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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd-zfs/osd_object.c
38 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
39 * Author: Mike Pershin <tappro@whamcloud.com>
40 * Author: Johann Lombardi <johann@whamcloud.com>
43 #define DEBUG_SUBSYSTEM S_OSD
45 #include <lustre_ver.h>
46 #include <libcfs/libcfs.h>
47 #include <obd_support.h>
48 #include <lustre_net.h>
50 #include <obd_class.h>
51 #include <lustre_disk.h>
52 #include <lustre_fid.h>
54 #include "osd_internal.h"
56 #include <sys/dnode.h>
61 #include <sys/spa_impl.h>
62 #include <sys/zfs_znode.h>
63 #include <sys/dmu_tx.h>
64 #include <sys/dmu_objset.h>
65 #include <sys/dsl_prop.h>
66 #include <sys/sa_impl.h>
69 char *osd_obj_tag = "osd_object";
71 static struct dt_object_operations osd_obj_ops;
72 static struct lu_object_operations osd_lu_obj_ops;
73 extern struct dt_body_operations osd_body_ops;
74 static struct dt_object_operations osd_obj_otable_it_ops;
76 extern struct kmem_cache *osd_object_kmem;
79 osd_object_sa_fini(struct osd_object *obj)
82 sa_handle_destroy(obj->oo_sa_hdl);
83 obj->oo_sa_hdl = NULL;
88 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
92 LASSERT(obj->oo_sa_hdl == NULL);
93 LASSERT(obj->oo_db != NULL);
95 rc = -sa_handle_get(o->od_os, obj->oo_db->db_object, obj,
96 SA_HDL_PRIVATE, &obj->oo_sa_hdl);
100 /* Cache the xattr object id, valid for the life of the object */
101 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
103 obj->oo_xattr = ZFS_NO_OBJECT;
106 osd_object_sa_fini(obj);
113 * Add object to list of dirty objects in tx handle.
116 osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
118 if (!list_empty(&obj->oo_sa_linkage))
121 down(&oh->ot_sa_lock);
122 write_lock(&obj->oo_attr_lock);
123 if (likely(list_empty(&obj->oo_sa_linkage)))
124 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
125 write_unlock(&obj->oo_attr_lock);
130 * Release spill block dbuf hold for all dirty SAs.
132 void osd_object_sa_dirty_rele(struct osd_thandle *oh)
134 struct osd_object *obj;
136 down(&oh->ot_sa_lock);
137 while (!list_empty(&oh->ot_sa_list)) {
138 obj = list_entry(oh->ot_sa_list.next,
139 struct osd_object, oo_sa_linkage);
140 sa_spill_rele(obj->oo_sa_hdl);
141 write_lock(&obj->oo_attr_lock);
142 list_del_init(&obj->oo_sa_linkage);
143 write_unlock(&obj->oo_attr_lock);
149 * Update the SA and add the object to the dirty list.
151 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
152 void *buf, uint32_t buflen, struct osd_thandle *oh)
156 LASSERT(obj->oo_sa_hdl != NULL);
157 LASSERT(oh->ot_tx != NULL);
159 rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
160 osd_object_sa_dirty_add(obj, oh);
166 * Bulk update the SA and add the object to the dirty list.
169 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
170 int count, struct osd_thandle *oh)
174 LASSERT(obj->oo_sa_hdl != NULL);
175 LASSERT(oh->ot_tx != NULL);
177 rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
178 osd_object_sa_dirty_add(obj, oh);
184 * Retrieve the attributes of a DMU object
186 int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
187 struct osd_object *obj, struct lu_attr *la)
189 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
190 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
196 LASSERT(obj->oo_db != NULL);
198 rc = -sa_handle_get(o->od_os, obj->oo_db->db_object, NULL,
199 SA_HDL_PRIVATE, &sa_hdl);
203 la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
204 LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
206 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
207 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
208 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
209 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
210 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
211 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
212 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
213 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
214 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
215 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
217 rc = -sa_bulk_lookup(sa_hdl, bulk, cnt);
221 la->la_atime = osa->atime[0];
222 la->la_mtime = osa->mtime[0];
223 la->la_ctime = osa->ctime[0];
224 la->la_mode = osa->mode;
225 la->la_uid = osa->uid;
226 la->la_gid = osa->gid;
227 la->la_nlink = osa->nlink;
228 la->la_flags = attrs_zfs2fs(osa->flags);
229 la->la_size = osa->size;
231 /* Try to get extra flag from LMA. Right now, only LMAI_ORPHAN
232 * flags is stored in LMA, and it is only for orphan directory */
233 if (S_ISDIR(la->la_mode) && dt_object_exists(&obj->oo_dt)) {
234 struct osd_thread_info *info = osd_oti_get(env);
235 struct lustre_mdt_attrs *lma;
238 lma = (struct lustre_mdt_attrs *)info->oti_buf;
240 buf.lb_len = sizeof(info->oti_buf);
241 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
244 lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
246 lma_to_lustre_flags(lma->lma_incompat);
248 } else if (rc == -ENODATA) {
253 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
254 rc = -sa_lookup(sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
257 la->la_rdev = osa->rdev;
258 la->la_valid |= LA_RDEV;
261 sa_handle_destroy(sa_hdl);
266 int __osd_obj2dbuf(const struct lu_env *env, objset_t *os,
267 uint64_t oid, dmu_buf_t **dbp)
269 dmu_object_info_t *doi = &osd_oti_get(env)->oti_doi;
272 rc = -sa_buf_hold(os, oid, osd_obj_tag, dbp);
276 dmu_object_info_from_db(*dbp, doi);
277 if (unlikely (oid != DMU_USERUSED_OBJECT &&
278 oid != DMU_GROUPUSED_OBJECT && doi->doi_bonus_type != DMU_OT_SA)) {
279 sa_buf_rele(*dbp, osd_obj_tag);
285 LASSERT((*dbp)->db_object == oid);
286 LASSERT((*dbp)->db_offset == -1);
287 LASSERT((*dbp)->db_data != NULL);
293 * Concurrency: no concurrent access is possible that early in object
296 struct lu_object *osd_object_alloc(const struct lu_env *env,
297 const struct lu_object_header *hdr,
300 struct osd_object *mo;
302 OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
306 l = &mo->oo_dt.do_lu;
307 dt_object_init(&mo->oo_dt, NULL, d);
308 mo->oo_dt.do_ops = &osd_obj_ops;
309 l->lo_ops = &osd_lu_obj_ops;
310 INIT_LIST_HEAD(&mo->oo_sa_linkage);
311 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
312 init_rwsem(&mo->oo_sem);
313 init_rwsem(&mo->oo_guard);
314 rwlock_init(&mo->oo_attr_lock);
315 mo->oo_destroy = OSD_DESTROY_NONE;
323 * Concurrency: shouldn't matter.
325 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
327 struct osd_device *osd = osd_obj2dev(obj);
328 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
332 if (obj->oo_db == NULL)
337 rc = osd_object_sa_init(obj, osd);
341 /* cache attrs in object */
342 rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
346 if (likely(!fid_is_acct(fid)))
347 /* no body operations for accounting objects */
348 obj->oo_dt.do_body_ops = &osd_body_ops;
351 * initialize object before marking it existing
353 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
356 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
361 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
363 struct osd_thread_info *info = osd_oti_get(env);
366 struct lustre_mdt_attrs *lma;
369 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
370 lma = (struct lustre_mdt_attrs *)info->oti_buf;
372 buf.lb_len = sizeof(info->oti_buf);
374 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
377 lustre_lma_swab(lma);
378 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
379 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
380 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
381 "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
382 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
383 PFID(lu_object_fid(&obj->oo_dt.do_lu)));
386 } else if (rc == -ENODATA) {
387 /* haven't initialize LMA xattr */
395 * Concurrency: no concurrent access is possible that early in object
398 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
399 const struct lu_object_conf *conf)
401 struct osd_object *obj = osd_obj(l);
402 struct osd_device *osd = osd_obj2dev(obj);
407 LASSERT(osd_invariant(obj));
409 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
410 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
411 l->lo_header->loh_attr |= LOHA_EXISTS;
415 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
418 rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
420 LASSERT(obj->oo_db == NULL);
421 rc = __osd_obj2dbuf(env, osd->od_os, oid, &obj->oo_db);
422 /* EEXIST will be returned if object is being deleted in ZFS */
428 CERROR("%s: lookup "DFID"/"LPX64" failed: rc = %d\n",
429 osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
433 rc = osd_object_init0(env, obj);
437 rc = osd_check_lma(env, obj);
440 } else if (rc == -ENOENT) {
443 LASSERT(osd_invariant(obj));
449 * Concurrency: no concurrent access is possible that late in object
452 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
454 struct osd_object *obj = osd_obj(l);
456 LASSERT(osd_invariant(obj));
458 dt_object_fini(&obj->oo_dt);
459 OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
463 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
467 LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
469 /* the object is supposed to be exclusively locked by
470 * the caller (osd_object_destroy()), while the transaction
471 * (oh) is per-thread and not shared */
472 if (likely(list_empty(&obj->oo_unlinked_linkage))) {
473 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
480 /* Default to max data size covered by a level-1 indirect block */
481 static unsigned long osd_sync_destroy_max_size =
482 1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
483 module_param(osd_sync_destroy_max_size, ulong, 0444);
484 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
487 osd_object_set_destroy_type(struct osd_object *obj)
490 * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
491 * only once and use it consistently thereafter.
493 down_write(&obj->oo_guard);
494 if (obj->oo_destroy == OSD_DESTROY_NONE) {
495 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
496 obj->oo_destroy = OSD_DESTROY_SYNC;
497 else /* Larger objects are destroyed asynchronously */
498 obj->oo_destroy = OSD_DESTROY_ASYNC;
500 up_write(&obj->oo_guard);
503 static int osd_declare_object_destroy(const struct lu_env *env,
504 struct dt_object *dt,
507 char *buf = osd_oti_get(env)->oti_str;
508 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
509 struct osd_object *obj = osd_dt_obj(dt);
510 struct osd_device *osd = osd_obj2dev(obj);
511 struct osd_thandle *oh;
517 LASSERT(dt_object_exists(dt));
519 oh = container_of0(th, struct osd_thandle, ot_super);
520 LASSERT(oh->ot_tx != NULL);
522 /* declare that we'll remove object from fid-dnode mapping */
523 zapid = osd_get_name_n_idx(env, osd, fid, buf);
524 dmu_tx_hold_bonus(oh->ot_tx, zapid);
525 dmu_tx_hold_zap(oh->ot_tx, zapid, FALSE, buf);
527 osd_declare_xattrs_destroy(env, obj, oh);
529 /* declare that we'll remove object from inode accounting ZAPs */
530 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
531 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, FALSE, buf);
532 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
533 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, FALSE, buf);
536 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
537 obj->oo_attr.la_gid, -1, oh, false, NULL, false);
541 /* data to be truncated */
542 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
543 obj->oo_attr.la_gid, 0, oh, true, NULL, false);
547 osd_object_set_destroy_type(obj);
548 if (obj->oo_destroy == OSD_DESTROY_SYNC)
549 dmu_tx_hold_free(oh->ot_tx, obj->oo_db->db_object,
552 dmu_tx_hold_zap(oh->ot_tx, osd->od_unlinkedid, TRUE, NULL);
557 static int osd_object_destroy(const struct lu_env *env,
558 struct dt_object *dt, struct thandle *th)
560 char *buf = osd_oti_get(env)->oti_str;
561 struct osd_object *obj = osd_dt_obj(dt);
562 struct osd_device *osd = osd_obj2dev(obj);
563 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
564 struct osd_thandle *oh;
569 down_write(&obj->oo_guard);
571 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
572 GOTO(out, rc = -ENOENT);
574 LASSERT(obj->oo_db != NULL);
576 oh = container_of0(th, struct osd_thandle, ot_super);
578 LASSERT(oh->ot_tx != NULL);
580 /* remove obj ref from index dir (it depends) */
581 zapid = osd_get_name_n_idx(env, osd, fid, buf);
582 rc = -zap_remove(osd->od_os, zapid, buf, oh->ot_tx);
584 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
585 osd->od_svname, buf, rc);
589 rc = osd_xattrs_destroy(env, obj, oh);
591 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
592 osd->od_svname, buf, rc);
596 /* Remove object from inode accounting. It is not fatal for the destroy
597 * operation if something goes wrong while updating accounting, but we
598 * still log an error message to notify the administrator */
599 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
600 obj->oo_attr.la_uid, -1, oh->ot_tx);
602 CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
603 " %d: rc = %d\n", osd->od_svname, PFID(fid),
604 obj->oo_attr.la_uid, rc);
605 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
606 obj->oo_attr.la_gid, -1, oh->ot_tx);
608 CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
609 " %d: rc = %d\n", osd->od_svname, PFID(fid),
610 obj->oo_attr.la_gid, rc);
612 oid = obj->oo_db->db_object;
613 if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
614 /* this may happen if the destroy wasn't declared
615 * e.g. when the object is created and then destroyed
616 * in the same transaction - we don't need additional
617 * space for destroy specifically */
618 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
619 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
621 CERROR("%s: failed to free %s "LPU64": rc = %d\n",
622 osd->od_svname, buf, oid, rc);
623 } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
624 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
626 CERROR("%s: failed to free %s "LPU64": rc = %d\n",
627 osd->od_svname, buf, oid, rc);
628 } else { /* asynchronous destroy */
629 rc = osd_object_unlinked_add(obj, oh);
633 rc = -zap_add_int(osd->od_os, osd->od_unlinkedid,
636 CERROR("%s: zap_add_int() failed %s "LPU64": rc = %d\n",
637 osd->od_svname, buf, oid, rc);
641 /* not needed in the cache anymore */
642 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
644 obj->oo_destroyed = 1;
645 up_write(&obj->oo_guard);
649 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
651 struct osd_object *obj = osd_obj(l);
653 if (obj->oo_db != NULL) {
654 osd_object_sa_fini(obj);
655 if (obj->oo_sa_xattr) {
656 nvlist_free(obj->oo_sa_xattr);
657 obj->oo_sa_xattr = NULL;
659 sa_buf_rele(obj->oo_db, osd_obj_tag);
660 list_del(&obj->oo_sa_linkage);
666 * Concurrency: ->loo_object_release() is called under site spin-lock.
668 static void osd_object_release(const struct lu_env *env,
674 * Concurrency: shouldn't matter.
676 static int osd_object_print(const struct lu_env *env, void *cookie,
677 lu_printer_t p, const struct lu_object *l)
679 struct osd_object *o = osd_obj(l);
681 return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
684 static void osd_object_read_lock(const struct lu_env *env,
685 struct dt_object *dt, unsigned role)
687 struct osd_object *obj = osd_dt_obj(dt);
689 LASSERT(osd_invariant(obj));
691 down_read_nested(&obj->oo_sem, role);
694 static void osd_object_write_lock(const struct lu_env *env,
695 struct dt_object *dt, unsigned role)
697 struct osd_object *obj = osd_dt_obj(dt);
699 LASSERT(osd_invariant(obj));
701 down_write_nested(&obj->oo_sem, role);
704 static void osd_object_read_unlock(const struct lu_env *env,
705 struct dt_object *dt)
707 struct osd_object *obj = osd_dt_obj(dt);
709 LASSERT(osd_invariant(obj));
710 up_read(&obj->oo_sem);
713 static void osd_object_write_unlock(const struct lu_env *env,
714 struct dt_object *dt)
716 struct osd_object *obj = osd_dt_obj(dt);
718 LASSERT(osd_invariant(obj));
719 up_write(&obj->oo_sem);
722 static int osd_object_write_locked(const struct lu_env *env,
723 struct dt_object *dt)
725 struct osd_object *obj = osd_dt_obj(dt);
728 LASSERT(osd_invariant(obj));
730 if (down_write_trylock(&obj->oo_sem)) {
732 up_write(&obj->oo_sem);
737 static int osd_attr_get(const struct lu_env *env,
738 struct dt_object *dt,
739 struct lu_attr *attr)
741 struct osd_object *obj = osd_dt_obj(dt);
746 down_read(&obj->oo_guard);
748 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
749 GOTO(out, rc = -ENOENT);
751 LASSERT(osd_invariant(obj));
754 read_lock(&obj->oo_attr_lock);
755 *attr = obj->oo_attr;
756 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
757 attr->la_flags |= LUSTRE_ORPHAN_FL;
758 read_unlock(&obj->oo_attr_lock);
760 /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
761 * from within sa_object_size() can block on a mutex, so
762 * we can't call sa_object_size() holding rwlock */
763 sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
764 /* we do not control size of indices, so always calculate
765 * it from number of blocks reported by DMU */
766 if (S_ISDIR(attr->la_mode))
767 attr->la_size = 512 * blocks;
768 /* Block size may be not set; suggest maximal I/O transfers. */
770 blksize = osd_spa_maxblocksize(
771 dmu_objset_spa(osd_obj2dev(obj)->od_os));
773 attr->la_blksize = blksize;
774 attr->la_blocks = blocks;
775 attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
778 up_read(&obj->oo_guard);
782 /* Simple wrapper on top of qsd API which implement quota transfer for osd
783 * setattr needs. As a reminder, only the root user can change ownership of
784 * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
785 static inline int qsd_transfer(const struct lu_env *env,
786 struct qsd_instance *qsd,
787 struct lquota_trans *trans, int qtype,
788 __u64 orig_id, __u64 new_id, __u64 bspace,
789 struct lquota_id_info *qi)
793 if (unlikely(qsd == NULL))
796 LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
797 qi->lqi_type = qtype;
799 /* inode accounting */
800 qi->lqi_is_blk = false;
802 /* one more inode for the new owner ... */
803 qi->lqi_id.qid_uid = new_id;
805 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
806 if (rc == -EDQUOT || rc == -EINPROGRESS)
811 /* and one less inode for the current id */
812 qi->lqi_id.qid_uid = orig_id;;
814 /* can't get EDQUOT when reducing usage */
815 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
816 if (rc == -EINPROGRESS)
821 /* block accounting */
822 qi->lqi_is_blk = true;
824 /* more blocks for the new owner ... */
825 qi->lqi_id.qid_uid = new_id;
826 qi->lqi_space = bspace;
827 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
828 if (rc == -EDQUOT || rc == -EINPROGRESS)
833 /* and finally less blocks for the current owner */
834 qi->lqi_id.qid_uid = orig_id;
835 qi->lqi_space = -bspace;
836 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
837 /* can't get EDQUOT when reducing usage */
838 if (rc == -EINPROGRESS)
843 static int osd_declare_attr_set(const struct lu_env *env,
844 struct dt_object *dt,
845 const struct lu_attr *attr,
846 struct thandle *handle)
848 struct osd_thread_info *info = osd_oti_get(env);
849 char *buf = osd_oti_get(env)->oti_str;
850 struct osd_object *obj = osd_dt_obj(dt);
851 struct osd_device *osd = osd_obj2dev(obj);
852 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 dmu_tx_hold_sa(oh->ot_tx, obj->oo_sa_hdl, 0);
871 if (oh->ot_tx->tx_err != 0)
872 GOTO(out, rc = -oh->ot_tx->tx_err);
874 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
875 bspace = toqb(bspace * blksize);
877 __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
880 if (attr && attr->la_valid & LA_UID) {
881 /* account for user inode tracking ZAP update */
882 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
883 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
885 /* quota enforcement for user */
886 if (attr->la_uid != obj->oo_attr.la_uid) {
887 rc = qsd_transfer(env, osd->od_quota_slave,
888 &oh->ot_quota_trans, USRQUOTA,
889 obj->oo_attr.la_uid, attr->la_uid,
890 bspace, &info->oti_qi);
895 if (attr && attr->la_valid & LA_GID) {
896 /* account for user inode tracking ZAP update */
897 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
898 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
900 /* quota enforcement for group */
901 if (attr->la_gid != obj->oo_attr.la_gid) {
902 rc = qsd_transfer(env, osd->od_quota_slave,
903 &oh->ot_quota_trans, GRPQUOTA,
904 obj->oo_attr.la_gid, attr->la_gid,
905 bspace, &info->oti_qi);
912 up_read(&obj->oo_guard);
917 * Set the attributes of an object
919 * The transaction passed to this routine must have
920 * dmu_tx_hold_bonus(tx, oid) called and then assigned
921 * to a transaction group.
923 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
924 const struct lu_attr *la, struct thandle *handle)
926 struct osd_thread_info *info = osd_oti_get(env);
927 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
928 struct osd_object *obj = osd_dt_obj(dt);
929 struct osd_device *osd = osd_obj2dev(obj);
930 struct osd_thandle *oh;
931 struct osa_attr *osa = &info->oti_osa;
932 __u64 valid = la->la_valid;
938 down_read(&obj->oo_guard);
939 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
940 GOTO(out, rc = -ENOENT);
942 LASSERT(handle != NULL);
943 LASSERT(osd_invariant(obj));
944 LASSERT(obj->oo_sa_hdl);
946 oh = container_of0(handle, struct osd_thandle, ot_super);
947 /* Assert that the transaction has been assigned to a
948 transaction group. */
949 LASSERT(oh->ot_tx->tx_txg != 0);
951 /* Only allow set size for regular file */
952 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
953 valid &= ~(LA_SIZE | LA_BLOCKS);
955 if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
958 if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
961 if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
967 if (valid & LA_FLAGS) {
968 struct lustre_mdt_attrs *lma;
971 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
972 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
973 lma = (struct lustre_mdt_attrs *)&info->oti_buf;
975 buf.lb_len = sizeof(info->oti_buf);
976 rc = osd_xattr_get(env, &obj->oo_dt, &buf,
980 le32_to_cpu(lma->lma_incompat);
982 lustre_to_lma_flags(la->la_flags);
984 cpu_to_le32(lma->lma_incompat);
986 buf.lb_len = sizeof(*lma);
987 rc = osd_xattr_set_internal(env, obj, &buf,
993 CWARN("%s: failed to set LMA flags: rc = %d\n",
1000 /* do both accounting updates outside oo_attr_lock below */
1001 if ((valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
1002 /* Update user accounting. Failure isn't fatal, but we still
1003 * log an error message */
1004 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1005 la->la_uid, 1, oh->ot_tx);
1007 CERROR("%s: failed to update accounting ZAP for user "
1008 "%d (%d)\n", osd->od_svname, la->la_uid, rc);
1009 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1010 obj->oo_attr.la_uid, -1, oh->ot_tx);
1012 CERROR("%s: failed to update accounting ZAP for user "
1013 "%d (%d)\n", osd->od_svname,
1014 obj->oo_attr.la_uid, rc);
1016 if ((valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
1017 /* Update group accounting. Failure isn't fatal, but we still
1018 * log an error message */
1019 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1020 la->la_gid, 1, oh->ot_tx);
1022 CERROR("%s: failed to update accounting ZAP for user "
1023 "%d (%d)\n", osd->od_svname, la->la_gid, rc);
1024 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1025 obj->oo_attr.la_gid, -1, oh->ot_tx);
1027 CERROR("%s: failed to update accounting ZAP for user "
1028 "%d (%d)\n", osd->od_svname,
1029 obj->oo_attr.la_gid, rc);
1032 write_lock(&obj->oo_attr_lock);
1034 if (valid & LA_ATIME) {
1035 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1036 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1039 if (valid & LA_MTIME) {
1040 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1041 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1044 if (valid & LA_CTIME) {
1045 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1046 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1049 if (valid & LA_MODE) {
1050 /* mode is stored along with type, so read it first */
1051 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1052 (la->la_mode & ~S_IFMT);
1053 osa->mode = obj->oo_attr.la_mode;
1054 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1057 if (valid & LA_SIZE) {
1058 osa->size = obj->oo_attr.la_size = la->la_size;
1059 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1062 if (valid & LA_NLINK) {
1063 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1064 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1067 if (valid & LA_RDEV) {
1068 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1069 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1072 if (valid & LA_FLAGS) {
1073 osa->flags = attrs_fs2zfs(la->la_flags);
1074 /* many flags are not supported by zfs, so ensure a good cached
1076 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1077 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1080 if (valid & LA_UID) {
1081 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1082 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1085 if (valid & LA_GID) {
1086 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1087 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1090 obj->oo_attr.la_valid |= valid;
1091 write_unlock(&obj->oo_attr_lock);
1093 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1094 rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1097 up_read(&obj->oo_guard);
1104 * XXX temporary solution.
1107 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1108 struct dt_object *parent, struct dt_object *child,
1113 ah->dah_parent = parent;
1114 ah->dah_mode = child_mode;
1117 static int osd_declare_object_create(const struct lu_env *env,
1118 struct dt_object *dt,
1119 struct lu_attr *attr,
1120 struct dt_allocation_hint *hint,
1121 struct dt_object_format *dof,
1122 struct thandle *handle)
1124 char *buf = osd_oti_get(env)->oti_str;
1125 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1126 struct osd_object *obj = osd_dt_obj(dt);
1127 struct osd_device *osd = osd_obj2dev(obj);
1128 struct osd_thandle *oh;
1135 switch (dof->dof_type) {
1139 if (obj->oo_dt.do_body_ops == NULL)
1140 obj->oo_dt.do_body_ops = &osd_body_ops;
1146 LASSERT(handle != NULL);
1147 oh = container_of0(handle, struct osd_thandle, ot_super);
1148 LASSERT(oh->ot_tx != NULL);
1150 switch (dof->dof_type) {
1152 dt->do_index_ops = &osd_dir_ops;
1154 /* for zap create */
1155 dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, 1, NULL);
1160 /* first, we'll create new object */
1161 dmu_tx_hold_bonus(oh->ot_tx, DMU_NEW_OBJECT);
1169 /* and we'll add it to some mapping */
1170 zapid = osd_get_name_n_idx(env, osd, fid, buf);
1171 dmu_tx_hold_bonus(oh->ot_tx, zapid);
1172 dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, buf);
1174 /* we will also update inode accounting ZAPs */
1175 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
1176 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
1177 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
1178 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
1180 dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
1182 __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
1183 XATTR_NAME_LMA, oh);
1185 rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1186 false, NULL, false);
1190 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1191 uint64_t oid, dmu_tx_t *tx, struct lu_attr *la,
1194 sa_handle_t *sa_hdl;
1195 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1196 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1204 gen = dmu_tx_get_txg(tx);
1206 ZFS_TIME_ENCODE(&now, crtime);
1208 osa->atime[0] = la->la_atime;
1209 osa->ctime[0] = la->la_ctime;
1210 osa->mtime[0] = la->la_mtime;
1211 osa->mode = la->la_mode;
1212 osa->uid = la->la_uid;
1213 osa->gid = la->la_gid;
1214 osa->rdev = la->la_rdev;
1215 osa->nlink = la->la_nlink;
1216 osa->flags = attrs_fs2zfs(la->la_flags);
1217 osa->size = la->la_size;
1219 /* Now add in all of the "SA" attributes */
1220 rc = -sa_handle_get(osd->od_os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
1225 * we need to create all SA below upon object create.
1227 * XXX The attribute order matters since the accounting callback relies
1228 * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1229 * look up the UID/GID attributes. Moreover, the callback does not seem
1230 * to support the spill block.
1231 * We define attributes in the same order as SA_*_OFFSET in order to
1232 * work around the problem. See ORI-610.
1235 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1236 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1237 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1238 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1239 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1240 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1241 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1242 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1243 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1244 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1245 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1246 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1247 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1248 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1250 rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1252 sa_handle_destroy(sa_hdl);
1257 * The transaction passed to this routine must have
1258 * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1259 * to a transaction group.
1261 int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
1262 dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la,
1267 struct osd_device *osd = osd_obj2dev(obj);
1268 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1269 dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1271 /* Assert that the transaction has been assigned to a
1272 transaction group. */
1273 LASSERT(tx->tx_txg != 0);
1275 /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1276 * would get an additional ditto copy */
1277 if (unlikely(S_ISREG(la->la_mode) &&
1278 fid_seq_is_local_file(fid_seq(fid))))
1279 type = DMU_OTN_UINT8_METADATA;
1281 /* Create a new DMU object using the default dnode size. */
1282 oid = osd_dmu_object_alloc(osd->od_os, type, 0, 0, tx);
1283 rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, dbp);
1284 LASSERTF(rc == 0, "sa_buf_hold "LPU64" failed: %d\n", oid, rc);
1286 LASSERT(la->la_valid & LA_MODE);
1290 rc = __osd_attr_init(env, osd, oid, tx, la, parent);
1292 sa_buf_rele(*dbp, osd_obj_tag);
1294 dmu_object_free(osd->od_os, oid, tx);
1302 * The transaction passed to this routine must have
1303 * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1304 * to a transaction group.
1306 * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1307 * This is fine for directories today, because storing the FID in the dirent
1308 * will also require a FAT ZAP. If there is a new type of micro ZAP created
1309 * then we might need to re-evaluate the use of this flag and instead do
1310 * a conversion from the different internal ZAP hash formats being used. */
1311 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1312 dmu_buf_t **zap_dbp, dmu_tx_t *tx,
1313 struct lu_attr *la, uint64_t parent, zap_flags_t flags)
1318 /* Assert that the transaction has been assigned to a
1319 transaction group. */
1320 LASSERT(tx->tx_txg != 0);
1322 oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1323 DMU_OT_DIRECTORY_CONTENTS,
1324 14, /* == ZFS fzap_default_blockshift */
1325 DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1328 rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, zap_dbp);
1332 LASSERT(la->la_valid & LA_MODE);
1336 return __osd_attr_init(env, osd, oid, tx, la, parent);
1339 static dmu_buf_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1340 struct lu_attr *la, uint64_t parent,
1341 struct osd_thandle *oh)
1346 /* Index file should be created as regular file in order not to confuse
1347 * ZPL which could interpret them as directory.
1348 * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1350 LASSERT(S_ISREG(la->la_mode));
1351 rc = __osd_zap_create(env, osd_obj2dev(obj), &db, oh->ot_tx, la, parent,
1352 ZAP_FLAG_UINT64_KEY);
1358 static dmu_buf_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1359 struct lu_attr *la, uint64_t parent,
1360 struct osd_thandle *oh)
1365 LASSERT(S_ISDIR(la->la_mode));
1366 rc = __osd_zap_create(env, osd_obj2dev(obj), &db,
1367 oh->ot_tx, la, parent, 0);
1373 static dmu_buf_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1374 struct lu_attr *la, uint64_t parent,
1375 struct osd_thandle *oh)
1377 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1380 struct osd_device *osd = osd_obj2dev(obj);
1382 LASSERT(S_ISREG(la->la_mode));
1383 rc = __osd_object_create(env, obj, &db, oh->ot_tx, la, parent);
1388 * XXX: This heuristic is non-optimal. It would be better to
1389 * increase the blocksize up to osd->od_max_blksz during the write.
1390 * This is exactly how the ZPL behaves and it ensures that the right
1391 * blocksize is selected based on the file size rather than the
1392 * making broad assumptions based on the osd type.
1394 if ((fid_is_idif(fid) || fid_is_norm(fid)) && osd->od_is_ost) {
1395 rc = -dmu_object_set_blocksize(osd->od_os, db->db_object,
1396 osd->od_max_blksz, 0, oh->ot_tx);
1398 CERROR("%s: can't change blocksize: %d\n",
1399 osd->od_svname, rc);
1407 static dmu_buf_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1408 struct lu_attr *la, uint64_t parent,
1409 struct osd_thandle *oh)
1414 LASSERT(S_ISLNK(la->la_mode));
1415 rc = __osd_object_create(env, obj, &db, oh->ot_tx, la, parent);
1421 static dmu_buf_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1422 struct lu_attr *la, uint64_t parent,
1423 struct osd_thandle *oh)
1428 la->la_valid = LA_MODE;
1429 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1430 la->la_valid |= LA_RDEV;
1432 rc = __osd_object_create(env, obj, &db, oh->ot_tx, la, parent);
1438 typedef dmu_buf_t *(*osd_obj_type_f)(const struct lu_env *env,
1439 struct osd_object *obj,
1442 struct osd_thandle *oh);
1444 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1446 osd_obj_type_f result;
1472 * Primitives for directory (i.e. ZAP) handling
1474 static inline int osd_init_lma(const struct lu_env *env, struct osd_object *obj,
1475 const struct lu_fid *fid, struct osd_thandle *oh)
1477 struct osd_thread_info *info = osd_oti_get(env);
1478 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1482 lustre_lma_init(lma, fid, 0, 0);
1483 lustre_lma_swab(lma);
1485 buf.lb_len = sizeof(*lma);
1487 rc = osd_xattr_set_internal(env, obj, &buf, XATTR_NAME_LMA,
1488 LU_XATTR_CREATE, oh);
1494 * Concurrency: @dt is write locked.
1496 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1497 struct lu_attr *attr,
1498 struct dt_allocation_hint *hint,
1499 struct dt_object_format *dof,
1502 struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
1503 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1504 struct osd_object *obj = osd_dt_obj(dt);
1505 struct osd_device *osd = osd_obj2dev(obj);
1506 char *buf = osd_oti_get(env)->oti_str;
1507 struct osd_thandle *oh;
1514 /* concurrent create declarations should not see
1515 * the object inconsistent (db, attr, etc).
1516 * in regular cases acquisition should be cheap */
1517 down_write(&obj->oo_guard);
1519 if (unlikely(dt_object_exists(dt)))
1520 GOTO(out, rc = -EEXIST);
1522 LASSERT(osd_invariant(obj));
1523 LASSERT(dof != NULL);
1525 LASSERT(th != NULL);
1526 oh = container_of0(th, struct osd_thandle, ot_super);
1529 * XXX missing: Quote handling.
1532 LASSERT(obj->oo_db == NULL);
1534 /* to follow ZFS on-disk format we need
1535 * to initialize parent dnode properly */
1537 if (hint != NULL && hint->dah_parent != NULL &&
1538 !dt_object_remote(hint->dah_parent))
1539 zapid = osd_dt_obj(hint->dah_parent)->oo_db->db_object;
1541 db = osd_create_type_f(dof->dof_type)(env, obj, attr, zapid, oh);
1543 GOTO(out, rc = PTR_ERR(db));
1546 zde->zde_dnode = db->db_object;
1547 zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1549 zapid = osd_get_name_n_idx(env, osd, fid, buf);
1551 rc = -zap_add(osd->od_os, zapid, buf, 8, 1, zde, oh->ot_tx);
1555 /* Add new object to inode accounting.
1556 * Errors are not considered as fatal */
1557 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1558 (attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
1561 CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
1562 "(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
1563 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1564 (attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
1567 CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
1568 "(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
1570 /* configure new osd object */
1572 rc = osd_object_init0(env, obj);
1573 LASSERT(ergo(rc == 0, dt_object_exists(dt)));
1574 LASSERT(osd_invariant(obj));
1576 rc = osd_init_lma(env, obj, fid, oh);
1578 CERROR("%s: can not set LMA on "DFID": rc = %d\n",
1579 osd->od_svname, PFID(fid), rc);
1582 up_write(&obj->oo_guard);
1586 static int osd_declare_object_ref_add(const struct lu_env *env,
1587 struct dt_object *dt,
1590 return osd_declare_attr_set(env, dt, NULL, th);
1594 * Concurrency: @dt is write locked.
1596 static int osd_object_ref_add(const struct lu_env *env,
1597 struct dt_object *dt,
1598 struct thandle *handle)
1600 struct osd_object *obj = osd_dt_obj(dt);
1601 struct osd_thandle *oh;
1602 struct osd_device *osd = osd_obj2dev(obj);
1608 down_read(&obj->oo_guard);
1609 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1610 GOTO(out, rc = -ENOENT);
1612 LASSERT(osd_invariant(obj));
1613 LASSERT(obj->oo_sa_hdl != NULL);
1615 oh = container_of0(handle, struct osd_thandle, ot_super);
1617 write_lock(&obj->oo_attr_lock);
1618 nlink = ++obj->oo_attr.la_nlink;
1619 write_unlock(&obj->oo_attr_lock);
1621 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1624 up_read(&obj->oo_guard);
1628 static int osd_declare_object_ref_del(const struct lu_env *env,
1629 struct dt_object *dt,
1630 struct thandle *handle)
1632 return osd_declare_attr_set(env, dt, NULL, handle);
1636 * Concurrency: @dt is write locked.
1638 static int osd_object_ref_del(const struct lu_env *env,
1639 struct dt_object *dt,
1640 struct thandle *handle)
1642 struct osd_object *obj = osd_dt_obj(dt);
1643 struct osd_thandle *oh;
1644 struct osd_device *osd = osd_obj2dev(obj);
1650 down_read(&obj->oo_guard);
1652 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1653 GOTO(out, rc = -ENOENT);
1655 LASSERT(osd_invariant(obj));
1656 LASSERT(obj->oo_sa_hdl != NULL);
1658 oh = container_of0(handle, struct osd_thandle, ot_super);
1659 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1661 write_lock(&obj->oo_attr_lock);
1662 nlink = --obj->oo_attr.la_nlink;
1663 write_unlock(&obj->oo_attr_lock);
1665 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1668 up_read(&obj->oo_guard);
1672 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1673 __u64 start, __u64 end)
1675 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1678 /* XXX: no other option than syncing the whole filesystem until we
1679 * support ZIL. If the object tracked the txg that it was last
1680 * modified in, it could pass that txg here instead of "0". Maybe
1681 * the changes are already committed, so no wait is needed at all? */
1682 txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
1687 static struct dt_object_operations osd_obj_ops = {
1688 .do_read_lock = osd_object_read_lock,
1689 .do_write_lock = osd_object_write_lock,
1690 .do_read_unlock = osd_object_read_unlock,
1691 .do_write_unlock = osd_object_write_unlock,
1692 .do_write_locked = osd_object_write_locked,
1693 .do_attr_get = osd_attr_get,
1694 .do_declare_attr_set = osd_declare_attr_set,
1695 .do_attr_set = osd_attr_set,
1696 .do_ah_init = osd_ah_init,
1697 .do_declare_create = osd_declare_object_create,
1698 .do_create = osd_object_create,
1699 .do_declare_destroy = osd_declare_object_destroy,
1700 .do_destroy = osd_object_destroy,
1701 .do_index_try = osd_index_try,
1702 .do_declare_ref_add = osd_declare_object_ref_add,
1703 .do_ref_add = osd_object_ref_add,
1704 .do_declare_ref_del = osd_declare_object_ref_del,
1705 .do_ref_del = osd_object_ref_del,
1706 .do_xattr_get = osd_xattr_get,
1707 .do_declare_xattr_set = osd_declare_xattr_set,
1708 .do_xattr_set = osd_xattr_set,
1709 .do_declare_xattr_del = osd_declare_xattr_del,
1710 .do_xattr_del = osd_xattr_del,
1711 .do_xattr_list = osd_xattr_list,
1712 .do_object_sync = osd_object_sync,
1715 static struct lu_object_operations osd_lu_obj_ops = {
1716 .loo_object_init = osd_object_init,
1717 .loo_object_delete = osd_object_delete,
1718 .loo_object_release = osd_object_release,
1719 .loo_object_free = osd_object_free,
1720 .loo_object_print = osd_object_print,
1721 .loo_object_invariant = osd_object_invariant,
1724 static int osd_otable_it_attr_get(const struct lu_env *env,
1725 struct dt_object *dt,
1726 struct lu_attr *attr)
1732 static struct dt_object_operations osd_obj_otable_it_ops = {
1733 .do_attr_get = osd_otable_it_attr_get,
1734 .do_index_try = osd_index_try,