Whamcloud - gitweb
LU-7345 obdclass: annotate locks in __local_file_create
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_handler.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2015, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/osd/osd_handler.c
37  *
38  * Top-level entry points into osd module
39  *
40  * Author: Nikita Danilov <nikita@clusterfs.com>
41  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
42  */
43
44 #define DEBUG_SUBSYSTEM S_OSD
45
46 #include <linux/module.h>
47 #include <linux/user_namespace.h>
48 #ifdef HAVE_UIDGID_HEADER
49 # include <linux/uidgid.h>
50 #endif
51
52 /* LUSTRE_VERSION_CODE */
53 #include <lustre_ver.h>
54 /* prerequisite for linux/xattr.h */
55 #include <linux/types.h>
56 /* prerequisite for linux/xattr.h */
57 #include <linux/fs.h>
58 /* XATTR_{REPLACE,CREATE} */
59 #include <linux/xattr.h>
60
61 #include <ldiskfs/ldiskfs.h>
62 #include <ldiskfs/xattr.h>
63 #undef ENTRY
64 /*
65  * struct OBD_{ALLOC,FREE}*()
66  * OBD_FAIL_CHECK
67  */
68 #include <obd_support.h>
69 /* struct ptlrpc_thread */
70 #include <lustre_net.h>
71 #include <lustre_fid.h>
72 /* process_config */
73 #include <lustre_param.h>
74
75 #include "osd_internal.h"
76 #include "osd_dynlocks.h"
77
78 /* llo_* api support */
79 #include <md_object.h>
80 #include <lustre_quota.h>
81
82 #include <lustre_linkea.h>
83
84 int ldiskfs_pdo = 1;
85 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
86                 "ldiskfs with parallel directory operations");
87
88 int ldiskfs_track_declares_assert;
89 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
90                 "LBUG during tracking of declares");
91
92 /* Slab to allocate dynlocks */
93 struct kmem_cache *dynlock_cachep;
94
95 /* Slab to allocate osd_it_ea */
96 struct kmem_cache *osd_itea_cachep;
97
98 static struct lu_kmem_descr ldiskfs_caches[] = {
99         {
100                 .ckd_cache = &dynlock_cachep,
101                 .ckd_name  = "dynlock_cache",
102                 .ckd_size  = sizeof(struct dynlock_handle)
103         },
104         {
105                 .ckd_cache = &osd_itea_cachep,
106                 .ckd_name  = "osd_itea_cache",
107                 .ckd_size  = sizeof(struct osd_it_ea)
108         },
109         {
110                 .ckd_cache = NULL
111         }
112 };
113
114 static const char dot[] = ".";
115 static const char dotdot[] = "..";
116 static const char remote_obj_dir[] = "REM_OBJ_DIR";
117
118 static const struct lu_object_operations      osd_lu_obj_ops;
119 static const struct dt_object_operations      osd_obj_ops;
120 static const struct dt_object_operations      osd_obj_ea_ops;
121 static const struct dt_object_operations      osd_obj_otable_it_ops;
122 static const struct dt_index_operations       osd_index_iam_ops;
123 static const struct dt_index_operations       osd_index_ea_ops;
124
125 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
126                           const struct lu_fid *fid);
127 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
128                                                 struct osd_device *osd);
129
130 int osd_trans_declare_op2rb[] = {
131         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
132         [OSD_OT_PUNCH]          = OSD_OT_MAX,
133         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
134         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
135         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
136         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
137         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
138         [OSD_OT_WRITE]          = OSD_OT_WRITE,
139         [OSD_OT_INSERT]         = OSD_OT_DELETE,
140         [OSD_OT_DELETE]         = OSD_OT_INSERT,
141         [OSD_OT_QUOTA]          = OSD_OT_MAX,
142 };
143
144 static int osd_has_index(const struct osd_object *obj)
145 {
146         return obj->oo_dt.do_index_ops != NULL;
147 }
148
149 static int osd_object_invariant(const struct lu_object *l)
150 {
151         return osd_invariant(osd_obj(l));
152 }
153
154 /*
155  * Concurrency: doesn't matter
156  */
157
158 /*
159  * Concurrency: doesn't matter
160  */
161 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
162 {
163         struct osd_thread_info *oti = osd_oti_get(env);
164         return oti->oti_w_locks > 0 && o->oo_owner == env;
165 }
166
167 /*
168  * Concurrency: doesn't access mutable data
169  */
170 static int osd_root_get(const struct lu_env *env,
171                         struct dt_device *dev, struct lu_fid *f)
172 {
173         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
174         return 0;
175 }
176
177 /*
178  * OSD object methods.
179  */
180
181 /*
182  * Concurrency: no concurrent access is possible that early in object
183  * life-cycle.
184  */
185 static struct lu_object *osd_object_alloc(const struct lu_env *env,
186                                           const struct lu_object_header *hdr,
187                                           struct lu_device *d)
188 {
189         struct osd_object *mo;
190
191         OBD_ALLOC_PTR(mo);
192         if (mo != NULL) {
193                 struct lu_object *l;
194
195                 l = &mo->oo_dt.do_lu;
196                 dt_object_init(&mo->oo_dt, NULL, d);
197                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
198                 l->lo_ops = &osd_lu_obj_ops;
199                 init_rwsem(&mo->oo_sem);
200                 init_rwsem(&mo->oo_ext_idx_sem);
201                 spin_lock_init(&mo->oo_guard);
202                 return l;
203         } else {
204                 return NULL;
205         }
206 }
207
208 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
209                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
210 {
211         int rc;
212
213         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
214         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
215                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
216         if (rc > 0) {
217                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
218                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
219                 rc = 0;
220                 lustre_lma_swab(lma);
221                 /* Check LMA compatibility */
222                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
223                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
224                               "for fid = "DFID", ino = %lu\n",
225                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
226                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
227                               PFID(&lma->lma_self_fid), inode->i_ino);
228                         rc = -EOPNOTSUPP;
229                 }
230         } else if (rc == 0) {
231                 rc = -ENODATA;
232         }
233
234         return rc;
235 }
236
237 /*
238  * retrieve object from backend ext fs.
239  **/
240 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
241                        struct osd_inode_id *id)
242 {
243         struct inode *inode = NULL;
244
245         /* if we look for an inode withing a running
246          * transaction, then we risk to deadlock */
247         /* osd_dirent_check_repair() breaks this */
248         /*LASSERT(current->journal_info == NULL);*/
249
250         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
251         if (IS_ERR(inode)) {
252                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
253                        id->oii_ino, PTR_ERR(inode));
254         } else if (id->oii_gen != OSD_OII_NOGEN &&
255                    inode->i_generation != id->oii_gen) {
256                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
257                        "i_generation = %u\n",
258                        id->oii_ino, id->oii_gen, inode->i_generation);
259                 iput(inode);
260                 inode = ERR_PTR(-ESTALE);
261         } else if (inode->i_nlink == 0) {
262                 /* due to parallel readdir and unlink,
263                 * we can have dead inode here. */
264                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
265                 iput(inode);
266                 inode = ERR_PTR(-ESTALE);
267         } else if (is_bad_inode(inode)) {
268                 CWARN("%.16s: bad inode: ino = %u\n",
269                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
270                 iput(inode);
271                 inode = ERR_PTR(-ENOENT);
272         } else {
273                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
274                 if (id->oii_gen == OSD_OII_NOGEN)
275                         osd_id_gen(id, inode->i_ino, inode->i_generation);
276
277                 /* Do not update file c/mtime in ldiskfs.
278                  * NB: we don't have any lock to protect this because we don't
279                  * have reference on osd_object now, but contention with
280                  * another lookup + attr_set can't happen in the tiny window
281                  * between if (...) and set S_NOCMTIME. */
282                 if (!(inode->i_flags & S_NOCMTIME))
283                         inode->i_flags |= S_NOCMTIME;
284         }
285         return inode;
286 }
287
288 static struct inode *
289 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
290              struct osd_inode_id *id, struct lu_fid *fid)
291 {
292         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
293         struct inode            *inode;
294         int                      rc;
295
296         inode = osd_iget(info, dev, id);
297         if (IS_ERR(inode))
298                 return inode;
299
300         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
301         if (rc == 0) {
302                 *fid = lma->lma_self_fid;
303         } else if (rc == -ENODATA) {
304                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
305                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
306                 else
307                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
308         } else {
309                 iput(inode);
310                 inode = ERR_PTR(rc);
311         }
312         return inode;
313 }
314
315 static struct inode *osd_iget_check(struct osd_thread_info *info,
316                                     struct osd_device *dev,
317                                     const struct lu_fid *fid,
318                                     struct osd_inode_id *id,
319                                     bool cached)
320 {
321         struct inode    *inode;
322         int              rc     = 0;
323         ENTRY;
324
325         /* The cached OI mapping is trustable. If we cannot locate the inode
326          * via the cached OI mapping, then return the failure to the caller
327          * directly without further OI checking. */
328
329         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
330         if (IS_ERR(inode)) {
331                 rc = PTR_ERR(inode);
332                 if (cached || (rc != -ENOENT && rc != -ESTALE)) {
333                         CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
334                                id->oii_ino, rc);
335
336                         GOTO(put, rc);
337                 }
338
339                 goto check_oi;
340         }
341
342         if (is_bad_inode(inode)) {
343                 rc = -ENOENT;
344                 if (cached) {
345                         CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
346
347                         GOTO(put, rc);
348                 }
349
350                 goto check_oi;
351         }
352
353         if (id->oii_gen != OSD_OII_NOGEN &&
354             inode->i_generation != id->oii_gen) {
355                 rc = -ESTALE;
356                 if (cached) {
357                         CDEBUG(D_INODE, "unmatched inode: ino = %u, "
358                                "oii_gen = %u, i_generation = %u\n",
359                                id->oii_ino, id->oii_gen, inode->i_generation);
360
361                         GOTO(put, rc);
362                 }
363
364                 goto check_oi;
365         }
366
367         if (inode->i_nlink == 0) {
368                 rc = -ENOENT;
369                 if (cached) {
370                         CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
371
372                         GOTO(put, rc);
373                 }
374
375                 goto check_oi;
376         }
377
378         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
379
380 check_oi:
381         if (rc != 0) {
382                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
383
384                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
385                 /* XXX: There are four possible cases:
386                  *      1. rc = 0.
387                  *         Backup/restore caused the OI invalid.
388                  *      2. rc = 0.
389                  *         Someone unlinked the object but NOT removed
390                  *         the OI mapping, such as mount target device
391                  *         as ldiskfs, and modify something directly.
392                  *      3. rc = -ENOENT.
393                  *         Someone just removed the object between the
394                  *         former oi_lookup and the iget. It is normal.
395                  *      4. Other failure cases.
396                  *
397                  *      Generally, when the device is mounted, it will
398                  *      auto check whether the system is restored from
399                  *      file-level backup or not. We trust such detect
400                  *      to distinguish the 1st case from the 2nd case. */
401                 if (rc == 0) {
402                         if (!IS_ERR(inode) && inode->i_generation != 0 &&
403                             inode->i_generation == id->oii_gen)
404                                 /* "id->oii_gen != OSD_OII_NOGEN" is for
405                                  * "@cached == false" case. */
406                                 rc = -ENOENT;
407                         else
408                                 rc = -EREMCHG;
409                 } else {
410                         /* If the OI mapping was in OI file before the
411                          * osd_iget_check(), but now, it is disappear,
412                          * then it must be removed by race. That is a
413                          * normal race case. */
414                 }
415         } else {
416                 if (id->oii_gen == OSD_OII_NOGEN)
417                         osd_id_gen(id, inode->i_ino, inode->i_generation);
418
419                 /* Do not update file c/mtime in ldiskfs.
420                  * NB: we don't have any lock to protect this because we don't
421                  * have reference on osd_object now, but contention with
422                  * another lookup + attr_set can't happen in the tiny window
423                  * between if (...) and set S_NOCMTIME. */
424                 if (!(inode->i_flags & S_NOCMTIME))
425                         inode->i_flags |= S_NOCMTIME;
426         }
427
428         GOTO(put, rc);
429
430 put:
431         if (rc != 0) {
432                 if (!IS_ERR(inode))
433                         iput(inode);
434
435                 inode = ERR_PTR(rc);
436         }
437
438         return inode;
439 }
440
441 /**
442  * \retval +v: new filter_fid, does not contain self-fid
443  * \retval 0:  filter_fid_old, contains self-fid
444  * \retval -v: other failure cases
445  */
446 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
447                  struct dentry *dentry, struct lu_fid *fid)
448 {
449         struct filter_fid_old   *ff     = &info->oti_ff;
450         struct ost_id           *ostid  = &info->oti_ostid;
451         int                      rc;
452
453         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
454         if (rc == sizeof(*ff)) {
455                 rc = 0;
456                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
457                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
458                 /* XXX: use 0 as the index for compatibility, the caller will
459                  *      handle index related issues when necessarry. */
460                 ostid_to_fid(fid, ostid, 0);
461         } else if (rc == sizeof(struct filter_fid)) {
462                 rc = 1;
463         } else if (rc >= 0) {
464                 rc = -EINVAL;
465         }
466
467         return rc;
468 }
469
470 static int osd_lma_self_repair(struct osd_thread_info *info,
471                                struct osd_device *osd, struct inode *inode,
472                                const struct lu_fid *fid, __u32 compat)
473 {
474         handle_t *jh;
475         int       rc;
476
477         LASSERT(current->journal_info == NULL);
478
479         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
480                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
481         if (IS_ERR(jh)) {
482                 rc = PTR_ERR(jh);
483                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
484                       osd_name(osd), rc);
485                 return rc;
486         }
487
488         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
489         if (rc != 0)
490                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
491                       osd_name(osd), rc);
492         ldiskfs_journal_stop(jh);
493         return rc;
494 }
495
496 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
497 {
498         struct osd_thread_info  *info   = osd_oti_get(env);
499         struct osd_device       *osd    = osd_obj2dev(obj);
500         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
501         struct inode            *inode  = obj->oo_inode;
502         struct dentry           *dentry = &info->oti_obj_dentry;
503         struct lu_fid           *fid    = NULL;
504         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
505         int                      rc;
506         ENTRY;
507
508         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
509         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
510                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
511         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
512                 fid = &lma->lma_self_fid;
513                 rc = osd_get_idif(info, inode, dentry, fid);
514                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
515                         /* For the given OST-object, if it has neither LMA nor
516                          * FID in XATTR_NAME_FID, then the given FID (which is
517                          * contained in the @obj, from client RPC for locating
518                          * the OST-object) is trusted. We use it to generate
519                          * the LMA. */
520                         osd_lma_self_repair(info, osd, inode, rfid,
521                                             LMAC_FID_ON_OST);
522                         RETURN(0);
523                 }
524         }
525
526         if (rc < 0)
527                 RETURN(rc);
528
529         if (rc > 0) {
530                 rc = 0;
531                 lustre_lma_swab(lma);
532                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
533                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
534                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
535                               "fid = "DFID", ino = %lu\n", osd_name(osd),
536                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
537                               PFID(rfid), inode->i_ino);
538                         rc = -EOPNOTSUPP;
539                 } else {
540                         fid = &lma->lma_self_fid;
541                 }
542         }
543
544         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
545                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
546                         struct ost_id   *oi   = &info->oti_ostid;
547                         struct lu_fid   *fid1 = &info->oti_fid3;
548                         __u32            idx  = fid_idif_ost_idx(rfid);
549
550                         /* For old IDIF, the OST index is not part of the IDIF,
551                          * Means that different OSTs may have the same IDIFs.
552                          * Under such case, we need to make some compatible
553                          * check to make sure to trigger OI scrub properly. */
554                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
555                                 /* Given @rfid is new, LMA is old. */
556                                 fid_to_ostid(fid, oi);
557                                 ostid_to_fid(fid1, oi, idx);
558                                 if (lu_fid_eq(fid1, rfid)) {
559                                         if (osd->od_index_in_idif)
560                                                 osd_lma_self_repair(info, osd,
561                                                         inode, rfid,
562                                                         LMAC_FID_ON_OST);
563                                         RETURN(0);
564                                 }
565                         }
566                 }
567
568                 rc = -EREMCHG;
569         }
570
571         RETURN(rc);
572 }
573
574 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
575                           const struct lu_fid *fid,
576                           const struct lu_object_conf *conf)
577 {
578         struct osd_thread_info *info;
579         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
580         struct osd_device      *dev;
581         struct osd_idmap_cache *oic;
582         struct osd_inode_id    *id;
583         struct inode           *inode;
584         struct osd_scrub       *scrub;
585         struct scrub_file      *sf;
586         int                     result;
587         int                     saved  = 0;
588         bool                    cached  = true;
589         bool                    triggered = false;
590         ENTRY;
591
592         LINVRNT(osd_invariant(obj));
593         LASSERT(obj->oo_inode == NULL);
594         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
595
596         dev = osd_dev(ldev);
597         scrub = &dev->od_scrub;
598         sf = &scrub->os_file;
599         info = osd_oti_get(env);
600         LASSERT(info);
601         oic = &info->oti_cache;
602
603         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
604                 RETURN(-ENOENT);
605
606         /* For the object is created as locking anchor, or for the object to
607          * be created on disk. No need to osd_oi_lookup() at here because FID
608          * shouldn't never be re-used, if it's really a duplicate FID from
609          * unexpected reason, we should be able to detect it later by calling
610          * do_create->osd_oi_insert(). */
611         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
612                 GOTO(out, result = 0);
613
614         /* Search order: 1. per-thread cache. */
615         if (lu_fid_eq(fid, &oic->oic_fid) &&
616             likely(oic->oic_dev == dev)) {
617                 id = &oic->oic_lid;
618                 goto iget;
619         }
620
621         id = &info->oti_id;
622         if (!list_empty(&scrub->os_inconsistent_items)) {
623                 /* Search order: 2. OI scrub pending list. */
624                 result = osd_oii_lookup(dev, fid, id);
625                 if (result == 0)
626                         goto iget;
627         }
628
629         cached = false;
630         /* Search order: 3. OI files. */
631         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
632         if (result == -ENOENT) {
633                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
634                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
635                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
636                                       sf->sf_oi_bitmap))
637                         GOTO(out, result = 0);
638
639                 goto trigger;
640         }
641
642         if (result != 0)
643                 GOTO(out, result);
644
645 iget:
646         inode = osd_iget_check(info, dev, fid, id, cached);
647         if (IS_ERR(inode)) {
648                 result = PTR_ERR(inode);
649                 if (result == -ENOENT || result == -ESTALE)
650                         GOTO(out, result = -ENOENT);
651
652                 if (result == -EREMCHG) {
653
654 trigger:
655                         if (unlikely(triggered))
656                                 GOTO(out, result = saved);
657
658                         triggered = true;
659                         if (thread_is_running(&scrub->os_thread)) {
660                                 result = -EINPROGRESS;
661                         } else if (!dev->od_noscrub) {
662                                 result = osd_scrub_start(dev, SS_AUTO_FULL |
663                                         SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
664                                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
665                                               "for "DFID", rc = %d [1]\n",
666                                               osd_name(dev), PFID(fid), result);
667                                 if (result == 0 || result == -EALREADY)
668                                         result = -EINPROGRESS;
669                                 else
670                                         result = -EREMCHG;
671                         }
672
673                         if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD))
674                                 GOTO(out, result);
675
676                         /* We still have chance to get the valid inode: for the
677                          * object which is referenced by remote name entry, the
678                          * object on the local MDT will be linked under the dir
679                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
680                          *
681                          * We do not know whether the object for the given FID
682                          * is referenced by some remote name entry or not, and
683                          * especially for DNE II, a multiple-linked object may
684                          * have many name entries reside on many MDTs.
685                          *
686                          * To simplify the operation, OSD will not distinguish
687                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
688                          * only happened for the RPC from other MDT during the
689                          * OI scrub, or for the client side RPC with FID only,
690                          * such as FID to path, or from old connected client. */
691                         saved = result;
692                         result = osd_lookup_in_remote_parent(info, dev,
693                                                              fid, id);
694                         if (result == 0) {
695                                 cached = true;
696                                 goto iget;
697                         }
698
699                         result = saved;
700                 }
701
702                 GOTO(out, result);
703         }
704
705         obj->oo_inode = inode;
706         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
707
708         result = osd_check_lma(env, obj);
709         if (result != 0) {
710                 if (result == -ENODATA) {
711                         if (cached) {
712                                 result = osd_oi_lookup(info, dev, fid, id,
713                                                        OI_CHECK_FLD);
714                                 if (result != 0) {
715                                         /* result == -ENOENT means that the OI
716                                          * mapping has been removed by race,
717                                          * the target inode belongs to other
718                                          * object.
719                                          *
720                                          * Others error also can be returned
721                                          * directly. */
722                                         iput(inode);
723                                         obj->oo_inode = NULL;
724                                         GOTO(out, result);
725                                 } else {
726                                         /* result == 0 means the cached OI
727                                          * mapping is still in the OI file,
728                                          * the target the inode is valid. */
729                                 }
730                         } else {
731                                 /* The current OI mapping is from the OI file,
732                                  * since the inode has been found via
733                                  * osd_iget_check(), no need recheck OI. */
734                         }
735
736                         goto found;
737                 }
738
739                 iput(inode);
740                 obj->oo_inode = NULL;
741                 if (result != -EREMCHG)
742                         GOTO(out, result);
743
744                 if (cached) {
745                         result = osd_oi_lookup(info, dev, fid, id,
746                                                OI_CHECK_FLD);
747                         /* result == -ENOENT means the cached OI mapping
748                          * has been removed from the OI file by race,
749                          * above target inode belongs to other object.
750                          *
751                          * Others error also can be returned directly. */
752                         if (result != 0)
753                                 GOTO(out, result);
754
755                         /* result == 0, goto trigger */
756                 } else {
757                         /* The current OI mapping is from the OI file,
758                          * since the inode has been found via
759                          * osd_iget_check(), no need recheck OI. */
760                 }
761
762                 goto trigger;
763         }
764
765 found:
766         obj->oo_compat_dot_created = 1;
767         obj->oo_compat_dotdot_created = 1;
768
769         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
770                 GOTO(out, result = 0);
771
772         LASSERT(obj->oo_hl_head == NULL);
773         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
774         if (obj->oo_hl_head == NULL) {
775                 obj->oo_inode = NULL;
776                 iput(inode);
777                 GOTO(out, result = -ENOMEM);
778         }
779         GOTO(out, result = 0);
780
781 out:
782         if (result != 0 && cached)
783                 fid_zero(&oic->oic_fid);
784
785         LINVRNT(osd_invariant(obj));
786         return result;
787 }
788
789 /*
790  * Concurrency: shouldn't matter.
791  */
792 static void osd_object_init0(struct osd_object *obj)
793 {
794         LASSERT(obj->oo_inode != NULL);
795         obj->oo_dt.do_body_ops = &osd_body_ops;
796         obj->oo_dt.do_lu.lo_header->loh_attr |=
797                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
798 }
799
800 /*
801  * Concurrency: no concurrent access is possible that early in object
802  * life-cycle.
803  */
804 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
805                            const struct lu_object_conf *conf)
806 {
807         struct osd_object *obj = osd_obj(l);
808         int result;
809
810         LINVRNT(osd_invariant(obj));
811
812         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
813                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
814                 l->lo_header->loh_attr |= LOHA_EXISTS;
815                 return 0;
816         }
817
818         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
819         obj->oo_dt.do_body_ops = &osd_body_ops_new;
820         if (result == 0 && obj->oo_inode != NULL)
821                 osd_object_init0(obj);
822
823         LINVRNT(osd_invariant(obj));
824         return result;
825 }
826
827 /*
828  * Concurrency: no concurrent access is possible that late in object
829  * life-cycle.
830  */
831 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
832 {
833         struct osd_object *obj = osd_obj(l);
834
835         LINVRNT(osd_invariant(obj));
836
837         dt_object_fini(&obj->oo_dt);
838         if (obj->oo_hl_head != NULL)
839                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
840         OBD_FREE_PTR(obj);
841 }
842
843 /*
844  * Concurrency: no concurrent access is possible that late in object
845  * life-cycle.
846  */
847 static void osd_index_fini(struct osd_object *o)
848 {
849         struct iam_container *bag;
850
851         if (o->oo_dir != NULL) {
852                 bag = &o->oo_dir->od_container;
853                 if (o->oo_inode != NULL) {
854                         if (bag->ic_object == o->oo_inode)
855                                 iam_container_fini(bag);
856                 }
857                 OBD_FREE_PTR(o->oo_dir);
858                 o->oo_dir = NULL;
859         }
860 }
861
862 /*
863  * Concurrency: no concurrent access is possible that late in object
864  * life-cycle (for all existing callers, that is. New callers have to provide
865  * their own locking.)
866  */
867 static int osd_inode_unlinked(const struct inode *inode)
868 {
869         return inode->i_nlink == 0;
870 }
871
872 enum {
873         OSD_TXN_OI_DELETE_CREDITS    = 20,
874         OSD_TXN_INODE_DELETE_CREDITS = 20
875 };
876
877 /*
878  * Journal
879  */
880
881 #if OSD_THANDLE_STATS
882 /**
883  * Set time when the handle is allocated
884  */
885 static void osd_th_alloced(struct osd_thandle *oth)
886 {
887         oth->oth_alloced = cfs_time_current();
888 }
889
890 /**
891  * Set time when the handle started
892  */
893 static void osd_th_started(struct osd_thandle *oth)
894 {
895         oth->oth_started = cfs_time_current();
896 }
897
898 /**
899  * Helper function to convert time interval to microseconds packed in
900  * long int.
901  */
902 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
903 {
904         struct timeval val;
905
906         cfs_duration_usec(cfs_time_sub(end, start), &val);
907         return val.tv_sec * 1000000 + val.tv_usec;
908 }
909
910 /**
911  * Check whether the we deal with this handle for too long.
912  */
913 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
914                                 cfs_time_t alloced, cfs_time_t started,
915                                 cfs_time_t closed)
916 {
917         cfs_time_t now = cfs_time_current();
918
919         LASSERT(dev != NULL);
920
921         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
922                             interval_to_usec(alloced, started));
923         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
924                             interval_to_usec(started, closed));
925         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
926                             interval_to_usec(closed, now));
927
928         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
929                 CWARN("transaction handle %p was open for too long: "
930                       "now "CFS_TIME_T" ,"
931                       "alloced "CFS_TIME_T" ,"
932                       "started "CFS_TIME_T" ,"
933                       "closed "CFS_TIME_T"\n",
934                       oth, now, alloced, started, closed);
935                 libcfs_debug_dumpstack(NULL);
936         }
937 }
938
939 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
940 {                                                                       \
941         cfs_time_t __closed = cfs_time_current();                       \
942         cfs_time_t __alloced = oth->oth_alloced;                        \
943         cfs_time_t __started = oth->oth_started;                        \
944                                                                         \
945         expr;                                                           \
946         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
947 }
948
949 #else /* OSD_THANDLE_STATS */
950
951 #define osd_th_alloced(h)                  do {} while(0)
952 #define osd_th_started(h)                  do {} while(0)
953 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
954
955 #endif /* OSD_THANDLE_STATS */
956
957 /*
958  * Concurrency: doesn't access mutable data.
959  */
960 static int osd_param_is_not_sane(const struct osd_device *dev,
961                                  const struct thandle *th)
962 {
963         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
964
965         return oh->ot_credits > osd_transaction_size(dev);
966 }
967
968 /*
969  * Concurrency: shouldn't matter.
970  */
971 static void osd_trans_commit_cb(struct super_block *sb,
972                                 struct ldiskfs_journal_cb_entry *jcb, int error)
973 {
974         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
975         struct thandle     *th  = &oh->ot_super;
976         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
977         struct dt_txn_commit_cb *dcb, *tmp;
978
979         LASSERT(oh->ot_handle == NULL);
980
981         if (error)
982                 CERROR("transaction @0x%p commit error: %d\n", th, error);
983
984         dt_txn_hook_commit(th);
985
986         /* call per-transaction callbacks if any */
987         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
988                                  dcb_linkage) {
989                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
990                          "commit callback entry: magic=%x name='%s'\n",
991                          dcb->dcb_magic, dcb->dcb_name);
992                 list_del_init(&dcb->dcb_linkage);
993                 dcb->dcb_func(NULL, th, dcb, error);
994         }
995
996         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
997         lu_device_put(lud);
998         th->th_dev = NULL;
999
1000         lu_context_exit(&th->th_ctx);
1001         lu_context_fini(&th->th_ctx);
1002         OBD_FREE_PTR(oh);
1003 }
1004
1005 static struct thandle *osd_trans_create(const struct lu_env *env,
1006                                         struct dt_device *d)
1007 {
1008         struct osd_thread_info  *oti = osd_oti_get(env);
1009         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
1010         struct osd_thandle      *oh;
1011         struct thandle          *th;
1012         ENTRY;
1013
1014         /* on pending IO in this thread should left from prev. request */
1015         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1016
1017         th = ERR_PTR(-ENOMEM);
1018         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
1019         if (oh != NULL) {
1020                 oh->ot_quota_trans = &oti->oti_quota_trans;
1021                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1022                 th = &oh->ot_super;
1023                 th->th_dev = d;
1024                 th->th_result = 0;
1025                 th->th_tags = LCT_TX_HANDLE;
1026                 oh->ot_credits = 0;
1027                 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1028                 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1029                 osd_th_alloced(oh);
1030
1031                 memset(oti->oti_declare_ops, 0,
1032                        sizeof(oti->oti_declare_ops));
1033                 memset(oti->oti_declare_ops_cred, 0,
1034                        sizeof(oti->oti_declare_ops_cred));
1035                 memset(oti->oti_declare_ops_used, 0,
1036                        sizeof(oti->oti_declare_ops_used));
1037         }
1038         RETURN(th);
1039 }
1040
1041 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1042 {
1043         struct osd_thread_info  *oti = osd_oti_get(env);
1044         struct osd_thandle      *oh;
1045
1046         oh = container_of0(th, struct osd_thandle, ot_super);
1047         LASSERT(oh != NULL);
1048
1049         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1050               oti->oti_declare_ops[OSD_OT_CREATE],
1051               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1052               oti->oti_declare_ops_used[OSD_OT_CREATE],
1053               oti->oti_declare_ops[OSD_OT_DESTROY],
1054               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1055               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1056         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1057               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1058               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1059               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1060               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1061               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1062               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1063         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1064               oti->oti_declare_ops[OSD_OT_WRITE],
1065               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1066               oti->oti_declare_ops_used[OSD_OT_WRITE],
1067               oti->oti_declare_ops[OSD_OT_PUNCH],
1068               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1069               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1070               oti->oti_declare_ops[OSD_OT_QUOTA],
1071               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1072               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1073         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1074               oti->oti_declare_ops[OSD_OT_INSERT],
1075               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1076               oti->oti_declare_ops_used[OSD_OT_INSERT],
1077               oti->oti_declare_ops[OSD_OT_DELETE],
1078               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1079               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1080         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1081               oti->oti_declare_ops[OSD_OT_REF_ADD],
1082               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1083               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1084               oti->oti_declare_ops[OSD_OT_REF_DEL],
1085               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1086               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1087 }
1088
1089 /*
1090  * Concurrency: shouldn't matter.
1091  */
1092 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1093                            struct thandle *th)
1094 {
1095         struct osd_thread_info *oti = osd_oti_get(env);
1096         struct osd_device  *dev = osd_dt_dev(d);
1097         handle_t           *jh;
1098         struct osd_thandle *oh;
1099         int rc;
1100
1101         ENTRY;
1102
1103         LASSERT(current->journal_info == NULL);
1104
1105         oh = container_of0(th, struct osd_thandle, ot_super);
1106         LASSERT(oh != NULL);
1107         LASSERT(oh->ot_handle == NULL);
1108
1109         rc = dt_txn_hook_start(env, d, th);
1110         if (rc != 0)
1111                 GOTO(out, rc);
1112
1113         if (unlikely(osd_param_is_not_sane(dev, th))) {
1114                 static unsigned long last_printed;
1115                 static int last_credits;
1116
1117                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1118                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1119                       oh->ot_credits,
1120                       osd_journal(dev)->j_max_transaction_buffers);
1121
1122                 osd_trans_dump_creds(env, th);
1123
1124                 if (last_credits != oh->ot_credits &&
1125                     time_after(jiffies, last_printed +
1126                                msecs_to_jiffies(60 * MSEC_PER_SEC))) {
1127                         libcfs_debug_dumpstack(NULL);
1128                         last_credits = oh->ot_credits;
1129                         last_printed = jiffies;
1130                 }
1131                 /* XXX Limit the credits to 'max_transaction_buffers', and
1132                  *     let the underlying filesystem to catch the error if
1133                  *     we really need so many credits.
1134                  *
1135                  *     This should be removed when we can calculate the
1136                  *     credits precisely. */
1137                 oh->ot_credits = osd_transaction_size(dev);
1138         }
1139
1140         /*
1141          * XXX temporary stuff. Some abstraction layer should
1142          * be used.
1143          */
1144         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1145         osd_th_started(oh);
1146         if (!IS_ERR(jh)) {
1147                 oh->ot_handle = jh;
1148                 LASSERT(oti->oti_txns == 0);
1149                 lu_context_init(&th->th_ctx, th->th_tags);
1150                 lu_context_enter(&th->th_ctx);
1151
1152                 lu_device_get(&d->dd_lu_dev);
1153                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1154                               "osd-tx", th);
1155                 oti->oti_txns++;
1156                 rc = 0;
1157         } else {
1158                 rc = PTR_ERR(jh);
1159         }
1160 out:
1161         RETURN(rc);
1162 }
1163
1164 static int osd_seq_exists(const struct lu_env *env,
1165                           struct osd_device *osd, u64 seq)
1166 {
1167         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1168         struct seq_server_site  *ss = osd_seq_site(osd);
1169         int                     rc;
1170         ENTRY;
1171
1172         LASSERT(ss != NULL);
1173         LASSERT(ss->ss_server_fld != NULL);
1174
1175         rc = osd_fld_lookup(env, osd, seq, range);
1176         if (rc != 0) {
1177                 if (rc != -ENOENT)
1178                         CERROR("%s: can't lookup FLD sequence "LPX64
1179                                ": rc = %d\n", osd_name(osd), seq, rc);
1180                 RETURN(0);
1181         }
1182
1183         RETURN(ss->ss_node_id == range->lsr_index);
1184 }
1185
1186 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1187 {
1188         struct dt_txn_commit_cb *dcb;
1189         struct dt_txn_commit_cb *tmp;
1190
1191         /* call per-transaction stop callbacks if any */
1192         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1193                                  dcb_linkage) {
1194                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1195                          "commit callback entry: magic=%x name='%s'\n",
1196                          dcb->dcb_magic, dcb->dcb_name);
1197                 list_del_init(&dcb->dcb_linkage);
1198                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1199         }
1200 }
1201
1202 /*
1203  * Concurrency: shouldn't matter.
1204  */
1205 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1206                           struct thandle *th)
1207 {
1208         int                     rc = 0, remove_agents = 0;
1209         struct osd_thandle     *oh;
1210         struct osd_thread_info *oti = osd_oti_get(env);
1211         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1212         struct osd_device      *osd = osd_dt_dev(th->th_dev);
1213         struct qsd_instance    *qsd = osd->od_quota_slave;
1214         struct lquota_trans    *qtrans;
1215         ENTRY;
1216
1217         oh = container_of0(th, struct osd_thandle, ot_super);
1218
1219         remove_agents = oh->ot_remove_agents;
1220
1221         qtrans = oh->ot_quota_trans;
1222         oh->ot_quota_trans = NULL;
1223
1224         if (oh->ot_handle != NULL) {
1225                 handle_t *hdl = oh->ot_handle;
1226
1227                 /*
1228                  * add commit callback
1229                  * notice we don't do this in osd_trans_start()
1230                  * as underlying transaction can change during truncate
1231                  */
1232                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1233                                          &oh->ot_jcb);
1234
1235                 LASSERT(oti->oti_txns == 1);
1236                 oti->oti_txns--;
1237
1238                 rc = dt_txn_hook_stop(env, th);
1239                 if (rc != 0)
1240                         CERROR("%s: failed in transaction hook: rc = %d\n",
1241                                osd_name(osd), rc);
1242
1243                 osd_trans_stop_cb(oh, rc);
1244                 /* hook functions might modify th_sync */
1245                 hdl->h_sync = th->th_sync;
1246
1247                 oh->ot_handle = NULL;
1248                 OSD_CHECK_SLOW_TH(oh, osd, rc = ldiskfs_journal_stop(hdl));
1249                 if (rc != 0)
1250                         CERROR("%s: failed to stop transaction: rc = %d\n",
1251                                osd_name(osd), rc);
1252         } else {
1253                 osd_trans_stop_cb(oh, th->th_result);
1254                 OBD_FREE_PTR(oh);
1255         }
1256
1257         /* inform the quota slave device that the transaction is stopping */
1258         qsd_op_end(env, qsd, qtrans);
1259
1260         /* as we want IO to journal and data IO be concurrent, we don't block
1261          * awaiting data IO completion in osd_do_bio(), instead we wait here
1262          * once transaction is submitted to the journal. all reqular requests
1263          * don't do direct IO (except read/write), thus this wait_event becomes
1264          * no-op for them.
1265          *
1266          * IMPORTANT: we have to wait till any IO submited by the thread is
1267          * completed otherwise iobuf may be corrupted by different request
1268          */
1269         wait_event(iobuf->dr_wait,
1270                        atomic_read(&iobuf->dr_numreqs) == 0);
1271         osd_fini_iobuf(osd, iobuf);
1272         if (!rc)
1273                 rc = iobuf->dr_error;
1274
1275         if (unlikely(remove_agents != 0))
1276                 osd_process_scheduled_agent_removals(env, osd);
1277
1278         RETURN(rc);
1279 }
1280
1281 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1282 {
1283         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1284                                                ot_super);
1285
1286         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1287         LASSERT(&dcb->dcb_func != NULL);
1288         if (dcb->dcb_flags & DCB_TRANS_STOP)
1289                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1290         else
1291                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1292
1293         return 0;
1294 }
1295
1296 /*
1297  * Called just before object is freed. Releases all resources except for
1298  * object itself (that is released by osd_object_free()).
1299  *
1300  * Concurrency: no concurrent access is possible that late in object
1301  * life-cycle.
1302  */
1303 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1304 {
1305         struct osd_object *obj   = osd_obj(l);
1306         struct inode      *inode = obj->oo_inode;
1307
1308         LINVRNT(osd_invariant(obj));
1309
1310         /*
1311          * If object is unlinked remove fid->ino mapping from object index.
1312          */
1313
1314         osd_index_fini(obj);
1315         if (inode != NULL) {
1316                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1317                 qid_t                    uid = i_uid_read(inode);
1318                 qid_t                    gid = i_gid_read(inode);
1319
1320                 iput(inode);
1321                 obj->oo_inode = NULL;
1322
1323                 if (qsd != NULL) {
1324                         struct osd_thread_info  *info = osd_oti_get(env);
1325                         struct lquota_id_info   *qi = &info->oti_qi;
1326
1327                         /* Release granted quota to master if necessary */
1328                         qi->lqi_id.qid_uid = uid;
1329                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1330
1331                         qi->lqi_id.qid_uid = gid;
1332                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1333                 }
1334         }
1335 }
1336
1337 /*
1338  * Concurrency: ->loo_object_release() is called under site spin-lock.
1339  */
1340 static void osd_object_release(const struct lu_env *env,
1341                                struct lu_object *l)
1342 {
1343 }
1344
1345 /*
1346  * Concurrency: shouldn't matter.
1347  */
1348 static int osd_object_print(const struct lu_env *env, void *cookie,
1349                             lu_printer_t p, const struct lu_object *l)
1350 {
1351         struct osd_object *o = osd_obj(l);
1352         struct iam_descr  *d;
1353
1354         if (o->oo_dir != NULL)
1355                 d = o->oo_dir->od_container.ic_descr;
1356         else
1357                 d = NULL;
1358         return (*p)(env, cookie,
1359                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1360                     o, o->oo_inode,
1361                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1362                     o->oo_inode ? o->oo_inode->i_generation : 0,
1363                     d ? d->id_ops->id_name : "plain");
1364 }
1365
1366 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1367
1368 /*
1369  * Concurrency: shouldn't matter.
1370  */
1371 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1372                struct obd_statfs *sfs)
1373 {
1374         struct osd_device  *osd = osd_dt_dev(d);
1375         struct super_block *sb = osd_sb(osd);
1376         struct kstatfs     *ksfs;
1377         int result = 0;
1378
1379         if (unlikely(osd->od_mnt == NULL))
1380                 return -EINPROGRESS;
1381
1382         /* osd_lproc.c call this without env, allocate ksfs for that case */
1383         if (unlikely(env == NULL)) {
1384                 OBD_ALLOC_PTR(ksfs);
1385                 if (ksfs == NULL)
1386                         return -ENOMEM;
1387         } else {
1388                 ksfs = &osd_oti_get(env)->oti_ksfs;
1389         }
1390
1391         spin_lock(&osd->od_osfs_lock);
1392         result = sb->s_op->statfs(sb->s_root, ksfs);
1393         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1394                 statfs_pack(sfs, ksfs);
1395                 if (unlikely(sb->s_flags & MS_RDONLY))
1396                         sfs->os_state = OS_STATE_READONLY;
1397                 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
1398                                               LDISKFS_FEATURE_INCOMPAT_EXTENTS))
1399                         sfs->os_maxbytes = sb->s_maxbytes;
1400                 else
1401                         sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
1402         }
1403         spin_unlock(&osd->od_osfs_lock);
1404
1405         if (unlikely(env == NULL))
1406                 OBD_FREE_PTR(ksfs);
1407
1408         /* Reserve a small amount of space for local objects like last_rcvd,
1409          * llog, quota files, ... */
1410         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1411                 sfs->os_bavail = 0;
1412         } else {
1413                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1414                 /** Take out metadata overhead for indirect blocks */
1415                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1416         }
1417
1418         return result;
1419 }
1420
1421 /**
1422  * Estimate space needed for file creations. We assume the largest filename
1423  * which is 2^64 - 1, hence a filename of 20 chars.
1424  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1425  */
1426 #ifdef __LDISKFS_DIR_REC_LEN
1427 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1428 #else
1429 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1430 #endif
1431
1432 /*
1433  * Concurrency: doesn't access mutable data.
1434  */
1435 static void osd_conf_get(const struct lu_env *env,
1436                          const struct dt_device *dev,
1437                          struct dt_device_param *param)
1438 {
1439         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1440         int                ea_overhead;
1441
1442         /*
1443          * XXX should be taken from not-yet-existing fs abstraction layer.
1444          */
1445         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1446         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1447         param->ddp_block_shift  = sb->s_blocksize_bits;
1448         param->ddp_mount_type     = LDD_MT_LDISKFS;
1449         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
1450                 param->ddp_maxbytes = sb->s_maxbytes;
1451         else
1452                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
1453         /* Overhead estimate should be fairly accurate, so we really take a tiny
1454          * error margin which also avoids fragmenting the filesystem too much */
1455         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1456         /* inode are statically allocated, so per-inode space consumption
1457          * is the space consumed by the directory entry */
1458         param->ddp_inodespace     = PER_OBJ_USAGE;
1459         /* per-fragment overhead to be used by the client code */
1460         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1461         param->ddp_mntopts      = 0;
1462         if (test_opt(sb, XATTR_USER))
1463                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1464         if (test_opt(sb, POSIX_ACL))
1465                 param->ddp_mntopts |= MNTOPT_ACL;
1466
1467         /* LOD might calculate the max stripe count based on max_ea_size,
1468          * so we need take account in the overhead as well,
1469          * xattr_header + magic + xattr_entry_head */
1470         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1471                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1472
1473 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1474         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1475                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1476                                                                 ea_overhead;
1477         else
1478 #endif
1479                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1480 }
1481
1482 /*
1483  * Concurrency: shouldn't matter.
1484  */
1485 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1486 {
1487         int rc;
1488
1489         CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1490
1491         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1492
1493         CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
1494                LUSTRE_OSD_LDISKFS_NAME, rc);
1495
1496         return rc;
1497 }
1498
1499 /**
1500  * Start commit for OSD device.
1501  *
1502  * An implementation of dt_commit_async method for OSD device.
1503  * Asychronously starts underlayng fs sync and thereby a transaction
1504  * commit.
1505  *
1506  * \param env environment
1507  * \param d dt device
1508  *
1509  * \see dt_device_operations
1510  */
1511 static int osd_commit_async(const struct lu_env *env,
1512                             struct dt_device *d)
1513 {
1514         struct super_block *s = osd_sb(osd_dt_dev(d));
1515         ENTRY;
1516
1517         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1518         RETURN(s->s_op->sync_fs(s, 0));
1519 }
1520
1521 /*
1522  * Concurrency: shouldn't matter.
1523  */
1524
1525 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1526 {
1527         struct super_block *sb = osd_sb(osd_dt_dev(d));
1528         struct block_device *dev = sb->s_bdev;
1529 #ifdef HAVE_DEV_SET_RDONLY
1530         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1531         int rc = 0;
1532 #else
1533         int rc = -EOPNOTSUPP;
1534 #endif
1535         ENTRY;
1536
1537 #ifdef HAVE_DEV_SET_RDONLY
1538         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1539
1540         if (jdev && (jdev != dev)) {
1541                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1542                        (long)jdev);
1543                 dev_set_rdonly(jdev);
1544         }
1545         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1546         dev_set_rdonly(dev);
1547 #else
1548         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1549                osd_dt_dev(d)->od_svname, (long)dev, rc);
1550 #endif
1551         RETURN(rc);
1552 }
1553
1554 /**
1555  * Note: we do not count into QUOTA here.
1556  * If we mount with --data_journal we may need more.
1557  */
1558 const int osd_dto_credits_noquota[DTO_NR] = {
1559         /**
1560          * Insert.
1561          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1562          * SINGLEDATA_TRANS_BLOCKS(8)
1563          * XXX Note: maybe iam need more, since iam have more level than
1564          *           EXT3 htree.
1565          */
1566         [DTO_INDEX_INSERT]  = 16,
1567         /**
1568          * Delete
1569          * just modify a single entry, probably merge few within a block
1570          */
1571         [DTO_INDEX_DELETE]  = 1,
1572         /**
1573          * Used for OI scrub
1574          */
1575         [DTO_INDEX_UPDATE]  = 16,
1576         /**
1577          * 4(inode, inode bits, groups, GDT)
1578          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1579          */
1580         [DTO_OBJECT_CREATE] = 4,
1581         /**
1582          * 4(inode, inode bits, groups, GDT)
1583          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1584          */
1585         [DTO_OBJECT_DELETE] = 4,
1586         /**
1587          * Attr set credits (inode)
1588          */
1589         [DTO_ATTR_SET_BASE] = 1,
1590         /**
1591          * Xattr set. The same as xattr of EXT3.
1592          * DATA_TRANS_BLOCKS(14)
1593          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1594          * are also counted in. Do not know why?
1595          */
1596         [DTO_XATTR_SET]     = 14,
1597         /**
1598          * credits for inode change during write.
1599          */
1600         [DTO_WRITE_BASE]    = 3,
1601         /**
1602          * credits for single block write.
1603          */
1604         [DTO_WRITE_BLOCK]   = 14,
1605         /**
1606          * Attr set credits for chown.
1607          * This is extra credits for setattr, and it is null without quota
1608          */
1609         [DTO_ATTR_SET_CHOWN] = 0
1610 };
1611
1612 static const struct dt_device_operations osd_dt_ops = {
1613         .dt_root_get       = osd_root_get,
1614         .dt_statfs         = osd_statfs,
1615         .dt_trans_create   = osd_trans_create,
1616         .dt_trans_start    = osd_trans_start,
1617         .dt_trans_stop     = osd_trans_stop,
1618         .dt_trans_cb_add   = osd_trans_cb_add,
1619         .dt_conf_get       = osd_conf_get,
1620         .dt_sync           = osd_sync,
1621         .dt_ro             = osd_ro,
1622         .dt_commit_async   = osd_commit_async,
1623 };
1624
1625 static void osd_object_read_lock(const struct lu_env *env,
1626                                  struct dt_object *dt, unsigned role)
1627 {
1628         struct osd_object *obj = osd_dt_obj(dt);
1629         struct osd_thread_info *oti = osd_oti_get(env);
1630
1631         LINVRNT(osd_invariant(obj));
1632
1633         LASSERT(obj->oo_owner != env);
1634         down_read_nested(&obj->oo_sem, role);
1635
1636         LASSERT(obj->oo_owner == NULL);
1637         oti->oti_r_locks++;
1638 }
1639
1640 static void osd_object_write_lock(const struct lu_env *env,
1641                                   struct dt_object *dt, unsigned role)
1642 {
1643         struct osd_object *obj = osd_dt_obj(dt);
1644         struct osd_thread_info *oti = osd_oti_get(env);
1645
1646         LINVRNT(osd_invariant(obj));
1647
1648         LASSERT(obj->oo_owner != env);
1649         down_write_nested(&obj->oo_sem, role);
1650
1651         LASSERT(obj->oo_owner == NULL);
1652         obj->oo_owner = env;
1653         oti->oti_w_locks++;
1654 }
1655
1656 static void osd_object_read_unlock(const struct lu_env *env,
1657                                    struct dt_object *dt)
1658 {
1659         struct osd_object *obj = osd_dt_obj(dt);
1660         struct osd_thread_info *oti = osd_oti_get(env);
1661
1662         LINVRNT(osd_invariant(obj));
1663
1664         LASSERT(oti->oti_r_locks > 0);
1665         oti->oti_r_locks--;
1666         up_read(&obj->oo_sem);
1667 }
1668
1669 static void osd_object_write_unlock(const struct lu_env *env,
1670                                     struct dt_object *dt)
1671 {
1672         struct osd_object *obj = osd_dt_obj(dt);
1673         struct osd_thread_info *oti = osd_oti_get(env);
1674
1675         LINVRNT(osd_invariant(obj));
1676
1677         LASSERT(obj->oo_owner == env);
1678         LASSERT(oti->oti_w_locks > 0);
1679         oti->oti_w_locks--;
1680         obj->oo_owner = NULL;
1681         up_write(&obj->oo_sem);
1682 }
1683
1684 static int osd_object_write_locked(const struct lu_env *env,
1685                                    struct dt_object *dt)
1686 {
1687         struct osd_object *obj = osd_dt_obj(dt);
1688
1689         LINVRNT(osd_invariant(obj));
1690
1691         return obj->oo_owner == env;
1692 }
1693
1694 static struct timespec *osd_inode_time(const struct lu_env *env,
1695                                        struct inode *inode, __u64 seconds)
1696 {
1697         struct osd_thread_info  *oti = osd_oti_get(env);
1698         struct timespec         *t   = &oti->oti_time;
1699
1700         t->tv_sec = seconds;
1701         t->tv_nsec = 0;
1702         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1703         return t;
1704 }
1705
1706
1707 static void osd_inode_getattr(const struct lu_env *env,
1708                               struct inode *inode, struct lu_attr *attr)
1709 {
1710         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1711                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1712                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1713                            LA_TYPE;
1714
1715         attr->la_atime   = LTIME_S(inode->i_atime);
1716         attr->la_mtime   = LTIME_S(inode->i_mtime);
1717         attr->la_ctime   = LTIME_S(inode->i_ctime);
1718         attr->la_mode    = inode->i_mode;
1719         attr->la_size    = i_size_read(inode);
1720         attr->la_blocks  = inode->i_blocks;
1721         attr->la_uid     = i_uid_read(inode);
1722         attr->la_gid     = i_gid_read(inode);
1723         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1724         attr->la_nlink   = inode->i_nlink;
1725         attr->la_rdev    = inode->i_rdev;
1726         attr->la_blksize = 1 << inode->i_blkbits;
1727         attr->la_blkbits = inode->i_blkbits;
1728 }
1729
1730 static int osd_attr_get(const struct lu_env *env,
1731                         struct dt_object *dt,
1732                         struct lu_attr *attr)
1733 {
1734         struct osd_object *obj = osd_dt_obj(dt);
1735
1736         if (!dt_object_exists(dt))
1737                 return -ENOENT;
1738
1739         LASSERT(!dt_object_remote(dt));
1740         LINVRNT(osd_invariant(obj));
1741
1742         spin_lock(&obj->oo_guard);
1743         osd_inode_getattr(env, obj->oo_inode, attr);
1744         spin_unlock(&obj->oo_guard);
1745         return 0;
1746 }
1747
1748 static int osd_declare_attr_set(const struct lu_env *env,
1749                                 struct dt_object *dt,
1750                                 const struct lu_attr *attr,
1751                                 struct thandle *handle)
1752 {
1753         struct osd_thandle     *oh;
1754         struct osd_object      *obj;
1755         struct osd_thread_info *info = osd_oti_get(env);
1756         struct lquota_id_info  *qi = &info->oti_qi;
1757         qid_t                   uid;
1758         qid_t                   gid;
1759         long long               bspace;
1760         int                     rc = 0;
1761         bool                    enforce;
1762         ENTRY;
1763
1764         LASSERT(dt != NULL);
1765         LASSERT(handle != NULL);
1766
1767         obj = osd_dt_obj(dt);
1768         LASSERT(osd_invariant(obj));
1769
1770         oh = container_of0(handle, struct osd_thandle, ot_super);
1771         LASSERT(oh->ot_handle == NULL);
1772
1773         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1774                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1775
1776         if (attr == NULL || obj->oo_inode == NULL)
1777                 RETURN(rc);
1778
1779         bspace   = obj->oo_inode->i_blocks;
1780         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1781         bspace   = toqb(bspace);
1782
1783         /* Changing ownership is always preformed by super user, it should not
1784          * fail with EDQUOT.
1785          *
1786          * We still need to call the osd_declare_qid() to calculate the journal
1787          * credits for updating quota accounting files and to trigger quota
1788          * space adjustment once the operation is completed.*/
1789         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1790                 /* USERQUOTA */
1791                 uid = i_uid_read(obj->oo_inode);
1792                 qi->lqi_type = USRQUOTA;
1793                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1794                 /* inode accounting */
1795                 qi->lqi_is_blk = false;
1796
1797                 /* one more inode for the new uid ... */
1798                 qi->lqi_id.qid_uid = attr->la_uid;
1799                 qi->lqi_space      = 1;
1800                 /* Reserve credits for the new uid */
1801                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1802                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1803                         rc = 0;
1804                 if (rc)
1805                         RETURN(rc);
1806
1807                 /* and one less inode for the current uid */
1808                 qi->lqi_id.qid_uid = uid;
1809                 qi->lqi_space      = -1;
1810                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1811                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1812                         rc = 0;
1813                 if (rc)
1814                         RETURN(rc);
1815
1816                 /* block accounting */
1817                 qi->lqi_is_blk = true;
1818
1819                 /* more blocks for the new uid ... */
1820                 qi->lqi_id.qid_uid = attr->la_uid;
1821                 qi->lqi_space      = bspace;
1822                 /*
1823                  * Credits for the new uid has been reserved, re-use "obj"
1824                  * to save credit reservation.
1825                  */
1826                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1827                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1828                         rc = 0;
1829                 if (rc)
1830                         RETURN(rc);
1831
1832                 /* and finally less blocks for the current uid */
1833                 qi->lqi_id.qid_uid = uid;
1834                 qi->lqi_space      = -bspace;
1835                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1836                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1837                         rc = 0;
1838                 if (rc)
1839                         RETURN(rc);
1840
1841                 /* GROUP QUOTA */
1842                 gid = i_gid_read(obj->oo_inode);
1843                 qi->lqi_type = GRPQUOTA;
1844                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1845
1846                 /* inode accounting */
1847                 qi->lqi_is_blk = false;
1848
1849                 /* one more inode for the new gid ... */
1850                 qi->lqi_id.qid_gid = attr->la_gid;
1851                 qi->lqi_space      = 1;
1852                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1853                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1854                         rc = 0;
1855                 if (rc)
1856                         RETURN(rc);
1857
1858                 /* and one less inode for the current gid */
1859                 qi->lqi_id.qid_gid = gid;
1860                 qi->lqi_space      = -1;
1861                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1862                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1863                         rc = 0;
1864                 if (rc)
1865                         RETURN(rc);
1866
1867                 /* block accounting */
1868                 qi->lqi_is_blk = true;
1869
1870                 /* more blocks for the new gid ... */
1871                 qi->lqi_id.qid_gid = attr->la_gid;
1872                 qi->lqi_space      = bspace;
1873                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1874                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1875                         rc = 0;
1876                 if (rc)
1877                         RETURN(rc);
1878
1879                 /* and finally less blocks for the current gid */
1880                 qi->lqi_id.qid_gid = gid;
1881                 qi->lqi_space      = -bspace;
1882                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1883                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1884                         rc = 0;
1885                 if (rc)
1886                         RETURN(rc);
1887         }
1888
1889         RETURN(rc);
1890 }
1891
1892 static int osd_inode_setattr(const struct lu_env *env,
1893                              struct inode *inode, const struct lu_attr *attr)
1894 {
1895         __u64 bits = attr->la_valid;
1896
1897         /* Only allow set size for regular file */
1898         if (!S_ISREG(inode->i_mode))
1899                 bits &= ~(LA_SIZE | LA_BLOCKS);
1900
1901         if (bits == 0)
1902                 return 0;
1903
1904         if (bits & LA_ATIME)
1905                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1906         if (bits & LA_CTIME)
1907                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1908         if (bits & LA_MTIME)
1909                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1910         if (bits & LA_SIZE) {
1911                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1912                 i_size_write(inode, attr->la_size);
1913         }
1914
1915 #if 0
1916         /* OSD should not change "i_blocks" which is used by quota.
1917          * "i_blocks" should be changed by ldiskfs only. */
1918         if (bits & LA_BLOCKS)
1919                 inode->i_blocks = attr->la_blocks;
1920 #endif
1921         if (bits & LA_MODE)
1922                 inode->i_mode = (inode->i_mode & S_IFMT) |
1923                                 (attr->la_mode & ~S_IFMT);
1924         if (bits & LA_UID)
1925                 i_uid_write(inode, attr->la_uid);
1926         if (bits & LA_GID)
1927                 i_gid_write(inode, attr->la_gid);
1928         if (bits & LA_NLINK)
1929                 set_nlink(inode, attr->la_nlink);
1930         if (bits & LA_RDEV)
1931                 inode->i_rdev = attr->la_rdev;
1932
1933         if (bits & LA_FLAGS) {
1934                 /* always keep S_NOCMTIME */
1935                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1936                                  S_NOCMTIME;
1937         }
1938         return 0;
1939 }
1940
1941 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1942 {
1943         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1944             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1945                 struct iattr    iattr;
1946                 int             rc;
1947
1948                 ll_vfs_dq_init(inode);
1949                 iattr.ia_valid = 0;
1950                 if (attr->la_valid & LA_UID)
1951                         iattr.ia_valid |= ATTR_UID;
1952                 if (attr->la_valid & LA_GID)
1953                         iattr.ia_valid |= ATTR_GID;
1954                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1955                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1956
1957                 rc = ll_vfs_dq_transfer(inode, &iattr);
1958                 if (rc) {
1959                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1960                                "enforcement enabled on the ldiskfs "
1961                                "filesystem?\n", inode->i_sb->s_id, rc);
1962                         return rc;
1963                 }
1964         }
1965         return 0;
1966 }
1967
1968 static int osd_attr_set(const struct lu_env *env,
1969                         struct dt_object *dt,
1970                         const struct lu_attr *attr,
1971                         struct thandle *handle)
1972 {
1973         struct osd_object *obj = osd_dt_obj(dt);
1974         struct inode      *inode;
1975         int rc;
1976
1977         if (!dt_object_exists(dt))
1978                 return -ENOENT;
1979
1980         LASSERT(handle != NULL);
1981         LASSERT(!dt_object_remote(dt));
1982         LASSERT(osd_invariant(obj));
1983
1984         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1985
1986         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1987                 struct osd_thread_info  *oti  = osd_oti_get(env);
1988                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1989                 struct lu_fid           *fid1 = &oti->oti_fid;
1990                 struct osd_inode_id     *id   = &oti->oti_id;
1991                 struct iam_path_descr   *ipd;
1992                 struct iam_container    *bag;
1993                 struct osd_thandle      *oh;
1994                 int                      rc;
1995
1996                 fid_cpu_to_be(fid1, fid0);
1997                 memset(id, 1, sizeof(*id));
1998                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1999                                   fid0)->oi_dir.od_container;
2000                 ipd = osd_idx_ipd_get(env, bag);
2001                 if (unlikely(ipd == NULL))
2002                         RETURN(-ENOMEM);
2003
2004                 oh = container_of0(handle, struct osd_thandle, ot_super);
2005                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2006                                 (const struct iam_rec *)id, ipd);
2007                 osd_ipd_put(env, bag, ipd);
2008                 return(rc > 0 ? 0 : rc);
2009         }
2010
2011         inode = obj->oo_inode;
2012
2013         rc = osd_quota_transfer(inode, attr);
2014         if (rc)
2015                 return rc;
2016
2017         spin_lock(&obj->oo_guard);
2018         rc = osd_inode_setattr(env, inode, attr);
2019         spin_unlock(&obj->oo_guard);
2020
2021         if (!rc)
2022                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2023
2024         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2025
2026         return rc;
2027 }
2028
2029 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2030                                            struct osd_object *obj,
2031                                            const char *name, const int namelen)
2032 {
2033         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2034 }
2035
2036 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2037                       umode_t mode, struct dt_allocation_hint *hint,
2038                       struct thandle *th)
2039 {
2040         int result;
2041         struct osd_device  *osd = osd_obj2dev(obj);
2042         struct osd_thandle *oth;
2043         struct dt_object   *parent = NULL;
2044         struct inode       *inode;
2045
2046         LINVRNT(osd_invariant(obj));
2047         LASSERT(obj->oo_inode == NULL);
2048         LASSERT(obj->oo_hl_head == NULL);
2049
2050         if (S_ISDIR(mode) && ldiskfs_pdo) {
2051                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2052                 if (obj->oo_hl_head == NULL)
2053                         return -ENOMEM;
2054         }
2055
2056         oth = container_of(th, struct osd_thandle, ot_super);
2057         LASSERT(oth->ot_handle->h_transaction != NULL);
2058
2059         if (hint != NULL && hint->dah_parent != NULL &&
2060             !dt_object_remote(hint->dah_parent))
2061                 parent = hint->dah_parent;
2062
2063         inode = ldiskfs_create_inode(oth->ot_handle,
2064                                      parent ? osd_dt_obj(parent)->oo_inode :
2065                                               osd_sb(osd)->s_root->d_inode,
2066                                      mode);
2067         if (!IS_ERR(inode)) {
2068                 /* Do not update file c/mtime in ldiskfs. */
2069                 inode->i_flags |= S_NOCMTIME;
2070
2071                 /* For new created object, it must be consistent,
2072                  * and it is unnecessary to scrub against it. */
2073                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2074
2075                 obj->oo_inode = inode;
2076                 result = 0;
2077         } else {
2078                 if (obj->oo_hl_head != NULL) {
2079                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2080                         obj->oo_hl_head = NULL;
2081                 }
2082                 result = PTR_ERR(inode);
2083         }
2084         LINVRNT(osd_invariant(obj));
2085         return result;
2086 }
2087
2088 enum {
2089         OSD_NAME_LEN = 255
2090 };
2091
2092 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2093                      struct lu_attr *attr,
2094                      struct dt_allocation_hint *hint,
2095                      struct dt_object_format *dof,
2096                      struct thandle *th)
2097 {
2098         int result;
2099         struct osd_thandle *oth;
2100         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2101
2102         LASSERT(S_ISDIR(attr->la_mode));
2103
2104         oth = container_of(th, struct osd_thandle, ot_super);
2105         LASSERT(oth->ot_handle->h_transaction != NULL);
2106         result = osd_mkfile(info, obj, mode, hint, th);
2107
2108         return result;
2109 }
2110
2111 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2112                         struct lu_attr *attr,
2113                         struct dt_allocation_hint *hint,
2114                         struct dt_object_format *dof,
2115                         struct thandle *th)
2116 {
2117         int result;
2118         struct osd_thandle *oth;
2119         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2120
2121         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2122
2123         LASSERT(S_ISREG(attr->la_mode));
2124
2125         oth = container_of(th, struct osd_thandle, ot_super);
2126         LASSERT(oth->ot_handle->h_transaction != NULL);
2127
2128         result = osd_mkfile(info, obj, mode, hint, th);
2129         if (result == 0) {
2130                 LASSERT(obj->oo_inode != NULL);
2131                 if (feat->dif_flags & DT_IND_VARKEY)
2132                         result = iam_lvar_create(obj->oo_inode,
2133                                                  feat->dif_keysize_max,
2134                                                  feat->dif_ptrsize,
2135                                                  feat->dif_recsize_max,
2136                                                  oth->ot_handle);
2137                 else
2138                         result = iam_lfix_create(obj->oo_inode,
2139                                                  feat->dif_keysize_max,
2140                                                  feat->dif_ptrsize,
2141                                                  feat->dif_recsize_max,
2142                                                  oth->ot_handle);
2143
2144         }
2145         return result;
2146 }
2147
2148 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2149                      struct lu_attr *attr,
2150                      struct dt_allocation_hint *hint,
2151                      struct dt_object_format *dof,
2152                      struct thandle *th)
2153 {
2154         LASSERT(S_ISREG(attr->la_mode));
2155         return osd_mkfile(info, obj, (attr->la_mode &
2156                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2157 }
2158
2159 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2160                      struct lu_attr *attr,
2161                      struct dt_allocation_hint *hint,
2162                      struct dt_object_format *dof,
2163                      struct thandle *th)
2164 {
2165         LASSERT(S_ISLNK(attr->la_mode));
2166         return osd_mkfile(info, obj, (attr->la_mode &
2167                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2168 }
2169
2170 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2171                      struct lu_attr *attr,
2172                      struct dt_allocation_hint *hint,
2173                      struct dt_object_format *dof,
2174                      struct thandle *th)
2175 {
2176         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2177         int result;
2178
2179         LINVRNT(osd_invariant(obj));
2180         LASSERT(obj->oo_inode == NULL);
2181         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2182                 S_ISFIFO(mode) || S_ISSOCK(mode));
2183
2184         result = osd_mkfile(info, obj, mode, hint, th);
2185         if (result == 0) {
2186                 LASSERT(obj->oo_inode != NULL);
2187                 /*
2188                  * This inode should be marked dirty for i_rdev.  Currently
2189                  * that is done in the osd_attr_init().
2190                  */
2191                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2192                                    attr->la_rdev);
2193         }
2194         LINVRNT(osd_invariant(obj));
2195         return result;
2196 }
2197
2198 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2199                               struct lu_attr *,
2200                               struct dt_allocation_hint *hint,
2201                               struct dt_object_format *dof,
2202                               struct thandle *);
2203
2204 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2205 {
2206         osd_obj_type_f result;
2207
2208         switch (type) {
2209         case DFT_DIR:
2210                 result = osd_mkdir;
2211                 break;
2212         case DFT_REGULAR:
2213                 result = osd_mkreg;
2214                 break;
2215         case DFT_SYM:
2216                 result = osd_mksym;
2217                 break;
2218         case DFT_NODE:
2219                 result = osd_mknod;
2220                 break;
2221         case DFT_INDEX:
2222                 result = osd_mk_index;
2223                 break;
2224
2225         default:
2226                 LBUG();
2227                 break;
2228         }
2229         return result;
2230 }
2231
2232
2233 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2234                         struct dt_object *parent, struct dt_object *child,
2235                         umode_t child_mode)
2236 {
2237         LASSERT(ah);
2238
2239         ah->dah_parent = parent;
2240         ah->dah_mode = child_mode;
2241 }
2242
2243 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2244                           struct lu_attr *attr, struct dt_object_format *dof)
2245 {
2246         struct inode   *inode = obj->oo_inode;
2247         __u64           valid = attr->la_valid;
2248         int             result;
2249
2250         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2251
2252         if (dof->dof_type != DFT_NODE)
2253                 attr->la_valid &= ~LA_RDEV;
2254         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2255                 attr->la_valid &= ~LA_ATIME;
2256         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2257                 attr->la_valid &= ~LA_CTIME;
2258         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2259                 attr->la_valid &= ~LA_MTIME;
2260
2261         result = osd_quota_transfer(inode, attr);
2262         if (result)
2263                 return;
2264
2265         if (attr->la_valid != 0) {
2266                 result = osd_inode_setattr(info->oti_env, inode, attr);
2267                 /*
2268                  * The osd_inode_setattr() should always succeed here.  The
2269                  * only error that could be returned is EDQUOT when we are
2270                  * trying to change the UID or GID of the inode. However, this
2271                  * should not happen since quota enforcement is no longer
2272                  * enabled on ldiskfs (lquota takes care of it).
2273                  */
2274                 LASSERTF(result == 0, "%d\n", result);
2275                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2276         }
2277
2278         attr->la_valid = valid;
2279 }
2280
2281 /**
2282  * Helper function for osd_object_create()
2283  *
2284  * \retval 0, on success
2285  */
2286 static int __osd_object_create(struct osd_thread_info *info,
2287                                struct osd_object *obj, struct lu_attr *attr,
2288                                struct dt_allocation_hint *hint,
2289                                struct dt_object_format *dof,
2290                                struct thandle *th)
2291 {
2292         int     result;
2293         __u32   umask;
2294
2295         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
2296
2297         /* we drop umask so that permissions we pass are not affected */
2298         umask = current->fs->umask;
2299         current->fs->umask = 0;
2300
2301         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2302                                                   th);
2303         if (likely(obj->oo_inode != NULL)) {
2304                 LASSERT(obj->oo_inode->i_state & I_NEW);
2305
2306                 /* Unlock the inode before attr initialization to avoid
2307                  * unnecessary dqget operations. LU-6378 */
2308                 unlock_new_inode(obj->oo_inode);
2309         }
2310
2311         if (likely(result == 0)) {
2312                 osd_attr_init(info, obj, attr, dof);
2313                 osd_object_init0(obj);
2314         }
2315
2316         /* restore previous umask value */
2317         current->fs->umask = umask;
2318
2319         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
2320
2321         return result;
2322 }
2323
2324 /**
2325  * Helper function for osd_object_create()
2326  *
2327  * \retval 0, on success
2328  */
2329 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2330                            const struct lu_fid *fid, struct thandle *th)
2331 {
2332         struct osd_thread_info *info = osd_oti_get(env);
2333         struct osd_inode_id    *id   = &info->oti_id;
2334         struct osd_device      *osd  = osd_obj2dev(obj);
2335         struct osd_thandle     *oh;
2336         int                     rc;
2337
2338         LASSERT(obj->oo_inode != NULL);
2339
2340         oh = container_of0(th, struct osd_thandle, ot_super);
2341         LASSERT(oh->ot_handle);
2342         osd_trans_exec_op(env, th, OSD_OT_INSERT);
2343
2344         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2345         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2346         osd_trans_exec_check(env, th, OSD_OT_INSERT);
2347
2348         return rc;
2349 }
2350
2351 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2352                    u64 seq, struct lu_seq_range *range)
2353 {
2354         struct seq_server_site  *ss = osd_seq_site(osd);
2355
2356         if (fid_seq_is_idif(seq)) {
2357                 fld_range_set_ost(range);
2358                 range->lsr_index = idif_ost_idx(seq);
2359                 return 0;
2360         }
2361
2362         if (!fid_seq_in_fldb(seq)) {
2363                 fld_range_set_mdt(range);
2364                 if (ss != NULL)
2365                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2366                          * at all */
2367                         range->lsr_index = ss->ss_node_id;
2368                 return 0;
2369         }
2370
2371         LASSERT(ss != NULL);
2372         fld_range_set_any(range);
2373         /* OSD will only do local fld lookup */
2374         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2375 }
2376
2377 /*
2378  * Concurrency: no external locking is necessary.
2379  */
2380 static int osd_declare_object_create(const struct lu_env *env,
2381                                      struct dt_object *dt,
2382                                      struct lu_attr *attr,
2383                                      struct dt_allocation_hint *hint,
2384                                      struct dt_object_format *dof,
2385                                      struct thandle *handle)
2386 {
2387         struct osd_thandle      *oh;
2388         int                      rc;
2389         ENTRY;
2390
2391         LASSERT(handle != NULL);
2392
2393         oh = container_of0(handle, struct osd_thandle, ot_super);
2394         LASSERT(oh->ot_handle == NULL);
2395
2396         /* EA object consumes more credits than regular object: osd_mk_index
2397          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
2398          * leaf_node, could involves the block, block bitmap, groups, GDT
2399          * change for each block, so add 4 * 2 credits in that case. */
2400         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2401                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
2402                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
2403         /* Reuse idle OI block may cause additional one OI block
2404          * to be changed. */
2405         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2406                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2407
2408         if (!attr)
2409                 RETURN(0);
2410
2411         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2412                                    osd_dt_obj(dt), false, NULL, false);
2413         if (rc != 0)
2414                 RETURN(rc);
2415
2416         RETURN(rc);
2417 }
2418
2419 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2420                              struct lu_attr *attr,
2421                              struct dt_allocation_hint *hint,
2422                              struct dt_object_format *dof, struct thandle *th)
2423 {
2424         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2425         struct osd_object       *obj    = osd_dt_obj(dt);
2426         struct osd_thread_info  *info   = osd_oti_get(env);
2427         int result;
2428         ENTRY;
2429
2430         if (dt_object_exists(dt))
2431                 return -EEXIST;
2432
2433         LINVRNT(osd_invariant(obj));
2434         LASSERT(!dt_object_remote(dt));
2435         LASSERT(osd_write_locked(env, obj));
2436         LASSERT(th != NULL);
2437
2438         if (unlikely(fid_is_acct(fid)))
2439                 /* Quota files can't be created from the kernel any more,
2440                  * 'tune2fs -O quota' will take care of creating them */
2441                 RETURN(-EPERM);
2442
2443         result = __osd_object_create(info, obj, attr, hint, dof, th);
2444         if (result == 0) {
2445                 result = __osd_oi_insert(env, obj, fid, th);
2446                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
2447                         obj->oo_dt.do_body_ops = &osd_body_ops;
2448         }
2449         LASSERT(ergo(result == 0,
2450                 dt_object_exists(dt) && !dt_object_remote(dt)));
2451
2452         LASSERT(osd_invariant(obj));
2453         RETURN(result);
2454 }
2455
2456 /**
2457  * Called to destroy on-disk representation of the object
2458  *
2459  * Concurrency: must be locked
2460  */
2461 static int osd_declare_object_destroy(const struct lu_env *env,
2462                                       struct dt_object *dt,
2463                                       struct thandle *th)
2464 {
2465         struct osd_object  *obj = osd_dt_obj(dt);
2466         struct inode       *inode = obj->oo_inode;
2467         struct osd_thandle *oh;
2468         int                 rc;
2469         ENTRY;
2470
2471         oh = container_of0(th, struct osd_thandle, ot_super);
2472         LASSERT(oh->ot_handle == NULL);
2473         LASSERT(inode);
2474
2475         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2476                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2477         /* Recycle idle OI leaf may cause additional three OI blocks
2478          * to be changed. */
2479         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2480                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2481         /* one less inode */
2482         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2483                                    -1, oh, obj, false, NULL, false);
2484         if (rc)
2485                 RETURN(rc);
2486         /* data to be truncated */
2487         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2488                                    0, oh, obj, true, NULL, false);
2489         RETURN(rc);
2490 }
2491
2492 static int osd_object_destroy(const struct lu_env *env,
2493                               struct dt_object *dt,
2494                               struct thandle *th)
2495 {
2496         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2497         struct osd_object      *obj = osd_dt_obj(dt);
2498         struct inode           *inode = obj->oo_inode;
2499         struct osd_device      *osd = osd_obj2dev(obj);
2500         struct osd_thandle     *oh;
2501         int                     result;
2502         ENTRY;
2503
2504         oh = container_of0(th, struct osd_thandle, ot_super);
2505         LASSERT(oh->ot_handle);
2506         LASSERT(inode);
2507         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2508
2509         if (unlikely(fid_is_acct(fid)))
2510                 RETURN(-EPERM);
2511
2512         if (S_ISDIR(inode->i_mode)) {
2513                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2514                         inode->i_nlink == 2);
2515                 /* it will check/delete the inode from remote parent,
2516                  * how to optimize it? unlink performance impaction XXX */
2517                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2518                 if (result != 0 && result != -ENOENT) {
2519                         CERROR("%s: delete inode "DFID": rc = %d\n",
2520                                osd_name(osd), PFID(fid), result);
2521                 }
2522                 spin_lock(&obj->oo_guard);
2523                 clear_nlink(inode);
2524                 spin_unlock(&obj->oo_guard);
2525                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2526         }
2527
2528         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2529
2530         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
2531         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2532                                OI_CHECK_FLD);
2533
2534         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
2535         /* XXX: add to ext3 orphan list */
2536         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2537
2538         /* not needed in the cache anymore */
2539         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2540         obj->oo_destroyed = 1;
2541
2542         RETURN(0);
2543 }
2544
2545 /**
2546  * Put the fid into lustre_mdt_attrs, and then place the structure
2547  * inode's ea. This fid should not be altered during the life time
2548  * of the inode.
2549  *
2550  * \retval +ve, on success
2551  * \retval -ve, on error
2552  *
2553  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2554  */
2555 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2556                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2557 {
2558         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2559         int                      rc;
2560         ENTRY;
2561
2562         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2563                 RETURN(0);
2564
2565         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
2566                 rc = -ENOMEM;
2567
2568         lustre_lma_init(lma, fid, compat, incompat);
2569         lustre_lma_swab(lma);
2570
2571         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2572                              XATTR_CREATE);
2573         /* LMA may already exist, but we need to check that all the
2574          * desired compat/incompat flags have been added. */
2575         if (unlikely(rc == -EEXIST)) {
2576                 if (compat == 0 && incompat == 0)
2577                         RETURN(0);
2578
2579                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2580                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2581                                      LMA_OLD_SIZE);
2582                 if (rc <= 0)
2583                         RETURN(-EINVAL);
2584
2585                 lustre_lma_swab(lma);
2586                 if (!(~lma->lma_compat & compat) &&
2587                     !(~lma->lma_incompat & incompat))
2588                         RETURN(0);
2589
2590                 lma->lma_compat |= compat;
2591                 lma->lma_incompat |= incompat;
2592                 lustre_lma_swab(lma);
2593                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2594                                      sizeof(*lma), XATTR_REPLACE);
2595         }
2596
2597         RETURN(rc);
2598 }
2599
2600 /**
2601  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2602  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2603  * To have compatilibility with 1.8 ldiskfs driver we need to have
2604  * magic number at start of fid data.
2605  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2606  * its inmemory API.
2607  */
2608 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2609                                          const struct lu_fid *fid)
2610 {
2611         if (!fid_is_namespace_visible(fid) ||
2612             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2613                 param->edp_magic = 0;
2614                 return;
2615         }
2616
2617         param->edp_magic = LDISKFS_LUFID_MAGIC;
2618         param->edp_len =  sizeof(struct lu_fid) + 1;
2619         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2620 }
2621
2622 /**
2623  * Try to read the fid from inode ea into dt_rec.
2624  *
2625  * \param fid object fid.
2626  *
2627  * \retval 0 on success
2628  */
2629 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2630                           __u32 ino, struct lu_fid *fid,
2631                           struct osd_inode_id *id)
2632 {
2633         struct osd_thread_info *info  = osd_oti_get(env);
2634         struct inode           *inode;
2635         ENTRY;
2636
2637         osd_id_gen(id, ino, OSD_OII_NOGEN);
2638         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2639         if (IS_ERR(inode))
2640                 RETURN(PTR_ERR(inode));
2641
2642         iput(inode);
2643         RETURN(0);
2644 }
2645
2646 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2647                                         struct inode *dir,
2648                                         struct inode *parent_dir,
2649                                         const struct lu_fid *dot_fid,
2650                                         const struct lu_fid *dot_dot_fid,
2651                                         struct osd_thandle *oth)
2652 {
2653         struct ldiskfs_dentry_param *dot_ldp;
2654         struct ldiskfs_dentry_param *dot_dot_ldp;
2655
2656         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2657         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2658
2659         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2660         dot_ldp->edp_magic = 0;
2661
2662         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2663                                       dir, dot_ldp, dot_dot_ldp);
2664 }
2665
2666 /**
2667  * Create an local agent inode for remote entry
2668  */
2669 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2670                                                   struct osd_device *osd,
2671                                                   struct osd_object *pobj,
2672                                                   const struct lu_fid *fid,
2673                                                   __u32 type,
2674                                                   struct thandle *th)
2675 {
2676         struct osd_thread_info  *info = osd_oti_get(env);
2677         struct inode            *local;
2678         struct osd_thandle      *oh;
2679         int                     rc;
2680         ENTRY;
2681
2682         LASSERT(th);
2683         oh = container_of(th, struct osd_thandle, ot_super);
2684         LASSERT(oh->ot_handle->h_transaction != NULL);
2685
2686         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2687         if (IS_ERR(local)) {
2688                 CERROR("%s: create local error %d\n", osd_name(osd),
2689                        (int)PTR_ERR(local));
2690                 RETURN(local);
2691         }
2692
2693         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
2694         unlock_new_inode(local);
2695
2696         /* Set special LMA flag for local agent inode */
2697         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2698         if (rc != 0) {
2699                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2700                        osd_name(osd), PFID(fid), rc);
2701                 RETURN(ERR_PTR(rc));
2702         }
2703
2704         if (!S_ISDIR(type))
2705                 RETURN(local);
2706
2707         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2708                                          lu_object_fid(&pobj->oo_dt.do_lu),
2709                                          fid, oh);
2710         if (rc != 0) {
2711                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2712                         osd_name(osd), PFID(fid), rc);
2713                 RETURN(ERR_PTR(rc));
2714         }
2715
2716         RETURN(local);
2717 }
2718
2719 /**
2720  * when direntry is deleted, we have to take care of possible agent inode
2721  * referenced by that. unfortunately we can't do this at that point:
2722  * iget() within a running transaction leads to deadlock and we better do
2723  * not call that every delete declaration to save performance. so we put
2724  * a potention agent inode on a list and process that once the transaction
2725  * is over. Notice it's not any worse in terms of real orphans as regular
2726  * object destroy doesn't put inodes on the on-disk orphan list. this should
2727  * be addressed separately
2728  */
2729 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
2730                                             struct osd_thandle *oh,
2731                                             __u32 ino)
2732 {
2733         struct osd_device      *osd = osd_dt_dev(oh->ot_super.th_dev);
2734         struct osd_obj_orphan *oor;
2735
2736         OBD_ALLOC_PTR(oor);
2737         if (oor == NULL)
2738                 return -ENOMEM;
2739
2740         oor->oor_ino = ino;
2741         oor->oor_env = (struct lu_env *)env;
2742         spin_lock(&osd->od_osfs_lock);
2743         list_add(&oor->oor_list, &osd->od_orphan_list);
2744         spin_unlock(&osd->od_osfs_lock);
2745
2746         oh->ot_remove_agents = 1;
2747
2748         return 0;
2749
2750 }
2751
2752 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
2753                                                 struct osd_device *osd)
2754 {
2755         struct osd_thread_info *info = osd_oti_get(env);
2756         struct osd_obj_orphan *oor, *tmp;
2757         struct osd_inode_id id;
2758         struct list_head list;
2759         struct inode *inode;
2760         struct lu_fid fid;
2761         handle_t *jh;
2762         __u32 ino;
2763
2764         INIT_LIST_HEAD(&list);
2765
2766         spin_lock(&osd->od_osfs_lock);
2767         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
2768                 if (oor->oor_env == env) {
2769                         list_del(&oor->oor_list);
2770                         list_add(&oor->oor_list, &list);
2771                 }
2772         }
2773         spin_unlock(&osd->od_osfs_lock);
2774
2775         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
2776
2777                 ino = oor->oor_ino;
2778
2779                 list_del(&oor->oor_list);
2780                 OBD_FREE_PTR(oor);
2781
2782                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
2783                 inode = osd_iget_fid(info, osd, &id, &fid);
2784                 if (IS_ERR(inode))
2785                         continue;
2786
2787                 if (!osd_remote_fid(env, osd, &fid)) {
2788                         iput(inode);
2789                         continue;
2790                 }
2791
2792                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
2793                 clear_nlink(inode);
2794                 mark_inode_dirty(inode);
2795                 ldiskfs_journal_stop(jh);
2796                 iput(inode);
2797         }
2798
2799         return 0;
2800 }
2801
2802 /**
2803  * OSD layer object create function for interoperability mode (b11826).
2804  * This is mostly similar to osd_object_create(). Only difference being, fid is
2805  * inserted into inode ea here.
2806  *
2807  * \retval   0, on success
2808  * \retval -ve, on error
2809  */
2810 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2811                                 struct lu_attr *attr,
2812                                 struct dt_allocation_hint *hint,
2813                                 struct dt_object_format *dof,
2814                                 struct thandle *th)
2815 {
2816         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2817         struct osd_object       *obj    = osd_dt_obj(dt);
2818         struct osd_thread_info  *info   = osd_oti_get(env);
2819         int                      result;
2820
2821         ENTRY;
2822
2823         if (dt_object_exists(dt))
2824                 RETURN(-EEXIST);
2825
2826         LASSERT(osd_invariant(obj));
2827         LASSERT(!dt_object_remote(dt));
2828         LASSERT(osd_write_locked(env, obj));
2829         LASSERT(th != NULL);
2830
2831         if (unlikely(fid_is_acct(fid)))
2832                 /* Quota files can't be created from the kernel any more,
2833                  * 'tune2fs -O quota' will take care of creating them */
2834                 RETURN(-EPERM);
2835
2836         result = __osd_object_create(info, obj, attr, hint, dof, th);
2837         if (result == 0) {
2838                 if (fid_is_idif(fid) &&
2839                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
2840                         struct lu_fid *tfid = &info->oti_fid;
2841                         struct ost_id *oi   = &info->oti_ostid;
2842
2843                         fid_to_ostid(fid, oi);
2844                         ostid_to_fid(tfid, oi, 0);
2845                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
2846                                                 LMAC_FID_ON_OST, 0);
2847                 } else {
2848                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
2849                                 fid_is_on_ost(info, osd_obj2dev(obj),
2850                                               fid, OI_CHECK_FLD) ?
2851                                 LMAC_FID_ON_OST : 0, 0);
2852                 }
2853                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
2854                         obj->oo_dt.do_body_ops = &osd_body_ops;
2855         }
2856
2857         if (result == 0)
2858                 result = __osd_oi_insert(env, obj, fid, th);
2859
2860         LASSERT(ergo(result == 0,
2861                      dt_object_exists(dt) && !dt_object_remote(dt)));
2862         LINVRNT(osd_invariant(obj));
2863         RETURN(result);
2864 }
2865
2866 static int osd_declare_object_ref_add(const struct lu_env *env,
2867                                       struct dt_object *dt,
2868                                       struct thandle *handle)
2869 {
2870         struct osd_thandle       *oh;
2871
2872         /* it's possible that object doesn't exist yet */
2873         LASSERT(handle != NULL);
2874
2875         oh = container_of0(handle, struct osd_thandle, ot_super);
2876         LASSERT(oh->ot_handle == NULL);
2877
2878         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2879                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2880
2881         return 0;
2882 }
2883
2884 /*
2885  * Concurrency: @dt is write locked.
2886  */
2887 static int osd_object_ref_add(const struct lu_env *env,
2888                               struct dt_object *dt, struct thandle *th)
2889 {
2890         struct osd_object  *obj = osd_dt_obj(dt);
2891         struct inode       *inode = obj->oo_inode;
2892         struct osd_thandle *oh;
2893         int                 rc = 0;
2894
2895         if (!dt_object_exists(dt))
2896                 return -ENOENT;
2897
2898         LINVRNT(osd_invariant(obj));
2899         LASSERT(!dt_object_remote(dt));
2900         LASSERT(osd_write_locked(env, obj));
2901         LASSERT(th != NULL);
2902
2903         oh = container_of0(th, struct osd_thandle, ot_super);
2904         LASSERT(oh->ot_handle != NULL);
2905
2906         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2907
2908         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2909                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2910         /*
2911          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2912          * (65000) subdirectories by storing "1" in i_nlink if the link count
2913          * would otherwise overflow. Directory tranversal tools understand
2914          * that (st_nlink == 1) indicates that the filesystem dose not track
2915          * hard links count on the directory, and will not abort subdirectory
2916          * scanning early once (st_nlink - 2) subdirs have been found.
2917          *
2918          * This also has to properly handle the case of inodes with nlink == 0
2919          * in case they are being linked into the PENDING directory
2920          */
2921         spin_lock(&obj->oo_guard);
2922         if (unlikely(inode->i_nlink == 0))
2923                 /* inc_nlink from 0 may cause WARN_ON */
2924                 set_nlink(inode, 1);
2925         else {
2926                 ldiskfs_inc_count(oh->ot_handle, inode);
2927                 if (!S_ISDIR(inode->i_mode))
2928                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2929         }
2930         spin_unlock(&obj->oo_guard);
2931
2932         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2933         LINVRNT(osd_invariant(obj));
2934
2935         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
2936
2937         return rc;
2938 }
2939
2940 static int osd_declare_object_ref_del(const struct lu_env *env,
2941                                       struct dt_object *dt,
2942                                       struct thandle *handle)
2943 {
2944         struct osd_thandle *oh;
2945
2946         LASSERT(!dt_object_remote(dt));
2947         LASSERT(handle != NULL);
2948
2949         oh = container_of0(handle, struct osd_thandle, ot_super);
2950         LASSERT(oh->ot_handle == NULL);
2951
2952         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2953                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2954
2955         return 0;
2956 }
2957
2958 /*
2959  * Concurrency: @dt is write locked.
2960  */
2961 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2962                               struct thandle *th)
2963 {
2964         struct osd_object       *obj = osd_dt_obj(dt);
2965         struct inode            *inode = obj->oo_inode;
2966         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2967         struct osd_thandle      *oh;
2968
2969         if (!dt_object_exists(dt))
2970                 return -ENOENT;
2971
2972         LINVRNT(osd_invariant(obj));
2973         LASSERT(!dt_object_remote(dt));
2974         LASSERT(osd_write_locked(env, obj));
2975         LASSERT(th != NULL);
2976
2977         oh = container_of0(th, struct osd_thandle, ot_super);
2978         LASSERT(oh->ot_handle != NULL);
2979
2980         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2981
2982         spin_lock(&obj->oo_guard);
2983         /* That can be result of upgrade from old Lustre version and
2984          * applied only to local files.  Just skip this ref_del call.
2985          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2986         if (inode->i_nlink == 0) {
2987                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2988                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2989                              ", maybe an upgraded file? (LU-3915)\n",
2990                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2991                 spin_unlock(&obj->oo_guard);
2992                 return 0;
2993         }
2994
2995         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2996                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2997
2998         ldiskfs_dec_count(oh->ot_handle, inode);
2999         spin_unlock(&obj->oo_guard);
3000
3001         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3002         LINVRNT(osd_invariant(obj));
3003
3004         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3005
3006         return 0;
3007 }
3008
3009 /*
3010  * Get the 64-bit version for an inode.
3011  */
3012 static int osd_object_version_get(const struct lu_env *env,
3013                                   struct dt_object *dt, dt_obj_version_t *ver)
3014 {
3015         struct inode *inode = osd_dt_obj(dt)->oo_inode;
3016
3017         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
3018                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3019         *ver = LDISKFS_I(inode)->i_fs_version;
3020         return 0;
3021 }
3022
3023 /*
3024  * Concurrency: @dt is read locked.
3025  */
3026 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3027                          struct lu_buf *buf, const char *name)
3028 {
3029         struct osd_object      *obj    = osd_dt_obj(dt);
3030         struct inode           *inode  = obj->oo_inode;
3031         struct osd_thread_info *info   = osd_oti_get(env);
3032         struct dentry          *dentry = &info->oti_obj_dentry;
3033
3034         /* version get is not real XATTR but uses xattr API */
3035         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3036                 /* for version we are just using xattr API but change inode
3037                  * field instead */
3038                 if (buf->lb_len == 0)
3039                         return sizeof(dt_obj_version_t);
3040
3041                 if (buf->lb_len < sizeof(dt_obj_version_t))
3042                         return -ERANGE;
3043
3044                 osd_object_version_get(env, dt, buf->lb_buf);
3045
3046                 return sizeof(dt_obj_version_t);
3047         }
3048
3049         if (!dt_object_exists(dt))
3050                 return -ENOENT;
3051
3052         LASSERT(!dt_object_remote(dt));
3053         LASSERT(inode->i_op != NULL);
3054         LASSERT(inode->i_op->getxattr != NULL);
3055
3056         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
3057 }
3058
3059
3060 static int osd_declare_xattr_set(const struct lu_env *env,
3061                                  struct dt_object *dt,
3062                                  const struct lu_buf *buf, const char *name,
3063                                  int fl, struct thandle *handle)
3064 {
3065         struct osd_thandle *oh;
3066         int credits;
3067         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3068
3069         LASSERT(handle != NULL);
3070
3071         oh = container_of0(handle, struct osd_thandle, ot_super);
3072         LASSERT(oh->ot_handle == NULL);
3073
3074         if (strcmp(name, XATTR_NAME_LMA) == 0) {
3075                 /* For non-upgrading case, the LMA is set first and
3076                  * usually fit inode. But for upgrade case, the LMA
3077                  * may be in another separated EA block. */
3078                 if (!dt_object_exists(dt))
3079                         credits = 0;
3080                 else if (fl == LU_XATTR_REPLACE)
3081                         credits = 1;
3082                 else
3083                         goto upgrade;
3084         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3085                 credits = 1;
3086         } else {
3087 upgrade:
3088                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
3089
3090                 if (buf != NULL) {
3091                         ssize_t buflen;
3092
3093                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
3094                                 /* learn xattr size from osd_xattr_get if
3095                                    attribute has not been read yet */
3096                                 buflen = __osd_xattr_get(
3097                                     osd_dt_obj(dt)->oo_inode,
3098                                     &osd_oti_get(env)->oti_obj_dentry,
3099                                     name, NULL, 0);
3100                                 if (buflen < 0)
3101                                         buflen = 0;
3102                         } else {
3103                                 buflen = buf->lb_len;
3104                         }
3105
3106                         if (buflen > sb->s_blocksize) {
3107                                 credits += osd_calc_bkmap_credits(
3108                                     sb, NULL, 0, -1,
3109                                     (buflen + sb->s_blocksize - 1) >>
3110                                     sb->s_blocksize_bits);
3111                         }
3112                 }
3113                 /*
3114                  * xattr set may involve inode quota change, reserve credits for
3115                  * dquot_initialize()
3116                  */
3117                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3118         }
3119
3120         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3121
3122         return 0;
3123 }
3124
3125 /*
3126  * Set the 64-bit version for object
3127  */
3128 static void osd_object_version_set(const struct lu_env *env,
3129                                    struct dt_object *dt,
3130                                    dt_obj_version_t *new_version)
3131 {
3132         struct inode *inode = osd_dt_obj(dt)->oo_inode;
3133
3134         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
3135                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3136
3137         LDISKFS_I(inode)->i_fs_version = *new_version;
3138         /** Version is set after all inode operations are finished,
3139          *  so we should mark it dirty here */
3140         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3141 }
3142
3143 /*
3144  * Concurrency: @dt is write locked.
3145  */
3146 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
3147                          const struct lu_buf *buf, const char *name, int fl,
3148                          struct thandle *handle)
3149 {
3150         struct osd_object      *obj      = osd_dt_obj(dt);
3151         struct inode           *inode    = obj->oo_inode;
3152         struct osd_thread_info *info     = osd_oti_get(env);
3153         int                     fs_flags = 0;
3154         int                     rc;
3155         ENTRY;
3156
3157         LASSERT(handle != NULL);
3158
3159         /* version set is not real XATTR */
3160         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3161                 /* for version we are just using xattr API but change inode
3162                  * field instead */
3163                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3164                 osd_object_version_set(env, dt, buf->lb_buf);
3165                 return sizeof(dt_obj_version_t);
3166         }
3167
3168         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
3169                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3170
3171         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3172         if (fl & LU_XATTR_REPLACE)
3173                 fs_flags |= XATTR_REPLACE;
3174
3175         if (fl & LU_XATTR_CREATE)
3176                 fs_flags |= XATTR_CREATE;
3177
3178         if (strcmp(name, XATTR_NAME_LMV) == 0) {
3179                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3180                 int                      rc;
3181
3182                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3183                 if (rc != 0)
3184                         RETURN(rc);
3185
3186                 lma->lma_incompat |= LMAI_STRIPED;
3187                 lustre_lma_swab(lma);
3188                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3189                                      sizeof(*lma), XATTR_REPLACE);
3190                 if (rc != 0)
3191                         RETURN(rc);
3192         }
3193
3194         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
3195             strcmp(name, XATTR_NAME_LINK) == 0)
3196                 return -ENOSPC;
3197
3198         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3199                                fs_flags);
3200         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3201
3202         return rc;
3203 }
3204
3205 /*
3206  * Concurrency: @dt is read locked.
3207  */
3208 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3209                           const struct lu_buf *buf)
3210 {
3211         struct osd_object      *obj    = osd_dt_obj(dt);
3212         struct inode           *inode  = obj->oo_inode;
3213         struct osd_thread_info *info   = osd_oti_get(env);
3214         struct dentry          *dentry = &info->oti_obj_dentry;
3215
3216         if (!dt_object_exists(dt))
3217                 return -ENOENT;
3218
3219         LASSERT(!dt_object_remote(dt));
3220         LASSERT(inode->i_op != NULL);
3221         LASSERT(inode->i_op->listxattr != NULL);
3222
3223         dentry->d_inode = inode;
3224         dentry->d_sb = inode->i_sb;
3225         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3226 }
3227
3228 static int osd_declare_xattr_del(const struct lu_env *env,
3229                                  struct dt_object *dt, const char *name,
3230                                  struct thandle *handle)
3231 {
3232         struct osd_thandle *oh;
3233         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3234
3235         LASSERT(!dt_object_remote(dt));
3236         LASSERT(handle != NULL);
3237
3238         oh = container_of0(handle, struct osd_thandle, ot_super);
3239         LASSERT(oh->ot_handle == NULL);
3240
3241         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3242                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3243         /*
3244          * xattr del may involve inode quota change, reserve credits for
3245          * dquot_initialize()
3246          */
3247         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3248
3249         return 0;
3250 }
3251
3252 /*
3253  * Concurrency: @dt is write locked.
3254  */
3255 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3256                          const char *name, struct thandle *handle)
3257 {
3258         struct osd_object      *obj    = osd_dt_obj(dt);
3259         struct inode           *inode  = obj->oo_inode;
3260         struct osd_thread_info *info   = osd_oti_get(env);
3261         struct dentry          *dentry = &info->oti_obj_dentry;
3262         int                     rc;
3263
3264         if (!dt_object_exists(dt))
3265                 return -ENOENT;
3266
3267         LASSERT(!dt_object_remote(dt));
3268         LASSERT(inode->i_op != NULL);
3269         LASSERT(inode->i_op->removexattr != NULL);
3270         LASSERT(handle != NULL);
3271
3272         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3273
3274         ll_vfs_dq_init(inode);
3275         dentry->d_inode = inode;
3276         dentry->d_sb = inode->i_sb;
3277         rc = inode->i_op->removexattr(dentry, name);
3278         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3279         return rc;
3280 }
3281
3282 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3283                            __u64 start, __u64 end)
3284 {
3285         struct osd_object       *obj    = osd_dt_obj(dt);
3286         struct inode            *inode  = obj->oo_inode;
3287         struct osd_thread_info  *info   = osd_oti_get(env);
3288         struct dentry           *dentry = &info->oti_obj_dentry;
3289         struct file             *file   = &info->oti_file;
3290         int                     rc;
3291
3292         ENTRY;
3293
3294         dentry->d_inode = inode;
3295         dentry->d_sb = inode->i_sb;
3296         file->f_path.dentry = dentry;
3297         file->f_mapping = inode->i_mapping;
3298         file->f_op = inode->i_fop;
3299         set_file_inode(file, inode);
3300
3301         rc = ll_vfs_fsync_range(file, start, end, 0);
3302
3303         RETURN(rc);
3304 }
3305
3306 /*
3307  * Index operations.
3308  */
3309
3310 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3311                            const struct dt_index_features *feat)
3312 {
3313         struct iam_descr *descr;
3314
3315         if (osd_object_is_root(o))
3316                 return feat == &dt_directory_features;
3317
3318         LASSERT(o->oo_dir != NULL);
3319
3320         descr = o->oo_dir->od_container.ic_descr;
3321         if (feat == &dt_directory_features) {
3322                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3323                         return 1;
3324                 else
3325                         return 0;
3326         } else {
3327                 return
3328                         feat->dif_keysize_min <= descr->id_key_size &&
3329                         descr->id_key_size <= feat->dif_keysize_max &&
3330                         feat->dif_recsize_min <= descr->id_rec_size &&
3331                         descr->id_rec_size <= feat->dif_recsize_max &&
3332                         !(feat->dif_flags & (DT_IND_VARKEY |
3333                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3334                         ergo(feat->dif_flags & DT_IND_UPDATE,
3335                              1 /* XXX check that object (and file system) is
3336                                 * writable */);
3337         }
3338 }
3339
3340 static int osd_iam_container_init(const struct lu_env *env,
3341                                   struct osd_object *obj,
3342                                   struct osd_directory *dir)
3343 {
3344         struct iam_container *bag = &dir->od_container;
3345         int result;
3346
3347         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3348         if (result != 0)
3349                 return result;
3350
3351         result = iam_container_setup(bag);
3352         if (result == 0)
3353                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3354         else
3355                 iam_container_fini(bag);
3356
3357         return result;
3358 }
3359
3360
3361 /*
3362  * Concurrency: no external locking is necessary.
3363  */
3364 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3365                          const struct dt_index_features *feat)
3366 {
3367         int                      result;
3368         int                      skip_iam = 0;
3369         struct osd_object       *obj = osd_dt_obj(dt);
3370
3371         LINVRNT(osd_invariant(obj));
3372
3373         if (osd_object_is_root(obj)) {
3374                 dt->do_index_ops = &osd_index_ea_ops;
3375                 result = 0;
3376         } else if (feat == &dt_directory_features) {
3377                 dt->do_index_ops = &osd_index_ea_ops;
3378                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3379                         result = 0;
3380                 else
3381                         result = -ENOTDIR;
3382                 skip_iam = 1;
3383         } else if (unlikely(feat == &dt_otable_features)) {
3384                 dt->do_index_ops = &osd_otable_ops;
3385                 return 0;
3386         } else if (unlikely(feat == &dt_acct_features)) {
3387                 dt->do_index_ops = &osd_acct_index_ops;
3388                 result = 0;
3389                 skip_iam = 1;
3390         } else if (!osd_has_index(obj)) {
3391                 struct osd_directory *dir;
3392
3393                 OBD_ALLOC_PTR(dir);
3394                 if (dir != NULL) {
3395
3396                         spin_lock(&obj->oo_guard);
3397                         if (obj->oo_dir == NULL)
3398                                 obj->oo_dir = dir;
3399                         else
3400                                 /*
3401                                  * Concurrent thread allocated container data.
3402                                  */
3403                                 OBD_FREE_PTR(dir);
3404                         spin_unlock(&obj->oo_guard);
3405                         /*
3406                          * Now, that we have container data, serialize its
3407                          * initialization.
3408                          */
3409                         down_write(&obj->oo_ext_idx_sem);
3410                         /*
3411                          * recheck under lock.
3412                          */
3413                         if (!osd_has_index(obj))
3414                                 result = osd_iam_container_init(env, obj,
3415                                                                 obj->oo_dir);
3416                         else
3417                                 result = 0;
3418                         up_write(&obj->oo_ext_idx_sem);
3419                 } else {
3420                         result = -ENOMEM;
3421                 }
3422         } else {
3423                 result = 0;
3424         }
3425
3426         if (result == 0 && skip_iam == 0) {
3427                 if (!osd_iam_index_probe(env, obj, feat))
3428                         result = -ENOTDIR;
3429         }
3430         LINVRNT(osd_invariant(obj));
3431
3432         if (result == 0 && feat == &dt_quota_glb_features &&
3433             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3434                 result = osd_quota_migration(env, dt);
3435
3436         return result;
3437 }
3438
3439 static int osd_otable_it_attr_get(const struct lu_env *env,
3440                                  struct dt_object *dt,
3441                                  struct lu_attr *attr)
3442 {
3443         attr->la_valid = 0;
3444         return 0;
3445 }
3446
3447 static const struct dt_object_operations osd_obj_ops = {
3448         .do_read_lock         = osd_object_read_lock,
3449         .do_write_lock        = osd_object_write_lock,
3450         .do_read_unlock       = osd_object_read_unlock,
3451         .do_write_unlock      = osd_object_write_unlock,
3452         .do_write_locked      = osd_object_write_locked,
3453         .do_attr_get          = osd_attr_get,
3454         .do_declare_attr_set  = osd_declare_attr_set,
3455         .do_attr_set          = osd_attr_set,
3456         .do_ah_init           = osd_ah_init,
3457         .do_declare_create    = osd_declare_object_create,
3458         .do_create            = osd_object_create,
3459         .do_declare_destroy   = osd_declare_object_destroy,
3460         .do_destroy           = osd_object_destroy,
3461         .do_index_try         = osd_index_try,
3462         .do_declare_ref_add   = osd_declare_object_ref_add,
3463         .do_ref_add           = osd_object_ref_add,
3464         .do_declare_ref_del   = osd_declare_object_ref_del,
3465         .do_ref_del           = osd_object_ref_del,
3466         .do_xattr_get         = osd_xattr_get,
3467         .do_declare_xattr_set = osd_declare_xattr_set,
3468         .do_xattr_set         = osd_xattr_set,
3469         .do_declare_xattr_del = osd_declare_xattr_del,
3470         .do_xattr_del         = osd_xattr_del,
3471         .do_xattr_list        = osd_xattr_list,
3472         .do_object_sync       = osd_object_sync,
3473 };
3474
3475 /**
3476  * dt_object_operations for interoperability mode
3477  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3478  */
3479 static const struct dt_object_operations osd_obj_ea_ops = {
3480         .do_read_lock         = osd_object_read_lock,
3481         .do_write_lock        = osd_object_write_lock,
3482         .do_read_unlock       = osd_object_read_unlock,
3483         .do_write_unlock      = osd_object_write_unlock,
3484         .do_write_locked      = osd_object_write_locked,
3485         .do_attr_get          = osd_attr_get,
3486         .do_declare_attr_set  = osd_declare_attr_set,
3487         .do_attr_set          = osd_attr_set,
3488         .do_ah_init           = osd_ah_init,
3489         .do_declare_create    = osd_declare_object_create,
3490         .do_create            = osd_object_ea_create,
3491         .do_declare_destroy   = osd_declare_object_destroy,
3492         .do_destroy           = osd_object_destroy,
3493         .do_index_try         = osd_index_try,
3494         .do_declare_ref_add   = osd_declare_object_ref_add,
3495         .do_ref_add           = osd_object_ref_add,
3496         .do_declare_ref_del   = osd_declare_object_ref_del,
3497         .do_ref_del           = osd_object_ref_del,
3498         .do_xattr_get         = osd_xattr_get,
3499         .do_declare_xattr_set = osd_declare_xattr_set,
3500         .do_xattr_set         = osd_xattr_set,
3501         .do_declare_xattr_del = osd_declare_xattr_del,
3502         .do_xattr_del         = osd_xattr_del,
3503         .do_xattr_list        = osd_xattr_list,
3504         .do_object_sync       = osd_object_sync,
3505 };
3506
3507 static const struct dt_object_operations osd_obj_otable_it_ops = {
3508         .do_attr_get    = osd_otable_it_attr_get,
3509         .do_index_try   = osd_index_try,
3510 };
3511
3512 static int osd_index_declare_iam_delete(const struct lu_env *env,
3513                                         struct dt_object *dt,
3514                                         const struct dt_key *key,
3515                                         struct thandle *handle)
3516 {
3517         struct osd_thandle    *oh;
3518
3519         oh = container_of0(handle, struct osd_thandle, ot_super);
3520         LASSERT(oh->ot_handle == NULL);
3521
3522         /* Recycle  may cause additional three blocks to be changed. */
3523         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3524                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3525
3526         return 0;
3527 }
3528
3529 /**
3530  *      delete a (key, value) pair from index \a dt specified by \a key
3531  *
3532  *      \param  dt      osd index object
3533  *      \param  key     key for index
3534  *      \param  rec     record reference
3535  *      \param  handle  transaction handler
3536  *
3537  *      \retval  0  success
3538  *      \retval -ve   failure
3539  */
3540 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3541                                 const struct dt_key *key,
3542                                 struct thandle *handle)
3543 {
3544         struct osd_thread_info *oti = osd_oti_get(env);
3545         struct osd_object      *obj = osd_dt_obj(dt);
3546         struct osd_thandle     *oh;
3547         struct iam_path_descr  *ipd;
3548         struct iam_container   *bag = &obj->oo_dir->od_container;
3549         int                     rc;
3550         ENTRY;
3551
3552         if (!dt_object_exists(dt))
3553                 RETURN(-ENOENT);
3554
3555         LINVRNT(osd_invariant(obj));
3556         LASSERT(!dt_object_remote(dt));
3557         LASSERT(bag->ic_object == obj->oo_inode);
3558         LASSERT(handle != NULL);
3559
3560         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3561
3562         ipd = osd_idx_ipd_get(env, bag);
3563         if (unlikely(ipd == NULL))
3564                 RETURN(-ENOMEM);
3565
3566         oh = container_of0(handle, struct osd_thandle, ot_super);
3567         LASSERT(oh->ot_handle != NULL);
3568         LASSERT(oh->ot_handle->h_transaction != NULL);
3569
3570         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3571                 /* swab quota uid/gid provided by caller */
3572                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3573                 key = (const struct dt_key *)&oti->oti_quota_id;
3574         }
3575
3576         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3577         osd_ipd_put(env, bag, ipd);
3578         LINVRNT(osd_invariant(obj));
3579         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
3580         RETURN(rc);
3581 }
3582
3583 static int osd_index_declare_ea_delete(const struct lu_env *env,
3584                                        struct dt_object *dt,
3585                                        const struct dt_key *key,
3586                                        struct thandle *handle)
3587 {
3588         struct osd_thandle *oh;
3589         struct inode       *inode;
3590         int                 rc;
3591         ENTRY;
3592
3593         LASSERT(!dt_object_remote(dt));
3594         LASSERT(handle != NULL);
3595
3596         oh = container_of0(handle, struct osd_thandle, ot_super);
3597         LASSERT(oh->ot_handle == NULL);
3598
3599         /* due to DNE we may need to remove an agent inode */
3600         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3601                              osd_dto_credits_noquota[DTO_INDEX_DELETE] +
3602                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3603
3604         inode = osd_dt_obj(dt)->oo_inode;
3605         LASSERT(inode);
3606
3607         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3608                                    0, oh, osd_dt_obj(dt), true, NULL, false);
3609         RETURN(rc);
3610 }
3611
3612 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3613                                           struct dt_rec *fid)
3614 {
3615         struct osd_fid_pack *rec;
3616         int                  rc = -ENODATA;
3617
3618         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3619                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3620                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3621                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
3622                         rc = -EINVAL;
3623         }
3624         return rc;
3625 }
3626
3627 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3628                           const struct lu_fid *fid)
3629 {
3630         struct seq_server_site  *ss = osd_seq_site(osd);
3631         ENTRY;
3632
3633         /* FID seqs not in FLDB, must be local seq */
3634         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3635                 RETURN(0);
3636
3637         /* If FLD is not being initialized yet, it only happens during the
3638          * initialization, likely during mgs initialization, and we assume
3639          * this is local FID. */
3640         if (ss == NULL || ss->ss_server_fld == NULL)
3641                 RETURN(0);
3642
3643         /* Only check the local FLDB here */
3644         if (osd_seq_exists(env, osd, fid_seq(fid)))
3645                 RETURN(0);
3646
3647         RETURN(1);
3648 }
3649
3650 /**
3651  * Index delete function for interoperability mode (b11826).
3652  * It will remove the directory entry added by osd_index_ea_insert().
3653  * This entry is needed to maintain name->fid mapping.
3654  *
3655  * \param key,  key i.e. file entry to be deleted
3656  *
3657  * \retval   0, on success
3658  * \retval -ve, on error
3659  */
3660 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3661                                const struct dt_key *key, struct thandle *handle)
3662 {
3663         struct osd_object          *obj = osd_dt_obj(dt);
3664         struct inode               *dir = obj->oo_inode;
3665         struct dentry              *dentry;
3666         struct osd_thandle         *oh;
3667         struct ldiskfs_dir_entry_2 *de = NULL;
3668         struct buffer_head         *bh;
3669         struct htree_lock          *hlock = NULL;
3670         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3671         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3672         int                        rc;
3673         ENTRY;
3674
3675         if (!dt_object_exists(dt))
3676                 RETURN(-ENOENT);
3677
3678         LINVRNT(osd_invariant(obj));
3679         LASSERT(!dt_object_remote(dt));
3680         LASSERT(handle != NULL);
3681
3682         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3683
3684         oh = container_of(handle, struct osd_thandle, ot_super);
3685         LASSERT(oh->ot_handle != NULL);
3686         LASSERT(oh->ot_handle->h_transaction != NULL);
3687
3688         ll_vfs_dq_init(dir);
3689         dentry = osd_child_dentry_get(env, obj,
3690                                       (char *)key, strlen((char *)key));
3691
3692         if (obj->oo_hl_head != NULL) {
3693                 hlock = osd_oti_get(env)->oti_hlock;
3694                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3695                                    dir, LDISKFS_HLOCK_DEL);
3696         } else {
3697                 down_write(&obj->oo_ext_idx_sem);
3698         }
3699
3700         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3701         if (bh) {
3702                 /* If this is not the ".." entry, it might be a remote DNE
3703                  * entry and  we need to check if the FID is for a remote
3704                  * MDT.  If the FID is  not in the directory entry (e.g.
3705                  * upgraded 1.8 filesystem without dirdata enabled) then
3706                  * we need to get the FID from the LMA. For a remote directory
3707                  * there HAS to be an LMA, it cannot be an IGIF inode in this
3708                  * case.
3709                  *
3710                  * Delete the entry before the agent inode in order to
3711                  * simplify error handling.  At worst an error after deleting
3712                  * the entry first might leak the agent inode afterward. The
3713                  * reverse would need filesystem abort in case of error deleting
3714                  * the entry after the agent had been removed, or leave a
3715                  * dangling entry pointing at a random inode. */
3716                 if (strcmp((char *)key, dotdot) != 0) {
3717                         LASSERT(de != NULL);
3718                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3719                         if (rc == -ENODATA) {
3720                                 /* can't get FID, postpone to the end of the
3721                                  * transaction when iget() is safe */
3722                                 osd_schedule_agent_inode_removal(env, oh,
3723                                                 le32_to_cpu(de->inode));
3724                         } else if (rc == 0 &&
3725                                    unlikely(osd_remote_fid(env, osd, fid))) {
3726                                 osd_schedule_agent_inode_removal(env, oh,
3727                                                 le32_to_cpu(de->inode));
3728                         }
3729                 }
3730                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3731                 brelse(bh);
3732         } else {
3733                 rc = -ENOENT;
3734         }
3735         if (hlock != NULL)
3736                 ldiskfs_htree_unlock(hlock);
3737         else
3738                 up_write(&obj->oo_ext_idx_sem);
3739
3740         if (rc != 0)
3741                 GOTO(out, rc);
3742
3743         /* For inode on the remote MDT, .. will point to
3744          * /Agent directory, Check whether it needs to delete
3745          * from agent directory */
3746         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3747                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3748                                                    oh);
3749                 if (rc != 0 && rc != -ENOENT) {
3750                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3751                                osd_name(osd), PFID(fid), rc);
3752                 }
3753
3754                 if (rc == -ENOENT)
3755                         rc = 0;
3756
3757                 GOTO(out, rc);
3758         }
3759 out:
3760
3761         LASSERT(osd_invariant(obj));
3762         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
3763         RETURN(rc);
3764 }
3765
3766 /**
3767  *      Lookup index for \a key and copy record to \a rec.
3768  *
3769  *      \param  dt      osd index object
3770  *      \param  key     key for index
3771  *      \param  rec     record reference
3772  *
3773  *      \retval  +ve  success : exact mach
3774  *      \retval  0    return record with key not greater than \a key
3775  *      \retval -ve   failure
3776  */
3777 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3778                                 struct dt_rec *rec, const struct dt_key *key)
3779 {
3780         struct osd_object      *obj = osd_dt_obj(dt);
3781         struct iam_path_descr  *ipd;
3782         struct iam_container   *bag = &obj->oo_dir->od_container;
3783         struct osd_thread_info *oti = osd_oti_get(env);
3784         struct iam_iterator    *it = &oti->oti_idx_it;
3785         struct iam_rec         *iam_rec;
3786         int                     rc;
3787         ENTRY;
3788
3789         if (!dt_object_exists(dt))
3790                 RETURN(-ENOENT);
3791
3792         LASSERT(osd_invariant(obj));
3793         LASSERT(!dt_object_remote(dt));
3794         LASSERT(bag->ic_object == obj->oo_inode);
3795
3796         ipd = osd_idx_ipd_get(env, bag);
3797         if (IS_ERR(ipd))
3798                 RETURN(-ENOMEM);
3799
3800         /* got ipd now we can start iterator. */
3801         iam_it_init(it, bag, 0, ipd);
3802
3803         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3804                 /* swab quota uid/gid provided by caller */
3805                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3806                 key = (const struct dt_key *)&oti->oti_quota_id;
3807         }
3808
3809         rc = iam_it_get(it, (struct iam_key *)key);
3810         if (rc >= 0) {
3811                 if (S_ISDIR(obj->oo_inode->i_mode))
3812                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3813                 else
3814                         iam_rec = (struct iam_rec *) rec;
3815
3816                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3817
3818                 if (S_ISDIR(obj->oo_inode->i_mode))
3819                         osd_fid_unpack((struct lu_fid *) rec,
3820                                        (struct osd_fid_pack *)iam_rec);
3821                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3822                         osd_quota_unpack(obj, rec);
3823         }
3824
3825         iam_it_put(it);
3826         iam_it_fini(it);
3827         osd_ipd_put(env, bag, ipd);
3828
3829         LINVRNT(osd_invariant(obj));
3830
3831         RETURN(rc);
3832 }
3833
3834 static int osd_index_declare_iam_insert(const struct lu_env *env,
3835                                         struct dt_object *dt,
3836                                         const struct dt_rec *rec,
3837                                         const struct dt_key *key,
3838                                         struct thandle *handle)
3839 {
3840         struct osd_thandle *oh;
3841
3842         LASSERT(handle != NULL);
3843
3844         oh = container_of0(handle, struct osd_thandle, ot_super);
3845         LASSERT(oh->ot_handle == NULL);
3846
3847         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3848                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3849
3850         return 0;
3851 }
3852
3853 /**
3854  *      Inserts (key, value) pair in \a dt index object.
3855  *
3856  *      \param  dt      osd index object
3857  *      \param  key     key for index
3858  *      \param  rec     record reference
3859  *      \param  th      transaction handler
3860  *
3861  *      \retval  0  success
3862  *      \retval -ve failure
3863  */
3864 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3865                                 const struct dt_rec *rec,
3866                                 const struct dt_key *key, struct thandle *th,
3867                                 int ignore_quota)
3868 {
3869         struct osd_object     *obj = osd_dt_obj(dt);
3870         struct iam_path_descr *ipd;
3871         struct osd_thandle    *oh;
3872         struct iam_container  *bag;
3873         struct osd_thread_info *oti = osd_oti_get(env);
3874         struct iam_rec         *iam_rec;
3875         int                     rc;
3876         ENTRY;
3877
3878         if (!dt_object_exists(dt))
3879                 RETURN(-ENOENT);
3880
3881         LINVRNT(osd_invariant(obj));
3882         LASSERT(!dt_object_remote(dt));
3883
3884         bag = &obj->oo_dir->od_container;
3885         LASSERT(bag->ic_object == obj->oo_inode);
3886         LASSERT(th != NULL);
3887
3888         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3889
3890         ipd = osd_idx_ipd_get(env, bag);
3891         if (unlikely(ipd == NULL))
3892                 RETURN(-ENOMEM);
3893
3894         oh = container_of0(th, struct osd_thandle, ot_super);
3895         LASSERT(oh->ot_handle != NULL);
3896         LASSERT(oh->ot_handle->h_transaction != NULL);
3897         if (S_ISDIR(obj->oo_inode->i_mode)) {
3898                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3899                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3900         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3901                 /* pack quota uid/gid */
3902                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3903                 key = (const struct dt_key *)&oti->oti_quota_id;
3904                 /* pack quota record */
3905                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3906                 iam_rec = (struct iam_rec *)rec;
3907         } else {
3908                 iam_rec = (struct iam_rec *)rec;
3909         }
3910
3911         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3912                         iam_rec, ipd);
3913         osd_ipd_put(env, bag, ipd);
3914         LINVRNT(osd_invariant(obj));
3915         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3916         RETURN(rc);
3917 }
3918
3919 /**
3920  * Calls ldiskfs_add_entry() to add directory entry
3921  * into the directory. This is required for
3922  * interoperability mode (b11826)
3923  *
3924  * \retval   0, on success
3925  * \retval -ve, on error
3926  */
3927 static int __osd_ea_add_rec(struct osd_thread_info *info,
3928                             struct osd_object *pobj, struct inode  *cinode,
3929                             const char *name, const struct lu_fid *fid,
3930                             struct htree_lock *hlock, struct thandle *th)
3931 {
3932         struct ldiskfs_dentry_param *ldp;
3933         struct dentry               *child;
3934         struct osd_thandle          *oth;
3935         int                          rc;
3936
3937         oth = container_of(th, struct osd_thandle, ot_super);
3938         LASSERT(oth->ot_handle != NULL);
3939         LASSERT(oth->ot_handle->h_transaction != NULL);
3940         LASSERT(pobj->oo_inode);
3941
3942         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3943         if (unlikely(pobj->oo_inode ==
3944                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3945                 ldp->edp_magic = 0;
3946         else
3947                 osd_get_ldiskfs_dirent_param(ldp, fid);
3948         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3949         child->d_fsdata = (void *)ldp;
3950         ll_vfs_dq_init(pobj->oo_inode);
3951         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3952         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
3953                 struct ldiskfs_dir_entry_2      *de;
3954                 struct buffer_head              *bh;
3955                 int                              rc1;
3956
3957                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
3958                                             NULL, hlock);
3959                 if (bh != NULL) {
3960                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
3961                                                                bh);
3962                         if (rc1 == 0) {
3963                                 if (S_ISDIR(cinode->i_mode))
3964                                         de->file_type = LDISKFS_DIRENT_LUFID |
3965                                                         LDISKFS_FT_REG_FILE;
3966                                 else
3967                                         de->file_type = LDISKFS_DIRENT_LUFID |
3968                                                         LDISKFS_FT_DIR;
3969                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
3970                                                               NULL, bh);
3971                                 brelse(bh);
3972                         }
3973                 }
3974         }
3975
3976         RETURN(rc);
3977 }
3978
3979 /**
3980  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3981  * into the directory.Also sets flags into osd object to
3982  * indicate dot and dotdot are created. This is required for
3983  * interoperability mode (b11826)
3984  *
3985  * \param dir   directory for dot and dotdot fixup.
3986  * \param obj   child object for linking
3987  *
3988  * \retval   0, on success
3989  * \retval -ve, on error
3990  */
3991 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3992                               struct osd_object *dir,
3993                               struct inode *parent_dir, const char *name,
3994                               const struct lu_fid *dot_fid,
3995                               const struct lu_fid *dot_dot_fid,
3996                               struct thandle *th)
3997 {
3998         struct inode                *inode = dir->oo_inode;
3999         struct osd_thandle          *oth;
4000         int result = 0;
4001
4002         oth = container_of(th, struct osd_thandle, ot_super);
4003         LASSERT(oth->ot_handle->h_transaction != NULL);
4004         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
4005
4006         if (strcmp(name, dot) == 0) {
4007                 if (dir->oo_compat_dot_created) {
4008                         result = -EEXIST;
4009                 } else {
4010                         LASSERT(inode == parent_dir);
4011                         dir->oo_compat_dot_created = 1;
4012                         result = 0;
4013                 }
4014         } else if (strcmp(name, dotdot) == 0) {
4015                 if (!dir->oo_compat_dot_created)
4016                         return -EINVAL;
4017                 /* in case of rename, dotdot is already created */
4018                 if (dir->oo_compat_dotdot_created) {
4019                         return __osd_ea_add_rec(info, dir, parent_dir, name,
4020                                                 dot_dot_fid, NULL, th);
4021                 }
4022
4023                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
4024                         struct lu_fid tfid = *dot_dot_fid;
4025
4026                         tfid.f_oid--;
4027                         result = osd_add_dot_dotdot_internal(info,
4028                                         dir->oo_inode, parent_dir, dot_fid,
4029                                         &tfid, oth);
4030                 } else {
4031                         result = osd_add_dot_dotdot_internal(info,
4032                                         dir->oo_inode, parent_dir, dot_fid,
4033                                         dot_dot_fid, oth);
4034                 }
4035
4036                 if (result == 0)
4037                         dir->oo_compat_dotdot_created = 1;
4038         }
4039
4040         return result;
4041 }
4042
4043
4044 /**
4045  * It will call the appropriate osd_add* function and return the
4046  * value, return by respective functions.
4047  */
4048 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
4049                           struct inode *cinode, const char *name,
4050                           const struct lu_fid *fid, struct thandle *th)
4051 {
4052         struct osd_thread_info *info   = osd_oti_get(env);
4053         struct htree_lock      *hlock;
4054         int                     rc;
4055
4056         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
4057
4058         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
4059                                                    name[2] =='\0'))) {
4060                 if (hlock != NULL) {
4061                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4062                                            pobj->oo_inode, 0);
4063                 } else {
4064                         down_write(&pobj->oo_ext_idx_sem);
4065                 }
4066
4067                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
4068                                         lu_object_fid(&pobj->oo_dt.do_lu),
4069                                         fid, th);
4070         } else {
4071                 if (hlock != NULL) {
4072                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
4073                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
4074                 } else {
4075                         down_write(&pobj->oo_ext_idx_sem);
4076                 }
4077
4078                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
4079                         struct lu_fid *tfid = &info->oti_fid;
4080
4081                         *tfid = *fid;
4082                         tfid->f_ver = ~0;
4083                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
4084                                               tfid, hlock, th);
4085                 } else {
4086                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
4087                                               hlock, th);
4088                 }
4089         }
4090         if (hlock != NULL)
4091                 ldiskfs_htree_unlock(hlock);
4092         else
4093                 up_write(&pobj->oo_ext_idx_sem);
4094
4095         return rc;
4096 }
4097
4098 static int
4099 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
4100                       struct osd_idmap_cache *oic)
4101 {
4102         struct osd_scrub    *scrub = &dev->od_scrub;
4103         struct lu_fid       *fid   = &oic->oic_fid;
4104         struct osd_inode_id *id    = &oti->oti_id;
4105         int                  once  = 0;
4106         int                  rc;
4107         ENTRY;
4108
4109         if (!fid_is_norm(fid) && !fid_is_igif(fid))
4110                 RETURN(0);
4111
4112         if (scrub->os_pos_current > id->oii_ino)
4113                 RETURN(0);
4114
4115 again:
4116         rc = osd_oi_lookup(oti, dev, fid, id, 0);
4117         if (rc == -ENOENT) {
4118                 struct inode *inode;
4119
4120                 *id = oic->oic_lid;
4121                 inode = osd_iget(oti, dev, &oic->oic_lid);
4122
4123                 /* The inode has been removed (by race maybe). */
4124                 if (IS_ERR(inode)) {
4125                         rc = PTR_ERR(inode);
4126
4127                         RETURN(rc == -ESTALE ? -ENOENT : rc);
4128                 }
4129
4130                 iput(inode);
4131                 /* The OI mapping is lost. */
4132                 if (id->oii_gen != OSD_OII_NOGEN)
4133                         goto trigger;
4134
4135                 /* The inode may has been reused by others, we do not know,
4136                  * leave it to be handled by subsequent osd_fid_lookup(). */
4137                 RETURN(0);
4138         } else if (rc != 0 || osd_id_eq(id, &oic->oic_lid)) {
4139                 RETURN(rc);
4140         }
4141
4142 trigger:
4143         if (thread_is_running(&scrub->os_thread)) {
4144                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
4145                 /* There is race condition between osd_oi_lookup and OI scrub.
4146                  * The OI scrub finished just after osd_oi_lookup() failure.
4147                  * Under such case, it is unnecessary to trigger OI scrub again,
4148                  * but try to call osd_oi_lookup() again. */
4149                 if (unlikely(rc == -EAGAIN))
4150                         goto again;
4151
4152                 RETURN(0);
4153         }
4154
4155         if (!dev->od_noscrub && ++once == 1) {
4156                 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
4157                                      SS_CLEAR_FAILOUT);
4158                 CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI scrub by RPC "
4159                        "for "DFID", rc = %d [2]\n",
4160                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
4161                        PFID(fid), rc);
4162                 if (rc == 0 || rc == -EALREADY)
4163                         goto again;
4164         }
4165
4166         RETURN(0);
4167 }
4168
4169 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
4170                                struct osd_device *dev,
4171                                struct osd_idmap_cache *oic,
4172                                struct lu_fid *fid, __u32 ino)
4173 {
4174         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
4175         struct inode            *inode;
4176         int                      rc;
4177
4178         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4179         inode = osd_iget(oti, dev, &oic->oic_lid);
4180         if (IS_ERR(inode)) {
4181                 fid_zero(&oic->oic_fid);
4182                 return PTR_ERR(inode);
4183         }
4184
4185         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
4186         iput(inode);
4187         if (rc != 0)
4188                 fid_zero(&oic->oic_fid);
4189         else
4190                 *fid = oic->oic_fid = lma->lma_self_fid;
4191         return rc;
4192 }
4193
4194 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
4195                       struct osd_inode_id *id, const struct lu_fid *fid)
4196 {
4197         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
4198                id->oii_ino, id->oii_gen, info);
4199         info->oti_cache.oic_lid = *id;
4200         info->oti_cache.oic_fid = *fid;
4201         info->oti_cache.oic_dev = osd;
4202 }
4203
4204 /**
4205  * Get parent FID from the linkEA.
4206  *
4207  * For a directory which parent resides on remote MDT, to satisfy the
4208  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
4209  * the other hand, to make the lookup(..) on the directory can return
4210  * the real parent FID, we append the real parent FID after its ".."
4211  * name entry in the /REMOTE_PARENT_DIR.
4212  *
4213  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4214  * backup. So after the restore, we cannot get the right parent FID from
4215  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4216  * we have stored the real parent FID in the directory object's linkEA,
4217  * we can parse the linkEA for the real parent FID.
4218  *
4219  * \param[in] env       pointer to the thread context
4220  * \param[in] obj       pointer to the object to be handled
4221  * \param[out]fid       pointer to the buffer to hold the parent FID
4222  *
4223  * \retval              0 for getting the real parent FID successfully
4224  * \retval              negative error number on failure
4225  */
4226 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4227                                     struct osd_object *obj,
4228                                     struct lu_fid *fid)
4229 {
4230         struct osd_thread_info  *oti    = osd_oti_get(env);
4231         struct lu_buf           *buf    = &oti->oti_big_buf;
4232         struct dentry           *dentry = &oti->oti_obj_dentry;
4233         struct inode            *inode  = obj->oo_inode;
4234         struct linkea_data       ldata  = { NULL };
4235         int                      rc;
4236         ENTRY;
4237
4238         fid_zero(fid);
4239         if (!S_ISDIR(inode->i_mode))
4240                 RETURN(-EIO);
4241
4242 again:
4243         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4244                              buf->lb_buf, buf->lb_len);
4245         if (rc == -ERANGE) {
4246                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4247                                      NULL, 0);
4248                 if (rc > 0) {
4249                         lu_buf_realloc(buf, rc);
4250                         if (buf->lb_buf == NULL)
4251                                 RETURN(-ENOMEM);
4252
4253                         goto again;
4254                 }
4255         }
4256
4257         if (unlikely(rc == 0))
4258                 RETURN(-ENODATA);
4259
4260         if (rc < 0)
4261                 RETURN(rc);
4262
4263         if (unlikely(buf->lb_buf == NULL)) {
4264                 lu_buf_realloc(buf, rc);
4265                 if (buf->lb_buf == NULL)
4266                         RETURN(-ENOMEM);
4267
4268                 goto again;
4269         }
4270
4271         ldata.ld_buf = buf;
4272         rc = linkea_init(&ldata);
4273         if (rc == 0) {
4274                 linkea_first_entry(&ldata);
4275                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
4276         }
4277
4278         RETURN(rc);
4279 }
4280
4281 /**
4282  * Calls ->lookup() to find dentry. From dentry get inode and
4283  * read inode's ea to get fid. This is required for  interoperability
4284  * mode (b11826)
4285  *
4286  * \retval   0, on success
4287  * \retval -ve, on error
4288  */
4289 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4290                              struct dt_rec *rec, const struct dt_key *key)
4291 {
4292         struct inode                    *dir    = obj->oo_inode;
4293         struct dentry                   *dentry;
4294         struct ldiskfs_dir_entry_2      *de;
4295         struct buffer_head              *bh;
4296         struct lu_fid                   *fid = (struct lu_fid *) rec;
4297         struct htree_lock               *hlock = NULL;
4298         int                             ino;
4299         int                             rc;
4300         ENTRY;
4301
4302         LASSERT(dir->i_op != NULL);
4303         LASSERT(dir->i_op->lookup != NULL);
4304
4305         dentry = osd_child_dentry_get(env, obj,
4306                                       (char *)key, strlen((char *)key));
4307
4308         if (obj->oo_hl_head != NULL) {
4309                 hlock = osd_oti_get(env)->oti_hlock;
4310                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4311                                    dir, LDISKFS_HLOCK_LOOKUP);
4312         } else {
4313                 down_read(&obj->oo_ext_idx_sem);
4314         }
4315
4316         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4317         if (bh) {
4318                 struct osd_thread_info *oti = osd_oti_get(env);
4319                 struct osd_inode_id *id = &oti->oti_id;
4320                 struct osd_idmap_cache *oic = &oti->oti_cache;
4321                 struct osd_device *dev = osd_obj2dev(obj);
4322
4323                 ino = le32_to_cpu(de->inode);
4324                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4325                         brelse(bh);
4326                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4327                         GOTO(out, rc);
4328                 }
4329
4330                 rc = osd_get_fid_from_dentry(de, rec);
4331
4332                 /* done with de, release bh */
4333                 brelse(bh);
4334                 if (rc != 0) {
4335                         if (unlikely(ino == osd_remote_parent_ino(dev))) {
4336                                 const char *name = (const char *)key;
4337
4338                                 /* If the parent is on remote MDT, and there
4339                                  * is no FID-in-dirent, then we have to get
4340                                  * the parent FID from the linkEA.  */
4341                                 if (likely(strlen(name) == 2 &&
4342                                            name[0] == '.' && name[1] == '.'))
4343                                         rc = osd_get_pfid_from_linkea(env, obj,
4344                                                                       fid);
4345                         } else {
4346                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4347                         }
4348                 } else {
4349                         osd_id_gen(id, ino, OSD_OII_NOGEN);
4350                 }
4351
4352                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
4353                         fid_zero(&oic->oic_fid);
4354
4355                         GOTO(out, rc);
4356                 }
4357
4358                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
4359                 rc = osd_consistency_check(oti, dev, oic);
4360                 if (rc != 0)
4361                         fid_zero(&oic->oic_fid);
4362         } else {
4363                 rc = -ENOENT;
4364         }
4365
4366         GOTO(out, rc);
4367
4368 out:
4369         if (hlock != NULL)
4370                 ldiskfs_htree_unlock(hlock);
4371         else
4372                 up_read(&obj->oo_ext_idx_sem);
4373         return rc;
4374 }
4375
4376 /**
4377  * Find the osd object for given fid.
4378  *
4379  * \param fid need to find the osd object having this fid
4380  *
4381  * \retval osd_object on success
4382  * \retval        -ve on error
4383  */
4384 static struct osd_object *osd_object_find(const struct lu_env *env,
4385                                           struct dt_object *dt,
4386                                           const struct lu_fid *fid)
4387 {
4388         struct lu_device  *ludev = dt->do_lu.lo_dev;
4389         struct osd_object *child = NULL;
4390         struct lu_object  *luch;
4391         struct lu_object  *lo;
4392
4393         /*
4394          * at this point topdev might not exist yet
4395          * (i.e. MGS is preparing profiles). so we can
4396          * not rely on topdev and instead lookup with
4397          * our device passed as topdev. this can't work
4398          * if the object isn't cached yet (as osd doesn't
4399          * allocate lu_header). IOW, the object must be
4400          * in the cache, otherwise lu_object_alloc() crashes
4401          * -bzzz
4402          */
4403         luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4404                                  ludev : ludev->ld_site->ls_top_dev,
4405                                  fid, NULL);
4406         if (!IS_ERR(luch)) {
4407                 if (lu_object_exists(luch)) {
4408                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4409                         if (lo != NULL)
4410                                 child = osd_obj(lo);
4411                         else
4412                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4413                                                 "lu_object can't be located"
4414                                                 DFID"\n", PFID(fid));
4415
4416                         if (child == NULL) {
4417                                 lu_object_put(env, luch);
4418                                 CERROR("Unable to get osd_object\n");
4419                                 child = ERR_PTR(-ENOENT);
4420                         }
4421                 } else {
4422                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
4423                                         "lu_object does not exists "DFID"\n",
4424                                         PFID(fid));
4425                         lu_object_put(env, luch);
4426                         child = ERR_PTR(-ENOENT);
4427                 }
4428         } else {
4429                 child = ERR_CAST(luch);
4430         }
4431
4432         return child;
4433 }
4434
4435 /**
4436  * Put the osd object once done with it.
4437  *
4438  * \param obj osd object that needs to be put
4439  */
4440 static inline void osd_object_put(const struct lu_env *env,
4441                                   struct osd_object *obj)
4442 {
4443         lu_object_put(env, &obj->oo_dt.do_lu);
4444 }
4445
4446 static int osd_index_declare_ea_insert(const struct lu_env *env,
4447                                        struct dt_object *dt,
4448                                        const struct dt_rec *rec,
4449                                        const struct dt_key *key,
4450                                        struct thandle *handle)
4451 {
4452         struct osd_thandle      *oh;
4453         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
4454         struct lu_fid           *fid = (struct lu_fid *)rec;
4455         int                      credits, rc = 0;
4456         ENTRY;
4457
4458         LASSERT(!dt_object_remote(dt));
4459         LASSERT(handle != NULL);
4460
4461         oh = container_of0(handle, struct osd_thandle, ot_super);
4462         LASSERT(oh->ot_handle == NULL);
4463
4464         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
4465         if (fid != NULL) {
4466                 rc = osd_remote_fid(env, osd, fid);
4467                 if (unlikely(rc < 0))
4468                         RETURN(rc);
4469                 if (rc > 0) {
4470                         /* a reference to remote inode is represented by an
4471                          * agent inode which we have to create */
4472                         credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
4473                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4474                 }
4475                 rc = 0;
4476         }
4477
4478         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
4479
4480         if (osd_dt_obj(dt)->oo_inode != NULL) {
4481                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4482
4483                 /* We ignore block quota on meta pool (MDTs), so needn't
4484                  * calculate how many blocks will be consumed by this index
4485                  * insert */
4486                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4487                                            i_gid_read(inode), 0, oh,
4488                                            osd_dt_obj(dt), true, NULL, false);
4489         }
4490
4491         RETURN(rc);
4492 }
4493
4494 /**
4495  * Index add function for interoperability mode (b11826).
4496  * It will add the directory entry.This entry is needed to
4497  * maintain name->fid mapping.
4498  *
4499  * \param key it is key i.e. file entry to be inserted
4500  * \param rec it is value of given key i.e. fid
4501  *
4502  * \retval   0, on success
4503  * \retval -ve, on error
4504  */
4505 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4506                                const struct dt_rec *rec,
4507                                const struct dt_key *key, struct thandle *th,
4508                                int ignore_quota)
4509 {
4510         struct osd_object       *obj = osd_dt_obj(dt);
4511         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
4512         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
4513         const struct lu_fid     *fid    = rec1->rec_fid;
4514         const char              *name = (const char *)key;
4515         struct osd_thread_info  *oti   = osd_oti_get(env);
4516         struct osd_inode_id     *id    = &oti->oti_id;
4517         struct inode            *child_inode = NULL;
4518         struct osd_object       *child = NULL;
4519         int                     rc;
4520         ENTRY;
4521
4522         if (!dt_object_exists(dt))
4523                 RETURN(-ENOENT);
4524
4525         LASSERT(osd_invariant(obj));
4526         LASSERT(!dt_object_remote(dt));
4527         LASSERT(th != NULL);
4528
4529         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4530
4531         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
4532
4533         rc = osd_remote_fid(env, osd, fid);
4534         if (rc < 0) {
4535                 CERROR("%s: Can not find object "DFID" rc %d\n",
4536                        osd_name(osd), PFID(fid), rc);
4537                 RETURN(rc);
4538         }
4539
4540         if (rc == 1) {
4541                 /* Insert remote entry */
4542                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4543                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
4544                         struct osd_thandle      *oh;
4545
4546                         /* If parent on remote MDT, we need put this object
4547                          * under AGENT */
4548                         oh = container_of(th, typeof(*oh), ot_super);
4549                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
4550                         if (rc != 0) {
4551                                 CERROR("%s: add "DFID" error: rc = %d\n",
4552                                        osd_name(osd),
4553                                        PFID(lu_object_fid(&dt->do_lu)), rc);
4554                                 RETURN(rc);
4555                         }
4556
4557                         child_inode = igrab(omm->omm_remote_parent->d_inode);
4558                 } else {
4559                         child_inode = osd_create_local_agent_inode(env, osd,
4560                                         obj, fid, rec1->rec_type & S_IFMT, th);
4561                         if (IS_ERR(child_inode))
4562                                 RETURN(PTR_ERR(child_inode));
4563                 }
4564         } else {
4565                 /* Insert local entry */
4566                 child = osd_object_find(env, dt, fid);
4567                 if (IS_ERR(child)) {
4568                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4569                                osd_name(osd), PFID(fid),
4570                                id->oii_ino, id->oii_gen,
4571                                (int)PTR_ERR(child));
4572                         RETURN(PTR_ERR(child));
4573                 }
4574                 child_inode = igrab(child->oo_inode);
4575         }
4576
4577         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
4578
4579         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4580                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4581
4582         iput(child_inode);
4583         if (child != NULL)
4584                 osd_object_put(env, child);
4585         LASSERT(osd_invariant(obj));
4586         osd_trans_exec_check(env, th, OSD_OT_INSERT);
4587         RETURN(rc);
4588 }
4589
4590 /**
4591  *  Initialize osd Iterator for given osd index object.
4592  *
4593  *  \param  dt      osd index object
4594  */
4595
4596 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4597                                      struct dt_object *dt,
4598                                      __u32 unused)
4599 {
4600         struct osd_it_iam      *it;
4601         struct osd_object      *obj = osd_dt_obj(dt);
4602         struct lu_object       *lo  = &dt->do_lu;
4603         struct iam_path_descr  *ipd;
4604         struct iam_container   *bag = &obj->oo_dir->od_container;
4605
4606         if (!dt_object_exists(dt))
4607                 return ERR_PTR(-ENOENT);
4608
4609         OBD_ALLOC_PTR(it);
4610         if (it == NULL)
4611                 return ERR_PTR(-ENOMEM);
4612
4613         ipd = osd_it_ipd_get(env, bag);
4614         if (likely(ipd != NULL)) {
4615                 it->oi_obj = obj;
4616                 it->oi_ipd = ipd;
4617                 lu_object_get(lo);
4618                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4619                 return (struct dt_it *)it;
4620         } else {
4621                 OBD_FREE_PTR(it);
4622                 return ERR_PTR(-ENOMEM);
4623         }
4624 }
4625
4626 /**
4627  * free given Iterator.
4628  */
4629
4630 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4631 {
4632         struct osd_it_iam       *it  = (struct osd_it_iam *)di;
4633         struct osd_object       *obj = it->oi_obj;
4634
4635         iam_it_fini(&it->oi_it);
4636         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4637         lu_object_put(env, &obj->oo_dt.do_lu);
4638         OBD_FREE_PTR(it);
4639 }
4640
4641 /**
4642  *  Move Iterator to record specified by \a key
4643  *
4644  *  \param  di      osd iterator
4645  *  \param  key     key for index
4646  *
4647  *  \retval +ve  di points to record with least key not larger than key
4648  *  \retval  0   di points to exact matched key
4649  *  \retval -ve  failure
4650  */
4651
4652 static int osd_it_iam_get(const struct lu_env *env,
4653                           struct dt_it *di, const struct dt_key *key)
4654 {
4655         struct osd_thread_info  *oti = osd_oti_get(env);
4656         struct osd_it_iam       *it = (struct osd_it_iam *)di;
4657
4658         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4659                 /* swab quota uid/gid */
4660                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4661                 key = (struct dt_key *)&oti->oti_quota_id;
4662         }
4663
4664         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4665 }
4666
4667 /**
4668  *  Release Iterator
4669  *
4670  *  \param  di      osd iterator
4671  */
4672 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4673 {
4674         struct osd_it_iam *it = (struct osd_it_iam *)di;
4675
4676         iam_it_put(&it->oi_it);
4677 }
4678
4679 /**
4680  *  Move iterator by one record
4681  *
4682  *  \param  di      osd iterator
4683  *
4684  *  \retval +1   end of container reached
4685  *  \retval  0   success
4686  *  \retval -ve  failure
4687  */
4688
4689 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4690 {
4691         struct osd_it_iam *it = (struct osd_it_iam *)di;
4692
4693         return iam_it_next(&it->oi_it);
4694 }
4695
4696 /**
4697  * Return pointer to the key under iterator.
4698  */
4699
4700 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4701                                  const struct dt_it *di)
4702 {
4703         struct osd_thread_info *oti = osd_oti_get(env);
4704         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4705         struct osd_object      *obj = it->oi_obj;
4706         struct dt_key          *key;
4707
4708         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4709
4710         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4711                 /* swab quota uid/gid */
4712                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4713                 key = (struct dt_key *)&oti->oti_quota_id;
4714         }
4715
4716         return key;
4717 }
4718
4719 /**
4720  * Return size of key under iterator (in bytes)
4721  */
4722
4723 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4724 {
4725         struct osd_it_iam *it = (struct osd_it_iam *)di;
4726
4727         return iam_it_key_size(&it->oi_it);
4728 }
4729
4730 static inline void
4731 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4732 {
4733         /* check if file type is required */
4734         if (ent->lde_attrs & LUDA_TYPE) {
4735                 struct luda_type *lt;
4736                 int align = sizeof(*lt) - 1;
4737
4738                 len = (len + align) & ~align;
4739                 lt = (struct luda_type *)(ent->lde_name + len);
4740                 lt->lt_type = cpu_to_le16(DTTOIF(type));
4741         }
4742
4743         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4744 }
4745
4746 /**
4747  * build lu direct from backend fs dirent.
4748  */
4749
4750 static inline void
4751 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4752                    char *name, __u16 namelen, __u16 type, __u32 attr)
4753 {
4754         ent->lde_attrs = attr | LUDA_FID;
4755         fid_cpu_to_le(&ent->lde_fid, fid);
4756
4757         ent->lde_hash = cpu_to_le64(offset);
4758         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4759
4760         strncpy(ent->lde_name, name, namelen);
4761         ent->lde_name[namelen] = '\0';
4762         ent->lde_namelen = cpu_to_le16(namelen);
4763
4764         /* append lustre attributes */
4765         osd_it_append_attrs(ent, namelen, type);
4766 }
4767
4768 /**
4769  * Return pointer to the record under iterator.
4770  */
4771 static int osd_it_iam_rec(const struct lu_env *env,
4772                           const struct dt_it *di,
4773                           struct dt_rec *dtrec, __u32 attr)
4774 {
4775         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4776         struct osd_thread_info *info = osd_oti_get(env);
4777         ENTRY;
4778
4779         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4780                 const struct osd_fid_pack *rec;
4781                 struct lu_fid             *fid = &info->oti_fid;
4782                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4783                 char                      *name;
4784                 int                        namelen;
4785                 __u64                      hash;
4786                 int                        rc;
4787
4788                 name = (char *)iam_it_key_get(&it->oi_it);
4789                 if (IS_ERR(name))
4790                         RETURN(PTR_ERR(name));
4791
4792                 namelen = iam_it_key_size(&it->oi_it);
4793
4794                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4795                 if (IS_ERR(rec))
4796                         RETURN(PTR_ERR(rec));
4797
4798                 rc = osd_fid_unpack(fid, rec);
4799                 if (rc)
4800                         RETURN(rc);
4801
4802                 hash = iam_it_store(&it->oi_it);
4803
4804                 /* IAM does not store object type in IAM index (dir) */
4805                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4806                                    0, LUDA_FID);
4807         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4808                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4809                            (struct iam_rec *)dtrec);
4810                 osd_quota_unpack(it->oi_obj, dtrec);
4811         } else {
4812                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4813                            (struct iam_rec *)dtrec);
4814         }
4815
4816         RETURN(0);
4817 }
4818
4819 /**
4820  * Returns cookie for current Iterator position.
4821  */
4822 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4823 {
4824         struct osd_it_iam *it = (struct osd_it_iam *)di;
4825
4826         return iam_it_store(&it->oi_it);
4827 }
4828
4829 /**
4830  * Restore iterator from cookie.
4831  *
4832  * \param  di      osd iterator
4833  * \param  hash    Iterator location cookie
4834  *
4835  * \retval +ve  di points to record with least key not larger than key.
4836  * \retval  0   di points to exact matched key
4837  * \retval -ve  failure
4838  */
4839
4840 static int osd_it_iam_load(const struct lu_env *env,
4841                            const struct dt_it *di, __u64 hash)
4842 {
4843         struct osd_it_iam *it = (struct osd_it_iam *)di;
4844
4845         return iam_it_load(&it->oi_it, hash);
4846 }
4847
4848 static const struct dt_index_operations osd_index_iam_ops = {
4849         .dio_lookup         = osd_index_iam_lookup,
4850         .dio_declare_insert = osd_index_declare_iam_insert,
4851         .dio_insert         = osd_index_iam_insert,
4852         .dio_declare_delete = osd_index_declare_iam_delete,
4853         .dio_delete         = osd_index_iam_delete,
4854         .dio_it     = {
4855                 .init     = osd_it_iam_init,
4856                 .fini     = osd_it_iam_fini,
4857                 .get      = osd_it_iam_get,
4858                 .put      = osd_it_iam_put,
4859                 .next     = osd_it_iam_next,
4860                 .key      = osd_it_iam_key,
4861                 .key_size = osd_it_iam_key_size,
4862                 .rec      = osd_it_iam_rec,
4863                 .store    = osd_it_iam_store,
4864                 .load     = osd_it_iam_load
4865         }
4866 };
4867
4868
4869 /**
4870  * Creates or initializes iterator context.
4871  *
4872  * \retval struct osd_it_ea, iterator structure on success
4873  *
4874  */
4875 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4876                                     struct dt_object *dt,
4877                                     __u32 attr)
4878 {
4879         struct osd_object       *obj  = osd_dt_obj(dt);
4880         struct osd_thread_info  *info = osd_oti_get(env);
4881         struct osd_it_ea        *oie;
4882         struct file             *file;
4883         struct lu_object        *lo   = &dt->do_lu;
4884         struct dentry           *obj_dentry;
4885         ENTRY;
4886
4887         if (!dt_object_exists(dt) || obj->oo_destroyed)
4888                 RETURN(ERR_PTR(-ENOENT));
4889
4890         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
4891         if (oie == NULL)
4892                 RETURN(ERR_PTR(-ENOMEM));
4893         obj_dentry = &oie->oie_dentry;
4894
4895         obj_dentry->d_inode = obj->oo_inode;
4896         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4897         obj_dentry->d_name.hash = 0;
4898
4899         oie->oie_rd_dirent       = 0;
4900         oie->oie_it_dirent       = 0;
4901         oie->oie_dirent          = NULL;
4902         if (unlikely(!info->oti_it_ea_buf_used)) {
4903                 oie->oie_buf = info->oti_it_ea_buf;
4904                 info->oti_it_ea_buf_used = 1;
4905         } else {
4906                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
4907                 if (oie->oie_buf == NULL)
4908                         RETURN(ERR_PTR(-ENOMEM));
4909         }
4910         oie->oie_obj             = obj;
4911
4912         file = &oie->oie_file;
4913
4914         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4915         if (attr & LUDA_64BITHASH)
4916                 file->f_mode    = FMODE_64BITHASH;
4917         else
4918                 file->f_mode    = FMODE_32BITHASH;
4919         file->f_path.dentry     = obj_dentry;
4920         file->f_mapping         = obj->oo_inode->i_mapping;
4921         file->f_op              = obj->oo_inode->i_fop;
4922         set_file_inode(file, obj->oo_inode);
4923
4924         lu_object_get(lo);
4925         RETURN((struct dt_it *) oie);
4926 }
4927
4928 /**
4929  * Destroy or finishes iterator context.
4930  *
4931  * \param di iterator structure to be destroyed
4932  */
4933 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4934 {
4935         struct osd_thread_info  *info = osd_oti_get(env);
4936         struct osd_it_ea        *oie    = (struct osd_it_ea *)di;
4937         struct osd_object       *obj    = oie->oie_obj;
4938         struct inode            *inode  = obj->oo_inode;
4939
4940         ENTRY;
4941         oie->oie_file.f_op->release(inode, &oie->oie_file);
4942         lu_object_put(env, &obj->oo_dt.do_lu);
4943         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
4944                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
4945         else
4946                 info->oti_it_ea_buf_used = 0;
4947         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
4948         EXIT;
4949 }
4950
4951 /**
4952  * It position the iterator at given key, so that next lookup continues from
4953  * that key Or it is similar to dio_it->load() but based on a key,
4954  * rather than file position.
4955  *
4956  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4957  * to the beginning.
4958  *
4959  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4960  */
4961 static int osd_it_ea_get(const struct lu_env *env,
4962                          struct dt_it *di, const struct dt_key *key)
4963 {
4964         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4965
4966         ENTRY;
4967         LASSERT(((const char *)key)[0] == '\0');
4968         it->oie_file.f_pos      = 0;
4969         it->oie_rd_dirent       = 0;
4970         it->oie_it_dirent       = 0;
4971         it->oie_dirent          = NULL;
4972
4973         RETURN(+1);
4974 }
4975
4976 /**
4977  * Does nothing
4978  */
4979 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4980 {
4981 }
4982
4983 struct osd_filldir_cbs {
4984 #ifdef HAVE_DIR_CONTEXT
4985         struct dir_context ctx;
4986 #endif
4987         struct osd_it_ea  *it;
4988 };
4989 /**
4990  * It is called internally by ->readdir(). It fills the
4991  * iterator's in-memory data structure with required
4992  * information i.e. name, namelen, rec_size etc.
4993  *
4994  * \param buf in which information to be filled in.
4995  * \param name name of the file in given dir
4996  *
4997  * \retval 0 on success
4998  * \retval 1 on buffer full
4999  */
5000 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
5001                                loff_t offset, __u64 ino,
5002                                unsigned d_type)
5003 {
5004         struct osd_it_ea        *it   = ((struct osd_filldir_cbs *)buf)->it;
5005         struct osd_object       *obj  = it->oie_obj;
5006         struct osd_it_ea_dirent *ent  = it->oie_dirent;
5007         struct lu_fid           *fid  = &ent->oied_fid;
5008         struct osd_fid_pack     *rec;
5009         ENTRY;
5010
5011         /* this should never happen */
5012         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
5013                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
5014                 RETURN(-EIO);
5015         }
5016
5017         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
5018             OSD_IT_EA_BUFSIZE)
5019                 RETURN(1);
5020
5021         /* "." is just the object itself. */
5022         if (namelen == 1 && name[0] == '.') {
5023                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5024         } else if (d_type & LDISKFS_DIRENT_LUFID) {
5025                 rec = (struct osd_fid_pack*) (name + namelen + 1);
5026                 if (osd_fid_unpack(fid, rec) != 0)
5027                         fid_zero(fid);
5028         } else {
5029                 fid_zero(fid);
5030         }
5031         d_type &= ~LDISKFS_DIRENT_LUFID;
5032
5033         /* NOT export local root. */
5034         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
5035                 ino = obj->oo_inode->i_ino;
5036                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
5037         }
5038
5039         ent->oied_ino     = ino;
5040         ent->oied_off     = offset;
5041         ent->oied_namelen = namelen;
5042         ent->oied_type    = d_type;
5043
5044         memcpy(ent->oied_name, name, namelen);
5045
5046         it->oie_rd_dirent++;
5047         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
5048         RETURN(0);
5049 }
5050
5051 /**
5052  * Calls ->readdir() to load a directory entry at a time
5053  * and stored it in iterator's in-memory data structure.
5054  *
5055  * \param di iterator's in memory structure
5056  *
5057  * \retval   0 on success
5058  * \retval -ve on error
5059  * \retval +1 reach the end of entry
5060  */
5061 static int osd_ldiskfs_it_fill(const struct lu_env *env,
5062                                const struct dt_it *di)
5063 {
5064         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
5065         struct osd_object  *obj   = it->oie_obj;
5066         struct inode       *inode = obj->oo_inode;
5067         struct htree_lock  *hlock = NULL;
5068         struct file        *filp  = &it->oie_file;
5069         int                 rc = 0;
5070         struct osd_filldir_cbs buf = {
5071 #ifdef HAVE_DIR_CONTEXT
5072                 .ctx.actor = osd_ldiskfs_filldir,
5073 #endif
5074                 .it = it
5075         };
5076
5077         ENTRY;
5078         it->oie_dirent = it->oie_buf;
5079         it->oie_rd_dirent = 0;
5080
5081         if (obj->oo_hl_head != NULL) {
5082                 hlock = osd_oti_get(env)->oti_hlock;
5083                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5084                                    inode, LDISKFS_HLOCK_READDIR);
5085         } else {
5086                 down_read(&obj->oo_ext_idx_sem);
5087         }
5088
5089 #ifdef HAVE_DIR_CONTEXT
5090         buf.ctx.pos = filp->f_pos;
5091         rc = inode->i_fop->iterate(filp, &buf.ctx);
5092         filp->f_pos = buf.ctx.pos;
5093 #else
5094         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
5095 #endif
5096
5097         if (hlock != NULL)
5098                 ldiskfs_htree_unlock(hlock);
5099         else
5100                 up_read(&obj->oo_ext_idx_sem);
5101
5102         if (it->oie_rd_dirent == 0) {
5103                 /*If it does not get any dirent, it means it has been reached
5104                  *to the end of the dir */
5105                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
5106                 if (rc == 0)
5107                         rc = 1;
5108         } else {
5109                 it->oie_dirent = it->oie_buf;
5110                 it->oie_it_dirent = 1;
5111         }
5112
5113         RETURN(rc);
5114 }
5115
5116 /**
5117  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5118  * to load a directory entry at a time and stored it in
5119  * iterator's in-memory data structure.
5120  *
5121  * \param di iterator's in memory structure
5122  *
5123  * \retval +ve iterator reached to end
5124  * \retval   0 iterator not reached to end
5125  * \retval -ve on error
5126  */
5127 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
5128 {
5129         struct osd_it_ea *it = (struct osd_it_ea *)di;
5130         int rc;
5131
5132         ENTRY;
5133
5134         if (it->oie_it_dirent < it->oie_rd_dirent) {
5135                 it->oie_dirent =
5136                         (void *) it->oie_dirent +
5137                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
5138                                        it->oie_dirent->oied_namelen);
5139                 it->oie_it_dirent++;
5140                 RETURN(0);
5141         } else {
5142                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
5143                         rc = +1;
5144                 else
5145                         rc = osd_ldiskfs_it_fill(env, di);
5146         }
5147
5148         RETURN(rc);
5149 }
5150
5151 /**
5152  * Returns the key at current position from iterator's in memory structure.
5153  *
5154  * \param di iterator's in memory structure
5155  *
5156  * \retval key i.e. struct dt_key on success
5157  */
5158 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
5159                                     const struct dt_it *di)
5160 {
5161         struct osd_it_ea *it = (struct osd_it_ea *)di;
5162
5163         return (struct dt_key *)it->oie_dirent->oied_name;
5164 }
5165
5166 /**
5167  * Returns the key's size at current position from iterator's in memory structure.
5168  *
5169  * \param di iterator's in memory structure
5170  *
5171  * \retval key_size i.e. struct dt_key on success
5172  */
5173 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
5174 {
5175         struct osd_it_ea *it = (struct osd_it_ea *)di;
5176
5177         return it->oie_dirent->oied_namelen;
5178 }
5179
5180 static inline bool
5181 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5182 {
5183         LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5184                  "dot_dotdot = %d\n", dot_dotdot);
5185
5186         if (LDISKFS_DIR_REC_LEN(de) >=
5187             __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5188                 return true;
5189
5190         return false;
5191 }
5192
5193 static inline bool
5194 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
5195                      unsigned blocksize, int dot_dotdot)
5196 {
5197         if (dot_dotdot > 0)
5198                 return osd_dot_dotdot_has_space(de, dot_dotdot);
5199
5200         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
5201             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
5202                 return true;
5203
5204         return false;
5205 }
5206
5207 static int
5208 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5209                     struct dentry *dentry, const struct lu_fid *fid,
5210                     struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5211                     struct htree_lock *hlock, int dot_dotdot)
5212 {
5213         struct inode                *dir        = dentry->d_parent->d_inode;
5214         struct inode                *inode      = dentry->d_inode;
5215         struct osd_fid_pack         *rec;
5216         struct ldiskfs_dentry_param *ldp;
5217         int                          namelen    = dentry->d_name.len;
5218         int                          rc;
5219         ENTRY;
5220
5221         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5222                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5223                 RETURN(0);
5224
5225         /* There is enough space to hold the FID-in-dirent. */
5226         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize,
5227                                  dot_dotdot)) {
5228                 rc = ldiskfs_journal_get_write_access(jh, bh);
5229                 if (rc != 0)
5230                         RETURN(rc);
5231
5232                 de->name[namelen] = 0;
5233                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
5234                 rec->fp_len = sizeof(struct lu_fid) + 1;
5235                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5236                 de->file_type |= LDISKFS_DIRENT_LUFID;
5237                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
5238
5239                 RETURN(rc);
5240         }
5241
5242         LASSERTF(dot_dotdot == 0, "dot_dotdot = %d\n", dot_dotdot);
5243
5244         rc = ldiskfs_delete_entry(jh, dir, de, bh);
5245         if (rc != 0)
5246                 RETURN(rc);
5247
5248         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5249         osd_get_ldiskfs_dirent_param(ldp, fid);
5250         dentry->d_fsdata = (void *)ldp;
5251         ll_vfs_dq_init(dir);
5252         rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5253         /* It is too bad, we cannot reinsert the name entry back.
5254          * That means we lose it! */
5255         if (rc != 0)
5256                 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5257                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5258                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5259                        dir->i_ino, dir->i_generation, namelen,
5260                        dentry->d_name.name, PFID(fid), rc);
5261
5262         RETURN(rc);
5263 }
5264
5265 static int
5266 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5267                         struct osd_it_ea *it, struct lu_fid *fid,
5268                         struct osd_inode_id *id, __u32 *attr)
5269 {
5270         struct osd_thread_info     *info        = osd_oti_get(env);
5271         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
5272         struct osd_device          *dev         = osd_obj2dev(obj);
5273         struct super_block         *sb          = osd_sb(dev);
5274         const char                 *devname     =
5275                                         LDISKFS_SB(sb)->s_es->s_volume_name;
5276         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
5277         struct inode               *dir         = obj->oo_inode;
5278         struct htree_lock          *hlock       = NULL;
5279         struct buffer_head         *bh          = NULL;
5280         handle_t                   *jh          = NULL;
5281         struct ldiskfs_dir_entry_2 *de;
5282         struct dentry              *dentry;
5283         struct inode               *inode;
5284         int                         credits;
5285         int                         rc;
5286         int                         dot_dotdot  = 0;
5287         bool                        dirty       = false;
5288         ENTRY;
5289
5290         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5291         inode = osd_iget(info, dev, id);
5292         if (IS_ERR(inode)) {
5293                 rc = PTR_ERR(inode);
5294                 if (rc == -ENOENT || rc == -ESTALE) {
5295                         *attr |= LUDA_UNKNOWN;
5296                         rc = 0;
5297                 } else {
5298                         CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
5299                                "check_repair, dir = %lu/%u, name = %.*s: "
5300                                "rc = %d\n",
5301                                devname, dir->i_ino, dir->i_generation,
5302                                ent->oied_namelen, ent->oied_name, rc);
5303                 }
5304
5305                 RETURN(rc);
5306         }
5307
5308         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5309                                            ent->oied_namelen);
5310         rc = osd_get_lma(info, inode, dentry, lma);
5311         if (rc == -ENODATA)
5312                 lma = NULL;
5313         else if (rc != 0)
5314                 GOTO(out, rc);
5315
5316         if (ent->oied_name[0] == '.') {
5317                 if (ent->oied_namelen == 1)
5318                         dot_dotdot = 1;
5319                 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5320                         dot_dotdot = 2;
5321         }
5322
5323         /* We need to ensure that the name entry is still valid.
5324          * Because it may be removed or renamed by other already.
5325          *
5326          * The unlink or rename operation will start journal before PDO lock,
5327          * so to avoid deadlock, here we need to start journal handle before
5328          * related PDO lock also. But because we do not know whether there
5329          * will be something to be repaired before PDO lock, we just start
5330          * journal without conditions.
5331          *
5332          * We may need to remove the name entry firstly, then insert back.
5333          * One credit is for user quota file update.
5334          * One credit is for group quota file update.
5335          * Two credits are for dirty inode. */
5336         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5337                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5338
5339         if (dev->od_dirent_journal != 0) {
5340
5341 again:
5342                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5343                 if (IS_ERR(jh)) {
5344                         rc = PTR_ERR(jh);
5345                         CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
5346                                "check_repair, dir = %lu/%u, credits = %d, "
5347                                "name = %.*s: rc = %d\n",
5348                                devname, dir->i_ino, dir->i_generation, credits,
5349                                ent->oied_namelen, ent->oied_name, rc);
5350
5351                         GOTO(out_inode, rc);
5352                 }
5353
5354                 if (obj->oo_hl_head != NULL) {
5355                         hlock = osd_oti_get(env)->oti_hlock;
5356                         /* "0" means exclusive lock for the whole directory.
5357                          * We need to prevent others access such name entry
5358                          * during the delete + insert. Neither HLOCK_ADD nor
5359                          * HLOCK_DEL cannot guarantee the atomicity. */
5360                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5361                 } else {
5362                         down_write(&obj->oo_ext_idx_sem);
5363                 }
5364         } else {
5365                 if (obj->oo_hl_head != NULL) {
5366                         hlock = osd_oti_get(env)->oti_hlock;
5367                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5368                                            LDISKFS_HLOCK_LOOKUP);
5369                 } else {
5370                         down_read(&obj->oo_ext_idx_sem);
5371                 }
5372         }
5373
5374         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5375         /* For dot/dotdot entry, if there is not enough space to hold the
5376          * FID-in-dirent, just keep them there. It only happens when the
5377          * device upgraded from 1.8 or restored from MDT file-level backup.
5378          * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5379          * and needs to get FID from LMA when readdir, it will not affect the
5380          * performance much. */
5381         if ((bh == NULL) || (le32_to_cpu(de->inode) != inode->i_ino) ||
5382             (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5383                 *attr |= LUDA_IGNORE;
5384
5385                 GOTO(out, rc = 0);
5386         }
5387
5388         if (lma != NULL) {
5389                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
5390                         struct lu_fid *tfid = &lma->lma_self_fid;
5391
5392                         *attr |= LUDA_IGNORE;
5393                         /* It must be REMOTE_PARENT_DIR and as the
5394                          * dotdot entry of remote directory */
5395                         if (unlikely(dot_dotdot != 2 ||
5396                                      fid_seq(tfid) != FID_SEQ_LOCAL_FILE ||
5397                                      fid_oid(tfid) != REMOTE_PARENT_DIR_OID)) {
5398                                 CDEBUG(D_LFSCK, "%.16s: expect remote agent "
5399                                        "parent directory, but got %.*s under "
5400                                        "dir = %lu/%u with the FID "DFID"\n",
5401                                        devname, ent->oied_namelen,
5402                                        ent->oied_name, dir->i_ino,
5403                                        dir->i_generation, PFID(tfid));
5404
5405                                 GOTO(out, rc = -EIO);
5406                         }
5407
5408                         GOTO(out, rc = 0);
5409                 }
5410
5411                 if (fid_is_sane(fid)) {
5412                         /* FID-in-dirent is valid. */
5413                         if (lu_fid_eq(fid, &lma->lma_self_fid))
5414                                 GOTO(out, rc = 0);
5415
5416                         /* Do not repair under dryrun mode. */
5417                         if (*attr & LUDA_VERIFY_DRYRUN) {
5418                                 *attr |= LUDA_REPAIR;
5419
5420                                 GOTO(out, rc = 0);
5421                         }
5422
5423                         if (jh == NULL) {
5424                                 brelse(bh);
5425                                 if (hlock != NULL)
5426                                         ldiskfs_htree_unlock(hlock);
5427                                 else
5428                                         up_read(&obj->oo_ext_idx_sem);
5429                                 dev->od_dirent_journal = 1;
5430
5431                                 goto again;
5432                         }
5433
5434                         *fid = lma->lma_self_fid;
5435                         dirty = true;
5436                         /* Update the FID-in-dirent. */
5437                         rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de,
5438                                                  hlock, dot_dotdot);
5439                         if (rc == 0)
5440                                 *attr |= LUDA_REPAIR;
5441                         else
5442                                 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
5443                                        "in the dirent, dir = %lu/%u, "
5444                                        "name = %.*s, "DFID": rc = %d\n",
5445                                        devname, dir->i_ino, dir->i_generation,
5446                                        ent->oied_namelen, ent->oied_name,
5447                                        PFID(fid), rc);
5448                 } else {
5449                         /* Do not repair under dryrun mode. */
5450                         if (*attr & LUDA_VERIFY_DRYRUN) {
5451                                 *fid = lma->lma_self_fid;
5452                                 *attr |= LUDA_REPAIR;
5453
5454                                 GOTO(out, rc = 0);
5455                         }
5456
5457                         if (jh == NULL) {
5458                                 brelse(bh);
5459                                 if (hlock != NULL)
5460                                         ldiskfs_htree_unlock(hlock);
5461                                 else
5462                                         up_read(&obj->oo_ext_idx_sem);
5463                                 dev->od_dirent_journal = 1;
5464
5465                                 goto again;
5466                         }
5467
5468                         *fid = lma->lma_self_fid;
5469                         dirty = true;
5470                         /* Append the FID-in-dirent. */
5471                         rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de,
5472                                                  hlock, dot_dotdot);
5473                         if (rc == 0)
5474                                 *attr |= LUDA_REPAIR;
5475                         else
5476                                 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
5477                                        "after the dirent, dir = %lu/%u, "
5478                                        "name = %.*s, "DFID": rc = %d\n",
5479                                        devname, dir->i_ino, dir->i_generation,
5480                                        ent->oied_namelen, ent->oied_name,
5481                                        PFID(fid), rc);
5482                 }
5483         } else {
5484                 /* Do not repair under dryrun mode. */
5485                 if (*attr & LUDA_VERIFY_DRYRUN) {
5486                         if (fid_is_sane(fid)) {
5487                                 *attr |= LUDA_REPAIR;
5488                         } else {
5489                                 lu_igif_build(fid, inode->i_ino,
5490                                               inode->i_generation);
5491                                 *attr |= LUDA_UPGRADE;
5492                         }
5493
5494                         GOTO(out, rc = 0);
5495                 }
5496
5497                 if (jh == NULL) {
5498                         brelse(bh);
5499                         if (hlock != NULL)
5500                                 ldiskfs_htree_unlock(hlock);
5501                         else
5502                                 up_read(&obj->oo_ext_idx_sem);
5503                         dev->od_dirent_journal = 1;
5504
5505                         goto again;
5506                 }
5507
5508                 dirty = true;
5509                 if (unlikely(fid_is_sane(fid))) {
5510                         /* FID-in-dirent exists, but FID-in-LMA is lost.
5511                          * Trust the FID-in-dirent, and add FID-in-LMA. */
5512                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5513                         if (rc == 0)
5514                                 *attr |= LUDA_REPAIR;
5515                         else
5516                                 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
5517                                        "update dirent, dir = %lu/%u, "
5518                                        "name = %.*s, "DFID": rc = %d\n",
5519                                        devname, dir->i_ino, dir->i_generation,
5520                                        ent->oied_namelen, ent->oied_name,
5521                                        PFID(fid), rc);
5522                 } else {
5523                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
5524                         /* It is probably IGIF object. Only aappend the
5525                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
5526                         rc = osd_dirent_reinsert(env, jh, dentry, fid, bh, de,
5527                                                  hlock, dot_dotdot);
5528                         if (rc == 0)
5529                                 *attr |= LUDA_UPGRADE;
5530                         else
5531                                 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
5532                                        "after the dirent, dir = %lu/%u, "
5533                                        "name = %.*s, "DFID": rc = %d\n",
5534                                        devname, dir->i_ino, dir->i_generation,
5535                                        ent->oied_namelen, ent->oied_name,
5536                                        PFID(fid), rc);
5537                 }
5538         }
5539
5540         GOTO(out, rc);
5541
5542 out:
5543         brelse(bh);
5544         if (hlock != NULL) {
5545                 ldiskfs_htree_unlock(hlock);
5546         } else {
5547                 if (dev->od_dirent_journal != 0)
5548                         up_write(&obj->oo_ext_idx_sem);
5549                 else
5550                         up_read(&obj->oo_ext_idx_sem);
5551         }
5552
5553         if (jh != NULL)
5554                 ldiskfs_journal_stop(jh);
5555
5556 out_inode:
5557         iput(inode);
5558         if (rc >= 0 && !dirty)
5559                 dev->od_dirent_journal = 0;
5560
5561         return rc;
5562 }
5563
5564 /**
5565  * Returns the value at current position from iterator's in memory structure.
5566  *
5567  * \param di struct osd_it_ea, iterator's in memory structure
5568  * \param attr attr requested for dirent.
5569  * \param lde lustre dirent
5570  *
5571  * \retval   0 no error and \param lde has correct lustre dirent.
5572  * \retval -ve on error
5573  */
5574 static inline int osd_it_ea_rec(const struct lu_env *env,
5575                                 const struct dt_it *di,
5576                                 struct dt_rec *dtrec, __u32 attr)
5577 {
5578         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
5579         struct osd_object      *obj   = it->oie_obj;
5580         struct osd_device      *dev   = osd_obj2dev(obj);
5581         struct osd_thread_info *oti   = osd_oti_get(env);
5582         struct osd_inode_id    *id    = &oti->oti_id;
5583         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
5584         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
5585         __u32                   ino   = it->oie_dirent->oied_ino;
5586         int                     rc    = 0;
5587         ENTRY;
5588
5589         LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
5590
5591         if (attr & LUDA_VERIFY) {
5592                 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5593                         attr |= LUDA_IGNORE;
5594                         /* If the parent is on remote MDT, and there
5595                          * is no FID-in-dirent, then we have to get
5596                          * the parent FID from the linkEA.  */
5597                         if (!fid_is_sane(fid) &&
5598                             it->oie_dirent->oied_namelen == 2 &&
5599                             it->oie_dirent->oied_name[0] == '.' &&
5600                             it->oie_dirent->oied_name[1] == '.')
5601                                 osd_get_pfid_from_linkea(env, obj, fid);
5602                 } else {
5603                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
5604                                                      &attr);
5605                 }
5606
5607                 if (!fid_is_sane(fid)) {
5608                         attr &= ~LUDA_IGNORE;
5609                         attr |= LUDA_UNKNOWN;
5610                 }
5611         } else {
5612                 attr &= ~LU_DIRENT_ATTRS_MASK;
5613                 if (!fid_is_sane(fid)) {
5614                         bool is_dotdot = false;
5615                         if (it->oie_dirent->oied_namelen == 2 &&
5616                             it->oie_dirent->oied_name[0] == '.' &&
5617                             it->oie_dirent->oied_name[1] == '.')
5618                                 is_dotdot = true;
5619                         /* If the parent is on remote MDT, and there
5620                          * is no FID-in-dirent, then we have to get
5621                          * the parent FID from the linkEA.  */
5622                         if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
5623                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
5624                         } else {
5625                                 if (is_dotdot == false &&
5626                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
5627                                         RETURN(-ENOENT);
5628
5629                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5630                         }
5631                 } else {
5632                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5633                 }
5634         }
5635
5636         /* Pack the entry anyway, at least the offset is right. */
5637         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5638                            it->oie_dirent->oied_name,
5639                            it->oie_dirent->oied_namelen,
5640                            it->oie_dirent->oied_type, attr);
5641
5642         if (rc < 0)
5643                 RETURN(rc);
5644
5645         if (osd_remote_fid(env, dev, fid))
5646                 RETURN(0);
5647
5648         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
5649                 osd_add_oi_cache(oti, dev, id, fid);
5650
5651         RETURN(rc > 0 ? 0 : rc);
5652 }
5653
5654 /**
5655  * Returns the record size size at current position.
5656  *
5657  * This function will return record(lu_dirent) size in bytes.
5658  *
5659  * \param[in] env       execution environment
5660  * \param[in] di        iterator's in memory structure
5661  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
5662  *                      calculate the lu_dirent size.
5663  *
5664  * \retval      record size(in bytes & in memory) of the current lu_dirent
5665  *              entry.
5666  */
5667 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5668                               __u32 attr)
5669 {
5670         struct osd_it_ea *it = (struct osd_it_ea *)di;
5671
5672         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5673 }
5674
5675 /**
5676  * Returns a cookie for current position of the iterator head, so that
5677  * user can use this cookie to load/start the iterator next time.
5678  *
5679  * \param di iterator's in memory structure
5680  *
5681  * \retval cookie for current position, on success
5682  */
5683 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5684 {
5685         struct osd_it_ea *it = (struct osd_it_ea *)di;
5686
5687         return it->oie_dirent->oied_off;
5688 }
5689
5690 /**
5691  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5692  * to load a directory entry at a time and stored it i inn,
5693  * in iterator's in-memory data structure.
5694  *
5695  * \param di struct osd_it_ea, iterator's in memory structure
5696  *
5697  * \retval +ve on success
5698  * \retval -ve on error
5699  */
5700 static int osd_it_ea_load(const struct lu_env *env,
5701                           const struct dt_it *di, __u64 hash)
5702 {
5703         struct osd_it_ea *it = (struct osd_it_ea *)di;
5704         int rc;
5705
5706         ENTRY;
5707         it->oie_file.f_pos = hash;
5708
5709         rc =  osd_ldiskfs_it_fill(env, di);
5710         if (rc > 0)
5711                 rc = -ENODATA;
5712
5713         if (rc == 0)
5714                 rc = +1;
5715
5716         RETURN(rc);
5717 }
5718
5719 /**
5720  * Index lookup function for interoperability mode (b11826).
5721  *
5722  * \param key,  key i.e. file name to be searched
5723  *
5724  * \retval +ve, on success
5725  * \retval -ve, on error
5726  */
5727 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5728                                struct dt_rec *rec, const struct dt_key *key)
5729 {
5730         struct osd_object *obj = osd_dt_obj(dt);
5731         int rc = 0;
5732
5733         ENTRY;
5734
5735         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5736         LINVRNT(osd_invariant(obj));
5737
5738         rc = osd_ea_lookup_rec(env, obj, rec, key);
5739         if (rc == 0)
5740                 rc = +1;
5741         RETURN(rc);
5742 }
5743
5744 /**
5745  * Index and Iterator operations for interoperability
5746  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5747  */
5748 static const struct dt_index_operations osd_index_ea_ops = {
5749         .dio_lookup         = osd_index_ea_lookup,
5750         .dio_declare_insert = osd_index_declare_ea_insert,
5751         .dio_insert         = osd_index_ea_insert,
5752         .dio_declare_delete = osd_index_declare_ea_delete,
5753         .dio_delete         = osd_index_ea_delete,
5754         .dio_it     = {
5755                 .init     = osd_it_ea_init,
5756                 .fini     = osd_it_ea_fini,
5757                 .get      = osd_it_ea_get,
5758                 .put      = osd_it_ea_put,
5759                 .next     = osd_it_ea_next,
5760                 .key      = osd_it_ea_key,
5761                 .key_size = osd_it_ea_key_size,
5762                 .rec      = osd_it_ea_rec,
5763                 .rec_size = osd_it_ea_rec_size,
5764                 .store    = osd_it_ea_store,
5765                 .load     = osd_it_ea_load
5766         }
5767 };
5768
5769 static void *osd_key_init(const struct lu_context *ctx,
5770                           struct lu_context_key *key)
5771 {
5772         struct osd_thread_info *info;
5773
5774         OBD_ALLOC_PTR(info);
5775         if (info == NULL)
5776                 return ERR_PTR(-ENOMEM);
5777
5778         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5779         if (info->oti_it_ea_buf == NULL)
5780                 goto out_free_info;
5781
5782         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5783
5784         info->oti_hlock = ldiskfs_htree_lock_alloc();
5785         if (info->oti_hlock == NULL)
5786                 goto out_free_ea;
5787
5788         return info;
5789
5790  out_free_ea:
5791         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5792  out_free_info:
5793         OBD_FREE_PTR(info);
5794         return ERR_PTR(-ENOMEM);
5795 }
5796
5797 static void osd_key_fini(const struct lu_context *ctx,
5798                          struct lu_context_key *key, void* data)
5799 {
5800         struct osd_thread_info *info = data;
5801
5802         if (info->oti_inode != NULL)
5803                 OBD_FREE_PTR(info->oti_inode);
5804         if (info->oti_hlock != NULL)
5805                 ldiskfs_htree_lock_free(info->oti_hlock);
5806         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5807         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5808         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5809         lu_buf_free(&info->oti_big_buf);
5810         OBD_FREE_PTR(info);
5811 }
5812
5813 static void osd_key_exit(const struct lu_context *ctx,
5814                          struct lu_context_key *key, void *data)
5815 {
5816         struct osd_thread_info *info = data;
5817
5818         LASSERT(info->oti_r_locks == 0);
5819         LASSERT(info->oti_w_locks == 0);
5820         LASSERT(info->oti_txns    == 0);
5821 }
5822
5823 /* type constructor/destructor: osd_type_init, osd_type_fini */
5824 LU_TYPE_INIT_FINI(osd, &osd_key);
5825
5826 struct lu_context_key osd_key = {
5827         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5828         .lct_init = osd_key_init,
5829         .lct_fini = osd_key_fini,
5830         .lct_exit = osd_key_exit
5831 };
5832
5833
5834 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5835                            const char *name, struct lu_device *next)
5836 {
5837         struct osd_device *osd = osd_dev(d);
5838
5839         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5840             >= sizeof(osd->od_svname))
5841                 return -E2BIG;
5842         return osd_procfs_init(osd, name);
5843 }
5844
5845 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5846 {
5847         struct seq_server_site  *ss = osd_seq_site(osd);
5848         int                     rc;
5849         ENTRY;
5850
5851         if (osd->od_is_ost || osd->od_cl_seq != NULL)
5852                 RETURN(0);
5853
5854         if (unlikely(ss == NULL))
5855                 RETURN(-ENODEV);
5856
5857         OBD_ALLOC_PTR(osd->od_cl_seq);
5858         if (osd->od_cl_seq == NULL)
5859                 RETURN(-ENOMEM);
5860
5861         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5862                              osd->od_svname, ss->ss_server_seq);
5863         if (rc != 0) {
5864                 OBD_FREE_PTR(osd->od_cl_seq);
5865                 osd->od_cl_seq = NULL;
5866                 RETURN(rc);
5867         }
5868
5869         if (ss->ss_node_id == 0) {
5870                 /* If the OSD on the sequence controller(MDT0), then allocate
5871                  * sequence here, otherwise allocate sequence after connected
5872                  * to MDT0 (see mdt_register_lwp_callback()). */
5873                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
5874                                    &osd->od_cl_seq->lcs_space, env);
5875         }
5876
5877         RETURN(rc);
5878 }
5879
5880 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5881 {
5882         if (osd->od_cl_seq == NULL)
5883                 return;
5884
5885         seq_client_fini(osd->od_cl_seq);
5886         OBD_FREE_PTR(osd->od_cl_seq);
5887         osd->od_cl_seq = NULL;
5888 }
5889
5890 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5891 {
5892         ENTRY;
5893
5894         /* shutdown quota slave instance associated with the device */
5895         if (o->od_quota_slave != NULL) {
5896                 qsd_fini(env, o->od_quota_slave);
5897                 o->od_quota_slave = NULL;
5898         }
5899
5900         osd_fid_fini(env, o);
5901
5902         RETURN(0);
5903 }
5904
5905 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5906 {
5907         ENTRY;
5908
5909         if (o->od_mnt != NULL) {
5910                 shrink_dcache_sb(osd_sb(o));
5911                 osd_sync(env, &o->od_dt_dev);
5912
5913                 mntput(o->od_mnt);
5914                 o->od_mnt = NULL;
5915         }
5916
5917         EXIT;
5918 }
5919
5920 static int osd_mount(const struct lu_env *env,
5921                      struct osd_device *o, struct lustre_cfg *cfg)
5922 {
5923         const char              *name  = lustre_cfg_string(cfg, 0);
5924         const char              *dev  = lustre_cfg_string(cfg, 1);
5925         const char              *opts;
5926         unsigned long            page, s_flags, lmd_flags = 0;
5927         struct page             *__page;
5928         struct file_system_type *type;
5929         char                    *options = NULL;
5930         char                    *str;
5931         struct osd_thread_info  *info = osd_oti_get(env);
5932         struct lu_fid           *fid = &info->oti_fid;
5933         struct inode            *inode;
5934         int                      rc = 0, force_over_128tb = 0;
5935         ENTRY;
5936
5937         if (o->od_mnt != NULL)
5938                 RETURN(0);
5939
5940         if (strlen(dev) >= sizeof(o->od_mntdev))
5941                 RETURN(-E2BIG);
5942         strcpy(o->od_mntdev, dev);
5943
5944         str = lustre_cfg_string(cfg, 2);
5945         s_flags = simple_strtoul(str, NULL, 0);
5946         str = strstr(str, ":");
5947         if (str)
5948                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5949         opts = lustre_cfg_string(cfg, 3);
5950 #ifdef __BIG_ENDIAN
5951         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
5952                 CERROR("%s: device %s extents feature is not guaranteed to "
5953                        "work on big-endian systems. Use \"bigendian_extents\" "
5954                        "mount option to override.\n", name, dev);
5955                 RETURN(-EINVAL);
5956         }
5957 #endif
5958         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL)
5959                 force_over_128tb = 1;
5960
5961         __page = alloc_page(GFP_IOFS);
5962         if (__page == NULL)
5963                 GOTO(out, rc = -ENOMEM);
5964         page = (unsigned long)page_address(__page);
5965         options = (char *)page;
5966         *options = '\0';
5967         if (opts != NULL) {
5968                 /* strip out the options for back compatiblity */
5969                 static char *sout[] = {
5970                         "mballoc",
5971                         "iopen",
5972                         "noiopen",
5973                         "iopen_nopriv",
5974                         "extents",
5975                         "noextents",
5976                         /* strip out option we processed in osd */
5977                         "bigendian_extents",
5978                         "force_over_128tb",
5979                         NULL
5980                 };
5981                 strcat(options, opts);
5982                 for (rc = 0, str = options; sout[rc]; ) {
5983                         char *op = strstr(str, sout[rc]);
5984                         if (op == NULL) {
5985                                 rc++;
5986                                 str = options;
5987                                 continue;
5988                         }
5989                         if (op == options || *(op - 1) == ',') {
5990                                 str = op + strlen(sout[rc]);
5991                                 if (*str == ',' || *str == '\0') {
5992                                         *str == ',' ? str++ : str;
5993                                         memmove(op, str, strlen(str) + 1);
5994                                 }
5995                         }
5996                         for (str = op; *str != ',' && *str != '\0'; str++)
5997                                 ;
5998                 }
5999         } else {
6000                 strncat(options, "user_xattr,acl", 14);
6001         }
6002
6003         /* Glom up mount options */
6004         if (*options != '\0')
6005                 strcat(options, ",");
6006         strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
6007
6008         type = get_fs_type("ldiskfs");
6009         if (!type) {
6010                 CERROR("%s: cannot find ldiskfs module\n", name);
6011                 GOTO(out, rc = -ENODEV);
6012         }
6013
6014         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
6015         module_put(type->owner);
6016
6017         if (IS_ERR(o->od_mnt)) {
6018                 rc = PTR_ERR(o->od_mnt);
6019                 o->od_mnt = NULL;
6020                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
6021                 GOTO(out, rc);
6022         }
6023
6024         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) > (8ULL << 32) &&
6025             force_over_128tb == 0) {
6026                 CERROR("%s: device %s LDISKFS does not support filesystems "
6027                        "greater than 128TB and can cause data corruption. "
6028                        "Use \"force_over_128tb\" mount option to override.\n",
6029                        name, dev);
6030                 GOTO(out, rc = -EINVAL);
6031         }
6032
6033 #ifdef HAVE_DEV_SET_RDONLY
6034         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
6035                 CERROR("%s: underlying device %s is marked as read-only. "
6036                        "Setup failed\n", name, dev);
6037                 GOTO(out_mnt, rc = -EROFS);
6038         }
6039 #endif
6040
6041         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
6042                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
6043                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
6044                 GOTO(out_mnt, rc = -EINVAL);
6045         }
6046
6047 #ifdef LDISKFS_MOUNT_DIRDATA
6048         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
6049                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6050                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
6051         else if (!o->od_is_ost)
6052                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
6053                       "enabling the dirdata feature. If you do not want to "
6054                       "downgrade to Lustre-1.x again, you can enable it via "
6055                       "'tune2fs -O dirdata device'\n", name, dev);
6056 #endif
6057         inode = osd_sb(o)->s_root->d_inode;
6058         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
6059         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
6060         if (rc != 0) {
6061                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
6062                 GOTO(out_mnt, rc);
6063         }
6064
6065         if (lmd_flags & LMD_FLG_NOSCRUB)
6066                 o->od_noscrub = 1;
6067
6068         GOTO(out, rc = 0);
6069
6070 out_mnt:
6071         mntput(o->od_mnt);
6072         o->od_mnt = NULL;
6073
6074 out:
6075         if (__page)
6076                 __free_page(__page);
6077
6078         return rc;
6079 }
6080
6081 static struct lu_device *osd_device_fini(const struct lu_env *env,
6082                                          struct lu_device *d)
6083 {
6084         struct osd_device *o = osd_dev(d);
6085         ENTRY;
6086
6087         osd_shutdown(env, o);
6088         osd_procfs_fini(o);
6089         osd_scrub_cleanup(env, o);
6090         osd_obj_map_fini(o);
6091         osd_umount(env, o);
6092
6093         RETURN(NULL);
6094 }
6095
6096 static int osd_device_init0(const struct lu_env *env,
6097                             struct osd_device *o,
6098                             struct lustre_cfg *cfg)
6099 {
6100         struct lu_device        *l = osd2lu_dev(o);
6101         struct osd_thread_info *info;
6102         int                     rc;
6103         int                     cplen = 0;
6104
6105         /* if the module was re-loaded, env can loose its keys */
6106         rc = lu_env_refill((struct lu_env *) env);
6107         if (rc)
6108                 GOTO(out, rc);
6109         info = osd_oti_get(env);
6110         LASSERT(info);
6111
6112         l->ld_ops = &osd_lu_ops;
6113         o->od_dt_dev.dd_ops = &osd_dt_ops;
6114
6115         spin_lock_init(&o->od_osfs_lock);
6116         mutex_init(&o->od_otable_mutex);
6117         INIT_LIST_HEAD(&o->od_orphan_list);
6118
6119         o->od_read_cache = 1;
6120         o->od_writethrough_cache = 1;
6121         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
6122
6123         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
6124                         sizeof(o->od_svname));
6125         if (cplen >= sizeof(o->od_svname)) {
6126                 rc = -E2BIG;
6127                 GOTO(out, rc);
6128         }
6129
6130         o->od_index = -1; /* -1 means index is invalid */
6131         rc = server_name2index(o->od_svname, &o->od_index, NULL);
6132         if (rc == LDD_F_SV_TYPE_OST)
6133                 o->od_is_ost = 1;
6134
6135         o->od_full_scrub_ratio = OFSR_DEFAULT;
6136         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
6137         rc = osd_mount(env, o, cfg);
6138         if (rc != 0)
6139                 GOTO(out, rc);
6140
6141         rc = osd_obj_map_init(env, o);
6142         if (rc != 0)
6143                 GOTO(out_mnt, rc);
6144
6145         rc = lu_site_init(&o->od_site, l);
6146         if (rc != 0)
6147                 GOTO(out_compat, rc);
6148         o->od_site.ls_bottom_dev = l;
6149
6150         rc = lu_site_init_finish(&o->od_site);
6151         if (rc != 0)
6152                 GOTO(out_site, rc);
6153
6154         INIT_LIST_HEAD(&o->od_ios_list);
6155         /* setup scrub, including OI files initialization */
6156         rc = osd_scrub_setup(env, o);
6157         if (rc < 0)
6158                 GOTO(out_site, rc);
6159
6160         rc = osd_procfs_init(o, o->od_svname);
6161         if (rc != 0) {
6162                 CERROR("%s: can't initialize procfs: rc = %d\n",
6163                        o->od_svname, rc);
6164                 GOTO(out_scrub, rc);
6165         }
6166
6167         LASSERT(l->ld_site->ls_linkage.next != NULL);
6168         LASSERT(l->ld_site->ls_linkage.prev != NULL);
6169
6170         /* initialize quota slave instance */
6171         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
6172                                      o->od_proc_entry);
6173         if (IS_ERR(o->od_quota_slave)) {
6174                 rc = PTR_ERR(o->od_quota_slave);
6175                 o->od_quota_slave = NULL;
6176                 GOTO(out_procfs, rc);
6177         }
6178
6179         RETURN(0);
6180
6181 out_procfs:
6182         osd_procfs_fini(o);
6183 out_scrub:
6184         osd_scrub_cleanup(env, o);
6185 out_site:
6186         lu_site_fini(&o->od_site);
6187 out_compat:
6188         osd_obj_map_fini(o);
6189 out_mnt:
6190         osd_umount(env, o);
6191 out:
6192         return rc;
6193 }
6194
6195 static struct lu_device *osd_device_alloc(const struct lu_env *env,
6196                                           struct lu_device_type *t,
6197                                           struct lustre_cfg *cfg)
6198 {
6199         struct osd_device *o;
6200         int                rc;
6201
6202         OBD_ALLOC_PTR(o);
6203         if (o == NULL)
6204                 return ERR_PTR(-ENOMEM);
6205
6206         rc = dt_device_init(&o->od_dt_dev, t);
6207         if (rc == 0) {
6208                 /* Because the ctx might be revived in dt_device_init,
6209                  * refill the env here */
6210                 lu_env_refill((struct lu_env *)env);
6211                 rc = osd_device_init0(env, o, cfg);
6212                 if (rc)
6213                         dt_device_fini(&o->od_dt_dev);
6214         }
6215
6216         if (unlikely(rc != 0))
6217                 OBD_FREE_PTR(o);
6218
6219         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
6220 }
6221
6222 static struct lu_device *osd_device_free(const struct lu_env *env,
6223                                          struct lu_device *d)
6224 {
6225         struct osd_device *o = osd_dev(d);
6226         ENTRY;
6227
6228         /* XXX: make osd top device in order to release reference */
6229         d->ld_site->ls_top_dev = d;
6230         lu_site_purge(env, d->ld_site, -1);
6231         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
6232                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
6233                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
6234         }
6235         lu_site_fini(&o->od_site);
6236         dt_device_fini(&o->od_dt_dev);
6237         OBD_FREE_PTR(o);
6238         RETURN(NULL);
6239 }
6240
6241 static int osd_process_config(const struct lu_env *env,
6242                               struct lu_device *d, struct lustre_cfg *cfg)
6243 {
6244         struct osd_device               *o = osd_dev(d);
6245         int                             rc;
6246         ENTRY;
6247
6248         switch (cfg->lcfg_command) {
6249         case LCFG_SETUP:
6250                 rc = osd_mount(env, o, cfg);
6251                 break;
6252         case LCFG_CLEANUP:
6253                 lu_dev_del_linkage(d->ld_site, d);
6254                 rc = osd_shutdown(env, o);
6255                 break;
6256         case LCFG_PARAM:
6257                 LASSERT(&o->od_dt_dev);
6258                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
6259                                               cfg, &o->od_dt_dev);
6260                 if (rc > 0 || rc == -ENOSYS)
6261                         rc = class_process_proc_param(PARAM_OST,
6262                                                       lprocfs_osd_obd_vars,
6263                                                       cfg, &o->od_dt_dev);
6264                 break;
6265         default:
6266                 rc = -ENOSYS;
6267         }
6268
6269         RETURN(rc);
6270 }
6271
6272 static int osd_recovery_complete(const struct lu_env *env,
6273                                  struct lu_device *d)
6274 {
6275         struct osd_device       *osd = osd_dev(d);
6276         int                      rc = 0;
6277         ENTRY;
6278
6279         if (osd->od_quota_slave == NULL)
6280                 RETURN(0);
6281
6282         /* start qsd instance on recovery completion, this notifies the quota
6283          * slave code that we are about to process new requests now */
6284         rc = qsd_start(env, osd->od_quota_slave);
6285         RETURN(rc);
6286 }
6287
6288 /*
6289  * we use exports to track all osd users
6290  */
6291 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6292                            struct obd_device *obd, struct obd_uuid *cluuid,
6293                            struct obd_connect_data *data, void *localdata)
6294 {
6295         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
6296         struct lustre_handle  conn;
6297         int                   rc;
6298         ENTRY;
6299
6300         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6301
6302         rc = class_connect(&conn, obd, cluuid);
6303         if (rc)
6304                 RETURN(rc);
6305
6306         *exp = class_conn2export(&conn);
6307
6308         spin_lock(&osd->od_osfs_lock);
6309         osd->od_connects++;
6310         spin_unlock(&osd->od_osfs_lock);
6311
6312         RETURN(0);
6313 }
6314
6315 /*
6316  * once last export (we don't count self-export) disappeared
6317  * osd can be released
6318  */
6319 static int osd_obd_disconnect(struct obd_export *exp)
6320 {
6321         struct obd_device *obd = exp->exp_obd;
6322         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6323         int                rc, release = 0;
6324         ENTRY;
6325
6326         /* Only disconnect the underlying layers on the final disconnect. */
6327         spin_lock(&osd->od_osfs_lock);
6328         osd->od_connects--;
6329         if (osd->od_connects == 0)
6330                 release = 1;
6331         spin_unlock(&osd->od_osfs_lock);
6332
6333         rc = class_disconnect(exp); /* bz 9811 */
6334
6335         if (rc == 0 && release)
6336                 class_manual_cleanup(obd);
6337         RETURN(rc);
6338 }
6339
6340 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6341                        struct lu_device *dev)
6342 {
6343         struct osd_device       *osd    = osd_dev(dev);
6344         struct lr_server_data   *lsd    =
6345                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
6346         int                      result = 0;
6347         ENTRY;
6348
6349         if (osd->od_quota_slave != NULL) {
6350                 /* set up quota slave objects */
6351                 result = qsd_prepare(env, osd->od_quota_slave);
6352                 if (result != 0)
6353                         RETURN(result);
6354         }
6355
6356         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
6357 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
6358                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
6359                         osd->od_index_in_idif = 1;
6360                 } else {
6361                         osd->od_index_in_idif = 0;
6362                         result = osd_register_proc_index_in_idif(osd);
6363                         if (result != 0)
6364                                 RETURN(result);
6365                 }
6366 #else
6367                 osd->od_index_in_idif = 1;
6368 #endif
6369         }
6370
6371         result = osd_fid_init(env, osd);
6372
6373         RETURN(result);
6374 }
6375
6376 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6377                          struct lu_fid *fid, struct md_op_data *op_data)
6378 {
6379         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6380
6381         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6382 }
6383
6384 static const struct lu_object_operations osd_lu_obj_ops = {
6385         .loo_object_init      = osd_object_init,
6386         .loo_object_delete    = osd_object_delete,
6387         .loo_object_release   = osd_object_release,
6388         .loo_object_free      = osd_object_free,
6389         .loo_object_print     = osd_object_print,
6390         .loo_object_invariant = osd_object_invariant
6391 };
6392
6393 const struct lu_device_operations osd_lu_ops = {
6394         .ldo_object_alloc      = osd_object_alloc,
6395         .ldo_process_config    = osd_process_config,
6396         .ldo_recovery_complete = osd_recovery_complete,
6397         .ldo_prepare           = osd_prepare,
6398 };
6399
6400 static const struct lu_device_type_operations osd_device_type_ops = {
6401         .ldto_init = osd_type_init,
6402         .ldto_fini = osd_type_fini,
6403
6404         .ldto_start = osd_type_start,
6405         .ldto_stop  = osd_type_stop,
6406
6407         .ldto_device_alloc = osd_device_alloc,
6408         .ldto_device_free  = osd_device_free,
6409
6410         .ldto_device_init    = osd_device_init,
6411         .ldto_device_fini    = osd_device_fini
6412 };
6413
6414 static struct lu_device_type osd_device_type = {
6415         .ldt_tags     = LU_DEVICE_DT,
6416         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
6417         .ldt_ops      = &osd_device_type_ops,
6418         .ldt_ctx_tags = LCT_LOCAL,
6419 };
6420
6421 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
6422 {
6423         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6424         struct super_block *sb = osd_sb(osd);
6425
6426         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
6427 }
6428
6429 /*
6430  * lprocfs legacy support.
6431  */
6432 static struct obd_ops osd_obd_device_ops = {
6433         .o_owner = THIS_MODULE,
6434         .o_connect      = osd_obd_connect,
6435         .o_disconnect   = osd_obd_disconnect,
6436         .o_fid_alloc    = osd_fid_alloc,
6437         .o_health_check = osd_health_check,
6438 };
6439
6440 static int __init osd_mod_init(void)
6441 {
6442         int rc;
6443
6444         LASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
6445 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
6446         /* please, try to keep osd_thread_info smaller than a page */
6447         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
6448 #endif
6449
6450         osd_oi_mod_init();
6451
6452         rc = lu_kmem_init(ldiskfs_caches);
6453         if (rc)
6454                 return rc;
6455
6456         rc = class_register_type(&osd_obd_device_ops, NULL, true,
6457                                  lprocfs_osd_module_vars,
6458                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6459         if (rc)
6460                 lu_kmem_fini(ldiskfs_caches);
6461         return rc;
6462 }
6463
6464 static void __exit osd_mod_exit(void)
6465 {
6466         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6467         lu_kmem_fini(ldiskfs_caches);
6468 }
6469
6470 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6471 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6472 MODULE_VERSION(LUSTRE_VERSION_STRING);
6473 MODULE_LICENSE("GPL");
6474
6475 module_init(osd_mod_init);
6476 module_exit(osd_mod_exit);