Whamcloud - gitweb
c41a42fc77b366f2625642bb5bc016996f38a819
[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 static struct dt_object_operations osd_obj_otable_it_ops;
70
71 static void
72 osd_object_sa_fini(struct osd_object *obj)
73 {
74         if (obj->oo_sa_hdl) {
75                 sa_handle_destroy(obj->oo_sa_hdl);
76                 obj->oo_sa_hdl = NULL;
77         }
78 }
79
80 static int
81 osd_object_sa_init(struct osd_object *obj, struct osd_device *o)
82 {
83         int rc;
84
85         LASSERT(obj->oo_sa_hdl == NULL);
86         LASSERT(obj->oo_dn != NULL);
87
88         rc = osd_sa_handle_get(obj);
89         if (rc)
90                 return rc;
91
92         /* Cache the xattr object id, valid for the life of the object */
93         rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_XATTR(o), &obj->oo_xattr, 8);
94         if (rc == -ENOENT) {
95                 obj->oo_xattr = ZFS_NO_OBJECT;
96                 rc = 0;
97         } else if (rc) {
98                 osd_object_sa_fini(obj);
99         }
100
101         return rc;
102 }
103
104 /*
105  * Add object to list of dirty objects in tx handle.
106  */
107 void osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
108 {
109         if (!list_empty(&obj->oo_sa_linkage))
110                 return;
111
112         write_lock(&obj->oo_attr_lock);
113         if (likely(list_empty(&obj->oo_sa_linkage)))
114                 list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
115         write_unlock(&obj->oo_attr_lock);
116 }
117
118 /*
119  * Release spill block dbuf hold for all dirty SAs.
120  */
121 void osd_object_sa_dirty_rele(const struct lu_env *env, struct osd_thandle *oh)
122 {
123         struct osd_object *obj;
124
125         while (!list_empty(&oh->ot_sa_list)) {
126                 obj = list_entry(oh->ot_sa_list.next,
127                                  struct osd_object, oo_sa_linkage);
128                 write_lock(&obj->oo_attr_lock);
129                 list_del_init(&obj->oo_sa_linkage);
130                 write_unlock(&obj->oo_attr_lock);
131                 if (obj->oo_late_xattr) {
132                         /*
133                          * take oo_guard to protect oo_sa_xattr buffer
134                          * from concurrent update by osd_xattr_set()
135                          */
136                         LASSERT(oh->ot_assigned != 0);
137                         down_write(&obj->oo_guard);
138                         if (obj->oo_late_attr_set)
139                                 __osd_sa_attr_init(env, obj, oh);
140                         else if (obj->oo_late_xattr)
141                                 __osd_sa_xattr_update(env, obj, oh);
142                         up_write(&obj->oo_guard);
143                 }
144                 sa_spill_rele(obj->oo_sa_hdl);
145         }
146 }
147
148 /*
149  * Update the SA and add the object to the dirty list.
150  */
151 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
152                          void *buf, uint32_t buflen, struct osd_thandle *oh)
153 {
154         int rc;
155
156         LASSERT(obj->oo_sa_hdl != NULL);
157         LASSERT(oh->ot_tx != NULL);
158
159         rc = -sa_update(obj->oo_sa_hdl, type, buf, buflen, oh->ot_tx);
160         osd_object_sa_dirty_add(obj, oh);
161
162         return rc;
163 }
164
165 /*
166  * Bulk update the SA and add the object to the dirty list.
167  */
168 static int
169 osd_object_sa_bulk_update(struct osd_object *obj, sa_bulk_attr_t *attrs,
170                           int count, struct osd_thandle *oh)
171 {
172         int rc;
173
174         LASSERT(obj->oo_sa_hdl != NULL);
175         LASSERT(oh->ot_tx != NULL);
176
177         rc = -sa_bulk_update(obj->oo_sa_hdl, attrs, count, oh->ot_tx);
178         osd_object_sa_dirty_add(obj, oh);
179
180         return rc;
181 }
182
183 /*
184  * Retrieve the attributes of a DMU object
185  */
186 static int __osd_object_attr_get(const struct lu_env *env, struct osd_device *o,
187                                  struct osd_object *obj, struct lu_attr *la)
188 {
189         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
190         sa_bulk_attr_t  *bulk = osd_oti_get(env)->oti_attr_bulk;
191         int              cnt = 0;
192         int              rc;
193         ENTRY;
194
195         LASSERT(obj->oo_dn != NULL);
196
197         la->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE | LA_TYPE |
198                         LA_SIZE | LA_UID | LA_GID | LA_FLAGS | LA_NLINK;
199
200         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(o), NULL, osa->atime, 16);
201         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(o), NULL, osa->mtime, 16);
202         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(o), NULL, osa->ctime, 16);
203         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(o), NULL, &osa->mode, 8);
204         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(o), NULL, &osa->size, 8);
205         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(o), NULL, &osa->nlink, 8);
206         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(o), NULL, &osa->uid, 8);
207         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(o), NULL, &osa->gid, 8);
208         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(o), NULL, &osa->flags, 8);
209         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
210
211         rc = -sa_bulk_lookup(obj->oo_sa_hdl, bulk, cnt);
212         if (rc)
213                 GOTO(out_sa, rc);
214
215 #ifdef ZFS_PROJINHERIT
216         if (o->od_projectused_dn && osa->flags & ZFS_PROJID) {
217                 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_PROJID(o),
218                                 &osa->projid, 8);
219                 if (rc)
220                         GOTO(out_sa, rc);
221
222                 la->la_projid = osa->projid;
223                 la->la_valid |= LA_PROJID;
224                 obj->oo_with_projid = 1;
225         } else {
226                 la->la_projid = ZFS_DEFAULT_PROJID;
227                 la->la_valid &= ~LA_PROJID;
228         }
229 #else
230         la->la_projid = 0;
231         la->la_valid &= ~LA_PROJID;
232 #endif
233
234         la->la_atime = osa->atime[0];
235         la->la_mtime = osa->mtime[0];
236         la->la_ctime = osa->ctime[0];
237         la->la_mode = osa->mode;
238         la->la_uid = osa->uid;
239         la->la_gid = osa->gid;
240         la->la_nlink = osa->nlink;
241         la->la_flags = attrs_zfs2fs(osa->flags);
242         la->la_size = osa->size;
243
244         /* Try to get extra flag from LMA. Right now, only LMAI_ORPHAN
245          * flags is stored in LMA, and it is only for orphan directory */
246         if (S_ISDIR(la->la_mode) && dt_object_exists(&obj->oo_dt)) {
247                 struct osd_thread_info *info = osd_oti_get(env);
248                 struct lustre_mdt_attrs *lma;
249                 struct lu_buf buf;
250
251                 lma = (struct lustre_mdt_attrs *)info->oti_buf;
252                 buf.lb_buf = lma;
253                 buf.lb_len = sizeof(info->oti_buf);
254                 rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
255                 if (rc > 0) {
256                         rc = 0;
257                         lma->lma_incompat = le32_to_cpu(lma->lma_incompat);
258                         obj->oo_lma_flags =
259                                 lma_to_lustre_flags(lma->lma_incompat);
260
261                 } else if (rc == -ENODATA) {
262                         rc = 0;
263                 }
264         }
265
266         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode)) {
267                 rc = -sa_lookup(obj->oo_sa_hdl, SA_ZPL_RDEV(o), &osa->rdev, 8);
268                 if (rc)
269                         GOTO(out_sa, rc);
270                 la->la_rdev = osa->rdev;
271                 la->la_valid |= LA_RDEV;
272         }
273 out_sa:
274
275         RETURN(rc);
276 }
277
278 int __osd_obj2dnode(objset_t *os, uint64_t oid, dnode_t **dnp)
279 {
280         dmu_buf_t *db;
281         dmu_buf_impl_t *dbi;
282         int rc;
283
284         rc = -dmu_bonus_hold(os, oid, osd_obj_tag, &db);
285         if (rc)
286                 return rc;
287
288         dbi = (dmu_buf_impl_t *)db;
289         DB_DNODE_ENTER(dbi);
290         *dnp = DB_DNODE(dbi);
291         DB_DNODE_EXIT(dbi);
292         LASSERT(*dnp != NULL);
293
294         return 0;
295 }
296
297 /*
298  * Concurrency: no concurrent access is possible that early in object
299  * life-cycle.
300  */
301 struct lu_object *osd_object_alloc(const struct lu_env *env,
302                                    const struct lu_object_header *hdr,
303                                    struct lu_device *d)
304 {
305         struct osd_object *mo;
306
307         OBD_SLAB_ALLOC_PTR_GFP(mo, osd_object_kmem, GFP_NOFS);
308         if (mo != NULL) {
309                 struct lu_object *l;
310                 struct lu_object_header *h;
311                 struct osd_device *o = osd_dev(d);
312
313                 l = &mo->oo_dt.do_lu;
314                 if (unlikely(o->od_in_init)) {
315                         OBD_ALLOC_PTR(h);
316                         if (!h) {
317                                 OBD_FREE_PTR(mo);
318                                 return NULL;
319                         }
320
321                         lu_object_header_init(h);
322                         lu_object_init(l, h, d);
323                         lu_object_add_top(h, l);
324                         mo->oo_header = h;
325                 } else {
326                         dt_object_init(&mo->oo_dt, NULL, d);
327                         mo->oo_header = NULL;
328                 }
329
330                 mo->oo_dt.do_ops = &osd_obj_ops;
331                 l->lo_ops = &osd_lu_obj_ops;
332                 INIT_LIST_HEAD(&mo->oo_sa_linkage);
333                 INIT_LIST_HEAD(&mo->oo_unlinked_linkage);
334                 init_rwsem(&mo->oo_sem);
335                 init_rwsem(&mo->oo_guard);
336                 rwlock_init(&mo->oo_attr_lock);
337                 mo->oo_destroy = OSD_DESTROY_NONE;
338                 return l;
339         } else {
340                 return NULL;
341         }
342 }
343
344 static void osd_obj_set_blksize(const struct lu_env *env,
345                                 struct osd_device *osd, struct osd_object *obj)
346 {
347         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
348         dmu_tx_t *tx;
349         dnode_t *dn = obj->oo_dn;
350         uint32_t blksz;
351         int rc = 0;
352         ENTRY;
353
354         LASSERT(!osd_oti_get(env)->oti_in_trans);
355
356         tx = dmu_tx_create(osd->od_os);
357         if (!tx) {
358                 CERROR("%s: fail to create tx to set blksize for "DFID"\n",
359                        osd->od_svname, PFID(fid));
360                 RETURN_EXIT;
361         }
362
363         dmu_tx_hold_bonus(tx, dn->dn_object);
364         rc = -dmu_tx_assign(tx, TXG_WAIT);
365         if (rc) {
366                 dmu_tx_abort(tx);
367                 CERROR("%s: fail to assign tx to set blksize for "DFID
368                        ": rc = %d\n", osd->od_svname, PFID(fid), rc);
369                 RETURN_EXIT;
370         }
371
372         down_write(&obj->oo_guard);
373         if (unlikely((1 << dn->dn_datablkshift) >= PAGE_SIZE))
374                 GOTO(out, rc = 1);
375
376         blksz = dn->dn_datablksz;
377         if (!is_power_of_2(blksz))
378                 blksz = size_roundup_power2(blksz);
379
380         if (blksz > osd->od_max_blksz)
381                 blksz = osd->od_max_blksz;
382         else if (blksz < PAGE_SIZE)
383                 blksz = PAGE_SIZE;
384         rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object, blksz, 0, tx);
385
386         GOTO(out, rc);
387
388 out:
389         up_write(&obj->oo_guard);
390         if (rc) {
391                 dmu_tx_abort(tx);
392                 if (unlikely(obj->oo_dn->dn_maxblkid > 0))
393                         rc = 1;
394                 if (rc < 0)
395                         CERROR("%s: fail to set blksize for "DFID": rc = %d\n",
396                                osd->od_svname, PFID(fid), rc);
397         } else {
398                 dmu_tx_commit(tx);
399                 CDEBUG(D_INODE, "%s: set blksize as %u for "DFID"\n",
400                        osd->od_svname, blksz, PFID(fid));
401         }
402 }
403
404 /*
405  * Concurrency: shouldn't matter.
406  */
407 static int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
408 {
409         struct osd_device       *osd = osd_obj2dev(obj);
410         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
411         int                      rc = 0;
412         ENTRY;
413
414         LASSERT(obj->oo_dn);
415
416         rc = osd_object_sa_init(obj, osd);
417         if (rc)
418                 RETURN(rc);
419
420         /* cache attrs in object */
421         rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
422         if (rc)
423                 RETURN(rc);
424
425         if (likely(!fid_is_acct(fid))) {
426                 /* no body operations for accounting objects */
427                 obj->oo_dt.do_body_ops = &osd_body_ops;
428
429                 if (S_ISREG(obj->oo_attr.la_mode) &&
430                     obj->oo_dn->dn_maxblkid == 0 &&
431                     (1 << obj->oo_dn->dn_datablkshift) < PAGE_SIZE &&
432                     (fid_is_idif(fid) || fid_is_norm(fid) ||
433                      fid_is_echo(fid)) &&
434                     osd->od_is_ost && !osd->od_dt_dev.dd_rdonly)
435                         osd_obj_set_blksize(env, osd, obj);
436         }
437
438         /*
439          * initialize object before marking it existing
440          */
441         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
442
443         smp_mb();
444         obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
445
446         RETURN(0);
447 }
448
449 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
450 {
451         struct osd_thread_info  *info = osd_oti_get(env);
452         struct lu_buf           buf;
453         int                     rc;
454         struct lustre_mdt_attrs *lma;
455         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
456         ENTRY;
457
458         BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*lma));
459         lma = (struct lustre_mdt_attrs *)info->oti_buf;
460         buf.lb_buf = lma;
461         buf.lb_len = sizeof(info->oti_buf);
462
463         rc = osd_xattr_get(env, &obj->oo_dt, &buf, XATTR_NAME_LMA);
464         if (rc > 0) {
465                 rc = 0;
466                 lustre_lma_swab(lma);
467                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
468                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
469                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
470                               "fid = "DFID"\n", osd_obj2dev(obj)->od_svname,
471                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
472                               PFID(rfid));
473                         rc = -EOPNOTSUPP;
474                 } else if (unlikely(!lu_fid_eq(rfid, &lma->lma_self_fid))) {
475                         CERROR("%s: FID-in-LMA "DFID" does not match the "
476                               "object self-fid "DFID"\n",
477                               osd_obj2dev(obj)->od_svname,
478                               PFID(&lma->lma_self_fid), PFID(rfid));
479                         rc = -EREMCHG;
480                 } else {
481                         struct osd_device *osd = osd_obj2dev(obj);
482
483                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
484                             osd->od_is_ost)
485                                 obj->oo_pfid_in_lma = 1;
486                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
487                             osd->od_remote_parent_dir != ZFS_NO_OBJECT)
488                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
489                 }
490         } else if (rc == -ENODATA) {
491                 /* haven't initialize LMA xattr */
492                 rc = 0;
493         }
494
495         RETURN(rc);
496 }
497
498 /**
499  * Helper function to retrieve DMU object id from fid for accounting object
500  */
501 static dnode_t *osd_quota_fid2dmu(const struct osd_device *osd,
502                                   const struct lu_fid *fid)
503 {
504         dnode_t *dn = NULL;
505
506         LASSERT(fid_is_acct(fid));
507
508         switch (fid_oid(fid)) {
509         case ACCT_USER_OID:
510                 dn = osd->od_userused_dn;
511                 break;
512         case ACCT_GROUP_OID:
513                 dn = osd->od_groupused_dn;
514                 break;
515 #ifdef ZFS_PROJINHERIT
516         case ACCT_PROJECT_OID:
517                 dn = osd->od_projectused_dn;
518                 break;
519 #endif
520         default:
521                 break;
522         }
523
524         return dn;
525 }
526
527 /*
528  * Concurrency: no concurrent access is possible that early in object
529  * life-cycle.
530  */
531 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
532                            const struct lu_object_conf *conf)
533 {
534         struct osd_object *obj = osd_obj(l);
535         struct osd_device *osd = osd_obj2dev(obj);
536         const struct lu_fid *fid = lu_object_fid(l);
537         struct lustre_scrub *scrub = &osd->od_scrub;
538         struct osd_thread_info *info = osd_oti_get(env);
539         struct luz_direntry *zde = &info->oti_zde;
540         struct osd_idmap_cache *idc;
541         char *name = info->oti_str;
542         uint64_t oid;
543         int rc = 0;
544         int rc1;
545         bool remote = false;
546         ENTRY;
547
548         LASSERT(osd_invariant(obj));
549
550         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
551                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
552                 l->lo_header->loh_attr |= LOHA_EXISTS;
553
554                 GOTO(out, rc = 0);
555         }
556
557         if (conf && conf->loc_flags & LOC_F_NEW)
558                 GOTO(out, rc = 0);
559
560         if (unlikely(fid_is_acct(fid))) {
561                 obj->oo_dn = osd_quota_fid2dmu(osd, fid);
562                 if (obj->oo_dn) {
563                         obj->oo_dt.do_index_ops = &osd_acct_index_ops;
564                         l->lo_header->loh_attr |= LOHA_EXISTS;
565                 }
566
567                 GOTO(out, rc = 0);
568         }
569
570         idc = osd_idc_find(env, osd, fid);
571         if (idc && !idc->oic_remote && idc->oic_dnode != ZFS_NO_OBJECT) {
572                 oid = idc->oic_dnode;
573                 goto zget;
574         }
575
576         rc = -ENOENT;
577         if (!list_empty(&osd->od_scrub.os_inconsistent_items))
578                 rc = osd_oii_lookup(osd, fid, &oid);
579
580         if (rc)
581                 rc = osd_fid_lookup(env, osd, fid, &oid);
582
583         if (rc == -ENOENT) {
584                 if (likely(!(fid_is_norm(fid) || fid_is_igif(fid)) ||
585                            fid_is_on_ost(env, osd, fid) ||
586                            !zfs_test_bit(osd_oi_fid2idx(osd, fid),
587                                          scrub->os_file.sf_oi_bitmap)))
588                         GOTO(out, rc = 0);
589
590                 rc = -EREMCHG;
591                 goto trigger;
592         }
593
594         if (rc)
595                 GOTO(out, rc);
596
597 zget:
598         LASSERT(obj->oo_dn == NULL);
599
600         rc = __osd_obj2dnode(osd->od_os, oid, &obj->oo_dn);
601         /* EEXIST will be returned if object is being deleted in ZFS */
602         if (rc == -EEXIST)
603                 GOTO(out, rc = 0);
604
605         if (rc) {
606                 CERROR("%s: lookup "DFID"/%#llx failed: rc = %d\n",
607                        osd->od_svname, PFID(lu_object_fid(l)), oid, rc);
608                 GOTO(out, rc);
609         }
610
611         rc = osd_object_init0(env, obj);
612         if (rc)
613                 GOTO(out, rc);
614
615         if (unlikely(obj->oo_header))
616                 GOTO(out, rc = 0);
617
618         rc = osd_check_lma(env, obj);
619         if ((!rc && !remote) || (rc != -EREMCHG))
620                 GOTO(out, rc);
621
622 trigger:
623         /* We still have chance to get the valid dnode: for the object that is
624          * referenced by remote name entry, the object on the local MDT will be
625          * linked under the dir /REMOTE_PARENT_DIR with its FID string as name.
626          *
627          * During the OI scrub, if we cannot find the OI mapping, we may still
628          * have change to map the FID to local OID via lookup the dir
629          * /REMOTE_PARENT_DIR. */
630         if (!remote && !fid_is_on_ost(env, osd, fid)) {
631                 osd_fid2str(name, fid, sizeof(info->oti_str));
632                 rc = osd_zap_lookup(osd, osd->od_remote_parent_dir,
633                                     NULL, name, 8, 3, (void *)zde);
634                 if (!rc) {
635                         oid = zde->lzd_reg.zde_dnode;
636                         osd_dnode_rele(obj->oo_dn);
637                         obj->oo_dn = NULL;
638                         remote = true;
639                         goto zget;
640                 }
641         }
642
643         /* The case someone triggered the OI scrub already. */
644         if (thread_is_running(&scrub->os_thread)) {
645                 if (!rc) {
646                         LASSERT(remote);
647
648                         lu_object_set_agent_entry(l);
649                         osd_oii_insert(env, osd, fid, oid, false);
650                 } else {
651                         rc = -EINPROGRESS;
652                 }
653
654                 GOTO(out, rc);
655         }
656
657         /* The case NOT allow to trigger OI scrub automatically. */
658         if (osd->od_auto_scrub_interval == AS_NEVER)
659                 GOTO(out, rc);
660
661         /* It is me to trigger the OI scrub. */
662         rc1 = osd_scrub_start(env, osd, SS_CLEAR_DRYRUN |
663                               SS_CLEAR_FAILOUT | SS_AUTO_FULL);
664         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the "DFID": rc = %d\n",
665                       osd_name(osd), PFID(fid), rc1);
666         if (!rc) {
667                 LASSERT(remote);
668
669                 lu_object_set_agent_entry(l);
670                 if (!rc1)
671                         osd_oii_insert(env, osd, fid, oid, false);
672         } else {
673                 if (!rc1)
674                         rc = -EINPROGRESS;
675                 else
676                         rc = -EREMCHG;
677         }
678
679         GOTO(out, rc);
680
681 out:
682         RETURN(rc);
683 }
684
685 /*
686  * Concurrency: no concurrent access is possible that late in object
687  * life-cycle.
688  */
689 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
690 {
691         struct osd_object *obj = osd_obj(l);
692         struct lu_object_header *h = obj->oo_header;
693
694         LASSERT(osd_invariant(obj));
695
696         dt_object_fini(&obj->oo_dt);
697         /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
698         OBD_SLAB_FREE_PTR(obj, osd_object_kmem);
699         if (unlikely(h)) {
700                 lu_object_header_fini(h);
701                 OBD_FREE_PRE(h, sizeof(*h), "kfreed");
702                 kfree_rcu(h, loh_rcu);
703         }
704 }
705
706 static int
707 osd_object_unlinked_add(struct osd_object *obj, struct osd_thandle *oh)
708 {
709         int rc = -EBUSY;
710
711         LASSERT(obj->oo_destroy == OSD_DESTROY_ASYNC);
712
713         /* the object is supposed to be exclusively locked by
714          * the caller (osd_destroy()), while the transaction
715          * (oh) is per-thread and not shared */
716         if (likely(list_empty(&obj->oo_unlinked_linkage))) {
717                 list_add(&obj->oo_unlinked_linkage, &oh->ot_unlinked_list);
718                 rc = 0;
719         }
720
721         return rc;
722 }
723
724 /* Default to max data size covered by a level-1 indirect block */
725 static unsigned long osd_sync_destroy_max_size =
726         1UL << (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT + SPA_MAXBLOCKSHIFT);
727 module_param(osd_sync_destroy_max_size, ulong, 0444);
728 MODULE_PARM_DESC(osd_sync_destroy_max_size, "Maximum object size to use synchronous destroy.");
729
730 static inline void
731 osd_object_set_destroy_type(struct osd_object *obj)
732 {
733         /*
734          * Lock-less OST_WRITE can race with OST_DESTROY, so set destroy type
735          * only once and use it consistently thereafter.
736          */
737         down_write(&obj->oo_guard);
738         if (obj->oo_destroy == OSD_DESTROY_NONE) {
739                 if (obj->oo_attr.la_size <= osd_sync_destroy_max_size)
740                         obj->oo_destroy = OSD_DESTROY_SYNC;
741                 else /* Larger objects are destroyed asynchronously */
742                         obj->oo_destroy = OSD_DESTROY_ASYNC;
743         }
744         up_write(&obj->oo_guard);
745 }
746
747 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
748                                struct thandle *th)
749 {
750         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
751         struct osd_object       *obj = osd_dt_obj(dt);
752         struct osd_device       *osd = osd_obj2dev(obj);
753         struct osd_thandle      *oh;
754         dnode_t *dn;
755         int                      rc;
756         uint64_t                 zapid;
757         ENTRY;
758
759         LASSERT(th != NULL);
760         LASSERT(dt_object_exists(dt));
761
762         oh = container_of0(th, struct osd_thandle, ot_super);
763         LASSERT(oh->ot_tx != NULL);
764
765         dmu_tx_mark_netfree(oh->ot_tx);
766
767         /* declare that we'll remove object from fid-dnode mapping */
768         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
769         osd_tx_hold_zap(oh->ot_tx, zapid, dn, FALSE, NULL);
770
771         osd_declare_xattrs_destroy(env, obj, oh);
772
773         /* one less inode */
774         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
775                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
776                                -1, oh, NULL, OSD_QID_INODE);
777         if (rc)
778                 RETURN(rc);
779
780         /* data to be truncated */
781         rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
782                                obj->oo_attr.la_gid, obj->oo_attr.la_projid,
783                                0, oh, NULL, OSD_QID_BLK);
784         if (rc)
785                 RETURN(rc);
786
787         osd_object_set_destroy_type(obj);
788         if (obj->oo_destroy == OSD_DESTROY_SYNC)
789                 dmu_tx_hold_free(oh->ot_tx, obj->oo_dn->dn_object,
790                                  0, DMU_OBJECT_END);
791         else
792                 osd_tx_hold_zap(oh->ot_tx, osd->od_unlinked->dn_object,
793                                 osd->od_unlinked, TRUE, NULL);
794
795         /* remove agent entry (if have) from remote parent */
796         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
797                 osd_tx_hold_zap(oh->ot_tx, osd->od_remote_parent_dir,
798                                 NULL, FALSE, NULL);
799
800         /* will help to find FID->ino when this object is being
801          * added to PENDING/ */
802         osd_idc_find_and_init(env, osd, obj);
803
804         RETURN(0);
805 }
806
807 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
808                        struct thandle *th)
809 {
810         struct osd_thread_info  *info = osd_oti_get(env);
811         char                    *buf = info->oti_str;
812         struct osd_object       *obj = osd_dt_obj(dt);
813         struct osd_device       *osd = osd_obj2dev(obj);
814         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
815         struct osd_thandle      *oh;
816         int                      rc;
817         uint64_t                 oid, zapid;
818         dnode_t *zdn;
819         ENTRY;
820
821         down_write(&obj->oo_guard);
822
823         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
824                 GOTO(out, rc = -ENOENT);
825
826         LASSERT(obj->oo_dn != NULL);
827
828         oh = container_of0(th, struct osd_thandle, ot_super);
829         LASSERT(oh != NULL);
830         LASSERT(oh->ot_tx != NULL);
831
832         /* remove obj ref from index dir (it depends) */
833         zapid = osd_get_name_n_idx(env, osd, fid, buf,
834                                    sizeof(info->oti_str), &zdn);
835         rc = osd_xattrs_destroy(env, obj, oh);
836         if (rc) {
837                 CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
838                        osd->od_svname, buf, rc);
839                 GOTO(out, rc);
840         }
841
842         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
843                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, true);
844                 if (rc)
845                         GOTO(out, rc);
846         }
847
848         oid = obj->oo_dn->dn_object;
849         if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
850                 /* this may happen if the destroy wasn't declared
851                  * e.g. when the object is created and then destroyed
852                  * in the same transaction - we don't need additional
853                  * space for destroy specifically */
854                 LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
855                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
856                 if (rc)
857                         CERROR("%s: failed to free %s %llu: rc = %d\n",
858                                osd->od_svname, buf, oid, rc);
859         } else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
860                 rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
861                 if (rc)
862                         CERROR("%s: failed to free %s %llu: rc = %d\n",
863                                osd->od_svname, buf, oid, rc);
864         } else { /* asynchronous destroy */
865                 char *key = info->oti_key;
866
867                 rc = osd_object_unlinked_add(obj, oh);
868                 if (rc)
869                         GOTO(out, rc);
870
871                 snprintf(key, sizeof(info->oti_key), "%llx", oid);
872                 rc = osd_zap_add(osd, osd->od_unlinked->dn_object,
873                                  osd->od_unlinked, key, 8, 1, &oid, oh->ot_tx);
874                 if (rc)
875                         CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
876                                osd->od_svname, buf, oid, rc);
877         }
878
879         /* Remove the OI mapping after the destroy to handle the race with
880          * OI scrub that may insert missed OI mapping during the interval. */
881         rc = osd_zap_remove(osd, zapid, zdn, buf, oh->ot_tx);
882         if (unlikely(rc == -ENOENT))
883                 rc = 0;
884         if (rc)
885                 CERROR("%s: zap_remove(%s) failed: rc = %d\n",
886                        osd->od_svname, buf, rc);
887
888         GOTO(out, rc);
889
890 out:
891         /* not needed in the cache anymore */
892         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
893         if (rc == 0)
894                 obj->oo_destroyed = 1;
895         up_write(&obj->oo_guard);
896         RETURN (0);
897 }
898
899 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
900 {
901         struct osd_object *obj = osd_obj(l);
902         const struct lu_fid *fid = lu_object_fid(l);
903
904         if (obj->oo_dn) {
905                 if (likely(!fid_is_acct(fid))) {
906                         osd_object_sa_fini(obj);
907                         if (obj->oo_sa_xattr) {
908                                 nvlist_free(obj->oo_sa_xattr);
909                                 obj->oo_sa_xattr = NULL;
910                         }
911                         osd_dnode_rele(obj->oo_dn);
912                         list_del(&obj->oo_sa_linkage);
913                 }
914                 obj->oo_dn = NULL;
915         }
916 }
917
918 /*
919  * Concurrency: ->loo_object_release() is called under site spin-lock.
920  */
921 static void osd_object_release(const struct lu_env *env,
922                                struct lu_object *l)
923 {
924 }
925
926 /*
927  * Concurrency: shouldn't matter.
928  */
929 static int osd_object_print(const struct lu_env *env, void *cookie,
930                             lu_printer_t p, const struct lu_object *l)
931 {
932         struct osd_object *o = osd_obj(l);
933
934         return (*p)(env, cookie, LUSTRE_OSD_ZFS_NAME"-object@%p", o);
935 }
936
937 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
938                           unsigned role)
939 {
940         struct osd_object *obj = osd_dt_obj(dt);
941
942         LASSERT(osd_invariant(obj));
943
944         down_read_nested(&obj->oo_sem, role);
945 }
946
947 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
948                            unsigned role)
949 {
950         struct osd_object *obj = osd_dt_obj(dt);
951
952         LASSERT(osd_invariant(obj));
953
954         down_write_nested(&obj->oo_sem, role);
955 }
956
957 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
958 {
959         struct osd_object *obj = osd_dt_obj(dt);
960
961         LASSERT(osd_invariant(obj));
962         up_read(&obj->oo_sem);
963 }
964
965 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
966 {
967         struct osd_object *obj = osd_dt_obj(dt);
968
969         LASSERT(osd_invariant(obj));
970         up_write(&obj->oo_sem);
971 }
972
973 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
974 {
975         struct osd_object *obj = osd_dt_obj(dt);
976         int rc = 1;
977
978         LASSERT(osd_invariant(obj));
979
980         if (down_write_trylock(&obj->oo_sem)) {
981                 rc = 0;
982                 up_write(&obj->oo_sem);
983         }
984         return rc;
985 }
986
987 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
988                         struct lu_attr *attr)
989 {
990         struct osd_object       *obj = osd_dt_obj(dt);
991         uint64_t                 blocks;
992         uint32_t                 blksize;
993         int                      rc = 0;
994
995         down_read(&obj->oo_guard);
996
997         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
998                 GOTO(out, rc = -ENOENT);
999
1000         if (unlikely(fid_is_acct(lu_object_fid(&dt->do_lu))))
1001                 GOTO(out, rc = 0);
1002
1003         LASSERT(osd_invariant(obj));
1004         LASSERT(obj->oo_dn);
1005
1006         read_lock(&obj->oo_attr_lock);
1007         *attr = obj->oo_attr;
1008         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
1009                 attr->la_valid |= LA_FLAGS;
1010                 attr->la_flags |= LUSTRE_ORPHAN_FL;
1011         }
1012         read_unlock(&obj->oo_attr_lock);
1013         if (attr->la_valid & LA_FLAGS && attr->la_flags & LUSTRE_ORPHAN_FL)
1014                 CDEBUG(D_INFO, "%s: set orphan flag on "DFID" (%llx/%x)\n",
1015                        osd_obj2dev(obj)->od_svname,
1016                        PFID(lu_object_fid(&dt->do_lu)),
1017                        attr->la_valid, obj->oo_lma_flags);
1018
1019         /* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
1020          * from within sa_object_size() can block on a mutex, so
1021          * we can't call sa_object_size() holding rwlock */
1022         sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
1023         /* we do not control size of indices, so always calculate
1024          * it from number of blocks reported by DMU */
1025         if (S_ISDIR(attr->la_mode))
1026                 attr->la_size = 512 * blocks;
1027         /* Block size may be not set; suggest maximal I/O transfers. */
1028         if (blksize == 0)
1029                 blksize = osd_spa_maxblocksize(
1030                         dmu_objset_spa(osd_obj2dev(obj)->od_os));
1031
1032         attr->la_blksize = blksize;
1033         attr->la_blocks = blocks;
1034         attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;
1035
1036 out:
1037         up_read(&obj->oo_guard);
1038         return rc;
1039 }
1040
1041 /* Simple wrapper on top of qsd API which implement quota transfer for osd
1042  * setattr needs. As a reminder, only the root user can change ownership of
1043  * a file, that's why EDQUOT & EINPROGRESS errors are discarded */
1044 static inline int qsd_transfer(const struct lu_env *env,
1045                                struct qsd_instance *qsd,
1046                                struct lquota_trans *trans, int qtype,
1047                                __u64 orig_id, __u64 new_id, __u64 bspace,
1048                                struct lquota_id_info *qi, bool ignore_edquot)
1049 {
1050         int     rc;
1051
1052         if (unlikely(qsd == NULL))
1053                 return 0;
1054
1055         LASSERT(qtype >= 0 && qtype < LL_MAXQUOTAS);
1056         qi->lqi_type = qtype;
1057
1058         /* inode accounting */
1059         qi->lqi_is_blk = false;
1060
1061         /* one more inode for the new owner ... */
1062         qi->lqi_id.qid_uid = new_id;
1063         qi->lqi_space      = 1;
1064         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1065         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1066                 rc = 0;
1067         if (rc)
1068                 return rc;
1069
1070         /* and one less inode for the current id */
1071         qi->lqi_id.qid_uid = orig_id;;
1072         qi->lqi_space      = -1;
1073         /* can't get EDQUOT when reducing usage */
1074         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1075         if (rc == -EINPROGRESS)
1076                 rc = 0;
1077         if (rc)
1078                 return rc;
1079
1080         /* block accounting */
1081         qi->lqi_is_blk = true;
1082
1083         /* more blocks for the new owner ... */
1084         qi->lqi_id.qid_uid = new_id;
1085         qi->lqi_space      = bspace;
1086         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1087         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
1088                 rc = 0;
1089         if (rc)
1090                 return rc;
1091
1092         /* and finally less blocks for the current owner */
1093         qi->lqi_id.qid_uid = orig_id;
1094         qi->lqi_space      = -bspace;
1095         rc = qsd_op_begin(env, qsd, trans, qi, NULL);
1096         /* can't get EDQUOT when reducing usage */
1097         if (rc == -EINPROGRESS)
1098                 rc = 0;
1099         return rc;
1100 }
1101
1102 static int osd_declare_attr_set(const struct lu_env *env,
1103                                 struct dt_object *dt,
1104                                 const struct lu_attr *attr,
1105                                 struct thandle *handle)
1106 {
1107         struct osd_thread_info  *info = osd_oti_get(env);
1108         struct osd_object       *obj = osd_dt_obj(dt);
1109         struct osd_device       *osd = osd_obj2dev(obj);
1110         dmu_tx_hold_t           *txh;
1111         struct osd_thandle      *oh;
1112         uint64_t                 bspace;
1113         uint32_t                 blksize;
1114         int                      rc = 0;
1115         bool                     found;
1116         ENTRY;
1117
1118
1119         LASSERT(handle != NULL);
1120         LASSERT(osd_invariant(obj));
1121
1122         oh = container_of0(handle, struct osd_thandle, ot_super);
1123
1124         down_read(&obj->oo_guard);
1125         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1126                 GOTO(out, rc = 0);
1127
1128         LASSERT(obj->oo_sa_hdl != NULL);
1129         LASSERT(oh->ot_tx != NULL);
1130         /* regular attributes are part of the bonus buffer */
1131         /* let's check whether this object is already part of
1132          * transaction.. */
1133         found = false;
1134         for (txh = list_head(&oh->ot_tx->tx_holds); txh;
1135              txh = list_next(&oh->ot_tx->tx_holds, txh)) {
1136                 if (txh->txh_dnode == NULL)
1137                         continue;
1138                 if (txh->txh_dnode->dn_object != obj->oo_dn->dn_object)
1139                         continue;
1140                 /* this object is part of the transaction already
1141                  * we don't need to declare bonus again */
1142                 found = true;
1143                 break;
1144         }
1145         if (!found)
1146                 dmu_tx_hold_bonus(oh->ot_tx, obj->oo_dn->dn_object);
1147         if (oh->ot_tx->tx_err != 0)
1148                 GOTO(out, rc = -oh->ot_tx->tx_err);
1149
1150         if (attr && attr->la_valid & LA_FLAGS) {
1151                 /* LMA is usually a part of bonus, no need to declare
1152                  * anything else */
1153         }
1154
1155         if (attr && (attr->la_valid & (LA_UID | LA_GID | LA_PROJID))) {
1156                 sa_object_size(obj->oo_sa_hdl, &blksize, &bspace);
1157                 bspace = toqb(bspace * 512);
1158
1159                 CDEBUG(D_QUOTA,
1160                        "%s: enforce quota on UID %u, GID %u, the quota space is %lld (%u)\n",
1161                        osd->od_svname,
1162                        attr->la_uid, attr->la_gid, bspace, blksize);
1163         }
1164
1165         if (attr && attr->la_valid & LA_UID) {
1166                 /* quota enforcement for user */
1167                 if (attr->la_uid != obj->oo_attr.la_uid) {
1168                         rc = qsd_transfer(env, osd_def_qsd(osd),
1169                                           &oh->ot_quota_trans, USRQUOTA,
1170                                           obj->oo_attr.la_uid, attr->la_uid,
1171                                           bspace, &info->oti_qi, true);
1172                         if (rc)
1173                                 GOTO(out, rc);
1174                 }
1175         }
1176         if (attr && attr->la_valid & LA_GID) {
1177                 /* quota enforcement for group */
1178                 if (attr->la_gid != obj->oo_attr.la_gid) {
1179                         rc = qsd_transfer(env, osd_def_qsd(osd),
1180                                           &oh->ot_quota_trans, GRPQUOTA,
1181                                           obj->oo_attr.la_gid, attr->la_gid,
1182                                           bspace, &info->oti_qi,
1183                                           !(attr->la_flags &
1184                                                         LUSTRE_SET_SYNC_FL));
1185                         if (rc)
1186                                 GOTO(out, rc);
1187                 }
1188         }
1189 #ifdef ZFS_PROJINHERIT
1190         if (attr && attr->la_valid & LA_PROJID) {
1191                 /* quota enforcement for project */
1192                 if (attr->la_projid != obj->oo_attr.la_projid) {
1193                         if (!osd->od_projectused_dn)
1194                                 GOTO(out, rc = -EOPNOTSUPP);
1195
1196                         /* Usually, if project quota is upgradable for the
1197                          * device, then the upgrade will be done before or when
1198                          * mount the device. So when we come here, this project
1199                          * should have project ID attribute already (that is
1200                          * zero by default).  Otherwise, there was something
1201                          * wrong during the former upgrade, let's return failure
1202                          * to report that.
1203                          *
1204                          * Please note that, different from other attributes,
1205                          * you can NOT simply set the project ID attribute under
1206                          * such case, because adding (NOT change) project ID
1207                          * attribute needs to change the object's attribute
1208                          * layout to match zfs backend quota accounting
1209                          * requirement. */
1210                         if (unlikely(!obj->oo_with_projid))
1211                                 GOTO(out, rc = -ENXIO);
1212
1213                         rc = qsd_transfer(env, osd_def_qsd(osd),
1214                                           &oh->ot_quota_trans, PRJQUOTA,
1215                                           obj->oo_attr.la_projid,
1216                                           attr->la_projid, bspace,
1217                                           &info->oti_qi, true);
1218                         if (rc)
1219                                 GOTO(out, rc);
1220                 }
1221         }
1222 #endif
1223 out:
1224         up_read(&obj->oo_guard);
1225         RETURN(rc);
1226 }
1227
1228 /*
1229  * Set the attributes of an object
1230  *
1231  * The transaction passed to this routine must have
1232  * dmu_tx_hold_bonus(tx, oid) called and then assigned
1233  * to a transaction group.
1234  */
1235 static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
1236                         const struct lu_attr *la, struct thandle *handle)
1237 {
1238         struct osd_thread_info  *info = osd_oti_get(env);
1239         sa_bulk_attr_t          *bulk = osd_oti_get(env)->oti_attr_bulk;
1240         struct osd_object       *obj = osd_dt_obj(dt);
1241         struct osd_device       *osd = osd_obj2dev(obj);
1242         struct osd_thandle      *oh;
1243         struct osa_attr         *osa = &info->oti_osa;
1244         __u64                    valid = la->la_valid;
1245         int                      cnt;
1246         int                      rc = 0;
1247
1248         ENTRY;
1249
1250         down_read(&obj->oo_guard);
1251         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
1252                 GOTO(out, rc = -ENOENT);
1253
1254         LASSERT(handle != NULL);
1255         LASSERT(osd_invariant(obj));
1256         LASSERT(obj->oo_sa_hdl);
1257
1258         oh = container_of0(handle, struct osd_thandle, ot_super);
1259         /* Assert that the transaction has been assigned to a
1260            transaction group. */
1261         LASSERT(oh->ot_tx->tx_txg != 0);
1262
1263         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) && !osd->od_is_ost) {
1264                 struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
1265                 char *buf = info->oti_str;
1266                 dnode_t *zdn = NULL;
1267                 uint64_t zapid;
1268
1269                 zapid = osd_get_name_n_idx(env, osd, lu_object_fid(&dt->do_lu),
1270                                            buf, sizeof(info->oti_str), &zdn);
1271                 rc = osd_zap_lookup(osd, zapid, zdn, buf, 8,
1272                                     sizeof(*zde) / 8, zde);
1273                 if (!rc) {
1274                         zde->zde_dnode -= 1;
1275                         rc = -zap_update(osd->od_os, zapid, buf, 8,
1276                                          sizeof(*zde) / 8, zde, oh->ot_tx);
1277                 }
1278                 if (rc > 0)
1279                         rc = 0;
1280                 GOTO(out, rc);
1281         }
1282
1283         /* Only allow set size for regular file */
1284         if (!S_ISREG(dt->do_lu.lo_header->loh_attr))
1285                 valid &= ~(LA_SIZE | LA_BLOCKS);
1286
1287         if (valid & LA_CTIME && la->la_ctime == obj->oo_attr.la_ctime)
1288                 valid &= ~LA_CTIME;
1289
1290         if (valid & LA_MTIME && la->la_mtime == obj->oo_attr.la_mtime)
1291                 valid &= ~LA_MTIME;
1292
1293         if (valid & LA_ATIME && la->la_atime == obj->oo_attr.la_atime)
1294                 valid &= ~LA_ATIME;
1295
1296         if (valid == 0)
1297                 GOTO(out, rc = 0);
1298
1299         if (valid & LA_FLAGS) {
1300                 struct lustre_mdt_attrs *lma;
1301                 struct lu_buf buf;
1302                 int size = 0;
1303
1304                 if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
1305                         LASSERT(!obj->oo_pfid_in_lma);
1306                         BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*lma));
1307                         lma = (struct lustre_mdt_attrs *)&info->oti_buf;
1308                         buf.lb_buf = lma;
1309                         buf.lb_len = sizeof(info->oti_buf);
1310
1311                         /* Please do NOT call osd_xattr_get() directly, that
1312                          * will cause recursive down_read() on oo_guard. */
1313                         rc = osd_xattr_get_internal(env, obj, &buf,
1314                                                     XATTR_NAME_LMA, &size);
1315                         if (!rc && unlikely(size < sizeof(*lma))) {
1316                                 rc = -EINVAL;
1317                         } else if (!rc) {
1318                                 lma->lma_incompat =
1319                                         le32_to_cpu(lma->lma_incompat);
1320                                 lma->lma_incompat |=
1321                                         lustre_to_lma_flags(la->la_flags);
1322                                 lma->lma_incompat =
1323                                         cpu_to_le32(lma->lma_incompat);
1324                                 buf.lb_buf = lma;
1325                                 buf.lb_len = sizeof(*lma);
1326                                 rc = osd_xattr_set_internal(env, obj, &buf,
1327                                                             XATTR_NAME_LMA,
1328                                                             LU_XATTR_REPLACE,
1329                                                             oh);
1330                         }
1331                         if (rc < 0) {
1332                                 CWARN("%s: failed to set LMA flags: rc = %d\n",
1333                                        osd->od_svname, rc);
1334                                 GOTO(out, rc);
1335                         }
1336                 }
1337         }
1338
1339         write_lock(&obj->oo_attr_lock);
1340         cnt = 0;
1341
1342         if (valid & LA_PROJID) {
1343 #ifdef ZFS_PROJINHERIT
1344                 if (osd->od_projectused_dn) {
1345                         LASSERT(obj->oo_with_projid);
1346
1347                         osa->projid = obj->oo_attr.la_projid = la->la_projid;
1348                         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1349                                          &osa->projid, 8);
1350                 } else
1351 #endif
1352                         valid &= ~LA_PROJID;
1353         }
1354
1355         if (valid & LA_ATIME) {
1356                 osa->atime[0] = obj->oo_attr.la_atime = la->la_atime;
1357                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL,
1358                                  osa->atime, 16);
1359         }
1360         if (valid & LA_MTIME) {
1361                 osa->mtime[0] = obj->oo_attr.la_mtime = la->la_mtime;
1362                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL,
1363                                  osa->mtime, 16);
1364         }
1365         if (valid & LA_CTIME) {
1366                 osa->ctime[0] = obj->oo_attr.la_ctime = la->la_ctime;
1367                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL,
1368                                  osa->ctime, 16);
1369         }
1370         if (valid & LA_MODE) {
1371                 /* mode is stored along with type, so read it first */
1372                 obj->oo_attr.la_mode = (obj->oo_attr.la_mode & S_IFMT) |
1373                         (la->la_mode & ~S_IFMT);
1374                 osa->mode = obj->oo_attr.la_mode;
1375                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL,
1376                                  &osa->mode, 8);
1377         }
1378         if (valid & LA_SIZE) {
1379                 osa->size = obj->oo_attr.la_size = la->la_size;
1380                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL,
1381                                  &osa->size, 8);
1382         }
1383         if (valid & LA_NLINK) {
1384                 osa->nlink = obj->oo_attr.la_nlink = la->la_nlink;
1385                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL,
1386                                  &osa->nlink, 8);
1387         }
1388         if (valid & LA_RDEV) {
1389                 osa->rdev = obj->oo_attr.la_rdev = la->la_rdev;
1390                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL,
1391                                  &osa->rdev, 8);
1392         }
1393         if (valid & LA_FLAGS) {
1394                 osa->flags = attrs_fs2zfs(la->la_flags);
1395                 /* many flags are not supported by zfs, so ensure a good cached
1396                  * copy */
1397                 obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
1398 #ifdef ZFS_PROJINHERIT
1399                 if (obj->oo_with_projid && osd->od_projectused_dn)
1400                         osa->flags |= ZFS_PROJID;
1401 #endif
1402                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
1403                                  &osa->flags, 8);
1404         }
1405         if (valid & LA_UID) {
1406                 osa->uid = obj->oo_attr.la_uid = la->la_uid;
1407                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL,
1408                                  &osa->uid, 8);
1409         }
1410         if (valid & LA_GID) {
1411                 osa->gid = obj->oo_attr.la_gid = la->la_gid;
1412                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL,
1413                                  &osa->gid, 8);
1414         }
1415         obj->oo_attr.la_valid |= valid;
1416         write_unlock(&obj->oo_attr_lock);
1417
1418         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1419         rc = osd_object_sa_bulk_update(obj, bulk, cnt, oh);
1420
1421 out:
1422         up_read(&obj->oo_guard);
1423         RETURN(rc);
1424 }
1425
1426 /*
1427  * Object creation.
1428  *
1429  * XXX temporary solution.
1430  */
1431
1432 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1433                         struct dt_object *parent, struct dt_object *child,
1434                         umode_t child_mode)
1435 {
1436         LASSERT(ah);
1437
1438         ah->dah_parent = parent;
1439         ah->dah_mode = child_mode;
1440
1441         if (parent != NULL && !dt_object_remote(parent)) {
1442                 /* will help to find FID->ino at dt_insert("..") */
1443                 struct osd_object *pobj = osd_dt_obj(parent);
1444
1445                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
1446         }
1447 }
1448
1449 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
1450                               struct lu_attr *attr,
1451                               struct dt_allocation_hint *hint,
1452                               struct dt_object_format *dof,
1453                               struct thandle *handle)
1454 {
1455         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1456         struct osd_object       *obj = osd_dt_obj(dt);
1457         struct osd_device       *osd = osd_obj2dev(obj);
1458         struct osd_thandle      *oh;
1459         uint64_t                 zapid;
1460         dnode_t                 *dn;
1461         int                      rc, dnode_size;
1462         ENTRY;
1463
1464         LASSERT(dof);
1465
1466         switch (dof->dof_type) {
1467                 case DFT_REGULAR:
1468                 case DFT_SYM:
1469                 case DFT_NODE:
1470                         if (obj->oo_dt.do_body_ops == NULL)
1471                                 obj->oo_dt.do_body_ops = &osd_body_ops;
1472                         break;
1473                 default:
1474                         break;
1475         }
1476
1477         LASSERT(handle != NULL);
1478         oh = container_of0(handle, struct osd_thandle, ot_super);
1479         LASSERT(oh->ot_tx != NULL);
1480
1481         /* this is the minimum set of EAs on every Lustre object */
1482         obj->oo_ea_in_bonus = OSD_BASE_EA_IN_BONUS;
1483         /* reserve 32 bytes for extra stuff like ACLs */
1484         dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
1485
1486         switch (dof->dof_type) {
1487                 case DFT_DIR:
1488                         dt->do_index_ops = &osd_dir_ops;
1489                         /* fallthrough */
1490                 case DFT_INDEX:
1491                         /* for zap create */
1492                         dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
1493                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1494                         break;
1495                 case DFT_REGULAR:
1496                 case DFT_SYM:
1497                 case DFT_NODE:
1498                         /* first, we'll create new object */
1499                         dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
1500                         break;
1501
1502                 default:
1503                         LBUG();
1504                         break;
1505         }
1506
1507         /* and we'll add it to some mapping */
1508         zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0, &dn);
1509         osd_tx_hold_zap(oh->ot_tx, zapid, dn, TRUE, NULL);
1510
1511         /* will help to find FID->ino mapping at dt_insert() */
1512         osd_idc_find_and_init(env, osd, obj);
1513
1514         rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid,
1515                                attr->la_projid, 1, oh, NULL, OSD_QID_INODE);
1516
1517         RETURN(rc);
1518 }
1519
1520 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
1521                     struct osd_object *obj, sa_handle_t *sa_hdl, dmu_tx_t *tx,
1522                     struct lu_attr *la, uint64_t parent,
1523                     nvlist_t *xattr)
1524 {
1525         sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
1526         struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
1527         uint64_t gen;
1528         uint64_t crtime[2];
1529         inode_timespec_t now;
1530         int cnt;
1531         int rc;
1532         char *dxattr = NULL;
1533         size_t sa_size;
1534
1535
1536         LASSERT(sa_hdl);
1537
1538         gen = dmu_tx_get_txg(tx);
1539         gethrestime(&now);
1540         ZFS_TIME_ENCODE(&now, crtime);
1541
1542         osa->atime[0] = la->la_atime;
1543         osa->ctime[0] = la->la_ctime;
1544         osa->mtime[0] = la->la_mtime;
1545         osa->mode = la->la_mode;
1546         osa->uid = la->la_uid;
1547         osa->gid = la->la_gid;
1548         osa->rdev = la->la_rdev;
1549         osa->nlink = la->la_nlink;
1550         if (la->la_valid & LA_FLAGS)
1551                 osa->flags = attrs_fs2zfs(la->la_flags);
1552         else
1553                 osa->flags = 0;
1554         osa->size  = la->la_size;
1555 #ifdef ZFS_PROJINHERIT
1556         if (osd->od_projectused_dn) {
1557                 if (la->la_valid & LA_PROJID)
1558                         osa->projid = la->la_projid;
1559                 else
1560                         osa->projid = ZFS_DEFAULT_PROJID;
1561                 osa->flags |= ZFS_PROJID;
1562                 if (obj)
1563                         obj->oo_with_projid = 1;
1564         } else {
1565                 osa->flags &= ~ZFS_PROJID;
1566         }
1567 #endif
1568
1569         /*
1570          * we need to create all SA below upon object create.
1571          *
1572          * XXX The attribute order matters since the accounting callback relies
1573          * on static offsets (i.e. SA_*_OFFSET, see zfs_space_delta_cb()) to
1574          * look up the UID/GID/PROJID attributes. Moreover, the callback does
1575          * not seem to support the spill block.
1576          * We define attributes in the same order as SA_*_OFFSET in order to
1577          * work around the problem. See ORI-610.
1578          */
1579         cnt = 0;
1580         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(osd), NULL, &osa->mode, 8);
1581         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_SIZE(osd), NULL, &osa->size, 8);
1582         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GEN(osd), NULL, &gen, 8);
1583         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_UID(osd), NULL, &osa->uid, 8);
1584         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_GID(osd), NULL, &osa->gid, 8);
1585         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PARENT(osd), NULL, &parent, 8);
1586         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL, &osa->flags, 8);
1587         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(osd), NULL, osa->atime, 16);
1588         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(osd), NULL, osa->mtime, 16);
1589         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(osd), NULL, osa->ctime, 16);
1590         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CRTIME(osd), NULL, crtime, 16);
1591         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_LINKS(osd), NULL, &osa->nlink, 8);
1592 #ifdef ZFS_PROJINHERIT
1593         if (osd->od_projectused_dn)
1594                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
1595                                  &osa->projid, 8);
1596 #endif
1597         SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_RDEV(osd), NULL, &osa->rdev, 8);
1598         LASSERT(cnt <= ARRAY_SIZE(osd_oti_get(env)->oti_attr_bulk));
1599
1600         if (xattr) {
1601                 rc = -nvlist_size(xattr, &sa_size, NV_ENCODE_XDR);
1602                 LASSERT(rc == 0);
1603
1604                 dxattr = osd_zio_buf_alloc(sa_size);
1605                 LASSERT(dxattr);
1606
1607                 rc = -nvlist_pack(xattr, &dxattr, &sa_size,
1608                                 NV_ENCODE_XDR, KM_SLEEP);
1609                 LASSERT(rc == 0);
1610
1611                 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_DXATTR(osd),
1612                                 NULL, dxattr, sa_size);
1613         }
1614
1615         rc = -sa_replace_all_by_template(sa_hdl, bulk, cnt, tx);
1616         if (dxattr)
1617                 osd_zio_buf_free(dxattr, sa_size);
1618
1619         return rc;
1620 }
1621
1622 int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
1623                        uint64_t oid, dnode_t **dnp)
1624 {
1625         dmu_tx_hold_t *txh;
1626         int rc = 0;
1627
1628         /* take dnode_t from tx to save on dnode#->dnode_t lookup */
1629         for (txh = list_tail(&tx->tx_holds); txh;
1630              txh = list_prev(&tx->tx_holds, txh)) {
1631                 dnode_t *dn = txh->txh_dnode;
1632                 dmu_buf_impl_t *db;
1633
1634                 if (dn == NULL)
1635                         continue;
1636                 if (dn->dn_object != oid)
1637                         continue;
1638                 db = dn->dn_bonus;
1639                 if (db == NULL) {
1640                         rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1641                         if (dn->dn_bonus == NULL)
1642                                 dbuf_create_bonus(dn);
1643                         rw_exit(&dn->dn_struct_rwlock);
1644                 }
1645                 db = dn->dn_bonus;
1646                 LASSERT(db);
1647                 LASSERT(dn->dn_handle);
1648                 DB_DNODE_ENTER(db);
1649                 if (zfs_refcount_add(&db->db_holds, osd_obj_tag) == 1) {
1650                         zfs_refcount_add(&dn->dn_holds, osd_obj_tag);
1651                         atomic_inc_32(&dn->dn_dbufs_count);
1652                 }
1653                 *dnp = dn;
1654                 DB_DNODE_EXIT(db);
1655                 dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
1656                 break;
1657         }
1658
1659         if (unlikely(*dnp == NULL))
1660                 rc = __osd_obj2dnode(tx->tx_objset, oid, dnp);
1661
1662         return rc;
1663 }
1664
1665 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
1666 int osd_find_dnsize(struct osd_device *osd, int ea_in_bonus)
1667 {
1668         int dnsize;
1669
1670         if (osd->od_dnsize == ZFS_DNSIZE_AUTO) {
1671                 dnsize = DNODE_MIN_SIZE;
1672                 do {
1673                         if (DN_BONUS_SIZE(dnsize) >= ea_in_bonus + 32)
1674                                 break;
1675                         dnsize <<= 1;
1676                 } while (dnsize < DNODE_MAX_SIZE);
1677                 if (dnsize > DNODE_MAX_SIZE)
1678                         dnsize = DNODE_MAX_SIZE;
1679         } else if (osd->od_dnsize == ZFS_DNSIZE_1K) {
1680                 dnsize = 1024;
1681         } else if (osd->od_dnsize == ZFS_DNSIZE_2K) {
1682                 dnsize = 2048;
1683         } else if (osd->od_dnsize == ZFS_DNSIZE_4K) {
1684                 dnsize = 4096;
1685         } else if (osd->od_dnsize == ZFS_DNSIZE_8K) {
1686                 dnsize = 8192;
1687         } else if (osd->od_dnsize == ZFS_DNSIZE_16K) {
1688                 dnsize = 16384;
1689         } else {
1690                 dnsize = DNODE_MIN_SIZE;
1691         }
1692         return dnsize;
1693 }
1694 #endif
1695
1696 /*
1697  * The transaction passed to this routine must have
1698  * dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
1699  * to a transaction group.
1700  */
1701 int __osd_object_create(const struct lu_env *env, struct osd_device *osd,
1702                         struct osd_object *obj, const struct lu_fid *fid,
1703                         dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la)
1704 {
1705         dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
1706         uint64_t oid;
1707         int size;
1708
1709         /* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
1710          * would get an additional ditto copy */
1711         if (unlikely(S_ISREG(la->la_mode) &&
1712                      fid_seq_is_local_file(fid_seq(fid))))
1713                 type = DMU_OTN_UINT8_METADATA;
1714
1715         /* Create a new DMU object using the default dnode size. */
1716         if (obj)
1717                 size = obj->oo_ea_in_bonus;
1718         else
1719                 size = OSD_BASE_EA_IN_BONUS;
1720         oid = osd_dmu_object_alloc(osd->od_os, type, 0,
1721                                    osd_find_dnsize(osd, size), tx);
1722
1723         LASSERT(la->la_valid & LA_MODE);
1724         la->la_size = 0;
1725         la->la_nlink = 1;
1726
1727         return osd_find_new_dnode(env, tx, oid, dnp);
1728 }
1729
1730 /*
1731  * The transaction passed to this routine must have
1732  * dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
1733  * to a transaction group.
1734  *
1735  * Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
1736  * This is fine for directories today, because storing the FID in the dirent
1737  * will also require a FAT ZAP.  If there is a new type of micro ZAP created
1738  * then we might need to re-evaluate the use of this flag and instead do
1739  * a conversion from the different internal ZAP hash formats being used. */
1740 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
1741                      dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la,
1742                      unsigned dnsize, zap_flags_t flags)
1743 {
1744         uint64_t oid;
1745
1746         /* Assert that the transaction has been assigned to a
1747            transaction group. */
1748         LASSERT(tx->tx_txg != 0);
1749         *dnp = NULL;
1750
1751         oid = osd_zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
1752                                    DMU_OT_DIRECTORY_CONTENTS,
1753                                    14, /* == ZFS fzap_default_blockshift */
1754                                    DN_MAX_INDBLKSHIFT, /* indirect blockshift */
1755                                    dnsize, tx);
1756
1757         la->la_size = 2;
1758         la->la_nlink = 1;
1759
1760         return osd_find_new_dnode(env, tx, oid, dnp);
1761 }
1762
1763 static dnode_t *osd_mkidx(const struct lu_env *env, struct osd_object *obj,
1764                           struct lu_attr *la, struct osd_thandle *oh)
1765 {
1766         struct osd_device *osd = osd_obj2dev(obj);
1767         dnode_t *dn;
1768         int rc;
1769
1770         /* Index file should be created as regular file in order not to confuse
1771          * ZPL which could interpret them as directory.
1772          * We set ZAP_FLAG_UINT64_KEY to let ZFS know than we are going to use
1773          * binary keys */
1774         LASSERT(S_ISREG(la->la_mode));
1775         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1776                 osd_find_dnsize(osd, obj->oo_ea_in_bonus), ZAP_FLAG_UINT64_KEY);
1777         if (rc)
1778                 return ERR_PTR(rc);
1779         return dn;
1780 }
1781
1782 static dnode_t *osd_mkdir(const struct lu_env *env, struct osd_object *obj,
1783                           struct lu_attr *la, struct osd_thandle *oh)
1784 {
1785         struct osd_device *osd = osd_obj2dev(obj);
1786         dnode_t *dn;
1787         int rc;
1788
1789         LASSERT(S_ISDIR(la->la_mode));
1790         rc = __osd_zap_create(env, osd, &dn, oh->ot_tx, la,
1791                               osd_find_dnsize(osd, obj->oo_ea_in_bonus), 0);
1792         if (rc)
1793                 return ERR_PTR(rc);
1794         return dn;
1795 }
1796
1797 static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
1798                           struct lu_attr *la, struct osd_thandle *oh)
1799 {
1800         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
1801         struct osd_device *osd = osd_obj2dev(obj);
1802         dnode_t *dn;
1803         int rc;
1804
1805         LASSERT(S_ISREG(la->la_mode));
1806         rc = __osd_object_create(env, osd, obj, fid, &dn, oh->ot_tx, la);
1807         if (rc)
1808                 return ERR_PTR(rc);
1809
1810         if ((fid_is_idif(fid) || fid_is_norm(fid) || fid_is_echo(fid))) {
1811                 /* The minimum block size must be at least page size otherwise
1812                  * it will break the assumption in tgt_thread_big_cache where
1813                  * the array size is PTLRPC_MAX_BRW_PAGES. It will also affect
1814                  * RDMA due to subpage transfer size */
1815                 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1816                                                PAGE_SIZE, 0, oh->ot_tx);
1817                 if (unlikely(rc)) {
1818                         CERROR("%s: can't change blocksize: %d\n",
1819                                osd->od_svname, rc);
1820                         return ERR_PTR(rc);
1821                 }
1822         } else if ((fid_is_llog(fid))) {
1823                 rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
1824                                                LLOG_MIN_CHUNK_SIZE, 0, oh->ot_tx);
1825                 if (unlikely(rc)) {
1826                         CERROR("%s: can't change blocksize: %d\n",
1827                                osd->od_svname, rc);
1828                         return ERR_PTR(rc);
1829                 }
1830         }
1831
1832         return dn;
1833 }
1834
1835 static dnode_t *osd_mksym(const struct lu_env *env, struct osd_object *obj,
1836                           struct lu_attr *la, struct osd_thandle *oh)
1837 {
1838         dnode_t *dn;
1839         int rc;
1840
1841         LASSERT(S_ISLNK(la->la_mode));
1842         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1843                                  lu_object_fid(&obj->oo_dt.do_lu),
1844                                  &dn, oh->ot_tx, la);
1845         if (rc)
1846                 return ERR_PTR(rc);
1847         return dn;
1848 }
1849
1850 static dnode_t *osd_mknod(const struct lu_env *env, struct osd_object *obj,
1851                           struct lu_attr *la, struct osd_thandle *oh)
1852 {
1853         dnode_t *dn;
1854         int rc;
1855
1856         if (S_ISCHR(la->la_mode) || S_ISBLK(la->la_mode))
1857                 la->la_valid |= LA_RDEV;
1858
1859         rc = __osd_object_create(env, osd_obj2dev(obj), obj,
1860                                  lu_object_fid(&obj->oo_dt.do_lu),
1861                                  &dn, oh->ot_tx, la);
1862         if (rc)
1863                 return ERR_PTR(rc);
1864         return dn;
1865 }
1866
1867 typedef dnode_t *(*osd_obj_type_f)(const struct lu_env *env,
1868                                    struct osd_object *obj,
1869                                    struct lu_attr *la,
1870                                    struct osd_thandle *oh);
1871
1872 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1873 {
1874         osd_obj_type_f result;
1875
1876         switch (type) {
1877         case DFT_DIR:
1878                 result = osd_mkdir;
1879                 break;
1880         case DFT_INDEX:
1881                 result = osd_mkidx;
1882                 break;
1883         case DFT_REGULAR:
1884                 result = osd_mkreg;
1885                 break;
1886         case DFT_SYM:
1887                 result = osd_mksym;
1888                 break;
1889         case DFT_NODE:
1890                 result = osd_mknod;
1891                 break;
1892         default:
1893                 LBUG();
1894                 break;
1895         }
1896         return result;
1897 }
1898
1899 /*
1900  * Concurrency: @dt is write locked.
1901  */
1902 static int osd_create(const struct lu_env *env, struct dt_object *dt,
1903                       struct lu_attr *attr, struct dt_allocation_hint *hint,
1904                       struct dt_object_format *dof, struct thandle *th)
1905 {
1906         struct osd_thread_info  *info = osd_oti_get(env);
1907         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1908         struct zpl_direntry     *zde = &info->oti_zde.lzd_reg;
1909         const struct lu_fid     *fid = lu_object_fid(&dt->do_lu);
1910         struct osd_object       *obj = osd_dt_obj(dt);
1911         struct osd_device       *osd = osd_obj2dev(obj);
1912         char                    *buf = info->oti_str;
1913         struct osd_thandle      *oh;
1914         dnode_t *dn = NULL, *zdn = NULL;
1915         uint64_t                 zapid, parent = 0;
1916         int                      rc;
1917         __u32 compat = 0;
1918
1919         ENTRY;
1920
1921         LASSERT(!fid_is_acct(fid));
1922
1923         /* concurrent create declarations should not see
1924          * the object inconsistent (db, attr, etc).
1925          * in regular cases acquisition should be cheap */
1926         down_write(&obj->oo_guard);
1927
1928         if (unlikely(dt_object_exists(dt)))
1929                 GOTO(out, rc = -EEXIST);
1930
1931         LASSERT(osd_invariant(obj));
1932         LASSERT(dof != NULL);
1933
1934         LASSERT(th != NULL);
1935         oh = container_of0(th, struct osd_thandle, ot_super);
1936
1937         LASSERT(obj->oo_dn == NULL);
1938
1939         /* to follow ZFS on-disk format we need
1940          * to initialize parent dnode properly */
1941         if (hint != NULL && hint->dah_parent != NULL &&
1942             !dt_object_remote(hint->dah_parent))
1943                 parent = osd_dt_obj(hint->dah_parent)->oo_dn->dn_object;
1944
1945         /* we may fix some attributes, better do not change the source */
1946         obj->oo_attr = *attr;
1947         obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
1948
1949 #ifdef ZFS_PROJINHERIT
1950         if (osd->od_projectused_dn) {
1951                 if (!(obj->oo_attr.la_valid & LA_PROJID))
1952                         obj->oo_attr.la_projid = ZFS_DEFAULT_PROJID;
1953                 obj->oo_with_projid = 1;
1954         }
1955 #endif
1956
1957         dn = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
1958         if (IS_ERR(dn)) {
1959                 rc = PTR_ERR(dn);
1960                 dn = NULL;
1961                 GOTO(out, rc);
1962         }
1963
1964         zde->zde_pad = 0;
1965         zde->zde_dnode = dn->dn_object;
1966         zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
1967
1968         zapid = osd_get_name_n_idx(env, osd, fid, buf,
1969                                    sizeof(info->oti_str), &zdn);
1970         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY) ||
1971             (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY)))
1972                 goto skip_add;
1973
1974         if (osd->od_is_ost && OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
1975                 zde->zde_dnode++;
1976
1977         rc = osd_zap_add(osd, zapid, zdn, buf, 8, 1, zde, oh->ot_tx);
1978         if (rc)
1979                 GOTO(out, rc);
1980
1981 skip_add:
1982         obj->oo_dn = dn;
1983         /* Now add in all of the "SA" attributes */
1984         rc = osd_sa_handle_get(obj);
1985         if (rc)
1986                 GOTO(out, rc);
1987
1988         rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
1989         if (rc)
1990                 GOTO(out, rc);
1991
1992         /* initialize LMA */
1993         if (fid_is_idif(fid) || (fid_is_norm(fid) && osd->od_is_ost))
1994                 compat |= LMAC_FID_ON_OST;
1995         lustre_lma_init(lma, fid, compat, 0);
1996         lustre_lma_swab(lma);
1997         rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
1998                                     (uchar_t *)lma, sizeof(*lma));
1999         if (rc)
2000                 GOTO(out, rc);
2001
2002         /* configure new osd object */
2003         obj->oo_parent = parent != 0 ? parent : zapid;
2004         obj->oo_late_attr_set = 1;
2005         rc = __osd_sa_xattr_schedule_update(env, obj, oh);
2006         if (rc)
2007                 GOTO(out, rc);
2008
2009         /* XXX: oo_lma_flags */
2010         obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
2011         if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
2012                 /* no body operations for accounting objects */
2013                 obj->oo_dt.do_body_ops = &osd_body_ops;
2014
2015         osd_idc_find_and_init(env, osd, obj);
2016
2017 out:
2018         if (unlikely(rc && dn)) {
2019                 dmu_object_free(osd->od_os, dn->dn_object, oh->ot_tx);
2020                 osd_dnode_rele(dn);
2021                 obj->oo_dn = NULL;
2022         } else if (!rc) {
2023                 obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
2024         }
2025         up_write(&obj->oo_guard);
2026         RETURN(rc);
2027 }
2028
2029 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
2030                                struct thandle *th)
2031 {
2032         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2033         return osd_declare_attr_set(env, dt, NULL, th);
2034 }
2035
2036 /*
2037  * Concurrency: @dt is write locked.
2038  */
2039 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
2040                        struct thandle *handle)
2041 {
2042         struct osd_object       *obj = osd_dt_obj(dt);
2043         struct osd_thandle      *oh;
2044         struct osd_device       *osd = osd_obj2dev(obj);
2045         uint64_t                 nlink;
2046         int rc;
2047
2048         ENTRY;
2049
2050         down_read(&obj->oo_guard);
2051         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2052                 GOTO(out, rc = -ENOENT);
2053
2054         LASSERT(osd_invariant(obj));
2055         LASSERT(obj->oo_sa_hdl != NULL);
2056
2057         oh = container_of0(handle, struct osd_thandle, ot_super);
2058
2059         write_lock(&obj->oo_attr_lock);
2060         nlink = ++obj->oo_attr.la_nlink;
2061         write_unlock(&obj->oo_attr_lock);
2062
2063         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2064
2065 out:
2066         up_read(&obj->oo_guard);
2067         RETURN(rc);
2068 }
2069
2070 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
2071                                struct thandle *handle)
2072 {
2073         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
2074         return osd_declare_attr_set(env, dt, NULL, handle);
2075 }
2076
2077 /*
2078  * Concurrency: @dt is write locked.
2079  */
2080 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
2081                        struct thandle *handle)
2082 {
2083         struct osd_object       *obj = osd_dt_obj(dt);
2084         struct osd_thandle      *oh;
2085         struct osd_device       *osd = osd_obj2dev(obj);
2086         uint64_t                 nlink;
2087         int                      rc;
2088
2089         ENTRY;
2090
2091         down_read(&obj->oo_guard);
2092
2093         if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
2094                 GOTO(out, rc = -ENOENT);
2095
2096         LASSERT(osd_invariant(obj));
2097         LASSERT(obj->oo_sa_hdl != NULL);
2098
2099         oh = container_of0(handle, struct osd_thandle, ot_super);
2100         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2101
2102         write_lock(&obj->oo_attr_lock);
2103         nlink = --obj->oo_attr.la_nlink;
2104         write_unlock(&obj->oo_attr_lock);
2105
2106         rc = osd_object_sa_update(obj, SA_ZPL_LINKS(osd), &nlink, 8, oh);
2107
2108 out:
2109         up_read(&obj->oo_guard);
2110         RETURN(rc);
2111 }
2112
2113 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
2114                            __u64 start, __u64 end)
2115 {
2116         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
2117         struct dmu_buf_impl *db = osd_dt_obj(dt)->oo_dn->dn_dbuf;
2118         uint64_t txg = 0;
2119         ENTRY;
2120
2121         if (osd->od_dt_dev.dd_rdonly)
2122                 RETURN(0);
2123
2124         mutex_enter(&db->db_mtx);
2125         if (db->db_last_dirty)
2126                 txg = db->db_last_dirty->dr_txg;
2127         mutex_exit(&db->db_mtx);
2128
2129         if (txg) {
2130                 /* the object is dirty or being synced */
2131                 if (osd_object_sync_delay_us < 0)
2132                         txg_wait_synced(dmu_objset_pool(osd->od_os), txg);
2133                 else
2134                         udelay(osd_object_sync_delay_us);
2135         }
2136
2137         RETURN(0);
2138 }
2139
2140 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
2141 {
2142         return 0;
2143 }
2144
2145 static struct dt_object_operations osd_obj_ops = {
2146         .do_read_lock           = osd_read_lock,
2147         .do_write_lock          = osd_write_lock,
2148         .do_read_unlock         = osd_read_unlock,
2149         .do_write_unlock        = osd_write_unlock,
2150         .do_write_locked        = osd_write_locked,
2151         .do_attr_get            = osd_attr_get,
2152         .do_declare_attr_set    = osd_declare_attr_set,
2153         .do_attr_set            = osd_attr_set,
2154         .do_ah_init             = osd_ah_init,
2155         .do_declare_create      = osd_declare_create,
2156         .do_create              = osd_create,
2157         .do_declare_destroy     = osd_declare_destroy,
2158         .do_destroy             = osd_destroy,
2159         .do_index_try           = osd_index_try,
2160         .do_declare_ref_add     = osd_declare_ref_add,
2161         .do_ref_add             = osd_ref_add,
2162         .do_declare_ref_del     = osd_declare_ref_del,
2163         .do_ref_del             = osd_ref_del,
2164         .do_xattr_get           = osd_xattr_get,
2165         .do_declare_xattr_set   = osd_declare_xattr_set,
2166         .do_xattr_set           = osd_xattr_set,
2167         .do_declare_xattr_del   = osd_declare_xattr_del,
2168         .do_xattr_del           = osd_xattr_del,
2169         .do_xattr_list          = osd_xattr_list,
2170         .do_object_sync         = osd_object_sync,
2171         .do_invalidate          = osd_invalidate,
2172 };
2173
2174 static struct lu_object_operations osd_lu_obj_ops = {
2175         .loo_object_init        = osd_object_init,
2176         .loo_object_delete      = osd_object_delete,
2177         .loo_object_release     = osd_object_release,
2178         .loo_object_free        = osd_object_free,
2179         .loo_object_print       = osd_object_print,
2180         .loo_object_invariant   = osd_object_invariant,
2181 };
2182
2183 static int osd_otable_it_attr_get(const struct lu_env *env,
2184                                 struct dt_object *dt,
2185                                 struct lu_attr *attr)
2186 {
2187         attr->la_valid = 0;
2188         return 0;
2189 }
2190
2191 static struct dt_object_operations osd_obj_otable_it_ops = {
2192         .do_attr_get            = osd_otable_it_attr_get,
2193         .do_index_try           = osd_index_try,
2194 };
2195
2196 module_param(osd_object_sync_delay_us, int, 0644);
2197 MODULE_PARM_DESC(osd_object_sync_delay_us,
2198                  "If zero or larger delay N usec instead of doing object sync");