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, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/osd-zfs/osd_object.c
34 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
35 * Author: Mike Pershin <tappro@whamcloud.com>
36 * Author: Johann Lombardi <johann@whamcloud.com>
39 #define DEBUG_SUBSYSTEM S_OSD
41 #include <lustre_ver.h>
42 #include <libcfs/libcfs.h>
43 #include <obd_support.h>
44 #include <lustre_net.h>
46 #include <obd_class.h>
47 #include <lustre_disk.h>
48 #include <lustre_fid.h>
50 #include "osd_internal.h"
52 #include <sys/dnode.h>
57 #include <sys/spa_impl.h>
58 #include <sys/zfs_znode.h>
59 #include <sys/dmu_tx.h>
60 #include <sys/dmu_objset.h>
61 #include <sys/dsl_prop.h>
62 #include <sys/sa_impl.h>
65 char *osd_obj_tag = "osd_object";
67 static struct dt_object_operations osd_obj_ops;
68 static struct lu_object_operations osd_lu_obj_ops;
69 extern struct dt_body_operations osd_body_ops;
70 static struct dt_object_operations osd_obj_otable_it_ops;
72 extern struct kmem_cache *osd_object_kmem;
75 osd_object_sa_fini(struct osd_object *obj)
78 sa_handle_destroy(obj->oo_sa_hdl);
79 obj->oo_sa_hdl = NULL;
84 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
88 LASSERT(obj->oo_sa_hdl == NULL);
89 LASSERT(obj->oo_dn != NULL);
91 rc = -sa_handle_get(o->od_os, obj->oo_dn->dn_object, obj,
92 SA_HDL_PRIVATE, &obj->oo_sa_hdl);
96 /* Cache the xattr object id, valid for the life of the object */
97 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
99 obj->oo_xattr = ZFS_NO_OBJECT;
102 osd_object_sa_fini(obj);
109 * Add object to list of dirty objects in tx handle.
112 osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
114 if (!list_empty(&obj->oo_sa_linkage))
117 down(&oh->ot_sa_lock);
118 write_lock(&obj->oo_attr_lock);
119 if (likely(list_empty(&obj->oo_sa_linkage)))
120 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
121 write_unlock(&obj->oo_attr_lock);
126 * Release spill block dbuf hold for all dirty SAs.
128 void osd_object_sa_dirty_rele(struct osd_thandle *oh)
130 struct osd_object *obj;
132 down(&oh->ot_sa_lock);
133 while (!list_empty(&oh->ot_sa_list)) {
134 obj = list_entry(oh->ot_sa_list.next,
135 struct osd_object, oo_sa_linkage);
136 sa_spill_rele(obj->oo_sa_hdl);
137 write_lock(&obj->oo_attr_lock);
138 list_del_init(&obj->oo_sa_linkage);
139 write_unlock(&obj->oo_attr_lock);
145 * Update the SA and add the object to the dirty list.
147 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
148 void *buf, uint32_t buflen, struct osd_thandle *oh)
152 LASSERT(obj->oo_sa_hdl != NULL);
153 LASSERT(oh->ot_tx != NULL);
155 rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
156 osd_object_sa_dirty_add(obj, oh);
162 * Bulk update the SA and add the object to the dirty list.
165 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
166 int count, struct osd_thandle *oh)
170 LASSERT(obj->oo_sa_hdl != NULL);
171 LASSERT(oh->ot_tx != NULL);
173 rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
174 osd_object_sa_dirty_add(obj, oh);
180 * Retrieve the attributes of a DMU object
182 int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
183 struct osd_object *obj, struct lu_attr *la)
185 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
186 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
192 LASSERT(obj->oo_dn != NULL);
194 rc = -sa_handle_get(o->od_os, obj->oo_dn->dn_object, NULL,
195 SA_HDL_PRIVATE, &sa_hdl);
199 la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
200 LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
202 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
203 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
204 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
205 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
206 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
207 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
208 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
209 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
210 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
211 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
213 rc = -sa_bulk_lookup(sa_hdl, bulk, cnt);
217 la->la_atime = osa->atime[0];
218 la->la_mtime = osa->mtime[0];
219 la->la_ctime = osa->ctime[0];
220 la->la_mode = osa->mode;
221 la->la_uid = osa->uid;
222 la->la_gid = osa->gid;
223 la->la_nlink = osa->nlink;
224 la->la_flags = attrs_zfs2fs(osa->flags);
225 la->la_size = osa->size;
227 /* Try to get extra flag from LMA. Right now, only LMAI_ORPHAN
228 * flags is stored in LMA, and it is only for orphan directory */
229 if (S_ISDIR(la->la_mode) && dt_object_exists(&obj->oo_dt)) {
230 struct osd_thread_info *info = osd_oti_get(env);
231 struct lustre_mdt_attrs *lma;
234 lma = (struct lustre_mdt_attrs *)info->oti_buf;
236 buf.lb_len = sizeof(info->oti_buf);
237 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
240 lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
242 lma_to_lustre_flags(lma->lma_incompat);
244 } else if (rc == -ENODATA) {
249 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
250 rc = -sa_lookup(sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
253 la->la_rdev = osa->rdev;
254 la->la_valid |= LA_RDEV;
257 sa_handle_destroy(sa_hdl);
262 int __osd_obj2dnode(const struct lu_env *env, objset_t *os,
263 uint64_t oid, dnode_t **dnp)
265 dmu_object_info_t *doi = &osd_oti_get(env)->oti_doi;
270 rc = dmu_bonus_hold(os, oid, osd_obj_tag, &db);
274 dbi = (dmu_buf_impl_t *)db;
276 *dnp = DB_DNODE(dbi);
278 LASSERT(*dnp != NULL);
279 dmu_object_info_from_dnode(*dnp, doi);
280 if (unlikely (oid != DMU_USERUSED_OBJECT &&
281 oid != DMU_GROUPUSED_OBJECT && doi->doi_bonus_type != DMU_OT_SA)) {
282 osd_dnode_rele(*dnp);
291 * Concurrency: no concurrent access is possible that early in object
294 struct lu_object *osd_object_alloc(const struct lu_env *env,
295 const struct lu_object_header *hdr,
298 struct osd_object *mo;
300 OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
304 l = &mo->oo_dt.do_lu;
305 dt_object_init(&mo->oo_dt, NULL, d);
306 mo->oo_dt.do_ops = &osd_obj_ops;
307 l->lo_ops = &osd_lu_obj_ops;
308 INIT_LIST_HEAD(&mo->oo_sa_linkage);
309 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
310 init_rwsem(&mo->oo_sem);
311 init_rwsem(&mo->oo_guard);
312 rwlock_init(&mo->oo_attr_lock);
313 mo->oo_destroy = OSD_DESTROY_NONE;
321 * Concurrency: shouldn't matter.
323 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
325 struct osd_device *osd = osd_obj2dev(obj);
326 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
330 if (obj->oo_dn == NULL)
335 rc = osd_object_sa_init(obj, osd);
339 /* cache attrs in object */
340 rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
344 if (likely(!fid_is_acct(fid)))
345 /* no body operations for accounting objects */
346 obj->oo_dt.do_body_ops = &osd_body_ops;
349 * initialize object before marking it existing
351 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
354 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
359 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
361 struct osd_thread_info *info = osd_oti_get(env);
364 struct lustre_mdt_attrs *lma;
367 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
368 lma = (struct lustre_mdt_attrs *)info->oti_buf;
370 buf.lb_len = sizeof(info->oti_buf);
372 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
375 lustre_lma_swab(lma);
376 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
377 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
378 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
379 "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
380 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
381 PFID(lu_object_fid(&obj->oo_dt.do_lu)));
384 } else if (rc == -ENODATA) {
385 /* haven't initialize LMA xattr */
393 * Concurrency: no concurrent access is possible that early in object
396 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
397 const struct lu_object_conf *conf)
399 struct osd_object *obj = osd_obj(l);
400 struct osd_device *osd = osd_obj2dev(obj);
405 LASSERT(osd_invariant(obj));
407 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
408 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
409 l->lo_header->loh_attr |= LOHA_EXISTS;
413 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
416 rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
418 LASSERT(obj->oo_dn == NULL);
419 rc = __osd_obj2dnode(env, osd->od_os, oid, &obj->oo_dn);
420 /* EEXIST will be returned if object is being deleted in ZFS */
426 CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
427 osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
431 rc = osd_object_init0(env, obj);
435 rc = osd_check_lma(env, obj);
438 } else if (rc == -ENOENT) {
441 LASSERT(osd_invariant(obj));
447 * Concurrency: no concurrent access is possible that late in object
450 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
452 struct osd_object *obj = osd_obj(l);
454 LASSERT(osd_invariant(obj));
456 dt_object_fini(&obj->oo_dt);
457 OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
461 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
465 LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
467 /* the object is supposed to be exclusively locked by
468 * the caller (osd_object_destroy()), while the transaction
469 * (oh) is per-thread and not shared */
470 if (likely(list_empty(&obj->oo_unlinked_linkage))) {
471 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
478 /* Default to max data size covered by a level-1 indirect block */
479 static unsigned long osd_sync_destroy_max_size =
480 1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
481 module_param(osd_sync_destroy_max_size, ulong, 0444);
482 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
485 osd_object_set_destroy_type(struct osd_object *obj)
488 * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
489 * only once and use it consistently thereafter.
491 down_write(&obj->oo_guard);
492 if (obj->oo_destroy == OSD_DESTROY_NONE) {
493 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
494 obj->oo_destroy = OSD_DESTROY_SYNC;
495 else /* Larger objects are destroyed asynchronously */
496 obj->oo_destroy = OSD_DESTROY_ASYNC;
498 up_write(&obj->oo_guard);
501 static int osd_declare_object_destroy(const struct lu_env *env,
502 struct dt_object *dt,
505 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
506 struct osd_object *obj = osd_dt_obj(dt);
507 struct osd_device *osd = osd_obj2dev(obj);
508 struct osd_thandle *oh;
514 LASSERT(dt_object_exists(dt));
516 oh = container_of0(th, struct osd_thandle, ot_super);
517 LASSERT(oh->ot_tx != NULL);
519 /* declare that we'll remove object from fid-dnode mapping */
520 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0);
521 dmu_tx_hold_zap(oh->ot_tx, zapid, FALSE, NULL);
523 osd_declare_xattrs_destroy(env, obj, oh);
525 /* declare that we'll remove object from inode accounting ZAPs */
526 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, FALSE, NULL);
527 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, FALSE, NULL);
530 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
531 obj->oo_attr.la_gid, -1, oh, false, NULL, false);
535 /* data to be truncated */
536 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
537 obj->oo_attr.la_gid, 0, oh, true, NULL, false);
541 osd_object_set_destroy_type(obj);
542 if (obj->oo_destroy == OSD_DESTROY_SYNC)
543 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
546 dmu_tx_hold_zap(oh->ot_tx, osd->od_unlinkedid, TRUE, NULL);
548 /* will help to find FID->ino when this object is being
549 * added to PENDING/ */
550 osd_idc_find_and_init(env, osd, obj);
555 static int osd_object_destroy(const struct lu_env *env,
556 struct dt_object *dt, struct thandle *th)
558 struct osd_thread_info *info = osd_oti_get(env);
559 char *buf = info->oti_str;
560 struct osd_object *obj = osd_dt_obj(dt);
561 struct osd_device *osd = osd_obj2dev(obj);
562 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
563 struct osd_thandle *oh;
568 down_write(&obj->oo_guard);
570 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
571 GOTO(out, rc = -ENOENT);
573 LASSERT(obj->oo_dn != NULL);
575 oh = container_of0(th, struct osd_thandle, ot_super);
577 LASSERT(oh->ot_tx != NULL);
579 /* remove obj ref from index dir (it depends) */
580 zapid = osd_get_name_n_idx(env, osd, fid, buf, sizeof(info->oti_str));
581 rc = -zap_remove(osd->od_os, zapid, buf, oh->ot_tx);
583 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
584 osd->od_svname, buf, rc);
588 rc = osd_xattrs_destroy(env, obj, oh);
590 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
591 osd->od_svname, buf, rc);
595 /* Remove object from inode accounting. It is not fatal for the destroy
596 * operation if something goes wrong while updating accounting, but we
597 * still log an error message to notify the administrator */
598 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
599 obj->oo_attr.la_uid, -1, oh->ot_tx);
601 CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
602 " %d: rc = %d\n", osd->od_svname, PFID(fid),
603 obj->oo_attr.la_uid, rc);
604 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
605 obj->oo_attr.la_gid, -1, oh->ot_tx);
607 CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
608 " %d: rc = %d\n", osd->od_svname, PFID(fid),
609 obj->oo_attr.la_gid, rc);
611 oid = obj->oo_dn->dn_object;
612 if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
613 /* this may happen if the destroy wasn't declared
614 * e.g. when the object is created and then destroyed
615 * in the same transaction - we don't need additional
616 * space for destroy specifically */
617 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
618 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
620 CERROR("%s: failed to free %s %llu: rc = %d\n",
621 osd->od_svname, buf, oid, rc);
622 } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
623 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
625 CERROR("%s: failed to free %s %llu: rc = %d\n",
626 osd->od_svname, buf, oid, rc);
627 } else { /* asynchronous destroy */
628 rc = osd_object_unlinked_add(obj, oh);
632 rc = -zap_add_int(osd->od_os, osd->od_unlinkedid,
635 CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
636 osd->od_svname, buf, oid, rc);
640 /* not needed in the cache anymore */
641 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
643 obj->oo_destroyed = 1;
644 up_write(&obj->oo_guard);
648 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
650 struct osd_object *obj = osd_obj(l);
652 if (obj->oo_dn != NULL) {
653 osd_object_sa_fini(obj);
654 if (obj->oo_sa_xattr) {
655 nvlist_free(obj->oo_sa_xattr);
656 obj->oo_sa_xattr = NULL;
658 osd_dnode_rele(obj->oo_dn);
659 list_del(&obj->oo_sa_linkage);
665 * Concurrency: ->loo_object_release() is called under site spin-lock.
667 static void osd_object_release(const struct lu_env *env,
673 * Concurrency: shouldn't matter.
675 static int osd_object_print(const struct lu_env *env, void *cookie,
676 lu_printer_t p, const struct lu_object *l)
678 struct osd_object *o = osd_obj(l);
680 return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
683 static void osd_object_read_lock(const struct lu_env *env,
684 struct dt_object *dt, unsigned role)
686 struct osd_object *obj = osd_dt_obj(dt);
688 LASSERT(osd_invariant(obj));
690 down_read_nested(&obj->oo_sem, role);
693 static void osd_object_write_lock(const struct lu_env *env,
694 struct dt_object *dt, unsigned role)
696 struct osd_object *obj = osd_dt_obj(dt);
698 LASSERT(osd_invariant(obj));
700 down_write_nested(&obj->oo_sem, role);
703 static void osd_object_read_unlock(const struct lu_env *env,
704 struct dt_object *dt)
706 struct osd_object *obj = osd_dt_obj(dt);
708 LASSERT(osd_invariant(obj));
709 up_read(&obj->oo_sem);
712 static void osd_object_write_unlock(const struct lu_env *env,
713 struct dt_object *dt)
715 struct osd_object *obj = osd_dt_obj(dt);
717 LASSERT(osd_invariant(obj));
718 up_write(&obj->oo_sem);
721 static int osd_object_write_locked(const struct lu_env *env,
722 struct dt_object *dt)
724 struct osd_object *obj = osd_dt_obj(dt);
727 LASSERT(osd_invariant(obj));
729 if (down_write_trylock(&obj->oo_sem)) {
731 up_write(&obj->oo_sem);
736 static int osd_attr_get(const struct lu_env *env,
737 struct dt_object *dt,
738 struct lu_attr *attr)
740 struct osd_object *obj = osd_dt_obj(dt);
745 down_read(&obj->oo_guard);
747 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
748 GOTO(out, rc = -ENOENT);
750 LASSERT(osd_invariant(obj));
753 read_lock(&obj->oo_attr_lock);
754 *attr = obj->oo_attr;
755 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
756 attr->la_flags |= LUSTRE_ORPHAN_FL;
757 read_unlock(&obj->oo_attr_lock);
759 /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
760 * from within sa_object_size() can block on a mutex, so
761 * we can't call sa_object_size() holding rwlock */
762 sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
763 /* we do not control size of indices, so always calculate
764 * it from number of blocks reported by DMU */
765 if (S_ISDIR(attr->la_mode))
766 attr->la_size = 512 * blocks;
767 /* Block size may be not set; suggest maximal I/O transfers. */
769 blksize = osd_spa_maxblocksize(
770 dmu_objset_spa(osd_obj2dev(obj)->od_os));
772 attr->la_blksize = blksize;
773 attr->la_blocks = blocks;
774 attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
777 up_read(&obj->oo_guard);
781 /* Simple wrapper on top of qsd API which implement quota transfer for osd
782 * setattr needs. As a reminder, only the root user can change ownership of
783 * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
784 static inline int qsd_transfer(const struct lu_env *env,
785 struct qsd_instance *qsd,
786 struct lquota_trans *trans, int qtype,
787 __u64 orig_id, __u64 new_id, __u64 bspace,
788 struct lquota_id_info *qi)
792 if (unlikely(qsd == NULL))
795 LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
796 qi->lqi_type = qtype;
798 /* inode accounting */
799 qi->lqi_is_blk = false;
801 /* one more inode for the new owner ... */
802 qi->lqi_id.qid_uid = new_id;
804 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
805 if (rc == -EDQUOT || rc == -EINPROGRESS)
810 /* and one less inode for the current id */
811 qi->lqi_id.qid_uid = orig_id;;
813 /* can't get EDQUOT when reducing usage */
814 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
815 if (rc == -EINPROGRESS)
820 /* block accounting */
821 qi->lqi_is_blk = true;
823 /* more blocks for the new owner ... */
824 qi->lqi_id.qid_uid = new_id;
825 qi->lqi_space = bspace;
826 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
827 if (rc == -EDQUOT || rc == -EINPROGRESS)
832 /* and finally less blocks for the current owner */
833 qi->lqi_id.qid_uid = orig_id;
834 qi->lqi_space = -bspace;
835 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
836 /* can't get EDQUOT when reducing usage */
837 if (rc == -EINPROGRESS)
842 static int osd_declare_attr_set(const struct lu_env *env,
843 struct dt_object *dt,
844 const struct lu_attr *attr,
845 struct thandle *handle)
847 struct osd_thread_info *info = osd_oti_get(env);
848 struct osd_object *obj = osd_dt_obj(dt);
849 struct osd_device *osd = osd_obj2dev(obj);
851 struct osd_thandle *oh;
859 LASSERT(handle != NULL);
860 LASSERT(osd_invariant(obj));
862 oh = container_of0(handle, struct osd_thandle, ot_super);
864 down_read(&obj->oo_guard);
865 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
868 LASSERT(obj->oo_sa_hdl != NULL);
869 LASSERT(oh->ot_tx != NULL);
870 /* regular attributes are part of the bonus buffer */
871 /* let's check whether this object is already part of
874 for (txh = list_head(&oh->ot_tx->tx_holds); txh;
875 txh = list_next(&oh->ot_tx->tx_holds, txh)) {
876 if (txh->txh_dnode == NULL)
878 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
880 /* this object is part of the transaction already
881 * we don't need to declare bonus again */
886 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
887 if (oh->ot_tx->tx_err != 0)
888 GOTO(out, rc = -oh->ot_tx->tx_err);
890 if (attr && attr->la_valid & LA_FLAGS) {
891 /* LMA is usually a part of bonus, no need to declare
895 if (attr && (attr->la_valid & (LA_UID | LA_GID))) {
896 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
897 bspace = toqb(bspace * blksize);
900 if (attr && attr->la_valid & LA_UID) {
901 /* account for user inode tracking ZAP update */
902 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, FALSE, NULL);
904 /* quota enforcement for user */
905 if (attr->la_uid != obj->oo_attr.la_uid) {
906 rc = qsd_transfer(env, osd->od_quota_slave,
907 &oh->ot_quota_trans, USRQUOTA,
908 obj->oo_attr.la_uid, attr->la_uid,
909 bspace, &info->oti_qi);
914 if (attr && attr->la_valid & LA_GID) {
915 /* account for user inode tracking ZAP update */
916 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, FALSE, NULL);
918 /* quota enforcement for group */
919 if (attr->la_gid != obj->oo_attr.la_gid) {
920 rc = qsd_transfer(env, osd->od_quota_slave,
921 &oh->ot_quota_trans, GRPQUOTA,
922 obj->oo_attr.la_gid, attr->la_gid,
923 bspace, &info->oti_qi);
930 up_read(&obj->oo_guard);
935 * Set the attributes of an object
937 * The transaction passed to this routine must have
938 * dmu_tx_hold_bonus(tx, oid) called and then assigned
939 * to a transaction group.
941 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
942 const struct lu_attr *la, struct thandle *handle)
944 struct osd_thread_info *info = osd_oti_get(env);
945 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
946 struct osd_object *obj = osd_dt_obj(dt);
947 struct osd_device *osd = osd_obj2dev(obj);
948 struct osd_thandle *oh;
949 struct osa_attr *osa = &info->oti_osa;
950 __u64 valid = la->la_valid;
956 down_read(&obj->oo_guard);
957 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
958 GOTO(out, rc = -ENOENT);
960 LASSERT(handle != NULL);
961 LASSERT(osd_invariant(obj));
962 LASSERT(obj->oo_sa_hdl);
964 oh = container_of0(handle, struct osd_thandle, ot_super);
965 /* Assert that the transaction has been assigned to a
966 transaction group. */
967 LASSERT(oh->ot_tx->tx_txg != 0);
969 /* Only allow set size for regular file */
970 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
971 valid &= ~(LA_SIZE | LA_BLOCKS);
973 if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
976 if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
979 if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
985 if (valid & LA_FLAGS) {
986 struct lustre_mdt_attrs *lma;
989 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
990 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
991 lma = (struct lustre_mdt_attrs *)&info->oti_buf;
993 buf.lb_len = sizeof(info->oti_buf);
994 rc = osd_xattr_get(env, &obj->oo_dt, &buf,
998 le32_to_cpu(lma->lma_incompat);
1000 lustre_to_lma_flags(la->la_flags);
1002 cpu_to_le32(lma->lma_incompat);
1004 buf.lb_len = sizeof(*lma);
1005 rc = osd_xattr_set_internal(env, obj, &buf,
1011 CWARN("%s: failed to set LMA flags: rc = %d\n",
1012 osd->od_svname, rc);
1018 /* do both accounting updates outside oo_attr_lock below */
1019 if ((valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
1020 /* Update user accounting. Failure isn't fatal, but we still
1021 * log an error message */
1022 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1023 la->la_uid, 1, oh->ot_tx);
1025 CERROR("%s: failed to update accounting ZAP for user "
1026 "%d (%d)\n", osd->od_svname, la->la_uid, rc);
1027 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1028 obj->oo_attr.la_uid, -1, oh->ot_tx);
1030 CERROR("%s: failed to update accounting ZAP for user "
1031 "%d (%d)\n", osd->od_svname,
1032 obj->oo_attr.la_uid, rc);
1034 if ((valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
1035 /* Update group accounting. Failure isn't fatal, but we still
1036 * log an error message */
1037 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1038 la->la_gid, 1, oh->ot_tx);
1040 CERROR("%s: failed to update accounting ZAP for user "
1041 "%d (%d)\n", osd->od_svname, la->la_gid, rc);
1042 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1043 obj->oo_attr.la_gid, -1, oh->ot_tx);
1045 CERROR("%s: failed to update accounting ZAP for user "
1046 "%d (%d)\n", osd->od_svname,
1047 obj->oo_attr.la_gid, rc);
1050 write_lock(&obj->oo_attr_lock);
1052 if (valid & LA_ATIME) {
1053 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1054 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1057 if (valid & LA_MTIME) {
1058 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1059 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1062 if (valid & LA_CTIME) {
1063 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1064 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1067 if (valid & LA_MODE) {
1068 /* mode is stored along with type, so read it first */
1069 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1070 (la->la_mode & ~S_IFMT);
1071 osa->mode = obj->oo_attr.la_mode;
1072 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1075 if (valid & LA_SIZE) {
1076 osa->size = obj->oo_attr.la_size = la->la_size;
1077 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1080 if (valid & LA_NLINK) {
1081 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1082 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1085 if (valid & LA_RDEV) {
1086 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1087 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1090 if (valid & LA_FLAGS) {
1091 osa->flags = attrs_fs2zfs(la->la_flags);
1092 /* many flags are not supported by zfs, so ensure a good cached
1094 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1095 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1098 if (valid & LA_UID) {
1099 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1100 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1103 if (valid & LA_GID) {
1104 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1105 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1108 obj->oo_attr.la_valid |= valid;
1109 write_unlock(&obj->oo_attr_lock);
1111 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1112 rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1115 up_read(&obj->oo_guard);
1122 * XXX temporary solution.
1125 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1126 struct dt_object *parent, struct dt_object *child,
1131 ah->dah_parent = parent;
1132 ah->dah_mode = child_mode;
1134 if (parent != NULL && !dt_object_remote(parent)) {
1135 /* will help to find FID->ino at dt_insert("..") */
1136 struct osd_object *pobj = osd_dt_obj(parent);
1138 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1142 static int osd_declare_object_create(const struct lu_env *env,
1143 struct dt_object *dt,
1144 struct lu_attr *attr,
1145 struct dt_allocation_hint *hint,
1146 struct dt_object_format *dof,
1147 struct thandle *handle)
1149 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1150 struct osd_object *obj = osd_dt_obj(dt);
1151 struct osd_device *osd = osd_obj2dev(obj);
1152 struct osd_thandle *oh;
1159 switch (dof->dof_type) {
1163 if (obj->oo_dt.do_body_ops == NULL)
1164 obj->oo_dt.do_body_ops = &osd_body_ops;
1170 LASSERT(handle != NULL);
1171 oh = container_of0(handle, struct osd_thandle, ot_super);
1172 LASSERT(oh->ot_tx != NULL);
1174 /* this is the minimum set of EAs on every Lustre object */
1175 obj->oo_ea_in_bonus = ZFS_SA_BASE_ATTR_SIZE +
1176 sizeof(__u64) + /* VBR VERSION */
1177 sizeof(struct lustre_mdt_attrs); /* LMA */
1178 /* reserve 32 bytes for extra stuff like ACLs */
1179 dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1181 switch (dof->dof_type) {
1183 dt->do_index_ops = &osd_dir_ops;
1185 /* for zap create */
1186 dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1187 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1192 /* first, we'll create new object */
1193 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1201 /* and we'll add it to some mapping */
1202 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0);
1203 dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, NULL);
1205 /* we will also update inode accounting ZAPs */
1206 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, FALSE, NULL);
1207 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, FALSE, NULL);
1209 /* will help to find FID->ino mapping at dt_insert() */
1210 osd_idc_find_and_init(env, osd, obj);
1212 rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1213 false, NULL, false);
1218 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1219 sa_handle_t *sa_hdl, dmu_tx_t *tx,
1220 struct lu_attr *la, uint64_t parent)
1222 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1223 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1232 gen = dmu_tx_get_txg(tx);
1234 ZFS_TIME_ENCODE(&now, crtime);
1236 osa->atime[0] = la->la_atime;
1237 osa->ctime[0] = la->la_ctime;
1238 osa->mtime[0] = la->la_mtime;
1239 osa->mode = la->la_mode;
1240 osa->uid = la->la_uid;
1241 osa->gid = la->la_gid;
1242 osa->rdev = la->la_rdev;
1243 osa->nlink = la->la_nlink;
1244 osa->flags = attrs_fs2zfs(la->la_flags);
1245 osa->size = la->la_size;
1248 * we need to create all SA below upon object create.
1250 * XXX The attribute order matters since the accounting callback relies
1251 * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1252 * look up the UID/GID attributes. Moreover, the callback does not seem
1253 * to support the spill block.
1254 * We define attributes in the same order as SA_*_OFFSET in order to
1255 * work around the problem. See ORI-610.
1258 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1259 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1260 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1261 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1262 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1263 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1264 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1265 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1266 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1267 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1268 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1269 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1270 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1271 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1273 rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1278 static int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1279 uint64_t oid, dnode_t **dnp)
1284 /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1285 for (txh = list_tail(&tx->tx_holds); txh;
1286 txh = list_prev(&tx->tx_holds, txh)) {
1287 dnode_t *dn = txh->txh_dnode;
1292 if (dn->dn_object != oid)
1296 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1297 if (dn->dn_bonus == NULL)
1298 dbuf_create_bonus(dn);
1299 rw_exit(&dn->dn_struct_rwlock);
1303 LASSERT(dn->dn_handle);
1305 if (refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1306 refcount_add(&dn->dn_holds, tag);
1307 atomic_inc_32(&dn->dn_dbufs_count);
1313 if (unlikely(*dnp == NULL))
1314 rc = __osd_obj2dnode(env, tx->tx_objset, oid, dnp);
1320 * The transaction passed to this routine must have
1321 * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1322 * to a transaction group.
1324 int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
1325 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1327 struct osd_device *osd = osd_obj2dev(obj);
1328 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1329 dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1332 /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1333 * would get an additional ditto copy */
1334 if (unlikely(S_ISREG(la->la_mode) &&
1335 fid_seq_is_local_file(fid_seq(fid))))
1336 type = DMU_OTN_UINT8_METADATA;
1338 /* Create a new DMU object using the default dnode size. */
1339 oid = osd_dmu_object_alloc(osd->od_os, type, 0, 0, tx);
1341 LASSERT(la->la_valid & LA_MODE);
1345 return osd_find_new_dnode(env, tx, oid, dnp);
1349 * The transaction passed to this routine must have
1350 * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1351 * to a transaction group.
1353 * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1354 * This is fine for directories today, because storing the FID in the dirent
1355 * will also require a FAT ZAP. If there is a new type of micro ZAP created
1356 * then we might need to re-evaluate the use of this flag and instead do
1357 * a conversion from the different internal ZAP hash formats being used. */
1358 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1359 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1364 /* Assert that the transaction has been assigned to a
1365 transaction group. */
1366 LASSERT(tx->tx_txg != 0);
1369 oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1370 DMU_OT_DIRECTORY_CONTENTS,
1371 14, /* == ZFS fzap_default_blockshift */
1372 DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1378 return osd_find_new_dnode(env, tx, oid, dnp);
1381 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1382 struct lu_attr *la, struct osd_thandle *oh)
1387 /* Index file should be created as regular file in order not to confuse
1388 * ZPL which could interpret them as directory.
1389 * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1391 LASSERT(S_ISREG(la->la_mode));
1392 rc = __osd_zap_create(env, osd_obj2dev(obj), &dn, oh->ot_tx, la,
1393 ZAP_FLAG_UINT64_KEY);
1399 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1400 struct lu_attr *la, struct osd_thandle *oh)
1405 LASSERT(S_ISDIR(la->la_mode));
1406 rc = __osd_zap_create(env, osd_obj2dev(obj), &dn, oh->ot_tx, la, 0);
1412 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1413 struct lu_attr *la, struct osd_thandle *oh)
1415 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1416 struct osd_device *osd = osd_obj2dev(obj);
1420 LASSERT(S_ISREG(la->la_mode));
1421 rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1425 if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid)) &&
1427 /* The minimum block size must be at least page size otherwise
1428 * it will break the assumption in tgt_thread_big_cache where
1429 * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1430 * RDMA due to subpage transfer size */
1431 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1432 PAGE_SIZE, 0, oh->ot_tx);
1434 CERROR("%s: can't change blocksize: %d\n",
1435 osd->od_svname, rc);
1443 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1444 struct lu_attr *la, struct osd_thandle *oh)
1449 LASSERT(S_ISLNK(la->la_mode));
1450 rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1456 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1457 struct lu_attr *la, struct osd_thandle *oh)
1462 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1463 la->la_valid |= LA_RDEV;
1465 rc = __osd_object_create(env, obj, &dn, oh->ot_tx, la);
1471 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1472 struct osd_object *obj,
1474 struct osd_thandle *oh);
1476 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1478 osd_obj_type_f result;
1504 * Concurrency: @dt is write locked.
1506 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1507 struct lu_attr *attr,
1508 struct dt_allocation_hint *hint,
1509 struct dt_object_format *dof,
1512 struct osd_thread_info *info = osd_oti_get(env);
1513 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1514 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1515 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1516 struct osd_object *obj = osd_dt_obj(dt);
1517 struct osd_device *osd = osd_obj2dev(obj);
1518 char *buf = info->oti_str;
1519 struct osd_thandle *oh;
1521 uint64_t zapid, parent = 0;
1526 /* concurrent create declarations should not see
1527 * the object inconsistent (db, attr, etc).
1528 * in regular cases acquisition should be cheap */
1529 down_write(&obj->oo_guard);
1531 if (unlikely(dt_object_exists(dt)))
1532 GOTO(out, rc = -EEXIST);
1534 LASSERT(osd_invariant(obj));
1535 LASSERT(dof != NULL);
1537 LASSERT(th != NULL);
1538 oh = container_of0(th, struct osd_thandle, ot_super);
1541 * XXX missing: Quote handling.
1544 LASSERT(obj->oo_dn == NULL);
1546 /* to follow ZFS on-disk format we need
1547 * to initialize parent dnode properly */
1548 if (hint != NULL && hint->dah_parent != NULL &&
1549 !dt_object_remote(hint->dah_parent))
1550 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1552 /* we may fix some attributes, better do not change the source */
1553 obj->oo_attr = *attr;
1554 obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1556 dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1564 zde->zde_dnode = dn->dn_object;
1565 zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1567 zapid = osd_get_name_n_idx(env, osd, fid, buf, sizeof(info->oti_str));
1569 rc = -zap_add(osd->od_os, zapid, buf, 8, 1, zde, oh->ot_tx);
1573 /* Now add in all of the "SA" attributes */
1574 rc = -sa_handle_get(osd->od_os, dn->dn_object, NULL,
1575 SA_HDL_PRIVATE, &obj->oo_sa_hdl);
1579 /* configure new osd object */
1581 parent = parent != 0 ? parent : zapid;
1582 rc = __osd_attr_init(env, osd, obj->oo_sa_hdl, oh->ot_tx,
1583 &obj->oo_attr, parent);
1587 /* XXX: oo_lma_flags */
1588 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
1590 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
1591 if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
1592 /* no body operations for accounting objects */
1593 obj->oo_dt.do_body_ops = &osd_body_ops;
1595 rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1599 /* initialize LMA */
1600 lustre_lma_init(lma, lu_object_fid(&obj->oo_dt.do_lu), 0, 0);
1601 lustre_lma_swab(lma);
1602 rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
1603 (uchar_t *)lma, sizeof(*lma));
1606 rc = __osd_sa_xattr_update(env, obj, oh);
1609 osd_idc_find_and_init(env, osd, obj);
1611 /* Add new object to inode accounting.
1612 * Errors are not considered as fatal */
1613 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1614 (attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
1617 CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
1618 "(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
1619 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1620 (attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
1623 CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
1624 "(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
1627 if (unlikely(rc && dn)) {
1628 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
1632 up_write(&obj->oo_guard);
1636 static int osd_declare_object_ref_add(const struct lu_env *env,
1637 struct dt_object *dt,
1640 return osd_declare_attr_set(env, dt, NULL, th);
1644 * Concurrency: @dt is write locked.
1646 static int osd_object_ref_add(const struct lu_env *env,
1647 struct dt_object *dt,
1648 struct thandle *handle)
1650 struct osd_object *obj = osd_dt_obj(dt);
1651 struct osd_thandle *oh;
1652 struct osd_device *osd = osd_obj2dev(obj);
1658 down_read(&obj->oo_guard);
1659 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1660 GOTO(out, rc = -ENOENT);
1662 LASSERT(osd_invariant(obj));
1663 LASSERT(obj->oo_sa_hdl != NULL);
1665 oh = container_of0(handle, struct osd_thandle, ot_super);
1667 write_lock(&obj->oo_attr_lock);
1668 nlink = ++obj->oo_attr.la_nlink;
1669 write_unlock(&obj->oo_attr_lock);
1671 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1674 up_read(&obj->oo_guard);
1678 static int osd_declare_object_ref_del(const struct lu_env *env,
1679 struct dt_object *dt,
1680 struct thandle *handle)
1682 return osd_declare_attr_set(env, dt, NULL, handle);
1686 * Concurrency: @dt is write locked.
1688 static int osd_object_ref_del(const struct lu_env *env,
1689 struct dt_object *dt,
1690 struct thandle *handle)
1692 struct osd_object *obj = osd_dt_obj(dt);
1693 struct osd_thandle *oh;
1694 struct osd_device *osd = osd_obj2dev(obj);
1700 down_read(&obj->oo_guard);
1702 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1703 GOTO(out, rc = -ENOENT);
1705 LASSERT(osd_invariant(obj));
1706 LASSERT(obj->oo_sa_hdl != NULL);
1708 oh = container_of0(handle, struct osd_thandle, ot_super);
1709 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1711 write_lock(&obj->oo_attr_lock);
1712 nlink = --obj->oo_attr.la_nlink;
1713 write_unlock(&obj->oo_attr_lock);
1715 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1718 up_read(&obj->oo_guard);
1722 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1723 __u64 start, __u64 end)
1725 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1728 /* XXX: no other option than syncing the whole filesystem until we
1729 * support ZIL. If the object tracked the txg that it was last
1730 * modified in, it could pass that txg here instead of "0". Maybe
1731 * the changes are already committed, so no wait is needed at all? */
1732 if (!osd->od_dt_dev.dd_rdonly)
1733 txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
1738 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
1743 static struct dt_object_operations osd_obj_ops = {
1744 .do_read_lock = osd_object_read_lock,
1745 .do_write_lock = osd_object_write_lock,
1746 .do_read_unlock = osd_object_read_unlock,
1747 .do_write_unlock = osd_object_write_unlock,
1748 .do_write_locked = osd_object_write_locked,
1749 .do_attr_get = osd_attr_get,
1750 .do_declare_attr_set = osd_declare_attr_set,
1751 .do_attr_set = osd_attr_set,
1752 .do_ah_init = osd_ah_init,
1753 .do_declare_create = osd_declare_object_create,
1754 .do_create = osd_object_create,
1755 .do_declare_destroy = osd_declare_object_destroy,
1756 .do_destroy = osd_object_destroy,
1757 .do_index_try = osd_index_try,
1758 .do_declare_ref_add = osd_declare_object_ref_add,
1759 .do_ref_add = osd_object_ref_add,
1760 .do_declare_ref_del = osd_declare_object_ref_del,
1761 .do_ref_del = osd_object_ref_del,
1762 .do_xattr_get = osd_xattr_get,
1763 .do_declare_xattr_set = osd_declare_xattr_set,
1764 .do_xattr_set = osd_xattr_set,
1765 .do_declare_xattr_del = osd_declare_xattr_del,
1766 .do_xattr_del = osd_xattr_del,
1767 .do_xattr_list = osd_xattr_list,
1768 .do_object_sync = osd_object_sync,
1769 .do_invalidate = osd_invalidate,
1772 static struct lu_object_operations osd_lu_obj_ops = {
1773 .loo_object_init = osd_object_init,
1774 .loo_object_delete = osd_object_delete,
1775 .loo_object_release = osd_object_release,
1776 .loo_object_free = osd_object_free,
1777 .loo_object_print = osd_object_print,
1778 .loo_object_invariant = osd_object_invariant,
1781 static int osd_otable_it_attr_get(const struct lu_env *env,
1782 struct dt_object *dt,
1783 struct lu_attr *attr)
1789 static struct dt_object_operations osd_obj_otable_it_ops = {
1790 .do_attr_get = osd_otable_it_attr_get,
1791 .do_index_try = osd_index_try,