Whamcloud - gitweb
6bf8185af44214a83dd448e0e890e3f29b2b425f
[fs/lustre-release.git] / lustre / osd-zfs / osd_object.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
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.
9  *
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).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  */
30 /*
31  * Copyright (c) 2012, 2013, Intel Corporation.
32  * Use is subject to license terms.
33  */
34 /*
35  * This file is part of Lustre, http://www.lustre.org/
36  * Lustre is a trademark of Sun Microsystems, Inc.
37  *
38  * lustre/osd-zfs/osd_object.c
39  *
40  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
41  * Author: Mike Pershin <tappro@whamcloud.com>
42  * Author: Johann Lombardi <johann@whamcloud.com>
43  */
44
45 #define DEBUG_SUBSYSTEM S_OSD
46
47 #include <lustre_ver.h>
48 #include <libcfs/libcfs.h>
49 #include <obd_support.h>
50 #include <lustre_net.h>
51 #include <obd.h>
52 #include <obd_class.h>
53 #include <lustre_disk.h>
54 #include <lustre_fid.h>
55
56 #include "osd_internal.h"
57
58 #include <sys/dnode.h>
59 #include <sys/dbuf.h>
60 #include <sys/spa.h>
61 #include <sys/stat.h>
62 #include <sys/zap.h>
63 #include <sys/spa_impl.h>
64 #include <sys/zfs_znode.h>
65 #include <sys/dmu_tx.h>
66 #include <sys/dmu_objset.h>
67 #include <sys/dsl_prop.h>
68 #include <sys/sa_impl.h>
69 #include <sys/txg.h>
70
71 char *osd_obj_tag = "osd_object";
72
73 static struct dt_object_operations osd_obj_ops;
74 static struct lu_object_operations osd_lu_obj_ops;
75 extern struct dt_body_operations osd_body_ops;
76 static struct dt_object_operations osd_obj_otable_it_ops;
77
78 extern struct kmem_cache *osd_object_kmem;
79
80 static void
81 osd_object_sa_fini(struct osd_object *obj)
82 {
83         if (obj->oo_sa_hdl) {
84                 sa_handle_destroy(obj->oo_sa_hdl);
85                 obj->oo_sa_hdl = NULL;
86         }
87 }
88
89 static int
90 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
91 {
92         int rc;
93
94         LASSERT(obj->oo_sa_hdl == NULL);
95         LASSERT(obj->oo_db != NULL);
96
97         rc = -sa_handle_get(o->od_os, obj->oo_db->db_object, obj,
98                             SA_HDL_PRIVATE, &obj->oo_sa_hdl);
99         if (rc)
100                 return rc;
101
102         /* Cache the xattr object id, valid for the life of the object */
103         rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
104         if (rc == -ENOENT) {
105                 obj->oo_xattr = ZFS_NO_OBJECT;
106                 rc = 0;
107         } else if (rc) {
108                 osd_object_sa_fini(obj);
109         }
110
111         return rc;
112 }
113
114 /*
115  * Add object to list of dirty objects in tx handle.
116  */
117 static void
118 osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
119 {
120         if (!list_empty(&obj->oo_sa_linkage))
121                 return;
122
123         down(&oh->ot_sa_lock);
124         write_lock(&obj->oo_attr_lock);
125         if (likely(list_empty(&obj->oo_sa_linkage)))
126                 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
127         write_unlock(&obj->oo_attr_lock);
128         up(&oh->ot_sa_lock);
129 }
130
131 /*
132  * Release spill block dbuf hold for all dirty SAs.
133  */
134 void osd_object_sa_dirty_rele(struct osd_thandle *oh)
135 {
136         struct osd_object *obj;
137
138         down(&oh->ot_sa_lock);
139         while (!list_empty(&oh->ot_sa_list)) {
140                 obj = list_entry(oh->ot_sa_list.next,
141                                      struct osd_object, oo_sa_linkage);
142                 sa_spill_rele(obj->oo_sa_hdl);
143                 write_lock(&obj->oo_attr_lock);
144                 list_del_init(&obj->oo_sa_linkage);
145                 write_unlock(&obj->oo_attr_lock);
146         }
147         up(&oh->ot_sa_lock);
148 }
149
150 /*
151  * Update the SA and add the object to the dirty list.
152  */
153 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
154                          void *buf, uint32_t buflen, struct osd_thandle *oh)
155 {
156         int rc;
157
158         LASSERT(obj->oo_sa_hdl != NULL);
159         LASSERT(oh->ot_tx != NULL);
160
161         rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
162         osd_object_sa_dirty_add(obj, oh);
163
164         return rc;
165 }
166
167 /*
168  * Bulk update the SA and add the object to the dirty list.
169  */
170 static int
171 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
172                           int count, struct osd_thandle *oh)
173 {
174         int rc;
175
176         LASSERT(obj->oo_sa_hdl != NULL);
177         LASSERT(oh->ot_tx != NULL);
178
179         rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
180         osd_object_sa_dirty_add(obj, oh);
181
182         return rc;
183 }
184
185 /*
186  * Retrieve the attributes of a DMU object
187  */
188 int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
189                           struct osd_object *obj, struct lu_attr *la)
190 {
191         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
192         sa_handle_t     *sa_hdl;
193         sa_bulk_attr_t  *bulk;
194         int              cnt = 0;
195         int              rc;
196         ENTRY;
197
198         LASSERT(obj->oo_db != NULL);
199
200         rc = -sa_handle_get(o->od_os, obj->oo_db->db_object, NULL,
201                             SA_HDL_PRIVATE, &sa_hdl);
202         if (rc)
203                 RETURN(rc);
204
205         OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 9);
206         if (bulk == NULL)
207                 GOTO(out_sa, rc = -ENOMEM);
208
209         la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
210                         LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
211
212         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
213         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
214         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
215         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
216         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
217         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
218         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
219         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
220         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
221
222         rc = -sa_bulk_lookup(sa_hdl, bulk, cnt);
223         if (rc)
224                 GOTO(out_bulk, rc);
225
226         la->la_atime = osa->atime[0];
227         la->la_mtime = osa->mtime[0];
228         la->la_ctime = osa->ctime[0];
229         la->la_mode = osa->mode;
230         la->la_uid = osa->uid;
231         la->la_gid = osa->gid;
232         la->la_nlink = osa->nlink;
233         la->la_flags = attrs_zfs2fs(osa->flags);
234         la->la_size = osa->size;
235
236         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
237                 rc = -sa_lookup(sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
238                 if (rc)
239                         GOTO(out_bulk, rc);
240                 la->la_rdev = osa->rdev;
241                 la->la_valid |= LA_RDEV;
242         }
243 out_bulk:
244         OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 9);
245 out_sa:
246         sa_handle_destroy(sa_hdl);
247
248         RETURN(rc);
249 }
250
251 int __osd_obj2dbuf(const struct lu_env *env, objset_t *os,
252                    uint64_t oid, dmu_buf_t **dbp)
253 {
254         dmu_object_info_t *doi = &osd_oti_get(env)->oti_doi;
255         int rc;
256
257         rc = -sa_buf_hold(os, oid, osd_obj_tag, dbp);
258         if (rc)
259                 return rc;
260
261         dmu_object_info_from_db(*dbp, doi);
262         if (unlikely (oid != DMU_USERUSED_OBJECT &&
263             oid != DMU_GROUPUSED_OBJECT && doi->doi_bonus_type != DMU_OT_SA)) {
264                 sa_buf_rele(*dbp, osd_obj_tag);
265                 *dbp = NULL;
266                 return -EINVAL;
267         }
268
269         LASSERT(*dbp);
270         LASSERT((*dbp)->db_object == oid);
271         LASSERT((*dbp)->db_offset == -1);
272         LASSERT((*dbp)->db_data != NULL);
273
274         return 0;
275 }
276
277 /*
278  * Concurrency: no concurrent access is possible that early in object
279  * life-cycle.
280  */
281 struct lu_object *osd_object_alloc(const struct lu_env *env,
282                                    const struct lu_object_header *hdr,
283                                    struct lu_device *d)
284 {
285         struct osd_object *mo;
286
287         OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
288         if (mo != NULL) {
289                 struct lu_object *l;
290
291                 l = &mo->oo_dt.do_lu;
292                 dt_object_init(&mo->oo_dt, NULL, d);
293                 mo->oo_dt.do_ops = &osd_obj_ops;
294                 l->lo_ops = &osd_lu_obj_ops;
295                 INIT_LIST_HEAD(&mo->oo_sa_linkage);
296                 init_rwsem(&mo->oo_sem);
297                 sema_init(&mo->oo_guard, 1);
298                 rwlock_init(&mo->oo_attr_lock);
299                 return l;
300         } else {
301                 return NULL;
302         }
303 }
304
305 /*
306  * Concurrency: shouldn't matter.
307  */
308 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
309 {
310         struct osd_device       *osd = osd_obj2dev(obj);
311         const struct lu_fid     *fid  = lu_object_fid(&obj->oo_dt.do_lu);
312         int                      rc = 0;
313         ENTRY;
314
315         if (obj->oo_db == NULL)
316                 RETURN(0);
317
318         /* object exist */
319
320         rc = osd_object_sa_init(obj, osd);
321         if (rc)
322                 RETURN(rc);
323
324         /* cache attrs in object */
325         rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
326         if (rc)
327                 RETURN(rc);
328
329         if (likely(!fid_is_acct(fid)))
330                 /* no body operations for accounting objects */
331                 obj->oo_dt.do_body_ops = &osd_body_ops;
332
333         /*
334          * initialize object before marking it existing
335          */
336         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
337
338         smp_mb();
339         obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
340
341         RETURN(0);
342 }
343
344 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
345 {
346         struct osd_thread_info  *info = osd_oti_get(env);
347         struct lu_buf           buf;
348         int                     rc;
349         struct lustre_mdt_attrs *lma;
350         ENTRY;
351
352         CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
353         lma = (struct lustre_mdt_attrs *)info->oti_buf;
354         buf.lb_buf = lma;
355         buf.lb_len = sizeof(info->oti_buf);
356
357         rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA, BYPASS_CAPA);
358         if (rc > 0) {
359                 rc = 0;
360                 lustre_lma_swab(lma);
361                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
362                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
363                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
364                               "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
365                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
366                               PFID(lu_object_fid(&obj->oo_dt.do_lu)));
367                         rc = -EOPNOTSUPP;
368                 }
369         } else if (rc == -ENODATA) {
370                 /* haven't initialize LMA xattr */
371                 rc = 0;
372         }
373
374         RETURN(rc);
375 }
376
377 /*
378  * Concurrency: no concurrent access is possible that early in object
379  * life-cycle.
380  */
381 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
382                            const struct lu_object_conf *conf)
383 {
384         struct osd_object       *obj = osd_obj(l);
385         struct osd_device       *osd = osd_obj2dev(obj);
386         uint64_t                 oid;
387         int                      rc;
388         ENTRY;
389
390         LASSERT(osd_invariant(obj));
391
392         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
393                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
394                 l->lo_header->loh_attr |= LOHA_EXISTS;
395                 RETURN(0);
396         }
397
398         rc = osd_fid_lookup(env, osd, lu_object_fid(l), &oid);
399         if (rc == 0) {
400                 LASSERT(obj->oo_db == NULL);
401                 rc = __osd_obj2dbuf(env, osd->od_os, oid, &obj->oo_db);
402                 if (rc != 0) {
403                         CERROR("%s: lookup "DFID"/"LPX64" failed: rc = %d\n",
404                                osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
405                         GOTO(out, rc);
406                 }
407                 LASSERT(obj->oo_db);
408                 rc = osd_object_init0(env, obj);
409                 if (rc != 0)
410                         GOTO(out, rc);
411
412                 rc = osd_check_lma(env, obj);
413                 if (rc != 0)
414                         GOTO(out, rc);
415         } else if (rc == -ENOENT) {
416                 rc = 0;
417         }
418         LASSERT(osd_invariant(obj));
419 out:
420         RETURN(rc);
421 }
422
423 /*
424  * Concurrency: no concurrent access is possible that late in object
425  * life-cycle.
426  */
427 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
428 {
429         struct osd_object *obj = osd_obj(l);
430
431         LASSERT(osd_invariant(obj));
432
433         dt_object_fini(&obj->oo_dt);
434         OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
435 }
436
437 static void __osd_declare_object_destroy(const struct lu_env *env,
438                                          struct osd_object *obj,
439                                          struct osd_thandle *oh)
440 {
441         struct osd_device       *osd = osd_obj2dev(obj);
442         dmu_buf_t               *db = obj->oo_db;
443         zap_attribute_t         *za = &osd_oti_get(env)->oti_za;
444         uint64_t                 oid = db->db_object, xid;
445         dmu_tx_t                *tx = oh->ot_tx;
446         zap_cursor_t            *zc;
447         int                      rc = 0;
448
449         dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
450
451         /* zap holding xattrs */
452         if (obj->oo_xattr != ZFS_NO_OBJECT) {
453                 oid = obj->oo_xattr;
454
455                 dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
456
457                 rc = osd_zap_cursor_init(&zc, osd->od_os, oid, 0);
458                 if (rc)
459                         goto out;
460
461                 while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
462                         BUG_ON(za->za_integer_length != sizeof(uint64_t));
463                         BUG_ON(za->za_num_integers != 1);
464
465                         rc = -zap_lookup(osd->od_os, obj->oo_xattr, za->za_name,
466                                          sizeof(uint64_t), 1, &xid);
467                         if (rc) {
468                                 CERROR("%s: xattr lookup failed: rc = %d\n",
469                                        osd->od_svname, rc);
470                                 goto out_err;
471                         }
472                         dmu_tx_hold_free(tx, xid, 0, DMU_OBJECT_END);
473
474                         zap_cursor_advance(zc);
475                 }
476                 if (rc == -ENOENT)
477                         rc = 0;
478 out_err:
479                 osd_zap_cursor_fini(zc);
480         }
481 out:
482         if (rc && tx->tx_err == 0)
483                 tx->tx_err = -rc;
484 }
485
486 static int osd_declare_object_destroy(const struct lu_env *env,
487                                       struct dt_object *dt,
488                                       struct thandle *th)
489 {
490         char                    *buf = osd_oti_get(env)->oti_str;
491         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
492         struct osd_object       *obj = osd_dt_obj(dt);
493         struct osd_device       *osd = osd_obj2dev(obj);
494         struct osd_thandle      *oh;
495         uint64_t                 zapid;
496         int                      rc;
497         ENTRY;
498
499         LASSERT(th != NULL);
500         LASSERT(dt_object_exists(dt));
501
502         oh = container_of0(th, struct osd_thandle, ot_super);
503         LASSERT(oh->ot_tx != NULL);
504
505         /* declare that we'll destroy the object */
506         __osd_declare_object_destroy(env, obj, oh);
507
508         /* declare that we'll remove object from fid-dnode mapping */
509         zapid = osd_get_name_n_idx(env, osd, fid, buf);
510         dmu_tx_hold_bonus(oh->ot_tx, zapid);
511         dmu_tx_hold_zap(oh->ot_tx, zapid, 0, buf);
512
513         /* declare that we'll remove object from inode accounting ZAPs */
514         dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
515         dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, 0, buf);
516         dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
517         dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, 0, buf);
518
519         /* one less inode */
520         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
521                                obj->oo_attr.la_gid, -1, oh, false, NULL, false);
522         if (rc)
523                 RETURN(rc);
524
525         /* data to be truncated */
526         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
527                                obj->oo_attr.la_gid, 0, oh, true, NULL, false);
528         RETURN(rc);
529 }
530
531 /*
532  * Delete a DMU object
533  *
534  * The transaction passed to this routine must have
535  * dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END) called
536  * and then assigned to a transaction group.
537  *
538  * This will release db and set it to NULL to prevent further dbuf releases.
539  */
540 static int __osd_object_destroy(const struct lu_env *env,
541                                 struct osd_object *obj,
542                                 dmu_tx_t *tx, void *tag)
543 {
544         struct osd_device       *osd = osd_obj2dev(obj);
545         uint64_t                 xid;
546         zap_attribute_t         *za = &osd_oti_get(env)->oti_za;
547         zap_cursor_t            *zc;
548         int                      rc;
549
550         /* Assert that the transaction has been assigned to a
551            transaction group. */
552         LASSERT(tx->tx_txg != 0);
553
554         /* zap holding xattrs */
555         if (obj->oo_xattr != ZFS_NO_OBJECT) {
556                 rc = osd_zap_cursor_init(&zc, osd->od_os, obj->oo_xattr, 0);
557                 if (rc)
558                         return rc;
559                 while ((rc = -zap_cursor_retrieve(zc, za)) == 0) {
560                         BUG_ON(za->za_integer_length != sizeof(uint64_t));
561                         BUG_ON(za->za_num_integers != 1);
562
563                         rc = -zap_lookup(osd->od_os, obj->oo_xattr, za->za_name,
564                                          sizeof(uint64_t), 1, &xid);
565                         if (rc) {
566                                 CERROR("%s: lookup xattr %s failed: rc = %d\n",
567                                        osd->od_svname, za->za_name, rc);
568                                 continue;
569                         }
570                         rc = -dmu_object_free(osd->od_os, xid, tx);
571                         if (rc)
572                                 CERROR("%s: fetch xattr %s failed: rc = %d\n",
573                                        osd->od_svname, za->za_name, rc);
574                         zap_cursor_advance(zc);
575                 }
576                 osd_zap_cursor_fini(zc);
577
578                 rc = -dmu_object_free(osd->od_os, obj->oo_xattr, tx);
579                 if (rc)
580                         CERROR("%s: freeing xattr failed: rc = %d\n",
581                                osd->od_svname, rc);
582         }
583
584         return -dmu_object_free(osd->od_os, obj->oo_db->db_object, tx);
585 }
586
587 static int osd_object_destroy(const struct lu_env *env,
588                               struct dt_object *dt,
589                               struct thandle *th)
590 {
591         char                    *buf = osd_oti_get(env)->oti_str;
592         struct osd_object       *obj = osd_dt_obj(dt);
593         struct osd_device       *osd = osd_obj2dev(obj);
594         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
595         struct osd_thandle      *oh;
596         int                      rc;
597         uint64_t                 zapid;
598         ENTRY;
599
600         LASSERT(obj->oo_db != NULL);
601         LASSERT(dt_object_exists(dt));
602         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
603
604         oh = container_of0(th, struct osd_thandle, ot_super);
605         LASSERT(oh != NULL);
606         LASSERT(oh->ot_tx != NULL);
607
608         zapid = osd_get_name_n_idx(env, osd, fid, buf);
609
610         /* remove obj ref from index dir (it depends) */
611         rc = -zap_remove(osd->od_os, zapid, buf, oh->ot_tx);
612         if (rc) {
613                 CERROR("%s: zap_remove() failed: rc = %d\n",
614                        osd->od_svname, rc);
615                 GOTO(out, rc);
616         }
617
618         /* Remove object from inode accounting. It is not fatal for the destroy
619          * operation if something goes wrong while updating accounting, but we
620          * still log an error message to notify the administrator */
621         rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
622                         obj->oo_attr.la_uid, -1, oh->ot_tx);
623         if (rc)
624                 CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
625                         " %d: rc = %d\n", osd->od_svname, PFID(fid),
626                         obj->oo_attr.la_uid, rc);
627         rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
628                                 obj->oo_attr.la_gid, -1, oh->ot_tx);
629         if (rc)
630                 CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
631                         " %d: rc = %d\n", osd->od_svname, PFID(fid),
632                         obj->oo_attr.la_gid, rc);
633
634         /* kill object */
635         rc = __osd_object_destroy(env, obj, oh->ot_tx, osd_obj_tag);
636         if (rc) {
637                 CERROR("%s: __osd_object_destroy() failed: rc = %d\n",
638                        osd->od_svname, rc);
639                 GOTO(out, rc);
640         }
641
642 out:
643         /* not needed in the cache anymore */
644         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
645
646         RETURN (0);
647 }
648
649 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
650 {
651         struct osd_object *obj = osd_obj(l);
652
653         if (obj->oo_db != NULL) {
654                 osd_object_sa_fini(obj);
655                 if (obj->oo_sa_xattr) {
656                         nvlist_free(obj->oo_sa_xattr);
657                         obj->oo_sa_xattr = NULL;
658                 }
659                 sa_buf_rele(obj->oo_db, osd_obj_tag);
660                 list_del(&obj->oo_sa_linkage);
661                 obj->oo_db = NULL;
662         }
663 }
664
665 /*
666  * Concurrency: ->loo_object_release() is called under site spin-lock.
667  */
668 static void osd_object_release(const struct lu_env *env,
669                                struct lu_object *l)
670 {
671 }
672
673 /*
674  * Concurrency: shouldn't matter.
675  */
676 static int osd_object_print(const struct lu_env *env, void *cookie,
677                             lu_printer_t p, const struct lu_object *l)
678 {
679         struct osd_object *o = osd_obj(l);
680
681         return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
682 }
683
684 static void osd_object_read_lock(const struct lu_env *env,
685                                  struct dt_object *dt, unsigned role)
686 {
687         struct osd_object *obj = osd_dt_obj(dt);
688
689         LASSERT(osd_invariant(obj));
690
691         down_read(&obj->oo_sem);
692 }
693
694 static void osd_object_write_lock(const struct lu_env *env,
695                                   struct dt_object *dt, unsigned role)
696 {
697         struct osd_object *obj = osd_dt_obj(dt);
698
699         LASSERT(osd_invariant(obj));
700
701         down_write(&obj->oo_sem);
702 }
703
704 static void osd_object_read_unlock(const struct lu_env *env,
705                                    struct dt_object *dt)
706 {
707         struct osd_object *obj = osd_dt_obj(dt);
708
709         LASSERT(osd_invariant(obj));
710         up_read(&obj->oo_sem);
711 }
712
713 static void osd_object_write_unlock(const struct lu_env *env,
714                                     struct dt_object *dt)
715 {
716         struct osd_object *obj = osd_dt_obj(dt);
717
718         LASSERT(osd_invariant(obj));
719         up_write(&obj->oo_sem);
720 }
721
722 static int osd_object_write_locked(const struct lu_env *env,
723                                    struct dt_object *dt)
724 {
725         struct osd_object *obj = osd_dt_obj(dt);
726         int rc = 1;
727
728         LASSERT(osd_invariant(obj));
729
730         if (down_write_trylock(&obj->oo_sem)) {
731                 rc = 0;
732                 up_write(&obj->oo_sem);
733         }
734         return rc;
735 }
736
737 static int osd_attr_get(const struct lu_env *env,
738                         struct dt_object *dt,
739                         struct lu_attr *attr,
740                         struct lustre_capa *capa)
741 {
742         struct osd_object       *obj = osd_dt_obj(dt);
743         uint64_t                 blocks;
744         uint32_t                 blksize;
745
746         LASSERT(dt_object_exists(dt));
747         LASSERT(osd_invariant(obj));
748         LASSERT(obj->oo_db);
749
750         read_lock(&obj->oo_attr_lock);
751         *attr = obj->oo_attr;
752         read_unlock(&obj->oo_attr_lock);
753
754         /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
755          * from within sa_object_size() can block on a mutex, so
756          * we can't call sa_object_size() holding rwlock */
757         sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
758         /* we do not control size of indices, so always calculate
759          * it from number of blocks reported by DMU */
760         if (S_ISDIR(attr->la_mode))
761                 attr->la_size = 512 * blocks;
762         /* Block size may be not set; suggest maximal I/O transfers. */
763         if (blksize == 0)
764                 blksize = 1ULL << SPA_MAXBLOCKSHIFT;
765
766         attr->la_blksize = blksize;
767         attr->la_blocks = blocks;
768         attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
769
770         return 0;
771 }
772
773 /* Simple wrapper on top of qsd API which implement quota transfer for osd
774  * setattr needs. As a reminder, only the root user can change ownership of
775  * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
776 static inline int qsd_transfer(const struct lu_env *env,
777                                struct qsd_instance *qsd,
778                                struct lquota_trans *trans, int qtype,
779                                __u64 orig_id, __u64 new_id, __u64 bspace,
780                                struct lquota_id_info *qi)
781 {
782         int     rc;
783
784         if (unlikely(qsd == NULL))
785                 return 0;
786
787         LASSERT(qtype >= 0 && qtype < MAXQUOTAS);
788         qi->lqi_type = qtype;
789
790         /* inode accounting */
791         qi->lqi_is_blk = false;
792
793         /* one more inode for the new owner ... */
794         qi->lqi_id.qid_uid = new_id;
795         qi->lqi_space      = 1;
796         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
797         if (rc == -EDQUOT || rc == -EINPROGRESS)
798                 rc = 0;
799         if (rc)
800                 return rc;
801
802         /* and one less inode for the current id */
803         qi->lqi_id.qid_uid = orig_id;;
804         qi->lqi_space      = -1;
805         /* can't get EDQUOT when reducing usage */
806         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
807         if (rc == -EINPROGRESS)
808                 rc = 0;
809         if (rc)
810                 return rc;
811
812         /* block accounting */
813         qi->lqi_is_blk = true;
814
815         /* more blocks for the new owner ... */
816         qi->lqi_id.qid_uid = new_id;
817         qi->lqi_space      = bspace;
818         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
819         if (rc == -EDQUOT || rc == -EINPROGRESS)
820                 rc = 0;
821         if (rc)
822                 return rc;
823
824         /* and finally less blocks for the current owner */
825         qi->lqi_id.qid_uid = orig_id;
826         qi->lqi_space      = -bspace;
827         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
828         /* can't get EDQUOT when reducing usage */
829         if (rc == -EINPROGRESS)
830                 rc = 0;
831         return rc;
832 }
833
834 static int osd_declare_attr_set(const struct lu_env *env,
835                                 struct dt_object *dt,
836                                 const struct lu_attr *attr,
837                                 struct thandle *handle)
838 {
839         struct osd_thread_info  *info = osd_oti_get(env);
840         char                    *buf = osd_oti_get(env)->oti_str;
841         struct osd_object       *obj = osd_dt_obj(dt);
842         struct osd_device       *osd = osd_obj2dev(obj);
843         struct osd_thandle      *oh;
844         uint64_t                 bspace;
845         uint32_t                 blksize;
846         int                      rc;
847         ENTRY;
848
849         if (!dt_object_exists(dt)) {
850                 /* XXX: sanity check that object creation is declared */
851                 RETURN(0);
852         }
853
854         LASSERT(handle != NULL);
855         LASSERT(osd_invariant(obj));
856
857         oh = container_of0(handle, struct osd_thandle, ot_super);
858
859         LASSERT(obj->oo_sa_hdl != NULL);
860         LASSERT(oh->ot_tx != NULL);
861         dmu_tx_hold_sa(oh->ot_tx, obj->oo_sa_hdl, 0);
862         if (oh->ot_tx->tx_err != 0)
863                 RETURN(-oh->ot_tx->tx_err);
864
865         sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
866         bspace = toqb(bspace * blksize);
867
868         if (attr && attr->la_valid & LA_UID) {
869                 /* account for user inode tracking ZAP update */
870                 dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
871                 dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
872
873                 /* quota enforcement for user */
874                 if (attr->la_uid != obj->oo_attr.la_uid) {
875                         rc = qsd_transfer(env, osd->od_quota_slave,
876                                           &oh->ot_quota_trans, USRQUOTA,
877                                           obj->oo_attr.la_uid, attr->la_uid,
878                                           bspace, &info->oti_qi);
879                         if (rc)
880                                 RETURN(rc);
881                 }
882         }
883         if (attr && attr->la_valid & LA_GID) {
884                 /* account for user inode tracking ZAP update */
885                 dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
886                 dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
887
888                 /* quota enforcement for group */
889                 if (attr->la_gid != obj->oo_attr.la_gid) {
890                         rc = qsd_transfer(env, osd->od_quota_slave,
891                                           &oh->ot_quota_trans, GRPQUOTA,
892                                           obj->oo_attr.la_gid, attr->la_gid,
893                                           bspace, &info->oti_qi);
894                         if (rc)
895                                 RETURN(rc);
896                 }
897         }
898
899         RETURN(0);
900 }
901
902 /*
903  * Set the attributes of an object
904  *
905  * The transaction passed to this routine must have
906  * dmu_tx_hold_bonus(tx, oid) called and then assigned
907  * to a transaction group.
908  */
909 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
910                         const struct lu_attr *la, struct thandle *handle,
911                         struct lustre_capa *capa)
912 {
913         struct osd_object       *obj = osd_dt_obj(dt);
914         struct osd_device       *osd = osd_obj2dev(obj);
915         struct osd_thandle      *oh;
916         struct osa_attr         *osa = &osd_oti_get(env)->oti_osa;
917         sa_bulk_attr_t          *bulk;
918         __u64                    valid = la->la_valid;
919         int                      cnt;
920         int                      rc = 0;
921
922         ENTRY;
923         LASSERT(handle != NULL);
924         LASSERT(dt_object_exists(dt));
925         LASSERT(osd_invariant(obj));
926         LASSERT(obj->oo_sa_hdl);
927
928         oh = container_of0(handle, struct osd_thandle, ot_super);
929         /* Assert that the transaction has been assigned to a
930            transaction group. */
931         LASSERT(oh->ot_tx->tx_txg != 0);
932
933         /* Only allow set size for regular file */
934         if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
935                 valid &= ~(LA_SIZE | LA_BLOCKS);
936
937         if (valid == 0)
938                 RETURN(0);
939
940         OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 10);
941         if (bulk == NULL)
942                 RETURN(-ENOMEM);
943
944         /* do both accounting updates outside oo_attr_lock below */
945         if ((valid & LA_UID) && (la->la_uid != obj->oo_attr.la_uid)) {
946                 /* Update user accounting. Failure isn't fatal, but we still
947                  * log an error message */
948                 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
949                                         la->la_uid, 1, oh->ot_tx);
950                 if (rc)
951                         CERROR("%s: failed to update accounting ZAP for user "
952                                 "%d (%d)\n", osd->od_svname, la->la_uid, rc);
953                 rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
954                                         obj->oo_attr.la_uid, -1, oh->ot_tx);
955                 if (rc)
956                         CERROR("%s: failed to update accounting ZAP for user "
957                                 "%d (%d)\n", osd->od_svname,
958                                 obj->oo_attr.la_uid, rc);
959         }
960         if ((valid & LA_GID) && (la->la_gid != obj->oo_attr.la_gid)) {
961                 /* Update group accounting. Failure isn't fatal, but we still
962                  * log an error message */
963                 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
964                                         la->la_gid, 1, oh->ot_tx);
965                 if (rc)
966                         CERROR("%s: failed to update accounting ZAP for user "
967                                 "%d (%d)\n", osd->od_svname, la->la_gid, rc);
968                 rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
969                                         obj->oo_attr.la_gid, -1, oh->ot_tx);
970                 if (rc)
971                         CERROR("%s: failed to update accounting ZAP for user "
972                                 "%d (%d)\n", osd->od_svname,
973                                 obj->oo_attr.la_gid, rc);
974         }
975
976         write_lock(&obj->oo_attr_lock);
977         cnt = 0;
978         if (valid & LA_ATIME) {
979                 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
980                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
981                                  osa->atime, 16);
982         }
983         if (valid & LA_MTIME) {
984                 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
985                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
986                                  osa->mtime, 16);
987         }
988         if (valid & LA_CTIME) {
989                 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
990                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
991                                  osa->ctime, 16);
992         }
993         if (valid & LA_MODE) {
994                 /* mode is stored along with type, so read it first */
995                 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
996                         (la->la_mode & ~S_IFMT);
997                 osa->mode = obj->oo_attr.la_mode;
998                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
999                                  &osa->mode, 8);
1000         }
1001         if (valid & LA_SIZE) {
1002                 osa->size = obj->oo_attr.la_size = la->la_size;
1003                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1004                                  &osa->size, 8);
1005         }
1006         if (valid & LA_NLINK) {
1007                 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1008                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1009                                  &osa->nlink, 8);
1010         }
1011         if (valid & LA_RDEV) {
1012                 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1013                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1014                                  &osa->rdev, 8);
1015         }
1016         if (valid & LA_FLAGS) {
1017                 osa->flags = attrs_fs2zfs(la->la_flags);
1018                 /* many flags are not supported by zfs, so ensure a good cached
1019                  * copy */
1020                 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1021                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1022                                  &osa->flags, 8);
1023         }
1024         if (valid & LA_UID) {
1025                 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1026                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1027                                  &osa->uid, 8);
1028         }
1029         if (valid & LA_GID) {
1030                 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1031                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1032                                  &osa->gid, 8);
1033         }
1034         obj->oo_attr.la_valid |= valid;
1035         write_unlock(&obj->oo_attr_lock);
1036
1037         rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1038
1039         OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 10);
1040         RETURN(rc);
1041 }
1042
1043 /*
1044  * Object creation.
1045  *
1046  * XXX temporary solution.
1047  */
1048
1049 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1050                         struct dt_object *parent, struct dt_object *child,
1051                         umode_t child_mode)
1052 {
1053         LASSERT(ah);
1054
1055         memset(ah, 0, sizeof(*ah));
1056         ah->dah_parent = parent;
1057         ah->dah_mode = child_mode;
1058 }
1059
1060 static int osd_declare_object_create(const struct lu_env *env,
1061                                      struct dt_object *dt,
1062                                      struct lu_attr *attr,
1063                                      struct dt_allocation_hint *hint,
1064                                      struct dt_object_format *dof,
1065                                      struct thandle *handle)
1066 {
1067         char                    *buf = osd_oti_get(env)->oti_str;
1068         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1069         struct osd_object       *obj = osd_dt_obj(dt);
1070         struct osd_device       *osd = osd_obj2dev(obj);
1071         struct osd_thandle      *oh;
1072         uint64_t                 zapid;
1073         int                      rc;
1074         ENTRY;
1075
1076         LASSERT(dof);
1077
1078         switch (dof->dof_type) {
1079                 case DFT_REGULAR:
1080                 case DFT_SYM:
1081                 case DFT_NODE:
1082                         if (obj->oo_dt.do_body_ops == NULL)
1083                                 obj->oo_dt.do_body_ops = &osd_body_ops;
1084                         break;
1085                 default:
1086                         break;
1087         }
1088
1089         LASSERT(handle != NULL);
1090         oh = container_of0(handle, struct osd_thandle, ot_super);
1091         LASSERT(oh->ot_tx != NULL);
1092
1093         switch (dof->dof_type) {
1094                 case DFT_DIR:
1095                 case DFT_INDEX:
1096                         /* for zap create */
1097                         dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, 1, NULL);
1098                         break;
1099                 case DFT_REGULAR:
1100                 case DFT_SYM:
1101                 case DFT_NODE:
1102                         /* first, we'll create new object */
1103                         dmu_tx_hold_bonus(oh->ot_tx, DMU_NEW_OBJECT);
1104                         break;
1105
1106                 default:
1107                         LBUG();
1108                         break;
1109         }
1110
1111         /* and we'll add it to some mapping */
1112         zapid = osd_get_name_n_idx(env, osd, fid, buf);
1113         dmu_tx_hold_bonus(oh->ot_tx, zapid);
1114         dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, buf);
1115
1116         /* we will also update inode accounting ZAPs */
1117         dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
1118         dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
1119         dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
1120         dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
1121
1122         dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
1123
1124         __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
1125                                 XATTR_NAME_LMA, oh);
1126
1127         rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
1128                                false, NULL, false);
1129         RETURN(rc);
1130 }
1131
1132 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1133                     uint64_t oid, dmu_tx_t *tx, struct lu_attr *la,
1134                     uint64_t parent)
1135 {
1136         sa_bulk_attr_t  *bulk;
1137         sa_handle_t     *sa_hdl;
1138         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1139         uint64_t         gen;
1140         uint64_t         crtime[2];
1141         timestruc_t      now;
1142         int              cnt;
1143         int              rc;
1144
1145         gethrestime(&now);
1146         gen = dmu_tx_get_txg(tx);
1147
1148         ZFS_TIME_ENCODE(&now, crtime);
1149
1150         osa->atime[0] = la->la_atime;
1151         osa->ctime[0] = la->la_ctime;
1152         osa->mtime[0] = la->la_mtime;
1153         osa->mode = la->la_mode;
1154         osa->uid = la->la_uid;
1155         osa->gid = la->la_gid;
1156         osa->rdev = la->la_rdev;
1157         osa->nlink = la->la_nlink;
1158         osa->flags = attrs_fs2zfs(la->la_flags);
1159         osa->size  = la->la_size;
1160
1161         /* Now add in all of the "SA" attributes */
1162         rc = -sa_handle_get(osd->od_os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
1163         if (rc)
1164                 return rc;
1165
1166         OBD_ALLOC(bulk, sizeof(sa_bulk_attr_t) * 13);
1167         if (bulk == NULL) {
1168                 rc = -ENOMEM;
1169                 goto out;
1170         }
1171         /*
1172          * we need to create all SA below upon object create.
1173          *
1174          * XXX The attribute order matters since the accounting callback relies
1175          * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1176          * look up the UID/GID attributes. Moreover, the callback does not seem
1177          * to support the spill block.
1178          * We define attributes in the same order as SA_*_OFFSET in order to
1179          * work around the problem. See ORI-610.
1180          */
1181         cnt = 0;
1182         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1183         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1184         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1185         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1186         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1187         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1188         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1189         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1190         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1191         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1192         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1193         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1194         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1195
1196         rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1197
1198         OBD_FREE(bulk, sizeof(sa_bulk_attr_t) * 13);
1199 out:
1200         sa_handle_destroy(sa_hdl);
1201         return rc;
1202 }
1203
1204 /*
1205  * The transaction passed to this routine must have
1206  * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1207  * to a transaction group.
1208  */
1209 int __osd_object_create(const struct lu_env *env, struct osd_device *osd,
1210                         dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la,
1211                         uint64_t parent)
1212 {
1213         uint64_t oid;
1214         int      rc;
1215
1216         /* Assert that the transaction has been assigned to a
1217            transaction group. */
1218         LASSERT(tx->tx_txg != 0);
1219
1220         /* Create a new DMU object. */
1221         oid = dmu_object_alloc(osd->od_os, DMU_OT_PLAIN_FILE_CONTENTS, 0,
1222                                DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1223         rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, dbp);
1224         LASSERTF(rc == 0, "sa_buf_hold "LPU64" failed: %d\n", oid, rc);
1225
1226         LASSERT(la->la_valid & LA_MODE);
1227         la->la_size = 0;
1228         la->la_nlink = 1;
1229
1230         rc = __osd_attr_init(env, osd, oid, tx, la, parent);
1231         if (rc != 0) {
1232                 sa_buf_rele(*dbp, osd_obj_tag);
1233                 *dbp = NULL;
1234                 dmu_object_free(osd->od_os, oid, tx);
1235                 return rc;
1236         }
1237
1238         return 0;
1239 }
1240
1241 /*
1242  * The transaction passed to this routine must have
1243  * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1244  * to a transaction group.
1245  *
1246  * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1247  * This is fine for directories today, because storing the FID in the dirent
1248  * will also require a FAT ZAP.  If there is a new type of micro ZAP created
1249  * then we might need to re-evaluate the use of this flag and instead do
1250  * a conversion from the different internal ZAP hash formats being used. */
1251 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1252                      dmu_buf_t **zap_dbp, dmu_tx_t *tx,
1253                      struct lu_attr *la, uint64_t parent, zap_flags_t flags)
1254 {
1255         uint64_t oid;
1256         int      rc;
1257
1258         /* Assert that the transaction has been assigned to a
1259            transaction group. */
1260         LASSERT(tx->tx_txg != 0);
1261
1262         oid = zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1263                                DMU_OT_DIRECTORY_CONTENTS,
1264                                14, /* == ZFS fzap_default_block_shift */
1265                                DN_MAX_INDBLKSHIFT, /* indirect block shift */
1266                                DMU_OT_SA, DN_MAX_BONUSLEN, tx);
1267
1268         rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, zap_dbp);
1269         if (rc)
1270                 return rc;
1271
1272         LASSERT(la->la_valid & LA_MODE);
1273         la->la_size = 2;
1274         la->la_nlink = 1;
1275
1276         return __osd_attr_init(env, osd, oid, tx, la, parent);
1277 }
1278
1279 static dmu_buf_t *osd_mkidx(const struct lu_env *env, struct osd_device *osd,
1280                             struct lu_attr *la, uint64_t parent,
1281                             struct osd_thandle *oh)
1282 {
1283         dmu_buf_t *db;
1284         int        rc;
1285
1286         /* Index file should be created as regular file in order not to confuse
1287          * ZPL which could interpret them as directory.
1288          * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1289          * binary keys */
1290         LASSERT(S_ISREG(la->la_mode));
1291         rc = __osd_zap_create(env, osd, &db, oh->ot_tx, la, parent,
1292                               ZAP_FLAG_UINT64_KEY);
1293         if (rc)
1294                 return ERR_PTR(rc);
1295         return db;
1296 }
1297
1298 static dmu_buf_t *osd_mkdir(const struct lu_env *env, struct osd_device *osd,
1299                             struct lu_attr *la, uint64_t parent,
1300                             struct osd_thandle *oh)
1301 {
1302         dmu_buf_t *db;
1303         int        rc;
1304
1305         LASSERT(S_ISDIR(la->la_mode));
1306         rc = __osd_zap_create(env, osd, &db, oh->ot_tx, la, parent, 0);
1307         if (rc)
1308                 return ERR_PTR(rc);
1309         return db;
1310 }
1311
1312 static dmu_buf_t* osd_mkreg(const struct lu_env *env, struct osd_device *osd,
1313                             struct lu_attr *la, uint64_t parent,
1314                             struct osd_thandle *oh)
1315 {
1316         dmu_buf_t *db;
1317         int         rc;
1318
1319         LASSERT(S_ISREG(la->la_mode));
1320         rc = __osd_object_create(env, osd, &db, oh->ot_tx, la, parent);
1321         if (rc)
1322                 return ERR_PTR(rc);
1323
1324         /*
1325          * XXX: a hack, OST to use bigger blocksize. we need
1326          * a method in OSD API to control this from OFD/MDD
1327          */
1328         if (!lu_device_is_md(osd2lu_dev(osd))) {
1329                 rc = -dmu_object_set_blocksize(osd->od_os,
1330                                                db->db_object,
1331                                 128 << 10, 0, oh->ot_tx);
1332                 if (unlikely(rc)) {
1333                         CERROR("%s: can't change blocksize: %d\n",
1334                                osd->od_svname, rc);
1335                         return ERR_PTR(rc);
1336                 }
1337         }
1338
1339         return db;
1340 }
1341
1342 static dmu_buf_t *osd_mksym(const struct lu_env *env, struct osd_device *osd,
1343                             struct lu_attr *la, uint64_t parent,
1344                             struct osd_thandle *oh)
1345 {
1346         dmu_buf_t *db;
1347         int        rc;
1348
1349         LASSERT(S_ISLNK(la->la_mode));
1350         rc = __osd_object_create(env, osd, &db, oh->ot_tx, la, parent);
1351         if (rc)
1352                 return ERR_PTR(rc);
1353         return db;
1354 }
1355
1356 static dmu_buf_t *osd_mknod(const struct lu_env *env, struct osd_device *osd,
1357                             struct lu_attr *la, uint64_t parent,
1358                             struct osd_thandle *oh)
1359 {
1360         dmu_buf_t *db;
1361         int        rc;
1362
1363         la->la_valid = LA_MODE;
1364         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1365                 la->la_valid |= LA_RDEV;
1366
1367         rc = __osd_object_create(env, osd, &db, oh->ot_tx, la, parent);
1368         if (rc)
1369                 return ERR_PTR(rc);
1370         return db;
1371 }
1372
1373 typedef dmu_buf_t *(*osd_obj_type_f)(const struct lu_env *env,
1374                                      struct osd_device *osd,
1375                                      struct lu_attr *la,
1376                                      uint64_t parent,
1377                                      struct osd_thandle *oh);
1378
1379 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1380 {
1381         osd_obj_type_f result;
1382
1383         switch (type) {
1384         case DFT_DIR:
1385                 result = osd_mkdir;
1386                 break;
1387         case DFT_INDEX:
1388                 result = osd_mkidx;
1389                 break;
1390         case DFT_REGULAR:
1391                 result = osd_mkreg;
1392                 break;
1393         case DFT_SYM:
1394                 result = osd_mksym;
1395                 break;
1396         case DFT_NODE:
1397                 result = osd_mknod;
1398                 break;
1399         default:
1400                 LBUG();
1401                 break;
1402         }
1403         return result;
1404 }
1405
1406 /*
1407  * Primitives for directory (i.e. ZAP) handling
1408  */
1409 static inline int osd_init_lma(const struct lu_env *env, struct osd_object *obj,
1410                                const struct lu_fid *fid, struct osd_thandle *oh)
1411 {
1412         struct osd_thread_info  *info = osd_oti_get(env);
1413         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1414         struct lu_buf            buf;
1415         int rc;
1416
1417         lustre_lma_init(lma, fid, 0, 0);
1418         lustre_lma_swab(lma);
1419         buf.lb_buf = lma;
1420         buf.lb_len = sizeof(*lma);
1421
1422         rc = osd_xattr_set_internal(env, obj, &buf, XATTR_NAME_LMA,
1423                                     LU_XATTR_CREATE, oh, BYPASS_CAPA);
1424
1425         return rc;
1426 }
1427
1428 /*
1429  * Concurrency: @dt is write locked.
1430  */
1431 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1432                              struct lu_attr *attr,
1433                              struct dt_allocation_hint *hint,
1434                              struct dt_object_format *dof,
1435                              struct thandle *th)
1436 {
1437         struct zpl_direntry     *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
1438         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1439         struct osd_object       *obj = osd_dt_obj(dt);
1440         struct osd_device       *osd = osd_obj2dev(obj);
1441         char                    *buf = osd_oti_get(env)->oti_str;
1442         struct osd_thandle      *oh;
1443         dmu_buf_t               *db;
1444         uint64_t                 zapid;
1445         int                      rc;
1446
1447         ENTRY;
1448
1449         /* concurrent create declarations should not see
1450          * the object inconsistent (db, attr, etc).
1451          * in regular cases acquisition should be cheap */
1452         down(&obj->oo_guard);
1453
1454         LASSERT(osd_invariant(obj));
1455         LASSERT(!dt_object_exists(dt));
1456         LASSERT(dof != NULL);
1457
1458         LASSERT(th != NULL);
1459         oh = container_of0(th, struct osd_thandle, ot_super);
1460
1461         /*
1462          * XXX missing: Quote handling.
1463          */
1464
1465         LASSERT(obj->oo_db == NULL);
1466
1467         /* to follow ZFS on-disk format we need
1468          * to initialize parent dnode properly */
1469         zapid = 0;
1470         if (hint && hint->dah_parent)
1471                 zapid = osd_dt_obj(hint->dah_parent)->oo_db->db_object;
1472
1473         db = osd_create_type_f(dof->dof_type)(env, osd, attr, zapid, oh);
1474         if (IS_ERR(db))
1475                 GOTO(out, rc = PTR_ERR(db));
1476
1477         zde->zde_pad = 0;
1478         zde->zde_dnode = db->db_object;
1479         zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1480
1481         zapid = osd_get_name_n_idx(env, osd, fid, buf);
1482
1483         rc = -zap_add(osd->od_os, zapid, buf, 8, 1, zde, oh->ot_tx);
1484         if (rc)
1485                 GOTO(out, rc);
1486
1487         /* Add new object to inode accounting.
1488          * Errors are not considered as fatal */
1489         rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
1490                                 (attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
1491                                 oh->ot_tx);
1492         if (rc)
1493                 CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
1494                         "(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
1495         rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
1496                                 (attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
1497                                 oh->ot_tx);
1498         if (rc)
1499                 CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
1500                         "(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
1501
1502         /* configure new osd object */
1503         obj->oo_db = db;
1504         rc = osd_object_init0(env, obj);
1505         LASSERT(ergo(rc == 0, dt_object_exists(dt)));
1506         LASSERT(osd_invariant(obj));
1507
1508         rc = osd_init_lma(env, obj, fid, oh);
1509         if (rc) {
1510                 CERROR("%s: can not set LMA on "DFID": rc = %d\n",
1511                        osd->od_svname, PFID(fid), rc);
1512                 /* ignore errors during LMA initialization */
1513                 rc = 0;
1514         }
1515
1516 out:
1517         up(&obj->oo_guard);
1518         RETURN(rc);
1519 }
1520
1521 static int osd_declare_object_ref_add(const struct lu_env *env,
1522                                       struct dt_object *dt,
1523                                       struct thandle *th)
1524 {
1525         return osd_declare_attr_set(env, dt, NULL, th);
1526 }
1527
1528 /*
1529  * Concurrency: @dt is write locked.
1530  */
1531 static int osd_object_ref_add(const struct lu_env *env,
1532                               struct dt_object *dt,
1533                               struct thandle *handle)
1534 {
1535         struct osd_object       *obj = osd_dt_obj(dt);
1536         struct osd_thandle      *oh;
1537         struct osd_device       *osd = osd_obj2dev(obj);
1538         uint64_t                 nlink;
1539         int rc;
1540
1541         ENTRY;
1542
1543         LASSERT(osd_invariant(obj));
1544         LASSERT(dt_object_exists(dt));
1545         LASSERT(obj->oo_sa_hdl != NULL);
1546
1547         oh = container_of0(handle, struct osd_thandle, ot_super);
1548
1549         write_lock(&obj->oo_attr_lock);
1550         nlink = ++obj->oo_attr.la_nlink;
1551         write_unlock(&obj->oo_attr_lock);
1552
1553         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1554         return rc;
1555 }
1556
1557 static int osd_declare_object_ref_del(const struct lu_env *env,
1558                                       struct dt_object *dt,
1559                                       struct thandle *handle)
1560 {
1561         return osd_declare_attr_set(env, dt, NULL, handle);
1562 }
1563
1564 /*
1565  * Concurrency: @dt is write locked.
1566  */
1567 static int osd_object_ref_del(const struct lu_env *env,
1568                               struct dt_object *dt,
1569                               struct thandle *handle)
1570 {
1571         struct osd_object       *obj = osd_dt_obj(dt);
1572         struct osd_thandle      *oh;
1573         struct osd_device       *osd = osd_obj2dev(obj);
1574         uint64_t                 nlink;
1575         int                      rc;
1576
1577         ENTRY;
1578
1579         LASSERT(osd_invariant(obj));
1580         LASSERT(dt_object_exists(dt));
1581         LASSERT(obj->oo_sa_hdl != NULL);
1582
1583         oh = container_of0(handle, struct osd_thandle, ot_super);
1584         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1585
1586         write_lock(&obj->oo_attr_lock);
1587         nlink = --obj->oo_attr.la_nlink;
1588         write_unlock(&obj->oo_attr_lock);
1589
1590         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1591         return rc;
1592 }
1593
1594 static int capa_is_sane(const struct lu_env *env, struct osd_device *dev,
1595                         struct lustre_capa *capa, struct lustre_capa_key *keys)
1596 {
1597         struct osd_thread_info  *oti = osd_oti_get(env);
1598         struct obd_capa         *oc;
1599         int                      i, rc = 0;
1600         ENTRY;
1601
1602         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1603         if (oc) {
1604                 if (capa_is_expired(oc)) {
1605                         DEBUG_CAPA(D_ERROR, capa, "expired");
1606                         rc = -ESTALE;
1607                 }
1608                 capa_put(oc);
1609                 RETURN(rc);
1610         }
1611
1612         spin_lock(&capa_lock);
1613         for (i = 0; i < 2; i++) {
1614                 if (keys[i].lk_keyid == capa->lc_keyid) {
1615                         oti->oti_capa_key = keys[i];
1616                         break;
1617                 }
1618         }
1619         spin_unlock(&capa_lock);
1620
1621         if (i == 2) {
1622                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1623                 RETURN(-ESTALE);
1624         }
1625
1626         rc = capa_hmac(oti->oti_capa.lc_hmac, capa, oti->oti_capa_key.lk_key);
1627         if (rc)
1628                 RETURN(rc);
1629         if (memcmp(oti->oti_capa.lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac)))
1630         {
1631                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1632                 RETURN(-EACCES);
1633         }
1634
1635         oc = capa_add(dev->od_capa_hash, capa);
1636         capa_put(oc);
1637
1638         RETURN(0);
1639 }
1640
1641 static int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1642                            struct lustre_capa *capa, __u64 opc)
1643 {
1644         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1645         struct osd_device       *dev = osd_dev(dt->do_lu.lo_dev);
1646         int                      rc;
1647
1648         if (!dev->od_fl_capa)
1649                 return 0;
1650
1651         if (capa == BYPASS_CAPA)
1652                 return 0;
1653
1654         if (!capa) {
1655                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1656                 return -EACCES;
1657         }
1658
1659         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1660                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",PFID(fid));
1661                 return -EACCES;
1662         }
1663
1664         if (!capa_opc_supported(capa, opc)) {
1665                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1666                 return -EACCES;
1667         }
1668
1669         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1670                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1671                 return -EACCES;
1672         }
1673
1674         return 0;
1675 }
1676
1677 static struct obd_capa *osd_capa_get(const struct lu_env *env,
1678                                      struct dt_object *dt,
1679                                      struct lustre_capa *old,
1680                                      __u64 opc)
1681 {
1682         struct osd_thread_info  *info = osd_oti_get(env);
1683         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1684         struct osd_object       *obj = osd_dt_obj(dt);
1685         struct osd_device       *dev = osd_obj2dev(obj);
1686         struct lustre_capa_key  *key = &info->oti_capa_key;
1687         struct lustre_capa      *capa = &info->oti_capa;
1688         struct obd_capa         *oc;
1689         int                      rc;
1690         ENTRY;
1691
1692         if (!dev->od_fl_capa)
1693                 RETURN(ERR_PTR(-ENOENT));
1694
1695         LASSERT(dt_object_exists(dt));
1696         LASSERT(osd_invariant(obj));
1697
1698         /* renewal sanity check */
1699         if (old && osd_object_auth(env, dt, old, opc))
1700                 RETURN(ERR_PTR(-EACCES));
1701
1702         capa->lc_fid = *fid;
1703         capa->lc_opc = opc;
1704         capa->lc_uid = 0;
1705         capa->lc_flags = dev->od_capa_alg << 24;
1706         capa->lc_timeout = dev->od_capa_timeout;
1707         capa->lc_expiry = 0;
1708
1709         oc = capa_lookup(dev->od_capa_hash, capa, 1);
1710         if (oc) {
1711                 LASSERT(!capa_is_expired(oc));
1712                 RETURN(oc);
1713         }
1714
1715         spin_lock(&capa_lock);
1716         *key = dev->od_capa_keys[1];
1717         spin_unlock(&capa_lock);
1718
1719         capa->lc_keyid = key->lk_keyid;
1720         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
1721
1722         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
1723         if (rc) {
1724                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
1725                 LBUG();
1726                 RETURN(ERR_PTR(rc));
1727         }
1728
1729         oc = capa_add(dev->od_capa_hash, capa);
1730         RETURN(oc);
1731 }
1732
1733 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1734                            __u64 start, __u64 end)
1735 {
1736         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1737         ENTRY;
1738
1739         /* XXX: no other option than syncing the whole filesystem until we
1740          * support ZIL.  If the object tracked the txg that it was last
1741          * modified in, it could pass that txg here instead of "0".  Maybe
1742          * the changes are already committed, so no wait is needed at all? */
1743         txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
1744
1745         RETURN(0);
1746 }
1747
1748 static struct dt_object_operations osd_obj_ops = {
1749         .do_read_lock           = osd_object_read_lock,
1750         .do_write_lock          = osd_object_write_lock,
1751         .do_read_unlock         = osd_object_read_unlock,
1752         .do_write_unlock        = osd_object_write_unlock,
1753         .do_write_locked        = osd_object_write_locked,
1754         .do_attr_get            = osd_attr_get,
1755         .do_declare_attr_set    = osd_declare_attr_set,
1756         .do_attr_set            = osd_attr_set,
1757         .do_ah_init             = osd_ah_init,
1758         .do_declare_create      = osd_declare_object_create,
1759         .do_create              = osd_object_create,
1760         .do_declare_destroy     = osd_declare_object_destroy,
1761         .do_destroy             = osd_object_destroy,
1762         .do_index_try           = osd_index_try,
1763         .do_declare_ref_add     = osd_declare_object_ref_add,
1764         .do_ref_add             = osd_object_ref_add,
1765         .do_declare_ref_del     = osd_declare_object_ref_del,
1766         .do_ref_del             = osd_object_ref_del,
1767         .do_xattr_get           = osd_xattr_get,
1768         .do_declare_xattr_set   = osd_declare_xattr_set,
1769         .do_xattr_set           = osd_xattr_set,
1770         .do_declare_xattr_del   = osd_declare_xattr_del,
1771         .do_xattr_del           = osd_xattr_del,
1772         .do_xattr_list          = osd_xattr_list,
1773         .do_capa_get            = osd_capa_get,
1774         .do_object_sync         = osd_object_sync,
1775 };
1776
1777 static struct lu_object_operations osd_lu_obj_ops = {
1778         .loo_object_init        = osd_object_init,
1779         .loo_object_delete      = osd_object_delete,
1780         .loo_object_release     = osd_object_release,
1781         .loo_object_free        = osd_object_free,
1782         .loo_object_print       = osd_object_print,
1783         .loo_object_invariant   = osd_object_invariant,
1784 };
1785
1786 static int osd_otable_it_attr_get(const struct lu_env *env,
1787                                 struct dt_object *dt,
1788                                 struct lu_attr *attr,
1789                                 struct lustre_capa *capa)
1790 {
1791         attr->la_valid = 0;
1792         return 0;
1793 }
1794
1795 static struct dt_object_operations osd_obj_otable_it_ops = {
1796         .do_attr_get    = osd_otable_it_attr_get,
1797         .do_index_try   = osd_index_try,
1798 };
1799