct thandle *th)
1098 return dt_declare_ref_del(env, dt_object_child(dt), th);
1101 static int lod_ref_del(const struct lu_env *env,
1102 struct dt_object *dt, struct thandle *th)
1104 return dt_ref_del(env, dt_object_child(dt), th);
1107 static struct obd_capa *lod_capa_get(const struct lu_env *env,
1108 struct dt_object *dt,
1109 struct lustre_capa *old, __u64 opc)
1111 return dt_capa_get(env, dt_object_child(dt), old, opc);
1114 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt)
1116 return dt_object_sync(env, dt_object_child(dt));
1119 static int lod_object_lock(const struct lu_env *env,
1120 struct dt_object *dt, struct lustre_handle *lh,
1121 struct ldlm_enqueue_info *einfo,
1124 struct dt_object *next = dt_object_child(dt);
1129 * declare setattr on the local object
1131 rc = dt_object_lock(env, next, lh, einfo, policy);
1136 struct dt_object_operations lod_obj_ops = {
1137 .do_read_lock = lod_object_read_lock,
1138 .do_write_lock = lod_object_write_lock,
1139 .do_read_unlock = lod_object_read_unlock,
1140 .do_write_unlock = lod_object_write_unlock,
1141 .do_write_locked = lod_object_write_locked,
1142 .do_attr_get = lod_attr_get,
1143 .do_declare_attr_set = lod_declare_attr_set,
1144 .do_attr_set = lod_attr_set,
1145 .do_xattr_get = lod_xattr_get,
1146 .do_declare_xattr_set = lod_declare_xattr_set,
1147 .do_xattr_set = lod_xattr_set,
1148 .do_declare_xattr_del = lod_declare_xattr_del,
1149 .do_xattr_del = lod_xattr_del,
1150 .do_xattr_list = lod_xattr_list,
1151 .do_ah_init = lod_ah_init,
1152 .do_declare_create = lod_declare_object_create,
1153 .do_create = lod_object_create,
1154 .do_declare_destroy = lod_declare_object_destroy,
1155 .do_destroy = lod_object_destroy,
1156 .do_index_try = lod_index_try,
1157 .do_declare_ref_add = lod_declare_ref_add,
1158 .do_ref_add = lod_ref_add,
1159 .do_declare_ref_del = lod_declare_ref_del,
1160 .do_ref_del = lod_ref_del,
1161 .do_capa_get = lod_capa_get,
1162 .do_object_sync = lod_object_sync,
1163 .do_object_lock = lod_object_lock,
1166 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
1167 struct lu_buf *buf, loff_t *pos,
1168 struct lustre_capa *capa)
1170 struct dt_object *next = dt_object_child(dt);
1171 return next->do_body_ops->dbo_read(env, next, buf, pos, capa);
1174 static ssize_t lod_declare_write(const struct lu_env *env,
1175 struct dt_object *dt,
1176 const loff_t size, loff_t pos,
1179 return dt_declare_record_write(env, dt_object_child(dt),
1183 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
1184 const struct lu_buf *buf, loff_t *pos,
1185 struct thandle *th, struct lustre_capa *capa, int iq)
1187 struct dt_object *next = dt_object_child(dt);
1189 return next->do_body_ops->dbo_write(env, next, buf, pos, th, capa, iq);
1192 static const struct dt_body_operations lod_body_lnk_ops = {
1193 .dbo_read = lod_read,
1194 .dbo_declare_write = lod_declare_write,
1195 .dbo_write = lod_write
1198 static int lod_object_init(const struct lu_env *env, struct lu_object *o,
1199 const struct lu_object_conf *conf)
1201 struct lod_device *d = lu2lod_dev(o->lo_dev);
1202 struct lu_object *below;
1203 struct lu_device *under;
1207 * create local object
1209 under = &d->lod_child->dd_lu_dev;
1210 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
1214 lu_object_add(o, below);
1219 void lod_object_free_striping(const struct lu_env *env, struct lod_object *lo)
1223 if (lo->ldo_stripe) {
1224 LASSERT(lo->ldo_stripes_allocated > 0);
1226 for (i = 0; i < lo->ldo_stripenr; i++) {
1227 if (lo->ldo_stripe[i])
1228 lu_object_put(env, &lo->ldo_stripe[i]->do_lu);
1231 i = sizeof(struct dt_object *) * lo->ldo_stripes_allocated;
1232 OBD_FREE(lo->ldo_stripe, i);
1233 lo->ldo_stripe = NULL;
1234 lo->ldo_stripes_allocated = 0;
1236 lo->ldo_stripenr = 0;
1237 lo->ldo_pattern = 0;
1241 * ->start is called once all slices are initialized, including header's
1242 * cache for mode (object type). using the type we can initialize ops
1244 static int lod_object_start(const struct lu_env *env, struct lu_object *o)
1246 if (S_ISLNK(o->lo_header->loh_attr & S_IFMT))
1247 lu2lod_obj(o)->ldo_obj.do_body_ops = &lod_body_lnk_ops;
1251 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
1253 struct lod_object *mo = lu2lod_obj(o);
1256 * release all underlying object pinned
1259 lod_object_free_striping(env, mo);
1261 lod_object_set_pool(mo, NULL);
1264 OBD_SLAB_FREE_PTR(mo, lod_object_kmem);
1267 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
1269 /* XXX: shouldn't we release everything here in case if object
1270 * creation failed before? */
1273 static int lod_object_print(const struct lu_env *env, void *cookie,
1274 lu_printer_t p, const struct lu_object *l)
1276 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
1278 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
1281 struct lu_object_operations lod_lu_obj_ops = {
1282 .loo_object_init = lod_object_init,
1283 .loo_object_start = lod_object_start,
1284 .loo_object_free = lod_object_free,
1285 .loo_object_release = lod_object_release,
1286 .loo_object_print = lod_object_print,
1290 * Init remote lod object
1292 static int lod_robject_init(const struct lu_env *env, struct lu_object *lo,
1293 const struct lu_object_conf *conf)
1295 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
1296 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
1297 struct lu_device *c_dev = NULL;
1298 struct lu_object *c_obj;
1303 if (ltd->ltd_tgts_size > 0) {
1304 cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
1305 struct lod_tgt_desc *tgt;
1306 tgt = LTD_TGT(ltd, i);
1307 LASSERT(tgt && tgt->ltd_tgt);
1308 if (tgt->ltd_index ==
1309 lu2lod_obj(lo)->ldo_mds_num) {
1310 c_dev = &(tgt->ltd_tgt->dd_lu_dev);
1315 lod_putref(lod, ltd);
1317 if (unlikely(c_dev == NULL))
1320 c_obj = c_dev->ld_ops->ldo_object_alloc(env, lo->lo_header, c_dev);
1321 if (unlikely(c_obj == NULL))
1324 lu_object_add(lo, c_obj);
1329 struct lu_object_operations lod_lu_robj_ops = {
1330 .loo_object_init = lod_robject_init,
1331 .loo_object_start = lod_object_start,
1332 .loo_object_free = lod_object_free,
1333 .loo_object_release = lod_object_release,
1334 .loo_object_print = lod_object_print,