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 static struct dt_object_operations osd_obj_otable_it_ops;
72 osd_object_sa_fini(struct osd_object *obj)
75 sa_handle_destroy(obj->oo_sa_hdl);
76 obj->oo_sa_hdl = NULL;
81 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
85 LASSERT(obj->oo_sa_hdl == NULL);
86 LASSERT(obj->oo_dn != NULL);
88 rc = osd_sa_handle_get(obj);
92 /* Cache the xattr object id, valid for the life of the object */
93 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
95 obj->oo_xattr = ZFS_NO_OBJECT;
98 osd_object_sa_fini(obj);
105 * Add object to list of dirty objects in tx handle.
107 void osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
109 if (!list_empty(&obj->oo_sa_linkage))
112 write_lock(&obj->oo_attr_lock);
113 if (likely(list_empty(&obj->oo_sa_linkage)))
114 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
115 write_unlock(&obj->oo_attr_lock);
119 * Release spill block dbuf hold for all dirty SAs.
121 void osd_object_sa_dirty_rele(const struct lu_env *env, struct osd_thandle *oh)
123 struct osd_object *obj;
125 while (!list_empty(&oh->ot_sa_list)) {
126 obj = list_entry(oh->ot_sa_list.next,
127 struct osd_object, oo_sa_linkage);
128 write_lock(&obj->oo_attr_lock);
129 list_del_init(&obj->oo_sa_linkage);
130 write_unlock(&obj->oo_attr_lock);
131 if (obj->oo_late_xattr) {
133 * take oo_guard to protect oo_sa_xattr buffer
134 * from concurrent update by osd_xattr_set()
136 LASSERT(oh->ot_assigned != 0);
137 down_write(&obj->oo_guard);
138 if (obj->oo_late_attr_set)
139 __osd_sa_attr_init(env, obj, oh);
140 else if (obj->oo_late_xattr)
141 __osd_sa_xattr_update(env, obj, oh);
142 up_write(&obj->oo_guard);
144 sa_spill_rele(obj->oo_sa_hdl);
149 * Update the SA and add the object to the dirty list.
151 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
152 void *buf, uint32_t buflen, struct osd_thandle *oh)
156 LASSERT(obj->oo_sa_hdl != NULL);
157 LASSERT(oh->ot_tx != NULL);
159 rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
160 osd_object_sa_dirty_add(obj, oh);
166 * Bulk update the SA and add the object to the dirty list.
169 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
170 int count, struct osd_thandle *oh)
174 LASSERT(obj->oo_sa_hdl != NULL);
175 LASSERT(oh->ot_tx != NULL);
177 rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
178 osd_object_sa_dirty_add(obj, oh);
184 * Retrieve the attributes of a DMU object
186 static int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
187 struct osd_object *obj, struct lu_attr *la)
189 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
190 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
195 LASSERT(obj->oo_dn != NULL);
197 la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_BTIME | LA_MODE |
198 LA_TYPE | LA_SIZE | LA_UID | LA_GID | LA_FLAGS |
201 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
202 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
203 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
204 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(o), NULL, osa->btime, 16);
205 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
206 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
207 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
208 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
209 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
210 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
211 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
213 rc = -sa_bulk_lookup(obj->oo_sa_hdl, bulk, cnt);
217 #ifdef ZFS_PROJINHERIT
218 if (o->od_projectused_dn && osa->flags & ZFS_PROJID) {
219 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_PROJID(o),
224 la->la_projid = osa->projid;
225 la->la_valid |= LA_PROJID;
226 obj->oo_with_projid = 1;
228 la->la_projid = ZFS_DEFAULT_PROJID;
229 la->la_valid &= ~LA_PROJID;
233 la->la_valid &= ~LA_PROJID;
236 la->la_atime = osa->atime[0];
237 la->la_mtime = osa->mtime[0];
238 la->la_ctime = osa->ctime[0];
239 la->la_btime = osa->btime[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);
295 LASSERT(*dnp != NULL);
301 * Concurrency: no concurrent access is possible that early in object
304 struct lu_object *osd_object_alloc(const struct lu_env *env,
305 const struct lu_object_header *hdr,
308 struct osd_object *mo;
310 OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
313 struct lu_object_header *h;
314 struct osd_device *o = osd_dev(d);
316 l = &mo->oo_dt.do_lu;
317 if (unlikely(o->od_in_init)) {
324 lu_object_header_init(h);
325 lu_object_init(l, h, d);
326 lu_object_add_top(h, l);
329 dt_object_init(&mo->oo_dt, NULL, d);
330 mo->oo_header = NULL;
333 mo->oo_dt.do_ops = &osd_obj_ops;
334 l->lo_ops = &osd_lu_obj_ops;
335 INIT_LIST_HEAD(&mo->oo_sa_linkage);
336 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
337 init_rwsem(&mo->oo_sem);
338 init_rwsem(&mo->oo_guard);
339 rwlock_init(&mo->oo_attr_lock);
340 mo->oo_destroy = OSD_DESTROY_NONE;
347 static void osd_obj_set_blksize(const struct lu_env *env,
348 struct osd_device *osd, struct osd_object *obj)
350 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
352 dnode_t *dn = obj->oo_dn;
357 LASSERT(!osd_oti_get(env)->oti_in_trans);
359 tx = dmu_tx_create(osd->od_os);
361 CERROR("%s: fail to create tx to set blksize for "DFID"\n",
362 osd->od_svname, PFID(fid));
366 dmu_tx_hold_bonus(tx, dn->dn_object);
367 rc = -dmu_tx_assign(tx, TXG_WAIT);
370 CERROR("%s: fail to assign tx to set blksize for "DFID
371 ": rc = %d\n", osd->od_svname, PFID(fid), rc);
375 down_write(&obj->oo_guard);
376 if (unlikely((1 << dn->dn_datablkshift) >= PAGE_SIZE))
379 blksz = dn->dn_datablksz;
380 if (!is_power_of_2(blksz))
381 blksz = size_roundup_power2(blksz);
383 if (blksz > osd->od_max_blksz)
384 blksz = osd->od_max_blksz;
385 else if (blksz < PAGE_SIZE)
387 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object, blksz, 0, tx);
392 up_write(&obj->oo_guard);
395 if (unlikely(obj->oo_dn->dn_maxblkid > 0))
398 CERROR("%s: fail to set blksize for "DFID": rc = %d\n",
399 osd->od_svname, PFID(fid), rc);
402 CDEBUG(D_INODE, "%s: set blksize as %u for "DFID"\n",
403 osd->od_svname, blksz, PFID(fid));
408 * Concurrency: shouldn't matter.
410 static int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
412 struct osd_device *osd = osd_obj2dev(obj);
413 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
419 rc = osd_object_sa_init(obj, osd);
423 /* cache attrs in object */
424 rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
428 if (likely(!fid_is_acct(fid))) {
429 /* no body operations for accounting objects */
430 obj->oo_dt.do_body_ops = &osd_body_ops;
432 if (S_ISREG(obj->oo_attr.la_mode) &&
433 obj->oo_dn->dn_maxblkid == 0 &&
434 (1 << obj->oo_dn->dn_datablkshift) < PAGE_SIZE &&
435 (fid_is_idif(fid) || fid_is_norm(fid) ||
437 osd->od_is_ost && !osd->od_dt_dev.dd_rdonly)
438 osd_obj_set_blksize(env, osd, obj);
442 * initialize object before marking it existing
444 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
447 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
452 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
454 struct osd_thread_info *info = osd_oti_get(env);
457 struct lustre_mdt_attrs *lma;
458 const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
461 BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*lma));
462 lma = (struct lustre_mdt_attrs *)info->oti_buf;
464 buf.lb_len = sizeof(info->oti_buf);
466 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
469 lustre_lma_swab(lma);
470 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
471 CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
472 CWARN("%s: unsupported incompat LMA feature(s) %#x for "
473 "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
474 lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
477 } else if (unlikely(!lu_fid_eq(rfid, &lma->lma_self_fid))) {
478 CERROR("%s: FID-in-LMA "DFID" does not match the "
479 "object self-fid "DFID"\n",
480 osd_obj2dev(obj)->od_svname,
481 PFID(&lma->lma_self_fid), PFID(rfid));
484 struct osd_device *osd = osd_obj2dev(obj);
486 if (lma->lma_compat & LMAC_STRIPE_INFO &&
488 obj->oo_pfid_in_lma = 1;
489 if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
490 osd->od_remote_parent_dir != ZFS_NO_OBJECT)
491 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
493 } else if (rc == -ENODATA) {
494 /* haven't initialize LMA xattr */
502 * Helper function to retrieve DMU object id from fid for accounting object
504 static dnode_t *osd_quota_fid2dmu(const struct osd_device *osd,
505 const struct lu_fid *fid)
509 LASSERT(fid_is_acct(fid));
511 switch (fid_oid(fid)) {
513 dn = osd->od_userused_dn;
516 dn = osd->od_groupused_dn;
518 #ifdef ZFS_PROJINHERIT
519 case ACCT_PROJECT_OID:
520 dn = osd->od_projectused_dn;
531 * Concurrency: no concurrent access is possible that early in object
534 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
535 const struct lu_object_conf *conf)
537 struct osd_object *obj = osd_obj(l);
538 struct osd_device *osd = osd_obj2dev(obj);
539 const struct lu_fid *fid = lu_object_fid(l);
540 struct lustre_scrub *scrub = &osd->od_scrub;
541 struct osd_thread_info *info = osd_oti_get(env);
542 struct luz_direntry *zde = &info->oti_zde;
543 struct osd_idmap_cache *idc;
544 char *name = info->oti_str;
551 LASSERT(osd_invariant(obj));
553 if (fid_is_otable_it(&l->lo_header->loh_fid)) {
554 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
555 l->lo_header->loh_attr |= LOHA_EXISTS;
560 if (conf && conf->loc_flags & LOC_F_NEW)
563 if (unlikely(fid_is_acct(fid))) {
564 obj->oo_dn = osd_quota_fid2dmu(osd, fid);
566 obj->oo_dt.do_index_ops = &osd_acct_index_ops;
567 l->lo_header->loh_attr |= LOHA_EXISTS;
573 idc = osd_idc_find(env, osd, fid);
574 if (idc && !idc->oic_remote && idc->oic_dnode != ZFS_NO_OBJECT) {
575 oid = idc->oic_dnode;
580 if (!list_empty(&osd->od_scrub.os_inconsistent_items))
581 rc = osd_oii_lookup(osd, fid, &oid);
584 rc = osd_fid_lookup(env, osd, fid, &oid);
587 if (likely(!(fid_is_norm(fid) || fid_is_igif(fid)) ||
588 fid_is_on_ost(env, osd, fid) ||
589 !zfs_test_bit(osd_oi_fid2idx(osd, fid),
590 scrub->os_file.sf_oi_bitmap)))
601 LASSERT(obj->oo_dn == NULL);
603 rc = __osd_obj2dnode(osd->od_os, oid, &obj->oo_dn);
604 /* EEXIST will be returned if object is being deleted in ZFS */
609 CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
610 osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
614 rc = osd_object_init0(env, obj);
618 if (unlikely(obj->oo_header))
621 rc = osd_check_lma(env, obj);
622 if ((!rc && !remote) || (rc != -EREMCHG))
626 /* We still have chance to get the valid dnode: for the object that is
627 * referenced by remote name entry, the object on the local MDT will be
628 * linked under the dir /REMOTE_PARENT_DIR with its FID string as name.
630 * During the OI scrub, if we cannot find the OI mapping, we may still
631 * have change to map the FID to local OID via lookup the dir
632 * /REMOTE_PARENT_DIR. */
633 if (!remote && !fid_is_on_ost(env, osd, fid)) {
634 osd_fid2str(name, fid, sizeof(info->oti_str));
635 rc = osd_zap_lookup(osd, osd->od_remote_parent_dir,
636 NULL, name, 8, 3, (void *)zde);
638 oid = zde->lzd_reg.zde_dnode;
639 osd_dnode_rele(obj->oo_dn);
646 /* The case someone triggered the OI scrub already. */
647 if (thread_is_running(&scrub->os_thread)) {
651 lu_object_set_agent_entry(l);
652 osd_oii_insert(env, osd, fid, oid, false);
660 /* The case NOT allow to trigger OI scrub automatically. */
661 if (osd->od_auto_scrub_interval == AS_NEVER)
664 /* It is me to trigger the OI scrub. */
665 rc1 = osd_scrub_start(env, osd, SS_CLEAR_DRYRUN |
666 SS_CLEAR_FAILOUT | SS_AUTO_FULL);
667 LCONSOLE_WARN("%s: trigger OI scrub by RPC for the "DFID": rc = %d\n",
668 osd_name(osd), PFID(fid), rc1);
672 lu_object_set_agent_entry(l);
674 osd_oii_insert(env, osd, fid, oid, false);
689 * Concurrency: no concurrent access is possible that late in object
692 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
694 struct osd_object *obj = osd_obj(l);
695 struct lu_object_header *h = obj->oo_header;
697 LASSERT(osd_invariant(obj));
699 dt_object_fini(&obj->oo_dt);
700 /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
701 OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
703 lu_object_header_fini(h);
704 OBD_FREE_PRE(h, sizeof(*h), "kfreed");
705 kfree_rcu(h, loh_rcu);
710 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
714 LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
716 /* the object is supposed to be exclusively locked by
717 * the caller (osd_destroy()), while the transaction
718 * (oh) is per-thread and not shared */
719 if (likely(list_empty(&obj->oo_unlinked_linkage))) {
720 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
727 /* Default to max data size covered by a level-1 indirect block */
728 static unsigned long osd_sync_destroy_max_size =
729 1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
730 module_param(osd_sync_destroy_max_size, ulong, 0444);
731 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
734 osd_object_set_destroy_type(struct osd_object *obj)
737 * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
738 * only once and use it consistently thereafter.
740 down_write(&obj->oo_guard);
741 if (obj->oo_destroy == OSD_DESTROY_NONE) {
742 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
743 obj->oo_destroy = OSD_DESTROY_SYNC;
744 else /* Larger objects are destroyed asynchronously */
745 obj->oo_destroy = OSD_DESTROY_ASYNC;
747 up_write(&obj->oo_guard);
750 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
753 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
754 struct osd_object *obj = osd_dt_obj(dt);
755 struct osd_device *osd = osd_obj2dev(obj);
756 struct osd_thandle *oh;
763 LASSERT(dt_object_exists(dt));
765 oh = container_of(th, struct osd_thandle, ot_super);
766 LASSERT(oh->ot_tx != NULL);
768 dmu_tx_mark_netfree(oh->ot_tx);
770 /* declare that we'll remove object from fid-dnode mapping */
771 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
772 osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
774 osd_declare_xattrs_destroy(env, obj, oh);
777 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
778 obj->oo_attr.la_gid, obj->oo_attr.la_projid,
779 -1, oh, NULL, OSD_QID_INODE);
783 /* data to be truncated */
784 rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
785 obj->oo_attr.la_gid, obj->oo_attr.la_projid,
786 0, oh, NULL, OSD_QID_BLK);
790 osd_object_set_destroy_type(obj);
791 if (obj->oo_destroy == OSD_DESTROY_SYNC)
792 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
795 osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
796 osd->od_unlinked, TRUE, NULL);
798 /* remove agent entry (if have) from remote parent */
799 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
800 osd_tx_hold_zap(oh->ot_tx, osd->od_remote_parent_dir,
803 /* will help to find FID->ino when this object is being
804 * added to PENDING/ */
805 osd_idc_find_and_init(env, osd, obj);
810 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
813 struct osd_thread_info *info = osd_oti_get(env);
814 char *buf = info->oti_str;
815 struct osd_object *obj = osd_dt_obj(dt);
816 struct osd_device *osd = osd_obj2dev(obj);
817 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
818 struct osd_thandle *oh;
824 down_write(&obj->oo_guard);
826 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
827 GOTO(out, rc = -ENOENT);
829 LASSERT(obj->oo_dn != NULL);
831 oh = container_of(th, struct osd_thandle, ot_super);
833 LASSERT(oh->ot_tx != NULL);
835 /* remove obj ref from index dir (it depends) */
836 zapid = osd_get_name_n_idx(env, osd, fid, buf,
837 sizeof(info->oti_str), &zdn);
838 rc = osd_xattrs_destroy(env, obj, oh);
840 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
841 osd->od_svname, buf, rc);
845 if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
846 rc = osd_delete_from_remote_parent(env, osd, obj, oh, true);
851 oid = obj->oo_dn->dn_object;
852 if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
853 /* this may happen if the destroy wasn't declared
854 * e.g. when the object is created and then destroyed
855 * in the same transaction - we don't need additional
856 * space for destroy specifically */
857 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
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 if (obj->oo_destroy == OSD_DESTROY_SYNC) {
863 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
865 CERROR("%s: failed to free %s %llu: rc = %d\n",
866 osd->od_svname, buf, oid, rc);
867 } else { /* asynchronous destroy */
868 char *key = info->oti_key;
870 rc = osd_object_unlinked_add(obj, oh);
874 snprintf(key, sizeof(info->oti_key), "%llx", oid);
875 rc = osd_zap_add(osd, osd->od_unlinked->dn_object,
876 osd->od_unlinked, key, 8, 1, &oid, oh->ot_tx);
878 CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
879 osd->od_svname, buf, oid, rc);
882 /* Remove the OI mapping after the destroy to handle the race with
883 * OI scrub that may insert missed OI mapping during the interval. */
884 rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
885 if (unlikely(rc == -ENOENT))
888 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
889 osd->od_svname, buf, rc);
894 /* not needed in the cache anymore */
895 set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
897 obj->oo_destroyed = 1;
898 up_write(&obj->oo_guard);
902 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
904 struct osd_object *obj = osd_obj(l);
905 const struct lu_fid *fid = lu_object_fid(l);
908 if (likely(!fid_is_acct(fid))) {
909 osd_object_sa_fini(obj);
910 if (obj->oo_sa_xattr) {
911 nvlist_free(obj->oo_sa_xattr);
912 obj->oo_sa_xattr = NULL;
914 osd_dnode_rele(obj->oo_dn);
915 list_del(&obj->oo_sa_linkage);
922 * Concurrency: ->loo_object_release() is called under site spin-lock.
924 static void osd_object_release(const struct lu_env *env,
930 * Concurrency: shouldn't matter.
932 static int osd_object_print(const struct lu_env *env, void *cookie,
933 lu_printer_t p, const struct lu_object *l)
935 struct osd_object *o = osd_obj(l);
937 return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
940 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
943 struct osd_object *obj = osd_dt_obj(dt);
945 LASSERT(osd_invariant(obj));
947 down_read_nested(&obj->oo_sem, role);
950 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
953 struct osd_object *obj = osd_dt_obj(dt);
955 LASSERT(osd_invariant(obj));
957 down_write_nested(&obj->oo_sem, role);
960 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
962 struct osd_object *obj = osd_dt_obj(dt);
964 LASSERT(osd_invariant(obj));
965 up_read(&obj->oo_sem);
968 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
970 struct osd_object *obj = osd_dt_obj(dt);
972 LASSERT(osd_invariant(obj));
973 up_write(&obj->oo_sem);
976 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
978 struct osd_object *obj = osd_dt_obj(dt);
981 LASSERT(osd_invariant(obj));
983 if (down_write_trylock(&obj->oo_sem)) {
985 up_write(&obj->oo_sem);
990 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
991 struct lu_attr *attr)
993 struct osd_object *obj = osd_dt_obj(dt);
994 struct osd_device *osd = osd_obj2dev(obj);
999 down_read(&obj->oo_guard);
1001 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1002 GOTO(out, rc = -ENOENT);
1004 if (unlikely(fid_is_acct(lu_object_fid(&dt->do_lu))))
1007 LASSERT(osd_invariant(obj));
1008 LASSERT(obj->oo_dn);
1010 read_lock(&obj->oo_attr_lock);
1011 *attr = obj->oo_attr;
1012 if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
1013 attr->la_valid |= LA_FLAGS;
1014 attr->la_flags |= LUSTRE_ORPHAN_FL;
1016 read_unlock(&obj->oo_attr_lock);
1017 if (attr->la_valid & LA_FLAGS && attr->la_flags & LUSTRE_ORPHAN_FL)
1018 CDEBUG(D_INFO, "%s: set orphan flag on "DFID" (%llx/%x)\n",
1019 osd_obj2dev(obj)->od_svname,
1020 PFID(lu_object_fid(&dt->do_lu)),
1021 attr->la_valid, obj->oo_lma_flags);
1023 /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
1024 * from within sa_object_size() can block on a mutex, so
1025 * we can't call sa_object_size() holding rwlock */
1026 sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
1027 /* we do not control size of indices, so always calculate
1028 * it from number of blocks reported by DMU */
1029 if (S_ISDIR(attr->la_mode)) {
1030 attr->la_size = 512 * blocks;
1031 rc = -zap_count(osd->od_os, obj->oo_dn->dn_object,
1032 &attr->la_dirent_count);
1034 /* Block size may be not set; suggest maximal I/O transfers. */
1036 blksize = osd_spa_maxblocksize(
1037 dmu_objset_spa(osd_obj2dev(obj)->od_os));
1039 attr->la_blksize = blksize;
1040 attr->la_blocks = blocks;
1041 attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
1044 up_read(&obj->oo_guard);
1048 /* Simple wrapper on top of qsd API which implement quota transfer for osd
1049 * setattr needs. As a reminder, only the root user can change ownership of
1050 * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
1051 static inline int qsd_transfer(const struct lu_env *env,
1052 struct qsd_instance *qsd,
1053 struct lquota_trans *trans, int qtype,
1054 __u64 orig_id, __u64 new_id, __u64 bspace,
1055 struct lquota_id_info *qi, bool ignore_edquot)
1059 if (unlikely(qsd == NULL))
1062 LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
1063 qi->lqi_type = qtype;
1065 /* inode accounting */
1066 qi->lqi_is_blk = false;
1068 /* one more inode for the new owner ... */
1069 qi->lqi_id.qid_uid = new_id;
1071 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1072 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1077 /* and one less inode for the current id */
1078 qi->lqi_id.qid_uid = orig_id;;
1080 /* can't get EDQUOT when reducing usage */
1081 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1082 if (rc == -EINPROGRESS)
1087 /* block accounting */
1088 qi->lqi_is_blk = true;
1090 /* more blocks for the new owner ... */
1091 qi->lqi_id.qid_uid = new_id;
1092 qi->lqi_space = bspace;
1093 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1094 if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1099 /* and finally less blocks for the current owner */
1100 qi->lqi_id.qid_uid = orig_id;
1101 qi->lqi_space = -bspace;
1102 rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1103 /* can't get EDQUOT when reducing usage */
1104 if (rc == -EINPROGRESS)
1109 static int osd_declare_attr_set(const struct lu_env *env,
1110 struct dt_object *dt,
1111 const struct lu_attr *attr,
1112 struct thandle *handle)
1114 struct osd_thread_info *info = osd_oti_get(env);
1115 struct osd_object *obj = osd_dt_obj(dt);
1116 struct osd_device *osd = osd_obj2dev(obj);
1118 struct osd_thandle *oh;
1126 LASSERT(handle != NULL);
1127 LASSERT(osd_invariant(obj));
1129 oh = container_of(handle, struct osd_thandle, ot_super);
1131 down_read(&obj->oo_guard);
1132 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1135 LASSERT(obj->oo_sa_hdl != NULL);
1136 LASSERT(oh->ot_tx != NULL);
1137 /* regular attributes are part of the bonus buffer */
1138 /* let's check whether this object is already part of
1141 for (txh = list_head(&oh->ot_tx->tx_holds); txh;
1142 txh = list_next(&oh->ot_tx->tx_holds, txh)) {
1143 if (txh->txh_dnode == NULL)
1145 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
1147 /* this object is part of the transaction already
1148 * we don't need to declare bonus again */
1153 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
1154 if (oh->ot_tx->tx_err != 0)
1155 GOTO(out, rc = -oh->ot_tx->tx_err);
1157 if (attr && attr->la_valid & LA_FLAGS) {
1158 /* LMA is usually a part of bonus, no need to declare
1162 if (attr && (attr->la_valid & (LA_UID | LA_GID | LA_PROJID))) {
1163 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
1164 bspace = toqb(bspace * 512);
1167 "%s: enforce quota on UID %u, GID %u, the quota space is %lld (%u)\n",
1169 attr->la_uid, attr->la_gid, bspace, blksize);
1172 if (attr && attr->la_valid & LA_UID) {
1173 /* quota enforcement for user */
1174 if (attr->la_uid != obj->oo_attr.la_uid) {
1175 rc = qsd_transfer(env, osd_def_qsd(osd),
1176 &oh->ot_quota_trans, USRQUOTA,
1177 obj->oo_attr.la_uid, attr->la_uid,
1178 bspace, &info->oti_qi, true);
1183 if (attr && attr->la_valid & LA_GID) {
1184 /* quota enforcement for group */
1185 if (attr->la_gid != obj->oo_attr.la_gid) {
1186 rc = qsd_transfer(env, osd_def_qsd(osd),
1187 &oh->ot_quota_trans, GRPQUOTA,
1188 obj->oo_attr.la_gid, attr->la_gid,
1189 bspace, &info->oti_qi,
1191 LUSTRE_SET_SYNC_FL));
1196 #ifdef ZFS_PROJINHERIT
1197 if (attr && attr->la_valid & LA_PROJID) {
1198 /* quota enforcement for project */
1199 if (attr->la_projid != obj->oo_attr.la_projid) {
1200 if (!osd->od_projectused_dn)
1201 GOTO(out, rc = -EOPNOTSUPP);
1203 /* Usually, if project quota is upgradable for the
1204 * device, then the upgrade will be done before or when
1205 * mount the device. So when we come here, this project
1206 * should have project ID attribute already (that is
1207 * zero by default). Otherwise, there was something
1208 * wrong during the former upgrade, let's return failure
1211 * Please note that, different from other attributes,
1212 * you can NOT simply set the project ID attribute under
1213 * such case, because adding (NOT change) project ID
1214 * attribute needs to change the object's attribute
1215 * layout to match zfs backend quota accounting
1217 if (unlikely(!obj->oo_with_projid))
1218 GOTO(out, rc = -ENXIO);
1220 rc = qsd_transfer(env, osd_def_qsd(osd),
1221 &oh->ot_quota_trans, PRJQUOTA,
1222 obj->oo_attr.la_projid,
1223 attr->la_projid, bspace,
1224 &info->oti_qi, true);
1231 up_read(&obj->oo_guard);
1236 * Set the attributes of an object
1238 * The transaction passed to this routine must have
1239 * dmu_tx_hold_bonus(tx, oid) called and then assigned
1240 * to a transaction group.
1242 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
1243 const struct lu_attr *la, struct thandle *handle)
1245 struct osd_thread_info *info = osd_oti_get(env);
1246 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1247 struct osd_object *obj = osd_dt_obj(dt);
1248 struct osd_device *osd = osd_obj2dev(obj);
1249 struct osd_thandle *oh;
1250 struct osa_attr *osa = &info->oti_osa;
1251 __u64 valid = la->la_valid;
1257 down_read(&obj->oo_guard);
1258 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1259 GOTO(out, rc = -ENOENT);
1261 LASSERT(handle != NULL);
1262 LASSERT(osd_invariant(obj));
1263 LASSERT(obj->oo_sa_hdl);
1265 oh = container_of(handle, struct osd_thandle, ot_super);
1266 /* Assert that the transaction has been assigned to a
1267 transaction group. */
1268 LASSERT(oh->ot_tx->tx_txg != 0);
1270 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) && !osd->od_is_ost) {
1271 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1272 char *buf = info->oti_str;
1273 dnode_t *zdn = NULL;
1276 zapid = osd_get_name_n_idx(env, osd, lu_object_fid(&dt->do_lu),
1277 buf, sizeof(info->oti_str), &zdn);
1278 rc = osd_zap_lookup(osd, zapid, zdn, buf, 8,
1279 sizeof(*zde) / 8, zde);
1281 zde->zde_dnode -= 1;
1282 rc = -zap_update(osd->od_os, zapid, buf, 8,
1283 sizeof(*zde) / 8, zde, oh->ot_tx);
1290 /* Only allow set size for regular file */
1291 if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
1292 valid &= ~(LA_SIZE | LA_BLOCKS);
1294 if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
1297 if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
1300 if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
1306 if (valid & LA_FLAGS) {
1307 struct lustre_mdt_attrs *lma;
1311 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
1312 LASSERT(!obj->oo_pfid_in_lma);
1313 BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*lma));
1314 lma = (struct lustre_mdt_attrs *)&info->oti_buf;
1316 buf.lb_len = sizeof(info->oti_buf);
1318 /* Please do NOT call osd_xattr_get() directly, that
1319 * will cause recursive down_read() on oo_guard. */
1320 rc = osd_xattr_get_internal(env, obj, &buf,
1321 XATTR_NAME_LMA, &size);
1322 if (!rc && unlikely(size < sizeof(*lma))) {
1326 le32_to_cpu(lma->lma_incompat);
1327 lma->lma_incompat |=
1328 lustre_to_lma_flags(la->la_flags);
1330 cpu_to_le32(lma->lma_incompat);
1332 buf.lb_len = sizeof(*lma);
1333 rc = osd_xattr_set_internal(env, obj, &buf,
1339 CWARN("%s: failed to set LMA flags: rc = %d\n",
1340 osd->od_svname, rc);
1346 write_lock(&obj->oo_attr_lock);
1349 if (valid & LA_PROJID) {
1350 #ifdef ZFS_PROJINHERIT
1351 if (osd->od_projectused_dn) {
1352 LASSERT(obj->oo_with_projid);
1354 osa->projid = obj->oo_attr.la_projid = la->la_projid;
1355 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1359 valid &= ~LA_PROJID;
1362 if (valid & LA_ATIME) {
1363 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1364 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1367 if (valid & LA_MTIME) {
1368 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1369 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1372 if (valid & LA_CTIME) {
1373 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1374 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1377 if (valid & LA_MODE) {
1378 /* mode is stored along with type, so read it first */
1379 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1380 (la->la_mode & ~S_IFMT);
1381 osa->mode = obj->oo_attr.la_mode;
1382 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1385 if (valid & LA_SIZE) {
1386 osa->size = obj->oo_attr.la_size = la->la_size;
1387 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1390 if (valid & LA_NLINK) {
1391 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1392 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1395 if (valid & LA_RDEV) {
1396 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1397 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1400 if (valid & LA_FLAGS) {
1401 osa->flags = attrs_fs2zfs(la->la_flags);
1402 /* many flags are not supported by zfs, so ensure a good cached
1404 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1405 #ifdef ZFS_PROJINHERIT
1406 if (obj->oo_with_projid && osd->od_projectused_dn)
1407 osa->flags |= ZFS_PROJID;
1409 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1412 if (valid & LA_UID) {
1413 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1414 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1417 if (valid & LA_GID) {
1418 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1419 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1422 obj->oo_attr.la_valid |= valid;
1423 write_unlock(&obj->oo_attr_lock);
1425 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1426 rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1429 up_read(&obj->oo_guard);
1436 * XXX temporary solution.
1439 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1440 struct dt_object *parent, struct dt_object *child,
1445 ah->dah_parent = parent;
1446 ah->dah_mode = child_mode;
1448 if (parent != NULL && !dt_object_remote(parent)) {
1449 /* will help to find FID->ino at dt_insert("..") */
1450 struct osd_object *pobj = osd_dt_obj(parent);
1452 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1456 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
1457 struct lu_attr *attr,
1458 struct dt_allocation_hint *hint,
1459 struct dt_object_format *dof,
1460 struct thandle *handle)
1462 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1463 struct osd_object *obj = osd_dt_obj(dt);
1464 struct osd_device *osd = osd_obj2dev(obj);
1465 struct osd_thandle *oh;
1473 switch (dof->dof_type) {
1477 if (obj->oo_dt.do_body_ops == NULL)
1478 obj->oo_dt.do_body_ops = &osd_body_ops;
1484 LASSERT(handle != NULL);
1485 oh = container_of(handle, struct osd_thandle, ot_super);
1486 LASSERT(oh->ot_tx != NULL);
1488 /* this is the minimum set of EAs on every Lustre object */
1489 obj->oo_ea_in_bonus = OSD_BASE_EA_IN_BONUS;
1490 /* reserve 32 bytes for extra stuff like ACLs */
1491 dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1493 switch (dof->dof_type) {
1495 dt->do_index_ops = &osd_dir_ops;
1498 /* for zap create */
1499 dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1500 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1505 /* first, we'll create new object */
1506 dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1514 /* and we'll add it to some mapping */
1515 zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
1516 osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
1518 /* will help to find FID->ino mapping at dt_insert() */
1519 osd_idc_find_and_init(env, osd, obj);
1521 rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid,
1522 attr->la_projid, 1, oh, NULL, OSD_QID_INODE);
1527 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1528 struct osd_object *obj, sa_handle_t *sa_hdl, dmu_tx_t *tx,
1529 struct lu_attr *la, uint64_t parent,
1532 sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1533 struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1535 inode_timespec_t now;
1538 char *dxattr = NULL;
1544 gen = dmu_tx_get_txg(tx);
1546 ZFS_TIME_ENCODE(&now, osa->btime);
1548 osa->atime[0] = la->la_atime;
1549 osa->ctime[0] = la->la_ctime;
1550 osa->mtime[0] = la->la_mtime;
1551 osa->mode = la->la_mode;
1552 osa->uid = la->la_uid;
1553 osa->gid = la->la_gid;
1554 osa->rdev = la->la_rdev;
1555 osa->nlink = la->la_nlink;
1556 if (la->la_valid & LA_FLAGS)
1557 osa->flags = attrs_fs2zfs(la->la_flags);
1560 osa->size = la->la_size;
1561 #ifdef ZFS_PROJINHERIT
1562 if (osd->od_projectused_dn) {
1563 if (la->la_valid & LA_PROJID)
1564 osa->projid = la->la_projid;
1566 osa->projid = ZFS_DEFAULT_PROJID;
1567 osa->flags |= ZFS_PROJID;
1569 obj->oo_with_projid = 1;
1571 osa->flags &= ~ZFS_PROJID;
1576 * we need to create all SA below upon object create.
1578 * XXX The attribute order matters since the accounting callback relies
1579 * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1580 * look up the UID/GID/PROJID attributes. Moreover, the callback does
1581 * not seem to support the spill block.
1582 * We define attributes in the same order as SA_*_OFFSET in order to
1583 * work around the problem. See ORI-610.
1586 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1587 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1588 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1589 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1590 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1591 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1592 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1593 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1594 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1595 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1596 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, osa->btime, 16);
1597 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1598 #ifdef ZFS_PROJINHERIT
1599 if (osd->od_projectused_dn)
1600 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1603 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1604 LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1607 rc = -nvlist_size(xattr, &sa_size, NV_ENCODE_XDR);
1610 dxattr = osd_zio_buf_alloc(sa_size);
1613 rc = -nvlist_pack(xattr, &dxattr, &sa_size,
1614 NV_ENCODE_XDR, KM_SLEEP);
1617 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd),
1618 NULL, dxattr, sa_size);
1621 rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1623 osd_zio_buf_free(dxattr, sa_size);
1628 int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1629 uint64_t oid, dnode_t **dnp)
1634 /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1635 for (txh = list_tail(&tx->tx_holds); txh;
1636 txh = list_prev(&tx->tx_holds, txh)) {
1637 dnode_t *dn = txh->txh_dnode;
1642 if (dn->dn_object != oid)
1646 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1647 if (dn->dn_bonus == NULL)
1648 dbuf_create_bonus(dn);
1649 rw_exit(&dn->dn_struct_rwlock);
1653 LASSERT(dn->dn_handle);
1655 if (zfs_refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1656 zfs_refcount_add(&dn->dn_holds, osd_obj_tag);
1657 atomic_inc_32(&dn->dn_dbufs_count);
1661 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1665 if (unlikely(*dnp == NULL))
1666 rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
1671 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
1672 int osd_find_dnsize(struct osd_device *osd, int ea_in_bonus)
1676 if (osd->od_dnsize == ZFS_DNSIZE_AUTO) {
1677 dnsize = DNODE_MIN_SIZE;
1679 if (DN_BONUS_SIZE(dnsize) >= ea_in_bonus + 32)
1682 } while (dnsize < DNODE_MAX_SIZE);
1683 if (dnsize > DNODE_MAX_SIZE)
1684 dnsize = DNODE_MAX_SIZE;
1685 } else if (osd->od_dnsize == ZFS_DNSIZE_1K) {
1687 } else if (osd->od_dnsize == ZFS_DNSIZE_2K) {
1689 } else if (osd->od_dnsize == ZFS_DNSIZE_4K) {
1691 } else if (osd->od_dnsize == ZFS_DNSIZE_8K) {
1693 } else if (osd->od_dnsize == ZFS_DNSIZE_16K) {
1696 dnsize = DNODE_MIN_SIZE;
1703 * The transaction passed to this routine must have
1704 * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1705 * to a transaction group.
1707 int __osd_object_create(const struct lu_env *env, struct osd_device *osd,
1708 struct osd_object *obj, const struct lu_fid *fid,
1709 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1711 dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1715 /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1716 * would get an additional ditto copy */
1717 if (unlikely(S_ISREG(la->la_mode) &&
1718 fid_seq_is_local_file(fid_seq(fid))))
1719 type = DMU_OTN_UINT8_METADATA;
1721 /* Create a new DMU object using the default dnode size. */
1723 size = obj->oo_ea_in_bonus;
1725 size = OSD_BASE_EA_IN_BONUS;
1726 oid = osd_dmu_object_alloc(osd->od_os, type, 0,
1727 osd_find_dnsize(osd, size), tx);
1729 LASSERT(la->la_valid & LA_MODE);
1733 return osd_find_new_dnode(env, tx, oid, dnp);
1737 * The transaction passed to this routine must have
1738 * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1739 * to a transaction group.
1741 * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1742 * This is fine for directories today, because storing the FID in the dirent
1743 * will also require a FAT ZAP. If there is a new type of micro ZAP created
1744 * then we might need to re-evaluate the use of this flag and instead do
1745 * a conversion from the different internal ZAP hash formats being used. */
1746 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1747 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1748 unsigned dnsize, zap_flags_t flags)
1752 /* Assert that the transaction has been assigned to a
1753 transaction group. */
1754 LASSERT(tx->tx_txg != 0);
1757 oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1758 DMU_OT_DIRECTORY_CONTENTS,
1759 14, /* == ZFS fzap_default_blockshift */
1760 DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1766 return osd_find_new_dnode(env, tx, oid, dnp);
1769 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1770 struct lu_attr *la, struct osd_thandle *oh)
1772 struct osd_device *osd = osd_obj2dev(obj);
1776 /* Index file should be created as regular file in order not to confuse
1777 * ZPL which could interpret them as directory.
1778 * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1780 LASSERT(S_ISREG(la->la_mode));
1781 rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1782 osd_find_dnsize(osd, obj->oo_ea_in_bonus), ZAP_FLAG_UINT64_KEY);
1788 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1789 struct lu_attr *la, struct osd_thandle *oh)
1791 struct osd_device *osd = osd_obj2dev(obj);
1795 LASSERT(S_ISDIR(la->la_mode));
1796 rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1797 osd_find_dnsize(osd, obj->oo_ea_in_bonus), 0);
1803 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1804 struct lu_attr *la, struct osd_thandle *oh)
1806 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1807 struct osd_device *osd = osd_obj2dev(obj);
1811 LASSERT(S_ISREG(la->la_mode));
1812 rc = __osd_object_create(env, osd, obj, fid, &dn, oh->ot_tx, la);
1816 if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid))) {
1817 /* The minimum block size must be at least page size otherwise
1818 * it will break the assumption in tgt_thread_big_cache where
1819 * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1820 * RDMA due to subpage transfer size */
1821 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1822 PAGE_SIZE, 0, oh->ot_tx);
1824 CERROR("%s: can't change blocksize: %d\n",
1825 osd->od_svname, rc);
1828 } else if ((fid_is_llog(fid))) {
1829 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1830 LLOG_MIN_CHUNK_SIZE, 0, oh->ot_tx);
1832 CERROR("%s: can't change blocksize: %d\n",
1833 osd->od_svname, rc);
1841 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1842 struct lu_attr *la, struct osd_thandle *oh)
1847 LASSERT(S_ISLNK(la->la_mode));
1848 rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1849 lu_object_fid(&obj->oo_dt.do_lu),
1850 &dn, oh->ot_tx, la);
1856 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1857 struct lu_attr *la, struct osd_thandle *oh)
1862 if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1863 la->la_valid |= LA_RDEV;
1865 rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1866 lu_object_fid(&obj->oo_dt.do_lu),
1867 &dn, oh->ot_tx, la);
1873 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1874 struct osd_object *obj,
1876 struct osd_thandle *oh);
1878 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1880 osd_obj_type_f result;
1906 * Concurrency: @dt is write locked.
1908 static int osd_create(const struct lu_env *env, struct dt_object *dt,
1909 struct lu_attr *attr, struct dt_allocation_hint *hint,
1910 struct dt_object_format *dof, struct thandle *th)
1912 struct osd_thread_info *info = osd_oti_get(env);
1913 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1914 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1915 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1916 struct osd_object *obj = osd_dt_obj(dt);
1917 struct osd_device *osd = osd_obj2dev(obj);
1918 char *buf = info->oti_str;
1919 struct osd_thandle *oh;
1920 dnode_t *dn = NULL, *zdn = NULL;
1921 uint64_t zapid, parent = 0;
1927 LASSERT(!fid_is_acct(fid));
1929 /* concurrent create declarations should not see
1930 * the object inconsistent (db, attr, etc).
1931 * in regular cases acquisition should be cheap */
1932 down_write(&obj->oo_guard);
1934 if (unlikely(dt_object_exists(dt)))
1935 GOTO(out, rc = -EEXIST);
1937 LASSERT(osd_invariant(obj));
1938 LASSERT(dof != NULL);
1940 LASSERT(th != NULL);
1941 oh = container_of(th, struct osd_thandle, ot_super);
1943 LASSERT(obj->oo_dn == NULL);
1945 /* to follow ZFS on-disk format we need
1946 * to initialize parent dnode properly */
1947 if (hint != NULL && hint->dah_parent != NULL &&
1948 !dt_object_remote(hint->dah_parent))
1949 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1951 /* we may fix some attributes, better do not change the source */
1952 obj->oo_attr = *attr;
1953 obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1955 #ifdef ZFS_PROJINHERIT
1956 if (osd->od_projectused_dn) {
1957 if (!(obj->oo_attr.la_valid & LA_PROJID))
1958 obj->oo_attr.la_projid = ZFS_DEFAULT_PROJID;
1959 obj->oo_with_projid = 1;
1963 dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1971 zde->zde_dnode = dn->dn_object;
1972 zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1974 zapid = osd_get_name_n_idx(env, osd, fid, buf,
1975 sizeof(info->oti_str), &zdn);
1976 if (CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY) ||
1977 (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY)))
1980 if (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
1983 rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
1989 /* Now add in all of the "SA" attributes */
1990 rc = osd_sa_handle_get(obj);
1994 rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1998 /* initialize LMA */
1999 if (fid_is_idif(fid) || (fid_is_norm(fid) && osd->od_is_ost))
2000 compat |= LMAC_FID_ON_OST;
2001 lustre_lma_init(lma, fid, compat, 0);
2002 lustre_lma_swab(lma);
2003 rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
2004 (uchar_t *)lma, sizeof(*lma));
2008 /* configure new osd object */
2009 obj->oo_parent = parent != 0 ? parent : zapid;
2010 obj->oo_late_attr_set = 1;
2011 rc = __osd_sa_xattr_schedule_update(env, obj, oh);
2015 /* XXX: oo_lma_flags */
2016 obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
2017 if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
2018 /* no body operations for accounting objects */
2019 obj->oo_dt.do_body_ops = &osd_body_ops;
2021 osd_idc_find_and_init(env, osd, obj);
2024 if (unlikely(rc && dn)) {
2025 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
2029 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
2031 up_write(&obj->oo_guard);
2035 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
2038 osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2039 return osd_declare_attr_set(env, dt, NULL, th);
2043 * Concurrency: @dt is write locked.
2045 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
2046 struct thandle *handle)
2048 struct osd_object *obj = osd_dt_obj(dt);
2049 struct osd_thandle *oh;
2050 struct osd_device *osd = osd_obj2dev(obj);
2056 down_read(&obj->oo_guard);
2057 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2058 GOTO(out, rc = -ENOENT);
2060 LASSERT(osd_invariant(obj));
2061 LASSERT(obj->oo_sa_hdl != NULL);
2063 oh = container_of(handle, struct osd_thandle, ot_super);
2065 write_lock(&obj->oo_attr_lock);
2066 nlink = ++obj->oo_attr.la_nlink;
2067 write_unlock(&obj->oo_attr_lock);
2069 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2072 up_read(&obj->oo_guard);
2076 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
2077 struct thandle *handle)
2079 osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2080 return osd_declare_attr_set(env, dt, NULL, handle);
2084 * Concurrency: @dt is write locked.
2086 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
2087 struct thandle *handle)
2089 struct osd_object *obj = osd_dt_obj(dt);
2090 struct osd_thandle *oh;
2091 struct osd_device *osd = osd_obj2dev(obj);
2097 down_read(&obj->oo_guard);
2099 if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2100 GOTO(out, rc = -ENOENT);
2102 LASSERT(osd_invariant(obj));
2103 LASSERT(obj->oo_sa_hdl != NULL);
2105 oh = container_of(handle, struct osd_thandle, ot_super);
2106 LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2108 write_lock(&obj->oo_attr_lock);
2109 nlink = --obj->oo_attr.la_nlink;
2110 write_unlock(&obj->oo_attr_lock);
2112 rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2115 up_read(&obj->oo_guard);
2119 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
2120 __u64 start, __u64 end)
2122 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
2123 struct dmu_buf_impl *db = osd_dt_obj(dt)->oo_dn->dn_dbuf;
2127 if (osd->od_dt_dev.dd_rdonly)
2130 mutex_enter(&db->db_mtx);
2131 if (db->db_last_dirty)
2132 txg = db->db_last_dirty->dr_txg;
2133 mutex_exit(&db->db_mtx);
2136 /* the object is dirty or being synced */
2137 if (osd_object_sync_delay_us < 0)
2138 txg_wait_synced(dmu_objset_pool(osd->od_os), txg);
2140 udelay(osd_object_sync_delay_us);
2146 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
2151 static struct dt_object_operations osd_obj_ops = {
2152 .do_read_lock = osd_read_lock,
2153 .do_write_lock = osd_write_lock,
2154 .do_read_unlock = osd_read_unlock,
2155 .do_write_unlock = osd_write_unlock,
2156 .do_write_locked = osd_write_locked,
2157 .do_attr_get = osd_attr_get,
2158 .do_declare_attr_set = osd_declare_attr_set,
2159 .do_attr_set = osd_attr_set,
2160 .do_ah_init = osd_ah_init,
2161 .do_declare_create = osd_declare_create,
2162 .do_create = osd_create,
2163 .do_declare_destroy = osd_declare_destroy,
2164 .do_destroy = osd_destroy,
2165 .do_index_try = osd_index_try,
2166 .do_declare_ref_add = osd_declare_ref_add,
2167 .do_ref_add = osd_ref_add,
2168 .do_declare_ref_del = osd_declare_ref_del,
2169 .do_ref_del = osd_ref_del,
2170 .do_xattr_get = osd_xattr_get,
2171 .do_declare_xattr_set = osd_declare_xattr_set,
2172 .do_xattr_set = osd_xattr_set,
2173 .do_declare_xattr_del = osd_declare_xattr_del,
2174 .do_xattr_del = osd_xattr_del,
2175 .do_xattr_list = osd_xattr_list,
2176 .do_object_sync = osd_object_sync,
2177 .do_invalidate = osd_invalidate,
2180 static struct lu_object_operations osd_lu_obj_ops = {
2181 .loo_object_init = osd_object_init,
2182 .loo_object_delete = osd_object_delete,
2183 .loo_object_release = osd_object_release,
2184 .loo_object_free = osd_object_free,
2185 .loo_object_print = osd_object_print,
2186 .loo_object_invariant = osd_object_invariant,
2189 static int osd_otable_it_attr_get(const struct lu_env *env,
2190 struct dt_object *dt,
2191 struct lu_attr *attr)
2197 static struct dt_object_operations osd_obj_otable_it_ops = {
2198 .do_attr_get = osd_otable_it_attr_get,
2199 .do_index_try = osd_index_try,
2202 module_param(osd_object_sync_delay_us, int, 0644);
2203 MODULE_PARM_DESC(osd_object_sync_delay_us,
2204 "If zero or larger delay N usec instead of doing object sync");