Whamcloud - gitweb
LU-6846 osd: reset do_body_ops after creation
[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, 2014, 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 int osd_trans_declare_op2rb[] = {
126         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
127         [OSD_OT_PUNCH]          = OSD_OT_MAX,
128         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
129         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
130         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
131         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
132         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
133         [OSD_OT_WRITE]          = OSD_OT_WRITE,
134         [OSD_OT_INSERT]         = OSD_OT_DELETE,
135         [OSD_OT_DELETE]         = OSD_OT_INSERT,
136         [OSD_OT_QUOTA]          = OSD_OT_MAX,
137 };
138
139 static int osd_has_index(const struct osd_object *obj)
140 {
141         return obj->oo_dt.do_index_ops != NULL;
142 }
143
144 static int osd_object_invariant(const struct lu_object *l)
145 {
146         return osd_invariant(osd_obj(l));
147 }
148
149 /*
150  * Concurrency: doesn't matter
151  */
152
153 /*
154  * Concurrency: doesn't matter
155  */
156 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
157 {
158         struct osd_thread_info *oti = osd_oti_get(env);
159         return oti->oti_w_locks > 0 && o->oo_owner == env;
160 }
161
162 /*
163  * Concurrency: doesn't access mutable data
164  */
165 static int osd_root_get(const struct lu_env *env,
166                         struct dt_device *dev, struct lu_fid *f)
167 {
168         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
169         return 0;
170 }
171
172 /*
173  * OSD object methods.
174  */
175
176 /*
177  * Concurrency: no concurrent access is possible that early in object
178  * life-cycle.
179  */
180 static struct lu_object *osd_object_alloc(const struct lu_env *env,
181                                           const struct lu_object_header *hdr,
182                                           struct lu_device *d)
183 {
184         struct osd_object *mo;
185
186         OBD_ALLOC_PTR(mo);
187         if (mo != NULL) {
188                 struct lu_object *l;
189
190                 l = &mo->oo_dt.do_lu;
191                 dt_object_init(&mo->oo_dt, NULL, d);
192                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
193                 l->lo_ops = &osd_lu_obj_ops;
194                 init_rwsem(&mo->oo_sem);
195                 init_rwsem(&mo->oo_ext_idx_sem);
196                 spin_lock_init(&mo->oo_guard);
197                 return l;
198         } else {
199                 return NULL;
200         }
201 }
202
203 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
204                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
205 {
206         int rc;
207
208         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
209         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
210                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
211         if (rc > 0) {
212                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
213                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
214                 rc = 0;
215                 lustre_lma_swab(lma);
216                 /* Check LMA compatibility */
217                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
218                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
219                               "for fid = "DFID", ino = %lu\n",
220                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
221                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
222                               PFID(&lma->lma_self_fid), inode->i_ino);
223                         rc = -EOPNOTSUPP;
224                 }
225         } else if (rc == 0) {
226                 rc = -ENODATA;
227         }
228
229         return rc;
230 }
231
232 /*
233  * retrieve object from backend ext fs.
234  **/
235 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
236                        struct osd_inode_id *id)
237 {
238         struct inode *inode = NULL;
239
240         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
241         if (IS_ERR(inode)) {
242                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
243                        id->oii_ino, PTR_ERR(inode));
244         } else if (id->oii_gen != OSD_OII_NOGEN &&
245                    inode->i_generation != id->oii_gen) {
246                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
247                        "i_generation = %u\n",
248                        id->oii_ino, id->oii_gen, inode->i_generation);
249                 iput(inode);
250                 inode = ERR_PTR(-ESTALE);
251         } else if (inode->i_nlink == 0) {
252                 /* due to parallel readdir and unlink,
253                 * we can have dead inode here. */
254                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
255                 iput(inode);
256                 inode = ERR_PTR(-ESTALE);
257         } else if (is_bad_inode(inode)) {
258                 CWARN("%.16s: bad inode: ino = %u\n",
259                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
260                 iput(inode);
261                 inode = ERR_PTR(-ENOENT);
262         } else {
263                 if (id->oii_gen == OSD_OII_NOGEN)
264                         osd_id_gen(id, inode->i_ino, inode->i_generation);
265
266                 /* Do not update file c/mtime in ldiskfs.
267                  * NB: we don't have any lock to protect this because we don't
268                  * have reference on osd_object now, but contention with
269                  * another lookup + attr_set can't happen in the tiny window
270                  * between if (...) and set S_NOCMTIME. */
271                 if (!(inode->i_flags & S_NOCMTIME))
272                         inode->i_flags |= S_NOCMTIME;
273         }
274         return inode;
275 }
276
277 static struct inode *
278 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
279              struct osd_inode_id *id, struct lu_fid *fid)
280 {
281         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
282         struct inode            *inode;
283         int                      rc;
284
285         inode = osd_iget(info, dev, id);
286         if (IS_ERR(inode))
287                 return inode;
288
289         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
290         if (rc == 0) {
291                 *fid = lma->lma_self_fid;
292         } else if (rc == -ENODATA) {
293                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
294                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
295                 else
296                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
297         } else {
298                 iput(inode);
299                 inode = ERR_PTR(rc);
300         }
301         return inode;
302 }
303
304 /**
305  * \retval +v: new filter_fid, does not contain self-fid
306  * \retval 0:  filter_fid_old, contains self-fid
307  * \retval -v: other failure cases
308  */
309 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
310                  struct dentry *dentry, struct lu_fid *fid)
311 {
312         struct filter_fid_old   *ff     = &info->oti_ff;
313         struct ost_id           *ostid  = &info->oti_ostid;
314         int                      rc;
315
316         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
317         if (rc == sizeof(*ff)) {
318                 rc = 0;
319                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
320                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
321                 /* XXX: should use real OST index in the future. LU-3569 */
322                 ostid_to_fid(fid, ostid, 0);
323         } else if (rc == sizeof(struct filter_fid)) {
324                 rc = 1;
325         } else if (rc >= 0) {
326                 rc = -EINVAL;
327         }
328
329         return rc;
330 }
331
332 static int osd_lma_self_repair(struct osd_thread_info *info,
333                                struct osd_device *osd, struct inode *inode,
334                                const struct lu_fid *fid, __u32 compat)
335 {
336         handle_t *jh;
337         int       rc;
338
339         LASSERT(current->journal_info == NULL);
340
341         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
342                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
343         if (IS_ERR(jh)) {
344                 rc = PTR_ERR(jh);
345                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
346                       osd_name(osd), rc);
347                 return rc;
348         }
349
350         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
351         if (rc != 0)
352                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
353                       osd_name(osd), rc);
354         ldiskfs_journal_stop(jh);
355         return rc;
356 }
357
358 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
359 {
360         struct osd_thread_info  *info   = osd_oti_get(env);
361         struct osd_device       *osd    = osd_obj2dev(obj);
362         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
363         struct inode            *inode  = obj->oo_inode;
364         struct dentry           *dentry = &info->oti_obj_dentry;
365         struct lu_fid           *fid    = NULL;
366         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
367         int                      rc;
368         ENTRY;
369
370         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
371         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
372                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
373         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
374                 fid = &lma->lma_self_fid;
375                 rc = osd_get_idif(info, inode, dentry, fid);
376                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
377                         /* For the given OST-object, if it has neither LMA nor
378                          * FID in XATTR_NAME_FID, then the given FID (which is
379                          * contained in the @obj, from client RPC for locating
380                          * the OST-object) is trusted. We use it to generate
381                          * the LMA. */
382                         osd_lma_self_repair(info, osd, inode, rfid,
383                                             LMAC_FID_ON_OST);
384                         RETURN(0);
385                 }
386         }
387
388         if (unlikely(rc == -ENODATA))
389                 RETURN(0);
390
391         if (rc < 0)
392                 RETURN(rc);
393
394         if (rc > 0) {
395                 rc = 0;
396                 lustre_lma_swab(lma);
397                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
398                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
399                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
400                               "fid = "DFID", ino = %lu\n", osd_name(osd),
401                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
402                               PFID(rfid), inode->i_ino);
403                         rc = -EOPNOTSUPP;
404                 } else {
405                         fid = &lma->lma_self_fid;
406                 }
407         }
408
409         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
410                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
411                         struct ost_id   *oi   = &info->oti_ostid;
412                         struct lu_fid   *fid1 = &info->oti_fid3;
413                         __u32            idx  = fid_idif_ost_idx(rfid);
414
415                         /* For old IDIF, the OST index is not part of the IDIF,
416                          * Means that different OSTs may have the same IDIFs.
417                          * Under such case, we need to make some compatible
418                          * check to make sure to trigger OI scrub properly. */
419                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
420                                 /* Given @rfid is new, LMA is old. */
421                                 fid_to_ostid(fid, oi);
422                                 ostid_to_fid(fid1, oi, idx);
423                                 if (lu_fid_eq(fid1, rfid)) {
424                                         if (osd->od_index_in_idif)
425                                                 osd_lma_self_repair(info, osd,
426                                                         inode, rfid,
427                                                         LMAC_FID_ON_OST);
428                                         RETURN(0);
429                                 }
430                         }
431                 }
432
433                 rc = -EREMCHG;
434         }
435
436         RETURN(rc);
437 }
438
439 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
440                           const struct lu_fid *fid,
441                           const struct lu_object_conf *conf)
442 {
443         struct osd_thread_info *info;
444         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
445         struct osd_device      *dev;
446         struct osd_idmap_cache *oic;
447         struct osd_inode_id    *id;
448         struct inode           *inode;
449         struct osd_scrub       *scrub;
450         struct scrub_file      *sf;
451         int                     result;
452         int                     saved  = 0;
453         bool                    in_oi  = false;
454         bool                    in_cache = false;
455         bool                    triggered = false;
456         ENTRY;
457
458         LINVRNT(osd_invariant(obj));
459         LASSERT(obj->oo_inode == NULL);
460         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
461
462         dev = osd_dev(ldev);
463         scrub = &dev->od_scrub;
464         sf = &scrub->os_file;
465         info = osd_oti_get(env);
466         LASSERT(info);
467         oic = &info->oti_cache;
468
469         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
470                 RETURN(-ENOENT);
471
472         /* For the object is created as locking anchor, or for the object to
473          * be created on disk. No need to osd_oi_lookup() at here because FID
474          * shouldn't never be re-used, if it's really a duplicate FID from
475          * unexpected reason, we should be able to detect it later by calling
476          * do_create->osd_oi_insert(). */
477         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
478                 GOTO(out, result = 0);
479
480         /* Search order: 1. per-thread cache. */
481         if (lu_fid_eq(fid, &oic->oic_fid) &&
482             likely(oic->oic_dev == dev)) {
483                 id = &oic->oic_lid;
484                 in_cache = true;
485                 goto iget;
486         }
487
488         id = &info->oti_id;
489         if (!list_empty(&scrub->os_inconsistent_items)) {
490                 /* Search order: 2. OI scrub pending list. */
491                 result = osd_oii_lookup(dev, fid, id);
492                 if (result == 0)
493                         goto iget;
494         }
495
496         /* Search order: 3. OI files. */
497         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
498         if (result == -ENOENT) {
499                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
500                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
501                                       sf->sf_oi_bitmap))
502                         GOTO(out, result = 0);
503
504                 goto trigger;
505         }
506
507         if (result != 0)
508                 GOTO(out, result);
509
510         in_oi = true;
511
512 iget:
513         inode = osd_iget(info, dev, id);
514         if (IS_ERR(inode)) {
515                 result = PTR_ERR(inode);
516                 if (result != -ENOENT && result != -ESTALE)
517                         GOTO(out, result);
518
519                 if (in_cache)
520                         fid_zero(&oic->oic_fid);
521
522                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
523                 if (result != 0)
524                         GOTO(out, result = (result == -ENOENT ? 0 : result));
525
526                 /* The OI mapping is there, but the inode is NOT there.
527                  * Two possible cases for that:
528                  *
529                  * 1) Backup/restore caused the OI invalid.
530                  * 2) Someone unlinked the object but NOT removed
531                  *    the OI mapping, such as mount target device
532                  *    as ldiskfs, and modify something directly.
533                  *
534                  * Generally, when the device is mounted, it will
535                  * auto check whether the system is restored from
536                  * file-level backup or not. We trust such detect
537                  * to distinguish the 1st case from the 2nd case. */
538                 if (!(scrub->os_file.sf_flags & SF_INCONSISTENT))
539                         GOTO(out, result = 0);
540
541 trigger:
542                 if (unlikely(triggered))
543                         GOTO(out, result = saved);
544
545                 triggered = true;
546                 if (thread_is_running(&scrub->os_thread)) {
547                         result = -EINPROGRESS;
548                 } else if (!dev->od_noscrub) {
549                         /* Since we do not know the right OI mapping, we have
550                          * to trigger OI scrub to scan the whole device. */
551                         result = osd_scrub_start(dev, SS_AUTO_FULL |
552                                 SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
553                         CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI "
554                                "scrub by RPC for "DFID", rc = %d [1]\n",
555                                osd_name(dev), PFID(fid), result);
556                         if (result == 0 || result == -EALREADY)
557                                 result = -EINPROGRESS;
558                         else
559                                 result = -EREMCHG;
560                 }
561
562                 /* We still have chance to get the valid inode: for the
563                  * object which is referenced by remote name entry, the
564                  * object on the local MDT will be linked under the dir
565                  * of "/REMOTE_PARENT_DIR" with its FID string as name.
566                  *
567                  * We do not know whether the object for the given FID
568                  * is referenced by some remote name entry or not, and
569                  * especially for DNE II, a multiple-linked object may
570                  * have many name entries reside on many MDTs.
571                  *
572                  * To simplify the operation, OSD will not distinguish
573                  * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
574                  * only happened for the RPC from other MDT during the
575                  * OI scrub, or for the client side RPC with FID only,
576                  * such as FID to path, or from old connected client. */
577                 saved = result;
578                 result = osd_lookup_in_remote_parent(info, dev, fid, id);
579                 if (result == 0) {
580                         in_oi = false;
581                         goto iget;
582                 }
583
584                 GOTO(out, result = saved);
585         }
586
587         obj->oo_inode = inode;
588         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
589
590         result = osd_check_lma(env, obj);
591         if (result != 0) {
592                 iput(inode);
593                 obj->oo_inode = NULL;
594
595                 if (result != -EREMCHG)
596                         GOTO(out, result);
597
598                 if (in_cache)
599                         fid_zero(&oic->oic_fid);
600
601                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
602                 if (result == 0)
603                         goto trigger;
604
605                 if (result != -ENOENT)
606                         GOTO(out, result);
607
608                 if (!in_oi && (fid_is_norm(fid) || fid_is_igif(fid)) &&
609                     ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
610                                      sf->sf_oi_bitmap))
611                         goto trigger;
612
613                 GOTO(out, result = 0);
614         }
615
616         obj->oo_compat_dot_created = 1;
617         obj->oo_compat_dotdot_created = 1;
618
619         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
620                 GOTO(out, result = 0);
621
622         LASSERT(obj->oo_hl_head == NULL);
623         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
624         if (obj->oo_hl_head == NULL) {
625                 obj->oo_inode = NULL;
626                 iput(inode);
627                 GOTO(out, result = -ENOMEM);
628         }
629         GOTO(out, result = 0);
630
631 out:
632         LINVRNT(osd_invariant(obj));
633         return result;
634 }
635
636 /*
637  * Concurrency: shouldn't matter.
638  */
639 static void osd_object_init0(struct osd_object *obj)
640 {
641         LASSERT(obj->oo_inode != NULL);
642         obj->oo_dt.do_body_ops = &osd_body_ops;
643         obj->oo_dt.do_lu.lo_header->loh_attr |=
644                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
645 }
646
647 /*
648  * Concurrency: no concurrent access is possible that early in object
649  * life-cycle.
650  */
651 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
652                            const struct lu_object_conf *conf)
653 {
654         struct osd_object *obj = osd_obj(l);
655         int result;
656
657         LINVRNT(osd_invariant(obj));
658
659         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
660                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
661                 l->lo_header->loh_attr |= LOHA_EXISTS;
662                 return 0;
663         }
664
665         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
666         obj->oo_dt.do_body_ops = &osd_body_ops_new;
667         if (result == 0 && obj->oo_inode != NULL)
668                 osd_object_init0(obj);
669
670         LINVRNT(osd_invariant(obj));
671         return result;
672 }
673
674 /*
675  * Concurrency: no concurrent access is possible that late in object
676  * life-cycle.
677  */
678 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
679 {
680         struct osd_object *obj = osd_obj(l);
681
682         LINVRNT(osd_invariant(obj));
683
684         dt_object_fini(&obj->oo_dt);
685         if (obj->oo_hl_head != NULL)
686                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
687         OBD_FREE_PTR(obj);
688 }
689
690 /*
691  * Concurrency: no concurrent access is possible that late in object
692  * life-cycle.
693  */
694 static void osd_index_fini(struct osd_object *o)
695 {
696         struct iam_container *bag;
697
698         if (o->oo_dir != NULL) {
699                 bag = &o->oo_dir->od_container;
700                 if (o->oo_inode != NULL) {
701                         if (bag->ic_object == o->oo_inode)
702                                 iam_container_fini(bag);
703                 }
704                 OBD_FREE_PTR(o->oo_dir);
705                 o->oo_dir = NULL;
706         }
707 }
708
709 /*
710  * Concurrency: no concurrent access is possible that late in object
711  * life-cycle (for all existing callers, that is. New callers have to provide
712  * their own locking.)
713  */
714 static int osd_inode_unlinked(const struct inode *inode)
715 {
716         return inode->i_nlink == 0;
717 }
718
719 enum {
720         OSD_TXN_OI_DELETE_CREDITS    = 20,
721         OSD_TXN_INODE_DELETE_CREDITS = 20
722 };
723
724 /*
725  * Journal
726  */
727
728 #if OSD_THANDLE_STATS
729 /**
730  * Set time when the handle is allocated
731  */
732 static void osd_th_alloced(struct osd_thandle *oth)
733 {
734         oth->oth_alloced = cfs_time_current();
735 }
736
737 /**
738  * Set time when the handle started
739  */
740 static void osd_th_started(struct osd_thandle *oth)
741 {
742         oth->oth_started = cfs_time_current();
743 }
744
745 /**
746  * Helper function to convert time interval to microseconds packed in
747  * long int.
748  */
749 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
750 {
751         struct timeval val;
752
753         cfs_duration_usec(cfs_time_sub(end, start), &val);
754         return val.tv_sec * 1000000 + val.tv_usec;
755 }
756
757 /**
758  * Check whether the we deal with this handle for too long.
759  */
760 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
761                                 cfs_time_t alloced, cfs_time_t started,
762                                 cfs_time_t closed)
763 {
764         cfs_time_t now = cfs_time_current();
765
766         LASSERT(dev != NULL);
767
768         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
769                             interval_to_usec(alloced, started));
770         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
771                             interval_to_usec(started, closed));
772         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
773                             interval_to_usec(closed, now));
774
775         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
776                 CWARN("transaction handle %p was open for too long: "
777                       "now "CFS_TIME_T" ,"
778                       "alloced "CFS_TIME_T" ,"
779                       "started "CFS_TIME_T" ,"
780                       "closed "CFS_TIME_T"\n",
781                       oth, now, alloced, started, closed);
782                 libcfs_debug_dumpstack(NULL);
783         }
784 }
785
786 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
787 {                                                                       \
788         cfs_time_t __closed = cfs_time_current();                       \
789         cfs_time_t __alloced = oth->oth_alloced;                        \
790         cfs_time_t __started = oth->oth_started;                        \
791                                                                         \
792         expr;                                                           \
793         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
794 }
795
796 #else /* OSD_THANDLE_STATS */
797
798 #define osd_th_alloced(h)                  do {} while(0)
799 #define osd_th_started(h)                  do {} while(0)
800 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
801
802 #endif /* OSD_THANDLE_STATS */
803
804 /*
805  * Concurrency: doesn't access mutable data.
806  */
807 static int osd_param_is_not_sane(const struct osd_device *dev,
808                                  const struct thandle *th)
809 {
810         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
811
812         return oh->ot_credits > osd_transaction_size(dev);
813 }
814
815 /*
816  * Concurrency: shouldn't matter.
817  */
818 static void osd_trans_commit_cb(struct super_block *sb,
819                                 struct ldiskfs_journal_cb_entry *jcb, int error)
820 {
821         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
822         struct thandle     *th  = &oh->ot_super;
823         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
824         struct dt_txn_commit_cb *dcb, *tmp;
825
826         LASSERT(oh->ot_handle == NULL);
827
828         if (error)
829                 CERROR("transaction @0x%p commit error: %d\n", th, error);
830
831         dt_txn_hook_commit(th);
832
833         /* call per-transaction callbacks if any */
834         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
835                                  dcb_linkage) {
836                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
837                          "commit callback entry: magic=%x name='%s'\n",
838                          dcb->dcb_magic, dcb->dcb_name);
839                 list_del_init(&dcb->dcb_linkage);
840                 dcb->dcb_func(NULL, th, dcb, error);
841         }
842
843         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
844         lu_device_put(lud);
845         th->th_dev = NULL;
846
847         lu_context_exit(&th->th_ctx);
848         lu_context_fini(&th->th_ctx);
849         OBD_FREE_PTR(oh);
850 }
851
852 static struct thandle *osd_trans_create(const struct lu_env *env,
853                                         struct dt_device *d)
854 {
855         struct osd_thread_info  *oti = osd_oti_get(env);
856         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
857         struct osd_thandle      *oh;
858         struct thandle          *th;
859         ENTRY;
860
861         /* on pending IO in this thread should left from prev. request */
862         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
863
864         th = ERR_PTR(-ENOMEM);
865         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
866         if (oh != NULL) {
867                 oh->ot_quota_trans = &oti->oti_quota_trans;
868                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
869                 th = &oh->ot_super;
870                 th->th_dev = d;
871                 th->th_result = 0;
872                 th->th_tags = LCT_TX_HANDLE;
873                 oh->ot_credits = 0;
874                 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
875                 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
876                 osd_th_alloced(oh);
877
878                 memset(oti->oti_declare_ops, 0,
879                        sizeof(oti->oti_declare_ops));
880                 memset(oti->oti_declare_ops_cred, 0,
881                        sizeof(oti->oti_declare_ops_cred));
882                 memset(oti->oti_declare_ops_used, 0,
883                        sizeof(oti->oti_declare_ops_used));
884         }
885         RETURN(th);
886 }
887
888 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
889 {
890         struct osd_thread_info  *oti = osd_oti_get(env);
891         struct osd_thandle      *oh;
892
893         oh = container_of0(th, struct osd_thandle, ot_super);
894         LASSERT(oh != NULL);
895
896         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
897               oti->oti_declare_ops[OSD_OT_CREATE],
898               oti->oti_declare_ops_cred[OSD_OT_CREATE],
899               oti->oti_declare_ops_used[OSD_OT_CREATE],
900               oti->oti_declare_ops[OSD_OT_DESTROY],
901               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
902               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
903         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
904               oti->oti_declare_ops[OSD_OT_ATTR_SET],
905               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
906               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
907               oti->oti_declare_ops[OSD_OT_XATTR_SET],
908               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
909               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
910         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
911               oti->oti_declare_ops[OSD_OT_WRITE],
912               oti->oti_declare_ops_cred[OSD_OT_WRITE],
913               oti->oti_declare_ops_used[OSD_OT_WRITE],
914               oti->oti_declare_ops[OSD_OT_PUNCH],
915               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
916               oti->oti_declare_ops_used[OSD_OT_PUNCH],
917               oti->oti_declare_ops[OSD_OT_QUOTA],
918               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
919               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
920         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
921               oti->oti_declare_ops[OSD_OT_INSERT],
922               oti->oti_declare_ops_cred[OSD_OT_INSERT],
923               oti->oti_declare_ops_used[OSD_OT_INSERT],
924               oti->oti_declare_ops[OSD_OT_DELETE],
925               oti->oti_declare_ops_cred[OSD_OT_DELETE],
926               oti->oti_declare_ops_used[OSD_OT_DELETE]);
927         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
928               oti->oti_declare_ops[OSD_OT_REF_ADD],
929               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
930               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
931               oti->oti_declare_ops[OSD_OT_REF_DEL],
932               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
933               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
934 }
935
936 /*
937  * Concurrency: shouldn't matter.
938  */
939 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
940                            struct thandle *th)
941 {
942         struct osd_thread_info *oti = osd_oti_get(env);
943         struct osd_device  *dev = osd_dt_dev(d);
944         handle_t           *jh;
945         struct osd_thandle *oh;
946         int rc;
947
948         ENTRY;
949
950         LASSERT(current->journal_info == NULL);
951
952         oh = container_of0(th, struct osd_thandle, ot_super);
953         LASSERT(oh != NULL);
954         LASSERT(oh->ot_handle == NULL);
955
956         rc = dt_txn_hook_start(env, d, th);
957         if (rc != 0)
958                 GOTO(out, rc);
959
960         if (unlikely(osd_param_is_not_sane(dev, th))) {
961                 static unsigned long last_printed;
962                 static int last_credits;
963
964                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
965                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
966                       oh->ot_credits,
967                       osd_journal(dev)->j_max_transaction_buffers);
968
969                 osd_trans_dump_creds(env, th);
970
971                 if (last_credits != oh->ot_credits &&
972                     time_after(jiffies, last_printed +
973                                msecs_to_jiffies(60 * MSEC_PER_SEC))) {
974                         libcfs_debug_dumpstack(NULL);
975                         last_credits = oh->ot_credits;
976                         last_printed = jiffies;
977                 }
978                 /* XXX Limit the credits to 'max_transaction_buffers', and
979                  *     let the underlying filesystem to catch the error if
980                  *     we really need so many credits.
981                  *
982                  *     This should be removed when we can calculate the
983                  *     credits precisely. */
984                 oh->ot_credits = osd_transaction_size(dev);
985         }
986
987         /*
988          * XXX temporary stuff. Some abstraction layer should
989          * be used.
990          */
991         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
992         osd_th_started(oh);
993         if (!IS_ERR(jh)) {
994                 oh->ot_handle = jh;
995                 LASSERT(oti->oti_txns == 0);
996                 lu_context_init(&th->th_ctx, th->th_tags);
997                 lu_context_enter(&th->th_ctx);
998
999                 lu_device_get(&d->dd_lu_dev);
1000                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1001                               "osd-tx", th);
1002                 oti->oti_txns++;
1003                 rc = 0;
1004         } else {
1005                 rc = PTR_ERR(jh);
1006         }
1007 out:
1008         RETURN(rc);
1009 }
1010
1011 static int osd_seq_exists(const struct lu_env *env,
1012                           struct osd_device *osd, u64 seq)
1013 {
1014         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1015         struct seq_server_site  *ss = osd_seq_site(osd);
1016         int                     rc;
1017         ENTRY;
1018
1019         LASSERT(ss != NULL);
1020         LASSERT(ss->ss_server_fld != NULL);
1021
1022         rc = osd_fld_lookup(env, osd, seq, range);
1023         if (rc != 0) {
1024                 if (rc != -ENOENT)
1025                         CERROR("%s: can't lookup FLD sequence "LPX64
1026                                ": rc = %d\n", osd_name(osd), seq, rc);
1027                 RETURN(0);
1028         }
1029
1030         RETURN(ss->ss_node_id == range->lsr_index);
1031 }
1032
1033 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1034 {
1035         struct dt_txn_commit_cb *dcb;
1036         struct dt_txn_commit_cb *tmp;
1037
1038         /* call per-transaction stop callbacks if any */
1039         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1040                                  dcb_linkage) {
1041                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1042                          "commit callback entry: magic=%x name='%s'\n",
1043                          dcb->dcb_magic, dcb->dcb_name);
1044                 list_del_init(&dcb->dcb_linkage);
1045                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1046         }
1047 }
1048
1049 /*
1050  * Concurrency: shouldn't matter.
1051  */
1052 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1053                           struct thandle *th)
1054 {
1055         int                     rc = 0;
1056         struct osd_thandle     *oh;
1057         struct osd_thread_info *oti = osd_oti_get(env);
1058         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1059         struct osd_device      *osd = osd_dt_dev(th->th_dev);
1060         struct qsd_instance    *qsd = osd->od_quota_slave;
1061         struct lquota_trans    *qtrans;
1062         ENTRY;
1063
1064         oh = container_of0(th, struct osd_thandle, ot_super);
1065
1066         qtrans = oh->ot_quota_trans;
1067         oh->ot_quota_trans = NULL;
1068
1069         if (oh->ot_handle != NULL) {
1070                 handle_t *hdl = oh->ot_handle;
1071
1072                 /*
1073                  * add commit callback
1074                  * notice we don't do this in osd_trans_start()
1075                  * as underlying transaction can change during truncate
1076                  */
1077                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1078                                          &oh->ot_jcb);
1079
1080                 LASSERT(oti->oti_txns == 1);
1081                 oti->oti_txns--;
1082
1083                 rc = dt_txn_hook_stop(env, th);
1084                 if (rc != 0)
1085                         CERROR("%s: failed in transaction hook: rc = %d\n",
1086                                osd_name(osd), rc);
1087
1088                 osd_trans_stop_cb(oh, rc);
1089                 /* hook functions might modify th_sync */
1090                 hdl->h_sync = th->th_sync;
1091
1092                 oh->ot_handle = NULL;
1093                 OSD_CHECK_SLOW_TH(oh, osd, rc = ldiskfs_journal_stop(hdl));
1094                 if (rc != 0)
1095                         CERROR("%s: failed to stop transaction: rc = %d\n",
1096                                osd_name(osd), rc);
1097         } else {
1098                 osd_trans_stop_cb(oh, th->th_result);
1099                 OBD_FREE_PTR(oh);
1100         }
1101
1102         /* inform the quota slave device that the transaction is stopping */
1103         qsd_op_end(env, qsd, qtrans);
1104
1105         /* as we want IO to journal and data IO be concurrent, we don't block
1106          * awaiting data IO completion in osd_do_bio(), instead we wait here
1107          * once transaction is submitted to the journal. all reqular requests
1108          * don't do direct IO (except read/write), thus this wait_event becomes
1109          * no-op for them.
1110          *
1111          * IMPORTANT: we have to wait till any IO submited by the thread is
1112          * completed otherwise iobuf may be corrupted by different request
1113          */
1114         wait_event(iobuf->dr_wait,
1115                        atomic_read(&iobuf->dr_numreqs) == 0);
1116         osd_fini_iobuf(osd, iobuf);
1117         if (!rc)
1118                 rc = iobuf->dr_error;
1119
1120         RETURN(rc);
1121 }
1122
1123 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1124 {
1125         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1126                                                ot_super);
1127
1128         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1129         LASSERT(&dcb->dcb_func != NULL);
1130         if (dcb->dcb_flags & DCB_TRANS_STOP)
1131                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1132         else
1133                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1134
1135         return 0;
1136 }
1137
1138 /*
1139  * Called just before object is freed. Releases all resources except for
1140  * object itself (that is released by osd_object_free()).
1141  *
1142  * Concurrency: no concurrent access is possible that late in object
1143  * life-cycle.
1144  */
1145 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1146 {
1147         struct osd_object *obj   = osd_obj(l);
1148         struct inode      *inode = obj->oo_inode;
1149
1150         LINVRNT(osd_invariant(obj));
1151
1152         /*
1153          * If object is unlinked remove fid->ino mapping from object index.
1154          */
1155
1156         osd_index_fini(obj);
1157         if (inode != NULL) {
1158                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1159                 qid_t                    uid = i_uid_read(inode);
1160                 qid_t                    gid = i_gid_read(inode);
1161
1162                 iput(inode);
1163                 obj->oo_inode = NULL;
1164
1165                 if (qsd != NULL) {
1166                         struct osd_thread_info  *info = osd_oti_get(env);
1167                         struct lquota_id_info   *qi = &info->oti_qi;
1168
1169                         /* Release granted quota to master if necessary */
1170                         qi->lqi_id.qid_uid = uid;
1171                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1172
1173                         qi->lqi_id.qid_uid = gid;
1174                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1175                 }
1176         }
1177 }
1178
1179 /*
1180  * Concurrency: ->loo_object_release() is called under site spin-lock.
1181  */
1182 static void osd_object_release(const struct lu_env *env,
1183                                struct lu_object *l)
1184 {
1185 }
1186
1187 /*
1188  * Concurrency: shouldn't matter.
1189  */
1190 static int osd_object_print(const struct lu_env *env, void *cookie,
1191                             lu_printer_t p, const struct lu_object *l)
1192 {
1193         struct osd_object *o = osd_obj(l);
1194         struct iam_descr  *d;
1195
1196         if (o->oo_dir != NULL)
1197                 d = o->oo_dir->od_container.ic_descr;
1198         else
1199                 d = NULL;
1200         return (*p)(env, cookie,
1201                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1202                     o, o->oo_inode,
1203                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1204                     o->oo_inode ? o->oo_inode->i_generation : 0,
1205                     d ? d->id_ops->id_name : "plain");
1206 }
1207
1208 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1209
1210 /*
1211  * Concurrency: shouldn't matter.
1212  */
1213 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1214                struct obd_statfs *sfs)
1215 {
1216         struct osd_device  *osd = osd_dt_dev(d);
1217         struct super_block *sb = osd_sb(osd);
1218         struct kstatfs     *ksfs;
1219         int result = 0;
1220
1221         if (unlikely(osd->od_mnt == NULL))
1222                 return -EINPROGRESS;
1223
1224         /* osd_lproc.c call this without env, allocate ksfs for that case */
1225         if (unlikely(env == NULL)) {
1226                 OBD_ALLOC_PTR(ksfs);
1227                 if (ksfs == NULL)
1228                         return -ENOMEM;
1229         } else {
1230                 ksfs = &osd_oti_get(env)->oti_ksfs;
1231         }
1232
1233         spin_lock(&osd->od_osfs_lock);
1234         result = sb->s_op->statfs(sb->s_root, ksfs);
1235         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1236                 statfs_pack(sfs, ksfs);
1237                 if (unlikely(sb->s_flags & MS_RDONLY))
1238                         sfs->os_state = OS_STATE_READONLY;
1239                 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
1240                                               LDISKFS_FEATURE_INCOMPAT_EXTENTS))
1241                         sfs->os_maxbytes = sb->s_maxbytes;
1242                 else
1243                         sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
1244         }
1245         spin_unlock(&osd->od_osfs_lock);
1246
1247         if (unlikely(env == NULL))
1248                 OBD_FREE_PTR(ksfs);
1249
1250         /* Reserve a small amount of space for local objects like last_rcvd,
1251          * llog, quota files, ... */
1252         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1253                 sfs->os_bavail = 0;
1254         } else {
1255                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1256                 /** Take out metadata overhead for indirect blocks */
1257                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1258         }
1259
1260         return result;
1261 }
1262
1263 /**
1264  * Estimate space needed for file creations. We assume the largest filename
1265  * which is 2^64 - 1, hence a filename of 20 chars.
1266  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1267  */
1268 #ifdef __LDISKFS_DIR_REC_LEN
1269 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1270 #else
1271 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1272 #endif
1273
1274 /*
1275  * Concurrency: doesn't access mutable data.
1276  */
1277 static void osd_conf_get(const struct lu_env *env,
1278                          const struct dt_device *dev,
1279                          struct dt_device_param *param)
1280 {
1281         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1282         int                ea_overhead;
1283
1284         /*
1285          * XXX should be taken from not-yet-existing fs abstraction layer.
1286          */
1287         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1288         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1289         param->ddp_block_shift  = sb->s_blocksize_bits;
1290         param->ddp_mount_type     = LDD_MT_LDISKFS;
1291         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
1292                 param->ddp_maxbytes = sb->s_maxbytes;
1293         else
1294                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
1295         /* Overhead estimate should be fairly accurate, so we really take a tiny
1296          * error margin which also avoids fragmenting the filesystem too much */
1297         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1298         /* inode are statically allocated, so per-inode space consumption
1299          * is the space consumed by the directory entry */
1300         param->ddp_inodespace     = PER_OBJ_USAGE;
1301         /* per-fragment overhead to be used by the client code */
1302         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1303         param->ddp_mntopts      = 0;
1304         if (test_opt(sb, XATTR_USER))
1305                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1306         if (test_opt(sb, POSIX_ACL))
1307                 param->ddp_mntopts |= MNTOPT_ACL;
1308
1309         /* LOD might calculate the max stripe count based on max_ea_size,
1310          * so we need take account in the overhead as well,
1311          * xattr_header + magic + xattr_entry_head */
1312         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1313                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1314
1315 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1316         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1317                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1318                                                                 ea_overhead;
1319         else
1320 #endif
1321                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1322 }
1323
1324 /*
1325  * Concurrency: shouldn't matter.
1326  */
1327 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1328 {
1329         int rc;
1330
1331         CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1332
1333         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1334
1335         CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
1336                LUSTRE_OSD_LDISKFS_NAME, rc);
1337
1338         return rc;
1339 }
1340
1341 /**
1342  * Start commit for OSD device.
1343  *
1344  * An implementation of dt_commit_async method for OSD device.
1345  * Asychronously starts underlayng fs sync and thereby a transaction
1346  * commit.
1347  *
1348  * \param env environment
1349  * \param d dt device
1350  *
1351  * \see dt_device_operations
1352  */
1353 static int osd_commit_async(const struct lu_env *env,
1354                             struct dt_device *d)
1355 {
1356         struct super_block *s = osd_sb(osd_dt_dev(d));
1357         ENTRY;
1358
1359         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1360         RETURN(s->s_op->sync_fs(s, 0));
1361 }
1362
1363 /*
1364  * Concurrency: shouldn't matter.
1365  */
1366
1367 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1368 {
1369         struct super_block *sb = osd_sb(osd_dt_dev(d));
1370         struct block_device *dev = sb->s_bdev;
1371 #ifdef HAVE_DEV_SET_RDONLY
1372         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1373         int rc = 0;
1374 #else
1375         int rc = -EOPNOTSUPP;
1376 #endif
1377         ENTRY;
1378
1379 #ifdef HAVE_DEV_SET_RDONLY
1380         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1381
1382         if (jdev && (jdev != dev)) {
1383                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1384                        (long)jdev);
1385                 dev_set_rdonly(jdev);
1386         }
1387         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1388         dev_set_rdonly(dev);
1389 #else
1390         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1391                osd_dt_dev(d)->od_svname, (long)dev, rc);
1392 #endif
1393         RETURN(rc);
1394 }
1395
1396 /**
1397  * Note: we do not count into QUOTA here.
1398  * If we mount with --data_journal we may need more.
1399  */
1400 const int osd_dto_credits_noquota[DTO_NR] = {
1401         /**
1402          * Insert.
1403          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1404          * SINGLEDATA_TRANS_BLOCKS(8)
1405          * XXX Note: maybe iam need more, since iam have more level than
1406          *           EXT3 htree.
1407          */
1408         [DTO_INDEX_INSERT]  = 16,
1409         /**
1410          * Delete
1411          * just modify a single entry, probably merge few within a block
1412          */
1413         [DTO_INDEX_DELETE]  = 1,
1414         /**
1415          * Used for OI scrub
1416          */
1417         [DTO_INDEX_UPDATE]  = 16,
1418         /**
1419          * 4(inode, inode bits, groups, GDT)
1420          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1421          */
1422         [DTO_OBJECT_CREATE] = 4,
1423         /**
1424          * 4(inode, inode bits, groups, GDT)
1425          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1426          */
1427         [DTO_OBJECT_DELETE] = 4,
1428         /**
1429          * Attr set credits (inode)
1430          */
1431         [DTO_ATTR_SET_BASE] = 1,
1432         /**
1433          * Xattr set. The same as xattr of EXT3.
1434          * DATA_TRANS_BLOCKS(14)
1435          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1436          * are also counted in. Do not know why?
1437          */
1438         [DTO_XATTR_SET]     = 14,
1439         /**
1440          * credits for inode change during write.
1441          */
1442         [DTO_WRITE_BASE]    = 3,
1443         /**
1444          * credits for single block write.
1445          */
1446         [DTO_WRITE_BLOCK]   = 14,
1447         /**
1448          * Attr set credits for chown.
1449          * This is extra credits for setattr, and it is null without quota
1450          */
1451         [DTO_ATTR_SET_CHOWN] = 0
1452 };
1453
1454 static const struct dt_device_operations osd_dt_ops = {
1455         .dt_root_get       = osd_root_get,
1456         .dt_statfs         = osd_statfs,
1457         .dt_trans_create   = osd_trans_create,
1458         .dt_trans_start    = osd_trans_start,
1459         .dt_trans_stop     = osd_trans_stop,
1460         .dt_trans_cb_add   = osd_trans_cb_add,
1461         .dt_conf_get       = osd_conf_get,
1462         .dt_sync           = osd_sync,
1463         .dt_ro             = osd_ro,
1464         .dt_commit_async   = osd_commit_async,
1465 };
1466
1467 static void osd_object_read_lock(const struct lu_env *env,
1468                                  struct dt_object *dt, unsigned role)
1469 {
1470         struct osd_object *obj = osd_dt_obj(dt);
1471         struct osd_thread_info *oti = osd_oti_get(env);
1472
1473         LINVRNT(osd_invariant(obj));
1474
1475         LASSERT(obj->oo_owner != env);
1476         down_read_nested(&obj->oo_sem, role);
1477
1478         LASSERT(obj->oo_owner == NULL);
1479         oti->oti_r_locks++;
1480 }
1481
1482 static void osd_object_write_lock(const struct lu_env *env,
1483                                   struct dt_object *dt, unsigned role)
1484 {
1485         struct osd_object *obj = osd_dt_obj(dt);
1486         struct osd_thread_info *oti = osd_oti_get(env);
1487
1488         LINVRNT(osd_invariant(obj));
1489
1490         LASSERT(obj->oo_owner != env);
1491         down_write_nested(&obj->oo_sem, role);
1492
1493         LASSERT(obj->oo_owner == NULL);
1494         obj->oo_owner = env;
1495         oti->oti_w_locks++;
1496 }
1497
1498 static void osd_object_read_unlock(const struct lu_env *env,
1499                                    struct dt_object *dt)
1500 {
1501         struct osd_object *obj = osd_dt_obj(dt);
1502         struct osd_thread_info *oti = osd_oti_get(env);
1503
1504         LINVRNT(osd_invariant(obj));
1505
1506         LASSERT(oti->oti_r_locks > 0);
1507         oti->oti_r_locks--;
1508         up_read(&obj->oo_sem);
1509 }
1510
1511 static void osd_object_write_unlock(const struct lu_env *env,
1512                                     struct dt_object *dt)
1513 {
1514         struct osd_object *obj = osd_dt_obj(dt);
1515         struct osd_thread_info *oti = osd_oti_get(env);
1516
1517         LINVRNT(osd_invariant(obj));
1518
1519         LASSERT(obj->oo_owner == env);
1520         LASSERT(oti->oti_w_locks > 0);
1521         oti->oti_w_locks--;
1522         obj->oo_owner = NULL;
1523         up_write(&obj->oo_sem);
1524 }
1525
1526 static int osd_object_write_locked(const struct lu_env *env,
1527                                    struct dt_object *dt)
1528 {
1529         struct osd_object *obj = osd_dt_obj(dt);
1530
1531         LINVRNT(osd_invariant(obj));
1532
1533         return obj->oo_owner == env;
1534 }
1535
1536 static struct timespec *osd_inode_time(const struct lu_env *env,
1537                                        struct inode *inode, __u64 seconds)
1538 {
1539         struct osd_thread_info  *oti = osd_oti_get(env);
1540         struct timespec         *t   = &oti->oti_time;
1541
1542         t->tv_sec = seconds;
1543         t->tv_nsec = 0;
1544         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1545         return t;
1546 }
1547
1548
1549 static void osd_inode_getattr(const struct lu_env *env,
1550                               struct inode *inode, struct lu_attr *attr)
1551 {
1552         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1553                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1554                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1555                            LA_TYPE;
1556
1557         attr->la_atime   = LTIME_S(inode->i_atime);
1558         attr->la_mtime   = LTIME_S(inode->i_mtime);
1559         attr->la_ctime   = LTIME_S(inode->i_ctime);
1560         attr->la_mode    = inode->i_mode;
1561         attr->la_size    = i_size_read(inode);
1562         attr->la_blocks  = inode->i_blocks;
1563         attr->la_uid     = i_uid_read(inode);
1564         attr->la_gid     = i_gid_read(inode);
1565         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1566         attr->la_nlink   = inode->i_nlink;
1567         attr->la_rdev    = inode->i_rdev;
1568         attr->la_blksize = 1 << inode->i_blkbits;
1569         attr->la_blkbits = inode->i_blkbits;
1570 }
1571
1572 static int osd_attr_get(const struct lu_env *env,
1573                         struct dt_object *dt,
1574                         struct lu_attr *attr)
1575 {
1576         struct osd_object *obj = osd_dt_obj(dt);
1577
1578         if (!dt_object_exists(dt))
1579                 return -ENOENT;
1580
1581         LASSERT(!dt_object_remote(dt));
1582         LINVRNT(osd_invariant(obj));
1583
1584         spin_lock(&obj->oo_guard);
1585         osd_inode_getattr(env, obj->oo_inode, attr);
1586         spin_unlock(&obj->oo_guard);
1587         return 0;
1588 }
1589
1590 static int osd_declare_attr_set(const struct lu_env *env,
1591                                 struct dt_object *dt,
1592                                 const struct lu_attr *attr,
1593                                 struct thandle *handle)
1594 {
1595         struct osd_thandle     *oh;
1596         struct osd_object      *obj;
1597         struct osd_thread_info *info = osd_oti_get(env);
1598         struct lquota_id_info  *qi = &info->oti_qi;
1599         qid_t                   uid;
1600         qid_t                   gid;
1601         long long               bspace;
1602         int                     rc = 0;
1603         bool                    enforce;
1604         ENTRY;
1605
1606         LASSERT(dt != NULL);
1607         LASSERT(handle != NULL);
1608
1609         obj = osd_dt_obj(dt);
1610         LASSERT(osd_invariant(obj));
1611
1612         oh = container_of0(handle, struct osd_thandle, ot_super);
1613         LASSERT(oh->ot_handle == NULL);
1614
1615         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1616                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1617
1618         if (attr == NULL || obj->oo_inode == NULL)
1619                 RETURN(rc);
1620
1621         bspace   = obj->oo_inode->i_blocks;
1622         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1623         bspace   = toqb(bspace);
1624
1625         /* Changing ownership is always preformed by super user, it should not
1626          * fail with EDQUOT.
1627          *
1628          * We still need to call the osd_declare_qid() to calculate the journal
1629          * credits for updating quota accounting files and to trigger quota
1630          * space adjustment once the operation is completed.*/
1631         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1632                 /* USERQUOTA */
1633                 uid = i_uid_read(obj->oo_inode);
1634                 qi->lqi_type = USRQUOTA;
1635                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1636                 /* inode accounting */
1637                 qi->lqi_is_blk = false;
1638
1639                 /* one more inode for the new uid ... */
1640                 qi->lqi_id.qid_uid = attr->la_uid;
1641                 qi->lqi_space      = 1;
1642                 /* Reserve credits for the new uid */
1643                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1644                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1645                         rc = 0;
1646                 if (rc)
1647                         RETURN(rc);
1648
1649                 /* and one less inode for the current uid */
1650                 qi->lqi_id.qid_uid = uid;
1651                 qi->lqi_space      = -1;
1652                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1653                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1654                         rc = 0;
1655                 if (rc)
1656                         RETURN(rc);
1657
1658                 /* block accounting */
1659                 qi->lqi_is_blk = true;
1660
1661                 /* more blocks for the new uid ... */
1662                 qi->lqi_id.qid_uid = attr->la_uid;
1663                 qi->lqi_space      = bspace;
1664                 /*
1665                  * Credits for the new uid has been reserved, re-use "obj"
1666                  * to save credit reservation.
1667                  */
1668                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1669                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1670                         rc = 0;
1671                 if (rc)
1672                         RETURN(rc);
1673
1674                 /* and finally less blocks for the current uid */
1675                 qi->lqi_id.qid_uid = uid;
1676                 qi->lqi_space      = -bspace;
1677                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1678                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1679                         rc = 0;
1680                 if (rc)
1681                         RETURN(rc);
1682
1683                 /* GROUP QUOTA */
1684                 gid = i_gid_read(obj->oo_inode);
1685                 qi->lqi_type = GRPQUOTA;
1686                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1687
1688                 /* inode accounting */
1689                 qi->lqi_is_blk = false;
1690
1691                 /* one more inode for the new gid ... */
1692                 qi->lqi_id.qid_gid = attr->la_gid;
1693                 qi->lqi_space      = 1;
1694                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1695                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1696                         rc = 0;
1697                 if (rc)
1698                         RETURN(rc);
1699
1700                 /* and one less inode for the current gid */
1701                 qi->lqi_id.qid_gid = gid;
1702                 qi->lqi_space      = -1;
1703                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1704                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1705                         rc = 0;
1706                 if (rc)
1707                         RETURN(rc);
1708
1709                 /* block accounting */
1710                 qi->lqi_is_blk = true;
1711
1712                 /* more blocks for the new gid ... */
1713                 qi->lqi_id.qid_gid = attr->la_gid;
1714                 qi->lqi_space      = bspace;
1715                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1716                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1717                         rc = 0;
1718                 if (rc)
1719                         RETURN(rc);
1720
1721                 /* and finally less blocks for the current gid */
1722                 qi->lqi_id.qid_gid = gid;
1723                 qi->lqi_space      = -bspace;
1724                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1725                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1726                         rc = 0;
1727                 if (rc)
1728                         RETURN(rc);
1729         }
1730
1731         RETURN(rc);
1732 }
1733
1734 static int osd_inode_setattr(const struct lu_env *env,
1735                              struct inode *inode, const struct lu_attr *attr)
1736 {
1737         __u64 bits = attr->la_valid;
1738
1739         /* Only allow set size for regular file */
1740         if (!S_ISREG(inode->i_mode))
1741                 bits &= ~(LA_SIZE | LA_BLOCKS);
1742
1743         if (bits == 0)
1744                 return 0;
1745
1746         if (bits & LA_ATIME)
1747                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1748         if (bits & LA_CTIME)
1749                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1750         if (bits & LA_MTIME)
1751                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1752         if (bits & LA_SIZE) {
1753                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1754                 i_size_write(inode, attr->la_size);
1755         }
1756
1757 #if 0
1758         /* OSD should not change "i_blocks" which is used by quota.
1759          * "i_blocks" should be changed by ldiskfs only. */
1760         if (bits & LA_BLOCKS)
1761                 inode->i_blocks = attr->la_blocks;
1762 #endif
1763         if (bits & LA_MODE)
1764                 inode->i_mode = (inode->i_mode & S_IFMT) |
1765                                 (attr->la_mode & ~S_IFMT);
1766         if (bits & LA_UID)
1767                 i_uid_write(inode, attr->la_uid);
1768         if (bits & LA_GID)
1769                 i_gid_write(inode, attr->la_gid);
1770         if (bits & LA_NLINK)
1771                 set_nlink(inode, attr->la_nlink);
1772         if (bits & LA_RDEV)
1773                 inode->i_rdev = attr->la_rdev;
1774
1775         if (bits & LA_FLAGS) {
1776                 /* always keep S_NOCMTIME */
1777                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1778                                  S_NOCMTIME;
1779         }
1780         return 0;
1781 }
1782
1783 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1784 {
1785         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1786             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1787                 struct iattr    iattr;
1788                 int             rc;
1789
1790                 ll_vfs_dq_init(inode);
1791                 iattr.ia_valid = 0;
1792                 if (attr->la_valid & LA_UID)
1793                         iattr.ia_valid |= ATTR_UID;
1794                 if (attr->la_valid & LA_GID)
1795                         iattr.ia_valid |= ATTR_GID;
1796                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1797                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1798
1799                 rc = ll_vfs_dq_transfer(inode, &iattr);
1800                 if (rc) {
1801                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1802                                "enforcement enabled on the ldiskfs "
1803                                "filesystem?\n", inode->i_sb->s_id, rc);
1804                         return rc;
1805                 }
1806         }
1807         return 0;
1808 }
1809
1810 static int osd_attr_set(const struct lu_env *env,
1811                         struct dt_object *dt,
1812                         const struct lu_attr *attr,
1813                         struct thandle *handle)
1814 {
1815         struct osd_object *obj = osd_dt_obj(dt);
1816         struct inode      *inode;
1817         int rc;
1818
1819         if (!dt_object_exists(dt))
1820                 return -ENOENT;
1821
1822         LASSERT(handle != NULL);
1823         LASSERT(!dt_object_remote(dt));
1824         LASSERT(osd_invariant(obj));
1825
1826         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1827
1828         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1829                 struct osd_thread_info  *oti  = osd_oti_get(env);
1830                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1831                 struct lu_fid           *fid1 = &oti->oti_fid;
1832                 struct osd_inode_id     *id   = &oti->oti_id;
1833                 struct iam_path_descr   *ipd;
1834                 struct iam_container    *bag;
1835                 struct osd_thandle      *oh;
1836                 int                      rc;
1837
1838                 fid_cpu_to_be(fid1, fid0);
1839                 memset(id, 1, sizeof(*id));
1840                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1841                                   fid0)->oi_dir.od_container;
1842                 ipd = osd_idx_ipd_get(env, bag);
1843                 if (unlikely(ipd == NULL))
1844                         RETURN(-ENOMEM);
1845
1846                 oh = container_of0(handle, struct osd_thandle, ot_super);
1847                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1848                                 (const struct iam_rec *)id, ipd);
1849                 osd_ipd_put(env, bag, ipd);
1850                 return(rc > 0 ? 0 : rc);
1851         }
1852
1853         inode = obj->oo_inode;
1854
1855         rc = osd_quota_transfer(inode, attr);
1856         if (rc)
1857                 return rc;
1858
1859         spin_lock(&obj->oo_guard);
1860         rc = osd_inode_setattr(env, inode, attr);
1861         spin_unlock(&obj->oo_guard);
1862
1863         if (!rc)
1864                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1865
1866         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
1867
1868         return rc;
1869 }
1870
1871 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
1872                                            struct osd_object *obj,
1873                                            const char *name, const int namelen)
1874 {
1875         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1876 }
1877
1878 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1879                       umode_t mode, struct dt_allocation_hint *hint,
1880                       struct thandle *th)
1881 {
1882         int result;
1883         struct osd_device  *osd = osd_obj2dev(obj);
1884         struct osd_thandle *oth;
1885         struct dt_object   *parent = NULL;
1886         struct inode       *inode;
1887
1888         LINVRNT(osd_invariant(obj));
1889         LASSERT(obj->oo_inode == NULL);
1890         LASSERT(obj->oo_hl_head == NULL);
1891
1892         if (S_ISDIR(mode) && ldiskfs_pdo) {
1893                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1894                 if (obj->oo_hl_head == NULL)
1895                         return -ENOMEM;
1896         }
1897
1898         oth = container_of(th, struct osd_thandle, ot_super);
1899         LASSERT(oth->ot_handle->h_transaction != NULL);
1900
1901         if (hint != NULL && hint->dah_parent != NULL &&
1902             !dt_object_remote(hint->dah_parent))
1903                 parent = hint->dah_parent;
1904
1905         inode = ldiskfs_create_inode(oth->ot_handle,
1906                                      parent ? osd_dt_obj(parent)->oo_inode :
1907                                               osd_sb(osd)->s_root->d_inode,
1908                                      mode);
1909         if (!IS_ERR(inode)) {
1910                 /* Do not update file c/mtime in ldiskfs. */
1911                 inode->i_flags |= S_NOCMTIME;
1912
1913                 /* For new created object, it must be consistent,
1914                  * and it is unnecessary to scrub against it. */
1915                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
1916
1917                 obj->oo_inode = inode;
1918                 result = 0;
1919         } else {
1920                 if (obj->oo_hl_head != NULL) {
1921                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1922                         obj->oo_hl_head = NULL;
1923                 }
1924                 result = PTR_ERR(inode);
1925         }
1926         LINVRNT(osd_invariant(obj));
1927         return result;
1928 }
1929
1930 enum {
1931         OSD_NAME_LEN = 255
1932 };
1933
1934 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1935                      struct lu_attr *attr,
1936                      struct dt_allocation_hint *hint,
1937                      struct dt_object_format *dof,
1938                      struct thandle *th)
1939 {
1940         int result;
1941         struct osd_thandle *oth;
1942         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1943
1944         LASSERT(S_ISDIR(attr->la_mode));
1945
1946         oth = container_of(th, struct osd_thandle, ot_super);
1947         LASSERT(oth->ot_handle->h_transaction != NULL);
1948         result = osd_mkfile(info, obj, mode, hint, th);
1949
1950         return result;
1951 }
1952
1953 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1954                         struct lu_attr *attr,
1955                         struct dt_allocation_hint *hint,
1956                         struct dt_object_format *dof,
1957                         struct thandle *th)
1958 {
1959         int result;
1960         struct osd_thandle *oth;
1961         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1962
1963         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1964
1965         LASSERT(S_ISREG(attr->la_mode));
1966
1967         oth = container_of(th, struct osd_thandle, ot_super);
1968         LASSERT(oth->ot_handle->h_transaction != NULL);
1969
1970         result = osd_mkfile(info, obj, mode, hint, th);
1971         if (result == 0) {
1972                 LASSERT(obj->oo_inode != NULL);
1973                 if (feat->dif_flags & DT_IND_VARKEY)
1974                         result = iam_lvar_create(obj->oo_inode,
1975                                                  feat->dif_keysize_max,
1976                                                  feat->dif_ptrsize,
1977                                                  feat->dif_recsize_max,
1978                                                  oth->ot_handle);
1979                 else
1980                         result = iam_lfix_create(obj->oo_inode,
1981                                                  feat->dif_keysize_max,
1982                                                  feat->dif_ptrsize,
1983                                                  feat->dif_recsize_max,
1984                                                  oth->ot_handle);
1985
1986         }
1987         return result;
1988 }
1989
1990 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1991                      struct lu_attr *attr,
1992                      struct dt_allocation_hint *hint,
1993                      struct dt_object_format *dof,
1994                      struct thandle *th)
1995 {
1996         LASSERT(S_ISREG(attr->la_mode));
1997         return osd_mkfile(info, obj, (attr->la_mode &
1998                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1999 }
2000
2001 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2002                      struct lu_attr *attr,
2003                      struct dt_allocation_hint *hint,
2004                      struct dt_object_format *dof,
2005                      struct thandle *th)
2006 {
2007         LASSERT(S_ISLNK(attr->la_mode));
2008         return osd_mkfile(info, obj, (attr->la_mode &
2009                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2010 }
2011
2012 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2013                      struct lu_attr *attr,
2014                      struct dt_allocation_hint *hint,
2015                      struct dt_object_format *dof,
2016                      struct thandle *th)
2017 {
2018         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2019         int result;
2020
2021         LINVRNT(osd_invariant(obj));
2022         LASSERT(obj->oo_inode == NULL);
2023         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2024                 S_ISFIFO(mode) || S_ISSOCK(mode));
2025
2026         result = osd_mkfile(info, obj, mode, hint, th);
2027         if (result == 0) {
2028                 LASSERT(obj->oo_inode != NULL);
2029                 /*
2030                  * This inode should be marked dirty for i_rdev.  Currently
2031                  * that is done in the osd_attr_init().
2032                  */
2033                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2034                                    attr->la_rdev);
2035         }
2036         LINVRNT(osd_invariant(obj));
2037         return result;
2038 }
2039
2040 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2041                               struct lu_attr *,
2042                               struct dt_allocation_hint *hint,
2043                               struct dt_object_format *dof,
2044                               struct thandle *);
2045
2046 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2047 {
2048         osd_obj_type_f result;
2049
2050         switch (type) {
2051         case DFT_DIR:
2052                 result = osd_mkdir;
2053                 break;
2054         case DFT_REGULAR:
2055                 result = osd_mkreg;
2056                 break;
2057         case DFT_SYM:
2058                 result = osd_mksym;
2059                 break;
2060         case DFT_NODE:
2061                 result = osd_mknod;
2062                 break;
2063         case DFT_INDEX:
2064                 result = osd_mk_index;
2065                 break;
2066
2067         default:
2068                 LBUG();
2069                 break;
2070         }
2071         return result;
2072 }
2073
2074
2075 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2076                         struct dt_object *parent, struct dt_object *child,
2077                         umode_t child_mode)
2078 {
2079         LASSERT(ah);
2080
2081         ah->dah_parent = parent;
2082         ah->dah_mode = child_mode;
2083 }
2084
2085 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2086                           struct lu_attr *attr, struct dt_object_format *dof)
2087 {
2088         struct inode   *inode = obj->oo_inode;
2089         __u64           valid = attr->la_valid;
2090         int             result;
2091
2092         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2093
2094         if (dof->dof_type != DFT_NODE)
2095                 attr->la_valid &= ~LA_RDEV;
2096         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2097                 attr->la_valid &= ~LA_ATIME;
2098         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2099                 attr->la_valid &= ~LA_CTIME;
2100         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2101                 attr->la_valid &= ~LA_MTIME;
2102
2103         result = osd_quota_transfer(inode, attr);
2104         if (result)
2105                 return;
2106
2107         if (attr->la_valid != 0) {
2108                 result = osd_inode_setattr(info->oti_env, inode, attr);
2109                 /*
2110                  * The osd_inode_setattr() should always succeed here.  The
2111                  * only error that could be returned is EDQUOT when we are
2112                  * trying to change the UID or GID of the inode. However, this
2113                  * should not happen since quota enforcement is no longer
2114                  * enabled on ldiskfs (lquota takes care of it).
2115                  */
2116                 LASSERTF(result == 0, "%d\n", result);
2117                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2118         }
2119
2120         attr->la_valid = valid;
2121 }
2122
2123 /**
2124  * Helper function for osd_object_create()
2125  *
2126  * \retval 0, on success
2127  */
2128 static int __osd_object_create(struct osd_thread_info *info,
2129                                struct osd_object *obj, struct lu_attr *attr,
2130                                struct dt_allocation_hint *hint,
2131                                struct dt_object_format *dof,
2132                                struct thandle *th)
2133 {
2134         int     result;
2135         __u32   umask;
2136
2137         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
2138
2139         /* we drop umask so that permissions we pass are not affected */
2140         umask = current->fs->umask;
2141         current->fs->umask = 0;
2142
2143         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2144                                                   th);
2145         if (likely(obj->oo_inode != NULL)) {
2146                 LASSERT(obj->oo_inode->i_state & I_NEW);
2147
2148                 /* Unlock the inode before attr initialization to avoid
2149                  * unnecessary dqget operations. LU-6378 */
2150                 unlock_new_inode(obj->oo_inode);
2151         }
2152
2153         if (likely(result == 0)) {
2154                 osd_attr_init(info, obj, attr, dof);
2155                 osd_object_init0(obj);
2156         }
2157
2158         /* restore previous umask value */
2159         current->fs->umask = umask;
2160
2161         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
2162
2163         return result;
2164 }
2165
2166 /**
2167  * Helper function for osd_object_create()
2168  *
2169  * \retval 0, on success
2170  */
2171 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2172                            const struct lu_fid *fid, struct thandle *th)
2173 {
2174         struct osd_thread_info *info = osd_oti_get(env);
2175         struct osd_inode_id    *id   = &info->oti_id;
2176         struct osd_device      *osd  = osd_obj2dev(obj);
2177         struct osd_thandle     *oh;
2178         int                     rc;
2179
2180         LASSERT(obj->oo_inode != NULL);
2181
2182         oh = container_of0(th, struct osd_thandle, ot_super);
2183         LASSERT(oh->ot_handle);
2184         osd_trans_exec_op(env, th, OSD_OT_INSERT);
2185
2186         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2187         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2188         osd_trans_exec_check(env, th, OSD_OT_INSERT);
2189
2190         return rc;
2191 }
2192
2193 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2194                    u64 seq, struct lu_seq_range *range)
2195 {
2196         struct seq_server_site  *ss = osd_seq_site(osd);
2197
2198         if (fid_seq_is_idif(seq)) {
2199                 fld_range_set_ost(range);
2200                 range->lsr_index = idif_ost_idx(seq);
2201                 return 0;
2202         }
2203
2204         if (!fid_seq_in_fldb(seq)) {
2205                 fld_range_set_mdt(range);
2206                 if (ss != NULL)
2207                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2208                          * at all */
2209                         range->lsr_index = ss->ss_node_id;
2210                 return 0;
2211         }
2212
2213         LASSERT(ss != NULL);
2214         fld_range_set_any(range);
2215         /* OSD will only do local fld lookup */
2216         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2217 }
2218
2219 /*
2220  * Concurrency: no external locking is necessary.
2221  */
2222 static int osd_declare_object_create(const struct lu_env *env,
2223                                      struct dt_object *dt,
2224                                      struct lu_attr *attr,
2225                                      struct dt_allocation_hint *hint,
2226                                      struct dt_object_format *dof,
2227                                      struct thandle *handle)
2228 {
2229         struct osd_thandle      *oh;
2230         int                      rc;
2231         ENTRY;
2232
2233         LASSERT(handle != NULL);
2234
2235         oh = container_of0(handle, struct osd_thandle, ot_super);
2236         LASSERT(oh->ot_handle == NULL);
2237
2238         /* EA object consumes more credits than regular object: osd_mk_index
2239          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
2240          * leaf_node, could involves the block, block bitmap, groups, GDT
2241          * change for each block, so add 4 * 2 credits in that case. */
2242         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2243                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
2244                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
2245         /* Reuse idle OI block may cause additional one OI block
2246          * to be changed. */
2247         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2248                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2249
2250         if (!attr)
2251                 RETURN(0);
2252
2253         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2254                                    osd_dt_obj(dt), false, NULL, false);
2255         if (rc != 0)
2256                 RETURN(rc);
2257
2258         RETURN(rc);
2259 }
2260
2261 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2262                              struct lu_attr *attr,
2263                              struct dt_allocation_hint *hint,
2264                              struct dt_object_format *dof, struct thandle *th)
2265 {
2266         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2267         struct osd_object       *obj    = osd_dt_obj(dt);
2268         struct osd_thread_info  *info   = osd_oti_get(env);
2269         int result;
2270         ENTRY;
2271
2272         if (dt_object_exists(dt))
2273                 return -EEXIST;
2274
2275         LINVRNT(osd_invariant(obj));
2276         LASSERT(!dt_object_remote(dt));
2277         LASSERT(osd_write_locked(env, obj));
2278         LASSERT(th != NULL);
2279
2280         if (unlikely(fid_is_acct(fid)))
2281                 /* Quota files can't be created from the kernel any more,
2282                  * 'tune2fs -O quota' will take care of creating them */
2283                 RETURN(-EPERM);
2284
2285         result = __osd_object_create(info, obj, attr, hint, dof, th);
2286         if (result == 0) {
2287                 result = __osd_oi_insert(env, obj, fid, th);
2288                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
2289                         obj->oo_dt.do_body_ops = &osd_body_ops;
2290         }
2291         LASSERT(ergo(result == 0,
2292                 dt_object_exists(dt) && !dt_object_remote(dt)));
2293
2294         LASSERT(osd_invariant(obj));
2295         RETURN(result);
2296 }
2297
2298 /**
2299  * Called to destroy on-disk representation of the object
2300  *
2301  * Concurrency: must be locked
2302  */
2303 static int osd_declare_object_destroy(const struct lu_env *env,
2304                                       struct dt_object *dt,
2305                                       struct thandle *th)
2306 {
2307         struct osd_object  *obj = osd_dt_obj(dt);
2308         struct inode       *inode = obj->oo_inode;
2309         struct osd_thandle *oh;
2310         int                 rc;
2311         ENTRY;
2312
2313         oh = container_of0(th, struct osd_thandle, ot_super);
2314         LASSERT(oh->ot_handle == NULL);
2315         LASSERT(inode);
2316
2317         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2318                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2319         /* Recycle idle OI leaf may cause additional three OI blocks
2320          * to be changed. */
2321         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2322                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2323         /* one less inode */
2324         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2325                                    -1, oh, obj, false, NULL, false);
2326         if (rc)
2327                 RETURN(rc);
2328         /* data to be truncated */
2329         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2330                                    0, oh, obj, true, NULL, false);
2331         RETURN(rc);
2332 }
2333
2334 static int osd_object_destroy(const struct lu_env *env,
2335                               struct dt_object *dt,
2336                               struct thandle *th)
2337 {
2338         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2339         struct osd_object      *obj = osd_dt_obj(dt);
2340         struct inode           *inode = obj->oo_inode;
2341         struct osd_device      *osd = osd_obj2dev(obj);
2342         struct osd_thandle     *oh;
2343         int                     result;
2344         ENTRY;
2345
2346         oh = container_of0(th, struct osd_thandle, ot_super);
2347         LASSERT(oh->ot_handle);
2348         LASSERT(inode);
2349         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2350
2351         if (unlikely(fid_is_acct(fid)))
2352                 RETURN(-EPERM);
2353
2354         if (S_ISDIR(inode->i_mode)) {
2355                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2356                         inode->i_nlink == 2);
2357                 /* it will check/delete the inode from remote parent,
2358                  * how to optimize it? unlink performance impaction XXX */
2359                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2360                 if (result != 0 && result != -ENOENT) {
2361                         CERROR("%s: delete inode "DFID": rc = %d\n",
2362                                osd_name(osd), PFID(fid), result);
2363                 }
2364                 spin_lock(&obj->oo_guard);
2365                 clear_nlink(inode);
2366                 spin_unlock(&obj->oo_guard);
2367                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2368         }
2369
2370         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2371
2372         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2373                                OI_CHECK_FLD);
2374
2375         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
2376         /* XXX: add to ext3 orphan list */
2377         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2378
2379         /* not needed in the cache anymore */
2380         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2381         obj->oo_destroyed = 1;
2382
2383         RETURN(0);
2384 }
2385
2386 /**
2387  * Put the fid into lustre_mdt_attrs, and then place the structure
2388  * inode's ea. This fid should not be altered during the life time
2389  * of the inode.
2390  *
2391  * \retval +ve, on success
2392  * \retval -ve, on error
2393  *
2394  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2395  */
2396 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2397                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2398 {
2399         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2400         int                      rc;
2401         ENTRY;
2402
2403         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2404                 RETURN(0);
2405
2406         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
2407                 rc = -ENOMEM;
2408
2409         lustre_lma_init(lma, fid, compat, incompat);
2410         lustre_lma_swab(lma);
2411
2412         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2413                              XATTR_CREATE);
2414         /* LMA may already exist, but we need to check that all the
2415          * desired compat/incompat flags have been added. */
2416         if (unlikely(rc == -EEXIST)) {
2417                 if (compat == 0 && incompat == 0)
2418                         RETURN(0);
2419
2420                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2421                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2422                                      LMA_OLD_SIZE);
2423                 if (rc <= 0)
2424                         RETURN(-EINVAL);
2425
2426                 lustre_lma_swab(lma);
2427                 if (!(~lma->lma_compat & compat) &&
2428                     !(~lma->lma_incompat & incompat))
2429                         RETURN(0);
2430
2431                 lma->lma_compat |= compat;
2432                 lma->lma_incompat |= incompat;
2433                 lustre_lma_swab(lma);
2434                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2435                                      sizeof(*lma), XATTR_REPLACE);
2436         }
2437
2438         RETURN(rc);
2439 }
2440
2441 /**
2442  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2443  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2444  * To have compatilibility with 1.8 ldiskfs driver we need to have
2445  * magic number at start of fid data.
2446  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2447  * its inmemory API.
2448  */
2449 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2450                                          const struct lu_fid *fid)
2451 {
2452         if (!fid_is_namespace_visible(fid) ||
2453             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2454                 param->edp_magic = 0;
2455                 return;
2456         }
2457
2458         param->edp_magic = LDISKFS_LUFID_MAGIC;
2459         param->edp_len =  sizeof(struct lu_fid) + 1;
2460         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2461 }
2462
2463 /**
2464  * Try to read the fid from inode ea into dt_rec.
2465  *
2466  * \param fid object fid.
2467  *
2468  * \retval 0 on success
2469  */
2470 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2471                           __u32 ino, struct lu_fid *fid,
2472                           struct osd_inode_id *id)
2473 {
2474         struct osd_thread_info *info  = osd_oti_get(env);
2475         struct inode           *inode;
2476         ENTRY;
2477
2478         osd_id_gen(id, ino, OSD_OII_NOGEN);
2479         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2480         if (IS_ERR(inode))
2481                 RETURN(PTR_ERR(inode));
2482
2483         iput(inode);
2484         RETURN(0);
2485 }
2486
2487 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2488                                         struct inode *dir,
2489                                         struct inode *parent_dir,
2490                                         const struct lu_fid *dot_fid,
2491                                         const struct lu_fid *dot_dot_fid,
2492                                         struct osd_thandle *oth)
2493 {
2494         struct ldiskfs_dentry_param *dot_ldp;
2495         struct ldiskfs_dentry_param *dot_dot_ldp;
2496
2497         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2498         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2499
2500         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2501         dot_ldp->edp_magic = 0;
2502
2503         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2504                                       dir, dot_ldp, dot_dot_ldp);
2505 }
2506
2507 /**
2508  * Create an local agent inode for remote entry
2509  */
2510 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2511                                                   struct osd_device *osd,
2512                                                   struct osd_object *pobj,
2513                                                   const struct lu_fid *fid,
2514                                                   __u32 type,
2515                                                   struct thandle *th)
2516 {
2517         struct osd_thread_info  *info = osd_oti_get(env);
2518         struct inode            *local;
2519         struct osd_thandle      *oh;
2520         int                     rc;
2521         ENTRY;
2522
2523         LASSERT(th);
2524         oh = container_of(th, struct osd_thandle, ot_super);
2525         LASSERT(oh->ot_handle->h_transaction != NULL);
2526
2527         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2528         if (IS_ERR(local)) {
2529                 CERROR("%s: create local error %d\n", osd_name(osd),
2530                        (int)PTR_ERR(local));
2531                 RETURN(local);
2532         }
2533
2534         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
2535         unlock_new_inode(local);
2536
2537         /* Set special LMA flag for local agent inode */
2538         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2539         if (rc != 0) {
2540                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2541                        osd_name(osd), PFID(fid), rc);
2542                 RETURN(ERR_PTR(rc));
2543         }
2544
2545         if (!S_ISDIR(type))
2546                 RETURN(local);
2547
2548         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2549                                          lu_object_fid(&pobj->oo_dt.do_lu),
2550                                          fid, oh);
2551         if (rc != 0) {
2552                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2553                         osd_name(osd), PFID(fid), rc);
2554                 RETURN(ERR_PTR(rc));
2555         }
2556
2557         RETURN(local);
2558 }
2559
2560 /**
2561  * Delete local agent inode for remote entry
2562  */
2563 static int osd_delete_local_agent_inode(const struct lu_env *env,
2564                                         struct osd_device *osd,
2565                                         const struct lu_fid *fid,
2566                                         __u32 ino, struct osd_thandle *oh)
2567 {
2568         struct osd_thread_info  *oti = osd_oti_get(env);
2569         struct osd_inode_id     *id = &oti->oti_id;
2570         struct inode            *inode;
2571         ENTRY;
2572
2573         id->oii_ino = le32_to_cpu(ino);
2574         id->oii_gen = OSD_OII_NOGEN;
2575         inode = osd_iget(oti, osd, id);
2576         if (IS_ERR(inode)) {
2577                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2578                        PFID(fid), id->oii_ino, id->oii_gen);
2579                 RETURN(PTR_ERR(inode));
2580         }
2581
2582         clear_nlink(inode);
2583         mark_inode_dirty(inode);
2584         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2585                 osd_name(osd), PFID(fid), inode->i_ino);
2586         iput(inode);
2587         RETURN(0);
2588 }
2589
2590 /**
2591  * OSD layer object create function for interoperability mode (b11826).
2592  * This is mostly similar to osd_object_create(). Only difference being, fid is
2593  * inserted into inode ea here.
2594  *
2595  * \retval   0, on success
2596  * \retval -ve, on error
2597  */
2598 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2599                                 struct lu_attr *attr,
2600                                 struct dt_allocation_hint *hint,
2601                                 struct dt_object_format *dof,
2602                                 struct thandle *th)
2603 {
2604         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2605         struct osd_object       *obj    = osd_dt_obj(dt);
2606         struct osd_thread_info  *info   = osd_oti_get(env);
2607         int                      result;
2608
2609         ENTRY;
2610
2611         if (dt_object_exists(dt))
2612                 RETURN(-EEXIST);
2613
2614         LASSERT(osd_invariant(obj));
2615         LASSERT(!dt_object_remote(dt));
2616         LASSERT(osd_write_locked(env, obj));
2617         LASSERT(th != NULL);
2618
2619         if (unlikely(fid_is_acct(fid)))
2620                 /* Quota files can't be created from the kernel any more,
2621                  * 'tune2fs -O quota' will take care of creating them */
2622                 RETURN(-EPERM);
2623
2624         result = __osd_object_create(info, obj, attr, hint, dof, th);
2625         if (result == 0) {
2626                 if (fid_is_idif(fid) &&
2627                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
2628                         struct lu_fid *tfid = &info->oti_fid;
2629                         struct ost_id *oi   = &info->oti_ostid;
2630
2631                         fid_to_ostid(fid, oi);
2632                         ostid_to_fid(tfid, oi, 0);
2633                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
2634                                                 LMAC_FID_ON_OST, 0);
2635                 } else {
2636                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
2637                                 fid_is_on_ost(info, osd_obj2dev(obj),
2638                                               fid, OI_CHECK_FLD) ?
2639                                 LMAC_FID_ON_OST : 0, 0);
2640                 }
2641                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
2642                         obj->oo_dt.do_body_ops = &osd_body_ops;
2643         }
2644
2645         if (result == 0)
2646                 result = __osd_oi_insert(env, obj, fid, th);
2647
2648         LASSERT(ergo(result == 0,
2649                      dt_object_exists(dt) && !dt_object_remote(dt)));
2650         LINVRNT(osd_invariant(obj));
2651         RETURN(result);
2652 }
2653
2654 static int osd_declare_object_ref_add(const struct lu_env *env,
2655                                       struct dt_object *dt,
2656                                       struct thandle *handle)
2657 {
2658         struct osd_thandle       *oh;
2659
2660         /* it's possible that object doesn't exist yet */
2661         LASSERT(handle != NULL);
2662
2663         oh = container_of0(handle, struct osd_thandle, ot_super);
2664         LASSERT(oh->ot_handle == NULL);
2665
2666         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2667                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2668
2669         return 0;
2670 }
2671
2672 /*
2673  * Concurrency: @dt is write locked.
2674  */
2675 static int osd_object_ref_add(const struct lu_env *env,
2676                               struct dt_object *dt, struct thandle *th)
2677 {
2678         struct osd_object  *obj = osd_dt_obj(dt);
2679         struct inode       *inode = obj->oo_inode;
2680         struct osd_thandle *oh;
2681         int                 rc = 0;
2682
2683         if (!dt_object_exists(dt))
2684                 return -ENOENT;
2685
2686         LINVRNT(osd_invariant(obj));
2687         LASSERT(!dt_object_remote(dt));
2688         LASSERT(osd_write_locked(env, obj));
2689         LASSERT(th != NULL);
2690
2691         oh = container_of0(th, struct osd_thandle, ot_super);
2692         LASSERT(oh->ot_handle != NULL);
2693
2694         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2695
2696         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2697                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2698         /*
2699          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2700          * (65000) subdirectories by storing "1" in i_nlink if the link count
2701          * would otherwise overflow. Directory tranversal tools understand
2702          * that (st_nlink == 1) indicates that the filesystem dose not track
2703          * hard links count on the directory, and will not abort subdirectory
2704          * scanning early once (st_nlink - 2) subdirs have been found.
2705          *
2706          * This also has to properly handle the case of inodes with nlink == 0
2707          * in case they are being linked into the PENDING directory
2708          */
2709         spin_lock(&obj->oo_guard);
2710         if (unlikely(inode->i_nlink == 0))
2711                 /* inc_nlink from 0 may cause WARN_ON */
2712                 set_nlink(inode, 1);
2713         else {
2714                 ldiskfs_inc_count(oh->ot_handle, inode);
2715                 if (!S_ISDIR(inode->i_mode))
2716                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2717         }
2718         spin_unlock(&obj->oo_guard);
2719
2720         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2721         LINVRNT(osd_invariant(obj));
2722
2723         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
2724
2725         return rc;
2726 }
2727
2728 static int osd_declare_object_ref_del(const struct lu_env *env,
2729                                       struct dt_object *dt,
2730                                       struct thandle *handle)
2731 {
2732         struct osd_thandle *oh;
2733
2734         LASSERT(!dt_object_remote(dt));
2735         LASSERT(handle != NULL);
2736
2737         oh = container_of0(handle, struct osd_thandle, ot_super);
2738         LASSERT(oh->ot_handle == NULL);
2739
2740         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2741                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2742
2743         return 0;
2744 }
2745
2746 /*
2747  * Concurrency: @dt is write locked.
2748  */
2749 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2750                               struct thandle *th)
2751 {
2752         struct osd_object       *obj = osd_dt_obj(dt);
2753         struct inode            *inode = obj->oo_inode;
2754         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2755         struct osd_thandle      *oh;
2756
2757         if (!dt_object_exists(dt))
2758                 return -ENOENT;
2759
2760         LINVRNT(osd_invariant(obj));
2761         LASSERT(!dt_object_remote(dt));
2762         LASSERT(osd_write_locked(env, obj));
2763         LASSERT(th != NULL);
2764
2765         oh = container_of0(th, struct osd_thandle, ot_super);
2766         LASSERT(oh->ot_handle != NULL);
2767
2768         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2769
2770         spin_lock(&obj->oo_guard);
2771         /* That can be result of upgrade from old Lustre version and
2772          * applied only to local files.  Just skip this ref_del call.
2773          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2774         if (inode->i_nlink == 0) {
2775                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2776                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2777                              ", maybe an upgraded file? (LU-3915)\n",
2778                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2779                 spin_unlock(&obj->oo_guard);
2780                 return 0;
2781         }
2782
2783         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2784                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2785
2786         ldiskfs_dec_count(oh->ot_handle, inode);
2787         spin_unlock(&obj->oo_guard);
2788
2789         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2790         LINVRNT(osd_invariant(obj));
2791
2792         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
2793
2794         return 0;
2795 }
2796
2797 /*
2798  * Get the 64-bit version for an inode.
2799  */
2800 static int osd_object_version_get(const struct lu_env *env,
2801                                   struct dt_object *dt, dt_obj_version_t *ver)
2802 {
2803         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2804
2805         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2806                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2807         *ver = LDISKFS_I(inode)->i_fs_version;
2808         return 0;
2809 }
2810
2811 /*
2812  * Concurrency: @dt is read locked.
2813  */
2814 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2815                          struct lu_buf *buf, const char *name)
2816 {
2817         struct osd_object      *obj    = osd_dt_obj(dt);
2818         struct inode           *inode  = obj->oo_inode;
2819         struct osd_thread_info *info   = osd_oti_get(env);
2820         struct dentry          *dentry = &info->oti_obj_dentry;
2821
2822         /* version get is not real XATTR but uses xattr API */
2823         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2824                 /* for version we are just using xattr API but change inode
2825                  * field instead */
2826                 if (buf->lb_len == 0)
2827                         return sizeof(dt_obj_version_t);
2828
2829                 if (buf->lb_len < sizeof(dt_obj_version_t))
2830                         return -ERANGE;
2831
2832                 osd_object_version_get(env, dt, buf->lb_buf);
2833
2834                 return sizeof(dt_obj_version_t);
2835         }
2836
2837         if (!dt_object_exists(dt))
2838                 return -ENOENT;
2839
2840         LASSERT(!dt_object_remote(dt));
2841         LASSERT(inode->i_op != NULL);
2842         LASSERT(inode->i_op->getxattr != NULL);
2843
2844         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2845 }
2846
2847
2848 static int osd_declare_xattr_set(const struct lu_env *env,
2849                                  struct dt_object *dt,
2850                                  const struct lu_buf *buf, const char *name,
2851                                  int fl, struct thandle *handle)
2852 {
2853         struct osd_thandle *oh;
2854         int credits;
2855         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2856
2857         LASSERT(handle != NULL);
2858
2859         oh = container_of0(handle, struct osd_thandle, ot_super);
2860         LASSERT(oh->ot_handle == NULL);
2861
2862         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2863                 /* For non-upgrading case, the LMA is set first and
2864                  * usually fit inode. But for upgrade case, the LMA
2865                  * may be in another separated EA block. */
2866                 if (!dt_object_exists(dt))
2867                         credits = 0;
2868                 else if (fl == LU_XATTR_REPLACE)
2869                         credits = 1;
2870                 else
2871                         goto upgrade;
2872         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2873                 credits = 1;
2874         } else {
2875 upgrade:
2876                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2877                 if (buf && buf->lb_len > sb->s_blocksize) {
2878                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2879                                         sb->s_blocksize_bits;
2880                 }
2881                 /*
2882                  * xattr set may involve inode quota change, reserve credits for
2883                  * dquot_initialize()
2884                  */
2885                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
2886         }
2887
2888         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2889
2890         return 0;
2891 }
2892
2893 /*
2894  * Set the 64-bit version for object
2895  */
2896 static void osd_object_version_set(const struct lu_env *env,
2897                                    struct dt_object *dt,
2898                                    dt_obj_version_t *new_version)
2899 {
2900         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2901
2902         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2903                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2904
2905         LDISKFS_I(inode)->i_fs_version = *new_version;
2906         /** Version is set after all inode operations are finished,
2907          *  so we should mark it dirty here */
2908         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2909 }
2910
2911 /*
2912  * Concurrency: @dt is write locked.
2913  */
2914 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2915                          const struct lu_buf *buf, const char *name, int fl,
2916                          struct thandle *handle)
2917 {
2918         struct osd_object      *obj      = osd_dt_obj(dt);
2919         struct inode           *inode    = obj->oo_inode;
2920         struct osd_thread_info *info     = osd_oti_get(env);
2921         int                     fs_flags = 0;
2922         int                     rc;
2923         ENTRY;
2924
2925         LASSERT(handle != NULL);
2926
2927         /* version set is not real XATTR */
2928         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2929                 /* for version we are just using xattr API but change inode
2930                  * field instead */
2931                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2932                 osd_object_version_set(env, dt, buf->lb_buf);
2933                 return sizeof(dt_obj_version_t);
2934         }
2935
2936         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
2937                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
2938
2939         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2940         if (fl & LU_XATTR_REPLACE)
2941                 fs_flags |= XATTR_REPLACE;
2942
2943         if (fl & LU_XATTR_CREATE)
2944                 fs_flags |= XATTR_CREATE;
2945
2946         if (strcmp(name, XATTR_NAME_LMV) == 0) {
2947                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2948                 int                      rc;
2949
2950                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
2951                 if (rc != 0)
2952                         RETURN(rc);
2953
2954                 lma->lma_incompat |= LMAI_STRIPED;
2955                 lustre_lma_swab(lma);
2956                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2957                                      sizeof(*lma), XATTR_REPLACE);
2958                 if (rc != 0)
2959                         RETURN(rc);
2960         }
2961
2962         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
2963             strcmp(name, XATTR_NAME_LINK) == 0)
2964                 return -ENOSPC;
2965
2966         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
2967                                fs_flags);
2968         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2969
2970         return rc;
2971 }
2972
2973 /*
2974  * Concurrency: @dt is read locked.
2975  */
2976 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2977                           const struct lu_buf *buf)
2978 {
2979         struct osd_object      *obj    = osd_dt_obj(dt);
2980         struct inode           *inode  = obj->oo_inode;
2981         struct osd_thread_info *info   = osd_oti_get(env);
2982         struct dentry          *dentry = &info->oti_obj_dentry;
2983
2984         if (!dt_object_exists(dt))
2985                 return -ENOENT;
2986
2987         LASSERT(!dt_object_remote(dt));
2988         LASSERT(inode->i_op != NULL);
2989         LASSERT(inode->i_op->listxattr != NULL);
2990
2991         dentry->d_inode = inode;
2992         dentry->d_sb = inode->i_sb;
2993         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2994 }
2995
2996 static int osd_declare_xattr_del(const struct lu_env *env,
2997                                  struct dt_object *dt, const char *name,
2998                                  struct thandle *handle)
2999 {
3000         struct osd_thandle *oh;
3001         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3002
3003         LASSERT(!dt_object_remote(dt));
3004         LASSERT(handle != NULL);
3005
3006         oh = container_of0(handle, struct osd_thandle, ot_super);
3007         LASSERT(oh->ot_handle == NULL);
3008
3009         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3010                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3011         /*
3012          * xattr del may involve inode quota change, reserve credits for
3013          * dquot_initialize()
3014          */
3015         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3016
3017         return 0;
3018 }
3019
3020 /*
3021  * Concurrency: @dt is write locked.
3022  */
3023 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3024                          const char *name, struct thandle *handle)
3025 {
3026         struct osd_object      *obj    = osd_dt_obj(dt);
3027         struct inode           *inode  = obj->oo_inode;
3028         struct osd_thread_info *info   = osd_oti_get(env);
3029         struct dentry          *dentry = &info->oti_obj_dentry;
3030         int                     rc;
3031
3032         if (!dt_object_exists(dt))
3033                 return -ENOENT;
3034
3035         LASSERT(!dt_object_remote(dt));
3036         LASSERT(inode->i_op != NULL);
3037         LASSERT(inode->i_op->removexattr != NULL);
3038         LASSERT(handle != NULL);
3039
3040         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3041
3042         ll_vfs_dq_init(inode);
3043         dentry->d_inode = inode;
3044         dentry->d_sb = inode->i_sb;
3045         rc = inode->i_op->removexattr(dentry, name);
3046         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3047         return rc;
3048 }
3049
3050 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3051                            __u64 start, __u64 end)
3052 {
3053         struct osd_object       *obj    = osd_dt_obj(dt);
3054         struct inode            *inode  = obj->oo_inode;
3055         struct osd_thread_info  *info   = osd_oti_get(env);
3056         struct dentry           *dentry = &info->oti_obj_dentry;
3057         struct file             *file   = &info->oti_file;
3058         int                     rc;
3059
3060         ENTRY;
3061
3062         dentry->d_inode = inode;
3063         dentry->d_sb = inode->i_sb;
3064         file->f_path.dentry = dentry;
3065         file->f_mapping = inode->i_mapping;
3066         file->f_op = inode->i_fop;
3067         set_file_inode(file, inode);
3068
3069         rc = ll_vfs_fsync_range(file, start, end, 0);
3070
3071         RETURN(rc);
3072 }
3073
3074 /*
3075  * Index operations.
3076  */
3077
3078 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3079                            const struct dt_index_features *feat)
3080 {
3081         struct iam_descr *descr;
3082
3083         if (osd_object_is_root(o))
3084                 return feat == &dt_directory_features;
3085
3086         LASSERT(o->oo_dir != NULL);
3087
3088         descr = o->oo_dir->od_container.ic_descr;
3089         if (feat == &dt_directory_features) {
3090                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3091                         return 1;
3092                 else
3093                         return 0;
3094         } else {
3095                 return
3096                         feat->dif_keysize_min <= descr->id_key_size &&
3097                         descr->id_key_size <= feat->dif_keysize_max &&
3098                         feat->dif_recsize_min <= descr->id_rec_size &&
3099                         descr->id_rec_size <= feat->dif_recsize_max &&
3100                         !(feat->dif_flags & (DT_IND_VARKEY |
3101                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3102                         ergo(feat->dif_flags & DT_IND_UPDATE,
3103                              1 /* XXX check that object (and file system) is
3104                                 * writable */);
3105         }
3106 }
3107
3108 static int osd_iam_container_init(const struct lu_env *env,
3109                                   struct osd_object *obj,
3110                                   struct osd_directory *dir)
3111 {
3112         struct iam_container *bag = &dir->od_container;
3113         int result;
3114
3115         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3116         if (result != 0)
3117                 return result;
3118
3119         result = iam_container_setup(bag);
3120         if (result == 0)
3121                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3122         else
3123                 iam_container_fini(bag);
3124
3125         return result;
3126 }
3127
3128
3129 /*
3130  * Concurrency: no external locking is necessary.
3131  */
3132 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3133                          const struct dt_index_features *feat)
3134 {
3135         int                      result;
3136         int                      skip_iam = 0;
3137         struct osd_object       *obj = osd_dt_obj(dt);
3138
3139         LINVRNT(osd_invariant(obj));
3140
3141         if (osd_object_is_root(obj)) {
3142                 dt->do_index_ops = &osd_index_ea_ops;
3143                 result = 0;
3144         } else if (feat == &dt_directory_features) {
3145                 dt->do_index_ops = &osd_index_ea_ops;
3146                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3147                         result = 0;
3148                 else
3149                         result = -ENOTDIR;
3150                 skip_iam = 1;
3151         } else if (unlikely(feat == &dt_otable_features)) {
3152                 dt->do_index_ops = &osd_otable_ops;
3153                 return 0;
3154         } else if (unlikely(feat == &dt_acct_features)) {
3155                 dt->do_index_ops = &osd_acct_index_ops;
3156                 result = 0;
3157                 skip_iam = 1;
3158         } else if (!osd_has_index(obj)) {
3159                 struct osd_directory *dir;
3160
3161                 OBD_ALLOC_PTR(dir);
3162                 if (dir != NULL) {
3163
3164                         spin_lock(&obj->oo_guard);
3165                         if (obj->oo_dir == NULL)
3166                                 obj->oo_dir = dir;
3167                         else
3168                                 /*
3169                                  * Concurrent thread allocated container data.
3170                                  */
3171                                 OBD_FREE_PTR(dir);
3172                         spin_unlock(&obj->oo_guard);
3173                         /*
3174                          * Now, that we have container data, serialize its
3175                          * initialization.
3176                          */
3177                         down_write(&obj->oo_ext_idx_sem);
3178                         /*
3179                          * recheck under lock.
3180                          */
3181                         if (!osd_has_index(obj))
3182                                 result = osd_iam_container_init(env, obj,
3183                                 &