Whamcloud - gitweb
00b3f4cf2c0de84f46eb1bf8d8c83d95c5863363
[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.gnu.org/licenses/gpl-2.0.html
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2012, 2017, Intel Corporation.
27  */
28 /*
29  * This file is part of Lustre, http://www.lustre.org/
30  * Lustre is a trademark of Sun Microsystems, Inc.
31  *
32  * lustre/osd-zfs/osd_object.c
33  *
34  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
35  * Author: Mike Pershin <tappro@whamcloud.com>
36  * Author: Johann Lombardi <johann@whamcloud.com>
37  */
38
39 #define DEBUG_SUBSYSTEM S_OSD
40
41 #include <libcfs/libcfs.h>
42 #include <obd_support.h>
43 #include <lustre_net.h>
44 #include <obd.h>
45 #include <obd_class.h>
46 #include <lustre_disk.h>
47 #include <lustre_fid.h>
48
49 #include "osd_internal.h"
50
51 #include <sys/dnode.h>
52 #include <sys/dbuf.h>
53 #include <sys/spa.h>
54 #include <sys/stat.h>
55 #include <sys/zap.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>
62 #include <sys/txg.h>
63
64 char *osd_obj_tag = "osd_object";
65 static int osd_object_sync_delay_us = -1;
66
67 static struct dt_object_operations osd_obj_ops;
68 static struct lu_object_operations osd_lu_obj_ops;
69 extern struct dt_body_operations osd_body_ops;
70 static struct dt_object_operations osd_obj_otable_it_ops;
71
72 extern struct kmem_cache *osd_object_kmem;
73
74 static void
75 osd_object_sa_fini(struct osd_object *obj)
76 {
77         if (obj->oo_sa_hdl) {
78                 sa_handle_destroy(obj->oo_sa_hdl);
79                 obj->oo_sa_hdl = NULL;
80         }
81 }
82
83 static int
84 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
85 {
86         int rc;
87
88         LASSERT(obj->oo_sa_hdl == NULL);
89         LASSERT(obj->oo_dn != NULL);
90
91         rc = osd_sa_handle_get(obj);
92         if (rc)
93                 return rc;
94
95         /* Cache the xattr object id, valid for the life of the object */
96         rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
97         if (rc == -ENOENT) {
98                 obj->oo_xattr = ZFS_NO_OBJECT;
99                 rc = 0;
100         } else if (rc) {
101                 osd_object_sa_fini(obj);
102         }
103
104         return rc;
105 }
106
107 /*
108  * Add object to list of dirty objects in tx handle.
109  */
110 void osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
111 {
112         if (!list_empty(&obj->oo_sa_linkage))
113                 return;
114
115         write_lock(&obj->oo_attr_lock);
116         if (likely(list_empty(&obj->oo_sa_linkage)))
117                 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
118         write_unlock(&obj->oo_attr_lock);
119 }
120
121 /*
122  * Release spill block dbuf hold for all dirty SAs.
123  */
124 void osd_object_sa_dirty_rele(const struct lu_env *env, struct osd_thandle *oh)
125 {
126         struct osd_object *obj;
127
128         while (!list_empty(&oh->ot_sa_list)) {
129                 obj = list_entry(oh->ot_sa_list.next,
130                                  struct osd_object, oo_sa_linkage);
131                 write_lock(&obj->oo_attr_lock);
132                 list_del_init(&obj->oo_sa_linkage);
133                 write_unlock(&obj->oo_attr_lock);
134                 if (obj->oo_late_xattr) {
135                         /*
136                          * take oo_guard to protect oo_sa_xattr buffer
137                          * from concurrent update by osd_xattr_set()
138                          */
139                         LASSERT(oh->ot_assigned != 0);
140                         down_write(&obj->oo_guard);
141                         if (obj->oo_late_attr_set)
142                                 __osd_sa_attr_init(env, obj, oh);
143                         else if (obj->oo_late_xattr)
144                                 __osd_sa_xattr_update(env, obj, oh);
145                         up_write(&obj->oo_guard);
146                 }
147                 sa_spill_rele(obj->oo_sa_hdl);
148         }
149 }
150
151 /*
152  * Update the SA and add the object to the dirty list.
153  */
154 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
155                          void *buf, uint32_t buflen, struct osd_thandle *oh)
156 {
157         int rc;
158
159         LASSERT(obj->oo_sa_hdl != NULL);
160         LASSERT(oh->ot_tx != NULL);
161
162         rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
163         osd_object_sa_dirty_add(obj, oh);
164
165         return rc;
166 }
167
168 /*
169  * Bulk update the SA and add the object to the dirty list.
170  */
171 static int
172 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
173                           int count, struct osd_thandle *oh)
174 {
175         int rc;
176
177         LASSERT(obj->oo_sa_hdl != NULL);
178         LASSERT(oh->ot_tx != NULL);
179
180         rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
181         osd_object_sa_dirty_add(obj, oh);
182
183         return rc;
184 }
185
186 /*
187  * Retrieve the attributes of a DMU object
188  */
189 int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
190                           struct osd_object *obj, struct lu_attr *la)
191 {
192         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
193         sa_bulk_attr_t  *bulk = osd_oti_get(env)->oti_attr_bulk;
194         int              cnt = 0;
195         int              rc;
196         ENTRY;
197
198         LASSERT(obj->oo_dn != NULL);
199
200         la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
201                         LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
202
203         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
204         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
205         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
206         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
207         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
208         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
209         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
210         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
211         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
212         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
213
214         rc = -sa_bulk_lookup(obj->oo_sa_hdl, bulk, cnt);
215         if (rc)
216                 GOTO(out_sa, rc);
217
218 #ifdef ZFS_PROJINHERIT
219         if (o->od_projectused_dn && osa->flags & ZFS_PROJID) {
220                 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_PROJID(o),
221                                 &osa->projid, 8);
222                 if (rc)
223                         GOTO(out_sa, rc);
224
225                 la->la_projid = osa->projid;
226                 la->la_valid |= LA_PROJID;
227                 obj->oo_with_projid = 1;
228         } else {
229                 la->la_projid = ZFS_DEFAULT_PROJID;
230                 la->la_valid &= ~LA_PROJID;
231         }
232 #else
233         la->la_projid = 0;
234         la->la_valid &= ~LA_PROJID;
235 #endif
236
237         la->la_atime = osa->atime[0];
238         la->la_mtime = osa->mtime[0];
239         la->la_ctime = osa->ctime[0];
240         la->la_mode = osa->mode;
241         la->la_uid = osa->uid;
242         la->la_gid = osa->gid;
243         la->la_nlink = osa->nlink;
244         la->la_flags = attrs_zfs2fs(osa->flags);
245         la->la_size = osa->size;
246
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;
252                 struct lu_buf buf;
253
254                 lma = (struct lustre_mdt_attrs *)info->oti_buf;
255                 buf.lb_buf = lma;
256                 buf.lb_len = sizeof(info->oti_buf);
257                 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
258                 if (rc > 0) {
259                         rc = 0;
260                         lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
261                         obj->oo_lma_flags =
262                                 lma_to_lustre_flags(lma->lma_incompat);
263
264                 } else if (rc == -ENODATA) {
265                         rc = 0;
266                 }
267         }
268
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);
271                 if (rc)
272                         GOTO(out_sa, rc);
273                 la->la_rdev = osa->rdev;
274                 la->la_valid |= LA_RDEV;
275         }
276 out_sa:
277
278         RETURN(rc);
279 }
280
281 int __osd_obj2dnode(objset_t *os, uint64_t oid, dnode_t **dnp)
282 {
283         dmu_buf_t *db;
284         dmu_buf_impl_t *dbi;
285         int rc;
286
287         rc = -dmu_bonus_hold(os, oid, osd_obj_tag, &db);
288         if (rc)
289                 return rc;
290
291         dbi = (dmu_buf_impl_t *)db;
292         DB_DNODE_ENTER(dbi);
293         *dnp = DB_DNODE(dbi);
294         LASSERT(*dnp != NULL);
295
296         return 0;
297 }
298
299 /*
300  * Concurrency: no concurrent access is possible that early in object
301  * life-cycle.
302  */
303 struct lu_object *osd_object_alloc(const struct lu_env *env,
304                                    const struct lu_object_header *hdr,
305                                    struct lu_device *d)
306 {
307         struct osd_object *mo;
308
309         OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
310         if (mo != NULL) {
311                 struct lu_object *l;
312
313                 l = &mo->oo_dt.do_lu;
314                 dt_object_init(&mo->oo_dt, NULL, d);
315                 mo->oo_dt.do_ops = &osd_obj_ops;
316                 l->lo_ops = &osd_lu_obj_ops;
317                 INIT_LIST_HEAD(&mo->oo_sa_linkage);
318                 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
319                 init_rwsem(&mo->oo_sem);
320                 init_rwsem(&mo->oo_guard);
321                 rwlock_init(&mo->oo_attr_lock);
322                 mo->oo_destroy = OSD_DESTROY_NONE;
323                 return l;
324         } else {
325                 return NULL;
326         }
327 }
328
329 static void osd_obj_set_blksize(const struct lu_env *env,
330                                 struct osd_device *osd, struct osd_object *obj)
331 {
332         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
333         dmu_tx_t *tx;
334         dnode_t *dn = obj->oo_dn;
335         uint32_t blksz;
336         int rc = 0;
337         ENTRY;
338
339         LASSERT(!osd_oti_get(env)->oti_in_trans);
340
341         tx = dmu_tx_create(osd->od_os);
342         if (!tx) {
343                 CERROR("%s: fail to create tx to set blksize for "DFID"\n",
344                        osd->od_svname, PFID(fid));
345                 RETURN_EXIT;
346         }
347
348         dmu_tx_hold_bonus(tx, dn->dn_object);
349         rc = -dmu_tx_assign(tx, TXG_WAIT);
350         if (rc) {
351                 dmu_tx_abort(tx);
352                 CERROR("%s: fail to assign tx to set blksize for "DFID
353                        ": rc = %d\n", osd->od_svname, PFID(fid), rc);
354                 RETURN_EXIT;
355         }
356
357         down_write(&obj->oo_guard);
358         if (unlikely((1 << dn->dn_datablkshift) >= PAGE_SIZE))
359                 GOTO(out, rc = 1);
360
361         blksz = dn->dn_datablksz;
362         if (!is_power_of_2(blksz))
363                 blksz = size_roundup_power2(blksz);
364
365         if (blksz > osd->od_max_blksz)
366                 blksz = osd->od_max_blksz;
367         else if (blksz < PAGE_SIZE)
368                 blksz = PAGE_SIZE;
369         rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object, blksz, 0, tx);
370
371         GOTO(out, rc);
372
373 out:
374         up_write(&obj->oo_guard);
375         if (rc) {
376                 dmu_tx_abort(tx);
377                 if (unlikely(obj->oo_dn->dn_maxblkid > 0))
378                         rc = 1;
379                 if (rc < 0)
380                         CERROR("%s: fail to set blksize for "DFID": rc = %d\n",
381                                osd->od_svname, PFID(fid), rc);
382         } else {
383                 dmu_tx_commit(tx);
384                 CDEBUG(D_INODE, "%s: set blksize as %u for "DFID"\n",
385                        osd->od_svname, blksz, PFID(fid));
386         }
387 }
388
389 /*
390  * Concurrency: shouldn't matter.
391  */
392 int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
393 {
394         struct osd_device       *osd = osd_obj2dev(obj);
395         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
396         int                      rc = 0;
397         ENTRY;
398
399         LASSERT(obj->oo_dn);
400
401         rc = osd_object_sa_init(obj, osd);
402         if (rc)
403                 RETURN(rc);
404
405         /* cache attrs in object */
406         rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
407         if (rc)
408                 RETURN(rc);
409
410         if (likely(!fid_is_acct(fid))) {
411                 /* no body operations for accounting objects */
412                 obj->oo_dt.do_body_ops = &osd_body_ops;
413
414                 if (S_ISREG(obj->oo_attr.la_mode) &&
415                     obj->oo_dn->dn_maxblkid == 0 &&
416                     (1 << obj->oo_dn->dn_datablkshift) < PAGE_SIZE &&
417                     (fid_is_idif(fid) || fid_is_norm(fid) ||
418                      fid_is_echo(fid)) &&
419                     osd->od_is_ost && !osd->od_dt_dev.dd_rdonly)
420                         osd_obj_set_blksize(env, osd, obj);
421         }
422
423         /*
424          * initialize object before marking it existing
425          */
426         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
427
428         smp_mb();
429         obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
430
431         RETURN(0);
432 }
433
434 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
435 {
436         struct osd_thread_info  *info = osd_oti_get(env);
437         struct lu_buf           buf;
438         int                     rc;
439         struct lustre_mdt_attrs *lma;
440         ENTRY;
441
442         CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
443         lma = (struct lustre_mdt_attrs *)info->oti_buf;
444         buf.lb_buf = lma;
445         buf.lb_len = sizeof(info->oti_buf);
446
447         rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
448         if (rc > 0) {
449                 rc = 0;
450                 lustre_lma_swab(lma);
451                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
452                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
453                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
454                               "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
455                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
456                               PFID(lu_object_fid(&obj->oo_dt.do_lu)));
457                         rc = -EOPNOTSUPP;
458                 } else {
459                         struct osd_device *osd = osd_obj2dev(obj);
460
461                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
462                             osd->od_is_ost)
463                                 obj->oo_pfid_in_lma = 1;
464                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
465                             osd->od_remote_parent_dir != ZFS_NO_OBJECT)
466                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
467                 }
468         } else if (rc == -ENODATA) {
469                 /* haven't initialize LMA xattr */
470                 rc = 0;
471         }
472
473         RETURN(rc);
474 }
475
476 /**
477  * Helper function to retrieve DMU object id from fid for accounting object
478  */
479 static dnode_t *osd_quota_fid2dmu(const struct osd_device *osd,
480                                   const struct lu_fid *fid)
481 {
482         dnode_t *dn = NULL;
483
484         LASSERT(fid_is_acct(fid));
485
486         switch (fid_oid(fid)) {
487         case ACCT_USER_OID:
488                 dn = osd->od_userused_dn;
489                 break;
490         case ACCT_GROUP_OID:
491                 dn = osd->od_groupused_dn;
492                 break;
493 #ifdef ZFS_PROJINHERIT
494         case ACCT_PROJECT_OID:
495                 dn = osd->od_projectused_dn;
496                 break;
497 #endif
498         default:
499                 break;
500         }
501
502         return dn;
503 }
504
505 /*
506  * Concurrency: no concurrent access is possible that early in object
507  * life-cycle.
508  */
509 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
510                            const struct lu_object_conf *conf)
511 {
512         struct osd_object *obj = osd_obj(l);
513         struct osd_device *osd = osd_obj2dev(obj);
514         const struct lu_fid *fid = lu_object_fid(l);
515         uint64_t oid;
516         int rc = 0;
517         ENTRY;
518
519         LASSERT(osd_invariant(obj));
520
521         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
522                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
523                 l->lo_header->loh_attr |= LOHA_EXISTS;
524                 RETURN(0);
525         }
526
527         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
528                 GOTO(out, rc = 0);
529
530         if (unlikely(fid_is_acct(fid))) {
531                 obj->oo_dn = osd_quota_fid2dmu(osd, fid);
532                 if (obj->oo_dn) {
533                         obj->oo_dt.do_index_ops = &osd_acct_index_ops;
534                         l->lo_header->loh_attr |= LOHA_EXISTS;
535                 }
536
537                 GOTO(out, rc = 0);
538         }
539
540         rc = osd_fid_lookup(env, osd, fid, &oid);
541         if (rc == 0) {
542                 LASSERT(obj->oo_dn == NULL);
543                 rc = __osd_obj2dnode(osd->od_os, oid, &obj->oo_dn);
544                 /* EEXIST will be returned if object is being deleted in ZFS */
545                 if (rc == -EEXIST) {
546                         rc = 0;
547                         GOTO(out, rc);
548                 }
549                 if (rc != 0) {
550                         CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
551                                osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
552                         GOTO(out, rc);
553                 }
554                 rc = osd_object_init0(env, obj);
555                 if (rc != 0)
556                         GOTO(out, rc);
557
558                 rc = osd_check_lma(env, obj);
559                 if (rc != 0)
560                         GOTO(out, rc);
561         } else if (rc == -ENOENT) {
562                 rc = 0;
563         }
564         LASSERT(osd_invariant(obj));
565 out:
566         RETURN(rc);
567 }
568
569 /*
570  * Concurrency: no concurrent access is possible that late in object
571  * life-cycle.
572  */
573 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
574 {
575         struct osd_object *obj = osd_obj(l);
576
577         LASSERT(osd_invariant(obj));
578
579         dt_object_fini(&obj->oo_dt);
580         OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
581 }
582
583 static int
584 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
585 {
586         int rc = -EBUSY;
587
588         LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
589
590         /* the object is supposed to be exclusively locked by
591          * the caller (osd_destroy()), while the transaction
592          * (oh) is per-thread and not shared */
593         if (likely(list_empty(&obj->oo_unlinked_linkage))) {
594                 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
595                 rc = 0;
596         }
597
598         return rc;
599 }
600
601 /* Default to max data size covered by a level-1 indirect block */
602 static unsigned long osd_sync_destroy_max_size =
603         1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
604 module_param(osd_sync_destroy_max_size, ulong, 0444);
605 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
606
607 static inline void
608 osd_object_set_destroy_type(struct osd_object *obj)
609 {
610         /*
611          * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
612          * only once and use it consistently thereafter.
613          */
614         down_write(&obj->oo_guard);
615         if (obj->oo_destroy == OSD_DESTROY_NONE) {
616                 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
617                         obj->oo_destroy = OSD_DESTROY_SYNC;
618                 else /* Larger objects are destroyed asynchronously */
619                         obj->oo_destroy = OSD_DESTROY_ASYNC;
620         }
621         up_write(&obj->oo_guard);
622 }
623
624 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
625                                struct thandle *th)
626 {
627         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
628         struct osd_object       *obj = osd_dt_obj(dt);
629         struct osd_device       *osd = osd_obj2dev(obj);
630         struct osd_thandle      *oh;
631         dnode_t *dn;
632         int                      rc;
633         uint64_t                 zapid;
634         ENTRY;
635
636         LASSERT(th != NULL);
637         LASSERT(dt_object_exists(dt));
638
639         oh = container_of0(th, struct osd_thandle, ot_super);
640         LASSERT(oh->ot_tx != NULL);
641
642         /* declare that we'll remove object from fid-dnode mapping */
643         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
644         osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
645
646         osd_declare_xattrs_destroy(env, obj, oh);
647
648         /* one less inode */
649         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
650                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
651                                -1, oh, NULL, OSD_QID_INODE);
652         if (rc)
653                 RETURN(rc);
654
655         /* data to be truncated */
656         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
657                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
658                                0, oh, NULL, OSD_QID_BLK);
659         if (rc)
660                 RETURN(rc);
661
662         osd_object_set_destroy_type(obj);
663         if (obj->oo_destroy == OSD_DESTROY_SYNC)
664                 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
665                                  0, DMU_OBJECT_END);
666         else
667                 osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
668                                 osd->od_unlinked, TRUE, NULL);
669
670         /* remove agent entry (if have) from remote parent */
671         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
672                 osd_tx_hold_zap(oh->ot_tx, osd->od_remote_parent_dir,
673                                 NULL, FALSE, NULL);
674
675         /* will help to find FID->ino when this object is being
676          * added to PENDING/ */
677         osd_idc_find_and_init(env, osd, obj);
678
679         RETURN(0);
680 }
681
682 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
683                        struct thandle *th)
684 {
685         struct osd_thread_info  *info = osd_oti_get(env);
686         char                    *buf = info->oti_str;
687         struct osd_object       *obj = osd_dt_obj(dt);
688         struct osd_device       *osd = osd_obj2dev(obj);
689         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
690         struct osd_thandle      *oh;
691         int                      rc;
692         uint64_t                 oid, zapid;
693         dnode_t *zdn;
694         ENTRY;
695
696         down_write(&obj->oo_guard);
697
698         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
699                 GOTO(out, rc = -ENOENT);
700
701         LASSERT(obj->oo_dn != NULL);
702
703         oh = container_of0(th, struct osd_thandle, ot_super);
704         LASSERT(oh != NULL);
705         LASSERT(oh->ot_tx != NULL);
706
707         /* remove obj ref from index dir (it depends) */
708         zapid = osd_get_name_n_idx(env, osd, fid, buf,
709                                    sizeof(info->oti_str), &zdn);
710         rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
711         if (rc) {
712                 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
713                        osd->od_svname, buf, rc);
714                 GOTO(out, rc);
715         }
716
717         rc = osd_xattrs_destroy(env, obj, oh);
718         if (rc) {
719                 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
720                        osd->od_svname, buf, rc);
721                 GOTO(out, rc);
722         }
723
724         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
725                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, true);
726                 if (rc)
727                         GOTO(out, rc);
728         }
729
730         oid = obj->oo_dn->dn_object;
731         if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
732                 /* this may happen if the destroy wasn't declared
733                  * e.g. when the object is created and then destroyed
734                  * in the same transaction - we don't need additional
735                  * space for destroy specifically */
736                 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
737                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
738                 if (rc)
739                         CERROR("%s: failed to free %s %llu: rc = %d\n",
740                                osd->od_svname, buf, oid, rc);
741         } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
742                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
743                 if (rc)
744                         CERROR("%s: failed to free %s %llu: rc = %d\n",
745                                osd->od_svname, buf, oid, rc);
746         } else { /* asynchronous destroy */
747                 char *key = info->oti_key;
748
749                 rc = osd_object_unlinked_add(obj, oh);
750                 if (rc)
751                         GOTO(out, rc);
752
753                 snprintf(key, sizeof(info->oti_key), "%llx", oid);
754                 rc = osd_zap_add(osd, osd->od_unlinked->dn_object,
755                                  osd->od_unlinked, key, 8, 1, &oid, oh->ot_tx);
756                 if (rc)
757                         CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
758                                osd->od_svname, buf, oid, rc);
759         }
760
761 out:
762         /* not needed in the cache anymore */
763         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
764         if (rc == 0)
765                 obj->oo_destroyed = 1;
766         up_write(&obj->oo_guard);
767         RETURN (0);
768 }
769
770 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
771 {
772         struct osd_object *obj = osd_obj(l);
773         const struct lu_fid *fid = lu_object_fid(l);
774
775         if (obj->oo_dn) {
776                 if (likely(!fid_is_acct(fid))) {
777                         osd_object_sa_fini(obj);
778                         if (obj->oo_sa_xattr) {
779                                 nvlist_free(obj->oo_sa_xattr);
780                                 obj->oo_sa_xattr = NULL;
781                         }
782                         osd_dnode_rele(obj->oo_dn);
783                         list_del(&obj->oo_sa_linkage);
784                 }
785                 obj->oo_dn = NULL;
786         }
787 }
788
789 /*
790  * Concurrency: ->loo_object_release() is called under site spin-lock.
791  */
792 static void osd_object_release(const struct lu_env *env,
793                                struct lu_object *l)
794 {
795 }
796
797 /*
798  * Concurrency: shouldn't matter.
799  */
800 static int osd_object_print(const struct lu_env *env, void *cookie,
801                             lu_printer_t p, const struct lu_object *l)
802 {
803         struct osd_object *o = osd_obj(l);
804
805         return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
806 }
807
808 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
809                           unsigned role)
810 {
811         struct osd_object *obj = osd_dt_obj(dt);
812
813         LASSERT(osd_invariant(obj));
814
815         down_read_nested(&obj->oo_sem, role);
816 }
817
818 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
819                            unsigned role)
820 {
821         struct osd_object *obj = osd_dt_obj(dt);
822
823         LASSERT(osd_invariant(obj));
824
825         down_write_nested(&obj->oo_sem, role);
826 }
827
828 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
829 {
830         struct osd_object *obj = osd_dt_obj(dt);
831
832         LASSERT(osd_invariant(obj));
833         up_read(&obj->oo_sem);
834 }
835
836 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
837 {
838         struct osd_object *obj = osd_dt_obj(dt);
839
840         LASSERT(osd_invariant(obj));
841         up_write(&obj->oo_sem);
842 }
843
844 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
845 {
846         struct osd_object *obj = osd_dt_obj(dt);
847         int rc = 1;
848
849         LASSERT(osd_invariant(obj));
850
851         if (down_write_trylock(&obj->oo_sem)) {
852                 rc = 0;
853                 up_write(&obj->oo_sem);
854         }
855         return rc;
856 }
857
858 static int osd_attr_get(const struct lu_env *env,
859                         struct dt_object *dt,
860                         struct lu_attr *attr)
861 {
862         struct osd_object       *obj = osd_dt_obj(dt);
863         uint64_t                 blocks;
864         uint32_t                 blksize;
865         int                      rc = 0;
866
867         down_read(&obj->oo_guard);
868
869         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
870                 GOTO(out, rc = -ENOENT);
871
872         if (unlikely(fid_is_acct(lu_object_fid(&dt->do_lu))))
873                 GOTO(out, rc = 0);
874
875         LASSERT(osd_invariant(obj));
876         LASSERT(obj->oo_dn);
877
878         read_lock(&obj->oo_attr_lock);
879         *attr = obj->oo_attr;
880         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
881                 attr->la_flags |= LUSTRE_ORPHAN_FL;
882         read_unlock(&obj->oo_attr_lock);
883
884         /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
885          * from within sa_object_size() can block on a mutex, so
886          * we can't call sa_object_size() holding rwlock */
887         sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
888         /* we do not control size of indices, so always calculate
889          * it from number of blocks reported by DMU */
890         if (S_ISDIR(attr->la_mode))
891                 attr->la_size = 512 * blocks;
892         /* Block size may be not set; suggest maximal I/O transfers. */
893         if (blksize == 0)
894                 blksize = osd_spa_maxblocksize(
895                         dmu_objset_spa(osd_obj2dev(obj)->od_os));
896
897         attr->la_blksize = blksize;
898         attr->la_blocks = blocks;
899         attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
900
901 out:
902         up_read(&obj->oo_guard);
903         return rc;
904 }
905
906 /* Simple wrapper on top of qsd API which implement quota transfer for osd
907  * setattr needs. As a reminder, only the root user can change ownership of
908  * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
909 static inline int qsd_transfer(const struct lu_env *env,
910                                struct qsd_instance *qsd,
911                                struct lquota_trans *trans, int qtype,
912                                __u64 orig_id, __u64 new_id, __u64 bspace,
913                                struct lquota_id_info *qi)
914 {
915         int     rc;
916
917         if (unlikely(qsd == NULL))
918                 return 0;
919
920         LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
921         qi->lqi_type = qtype;
922
923         /* inode accounting */
924         qi->lqi_is_blk = false;
925
926         /* one more inode for the new owner ... */
927         qi->lqi_id.qid_uid = new_id;
928         qi->lqi_space      = 1;
929         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
930         if (rc == -EDQUOT || rc == -EINPROGRESS)
931                 rc = 0;
932         if (rc)
933                 return rc;
934
935         /* and one less inode for the current id */
936         qi->lqi_id.qid_uid = orig_id;;
937         qi->lqi_space      = -1;
938         /* can't get EDQUOT when reducing usage */
939         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
940         if (rc == -EINPROGRESS)
941                 rc = 0;
942         if (rc)
943                 return rc;
944
945         /* block accounting */
946         qi->lqi_is_blk = true;
947
948         /* more blocks for the new owner ... */
949         qi->lqi_id.qid_uid = new_id;
950         qi->lqi_space      = bspace;
951         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
952         if (rc == -EDQUOT || rc == -EINPROGRESS)
953                 rc = 0;
954         if (rc)
955                 return rc;
956
957         /* and finally less blocks for the current owner */
958         qi->lqi_id.qid_uid = orig_id;
959         qi->lqi_space      = -bspace;
960         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
961         /* can't get EDQUOT when reducing usage */
962         if (rc == -EINPROGRESS)
963                 rc = 0;
964         return rc;
965 }
966
967 static int osd_declare_attr_set(const struct lu_env *env,
968                                 struct dt_object *dt,
969                                 const struct lu_attr *attr,
970                                 struct thandle *handle)
971 {
972         struct osd_thread_info  *info = osd_oti_get(env);
973         struct osd_object       *obj = osd_dt_obj(dt);
974         struct osd_device       *osd = osd_obj2dev(obj);
975         dmu_tx_hold_t           *txh;
976         struct osd_thandle      *oh;
977         uint64_t                 bspace;
978         uint32_t                 blksize;
979         int                      rc = 0;
980         bool                     found;
981         ENTRY;
982
983
984         LASSERT(handle != NULL);
985         LASSERT(osd_invariant(obj));
986
987         oh = container_of0(handle, struct osd_thandle, ot_super);
988
989         down_read(&obj->oo_guard);
990         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
991                 GOTO(out, rc = 0);
992
993         LASSERT(obj->oo_sa_hdl != NULL);
994         LASSERT(oh->ot_tx != NULL);
995         /* regular attributes are part of the bonus buffer */
996         /* let's check whether this object is already part of
997          * transaction.. */
998         found = false;
999         for (txh = list_head(&oh->ot_tx->tx_holds); txh;
1000              txh = list_next(&oh->ot_tx->tx_holds, txh)) {
1001                 if (txh->txh_dnode == NULL)
1002                         continue;
1003                 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
1004                         continue;
1005                 /* this object is part of the transaction already
1006                  * we don't need to declare bonus again */
1007                 found = true;
1008                 break;
1009         }
1010         if (!found)
1011                 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
1012         if (oh->ot_tx->tx_err != 0)
1013                 GOTO(out, rc = -oh->ot_tx->tx_err);
1014
1015         if (attr && attr->la_valid & LA_FLAGS) {
1016                 /* LMA is usually a part of bonus, no need to declare
1017                  * anything else */
1018         }
1019
1020         if (attr && (attr->la_valid & (LA_UID | LA_GID | LA_PROJID))) {
1021                 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
1022                 bspace = toqb(bspace * blksize);
1023         }
1024
1025         if (attr && attr->la_valid & LA_UID) {
1026                 /* quota enforcement for user */
1027                 if (attr->la_uid != obj->oo_attr.la_uid) {
1028                         rc = qsd_transfer(env, osd->od_quota_slave,
1029                                           &oh->ot_quota_trans, USRQUOTA,
1030                                           obj->oo_attr.la_uid, attr->la_uid,
1031                                           bspace, &info->oti_qi);
1032                         if (rc)
1033                                 GOTO(out, rc);
1034                 }
1035         }
1036         if (attr && attr->la_valid & LA_GID) {
1037                 /* quota enforcement for group */
1038                 if (attr->la_gid != obj->oo_attr.la_gid) {
1039                         rc = qsd_transfer(env, osd->od_quota_slave,
1040                                           &oh->ot_quota_trans, GRPQUOTA,
1041                                           obj->oo_attr.la_gid, attr->la_gid,
1042                                           bspace, &info->oti_qi);
1043                         if (rc)
1044                                 GOTO(out, rc);
1045                 }
1046         }
1047 #ifdef ZFS_PROJINHERIT
1048         if (attr && attr->la_valid & LA_PROJID) {
1049                 if (!osd->od_projectused_dn)
1050                         GOTO(out, rc = -EOPNOTSUPP);
1051
1052                 /* Usually, if project quota is upgradable for the device,
1053                  * then the upgrade will be done before or when mount the
1054                  * device. So when we come here, this project should have
1055                  * project ID attribute already (that is zero by default).
1056                  * Otherwise, there was something wrong during the former
1057                  * upgrade, let's return failure to report that.
1058                  *
1059                  * Please note that, different from other attributes, you
1060                  * can NOT simply set the project ID attribute under such
1061                  * case, because adding (NOT change) project ID attribute
1062                  * needs to change the object's attribute layout to match
1063                  * zfs backend quota accounting requirement. */
1064                 if (unlikely(!obj->oo_with_projid))
1065                         GOTO(out, rc = -ENXIO);
1066
1067                 /* quota enforcement for project */
1068                 if (attr->la_projid != obj->oo_attr.la_projid) {
1069                         rc = qsd_transfer(env, osd->od_quota_slave,
1070                                           &oh->ot_quota_trans, PRJQUOTA,
1071                                           obj->oo_attr.la_projid,
1072                                           attr->la_projid, bspace,
1073                                           &info->oti_qi);
1074                         if (rc)
1075                                 GOTO(out, rc);
1076                 }
1077         }
1078 #endif
1079 out:
1080         up_read(&obj->oo_guard);
1081         RETURN(rc);
1082 }
1083
1084 /*
1085  * Set the attributes of an object
1086  *
1087  * The transaction passed to this routine must have
1088  * dmu_tx_hold_bonus(tx, oid) called and then assigned
1089  * to a transaction group.
1090  */
1091 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
1092                         const struct lu_attr *la, struct thandle *handle)
1093 {
1094         struct osd_thread_info  *info = osd_oti_get(env);
1095         sa_bulk_attr_t          *bulk = osd_oti_get(env)->oti_attr_bulk;
1096         struct osd_object       *obj = osd_dt_obj(dt);
1097         struct osd_device       *osd = osd_obj2dev(obj);
1098         struct osd_thandle      *oh;
1099         struct osa_attr         *osa = &info->oti_osa;
1100         __u64                    valid = la->la_valid;
1101         int                      cnt;
1102         int                      rc = 0;
1103
1104         ENTRY;
1105
1106         down_read(&obj->oo_guard);
1107         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1108                 GOTO(out, rc = -ENOENT);
1109
1110         LASSERT(handle != NULL);
1111         LASSERT(osd_invariant(obj));
1112         LASSERT(obj->oo_sa_hdl);
1113
1114         oh = container_of0(handle, struct osd_thandle, ot_super);
1115         /* Assert that the transaction has been assigned to a
1116            transaction group. */
1117         LASSERT(oh->ot_tx->tx_txg != 0);
1118
1119         /* Only allow set size for regular file */
1120         if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
1121                 valid &= ~(LA_SIZE | LA_BLOCKS);
1122
1123         if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
1124                 valid &= ~LA_CTIME;
1125
1126         if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
1127                 valid &= ~LA_MTIME;
1128
1129         if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
1130                 valid &= ~LA_ATIME;
1131
1132         if (valid == 0)
1133                 GOTO(out, rc = 0);
1134
1135         if (valid & LA_FLAGS) {
1136                 struct lustre_mdt_attrs *lma;
1137                 struct lu_buf buf;
1138
1139                 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
1140                         LASSERT(!obj->oo_pfid_in_lma);
1141                         CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
1142                         lma = (struct lustre_mdt_attrs *)&info->oti_buf;
1143                         buf.lb_buf = lma;
1144                         buf.lb_len = sizeof(info->oti_buf);
1145                         rc = osd_xattr_get(env, &obj->oo_dt, &buf,
1146                                            XATTR_NAME_LMA);
1147                         if (rc > 0) {
1148                                 lma->lma_incompat =
1149                                         le32_to_cpu(lma->lma_incompat);
1150                                 lma->lma_incompat |=
1151                                         lustre_to_lma_flags(la->la_flags);
1152                                 lma->lma_incompat =
1153                                         cpu_to_le32(lma->lma_incompat);
1154                                 buf.lb_buf = lma;
1155                                 buf.lb_len = sizeof(*lma);
1156                                 rc = osd_xattr_set_internal(env, obj, &buf,
1157                                                             XATTR_NAME_LMA,
1158                                                             LU_XATTR_REPLACE,
1159                                                             oh);
1160                         }
1161                         if (rc < 0) {
1162                                 CWARN("%s: failed to set LMA flags: rc = %d\n",
1163                                        osd->od_svname, rc);
1164                                 GOTO(out, rc);
1165                         }
1166                 }
1167         }
1168
1169         write_lock(&obj->oo_attr_lock);
1170         cnt = 0;
1171
1172         if (valid & LA_PROJID) {
1173 #ifdef ZFS_PROJINHERIT
1174                 /* osd_declare_attr_set() must be called firstly.
1175                  * If osd::od_projectused_dn is not set, then we
1176                  * can not arrive at here. */
1177                 LASSERT(osd->od_projectused_dn);
1178                 LASSERT(obj->oo_with_projid);
1179
1180                 osa->projid = obj->oo_attr.la_projid = la->la_projid;
1181                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1182                                  &osa->projid, 8);
1183 #else
1184                 valid &= ~LA_PROJID;
1185 #endif
1186         }
1187
1188         if (valid & LA_ATIME) {
1189                 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1190                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1191                                  osa->atime, 16);
1192         }
1193         if (valid & LA_MTIME) {
1194                 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1195                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1196                                  osa->mtime, 16);
1197         }
1198         if (valid & LA_CTIME) {
1199                 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1200                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1201                                  osa->ctime, 16);
1202         }
1203         if (valid & LA_MODE) {
1204                 /* mode is stored along with type, so read it first */
1205                 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1206                         (la->la_mode & ~S_IFMT);
1207                 osa->mode = obj->oo_attr.la_mode;
1208                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1209                                  &osa->mode, 8);
1210         }
1211         if (valid & LA_SIZE) {
1212                 osa->size = obj->oo_attr.la_size = la->la_size;
1213                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1214                                  &osa->size, 8);
1215         }
1216         if (valid & LA_NLINK) {
1217                 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1218                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1219                                  &osa->nlink, 8);
1220         }
1221         if (valid & LA_RDEV) {
1222                 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1223                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1224                                  &osa->rdev, 8);
1225         }
1226         if (valid & LA_FLAGS) {
1227                 osa->flags = attrs_fs2zfs(la->la_flags);
1228                 /* many flags are not supported by zfs, so ensure a good cached
1229                  * copy */
1230                 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1231 #ifdef ZFS_PROJINHERIT
1232                 if (obj->oo_with_projid)
1233                         osa->flags |= ZFS_PROJID;
1234 #endif
1235                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1236                                  &osa->flags, 8);
1237         }
1238         if (valid & LA_UID) {
1239                 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1240                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1241                                  &osa->uid, 8);
1242         }
1243         if (valid & LA_GID) {
1244                 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1245                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1246                                  &osa->gid, 8);
1247         }
1248         obj->oo_attr.la_valid |= valid;
1249         write_unlock(&obj->oo_attr_lock);
1250
1251         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1252         rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1253
1254 out:
1255         up_read(&obj->oo_guard);
1256         RETURN(rc);
1257 }
1258
1259 /*
1260  * Object creation.
1261  *
1262  * XXX temporary solution.
1263  */
1264
1265 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1266                         struct dt_object *parent, struct dt_object *child,
1267                         umode_t child_mode)
1268 {
1269         LASSERT(ah);
1270
1271         ah->dah_parent = parent;
1272         ah->dah_mode = child_mode;
1273
1274         if (parent != NULL && !dt_object_remote(parent)) {
1275                 /* will help to find FID->ino at dt_insert("..") */
1276                 struct osd_object *pobj = osd_dt_obj(parent);
1277
1278                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1279         }
1280 }
1281
1282 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
1283                               struct lu_attr *attr,
1284                               struct dt_allocation_hint *hint,
1285                               struct dt_object_format *dof,
1286                               struct thandle *handle)
1287 {
1288         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1289         struct osd_object       *obj = osd_dt_obj(dt);
1290         struct osd_device       *osd = osd_obj2dev(obj);
1291         struct osd_thandle      *oh;
1292         uint64_t                 zapid;
1293         dnode_t                 *dn;
1294         int                      rc, dnode_size;
1295         ENTRY;
1296
1297         LASSERT(dof);
1298
1299         switch (dof->dof_type) {
1300                 case DFT_REGULAR:
1301                 case DFT_SYM:
1302                 case DFT_NODE:
1303                         if (obj->oo_dt.do_body_ops == NULL)
1304                                 obj->oo_dt.do_body_ops = &osd_body_ops;
1305                         break;
1306                 default:
1307                         break;
1308         }
1309
1310         LASSERT(handle != NULL);
1311         oh = container_of0(handle, struct osd_thandle, ot_super);
1312         LASSERT(oh->ot_tx != NULL);
1313
1314         /* this is the minimum set of EAs on every Lustre object */
1315         obj->oo_ea_in_bonus = OSD_BASE_EA_IN_BONUS;
1316         /* reserve 32 bytes for extra stuff like ACLs */
1317         dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1318
1319         switch (dof->dof_type) {
1320                 case DFT_DIR:
1321                         dt->do_index_ops = &osd_dir_ops;
1322                 case DFT_INDEX:
1323                         /* for zap create */
1324                         dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1325                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1326                         break;
1327                 case DFT_REGULAR:
1328                 case DFT_SYM:
1329                 case DFT_NODE:
1330                         /* first, we'll create new object */
1331                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1332                         break;
1333
1334                 default:
1335                         LBUG();
1336                         break;
1337         }
1338
1339         /* and we'll add it to some mapping */
1340         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
1341         osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
1342
1343         /* will help to find FID->ino mapping at dt_insert() */
1344         osd_idc_find_and_init(env, osd, obj);
1345
1346         rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid,
1347                                attr->la_projid, 1, oh, NULL, OSD_QID_INODE);
1348
1349         RETURN(rc);
1350 }
1351
1352 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1353                     struct osd_object *obj, sa_handle_t *sa_hdl, dmu_tx_t *tx,
1354                     struct lu_attr *la, uint64_t parent,
1355                     nvlist_t *xattr)
1356 {
1357         sa_bulk_attr_t  *bulk = osd_oti_get(env)->oti_attr_bulk;
1358         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1359         uint64_t         gen;
1360         uint64_t         crtime[2];
1361         timestruc_t      now;
1362         int              cnt;
1363         int              rc;
1364         char *dxattr = NULL;
1365         size_t sa_size;
1366
1367
1368         LASSERT(sa_hdl);
1369
1370         gen = dmu_tx_get_txg(tx);
1371         gethrestime(&now);
1372         ZFS_TIME_ENCODE(&now, crtime);
1373
1374         osa->atime[0] = la->la_atime;
1375         osa->ctime[0] = la->la_ctime;
1376         osa->mtime[0] = la->la_mtime;
1377         osa->mode = la->la_mode;
1378         osa->uid = la->la_uid;
1379         osa->gid = la->la_gid;
1380         osa->rdev = la->la_rdev;
1381         osa->nlink = la->la_nlink;
1382         if (la->la_valid & LA_FLAGS)
1383                 osa->flags = attrs_fs2zfs(la->la_flags);
1384         else
1385                 osa->flags = 0;
1386         osa->size  = la->la_size;
1387 #ifdef ZFS_PROJINHERIT
1388         if (osd->od_projectused_dn) {
1389                 if (la->la_valid & LA_PROJID)
1390                         osa->projid = la->la_projid;
1391                 else
1392                         osa->projid = ZFS_DEFAULT_PROJID;
1393                 osa->flags |= ZFS_PROJID;
1394                 if (obj)
1395                         obj->oo_with_projid = 1;
1396         } else {
1397                 osa->flags &= ~ZFS_PROJID;
1398         }
1399 #endif
1400
1401         /*
1402          * we need to create all SA below upon object create.
1403          *
1404          * XXX The attribute order matters since the accounting callback relies
1405          * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1406          * look up the UID/GID/PROJID attributes. Moreover, the callback does
1407          * not seem to support the spill block.
1408          * We define attributes in the same order as SA_*_OFFSET in order to
1409          * work around the problem. See ORI-610.
1410          */
1411         cnt = 0;
1412         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1413         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1414         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1415         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1416         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1417         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1418         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1419         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1420         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1421         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1422         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1423         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1424 #ifdef ZFS_PROJINHERIT
1425         if (osd->od_projectused_dn)
1426                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1427                                  &osa->projid, 8);
1428 #endif
1429         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1430         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1431
1432         if (xattr) {
1433                 rc = -nvlist_size(xattr, &sa_size, NV_ENCODE_XDR);
1434                 LASSERT(rc == 0);
1435
1436                 dxattr = osd_zio_buf_alloc(sa_size);
1437                 LASSERT(dxattr);
1438
1439                 rc = -nvlist_pack(xattr, &dxattr, &sa_size,
1440                                 NV_ENCODE_XDR, KM_SLEEP);
1441                 LASSERT(rc == 0);
1442
1443                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd),
1444                                 NULL, dxattr, sa_size);
1445         }
1446
1447         rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1448         if (dxattr)
1449                 osd_zio_buf_free(dxattr, sa_size);
1450
1451         return rc;
1452 }
1453
1454 static int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1455                               uint64_t oid, dnode_t **dnp)
1456 {
1457         dmu_tx_hold_t *txh;
1458         int rc = 0;
1459
1460         /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1461         for (txh = list_tail(&tx->tx_holds); txh;
1462              txh = list_prev(&tx->tx_holds, txh)) {
1463                 dnode_t *dn = txh->txh_dnode;
1464                 dmu_buf_impl_t *db;
1465
1466                 if (dn == NULL)
1467                         continue;
1468                 if (dn->dn_object != oid)
1469                         continue;
1470                 db = dn->dn_bonus;
1471                 if (db == NULL) {
1472                         rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1473                         if (dn->dn_bonus == NULL)
1474                                 dbuf_create_bonus(dn);
1475                         rw_exit(&dn->dn_struct_rwlock);
1476                 }
1477                 db = dn->dn_bonus;
1478                 LASSERT(db);
1479                 LASSERT(dn->dn_handle);
1480                 DB_DNODE_ENTER(db);
1481                 if (refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1482                         refcount_add(&dn->dn_holds, tag);
1483                         atomic_inc_32(&dn->dn_dbufs_count);
1484                 }
1485                 *dnp = dn;
1486                 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1487                 break;
1488         }
1489
1490         if (unlikely(*dnp == NULL))
1491                 rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
1492
1493         return rc;
1494 }
1495
1496 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
1497 int osd_find_dnsize(struct osd_device *osd, int ea_in_bonus)
1498 {
1499         int dnsize;
1500
1501         if (osd->od_dnsize == ZFS_DNSIZE_AUTO) {
1502                 dnsize = DNODE_MIN_SIZE;
1503                 do {
1504                         if (DN_BONUS_SIZE(dnsize) >= ea_in_bonus + 32)
1505                                 break;
1506                         dnsize <<= 1;
1507                 } while (dnsize < DNODE_MAX_SIZE);
1508                 if (dnsize > DNODE_MAX_SIZE)
1509                         dnsize = DNODE_MAX_SIZE;
1510         } else if (osd->od_dnsize == ZFS_DNSIZE_1K) {
1511                 dnsize = 1024;
1512         } else if (osd->od_dnsize == ZFS_DNSIZE_2K) {
1513                 dnsize = 2048;
1514         } else if (osd->od_dnsize == ZFS_DNSIZE_4K) {
1515                 dnsize = 4096;
1516         } else if (osd->od_dnsize == ZFS_DNSIZE_8K) {
1517                 dnsize = 8192;
1518         } else if (osd->od_dnsize == ZFS_DNSIZE_16K) {
1519                 dnsize = 16384;
1520         } else {
1521                 dnsize = DNODE_MIN_SIZE;
1522         }
1523         return dnsize;
1524 }
1525 #endif
1526
1527 /*
1528  * The transaction passed to this routine must have
1529  * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1530  * to a transaction group.
1531  */
1532 int __osd_object_create(const struct lu_env *env, struct osd_device *osd,
1533                         struct osd_object *obj, const struct lu_fid *fid,
1534                         dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1535 {
1536         dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1537         uint64_t oid;
1538         int size;
1539
1540         /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1541          * would get an additional ditto copy */
1542         if (unlikely(S_ISREG(la->la_mode) &&
1543                      fid_seq_is_local_file(fid_seq(fid))))
1544                 type = DMU_OTN_UINT8_METADATA;
1545
1546         /* Create a new DMU object using the default dnode size. */
1547         if (obj)
1548                 size = obj->oo_ea_in_bonus;
1549         else
1550                 size = OSD_BASE_EA_IN_BONUS;
1551         oid = osd_dmu_object_alloc(osd->od_os, type, 0,
1552                                    osd_find_dnsize(osd, size), tx);
1553
1554         LASSERT(la->la_valid & LA_MODE);
1555         la->la_size = 0;
1556         la->la_nlink = 1;
1557
1558         return osd_find_new_dnode(env, tx, oid, dnp);
1559 }
1560
1561 /*
1562  * The transaction passed to this routine must have
1563  * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1564  * to a transaction group.
1565  *
1566  * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1567  * This is fine for directories today, because storing the FID in the dirent
1568  * will also require a FAT ZAP.  If there is a new type of micro ZAP created
1569  * then we might need to re-evaluate the use of this flag and instead do
1570  * a conversion from the different internal ZAP hash formats being used. */
1571 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1572                      dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1573                      unsigned dnsize, zap_flags_t flags)
1574 {
1575         uint64_t oid;
1576
1577         /* Assert that the transaction has been assigned to a
1578            transaction group. */
1579         LASSERT(tx->tx_txg != 0);
1580         *dnp = NULL;
1581
1582         oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1583                                    DMU_OT_DIRECTORY_CONTENTS,
1584                                    14, /* == ZFS fzap_default_blockshift */
1585                                    DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1586                                    dnsize, tx);
1587
1588         la->la_size = 2;
1589         la->la_nlink = 1;
1590
1591         return osd_find_new_dnode(env, tx, oid, dnp);
1592 }
1593
1594 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1595                           struct lu_attr *la, struct osd_thandle *oh)
1596 {
1597         struct osd_device *osd = osd_obj2dev(obj);
1598         dnode_t *dn;
1599         int rc;
1600
1601         /* Index file should be created as regular file in order not to confuse
1602          * ZPL which could interpret them as directory.
1603          * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1604          * binary keys */
1605         LASSERT(S_ISREG(la->la_mode));
1606         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1607                 osd_find_dnsize(osd, obj->oo_ea_in_bonus), ZAP_FLAG_UINT64_KEY);
1608         if (rc)
1609                 return ERR_PTR(rc);
1610         return dn;
1611 }
1612
1613 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1614                           struct lu_attr *la, struct osd_thandle *oh)
1615 {
1616         struct osd_device *osd = osd_obj2dev(obj);
1617         dnode_t *dn;
1618         int rc;
1619
1620         LASSERT(S_ISDIR(la->la_mode));
1621         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1622                               osd_find_dnsize(osd, obj->oo_ea_in_bonus), 0);
1623         if (rc)
1624                 return ERR_PTR(rc);
1625         return dn;
1626 }
1627
1628 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1629                           struct lu_attr *la, struct osd_thandle *oh)
1630 {
1631         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1632         struct osd_device *osd = osd_obj2dev(obj);
1633         dnode_t *dn;
1634         int rc;
1635
1636         LASSERT(S_ISREG(la->la_mode));
1637         rc = __osd_object_create(env, osd, obj, fid, &dn, oh->ot_tx, la);
1638         if (rc)
1639                 return ERR_PTR(rc);
1640
1641         if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid))) {
1642                 /* The minimum block size must be at least page size otherwise
1643                  * it will break the assumption in tgt_thread_big_cache where
1644                  * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1645                  * RDMA due to subpage transfer size */
1646                 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1647                                                PAGE_SIZE, 0, oh->ot_tx);
1648                 if (unlikely(rc)) {
1649                         CERROR("%s: can't change blocksize: %d\n",
1650                                osd->od_svname, rc);
1651                         return ERR_PTR(rc);
1652                 }
1653         }
1654
1655         return dn;
1656 }
1657
1658 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1659                           struct lu_attr *la, struct osd_thandle *oh)
1660 {
1661         dnode_t *dn;
1662         int rc;
1663
1664         LASSERT(S_ISLNK(la->la_mode));
1665         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1666                                  lu_object_fid(&obj->oo_dt.do_lu),
1667                                  &dn, oh->ot_tx, la);
1668         if (rc)
1669                 return ERR_PTR(rc);
1670         return dn;
1671 }
1672
1673 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1674                           struct lu_attr *la, struct osd_thandle *oh)
1675 {
1676         dnode_t *dn;
1677         int rc;
1678
1679         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1680                 la->la_valid |= LA_RDEV;
1681
1682         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1683                                  lu_object_fid(&obj->oo_dt.do_lu),
1684                                  &dn, oh->ot_tx, la);
1685         if (rc)
1686                 return ERR_PTR(rc);
1687         return dn;
1688 }
1689
1690 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1691                                    struct osd_object *obj,
1692                                    struct lu_attr *la,
1693                                    struct osd_thandle *oh);
1694
1695 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1696 {
1697         osd_obj_type_f result;
1698
1699         switch (type) {
1700         case DFT_DIR:
1701                 result = osd_mkdir;
1702                 break;
1703         case DFT_INDEX:
1704                 result = osd_mkidx;
1705                 break;
1706         case DFT_REGULAR:
1707                 result = osd_mkreg;
1708                 break;
1709         case DFT_SYM:
1710                 result = osd_mksym;
1711                 break;
1712         case DFT_NODE:
1713                 result = osd_mknod;
1714                 break;
1715         default:
1716                 LBUG();
1717                 break;
1718         }
1719         return result;
1720 }
1721
1722 /*
1723  * Concurrency: @dt is write locked.
1724  */
1725 static int osd_create(const struct lu_env *env, struct dt_object *dt,
1726                       struct lu_attr *attr, struct dt_allocation_hint *hint,
1727                       struct dt_object_format *dof, struct thandle *th)
1728 {
1729         struct osd_thread_info  *info = osd_oti_get(env);
1730         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1731         struct zpl_direntry     *zde = &info->oti_zde.lzd_reg;
1732         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1733         struct osd_object       *obj = osd_dt_obj(dt);
1734         struct osd_device       *osd = osd_obj2dev(obj);
1735         char                    *buf = info->oti_str;
1736         struct osd_thandle      *oh;
1737         dnode_t *dn = NULL, *zdn = NULL;
1738         uint64_t                 zapid, parent = 0;
1739         int                      rc;
1740
1741         ENTRY;
1742
1743         LASSERT(!fid_is_acct(fid));
1744
1745         /* concurrent create declarations should not see
1746          * the object inconsistent (db, attr, etc).
1747          * in regular cases acquisition should be cheap */
1748         down_write(&obj->oo_guard);
1749
1750         if (unlikely(dt_object_exists(dt)))
1751                 GOTO(out, rc = -EEXIST);
1752
1753         LASSERT(osd_invariant(obj));
1754         LASSERT(dof != NULL);
1755
1756         LASSERT(th != NULL);
1757         oh = container_of0(th, struct osd_thandle, ot_super);
1758
1759         LASSERT(obj->oo_dn == NULL);
1760
1761         /* to follow ZFS on-disk format we need
1762          * to initialize parent dnode properly */
1763         if (hint != NULL && hint->dah_parent != NULL &&
1764             !dt_object_remote(hint->dah_parent))
1765                 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1766
1767         /* we may fix some attributes, better do not change the source */
1768         obj->oo_attr = *attr;
1769         obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1770
1771 #ifdef ZFS_PROJINHERIT
1772         if (osd->od_projectused_dn) {
1773                 if (!(obj->oo_attr.la_valid & LA_PROJID))
1774                         obj->oo_attr.la_projid = ZFS_DEFAULT_PROJID;
1775                 obj->oo_with_projid = 1;
1776         }
1777 #endif
1778
1779         dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1780         if (IS_ERR(dn)) {
1781                 rc = PTR_ERR(dn);
1782                 dn = NULL;
1783                 GOTO(out, rc);
1784         }
1785
1786         zde->zde_pad = 0;
1787         zde->zde_dnode = dn->dn_object;
1788         zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1789
1790         zapid = osd_get_name_n_idx(env, osd, fid, buf,
1791                                    sizeof(info->oti_str), &zdn);
1792         rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
1793         if (rc)
1794                 GOTO(out, rc);
1795         obj->oo_dn = dn;
1796         /* Now add in all of the "SA" attributes */
1797         rc = osd_sa_handle_get(obj);
1798         if (rc)
1799                 GOTO(out, rc);
1800
1801         rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1802         if (rc)
1803                 GOTO(out, rc);
1804
1805         /* initialize LMA */
1806         lustre_lma_init(lma, fid, 0, 0);
1807         lustre_lma_swab(lma);
1808         rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
1809                                     (uchar_t *)lma, sizeof(*lma));
1810         if (rc)
1811                 GOTO(out, rc);
1812
1813         /* configure new osd object */
1814         obj->oo_parent = parent != 0 ? parent : zapid;
1815         obj->oo_late_attr_set = 1;
1816         rc = __osd_sa_xattr_schedule_update(env, obj, oh);
1817         if (rc)
1818                 GOTO(out, rc);
1819
1820         /* XXX: oo_lma_flags */
1821         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
1822         if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
1823                 /* no body operations for accounting objects */
1824                 obj->oo_dt.do_body_ops = &osd_body_ops;
1825
1826         osd_idc_find_and_init(env, osd, obj);
1827
1828 out:
1829         if (unlikely(rc && dn)) {
1830                 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
1831                 osd_dnode_rele(dn);
1832                 obj->oo_dn = NULL;
1833         } else if (!rc) {
1834                 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
1835         }
1836         up_write(&obj->oo_guard);
1837         RETURN(rc);
1838 }
1839
1840 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
1841                                struct thandle *th)
1842 {
1843         return osd_declare_attr_set(env, dt, NULL, th);
1844 }
1845
1846 /*
1847  * Concurrency: @dt is write locked.
1848  */
1849 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
1850                        struct thandle *handle)
1851 {
1852         struct osd_object       *obj = osd_dt_obj(dt);
1853         struct osd_thandle      *oh;
1854         struct osd_device       *osd = osd_obj2dev(obj);
1855         uint64_t                 nlink;
1856         int rc;
1857
1858         ENTRY;
1859
1860         down_read(&obj->oo_guard);
1861         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1862                 GOTO(out, rc = -ENOENT);
1863
1864         LASSERT(osd_invariant(obj));
1865         LASSERT(obj->oo_sa_hdl != NULL);
1866
1867         oh = container_of0(handle, struct osd_thandle, ot_super);
1868
1869         write_lock(&obj->oo_attr_lock);
1870         nlink = ++obj->oo_attr.la_nlink;
1871         write_unlock(&obj->oo_attr_lock);
1872
1873         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1874
1875 out:
1876         up_read(&obj->oo_guard);
1877         RETURN(rc);
1878 }
1879
1880 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
1881                                struct thandle *handle)
1882 {
1883         return osd_declare_attr_set(env, dt, NULL, handle);
1884 }
1885
1886 /*
1887  * Concurrency: @dt is write locked.
1888  */
1889 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
1890                        struct thandle *handle)
1891 {
1892         struct osd_object       *obj = osd_dt_obj(dt);
1893         struct osd_thandle      *oh;
1894         struct osd_device       *osd = osd_obj2dev(obj);
1895         uint64_t                 nlink;
1896         int                      rc;
1897
1898         ENTRY;
1899
1900         down_read(&obj->oo_guard);
1901
1902         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1903                 GOTO(out, rc = -ENOENT);
1904
1905         LASSERT(osd_invariant(obj));
1906         LASSERT(obj->oo_sa_hdl != NULL);
1907
1908         oh = container_of0(handle, struct osd_thandle, ot_super);
1909         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1910
1911         write_lock(&obj->oo_attr_lock);
1912         nlink = --obj->oo_attr.la_nlink;
1913         write_unlock(&obj->oo_attr_lock);
1914
1915         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
1916
1917 out:
1918         up_read(&obj->oo_guard);
1919         RETURN(rc);
1920 }
1921
1922 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
1923                            __u64 start, __u64 end)
1924 {
1925         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1926         ENTRY;
1927
1928         /* XXX: no other option than syncing the whole filesystem until we
1929          * support ZIL.  If the object tracked the txg that it was last
1930          * modified in, it could pass that txg here instead of "0".  Maybe
1931          * the changes are already committed, so no wait is needed at all? */
1932         if (!osd->od_dt_dev.dd_rdonly) {
1933                 if (osd_object_sync_delay_us < 0)
1934                         txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
1935                 else
1936                         udelay(osd_object_sync_delay_us);
1937         }
1938
1939         RETURN(0);
1940 }
1941
1942 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
1943 {
1944         return 0;
1945 }
1946
1947 static struct dt_object_operations osd_obj_ops = {
1948         .do_read_lock           = osd_read_lock,
1949         .do_write_lock          = osd_write_lock,
1950         .do_read_unlock         = osd_read_unlock,
1951         .do_write_unlock        = osd_write_unlock,
1952         .do_write_locked        = osd_write_locked,
1953         .do_attr_get            = osd_attr_get,
1954         .do_declare_attr_set    = osd_declare_attr_set,
1955         .do_attr_set            = osd_attr_set,
1956         .do_ah_init             = osd_ah_init,
1957         .do_declare_create      = osd_declare_create,
1958         .do_create              = osd_create,
1959         .do_declare_destroy     = osd_declare_destroy,
1960         .do_destroy             = osd_destroy,
1961         .do_index_try           = osd_index_try,
1962         .do_declare_ref_add     = osd_declare_ref_add,
1963         .do_ref_add             = osd_ref_add,
1964         .do_declare_ref_del     = osd_declare_ref_del,
1965         .do_ref_del             = osd_ref_del,
1966         .do_xattr_get           = osd_xattr_get,
1967         .do_declare_xattr_set   = osd_declare_xattr_set,
1968         .do_xattr_set           = osd_xattr_set,
1969         .do_declare_xattr_del   = osd_declare_xattr_del,
1970         .do_xattr_del           = osd_xattr_del,
1971         .do_xattr_list          = osd_xattr_list,
1972         .do_object_sync         = osd_object_sync,
1973         .do_invalidate          = osd_invalidate,
1974 };
1975
1976 static struct lu_object_operations osd_lu_obj_ops = {
1977         .loo_object_init        = osd_object_init,
1978         .loo_object_delete      = osd_object_delete,
1979         .loo_object_release     = osd_object_release,
1980         .loo_object_free        = osd_object_free,
1981         .loo_object_print       = osd_object_print,
1982         .loo_object_invariant   = osd_object_invariant,
1983 };
1984
1985 static int osd_otable_it_attr_get(const struct lu_env *env,
1986                                 struct dt_object *dt,
1987                                 struct lu_attr *attr)
1988 {
1989         attr->la_valid = 0;
1990         return 0;
1991 }
1992
1993 static struct dt_object_operations osd_obj_otable_it_ops = {
1994         .do_attr_get            = osd_otable_it_attr_get,
1995         .do_index_try           = osd_index_try,
1996 };
1997
1998 module_param(osd_object_sync_delay_us, int, 0644);
1999 MODULE_PARM_DESC(osd_object_sync_delay_us,
2000                  "If zero or larger delay N usec instead of doing object sync");