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, 2017, 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 <libcfs/libcfs.h>
42 #include <obd_support.h>
43 #include <lustre_net.h>
45 #include <obd_class.h>
46 #include <lustre_disk.h>
47 #include <lustre_fid.h>
49 #include "osd_internal.h"
51 #include <sys/dnode.h>
56 #include <sys/spa_impl.h>
57 #include <sys/zfs_znode.h>
58 #include <sys/dmu_tx.h>
59 #include <sys/dmu_objset.h>
60 #include <sys/dsl_prop.h>
61 #include <sys/sa_impl.h>
64 char *osd_obj_tag = "osd_object";
65 static int osd_object_sync_delay_us = -1;
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 = osd_sa_handle_get(obj);
95 /* Cache the xattr object id, valid for the life of the object */
96 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
98 obj->oo_xattr = ZFS_NO_OBJECT;
101 osd_object_sa_fini(obj);
108 * Add object to list of dirty objects in tx handle.
110 void osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
112 if (!list_empty(&obj->oo_sa_linkage))
115 write_lock(&obj->oo_attr_lock);
116 if (likely(list_empty(&obj->oo_sa_linkage)))
117 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
118 write_unlock(&obj->oo_attr_lock);
122 * Release spill block dbuf hold for all dirty SAs.
124 void osd_object_sa_dirty_rele(const struct lu_env *env, struct osd_thandle *oh)
126 struct osd_object *obj;
128 while (!list_empty(&oh->ot_sa_list)) {
129 obj = list_entry(oh->ot_sa_list.next,
130 struct osd_object, oo_sa_linkage);
131 write_lock(&obj->oo_attr_lock);
132 list_del_init(&obj->oo_sa_linkage);
133 write_unlock(&obj->oo_attr_lock);
134 if (obj->oo_late_xattr) {
136 * take oo_guard to protect oo_sa_xattr buffer
137 * from concurrent update by osd_xattr_set()
139 LASSERT(oh->ot_assigned != 0);
140 down_write(&obj->oo_guard);
141 if (obj->oo_late_attr_set)
142 __osd_sa_attr_init(env, obj, oh);
143 else if (obj->oo_late_xattr)
144 __osd_sa_xattr_update(env, obj, oh);
145 up_write(&obj->oo_guard);
147 sa_spill_rele(obj->oo_sa_hdl);
152 * Update the SA and add the object to the dirty list.
154 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
155 void *buf, uint32_t buflen, struct osd_thandle *oh)
159 LASSERT(obj->oo_sa_hdl != NULL);
160 LASSERT(oh->ot_tx != NULL);
162 rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
163 osd_object_sa_dirty_add(obj, oh);
169 * Bulk update the SA and add the object to the dirty list.
172 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
173 int count, struct osd_thandle *oh)
177 LASSERT(obj->oo_sa_hdl != NULL);
178 LASSERT(oh->ot_tx != NULL);
180 rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
181 osd_object_sa_dirty_add(obj, oh);
187 * Retrieve the attributes of a DMU object
189 int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
190 struct osd_object *obj, struct lu_attr *la)
192 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
193 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
198 LASSERT(obj->oo_dn != NULL);
200 la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
201 LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
203 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
204 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
205 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
206 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
207 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
208 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
209 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
210 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
211 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
212 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
214 rc = -sa_bulk_lookup(obj->oo_sa_hdl, bulk, cnt);
218 #ifdef ZFS_PROJINHERIT
219 if (o->od_projectused_dn && osa->flags & ZFS_PROJID) {
220 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_PROJID(o),
225 la->la_projid = osa->projid;
226 la->la_valid |= LA_PROJID;
227 obj->oo_with_projid = 1;
229 la->la_projid = ZFS_DEFAULT_PROJID;
230 la->la_valid &= ~LA_PROJID;
234 la->la_valid &= ~LA_PROJID;
237 la->la_atime = osa->atime[0];
238 la->la_mtime = osa->mtime[0];
239 la->la_ctime = osa->ctime[0];
240 la->la_mode = osa->mode;
241 la->la_uid = osa->uid;
242 la->la_gid = osa->gid;
243 la->la_nlink = osa->nlink;
244 la->la_flags = attrs_zfs2fs(osa->flags);
245 la->la_size = osa->size;
247 /* Try to get extra flag from LMA. Right now, only LMAI_ORPHAN
248 * flags is stored in LMA, and it is only for orphan directory */
249 if (S_ISDIR(la->la_mode) && dt_object_exists(&obj->oo_dt)) {
250 struct osd_thread_info *info = osd_oti_get(env);
251 struct lustre_mdt_attrs *lma;
254 lma = (struct lustre_mdt_attrs *)info->oti_buf;
256 buf.lb_len = sizeof(info->oti_buf);
257 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
260 lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
262 lma_to_lustre_flags(lma->lma_incompat);
264 } else if (rc == -ENODATA) {
269 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
270 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
273 la->la_rdev = osa->rdev;
274 la->la_valid |= LA_RDEV;
281 int __osd_obj2dnode(objset_t *os, uint64_t oid, dnode_t **dnp)
287 rc = -dmu_bonus_hold(os, oid, osd_obj_tag, &db);
291 dbi = (dmu_buf_impl_t *)db;
293 *dnp = DB_DNODE(dbi);
294 LASSERT(*dnp != NULL);
300 * Concurrency: no concurrent access is possible that early in object
303 struct lu_object *osd_object_alloc(const struct lu_env *env,
304 const struct lu_object_header *hdr,
307 struct osd_object *mo;
309 OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
313 l = &mo->oo_dt.do_lu;
314 dt_object_init(&mo->oo_dt, NULL, d);
315 mo->oo_dt.do_ops = &osd_obj_ops;
316 l->lo_ops = &osd_lu_obj_ops;
317 INIT_LIST_HEAD(&mo->oo_sa_linkage);
318 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
319 init_rwsem(&mo->oo_sem);
320 init_rwsem(&mo->oo_guard);
321 rwlock_init(&mo->oo_attr_lock);
322 mo->oo_destroy = OSD_DESTROY_NONE;
329 static void osd_obj_set_blksize(const struct lu_env *env,
330 struct osd_device *osd, struct osd_object *obj)
332 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
334 dnode_t *dn = obj->oo_dn;
339 LASSERT(!osd_oti_get(env)->oti_in_trans);
341 tx = dmu_tx_create(osd->od_os);
343 CERROR("%s: fail to create tx to set blksize for "DFID"\n",
344 osd->od_svname, PFID(fid));
348 dmu_tx_hold_bonus(tx, dn->dn_object);
349 rc = -dmu_tx_assign(tx, TXG_WAIT);
352 CERROR("%s: fail to assign tx to set blksize for "DFID
353 ": rc = %d\n", osd->od_svname, PFID(fid), rc);
357 down_write(&obj->oo_guard);
358 if (unlikely((1 << dn->dn_datablkshift) >= PAGE_SIZE))
361 blksz = dn->dn_datablksz;
362 if (!is_power_of_2(blksz))
363 blksz = size_roundup_power2(blksz);
365 if (blksz > osd->od_max_blksz)
366 blksz = osd->od_max_blksz;
367 else if (blksz < PAGE_SIZE)
369 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object, blksz, 0, tx);
374 up_write(&obj->oo_guard);
377 if (unlikely(obj->oo_dn->dn_maxblkid > 0))
380 CERROR("%s: fail to set blksize for "DFID": rc = %d\n",
381 osd->od_svname, PFID(fid), rc);
384 CDEBUG(D_INODE, "%s: set blksize as %u for "DFID"\n",
385 osd->od_svname, blksz, PFID(fid));
390 * Concurrency: shouldn't matter.
392 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
394 struct osd_device *osd = osd_obj2dev(obj);
395 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
401 rc = osd_object_sa_init(obj, osd);
405 /* cache attrs in object */
406 rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
410 if (likely(!fid_is_acct(fid))) {
411 /* no body operations for accounting objects */
412 obj->oo_dt.do_body_ops = &osd_body_ops;
414 if (S_ISREG(obj->oo_attr.la_mode) &&
415 obj->oo_dn->dn_maxblkid == 0 &&
416 (1 << obj->oo_dn->dn_datablkshift) < PAGE_SIZE &&
417 (fid_is_idif(fid) || fid_is_norm(fid) ||
419 osd->od_is_ost && !osd->od_dt_dev.dd_rdonly)
420 osd_obj_set_blksize(env, osd, obj);
424 * initialize object before marking it existing
426 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
429 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
434 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
436 struct osd_thread_info *info = osd_oti_get(env);
439 struct lustre_mdt_attrs *lma;
442 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
443 lma = (struct lustre_mdt_attrs *)info->oti_buf;
445 buf.lb_len = sizeof(info->oti_buf);
447 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
450 lustre_lma_swab(lma);
451 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
452 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
453 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
454 "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
455 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
456 PFID(lu_object_fid(&obj->oo_dt.do_lu)));
459 struct osd_device *osd = osd_obj2dev(obj);
461 if (lma->lma_compat & LMAC_STRIPE_INFO &&
463 obj->oo_pfid_in_lma = 1;
464 if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
465 osd->od_remote_parent_dir != ZFS_NO_OBJECT)
466 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
468 } else if (rc == -ENODATA) {
469 /* haven't initialize LMA xattr */
477 * Helper function to retrieve DMU object id from fid for accounting object
479 static dnode_t *osd_quota_fid2dmu(const struct osd_device *osd,
480 const struct lu_fid *fid)
484 LASSERT(fid_is_acct(fid));
486 switch (fid_oid(fid)) {
488 dn = osd->od_userused_dn;
491 dn = osd->od_groupused_dn;
493 #ifdef ZFS_PROJINHERIT
494 case ACCT_PROJECT_OID:
495 dn = osd->od_projectused_dn;
506 * Concurrency: no concurrent access is possible that early in object
509 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
510 const struct lu_object_conf *conf)
512 struct osd_object *obj = osd_obj(l);
513 struct osd_device *osd = osd_obj2dev(obj);
514 const struct lu_fid *fid = lu_object_fid(l);
519 LASSERT(osd_invariant(obj));
521 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
522 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
523 l->lo_header->loh_attr |= LOHA_EXISTS;
527 if (conf != NULL && conf->loc_flags & LOC_F_NEW)
530 if (unlikely(fid_is_acct(fid))) {
531 obj->oo_dn = osd_quota_fid2dmu(osd, fid);
533 obj->oo_dt.do_index_ops = &osd_acct_index_ops;
534 l->lo_header->loh_attr |= LOHA_EXISTS;
540 rc = osd_fid_lookup(env, osd, fid, &oid);
542 LASSERT(obj->oo_dn == NULL);
543 rc = __osd_obj2dnode(osd->od_os, oid, &obj->oo_dn);
544 /* EEXIST will be returned if object is being deleted in ZFS */
550 CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
551 osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
554 rc = osd_object_init0(env, obj);
558 rc = osd_check_lma(env, obj);
561 } else if (rc == -ENOENT) {
564 LASSERT(osd_invariant(obj));
570 * Concurrency: no concurrent access is possible that late in object
573 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
575 struct osd_object *obj = osd_obj(l);
577 LASSERT(osd_invariant(obj));
579 dt_object_fini(&obj->oo_dt);
580 OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
584 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
588 LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
590 /* the object is supposed to be exclusively locked by
591 * the caller (osd_destroy()), while the transaction
592 * (oh) is per-thread and not shared */
593 if (likely(list_empty(&obj->oo_unlinked_linkage))) {
594 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
601 /* Default to max data size covered by a level-1 indirect block */
602 static unsigned long osd_sync_destroy_max_size =
603 1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
604 module_param(osd_sync_destroy_max_size, ulong, 0444);
605 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
608 osd_object_set_destroy_type(struct osd_object *obj)
611 * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
612 * only once and use it consistently thereafter.
614 down_write(&obj->oo_guard);
615 if (obj->oo_destroy == OSD_DESTROY_NONE) {
616 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
617 obj->oo_destroy = OSD_DESTROY_SYNC;
618 else /* Larger objects are destroyed asynchronously */
619 obj->oo_destroy = OSD_DESTROY_ASYNC;
621 up_write(&obj->oo_guard);
624 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
627 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
628 struct osd_object *obj = osd_dt_obj(dt);
629 struct osd_device *osd = osd_obj2dev(obj);
630 struct osd_thandle *oh;
637 LASSERT(dt_object_exists(dt));
639 oh = container_of0(th, struct osd_thandle, ot_super);
640 LASSERT(oh->ot_tx != NULL);
642 /* declare that we'll remove object from fid-dnode mapping */
643 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
644 osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
646 osd_declare_xattrs_destroy(env, obj, oh);
649 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
650 obj->oo_attr.la_gid, obj->oo_attr.la_projid,
651 -1, oh, NULL, OSD_QID_INODE);
655 /* data to be truncated */
656 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
657 obj->oo_attr.la_gid, obj->oo_attr.la_projid,
658 0, oh, NULL, OSD_QID_BLK);
662 osd_object_set_destroy_type(obj);
663 if (obj->oo_destroy == OSD_DESTROY_SYNC)
664 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
667 osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
668 osd->od_unlinked, TRUE, NULL);
670 /* remove agent entry (if have) from remote parent */
671 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
672 osd_tx_hold_zap(oh->ot_tx, osd->od_remote_parent_dir,
675 /* will help to find FID->ino when this object is being
676 * added to PENDING/ */
677 osd_idc_find_and_init(env, osd, obj);
682 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
685 struct osd_thread_info *info = osd_oti_get(env);
686 char *buf = info->oti_str;
687 struct osd_object *obj = osd_dt_obj(dt);
688 struct osd_device *osd = osd_obj2dev(obj);
689 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
690 struct osd_thandle *oh;
696 down_write(&obj->oo_guard);
698 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
699 GOTO(out, rc = -ENOENT);
701 LASSERT(obj->oo_dn != NULL);
703 oh = container_of0(th, struct osd_thandle, ot_super);
705 LASSERT(oh->ot_tx != NULL);
707 /* remove obj ref from index dir (it depends) */
708 zapid = osd_get_name_n_idx(env, osd, fid, buf,
709 sizeof(info->oti_str), &zdn);
710 rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
712 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
713 osd->od_svname, buf, rc);
717 rc = osd_xattrs_destroy(env, obj, oh);
719 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
720 osd->od_svname, buf, rc);
724 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
725 rc = osd_delete_from_remote_parent(env, osd, obj, oh, true);
730 oid = obj->oo_dn->dn_object;
731 if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
732 /* this may happen if the destroy wasn't declared
733 * e.g. when the object is created and then destroyed
734 * in the same transaction - we don't need additional
735 * space for destroy specifically */
736 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
737 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
739 CERROR("%s: failed to free %s %llu: rc = %d\n",
740 osd->od_svname, buf, oid, rc);
741 } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
742 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
744 CERROR("%s: failed to free %s %llu: rc = %d\n",
745 osd->od_svname, buf, oid, rc);
746 } else { /* asynchronous destroy */
747 char *key = info->oti_key;
749 rc = osd_object_unlinked_add(obj, oh);
753 snprintf(key, sizeof(info->oti_key), "%llx", oid);
754 rc = osd_zap_add(osd, osd->od_unlinked->dn_object,
755 osd->od_unlinked, key, 8, 1, &oid, oh->ot_tx);
757 CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
758 osd->od_svname, buf, oid, rc);
762 /* not needed in the cache anymore */
763 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
765 obj->oo_destroyed = 1;
766 up_write(&obj->oo_guard);
770 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
772 struct osd_object *obj = osd_obj(l);
773 const struct lu_fid *fid = lu_object_fid(l);
776 if (likely(!fid_is_acct(fid))) {
777 osd_object_sa_fini(obj);
778 if (obj->oo_sa_xattr) {
779 nvlist_free(obj->oo_sa_xattr);
780 obj->oo_sa_xattr = NULL;
782 osd_dnode_rele(obj->oo_dn);
783 list_del(&obj->oo_sa_linkage);
790 * Concurrency: ->loo_object_release() is called under site spin-lock.
792 static void osd_object_release(const struct lu_env *env,
798 * Concurrency: shouldn't matter.
800 static int osd_object_print(const struct lu_env *env, void *cookie,
801 lu_printer_t p, const struct lu_object *l)
803 struct osd_object *o = osd_obj(l);
805 return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
808 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
811 struct osd_object *obj = osd_dt_obj(dt);
813 LASSERT(osd_invariant(obj));
815 down_read_nested(&obj->oo_sem, role);
818 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
821 struct osd_object *obj = osd_dt_obj(dt);
823 LASSERT(osd_invariant(obj));
825 down_write_nested(&obj->oo_sem, role);
828 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
830 struct osd_object *obj = osd_dt_obj(dt);
832 LASSERT(osd_invariant(obj));
833 up_read(&obj->oo_sem);
836 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
838 struct osd_object *obj = osd_dt_obj(dt);
840 LASSERT(osd_invariant(obj));
841 up_write(&obj->oo_sem);
844 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
846 struct osd_object *obj = osd_dt_obj(dt);
849 LASSERT(osd_invariant(obj));
851 if (down_write_trylock(&obj->oo_sem)) {
853 up_write(&obj->oo_sem);
858 static int osd_attr_get(const struct lu_env *env,
859 struct dt_object *dt,
860 struct lu_attr *attr)
862 struct osd_object *obj = osd_dt_obj(dt);
867 down_read(&obj->oo_guard);
869 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
870 GOTO(out, rc = -ENOENT);
872 if (unlikely(fid_is_acct(lu_object_fid(&dt->do_lu))))
875 LASSERT(osd_invariant(obj));
878 read_lock(&obj->oo_attr_lock);
879 *attr = obj->oo_attr;
880 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
881 attr->la_flags |= LUSTRE_ORPHAN_FL;
882 read_unlock(&obj->oo_attr_lock);
884 /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
885 * from within sa_object_size() can block on a mutex, so
886 * we can't call sa_object_size() holding rwlock */
887 sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
888 /* we do not control size of indices, so always calculate
889 * it from number of blocks reported by DMU */
890 if (S_ISDIR(attr->la_mode))
891 attr->la_size = 512 * blocks;
892 /* Block size may be not set; suggest maximal I/O transfers. */
894 blksize = osd_spa_maxblocksize(
895 dmu_objset_spa(osd_obj2dev(obj)->od_os));
897 attr->la_blksize = blksize;
898 attr->la_blocks = blocks;
899 attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
902 up_read(&obj->oo_guard);
906 /* Simple wrapper on top of qsd API which implement quota transfer for osd
907 * setattr needs. As a reminder, only the root user can change ownership of
908 * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
909 static inline int qsd_transfer(const struct lu_env *env,
910 struct qsd_instance *qsd,
911 struct lquota_trans *trans, int qtype,
912 __u64 orig_id, __u64 new_id, __u64 bspace,
913 struct lquota_id_info *qi)
917 if (unlikely(qsd == NULL))
920 LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
921 qi->lqi_type = qtype;
923 /* inode accounting */
924 qi->lqi_is_blk = false;
926 /* one more inode for the new owner ... */
927 qi->lqi_id.qid_uid = new_id;
929 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
930 if (rc == -EDQUOT || rc == -EINPROGRESS)
935 /* and one less inode for the current id */
936 qi->lqi_id.qid_uid = orig_id;;
938 /* can't get EDQUOT when reducing usage */
939 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
940 if (rc == -EINPROGRESS)
945 /* block accounting */
946 qi->lqi_is_blk = true;
948 /* more blocks for the new owner ... */
949 qi->lqi_id.qid_uid = new_id;
950 qi->lqi_space = bspace;
951 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
952 if (rc == -EDQUOT || rc == -EINPROGRESS)
957 /* and finally less blocks for the current owner */
958 qi->lqi_id.qid_uid = orig_id;
959 qi->lqi_space = -bspace;
960 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
961 /* can't get EDQUOT when reducing usage */
962 if (rc == -EINPROGRESS)
967 static int osd_declare_attr_set(const struct lu_env *env,
968 struct dt_object *dt,
969 const struct lu_attr *attr,
970 struct thandle *handle)
972 struct osd_thread_info *info = osd_oti_get(env);
973 struct osd_object *obj = osd_dt_obj(dt);
974 struct osd_device *osd = osd_obj2dev(obj);
976 struct osd_thandle *oh;
984 LASSERT(handle != NULL);
985 LASSERT(osd_invariant(obj));
987 oh = container_of0(handle, struct osd_thandle, ot_super);
989 down_read(&obj->oo_guard);
990 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
993 LASSERT(obj->oo_sa_hdl != NULL);
994 LASSERT(oh->ot_tx != NULL);
995 /* regular attributes are part of the bonus buffer */
996 /* let's check whether this object is already part of
999 for (txh = list_head(&oh->ot_tx->tx_holds); txh;
1000 txh = list_next(&oh->ot_tx->tx_holds, txh)) {
1001 if (txh->txh_dnode == NULL)
1003 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
1005 /* this object is part of the transaction already
1006 * we don't need to declare bonus again */
1011 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
1012 if (oh->ot_tx->tx_err != 0)
1013 GOTO(out, rc = -oh->ot_tx->tx_err);
1015 if (attr && attr->la_valid & LA_FLAGS) {
1016 /* LMA is usually a part of bonus, no need to declare
1020 if (attr && (attr->la_valid & (LA_UID | LA_GID | LA_PROJID))) {
1021 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
1022 bspace = toqb(bspace * blksize);
1025 if (attr && attr->la_valid & LA_UID) {
1026 /* quota enforcement for user */
1027 if (attr->la_uid != obj->oo_attr.la_uid) {
1028 rc = qsd_transfer(env, osd->od_quota_slave,
1029 &oh->ot_quota_trans, USRQUOTA,
1030 obj->oo_attr.la_uid, attr->la_uid,
1031 bspace, &info->oti_qi);
1036 if (attr && attr->la_valid & LA_GID) {
1037 /* quota enforcement for group */
1038 if (attr->la_gid != obj->oo_attr.la_gid) {
1039 rc = qsd_transfer(env, osd->od_quota_slave,
1040 &oh->ot_quota_trans, GRPQUOTA,
1041 obj->oo_attr.la_gid, attr->la_gid,
1042 bspace, &info->oti_qi);
1047 #ifdef ZFS_PROJINHERIT
1048 if (attr && attr->la_valid & LA_PROJID) {
1049 if (!osd->od_projectused_dn)
1050 GOTO(out, rc = -EOPNOTSUPP);
1052 /* Usually, if project quota is upgradable for the device,
1053 * then the upgrade will be done before or when mount the
1054 * device. So when we come here, this project should have
1055 * project ID attribute already (that is zero by default).
1056 * Otherwise, there was something wrong during the former
1057 * upgrade, let's return failure to report that.
1059 * Please note that, different from other attributes, you
1060 * can NOT simply set the project ID attribute under such
1061 * case, because adding (NOT change) project ID attribute
1062 * needs to change the object's attribute layout to match
1063 * zfs backend quota accounting requirement. */
1064 if (unlikely(!obj->oo_with_projid))
1065 GOTO(out, rc = -ENXIO);
1067 /* quota enforcement for project */
1068 if (attr->la_projid != obj->oo_attr.la_projid) {
1069 rc = qsd_transfer(env, osd->od_quota_slave,
1070 &oh->ot_quota_trans, PRJQUOTA,
1071 obj->oo_attr.la_projid,
1072 attr->la_projid, bspace,
1080 up_read(&obj->oo_guard);
1085 * Set the attributes of an object
1087 * The transaction passed to this routine must have
1088 * dmu_tx_hold_bonus(tx, oid) called and then assigned
1089 * to a transaction group.
1091 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
1092 const struct lu_attr *la, struct thandle *handle)
1094 struct osd_thread_info *info = osd_oti_get(env);
1095 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1096 struct osd_object *obj = osd_dt_obj(dt);
1097 struct osd_device *osd = osd_obj2dev(obj);
1098 struct osd_thandle *oh;
1099 struct osa_attr *osa = &info->oti_osa;
1100 __u64 valid = la->la_valid;
1106 down_read(&obj->oo_guard);
1107 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1108 GOTO(out, rc = -ENOENT);
1110 LASSERT(handle != NULL);
1111 LASSERT(osd_invariant(obj));
1112 LASSERT(obj->oo_sa_hdl);
1114 oh = container_of0(handle, struct osd_thandle, ot_super);
1115 /* Assert that the transaction has been assigned to a
1116 transaction group. */
1117 LASSERT(oh->ot_tx->tx_txg != 0);
1119 /* Only allow set size for regular file */
1120 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
1121 valid &= ~(LA_SIZE | LA_BLOCKS);
1123 if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
1126 if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
1129 if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
1135 if (valid & LA_FLAGS) {
1136 struct lustre_mdt_attrs *lma;
1139 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
1140 LASSERT(!obj->oo_pfid_in_lma);
1141 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
1142 lma = (struct lustre_mdt_attrs *)&info->oti_buf;
1144 buf.lb_len = sizeof(info->oti_buf);
1145 rc = osd_xattr_get(env, &obj->oo_dt, &buf,
1149 le32_to_cpu(lma->lma_incompat);
1150 lma->lma_incompat |=
1151 lustre_to_lma_flags(la->la_flags);
1153 cpu_to_le32(lma->lma_incompat);
1155 buf.lb_len = sizeof(*lma);
1156 rc = osd_xattr_set_internal(env, obj, &buf,
1162 CWARN("%s: failed to set LMA flags: rc = %d\n",
1163 osd->od_svname, rc);
1169 write_lock(&obj->oo_attr_lock);
1172 if (valid & LA_PROJID) {
1173 #ifdef ZFS_PROJINHERIT
1174 /* osd_declare_attr_set() must be called firstly.
1175 * If osd::od_projectused_dn is not set, then we
1176 * can not arrive at here. */
1177 LASSERT(osd->od_projectused_dn);
1178 LASSERT(obj->oo_with_projid);
1180 osa->projid = obj->oo_attr.la_projid = la->la_projid;
1181 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1184 valid &= ~LA_PROJID;
1188 if (valid & LA_ATIME) {
1189 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1190 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1193 if (valid & LA_MTIME) {
1194 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1195 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1198 if (valid & LA_CTIME) {
1199 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1200 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1203 if (valid & LA_MODE) {
1204 /* mode is stored along with type, so read it first */
1205 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1206 (la->la_mode & ~S_IFMT);
1207 osa->mode = obj->oo_attr.la_mode;
1208 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1211 if (valid & LA_SIZE) {
1212 osa->size = obj->oo_attr.la_size = la->la_size;
1213 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1216 if (valid & LA_NLINK) {
1217 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1218 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1221 if (valid & LA_RDEV) {
1222 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1223 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1226 if (valid & LA_FLAGS) {
1227 osa->flags = attrs_fs2zfs(la->la_flags);
1228 /* many flags are not supported by zfs, so ensure a good cached
1230 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1231 #ifdef ZFS_PROJINHERIT
1232 if (obj->oo_with_projid)
1233 osa->flags |= ZFS_PROJID;
1235 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1238 if (valid & LA_UID) {
1239 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1240 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1243 if (valid & LA_GID) {
1244 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1245 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1248 obj->oo_attr.la_valid |= valid;
1249 write_unlock(&obj->oo_attr_lock);
1251 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1252 rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1255 up_read(&obj->oo_guard);
1262 * XXX temporary solution.
1265 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1266 struct dt_object *parent, struct dt_object *child,
1271 ah->dah_parent = parent;
1272 ah->dah_mode = child_mode;
1274 if (parent != NULL && !dt_object_remote(parent)) {
1275 /* will help to find FID->ino at dt_insert("..") */
1276 struct osd_object *pobj = osd_dt_obj(parent);
1278 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1282 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
1283 struct lu_attr *attr,
1284 struct dt_allocation_hint *hint,
1285 struct dt_object_format *dof,
1286 struct thandle *handle)
1288 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1289 struct osd_object *obj = osd_dt_obj(dt);
1290 struct osd_device *osd = osd_obj2dev(obj);
1291 struct osd_thandle *oh;
1299 switch (dof->dof_type) {
1303 if (obj->oo_dt.do_body_ops == NULL)
1304 obj->oo_dt.do_body_ops = &osd_body_ops;
1310 LASSERT(handle != NULL);
1311 oh = container_of0(handle, struct osd_thandle, ot_super);
1312 LASSERT(oh->ot_tx != NULL);
1314 /* this is the minimum set of EAs on every Lustre object */
1315 obj->oo_ea_in_bonus = OSD_BASE_EA_IN_BONUS;
1316 /* reserve 32 bytes for extra stuff like ACLs */
1317 dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1319 switch (dof->dof_type) {
1321 dt->do_index_ops = &osd_dir_ops;
1323 /* for zap create */
1324 dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1325 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1330 /* first, we'll create new object */
1331 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1339 /* and we'll add it to some mapping */
1340 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
1341 osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
1343 /* will help to find FID->ino mapping at dt_insert() */
1344 osd_idc_find_and_init(env, osd, obj);
1346 rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid,
1347 attr->la_projid, 1, oh, NULL, OSD_QID_INODE);
1352 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1353 struct osd_object *obj, sa_handle_t *sa_hdl, dmu_tx_t *tx,
1354 struct lu_attr *la, uint64_t parent,
1357 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1358 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1364 char *dxattr = NULL;
1370 gen = dmu_tx_get_txg(tx);
1372 ZFS_TIME_ENCODE(&now, crtime);
1374 osa->atime[0] = la->la_atime;
1375 osa->ctime[0] = la->la_ctime;
1376 osa->mtime[0] = la->la_mtime;
1377 osa->mode = la->la_mode;
1378 osa->uid = la->la_uid;
1379 osa->gid = la->la_gid;
1380 osa->rdev = la->la_rdev;
1381 osa->nlink = la->la_nlink;
1382 if (la->la_valid & LA_FLAGS)
1383 osa->flags = attrs_fs2zfs(la->la_flags);
1386 osa->size = la->la_size;
1387 #ifdef ZFS_PROJINHERIT
1388 if (osd->od_projectused_dn) {
1389 if (la->la_valid & LA_PROJID)
1390 osa->projid = la->la_projid;
1392 osa->projid = ZFS_DEFAULT_PROJID;
1393 osa->flags |= ZFS_PROJID;
1395 obj->oo_with_projid = 1;
1397 osa->flags &= ~ZFS_PROJID;
1402 * we need to create all SA below upon object create.
1404 * XXX The attribute order matters since the accounting callback relies
1405 * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1406 * look up the UID/GID/PROJID attributes. Moreover, the callback does
1407 * not seem to support the spill block.
1408 * We define attributes in the same order as SA_*_OFFSET in order to
1409 * work around the problem. See ORI-610.
1412 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1413 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1414 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1415 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1416 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1417 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1418 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1419 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1420 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1421 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1422 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1423 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1424 #ifdef ZFS_PROJINHERIT
1425 if (osd->od_projectused_dn)
1426 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1429 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1430 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1433 rc = -nvlist_size(xattr, &sa_size, NV_ENCODE_XDR);
1436 dxattr = osd_zio_buf_alloc(sa_size);
1439 rc = -nvlist_pack(xattr, &dxattr, &sa_size,
1440 NV_ENCODE_XDR, KM_SLEEP);
1443 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd),
1444 NULL, dxattr, sa_size);
1447 rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1449 osd_zio_buf_free(dxattr, sa_size);
1454 static int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1455 uint64_t oid, dnode_t **dnp)
1460 /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1461 for (txh = list_tail(&tx->tx_holds); txh;
1462 txh = list_prev(&tx->tx_holds, txh)) {
1463 dnode_t *dn = txh->txh_dnode;
1468 if (dn->dn_object != oid)
1472 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1473 if (dn->dn_bonus == NULL)
1474 dbuf_create_bonus(dn);
1475 rw_exit(&dn->dn_struct_rwlock);
1479 LASSERT(dn->dn_handle);
1481 if (refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1482 refcount_add(&dn->dn_holds, tag);
1483 atomic_inc_32(&dn->dn_dbufs_count);
1486 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1490 if (unlikely(*dnp == NULL))
1491 rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
1496 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
1497 int osd_find_dnsize(struct osd_device *osd, int ea_in_bonus)
1501 if (osd->od_dnsize == ZFS_DNSIZE_AUTO) {
1502 dnsize = DNODE_MIN_SIZE;
1504 if (DN_BONUS_SIZE(dnsize) >= ea_in_bonus + 32)
1507 } while (dnsize < DNODE_MAX_SIZE);
1508 if (dnsize > DNODE_MAX_SIZE)
1509 dnsize = DNODE_MAX_SIZE;
1510 } else if (osd->od_dnsize == ZFS_DNSIZE_1K) {
1512 } else if (osd->od_dnsize == ZFS_DNSIZE_2K) {
1514 } else if (osd->od_dnsize == ZFS_DNSIZE_4K) {
1516 } else if (osd->od_dnsize == ZFS_DNSIZE_8K) {
1518 } else if (osd->od_dnsize == ZFS_DNSIZE_16K) {
1521 dnsize = DNODE_MIN_SIZE;
1528 * The transaction passed to this routine must have
1529 * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1530 * to a transaction group.
1532 int __osd_object_create(const struct lu_env *env, struct osd_device *osd,
1533 struct osd_object *obj, const struct lu_fid *fid,
1534 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1536 dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1540 /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1541 * would get an additional ditto copy */
1542 if (unlikely(S_ISREG(la->la_mode) &&
1543 fid_seq_is_local_file(fid_seq(fid))))
1544 type = DMU_OTN_UINT8_METADATA;
1546 /* Create a new DMU object using the default dnode size. */
1548 size = obj->oo_ea_in_bonus;
1550 size = OSD_BASE_EA_IN_BONUS;
1551 oid = osd_dmu_object_alloc(osd->od_os, type, 0,
1552 osd_find_dnsize(osd, size), tx);
1554 LASSERT(la->la_valid & LA_MODE);
1558 return osd_find_new_dnode(env, tx, oid, dnp);
1562 * The transaction passed to this routine must have
1563 * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1564 * to a transaction group.
1566 * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1567 * This is fine for directories today, because storing the FID in the dirent
1568 * will also require a FAT ZAP. If there is a new type of micro ZAP created
1569 * then we might need to re-evaluate the use of this flag and instead do
1570 * a conversion from the different internal ZAP hash formats being used. */
1571 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1572 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1573 unsigned dnsize, zap_flags_t flags)
1577 /* Assert that the transaction has been assigned to a
1578 transaction group. */
1579 LASSERT(tx->tx_txg != 0);
1582 oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1583 DMU_OT_DIRECTORY_CONTENTS,
1584 14, /* == ZFS fzap_default_blockshift */
1585 DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1591 return osd_find_new_dnode(env, tx, oid, dnp);
1594 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1595 struct lu_attr *la, struct osd_thandle *oh)
1597 struct osd_device *osd = osd_obj2dev(obj);
1601 /* Index file should be created as regular file in order not to confuse
1602 * ZPL which could interpret them as directory.
1603 * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1605 LASSERT(S_ISREG(la->la_mode));
1606 rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1607 osd_find_dnsize(osd, obj->oo_ea_in_bonus), ZAP_FLAG_UINT64_KEY);
1613 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1614 struct lu_attr *la, struct osd_thandle *oh)
1616 struct osd_device *osd = osd_obj2dev(obj);
1620 LASSERT(S_ISDIR(la->la_mode));
1621 rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1622 osd_find_dnsize(osd, obj->oo_ea_in_bonus), 0);
1628 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1629 struct lu_attr *la, struct osd_thandle *oh)
1631 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1632 struct osd_device *osd = osd_obj2dev(obj);
1636 LASSERT(S_ISREG(la->la_mode));
1637 rc = __osd_object_create(env, osd, obj, fid, &dn, oh->ot_tx, la);
1641 if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid))) {
1642 /* The minimum block size must be at least page size otherwise
1643 * it will break the assumption in tgt_thread_big_cache where
1644 * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1645 * RDMA due to subpage transfer size */
1646 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1647 PAGE_SIZE, 0, oh->ot_tx);
1649 CERROR("%s: can't change blocksize: %d\n",
1650 osd->od_svname, rc);
1658 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1659 struct lu_attr *la, struct osd_thandle *oh)
1664 LASSERT(S_ISLNK(la->la_mode));
1665 rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1666 lu_object_fid(&obj->oo_dt.do_lu),
1667 &dn, oh->ot_tx, la);
1673 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1674 struct lu_attr *la, struct osd_thandle *oh)
1679 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1680 la->la_valid |= LA_RDEV;
1682 rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1683 lu_object_fid(&obj->oo_dt.do_lu),
1684 &dn, oh->ot_tx, la);
1690 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1691 struct osd_object *obj,
1693 struct osd_thandle *oh);
1695 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1697 osd_obj_type_f result;
1723 * Concurrency: @dt is write locked.
1725 static int osd_create(const struct lu_env *env, struct dt_object *dt,
1726 struct lu_attr *attr, struct dt_allocation_hint *hint,
1727 struct dt_object_format *dof, struct thandle *th)
1729 struct osd_thread_info *info = osd_oti_get(env);
1730 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1731 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1732 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1733 struct osd_object *obj = osd_dt_obj(dt);
1734 struct osd_device *osd = osd_obj2dev(obj);
1735 char *buf = info->oti_str;
1736 struct osd_thandle *oh;
1737 dnode_t *dn = NULL, *zdn = NULL;
1738 uint64_t zapid, parent = 0;
1743 LASSERT(!fid_is_acct(fid));
1745 /* concurrent create declarations should not see
1746 * the object inconsistent (db, attr, etc).
1747 * in regular cases acquisition should be cheap */
1748 down_write(&obj->oo_guard);
1750 if (unlikely(dt_object_exists(dt)))
1751 GOTO(out, rc = -EEXIST);
1753 LASSERT(osd_invariant(obj));
1754 LASSERT(dof != NULL);
1756 LASSERT(th != NULL);
1757 oh = container_of0(th, struct osd_thandle, ot_super);
1759 LASSERT(obj->oo_dn == NULL);
1761 /* to follow ZFS on-disk format we need
1762 * to initialize parent dnode properly */
1763 if (hint != NULL && hint->dah_parent != NULL &&
1764 !dt_object_remote(hint->dah_parent))
1765 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1767 /* we may fix some attributes, better do not change the source */
1768 obj->oo_attr = *attr;
1769 obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1771 #ifdef ZFS_PROJINHERIT
1772 if (osd->od_projectused_dn) {
1773 if (!(obj->oo_attr.la_valid & LA_PROJID))
1774 obj->oo_attr.la_projid = ZFS_DEFAULT_PROJID;
1775 obj->oo_with_projid = 1;
1779 dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1787 zde->zde_dnode = dn->dn_object;
1788 zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1790 zapid = osd_get_name_n_idx(env, osd, fid, buf,
1791 sizeof(info->oti_str), &zdn);
1792 rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
1796 /* Now add in all of the "SA" attributes */
1797 rc = osd_sa_handle_get(obj);
1801 rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1805 /* initialize LMA */
1806 lustre_lma_init(lma, fid, 0, 0);
1807 lustre_lma_swab(lma);
1808 rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
1809 (uchar_t *)lma, sizeof(*lma));
1813 /* configure new osd object */
1814 obj->oo_parent = parent != 0 ? parent : zapid;
1815 obj->oo_late_attr_set = 1;
1816 rc = __osd_sa_xattr_schedule_update(env, obj, oh);
1820 /* XXX: oo_lma_flags */
1821 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
1822 if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
1823 /* no body operations for accounting objects */
1824 obj->oo_dt.do_body_ops = &osd_body_ops;
1826 osd_idc_find_and_init(env, osd, obj);
1829 if (unlikely(rc && dn)) {
1830 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
1834 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
1836 up_write(&obj->oo_guard);
1840 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
1843 return osd_declare_attr_set(env, dt, NULL, th);
1847 * Concurrency: @dt is write locked.
1849 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
1850 struct thandle *handle)
1852 struct osd_object *obj = osd_dt_obj(dt);
1853 struct osd_thandle *oh;
1854 struct osd_device *osd = osd_obj2dev(obj);
1860 down_read(&obj->oo_guard);
1861 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1862 GOTO(out, rc = -ENOENT);
1864 LASSERT(osd_invariant(obj));
1865 LASSERT(obj->oo_sa_hdl != NULL);
1867 oh = container_of0(handle, struct osd_thandle, ot_super);
1869 write_lock(&obj->oo_attr_lock);
1870 nlink = ++obj->oo_attr.la_nlink;
1871 write_unlock(&obj->oo_attr_lock);
1873 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1876 up_read(&obj->oo_guard);
1880 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
1881 struct thandle *handle)
1883 return osd_declare_attr_set(env, dt, NULL, handle);
1887 * Concurrency: @dt is write locked.
1889 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
1890 struct thandle *handle)
1892 struct osd_object *obj = osd_dt_obj(dt);
1893 struct osd_thandle *oh;
1894 struct osd_device *osd = osd_obj2dev(obj);
1900 down_read(&obj->oo_guard);
1902 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1903 GOTO(out, rc = -ENOENT);
1905 LASSERT(osd_invariant(obj));
1906 LASSERT(obj->oo_sa_hdl != NULL);
1908 oh = container_of0(handle, struct osd_thandle, ot_super);
1909 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1911 write_lock(&obj->oo_attr_lock);
1912 nlink = --obj->oo_attr.la_nlink;
1913 write_unlock(&obj->oo_attr_lock);
1915 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1918 up_read(&obj->oo_guard);
1922 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1923 __u64 start, __u64 end)
1925 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1928 /* XXX: no other option than syncing the whole filesystem until we
1929 * support ZIL. If the object tracked the txg that it was last
1930 * modified in, it could pass that txg here instead of "0". Maybe
1931 * the changes are already committed, so no wait is needed at all? */
1932 if (!osd->od_dt_dev.dd_rdonly) {
1933 if (osd_object_sync_delay_us < 0)
1934 txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
1936 udelay(osd_object_sync_delay_us);
1942 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
1947 static struct dt_object_operations osd_obj_ops = {
1948 .do_read_lock = osd_read_lock,
1949 .do_write_lock = osd_write_lock,
1950 .do_read_unlock = osd_read_unlock,
1951 .do_write_unlock = osd_write_unlock,
1952 .do_write_locked = osd_write_locked,
1953 .do_attr_get = osd_attr_get,
1954 .do_declare_attr_set = osd_declare_attr_set,
1955 .do_attr_set = osd_attr_set,
1956 .do_ah_init = osd_ah_init,
1957 .do_declare_create = osd_declare_create,
1958 .do_create = osd_create,
1959 .do_declare_destroy = osd_declare_destroy,
1960 .do_destroy = osd_destroy,
1961 .do_index_try = osd_index_try,
1962 .do_declare_ref_add = osd_declare_ref_add,
1963 .do_ref_add = osd_ref_add,
1964 .do_declare_ref_del = osd_declare_ref_del,
1965 .do_ref_del = osd_ref_del,
1966 .do_xattr_get = osd_xattr_get,
1967 .do_declare_xattr_set = osd_declare_xattr_set,
1968 .do_xattr_set = osd_xattr_set,
1969 .do_declare_xattr_del = osd_declare_xattr_del,
1970 .do_xattr_del = osd_xattr_del,
1971 .do_xattr_list = osd_xattr_list,
1972 .do_object_sync = osd_object_sync,
1973 .do_invalidate = osd_invalidate,
1976 static struct lu_object_operations osd_lu_obj_ops = {
1977 .loo_object_init = osd_object_init,
1978 .loo_object_delete = osd_object_delete,
1979 .loo_object_release = osd_object_release,
1980 .loo_object_free = osd_object_free,
1981 .loo_object_print = osd_object_print,
1982 .loo_object_invariant = osd_object_invariant,
1985 static int osd_otable_it_attr_get(const struct lu_env *env,
1986 struct dt_object *dt,
1987 struct lu_attr *attr)
1993 static struct dt_object_operations osd_obj_otable_it_ops = {
1994 .do_attr_get = osd_otable_it_attr_get,
1995 .do_index_try = osd_index_try,
1998 module_param(osd_object_sync_delay_us, int, 0644);
1999 MODULE_PARM_DESC(osd_object_sync_delay_us,
2000 "If zero or larger delay N usec instead of doing object sync");