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);
312 struct lu_object_header *h;
313 struct osd_device *o = osd_dev(d);
315 l = &mo->oo_dt.do_lu;
316 if (unlikely(o->od_in_init)) {
323 lu_object_header_init(h);
324 lu_object_init(l, h, d);
325 lu_object_add_top(h, l);
328 dt_object_init(&mo->oo_dt, NULL, d);
329 mo->oo_header = NULL;
332 mo->oo_dt.do_ops = &osd_obj_ops;
333 l->lo_ops = &osd_lu_obj_ops;
334 INIT_LIST_HEAD(&mo->oo_sa_linkage);
335 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
336 init_rwsem(&mo->oo_sem);
337 init_rwsem(&mo->oo_guard);
338 rwlock_init(&mo->oo_attr_lock);
339 mo->oo_destroy = OSD_DESTROY_NONE;
346 static void osd_obj_set_blksize(const struct lu_env *env,
347 struct osd_device *osd, struct osd_object *obj)
349 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
351 dnode_t *dn = obj->oo_dn;
356 LASSERT(!osd_oti_get(env)->oti_in_trans);
358 tx = dmu_tx_create(osd->od_os);
360 CERROR("%s: fail to create tx to set blksize for "DFID"\n",
361 osd->od_svname, PFID(fid));
365 dmu_tx_hold_bonus(tx, dn->dn_object);
366 rc = -dmu_tx_assign(tx, TXG_WAIT);
369 CERROR("%s: fail to assign tx to set blksize for "DFID
370 ": rc = %d\n", osd->od_svname, PFID(fid), rc);
374 down_write(&obj->oo_guard);
375 if (unlikely((1 << dn->dn_datablkshift) >= PAGE_SIZE))
378 blksz = dn->dn_datablksz;
379 if (!is_power_of_2(blksz))
380 blksz = size_roundup_power2(blksz);
382 if (blksz > osd->od_max_blksz)
383 blksz = osd->od_max_blksz;
384 else if (blksz < PAGE_SIZE)
386 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object, blksz, 0, tx);
391 up_write(&obj->oo_guard);
394 if (unlikely(obj->oo_dn->dn_maxblkid > 0))
397 CERROR("%s: fail to set blksize for "DFID": rc = %d\n",
398 osd->od_svname, PFID(fid), rc);
401 CDEBUG(D_INODE, "%s: set blksize as %u for "DFID"\n",
402 osd->od_svname, blksz, PFID(fid));
407 * Concurrency: shouldn't matter.
409 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
411 struct osd_device *osd = osd_obj2dev(obj);
412 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
418 rc = osd_object_sa_init(obj, osd);
422 /* cache attrs in object */
423 rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
427 if (likely(!fid_is_acct(fid))) {
428 /* no body operations for accounting objects */
429 obj->oo_dt.do_body_ops = &osd_body_ops;
431 if (S_ISREG(obj->oo_attr.la_mode) &&
432 obj->oo_dn->dn_maxblkid == 0 &&
433 (1 << obj->oo_dn->dn_datablkshift) < PAGE_SIZE &&
434 (fid_is_idif(fid) || fid_is_norm(fid) ||
436 osd->od_is_ost && !osd->od_dt_dev.dd_rdonly)
437 osd_obj_set_blksize(env, osd, obj);
441 * initialize object before marking it existing
443 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
446 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
451 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
453 struct osd_thread_info *info = osd_oti_get(env);
456 struct lustre_mdt_attrs *lma;
457 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
460 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
461 lma = (struct lustre_mdt_attrs *)info->oti_buf;
463 buf.lb_len = sizeof(info->oti_buf);
465 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
468 lustre_lma_swab(lma);
469 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
470 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
471 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
472 "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
473 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
476 } else if (unlikely(!lu_fid_eq(rfid, &lma->lma_self_fid))) {
477 CERROR("%s: FID-in-LMA "DFID" does not match the "
478 "object self-fid "DFID"\n",
479 osd_obj2dev(obj)->od_svname,
480 PFID(&lma->lma_self_fid), PFID(rfid));
483 struct osd_device *osd = osd_obj2dev(obj);
485 if (lma->lma_compat & LMAC_STRIPE_INFO &&
487 obj->oo_pfid_in_lma = 1;
488 if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
489 osd->od_remote_parent_dir != ZFS_NO_OBJECT)
490 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
492 } else if (rc == -ENODATA) {
493 /* haven't initialize LMA xattr */
501 * Helper function to retrieve DMU object id from fid for accounting object
503 static dnode_t *osd_quota_fid2dmu(const struct osd_device *osd,
504 const struct lu_fid *fid)
508 LASSERT(fid_is_acct(fid));
510 switch (fid_oid(fid)) {
512 dn = osd->od_userused_dn;
515 dn = osd->od_groupused_dn;
517 #ifdef ZFS_PROJINHERIT
518 case ACCT_PROJECT_OID:
519 dn = osd->od_projectused_dn;
530 * Concurrency: no concurrent access is possible that early in object
533 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
534 const struct lu_object_conf *conf)
536 struct osd_object *obj = osd_obj(l);
537 struct osd_device *osd = osd_obj2dev(obj);
538 const struct lu_fid *fid = lu_object_fid(l);
539 struct lustre_scrub *scrub = &osd->od_scrub;
540 struct osd_thread_info *info = osd_oti_get(env);
541 struct luz_direntry *zde = &info->oti_zde;
542 struct osd_idmap_cache *idc;
543 char *name = info->oti_str;
550 LASSERT(osd_invariant(obj));
552 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
553 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
554 l->lo_header->loh_attr |= LOHA_EXISTS;
559 if (conf && conf->loc_flags & LOC_F_NEW)
562 if (unlikely(fid_is_acct(fid))) {
563 obj->oo_dn = osd_quota_fid2dmu(osd, fid);
565 obj->oo_dt.do_index_ops = &osd_acct_index_ops;
566 l->lo_header->loh_attr |= LOHA_EXISTS;
572 idc = osd_idc_find(env, osd, fid);
573 if (idc && !idc->oic_remote && idc->oic_dnode != ZFS_NO_OBJECT) {
574 oid = idc->oic_dnode;
579 if (!list_empty(&osd->od_scrub.os_inconsistent_items))
580 rc = osd_oii_lookup(osd, fid, &oid);
583 rc = osd_fid_lookup(env, osd, fid, &oid);
586 if (likely(!(fid_is_norm(fid) || fid_is_igif(fid)) ||
587 fid_is_on_ost(env, osd, fid) ||
588 !zfs_test_bit(osd_oi_fid2idx(osd, fid),
589 scrub->os_file.sf_oi_bitmap)))
600 LASSERT(obj->oo_dn == NULL);
602 rc = __osd_obj2dnode(osd->od_os, oid, &obj->oo_dn);
603 /* EEXIST will be returned if object is being deleted in ZFS */
608 CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
609 osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
613 rc = osd_object_init0(env, obj);
617 if (unlikely(obj->oo_header))
620 rc = osd_check_lma(env, obj);
621 if ((!rc && !remote) || (rc != -EREMCHG))
625 /* We still have chance to get the valid dnode: for the object that is
626 * referenced by remote name entry, the object on the local MDT will be
627 * linked under the dir /REMOTE_PARENT_DIR with its FID string as name.
629 * During the OI scrub, if we cannot find the OI mapping, we may still
630 * have change to map the FID to local OID via lookup the dir
631 * /REMOTE_PARENT_DIR. */
632 if (!remote && !fid_is_on_ost(env, osd, fid)) {
633 osd_fid2str(name, fid, sizeof(info->oti_str));
634 rc = osd_zap_lookup(osd, osd->od_remote_parent_dir,
635 NULL, name, 8, 3, (void *)zde);
637 oid = zde->lzd_reg.zde_dnode;
638 osd_dnode_rele(obj->oo_dn);
645 /* The case someone triggered the OI scrub already. */
646 if (thread_is_running(&scrub->os_thread)) {
650 lu_object_set_agent_entry(l);
651 osd_oii_insert(env, osd, fid, oid, false);
659 /* The case NOT allow to trigger OI scrub automatically. */
660 if (osd->od_auto_scrub_interval == AS_NEVER)
663 /* It is me to trigger the OI scrub. */
664 rc1 = osd_scrub_start(env, osd, SS_CLEAR_DRYRUN |
665 SS_CLEAR_FAILOUT | SS_AUTO_FULL);
666 LCONSOLE_WARN("%s: trigger OI scrub by RPC for the "DFID": rc = %d\n",
667 osd_name(osd), PFID(fid), rc1);
671 lu_object_set_agent_entry(l);
673 osd_oii_insert(env, osd, fid, oid, false);
688 * Concurrency: no concurrent access is possible that late in object
691 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
693 struct osd_object *obj = osd_obj(l);
694 struct lu_object_header *h = obj->oo_header;
696 LASSERT(osd_invariant(obj));
698 dt_object_fini(&obj->oo_dt);
699 OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
701 lu_object_header_fini(h);
707 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
711 LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
713 /* the object is supposed to be exclusively locked by
714 * the caller (osd_destroy()), while the transaction
715 * (oh) is per-thread and not shared */
716 if (likely(list_empty(&obj->oo_unlinked_linkage))) {
717 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
724 /* Default to max data size covered by a level-1 indirect block */
725 static unsigned long osd_sync_destroy_max_size =
726 1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
727 module_param(osd_sync_destroy_max_size, ulong, 0444);
728 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
731 osd_object_set_destroy_type(struct osd_object *obj)
734 * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
735 * only once and use it consistently thereafter.
737 down_write(&obj->oo_guard);
738 if (obj->oo_destroy == OSD_DESTROY_NONE) {
739 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
740 obj->oo_destroy = OSD_DESTROY_SYNC;
741 else /* Larger objects are destroyed asynchronously */
742 obj->oo_destroy = OSD_DESTROY_ASYNC;
744 up_write(&obj->oo_guard);
747 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
750 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
751 struct osd_object *obj = osd_dt_obj(dt);
752 struct osd_device *osd = osd_obj2dev(obj);
753 struct osd_thandle *oh;
760 LASSERT(dt_object_exists(dt));
762 oh = container_of0(th, struct osd_thandle, ot_super);
763 LASSERT(oh->ot_tx != NULL);
765 /* declare that we'll remove object from fid-dnode mapping */
766 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
767 osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
769 osd_declare_xattrs_destroy(env, obj, oh);
772 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
773 obj->oo_attr.la_gid, obj->oo_attr.la_projid,
774 -1, oh, NULL, OSD_QID_INODE);
778 /* data to be truncated */
779 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
780 obj->oo_attr.la_gid, obj->oo_attr.la_projid,
781 0, oh, NULL, OSD_QID_BLK);
785 osd_object_set_destroy_type(obj);
786 if (obj->oo_destroy == OSD_DESTROY_SYNC)
787 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
790 osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
791 osd->od_unlinked, TRUE, NULL);
793 /* remove agent entry (if have) from remote parent */
794 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
795 osd_tx_hold_zap(oh->ot_tx, osd->od_remote_parent_dir,
798 /* will help to find FID->ino when this object is being
799 * added to PENDING/ */
800 osd_idc_find_and_init(env, osd, obj);
805 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
808 struct osd_thread_info *info = osd_oti_get(env);
809 char *buf = info->oti_str;
810 struct osd_object *obj = osd_dt_obj(dt);
811 struct osd_device *osd = osd_obj2dev(obj);
812 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
813 struct osd_thandle *oh;
819 down_write(&obj->oo_guard);
821 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
822 GOTO(out, rc = -ENOENT);
824 LASSERT(obj->oo_dn != NULL);
826 oh = container_of0(th, struct osd_thandle, ot_super);
828 LASSERT(oh->ot_tx != NULL);
830 /* remove obj ref from index dir (it depends) */
831 zapid = osd_get_name_n_idx(env, osd, fid, buf,
832 sizeof(info->oti_str), &zdn);
833 rc = osd_xattrs_destroy(env, obj, oh);
835 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
836 osd->od_svname, buf, rc);
840 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
841 rc = osd_delete_from_remote_parent(env, osd, obj, oh, true);
846 oid = obj->oo_dn->dn_object;
847 if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
848 /* this may happen if the destroy wasn't declared
849 * e.g. when the object is created and then destroyed
850 * in the same transaction - we don't need additional
851 * space for destroy specifically */
852 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
853 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
855 CERROR("%s: failed to free %s %llu: rc = %d\n",
856 osd->od_svname, buf, oid, rc);
857 } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
858 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
860 CERROR("%s: failed to free %s %llu: rc = %d\n",
861 osd->od_svname, buf, oid, rc);
862 } else { /* asynchronous destroy */
863 char *key = info->oti_key;
865 rc = osd_object_unlinked_add(obj, oh);
869 snprintf(key, sizeof(info->oti_key), "%llx", oid);
870 rc = osd_zap_add(osd, osd->od_unlinked->dn_object,
871 osd->od_unlinked, key, 8, 1, &oid, oh->ot_tx);
873 CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
874 osd->od_svname, buf, oid, rc);
877 /* Remove the OI mapping after the destroy to handle the race with
878 * OI scrub that may insert missed OI mapping during the interval. */
879 rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
880 if (unlikely(rc == -ENOENT))
883 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
884 osd->od_svname, buf, rc);
889 /* not needed in the cache anymore */
890 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
892 obj->oo_destroyed = 1;
893 up_write(&obj->oo_guard);
897 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
899 struct osd_object *obj = osd_obj(l);
900 const struct lu_fid *fid = lu_object_fid(l);
903 if (likely(!fid_is_acct(fid))) {
904 osd_object_sa_fini(obj);
905 if (obj->oo_sa_xattr) {
906 nvlist_free(obj->oo_sa_xattr);
907 obj->oo_sa_xattr = NULL;
909 osd_dnode_rele(obj->oo_dn);
910 list_del(&obj->oo_sa_linkage);
917 * Concurrency: ->loo_object_release() is called under site spin-lock.
919 static void osd_object_release(const struct lu_env *env,
925 * Concurrency: shouldn't matter.
927 static int osd_object_print(const struct lu_env *env, void *cookie,
928 lu_printer_t p, const struct lu_object *l)
930 struct osd_object *o = osd_obj(l);
932 return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
935 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
938 struct osd_object *obj = osd_dt_obj(dt);
940 LASSERT(osd_invariant(obj));
942 down_read_nested(&obj->oo_sem, role);
945 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
948 struct osd_object *obj = osd_dt_obj(dt);
950 LASSERT(osd_invariant(obj));
952 down_write_nested(&obj->oo_sem, role);
955 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
957 struct osd_object *obj = osd_dt_obj(dt);
959 LASSERT(osd_invariant(obj));
960 up_read(&obj->oo_sem);
963 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
965 struct osd_object *obj = osd_dt_obj(dt);
967 LASSERT(osd_invariant(obj));
968 up_write(&obj->oo_sem);
971 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
973 struct osd_object *obj = osd_dt_obj(dt);
976 LASSERT(osd_invariant(obj));
978 if (down_write_trylock(&obj->oo_sem)) {
980 up_write(&obj->oo_sem);
985 static int osd_attr_get(const struct lu_env *env,
986 struct dt_object *dt,
987 struct lu_attr *attr)
989 struct osd_object *obj = osd_dt_obj(dt);
994 down_read(&obj->oo_guard);
996 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
997 GOTO(out, rc = -ENOENT);
999 if (unlikely(fid_is_acct(lu_object_fid(&dt->do_lu))))
1002 LASSERT(osd_invariant(obj));
1003 LASSERT(obj->oo_dn);
1005 read_lock(&obj->oo_attr_lock);
1006 *attr = obj->oo_attr;
1007 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
1008 attr->la_flags |= LUSTRE_ORPHAN_FL;
1009 read_unlock(&obj->oo_attr_lock);
1011 /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
1012 * from within sa_object_size() can block on a mutex, so
1013 * we can't call sa_object_size() holding rwlock */
1014 sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
1015 /* we do not control size of indices, so always calculate
1016 * it from number of blocks reported by DMU */
1017 if (S_ISDIR(attr->la_mode))
1018 attr->la_size = 512 * blocks;
1019 /* Block size may be not set; suggest maximal I/O transfers. */
1021 blksize = osd_spa_maxblocksize(
1022 dmu_objset_spa(osd_obj2dev(obj)->od_os));
1024 attr->la_blksize = blksize;
1025 attr->la_blocks = blocks;
1026 attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
1029 up_read(&obj->oo_guard);
1033 /* Simple wrapper on top of qsd API which implement quota transfer for osd
1034 * setattr needs. As a reminder, only the root user can change ownership of
1035 * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
1036 static inline int qsd_transfer(const struct lu_env *env,
1037 struct qsd_instance *qsd,
1038 struct lquota_trans *trans, int qtype,
1039 __u64 orig_id, __u64 new_id, __u64 bspace,
1040 struct lquota_id_info *qi, bool ignore_edquot)
1044 if (unlikely(qsd == NULL))
1047 LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
1048 qi->lqi_type = qtype;
1050 /* inode accounting */
1051 qi->lqi_is_blk = false;
1053 /* one more inode for the new owner ... */
1054 qi->lqi_id.qid_uid = new_id;
1056 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1057 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1062 /* and one less inode for the current id */
1063 qi->lqi_id.qid_uid = orig_id;;
1065 /* can't get EDQUOT when reducing usage */
1066 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1067 if (rc == -EINPROGRESS)
1072 /* block accounting */
1073 qi->lqi_is_blk = true;
1075 /* more blocks for the new owner ... */
1076 qi->lqi_id.qid_uid = new_id;
1077 qi->lqi_space = bspace;
1078 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1079 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1084 /* and finally less blocks for the current owner */
1085 qi->lqi_id.qid_uid = orig_id;
1086 qi->lqi_space = -bspace;
1087 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1088 /* can't get EDQUOT when reducing usage */
1089 if (rc == -EINPROGRESS)
1094 static int osd_declare_attr_set(const struct lu_env *env,
1095 struct dt_object *dt,
1096 const struct lu_attr *attr,
1097 struct thandle *handle)
1099 struct osd_thread_info *info = osd_oti_get(env);
1100 struct osd_object *obj = osd_dt_obj(dt);
1101 struct osd_device *osd = osd_obj2dev(obj);
1103 struct osd_thandle *oh;
1111 LASSERT(handle != NULL);
1112 LASSERT(osd_invariant(obj));
1114 oh = container_of0(handle, struct osd_thandle, ot_super);
1116 down_read(&obj->oo_guard);
1117 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1120 LASSERT(obj->oo_sa_hdl != NULL);
1121 LASSERT(oh->ot_tx != NULL);
1122 /* regular attributes are part of the bonus buffer */
1123 /* let's check whether this object is already part of
1126 for (txh = list_head(&oh->ot_tx->tx_holds); txh;
1127 txh = list_next(&oh->ot_tx->tx_holds, txh)) {
1128 if (txh->txh_dnode == NULL)
1130 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
1132 /* this object is part of the transaction already
1133 * we don't need to declare bonus again */
1138 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
1139 if (oh->ot_tx->tx_err != 0)
1140 GOTO(out, rc = -oh->ot_tx->tx_err);
1142 if (attr && attr->la_valid & LA_FLAGS) {
1143 /* LMA is usually a part of bonus, no need to declare
1147 if (attr && (attr->la_valid & (LA_UID | LA_GID | LA_PROJID))) {
1148 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
1149 bspace = toqb(bspace * 512);
1151 CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u,"
1152 "the quota space is %lld (%u)\n", osd->od_svname,
1153 attr->la_uid, attr->la_gid, bspace, blksize);
1156 if (attr && attr->la_valid & LA_UID) {
1157 /* quota enforcement for user */
1158 if (attr->la_uid != obj->oo_attr.la_uid) {
1159 rc = qsd_transfer(env, osd->od_quota_slave,
1160 &oh->ot_quota_trans, USRQUOTA,
1161 obj->oo_attr.la_uid, attr->la_uid,
1162 bspace, &info->oti_qi, true);
1167 if (attr && attr->la_valid & LA_GID) {
1168 /* quota enforcement for group */
1169 if (attr->la_gid != obj->oo_attr.la_gid) {
1170 rc = qsd_transfer(env, osd->od_quota_slave,
1171 &oh->ot_quota_trans, GRPQUOTA,
1172 obj->oo_attr.la_gid, attr->la_gid,
1173 bspace, &info->oti_qi,
1175 LUSTRE_SET_SYNC_FL));
1180 #ifdef ZFS_PROJINHERIT
1181 if (attr && attr->la_valid & LA_PROJID) {
1182 if (!osd->od_projectused_dn)
1183 GOTO(out, rc = -EOPNOTSUPP);
1185 /* Usually, if project quota is upgradable for the device,
1186 * then the upgrade will be done before or when mount the
1187 * device. So when we come here, this project should have
1188 * project ID attribute already (that is zero by default).
1189 * Otherwise, there was something wrong during the former
1190 * upgrade, let's return failure to report that.
1192 * Please note that, different from other attributes, you
1193 * can NOT simply set the project ID attribute under such
1194 * case, because adding (NOT change) project ID attribute
1195 * needs to change the object's attribute layout to match
1196 * zfs backend quota accounting requirement. */
1197 if (unlikely(!obj->oo_with_projid))
1198 GOTO(out, rc = -ENXIO);
1200 /* quota enforcement for project */
1201 if (attr->la_projid != obj->oo_attr.la_projid) {
1202 rc = qsd_transfer(env, osd->od_quota_slave,
1203 &oh->ot_quota_trans, PRJQUOTA,
1204 obj->oo_attr.la_projid,
1205 attr->la_projid, bspace,
1206 &info->oti_qi, true);
1213 up_read(&obj->oo_guard);
1218 * Set the attributes of an object
1220 * The transaction passed to this routine must have
1221 * dmu_tx_hold_bonus(tx, oid) called and then assigned
1222 * to a transaction group.
1224 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
1225 const struct lu_attr *la, struct thandle *handle)
1227 struct osd_thread_info *info = osd_oti_get(env);
1228 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1229 struct osd_object *obj = osd_dt_obj(dt);
1230 struct osd_device *osd = osd_obj2dev(obj);
1231 struct osd_thandle *oh;
1232 struct osa_attr *osa = &info->oti_osa;
1233 __u64 valid = la->la_valid;
1239 down_read(&obj->oo_guard);
1240 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1241 GOTO(out, rc = -ENOENT);
1243 LASSERT(handle != NULL);
1244 LASSERT(osd_invariant(obj));
1245 LASSERT(obj->oo_sa_hdl);
1247 oh = container_of0(handle, struct osd_thandle, ot_super);
1248 /* Assert that the transaction has been assigned to a
1249 transaction group. */
1250 LASSERT(oh->ot_tx->tx_txg != 0);
1252 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) && !osd->od_is_ost) {
1253 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1254 char *buf = info->oti_str;
1255 dnode_t *zdn = NULL;
1258 zapid = osd_get_name_n_idx(env, osd, lu_object_fid(&dt->do_lu),
1259 buf, sizeof(info->oti_str), &zdn);
1260 rc = osd_zap_lookup(osd, zapid, zdn, buf, 8,
1261 sizeof(*zde) / 8, zde);
1263 zde->zde_dnode -= 1;
1264 rc = -zap_update(osd->od_os, zapid, buf, 8,
1265 sizeof(*zde) / 8, zde, oh->ot_tx);
1267 up_read(&obj->oo_guard);
1269 RETURN(rc > 0 ? 0 : rc);
1272 /* Only allow set size for regular file */
1273 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
1274 valid &= ~(LA_SIZE | LA_BLOCKS);
1276 if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
1279 if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
1282 if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
1288 if (valid & LA_FLAGS) {
1289 struct lustre_mdt_attrs *lma;
1292 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
1293 LASSERT(!obj->oo_pfid_in_lma);
1294 CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
1295 lma = (struct lustre_mdt_attrs *)&info->oti_buf;
1297 buf.lb_len = sizeof(info->oti_buf);
1298 rc = osd_xattr_get(env, &obj->oo_dt, &buf,
1302 le32_to_cpu(lma->lma_incompat);
1303 lma->lma_incompat |=
1304 lustre_to_lma_flags(la->la_flags);
1306 cpu_to_le32(lma->lma_incompat);
1308 buf.lb_len = sizeof(*lma);
1309 rc = osd_xattr_set_internal(env, obj, &buf,
1315 CWARN("%s: failed to set LMA flags: rc = %d\n",
1316 osd->od_svname, rc);
1322 write_lock(&obj->oo_attr_lock);
1325 if (valid & LA_PROJID) {
1326 #ifdef ZFS_PROJINHERIT
1327 /* osd_declare_attr_set() must be called firstly.
1328 * If osd::od_projectused_dn is not set, then we
1329 * can not arrive at here. */
1330 LASSERT(osd->od_projectused_dn);
1331 LASSERT(obj->oo_with_projid);
1333 osa->projid = obj->oo_attr.la_projid = la->la_projid;
1334 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1337 valid &= ~LA_PROJID;
1341 if (valid & LA_ATIME) {
1342 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1343 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1346 if (valid & LA_MTIME) {
1347 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1348 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1351 if (valid & LA_CTIME) {
1352 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1353 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1356 if (valid & LA_MODE) {
1357 /* mode is stored along with type, so read it first */
1358 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1359 (la->la_mode & ~S_IFMT);
1360 osa->mode = obj->oo_attr.la_mode;
1361 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1364 if (valid & LA_SIZE) {
1365 osa->size = obj->oo_attr.la_size = la->la_size;
1366 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1369 if (valid & LA_NLINK) {
1370 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1371 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1374 if (valid & LA_RDEV) {
1375 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1376 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1379 if (valid & LA_FLAGS) {
1380 osa->flags = attrs_fs2zfs(la->la_flags);
1381 /* many flags are not supported by zfs, so ensure a good cached
1383 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1384 #ifdef ZFS_PROJINHERIT
1385 if (obj->oo_with_projid)
1386 osa->flags |= ZFS_PROJID;
1388 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1391 if (valid & LA_UID) {
1392 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1393 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1396 if (valid & LA_GID) {
1397 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1398 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1401 obj->oo_attr.la_valid |= valid;
1402 write_unlock(&obj->oo_attr_lock);
1404 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1405 rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1408 up_read(&obj->oo_guard);
1415 * XXX temporary solution.
1418 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1419 struct dt_object *parent, struct dt_object *child,
1424 ah->dah_parent = parent;
1425 ah->dah_mode = child_mode;
1427 if (parent != NULL && !dt_object_remote(parent)) {
1428 /* will help to find FID->ino at dt_insert("..") */
1429 struct osd_object *pobj = osd_dt_obj(parent);
1431 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1435 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
1436 struct lu_attr *attr,
1437 struct dt_allocation_hint *hint,
1438 struct dt_object_format *dof,
1439 struct thandle *handle)
1441 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1442 struct osd_object *obj = osd_dt_obj(dt);
1443 struct osd_device *osd = osd_obj2dev(obj);
1444 struct osd_thandle *oh;
1452 switch (dof->dof_type) {
1456 if (obj->oo_dt.do_body_ops == NULL)
1457 obj->oo_dt.do_body_ops = &osd_body_ops;
1463 LASSERT(handle != NULL);
1464 oh = container_of0(handle, struct osd_thandle, ot_super);
1465 LASSERT(oh->ot_tx != NULL);
1467 /* this is the minimum set of EAs on every Lustre object */
1468 obj->oo_ea_in_bonus = OSD_BASE_EA_IN_BONUS;
1469 /* reserve 32 bytes for extra stuff like ACLs */
1470 dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1472 switch (dof->dof_type) {
1474 dt->do_index_ops = &osd_dir_ops;
1476 /* for zap create */
1477 dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1478 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1483 /* first, we'll create new object */
1484 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1492 /* and we'll add it to some mapping */
1493 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
1494 osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
1496 /* will help to find FID->ino mapping at dt_insert() */
1497 osd_idc_find_and_init(env, osd, obj);
1499 rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid,
1500 attr->la_projid, 1, oh, NULL, OSD_QID_INODE);
1505 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1506 struct osd_object *obj, sa_handle_t *sa_hdl, dmu_tx_t *tx,
1507 struct lu_attr *la, uint64_t parent,
1510 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1511 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1517 char *dxattr = NULL;
1523 gen = dmu_tx_get_txg(tx);
1525 ZFS_TIME_ENCODE(&now, crtime);
1527 osa->atime[0] = la->la_atime;
1528 osa->ctime[0] = la->la_ctime;
1529 osa->mtime[0] = la->la_mtime;
1530 osa->mode = la->la_mode;
1531 osa->uid = la->la_uid;
1532 osa->gid = la->la_gid;
1533 osa->rdev = la->la_rdev;
1534 osa->nlink = la->la_nlink;
1535 if (la->la_valid & LA_FLAGS)
1536 osa->flags = attrs_fs2zfs(la->la_flags);
1539 osa->size = la->la_size;
1540 #ifdef ZFS_PROJINHERIT
1541 if (osd->od_projectused_dn) {
1542 if (la->la_valid & LA_PROJID)
1543 osa->projid = la->la_projid;
1545 osa->projid = ZFS_DEFAULT_PROJID;
1546 osa->flags |= ZFS_PROJID;
1548 obj->oo_with_projid = 1;
1550 osa->flags &= ~ZFS_PROJID;
1555 * we need to create all SA below upon object create.
1557 * XXX The attribute order matters since the accounting callback relies
1558 * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1559 * look up the UID/GID/PROJID attributes. Moreover, the callback does
1560 * not seem to support the spill block.
1561 * We define attributes in the same order as SA_*_OFFSET in order to
1562 * work around the problem. See ORI-610.
1565 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1566 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1567 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1568 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1569 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1570 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1571 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1572 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1573 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1574 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1575 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1576 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1577 #ifdef ZFS_PROJINHERIT
1578 if (osd->od_projectused_dn)
1579 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1582 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1583 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1586 rc = -nvlist_size(xattr, &sa_size, NV_ENCODE_XDR);
1589 dxattr = osd_zio_buf_alloc(sa_size);
1592 rc = -nvlist_pack(xattr, &dxattr, &sa_size,
1593 NV_ENCODE_XDR, KM_SLEEP);
1596 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd),
1597 NULL, dxattr, sa_size);
1600 rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1602 osd_zio_buf_free(dxattr, sa_size);
1607 int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1608 uint64_t oid, dnode_t **dnp)
1613 /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1614 for (txh = list_tail(&tx->tx_holds); txh;
1615 txh = list_prev(&tx->tx_holds, txh)) {
1616 dnode_t *dn = txh->txh_dnode;
1621 if (dn->dn_object != oid)
1625 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1626 if (dn->dn_bonus == NULL)
1627 dbuf_create_bonus(dn);
1628 rw_exit(&dn->dn_struct_rwlock);
1632 LASSERT(dn->dn_handle);
1634 if (refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1635 refcount_add(&dn->dn_holds, tag);
1636 atomic_inc_32(&dn->dn_dbufs_count);
1639 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1643 if (unlikely(*dnp == NULL))
1644 rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
1649 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
1650 int osd_find_dnsize(struct osd_device *osd, int ea_in_bonus)
1654 if (osd->od_dnsize == ZFS_DNSIZE_AUTO) {
1655 dnsize = DNODE_MIN_SIZE;
1657 if (DN_BONUS_SIZE(dnsize) >= ea_in_bonus + 32)
1660 } while (dnsize < DNODE_MAX_SIZE);
1661 if (dnsize > DNODE_MAX_SIZE)
1662 dnsize = DNODE_MAX_SIZE;
1663 } else if (osd->od_dnsize == ZFS_DNSIZE_1K) {
1665 } else if (osd->od_dnsize == ZFS_DNSIZE_2K) {
1667 } else if (osd->od_dnsize == ZFS_DNSIZE_4K) {
1669 } else if (osd->od_dnsize == ZFS_DNSIZE_8K) {
1671 } else if (osd->od_dnsize == ZFS_DNSIZE_16K) {
1674 dnsize = DNODE_MIN_SIZE;
1681 * The transaction passed to this routine must have
1682 * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1683 * to a transaction group.
1685 int __osd_object_create(const struct lu_env *env, struct osd_device *osd,
1686 struct osd_object *obj, const struct lu_fid *fid,
1687 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1689 dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1693 /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1694 * would get an additional ditto copy */
1695 if (unlikely(S_ISREG(la->la_mode) &&
1696 fid_seq_is_local_file(fid_seq(fid))))
1697 type = DMU_OTN_UINT8_METADATA;
1699 /* Create a new DMU object using the default dnode size. */
1701 size = obj->oo_ea_in_bonus;
1703 size = OSD_BASE_EA_IN_BONUS;
1704 oid = osd_dmu_object_alloc(osd->od_os, type, 0,
1705 osd_find_dnsize(osd, size), tx);
1707 LASSERT(la->la_valid & LA_MODE);
1711 return osd_find_new_dnode(env, tx, oid, dnp);
1715 * The transaction passed to this routine must have
1716 * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1717 * to a transaction group.
1719 * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1720 * This is fine for directories today, because storing the FID in the dirent
1721 * will also require a FAT ZAP. If there is a new type of micro ZAP created
1722 * then we might need to re-evaluate the use of this flag and instead do
1723 * a conversion from the different internal ZAP hash formats being used. */
1724 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1725 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1726 unsigned dnsize, zap_flags_t flags)
1730 /* Assert that the transaction has been assigned to a
1731 transaction group. */
1732 LASSERT(tx->tx_txg != 0);
1735 oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1736 DMU_OT_DIRECTORY_CONTENTS,
1737 14, /* == ZFS fzap_default_blockshift */
1738 DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1744 return osd_find_new_dnode(env, tx, oid, dnp);
1747 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1748 struct lu_attr *la, struct osd_thandle *oh)
1750 struct osd_device *osd = osd_obj2dev(obj);
1754 /* Index file should be created as regular file in order not to confuse
1755 * ZPL which could interpret them as directory.
1756 * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1758 LASSERT(S_ISREG(la->la_mode));
1759 rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1760 osd_find_dnsize(osd, obj->oo_ea_in_bonus), ZAP_FLAG_UINT64_KEY);
1766 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1767 struct lu_attr *la, struct osd_thandle *oh)
1769 struct osd_device *osd = osd_obj2dev(obj);
1773 LASSERT(S_ISDIR(la->la_mode));
1774 rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1775 osd_find_dnsize(osd, obj->oo_ea_in_bonus), 0);
1781 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1782 struct lu_attr *la, struct osd_thandle *oh)
1784 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1785 struct osd_device *osd = osd_obj2dev(obj);
1789 LASSERT(S_ISREG(la->la_mode));
1790 rc = __osd_object_create(env, osd, obj, fid, &dn, oh->ot_tx, la);
1794 if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid))) {
1795 /* The minimum block size must be at least page size otherwise
1796 * it will break the assumption in tgt_thread_big_cache where
1797 * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1798 * RDMA due to subpage transfer size */
1799 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1800 PAGE_SIZE, 0, oh->ot_tx);
1802 CERROR("%s: can't change blocksize: %d\n",
1803 osd->od_svname, rc);
1811 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1812 struct lu_attr *la, struct osd_thandle *oh)
1817 LASSERT(S_ISLNK(la->la_mode));
1818 rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1819 lu_object_fid(&obj->oo_dt.do_lu),
1820 &dn, oh->ot_tx, la);
1826 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1827 struct lu_attr *la, struct osd_thandle *oh)
1832 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1833 la->la_valid |= LA_RDEV;
1835 rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1836 lu_object_fid(&obj->oo_dt.do_lu),
1837 &dn, oh->ot_tx, la);
1843 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1844 struct osd_object *obj,
1846 struct osd_thandle *oh);
1848 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1850 osd_obj_type_f result;
1876 * Concurrency: @dt is write locked.
1878 static int osd_create(const struct lu_env *env, struct dt_object *dt,
1879 struct lu_attr *attr, struct dt_allocation_hint *hint,
1880 struct dt_object_format *dof, struct thandle *th)
1882 struct osd_thread_info *info = osd_oti_get(env);
1883 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1884 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1885 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1886 struct osd_object *obj = osd_dt_obj(dt);
1887 struct osd_device *osd = osd_obj2dev(obj);
1888 char *buf = info->oti_str;
1889 struct osd_thandle *oh;
1890 dnode_t *dn = NULL, *zdn = NULL;
1891 uint64_t zapid, parent = 0;
1897 LASSERT(!fid_is_acct(fid));
1899 /* concurrent create declarations should not see
1900 * the object inconsistent (db, attr, etc).
1901 * in regular cases acquisition should be cheap */
1902 down_write(&obj->oo_guard);
1904 if (unlikely(dt_object_exists(dt)))
1905 GOTO(out, rc = -EEXIST);
1907 LASSERT(osd_invariant(obj));
1908 LASSERT(dof != NULL);
1910 LASSERT(th != NULL);
1911 oh = container_of0(th, struct osd_thandle, ot_super);
1913 LASSERT(obj->oo_dn == NULL);
1915 /* to follow ZFS on-disk format we need
1916 * to initialize parent dnode properly */
1917 if (hint != NULL && hint->dah_parent != NULL &&
1918 !dt_object_remote(hint->dah_parent))
1919 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1921 /* we may fix some attributes, better do not change the source */
1922 obj->oo_attr = *attr;
1923 obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1925 #ifdef ZFS_PROJINHERIT
1926 if (osd->od_projectused_dn) {
1927 if (!(obj->oo_attr.la_valid & LA_PROJID))
1928 obj->oo_attr.la_projid = ZFS_DEFAULT_PROJID;
1929 obj->oo_with_projid = 1;
1933 dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1941 zde->zde_dnode = dn->dn_object;
1942 zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1944 zapid = osd_get_name_n_idx(env, osd, fid, buf,
1945 sizeof(info->oti_str), &zdn);
1946 if (CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY) ||
1947 (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY)))
1950 if (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
1953 rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
1959 /* Now add in all of the "SA" attributes */
1960 rc = osd_sa_handle_get(obj);
1964 rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1968 /* initialize LMA */
1969 if (fid_is_idif(fid) || (fid_is_norm(fid) && osd->od_is_ost))
1970 compat |= LMAC_FID_ON_OST;
1971 lustre_lma_init(lma, fid, compat, 0);
1972 lustre_lma_swab(lma);
1973 rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
1974 (uchar_t *)lma, sizeof(*lma));
1978 /* configure new osd object */
1979 obj->oo_parent = parent != 0 ? parent : zapid;
1980 obj->oo_late_attr_set = 1;
1981 rc = __osd_sa_xattr_schedule_update(env, obj, oh);
1985 /* XXX: oo_lma_flags */
1986 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
1987 if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
1988 /* no body operations for accounting objects */
1989 obj->oo_dt.do_body_ops = &osd_body_ops;
1991 osd_idc_find_and_init(env, osd, obj);
1994 if (unlikely(rc && dn)) {
1995 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
1999 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
2001 up_write(&obj->oo_guard);
2005 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
2008 osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2009 return osd_declare_attr_set(env, dt, NULL, th);
2013 * Concurrency: @dt is write locked.
2015 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
2016 struct thandle *handle)
2018 struct osd_object *obj = osd_dt_obj(dt);
2019 struct osd_thandle *oh;
2020 struct osd_device *osd = osd_obj2dev(obj);
2026 down_read(&obj->oo_guard);
2027 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2028 GOTO(out, rc = -ENOENT);
2030 LASSERT(osd_invariant(obj));
2031 LASSERT(obj->oo_sa_hdl != NULL);
2033 oh = container_of0(handle, struct osd_thandle, ot_super);
2035 write_lock(&obj->oo_attr_lock);
2036 nlink = ++obj->oo_attr.la_nlink;
2037 write_unlock(&obj->oo_attr_lock);
2039 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2042 up_read(&obj->oo_guard);
2046 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
2047 struct thandle *handle)
2049 osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2050 return osd_declare_attr_set(env, dt, NULL, handle);
2054 * Concurrency: @dt is write locked.
2056 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
2057 struct thandle *handle)
2059 struct osd_object *obj = osd_dt_obj(dt);
2060 struct osd_thandle *oh;
2061 struct osd_device *osd = osd_obj2dev(obj);
2067 down_read(&obj->oo_guard);
2069 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2070 GOTO(out, rc = -ENOENT);
2072 LASSERT(osd_invariant(obj));
2073 LASSERT(obj->oo_sa_hdl != NULL);
2075 oh = container_of0(handle, struct osd_thandle, ot_super);
2076 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2078 write_lock(&obj->oo_attr_lock);
2079 nlink = --obj->oo_attr.la_nlink;
2080 write_unlock(&obj->oo_attr_lock);
2082 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2085 up_read(&obj->oo_guard);
2089 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
2090 __u64 start, __u64 end)
2092 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
2095 /* XXX: no other option than syncing the whole filesystem until we
2096 * support ZIL. If the object tracked the txg that it was last
2097 * modified in, it could pass that txg here instead of "0". Maybe
2098 * the changes are already committed, so no wait is needed at all? */
2099 if (!osd->od_dt_dev.dd_rdonly) {
2100 if (osd_object_sync_delay_us < 0)
2101 txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
2103 udelay(osd_object_sync_delay_us);
2109 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
2114 static struct dt_object_operations osd_obj_ops = {
2115 .do_read_lock = osd_read_lock,
2116 .do_write_lock = osd_write_lock,
2117 .do_read_unlock = osd_read_unlock,
2118 .do_write_unlock = osd_write_unlock,
2119 .do_write_locked = osd_write_locked,
2120 .do_attr_get = osd_attr_get,
2121 .do_declare_attr_set = osd_declare_attr_set,
2122 .do_attr_set = osd_attr_set,
2123 .do_ah_init = osd_ah_init,
2124 .do_declare_create = osd_declare_create,
2125 .do_create = osd_create,
2126 .do_declare_destroy = osd_declare_destroy,
2127 .do_destroy = osd_destroy,
2128 .do_index_try = osd_index_try,
2129 .do_declare_ref_add = osd_declare_ref_add,
2130 .do_ref_add = osd_ref_add,
2131 .do_declare_ref_del = osd_declare_ref_del,
2132 .do_ref_del = osd_ref_del,
2133 .do_xattr_get = osd_xattr_get,
2134 .do_declare_xattr_set = osd_declare_xattr_set,
2135 .do_xattr_set = osd_xattr_set,
2136 .do_declare_xattr_del = osd_declare_xattr_del,
2137 .do_xattr_del = osd_xattr_del,
2138 .do_xattr_list = osd_xattr_list,
2139 .do_object_sync = osd_object_sync,
2140 .do_invalidate = osd_invalidate,
2143 static struct lu_object_operations osd_lu_obj_ops = {
2144 .loo_object_init = osd_object_init,
2145 .loo_object_delete = osd_object_delete,
2146 .loo_object_release = osd_object_release,
2147 .loo_object_free = osd_object_free,
2148 .loo_object_print = osd_object_print,
2149 .loo_object_invariant = osd_object_invariant,
2152 static int osd_otable_it_attr_get(const struct lu_env *env,
2153 struct dt_object *dt,
2154 struct lu_attr *attr)
2160 static struct dt_object_operations osd_obj_otable_it_ops = {
2161 .do_attr_get = osd_otable_it_attr_get,
2162 .do_index_try = osd_index_try,
2165 module_param(osd_object_sync_delay_us, int, 0644);
2166 MODULE_PARM_DESC(osd_object_sync_delay_us,
2167 "If zero or larger delay N usec instead of doing object sync");