Whamcloud - gitweb
c771ac559e011bc6080822948fe65c4471691baf
[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 static struct inode *osd_iget_check(struct osd_thread_info *info,
305                                     struct osd_device *dev,
306                                     const struct lu_fid *fid,
307                                     struct osd_inode_id *id,
308                                     bool in_oi)
309 {
310         struct inode    *inode;
311         int              rc     = 0;
312         ENTRY;
313
314         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
315         if (IS_ERR(inode)) {
316                 rc = PTR_ERR(inode);
317                 if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
318                         CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
319                                id->oii_ino, rc);
320
321                         GOTO(put, rc);
322                 }
323
324                 goto check_oi;
325         }
326
327         if (is_bad_inode(inode)) {
328                 rc = -ENOENT;
329                 if (!in_oi) {
330                         CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
331
332                         GOTO(put, rc);
333                 }
334
335                 goto check_oi;
336         }
337
338         if (id->oii_gen != OSD_OII_NOGEN &&
339             inode->i_generation != id->oii_gen) {
340                 rc = -ESTALE;
341                 if (!in_oi) {
342                         CDEBUG(D_INODE, "unmatched inode: ino = %u, "
343                                "oii_gen = %u, i_generation = %u\n",
344                                id->oii_ino, id->oii_gen, inode->i_generation);
345
346                         GOTO(put, rc);
347                 }
348
349                 goto check_oi;
350         }
351
352         if (inode->i_nlink == 0) {
353                 rc = -ENOENT;
354                 if (!in_oi) {
355                         CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
356
357                         GOTO(put, rc);
358                 }
359
360                 goto check_oi;
361         }
362
363 check_oi:
364         if (rc != 0) {
365                 struct osd_inode_id saved_id = *id;
366
367                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
368
369                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
370                 /* XXX: There are some possible cases:
371                  *      1. rc = 0.
372                  *         Backup/restore caused the OI invalid.
373                  *      2. rc = 0.
374                  *         Someone unlinked the object but NOT removed
375                  *         the OI mapping, such as mount target device
376                  *         as ldiskfs, and modify something directly.
377                  *      3. rc = -ENOENT.
378                  *         Someone just removed the object between the
379                  *         former oi_lookup and the iget. It is normal.
380                  *      4. Other failure cases.
381                  *
382                  *      Generally, when the device is mounted, it will
383                  *      auto check whether the system is restored from
384                  *      file-level backup or not. We trust such detect
385                  *      to distinguish the 1st case from the 2nd case. */
386                 if (rc == 0) {
387                         if (!IS_ERR(inode) && inode->i_generation != 0 &&
388                             inode->i_generation == id->oii_gen) {
389                                 rc = -ENOENT;
390                         } else {
391                                 __u32 level = D_LFSCK;
392
393                                 rc = -EREMCHG;
394                                 if (!thread_is_running(&dev->od_scrub.os_thread))
395                                         level |= D_CONSOLE;
396
397                                 CDEBUG(level, "%s: the OI mapping for the FID "
398                                        DFID" become inconsistent, the given ID "
399                                        "%u/%u, the ID in OI mapping %u/%u\n",
400                                        osd_name(dev), PFID(fid),
401                                        saved_id.oii_ino, saved_id.oii_gen,
402                                        id->oii_ino, id->oii_ino);
403                         }
404                 }
405         } else {
406                 if (id->oii_gen == OSD_OII_NOGEN)
407                         osd_id_gen(id, inode->i_ino, inode->i_generation);
408
409                 /* Do not update file c/mtime in ldiskfs.
410                  * NB: we don't have any lock to protect this because we don't
411                  * have reference on osd_object now, but contention with
412                  * another lookup + attr_set can't happen in the tiny window
413                  * between if (...) and set S_NOCMTIME. */
414                 if (!(inode->i_flags & S_NOCMTIME))
415                         inode->i_flags |= S_NOCMTIME;
416         }
417
418         GOTO(put, rc);
419
420 put:
421         if (rc != 0) {
422                 if (!IS_ERR(inode))
423                         iput(inode);
424
425                 inode = ERR_PTR(rc);
426         }
427
428         return inode;
429 }
430
431 /**
432  * \retval +v: new filter_fid, does not contain self-fid
433  * \retval 0:  filter_fid_old, contains self-fid
434  * \retval -v: other failure cases
435  */
436 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
437                  struct dentry *dentry, struct lu_fid *fid)
438 {
439         struct filter_fid_old   *ff     = &info->oti_ff;
440         struct ost_id           *ostid  = &info->oti_ostid;
441         int                      rc;
442
443         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
444         if (rc == sizeof(*ff)) {
445                 rc = 0;
446                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
447                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
448                 /* XXX: should use real OST index in the future. LU-3569 */
449                 ostid_to_fid(fid, ostid, 0);
450         } else if (rc == sizeof(struct filter_fid)) {
451                 rc = 1;
452         } else if (rc >= 0) {
453                 rc = -EINVAL;
454         }
455
456         return rc;
457 }
458
459 static int osd_lma_self_repair(struct osd_thread_info *info,
460                                struct osd_device *osd, struct inode *inode,
461                                const struct lu_fid *fid, __u32 compat)
462 {
463         handle_t *jh;
464         int       rc;
465
466         LASSERT(current->journal_info == NULL);
467
468         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
469                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
470         if (IS_ERR(jh)) {
471                 rc = PTR_ERR(jh);
472                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
473                       osd_name(osd), rc);
474                 return rc;
475         }
476
477         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
478         if (rc != 0)
479                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
480                       osd_name(osd), rc);
481         ldiskfs_journal_stop(jh);
482         return rc;
483 }
484
485 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
486 {
487         struct osd_thread_info  *info   = osd_oti_get(env);
488         struct osd_device       *osd    = osd_obj2dev(obj);
489         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
490         struct inode            *inode  = obj->oo_inode;
491         struct dentry           *dentry = &info->oti_obj_dentry;
492         struct lu_fid           *fid    = NULL;
493         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
494         int                      rc;
495         ENTRY;
496
497         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
498         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
499                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
500         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
501                 fid = &lma->lma_self_fid;
502                 rc = osd_get_idif(info, inode, dentry, fid);
503                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
504                         /* For the given OST-object, if it has neither LMA nor
505                          * FID in XATTR_NAME_FID, then the given FID (which is
506                          * contained in the @obj, from client RPC for locating
507                          * the OST-object) is trusted. We use it to generate
508                          * the LMA. */
509                         osd_lma_self_repair(info, osd, inode, rfid,
510                                             LMAC_FID_ON_OST);
511                         RETURN(0);
512                 }
513         }
514
515         if (unlikely(rc == -ENODATA))
516                 RETURN(0);
517
518         if (rc < 0)
519                 RETURN(rc);
520
521         if (rc > 0) {
522                 rc = 0;
523                 lustre_lma_swab(lma);
524                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
525                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
526                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
527                               "fid = "DFID", ino = %lu\n", osd_name(osd),
528                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
529                               PFID(rfid), inode->i_ino);
530                         rc = -EOPNOTSUPP;
531                 } else {
532                         fid = &lma->lma_self_fid;
533                 }
534         }
535
536         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
537                 __u32 level = D_LFSCK;
538
539                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
540                         struct ost_id   *oi   = &info->oti_ostid;
541                         struct lu_fid   *fid1 = &info->oti_fid3;
542                         __u32            idx  = fid_idif_ost_idx(rfid);
543
544                         /* For old IDIF, the OST index is not part of the IDIF,
545                          * Means that different OSTs may have the same IDIFs.
546                          * Under such case, we need to make some compatible
547                          * check to make sure to trigger OI scrub properly. */
548                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
549                                 /* Given @rfid is new, LMA is old. */
550                                 fid_to_ostid(fid, oi);
551                                 ostid_to_fid(fid1, oi, idx);
552                                 if (lu_fid_eq(fid1, rfid)) {
553                                         if (osd->od_index_in_idif)
554                                                 osd_lma_self_repair(info, osd,
555                                                         inode, rfid,
556                                                         LMAC_FID_ON_OST);
557                                         RETURN(0);
558                                 }
559                         }
560                 }
561
562
563                 rc = -EREMCHG;
564                 if (!thread_is_running(&osd->od_scrub.os_thread))
565                         level |= D_CONSOLE;
566
567                 CDEBUG(level, "%s: FID "DFID" != self_fid "DFID"\n",
568                        osd_name(osd), PFID(rfid), PFID(fid));
569         }
570
571         RETURN(rc);
572 }
573
574 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
575                           const struct lu_fid *fid,
576                           const struct lu_object_conf *conf)
577 {
578         struct osd_thread_info *info;
579         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
580         struct osd_device      *dev;
581         struct osd_idmap_cache *oic;
582         struct osd_inode_id    *id;
583         struct inode           *inode;
584         struct osd_scrub       *scrub;
585         struct scrub_file      *sf;
586         int                     result;
587         int                     saved  = 0;
588         bool                    in_oi  = false;
589         bool                    triggered = false;
590         ENTRY;
591
592         LINVRNT(osd_invariant(obj));
593         LASSERT(obj->oo_inode == NULL);
594         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
595
596         dev = osd_dev(ldev);
597         scrub = &dev->od_scrub;
598         sf = &scrub->os_file;
599         info = osd_oti_get(env);
600         LASSERT(info);
601         oic = &info->oti_cache;
602
603         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
604                 RETURN(-ENOENT);
605
606         /* For the object is created as locking anchor, or for the object to
607          * be created on disk. No need to osd_oi_lookup() at here because FID
608          * shouldn't never be re-used, if it's really a duplicate FID from
609          * unexpected reason, we should be able to detect it later by calling
610          * do_create->osd_oi_insert(). */
611         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
612                 GOTO(out, result = 0);
613
614         /* Search order: 1. per-thread cache. */
615         if (lu_fid_eq(fid, &oic->oic_fid) &&
616             likely(oic->oic_dev == dev)) {
617                 id = &oic->oic_lid;
618                 goto iget;
619         }
620
621         id = &info->oti_id;
622         if (!list_empty(&scrub->os_inconsistent_items)) {
623                 /* Search order: 2. OI scrub pending list. */
624                 result = osd_oii_lookup(dev, fid, id);
625                 if (result == 0)
626                         goto iget;
627         }
628
629         /* Search order: 3. OI files. */
630         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
631         if (result == -ENOENT) {
632                 if (!fid_is_norm(fid) ||
633                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
634                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
635                                       sf->sf_oi_bitmap))
636                         GOTO(out, result = 0);
637
638                 goto trigger;
639         }
640
641         if (result != 0)
642                 GOTO(out, result);
643
644         in_oi = true;
645
646 iget:
647         inode = osd_iget_check(info, dev, fid, id, in_oi);
648         if (IS_ERR(inode)) {
649                 result = PTR_ERR(inode);
650                 if (result == -ENOENT || result == -ESTALE) {
651                         if (!in_oi)
652                                 fid_zero(&oic->oic_fid);
653
654                         GOTO(out, result = -ENOENT);
655                 } else if (result == -EREMCHG) {
656
657 trigger:
658                         if (!in_oi)
659                                 fid_zero(&oic->oic_fid);
660
661                         if (unlikely(triggered))
662                                 GOTO(out, result = saved);
663
664                         triggered = true;
665                         if (thread_is_running(&scrub->os_thread)) {
666                                 result = -EINPROGRESS;
667                         } else if (!dev->od_noscrub) {
668                                 /* Since we do not know the right OI mapping,
669                                  * we have to trigger OI scrub to scan the
670                                  * whole device. */
671                                 result = osd_scrub_start(dev, SS_AUTO_FULL |
672                                         SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
673                                 CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI "
674                                        "scrub by RPC for "DFID", rc = %d [1]\n",
675                                        osd_name(dev), PFID(fid),result);
676                                 if (result == 0 || result == -EALREADY)
677                                         result = -EINPROGRESS;
678                                 else
679                                         result = -EREMCHG;
680                         }
681
682                         /* We still have chance to get the valid inode: for the
683                          * object which is referenced by remote name entry, the
684                          * object on the local MDT will be linked under the dir
685                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
686                          *
687                          * We do not know whether the object for the given FID
688                          * is referenced by some remote name entry or not, and
689                          * especially for DNE II, a multiple-linked object may
690                          * have many name entries reside on many MDTs.
691                          *
692                          * To simplify the operation, OSD will not distinguish
693                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
694                          * only happened for the RPC from other MDT during the
695                          * OI scrub, or for the client side RPC with FID only,
696                          * such as FID to path, or from old connected client. */
697                         saved = result;
698                         result = osd_lookup_in_remote_parent(info, dev,
699                                                              fid, id);
700                         if (result == 0) {
701                                 in_oi = false;
702                                 goto iget;
703                         }
704
705                         result = saved;
706                 }
707
708                 GOTO(out, result);
709         }
710
711         obj->oo_inode = inode;
712         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
713
714         result = osd_check_lma(env, obj);
715         if (result != 0) {
716                 iput(inode);
717                 obj->oo_inode = NULL;
718                 if (result == -EREMCHG) {
719                         if (!in_oi) {
720                                 result = osd_oi_lookup(info, dev, fid, id,
721                                                        OI_CHECK_FLD);
722                                 if (result != 0) {
723                                         fid_zero(&oic->oic_fid);
724                                         GOTO(out, result);
725                                 }
726                         }
727
728                         goto trigger;
729                 }
730
731                 GOTO(out, result);
732         }
733
734         obj->oo_compat_dot_created = 1;
735         obj->oo_compat_dotdot_created = 1;
736
737         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
738                 GOTO(out, result = 0);
739
740         LASSERT(obj->oo_hl_head == NULL);
741         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
742         if (obj->oo_hl_head == NULL) {
743                 obj->oo_inode = NULL;
744                 iput(inode);
745                 GOTO(out, result = -ENOMEM);
746         }
747         GOTO(out, result = 0);
748
749 out:
750         LINVRNT(osd_invariant(obj));
751         return result;
752 }
753
754 /*
755  * Concurrency: shouldn't matter.
756  */
757 static void osd_object_init0(struct osd_object *obj)
758 {
759         LASSERT(obj->oo_inode != NULL);
760         obj->oo_dt.do_body_ops = &osd_body_ops;
761         obj->oo_dt.do_lu.lo_header->loh_attr |=
762                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
763 }
764
765 /*
766  * Concurrency: no concurrent access is possible that early in object
767  * life-cycle.
768  */
769 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
770                            const struct lu_object_conf *conf)
771 {
772         struct osd_object *obj = osd_obj(l);
773         int result;
774
775         LINVRNT(osd_invariant(obj));
776
777         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
778                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
779                 l->lo_header->loh_attr |= LOHA_EXISTS;
780                 return 0;
781         }
782
783         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
784         obj->oo_dt.do_body_ops = &osd_body_ops_new;
785         if (result == 0 && obj->oo_inode != NULL)
786                 osd_object_init0(obj);
787
788         LINVRNT(osd_invariant(obj));
789         return result;
790 }
791
792 /*
793  * Concurrency: no concurrent access is possible that late in object
794  * life-cycle.
795  */
796 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
797 {
798         struct osd_object *obj = osd_obj(l);
799
800         LINVRNT(osd_invariant(obj));
801
802         dt_object_fini(&obj->oo_dt);
803         if (obj->oo_hl_head != NULL)
804                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
805         OBD_FREE_PTR(obj);
806 }
807
808 /*
809  * Concurrency: no concurrent access is possible that late in object
810  * life-cycle.
811  */
812 static void osd_index_fini(struct osd_object *o)
813 {
814         struct iam_container *bag;
815
816         if (o->oo_dir != NULL) {
817                 bag = &o->oo_dir->od_container;
818                 if (o->oo_inode != NULL) {
819                         if (bag->ic_object == o->oo_inode)
820                                 iam_container_fini(bag);
821                 }
822                 OBD_FREE_PTR(o->oo_dir);
823                 o->oo_dir = NULL;
824         }
825 }
826
827 /*
828  * Concurrency: no concurrent access is possible that late in object
829  * life-cycle (for all existing callers, that is. New callers have to provide
830  * their own locking.)
831  */
832 static int osd_inode_unlinked(const struct inode *inode)
833 {
834         return inode->i_nlink == 0;
835 }
836
837 enum {
838         OSD_TXN_OI_DELETE_CREDITS    = 20,
839         OSD_TXN_INODE_DELETE_CREDITS = 20
840 };
841
842 /*
843  * Journal
844  */
845
846 #if OSD_THANDLE_STATS
847 /**
848  * Set time when the handle is allocated
849  */
850 static void osd_th_alloced(struct osd_thandle *oth)
851 {
852         oth->oth_alloced = cfs_time_current();
853 }
854
855 /**
856  * Set time when the handle started
857  */
858 static void osd_th_started(struct osd_thandle *oth)
859 {
860         oth->oth_started = cfs_time_current();
861 }
862
863 /**
864  * Helper function to convert time interval to microseconds packed in
865  * long int.
866  */
867 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
868 {
869         struct timeval val;
870
871         cfs_duration_usec(cfs_time_sub(end, start), &val);
872         return val.tv_sec * 1000000 + val.tv_usec;
873 }
874
875 /**
876  * Check whether the we deal with this handle for too long.
877  */
878 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
879                                 cfs_time_t alloced, cfs_time_t started,
880                                 cfs_time_t closed)
881 {
882         cfs_time_t now = cfs_time_current();
883
884         LASSERT(dev != NULL);
885
886         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
887                             interval_to_usec(alloced, started));
888         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
889                             interval_to_usec(started, closed));
890         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
891                             interval_to_usec(closed, now));
892
893         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
894                 CWARN("transaction handle %p was open for too long: "
895                       "now "CFS_TIME_T" ,"
896                       "alloced "CFS_TIME_T" ,"
897                       "started "CFS_TIME_T" ,"
898                       "closed "CFS_TIME_T"\n",
899                       oth, now, alloced, started, closed);
900                 libcfs_debug_dumpstack(NULL);
901         }
902 }
903
904 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
905 {                                                                       \
906         cfs_time_t __closed = cfs_time_current();                       \
907         cfs_time_t __alloced = oth->oth_alloced;                        \
908         cfs_time_t __started = oth->oth_started;                        \
909                                                                         \
910         expr;                                                           \
911         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
912 }
913
914 #else /* OSD_THANDLE_STATS */
915
916 #define osd_th_alloced(h)                  do {} while(0)
917 #define osd_th_started(h)                  do {} while(0)
918 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
919
920 #endif /* OSD_THANDLE_STATS */
921
922 /*
923  * Concurrency: doesn't access mutable data.
924  */
925 static int osd_param_is_not_sane(const struct osd_device *dev,
926                                  const struct thandle *th)
927 {
928         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
929
930         return oh->ot_credits > osd_transaction_size(dev);
931 }
932
933 /*
934  * Concurrency: shouldn't matter.
935  */
936 static void osd_trans_commit_cb(struct super_block *sb,
937                                 struct ldiskfs_journal_cb_entry *jcb, int error)
938 {
939         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
940         struct thandle     *th  = &oh->ot_super;
941         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
942         struct dt_txn_commit_cb *dcb, *tmp;
943
944         LASSERT(oh->ot_handle == NULL);
945
946         if (error)
947                 CERROR("transaction @0x%p commit error: %d\n", th, error);
948
949         dt_txn_hook_commit(th);
950
951         /* call per-transaction callbacks if any */
952         list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
953                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
954                          "commit callback entry: magic=%x name='%s'\n",
955                          dcb->dcb_magic, dcb->dcb_name);
956                 list_del_init(&dcb->dcb_linkage);
957                 dcb->dcb_func(NULL, th, dcb, error);
958         }
959
960         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
961         lu_device_put(lud);
962         th->th_dev = NULL;
963
964         lu_context_exit(&th->th_ctx);
965         lu_context_fini(&th->th_ctx);
966         OBD_FREE_PTR(oh);
967 }
968
969 static struct thandle *osd_trans_create(const struct lu_env *env,
970                                         struct dt_device *d)
971 {
972         struct osd_thread_info  *oti = osd_oti_get(env);
973         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
974         struct osd_thandle      *oh;
975         struct thandle          *th;
976         ENTRY;
977
978         /* on pending IO in this thread should left from prev. request */
979         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
980
981         th = ERR_PTR(-ENOMEM);
982         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
983         if (oh != NULL) {
984                 oh->ot_quota_trans = &oti->oti_quota_trans;
985                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
986                 th = &oh->ot_super;
987                 th->th_dev = d;
988                 th->th_result = 0;
989                 th->th_tags = LCT_TX_HANDLE;
990                 oh->ot_credits = 0;
991                 INIT_LIST_HEAD(&oh->ot_dcb_list);
992                 osd_th_alloced(oh);
993
994                 memset(oti->oti_declare_ops, 0,
995                                         sizeof(oti->oti_declare_ops));
996                 memset(oti->oti_declare_ops_rb, 0,
997                                         sizeof(oti->oti_declare_ops_rb));
998                 memset(oti->oti_declare_ops_cred, 0,
999                                         sizeof(oti->oti_declare_ops_cred));
1000                 oti->oti_rollback = false;
1001         }
1002         RETURN(th);
1003 }
1004
1005 /*
1006  * Concurrency: shouldn't matter.
1007  */
1008 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1009                            struct thandle *th)
1010 {
1011         struct osd_thread_info *oti = osd_oti_get(env);
1012         struct osd_device  *dev = osd_dt_dev(d);
1013         handle_t           *jh;
1014         struct osd_thandle *oh;
1015         int rc;
1016
1017         ENTRY;
1018
1019         LASSERT(current->journal_info == NULL);
1020
1021         oh = container_of0(th, struct osd_thandle, ot_super);
1022         LASSERT(oh != NULL);
1023         LASSERT(oh->ot_handle == NULL);
1024
1025         rc = dt_txn_hook_start(env, d, th);
1026         if (rc != 0)
1027                 GOTO(out, rc);
1028
1029         if (unlikely(osd_param_is_not_sane(dev, th))) {
1030                 static unsigned long last_printed;
1031                 static int last_credits;
1032
1033                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1034                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1035                       oh->ot_credits,
1036                       osd_journal(dev)->j_max_transaction_buffers);
1037                 CWARN("  create: %u/%u, destroy: %u/%u\n",
1038                       oti->oti_declare_ops[OSD_OT_CREATE],
1039                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
1040                       oti->oti_declare_ops[OSD_OT_DESTROY],
1041                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1042                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
1043                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
1044                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1045                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
1046                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1047                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
1048                       oti->oti_declare_ops[OSD_OT_WRITE],
1049                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
1050                       oti->oti_declare_ops[OSD_OT_PUNCH],
1051                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1052                       oti->oti_declare_ops[OSD_OT_QUOTA],
1053                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1054                 CWARN("  insert: %u/%u, delete: %u/%u\n",
1055                       oti->oti_declare_ops[OSD_OT_INSERT],
1056                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
1057                       oti->oti_declare_ops[OSD_OT_DELETE],
1058                       oti->oti_declare_ops_cred[OSD_OT_DELETE]);
1059                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
1060                       oti->oti_declare_ops[OSD_OT_REF_ADD],
1061                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1062                       oti->oti_declare_ops[OSD_OT_REF_DEL],
1063                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1064
1065                 if (last_credits != oh->ot_credits &&
1066                     time_after(jiffies, last_printed +
1067                                msecs_to_jiffies(60 * MSEC_PER_SEC))) {
1068                         libcfs_debug_dumpstack(NULL);
1069                         last_credits = oh->ot_credits;
1070                         last_printed = jiffies;
1071                 }
1072                 /* XXX Limit the credits to 'max_transaction_buffers', and
1073                  *     let the underlying filesystem to catch the error if
1074                  *     we really need so many credits.
1075                  *
1076                  *     This should be removed when we can calculate the
1077                  *     credits precisely. */
1078                 oh->ot_credits = osd_transaction_size(dev);
1079         }
1080
1081         /*
1082          * XXX temporary stuff. Some abstraction layer should
1083          * be used.
1084          */
1085         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1086         osd_th_started(oh);
1087         if (!IS_ERR(jh)) {
1088                 oh->ot_handle = jh;
1089                 LASSERT(oti->oti_txns == 0);
1090                 lu_context_init(&th->th_ctx, th->th_tags);
1091                 lu_context_enter(&th->th_ctx);
1092
1093                 lu_device_get(&d->dd_lu_dev);
1094                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1095                               "osd-tx", th);
1096                 oti->oti_txns++;
1097                 rc = 0;
1098         } else {
1099                 rc = PTR_ERR(jh);
1100         }
1101 out:
1102         RETURN(rc);
1103 }
1104
1105 static int osd_seq_exists(const struct lu_env *env,
1106                           struct osd_device *osd, u64 seq)
1107 {
1108         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1109         struct seq_server_site  *ss = osd_seq_site(osd);
1110         int                     rc;
1111         ENTRY;
1112
1113         LASSERT(ss != NULL);
1114         LASSERT(ss->ss_server_fld != NULL);
1115
1116         rc = osd_fld_lookup(env, osd, seq, range);
1117         if (rc != 0) {
1118                 if (rc != -ENOENT)
1119                         CERROR("%s: can't lookup FLD sequence "LPX64
1120                                ": rc = %d\n", osd_name(osd), seq, rc);
1121                 RETURN(0);
1122         }
1123
1124         RETURN(ss->ss_node_id == range->lsr_index);
1125 }
1126
1127 /*
1128  * Concurrency: shouldn't matter.
1129  */
1130 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1131                           struct thandle *th)
1132 {
1133         int                     rc = 0;
1134         struct osd_thandle     *oh;
1135         struct osd_thread_info *oti = osd_oti_get(env);
1136         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1137         struct osd_device      *osd = osd_dt_dev(th->th_dev);
1138         struct qsd_instance    *qsd = osd->od_quota_slave;
1139         struct lquota_trans    *qtrans;
1140         ENTRY;
1141
1142         oh = container_of0(th, struct osd_thandle, ot_super);
1143
1144         qtrans = oh->ot_quota_trans;
1145         oh->ot_quota_trans = NULL;
1146
1147         if (oh->ot_handle != NULL) {
1148                 handle_t *hdl = oh->ot_handle;
1149
1150                 /*
1151                  * add commit callback
1152                  * notice we don't do this in osd_trans_start()
1153                  * as underlying transaction can change during truncate
1154                  */
1155                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1156                                          &oh->ot_jcb);
1157
1158                 LASSERT(oti->oti_txns == 1);
1159                 oti->oti_txns--;
1160                 rc = dt_txn_hook_stop(env, th);
1161                 if (rc != 0)
1162                         CERROR("%s: failed in transaction hook: rc = %d\n",
1163                                osd_name(osd), rc);
1164
1165                 /* hook functions might modify th_sync */
1166                 hdl->h_sync = th->th_sync;
1167
1168                 oh->ot_handle = NULL;
1169                 OSD_CHECK_SLOW_TH(oh, osd, rc = ldiskfs_journal_stop(hdl));
1170                 if (rc != 0)
1171                         CERROR("%s: failed to stop transaction: rc = %d\n",
1172                                osd_name(osd), rc);
1173         } else {
1174                 OBD_FREE_PTR(oh);
1175         }
1176
1177         /* inform the quota slave device that the transaction is stopping */
1178         qsd_op_end(env, qsd, qtrans);
1179
1180         /* as we want IO to journal and data IO be concurrent, we don't block
1181          * awaiting data IO completion in osd_do_bio(), instead we wait here
1182          * once transaction is submitted to the journal. all reqular requests
1183          * don't do direct IO (except read/write), thus this wait_event becomes
1184          * no-op for them.
1185          *
1186          * IMPORTANT: we have to wait till any IO submited by the thread is
1187          * completed otherwise iobuf may be corrupted by different request
1188          */
1189         wait_event(iobuf->dr_wait,
1190                        atomic_read(&iobuf->dr_numreqs) == 0);
1191         osd_fini_iobuf(osd, iobuf);
1192         if (!rc)
1193                 rc = iobuf->dr_error;
1194
1195         RETURN(rc);
1196 }
1197
1198 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1199 {
1200         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1201                                                ot_super);
1202
1203         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1204         LASSERT(&dcb->dcb_func != NULL);
1205         list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1206
1207         return 0;
1208 }
1209
1210 /*
1211  * Called just before object is freed. Releases all resources except for
1212  * object itself (that is released by osd_object_free()).
1213  *
1214  * Concurrency: no concurrent access is possible that late in object
1215  * life-cycle.
1216  */
1217 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1218 {
1219         struct osd_object *obj   = osd_obj(l);
1220         struct inode      *inode = obj->oo_inode;
1221
1222         LINVRNT(osd_invariant(obj));
1223
1224         /*
1225          * If object is unlinked remove fid->ino mapping from object index.
1226          */
1227
1228         osd_index_fini(obj);
1229         if (inode != NULL) {
1230                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1231                 qid_t                    uid = i_uid_read(inode);
1232                 qid_t                    gid = i_gid_read(inode);
1233
1234                 iput(inode);
1235                 obj->oo_inode = NULL;
1236
1237                 if (qsd != NULL) {
1238                         struct osd_thread_info  *info = osd_oti_get(env);
1239                         struct lquota_id_info   *qi = &info->oti_qi;
1240
1241                         /* Release granted quota to master if necessary */
1242                         qi->lqi_id.qid_uid = uid;
1243                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1244
1245                         qi->lqi_id.qid_uid = gid;
1246                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1247                 }
1248         }
1249 }
1250
1251 /*
1252  * Concurrency: ->loo_object_release() is called under site spin-lock.
1253  */
1254 static void osd_object_release(const struct lu_env *env,
1255                                struct lu_object *l)
1256 {
1257 }
1258
1259 /*
1260  * Concurrency: shouldn't matter.
1261  */
1262 static int osd_object_print(const struct lu_env *env, void *cookie,
1263                             lu_printer_t p, const struct lu_object *l)
1264 {
1265         struct osd_object *o = osd_obj(l);
1266         struct iam_descr  *d;
1267
1268         if (o->oo_dir != NULL)
1269                 d = o->oo_dir->od_container.ic_descr;
1270         else
1271                 d = NULL;
1272         return (*p)(env, cookie,
1273                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1274                     o, o->oo_inode,
1275                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1276                     o->oo_inode ? o->oo_inode->i_generation : 0,
1277                     d ? d->id_ops->id_name : "plain");
1278 }
1279
1280 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1281
1282 /*
1283  * Concurrency: shouldn't matter.
1284  */
1285 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1286                struct obd_statfs *sfs)
1287 {
1288         struct osd_device  *osd = osd_dt_dev(d);
1289         struct super_block *sb = osd_sb(osd);
1290         struct kstatfs     *ksfs;
1291         int result = 0;
1292
1293         if (unlikely(osd->od_mnt == NULL))
1294                 return -EINPROGRESS;
1295
1296         /* osd_lproc.c call this without env, allocate ksfs for that case */
1297         if (unlikely(env == NULL)) {
1298                 OBD_ALLOC_PTR(ksfs);
1299                 if (ksfs == NULL)
1300                         return -ENOMEM;
1301         } else {
1302                 ksfs = &osd_oti_get(env)->oti_ksfs;
1303         }
1304
1305         spin_lock(&osd->od_osfs_lock);
1306         result = sb->s_op->statfs(sb->s_root, ksfs);
1307         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1308                 statfs_pack(sfs, ksfs);
1309                 if (sb->s_flags & MS_RDONLY)
1310                         sfs->os_state = OS_STATE_READONLY;
1311         }
1312
1313         spin_unlock(&osd->od_osfs_lock);
1314
1315         if (unlikely(env == NULL))
1316                 OBD_FREE_PTR(ksfs);
1317
1318         /* Reserve a small amount of space for local objects like last_rcvd,
1319          * llog, quota files, ... */
1320         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1321                 sfs->os_bavail = 0;
1322         } else {
1323                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1324                 /** Take out metadata overhead for indirect blocks */
1325                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1326         }
1327
1328         return result;
1329 }
1330
1331 /**
1332  * Estimate space needed for file creations. We assume the largest filename
1333  * which is 2^64 - 1, hence a filename of 20 chars.
1334  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1335  */
1336 #ifdef __LDISKFS_DIR_REC_LEN
1337 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1338 #else
1339 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1340 #endif
1341
1342 /*
1343  * Concurrency: doesn't access mutable data.
1344  */
1345 static void osd_conf_get(const struct lu_env *env,
1346                          const struct dt_device *dev,
1347                          struct dt_device_param *param)
1348 {
1349         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1350         int                ea_overhead;
1351
1352         /*
1353          * XXX should be taken from not-yet-existing fs abstraction layer.
1354          */
1355         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1356         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1357         param->ddp_block_shift  = sb->s_blocksize_bits;
1358         param->ddp_mount_type     = LDD_MT_LDISKFS;
1359         param->ddp_maxbytes       = sb->s_maxbytes;
1360         /* Overhead estimate should be fairly accurate, so we really take a tiny
1361          * error margin which also avoids fragmenting the filesystem too much */
1362         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1363         /* inode are statically allocated, so per-inode space consumption
1364          * is the space consumed by the directory entry */
1365         param->ddp_inodespace     = PER_OBJ_USAGE;
1366         /* per-fragment overhead to be used by the client code */
1367         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1368         param->ddp_mntopts      = 0;
1369         if (test_opt(sb, XATTR_USER))
1370                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1371         if (test_opt(sb, POSIX_ACL))
1372                 param->ddp_mntopts |= MNTOPT_ACL;
1373
1374         /* LOD might calculate the max stripe count based on max_ea_size,
1375          * so we need take account in the overhead as well,
1376          * xattr_header + magic + xattr_entry_head */
1377         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1378                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1379
1380 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1381         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1382                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1383                                                                 ea_overhead;
1384         else
1385 #endif
1386                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1387 }
1388
1389 /*
1390  * Concurrency: shouldn't matter.
1391  */
1392 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1393 {
1394         int rc;
1395
1396         CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1397
1398         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1399
1400         CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
1401                LUSTRE_OSD_LDISKFS_NAME, rc);
1402
1403         return rc;
1404 }
1405
1406 /**
1407  * Start commit for OSD device.
1408  *
1409  * An implementation of dt_commit_async method for OSD device.
1410  * Asychronously starts underlayng fs sync and thereby a transaction
1411  * commit.
1412  *
1413  * \param env environment
1414  * \param d dt device
1415  *
1416  * \see dt_device_operations
1417  */
1418 static int osd_commit_async(const struct lu_env *env,
1419                             struct dt_device *d)
1420 {
1421         struct super_block *s = osd_sb(osd_dt_dev(d));
1422         ENTRY;
1423
1424         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1425         RETURN(s->s_op->sync_fs(s, 0));
1426 }
1427
1428 /*
1429  * Concurrency: shouldn't matter.
1430  */
1431
1432 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1433 {
1434         struct super_block *sb = osd_sb(osd_dt_dev(d));
1435         struct block_device *dev = sb->s_bdev;
1436 #ifdef HAVE_DEV_SET_RDONLY
1437         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1438         int rc = 0;
1439 #else
1440         int rc = -EOPNOTSUPP;
1441 #endif
1442         ENTRY;
1443
1444 #ifdef HAVE_DEV_SET_RDONLY
1445         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1446
1447         if (jdev && (jdev != dev)) {
1448                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1449                        (long)jdev);
1450                 dev_set_rdonly(jdev);
1451         }
1452         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1453         dev_set_rdonly(dev);
1454 #else
1455         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1456                osd_dt_dev(d)->od_svname, (long)dev, rc);
1457 #endif
1458         RETURN(rc);
1459 }
1460
1461 /**
1462  * Note: we do not count into QUOTA here.
1463  * If we mount with --data_journal we may need more.
1464  */
1465 const int osd_dto_credits_noquota[DTO_NR] = {
1466         /**
1467          * Insert.
1468          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1469          * SINGLEDATA_TRANS_BLOCKS(8)
1470          * XXX Note: maybe iam need more, since iam have more level than
1471          *           EXT3 htree.
1472          */
1473         [DTO_INDEX_INSERT]  = 16,
1474         /**
1475          * Delete
1476          * just modify a single entry, probably merge few within a block
1477          */
1478         [DTO_INDEX_DELETE]  = 1,
1479         /**
1480          * Used for OI scrub
1481          */
1482         [DTO_INDEX_UPDATE]  = 16,
1483         /**
1484          * 4(inode, inode bits, groups, GDT)
1485          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1486          */
1487         [DTO_OBJECT_CREATE] = 4,
1488         /**
1489          * 4(inode, inode bits, groups, GDT)
1490          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1491          */
1492         [DTO_OBJECT_DELETE] = 4,
1493         /**
1494          * Attr set credits (inode)
1495          */
1496         [DTO_ATTR_SET_BASE] = 1,
1497         /**
1498          * Xattr set. The same as xattr of EXT3.
1499          * DATA_TRANS_BLOCKS(14)
1500          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1501          * are also counted in. Do not know why?
1502          */
1503         [DTO_XATTR_SET]     = 14,
1504         /**
1505          * credits for inode change during write.
1506          */
1507         [DTO_WRITE_BASE]    = 3,
1508         /**
1509          * credits for single block write.
1510          */
1511         [DTO_WRITE_BLOCK]   = 14,
1512         /**
1513          * Attr set credits for chown.
1514          * This is extra credits for setattr, and it is null without quota
1515          */
1516         [DTO_ATTR_SET_CHOWN] = 0
1517 };
1518
1519 static const struct dt_device_operations osd_dt_ops = {
1520         .dt_root_get       = osd_root_get,
1521         .dt_statfs         = osd_statfs,
1522         .dt_trans_create   = osd_trans_create,
1523         .dt_trans_start    = osd_trans_start,
1524         .dt_trans_stop     = osd_trans_stop,
1525         .dt_trans_cb_add   = osd_trans_cb_add,
1526         .dt_conf_get       = osd_conf_get,
1527         .dt_sync           = osd_sync,
1528         .dt_ro             = osd_ro,
1529         .dt_commit_async   = osd_commit_async,
1530 };
1531
1532 static void osd_object_read_lock(const struct lu_env *env,
1533                                  struct dt_object *dt, unsigned role)
1534 {
1535         struct osd_object *obj = osd_dt_obj(dt);
1536         struct osd_thread_info *oti = osd_oti_get(env);
1537
1538         LINVRNT(osd_invariant(obj));
1539
1540         LASSERT(obj->oo_owner != env);
1541         down_read_nested(&obj->oo_sem, role);
1542
1543         LASSERT(obj->oo_owner == NULL);
1544         oti->oti_r_locks++;
1545 }
1546
1547 static void osd_object_write_lock(const struct lu_env *env,
1548                                   struct dt_object *dt, unsigned role)
1549 {
1550         struct osd_object *obj = osd_dt_obj(dt);
1551         struct osd_thread_info *oti = osd_oti_get(env);
1552
1553         LINVRNT(osd_invariant(obj));
1554
1555         LASSERT(obj->oo_owner != env);
1556         down_write_nested(&obj->oo_sem, role);
1557
1558         LASSERT(obj->oo_owner == NULL);
1559         obj->oo_owner = env;
1560         oti->oti_w_locks++;
1561 }
1562
1563 static void osd_object_read_unlock(const struct lu_env *env,
1564                                    struct dt_object *dt)
1565 {
1566         struct osd_object *obj = osd_dt_obj(dt);
1567         struct osd_thread_info *oti = osd_oti_get(env);
1568
1569         LINVRNT(osd_invariant(obj));
1570
1571         LASSERT(oti->oti_r_locks > 0);
1572         oti->oti_r_locks--;
1573         up_read(&obj->oo_sem);
1574 }
1575
1576 static void osd_object_write_unlock(const struct lu_env *env,
1577                                     struct dt_object *dt)
1578 {
1579         struct osd_object *obj = osd_dt_obj(dt);
1580         struct osd_thread_info *oti = osd_oti_get(env);
1581
1582         LINVRNT(osd_invariant(obj));
1583
1584         LASSERT(obj->oo_owner == env);
1585         LASSERT(oti->oti_w_locks > 0);
1586         oti->oti_w_locks--;
1587         obj->oo_owner = NULL;
1588         up_write(&obj->oo_sem);
1589 }
1590
1591 static int osd_object_write_locked(const struct lu_env *env,
1592                                    struct dt_object *dt)
1593 {
1594         struct osd_object *obj = osd_dt_obj(dt);
1595
1596         LINVRNT(osd_invariant(obj));
1597
1598         return obj->oo_owner == env;
1599 }
1600
1601 static struct timespec *osd_inode_time(const struct lu_env *env,
1602                                        struct inode *inode, __u64 seconds)
1603 {
1604         struct osd_thread_info  *oti = osd_oti_get(env);
1605         struct timespec         *t   = &oti->oti_time;
1606
1607         t->tv_sec = seconds;
1608         t->tv_nsec = 0;
1609         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1610         return t;
1611 }
1612
1613
1614 static void osd_inode_getattr(const struct lu_env *env,
1615                               struct inode *inode, struct lu_attr *attr)
1616 {
1617         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1618                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1619                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1620                            LA_TYPE;
1621
1622         attr->la_atime   = LTIME_S(inode->i_atime);
1623         attr->la_mtime   = LTIME_S(inode->i_mtime);
1624         attr->la_ctime   = LTIME_S(inode->i_ctime);
1625         attr->la_mode    = inode->i_mode;
1626         attr->la_size    = i_size_read(inode);
1627         attr->la_blocks  = inode->i_blocks;
1628         attr->la_uid     = i_uid_read(inode);
1629         attr->la_gid     = i_gid_read(inode);
1630         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1631         attr->la_nlink   = inode->i_nlink;
1632         attr->la_rdev    = inode->i_rdev;
1633         attr->la_blksize = 1 << inode->i_blkbits;
1634         attr->la_blkbits = inode->i_blkbits;
1635 }
1636
1637 static int osd_attr_get(const struct lu_env *env,
1638                         struct dt_object *dt,
1639                         struct lu_attr *attr)
1640 {
1641         struct osd_object *obj = osd_dt_obj(dt);
1642
1643         if (!dt_object_exists(dt))
1644                 return -ENOENT;
1645
1646         LASSERT(!dt_object_remote(dt));
1647         LINVRNT(osd_invariant(obj));
1648
1649         spin_lock(&obj->oo_guard);
1650         osd_inode_getattr(env, obj->oo_inode, attr);
1651         spin_unlock(&obj->oo_guard);
1652         return 0;
1653 }
1654
1655 static int osd_declare_attr_set(const struct lu_env *env,
1656                                 struct dt_object *dt,
1657                                 const struct lu_attr *attr,
1658                                 struct thandle *handle)
1659 {
1660         struct osd_thandle     *oh;
1661         struct osd_object      *obj;
1662         struct osd_thread_info *info = osd_oti_get(env);
1663         struct lquota_id_info  *qi = &info->oti_qi;
1664         qid_t                   uid;
1665         qid_t                   gid;
1666         long long               bspace;
1667         int                     rc = 0;
1668         bool                    enforce;
1669         ENTRY;
1670
1671         LASSERT(dt != NULL);
1672         LASSERT(handle != NULL);
1673
1674         obj = osd_dt_obj(dt);
1675         LASSERT(osd_invariant(obj));
1676
1677         oh = container_of0(handle, struct osd_thandle, ot_super);
1678         LASSERT(oh->ot_handle == NULL);
1679
1680         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1681                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1682
1683         if (attr == NULL || obj->oo_inode == NULL)
1684                 RETURN(rc);
1685
1686         bspace   = obj->oo_inode->i_blocks;
1687         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1688         bspace   = toqb(bspace);
1689
1690         /* Changing ownership is always preformed by super user, it should not
1691          * fail with EDQUOT.
1692          *
1693          * We still need to call the osd_declare_qid() to calculate the journal
1694          * credits for updating quota accounting files and to trigger quota
1695          * space adjustment once the operation is completed.*/
1696         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1697                 /* USERQUOTA */
1698                 uid = i_uid_read(obj->oo_inode);
1699                 qi->lqi_type = USRQUOTA;
1700                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1701                 /* inode accounting */
1702                 qi->lqi_is_blk = false;
1703
1704                 /* one more inode for the new uid ... */
1705                 qi->lqi_id.qid_uid = attr->la_uid;
1706                 qi->lqi_space      = 1;
1707                 /* Reserve credits for the new uid */
1708                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1709                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1710                         rc = 0;
1711                 if (rc)
1712                         RETURN(rc);
1713
1714                 /* and one less inode for the current uid */
1715                 qi->lqi_id.qid_uid = uid;
1716                 qi->lqi_space      = -1;
1717                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1718                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1719                         rc = 0;
1720                 if (rc)
1721                         RETURN(rc);
1722
1723                 /* block accounting */
1724                 qi->lqi_is_blk = true;
1725
1726                 /* more blocks for the new uid ... */
1727                 qi->lqi_id.qid_uid = attr->la_uid;
1728                 qi->lqi_space      = bspace;
1729                 /*
1730                  * Credits for the new uid has been reserved, re-use "obj"
1731                  * to save credit reservation.
1732                  */
1733                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1734                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1735                         rc = 0;
1736                 if (rc)
1737                         RETURN(rc);
1738
1739                 /* and finally less blocks for the current uid */
1740                 qi->lqi_id.qid_uid = uid;
1741                 qi->lqi_space      = -bspace;
1742                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1743                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1744                         rc = 0;
1745                 if (rc)
1746                         RETURN(rc);
1747
1748                 /* GROUP QUOTA */
1749                 gid = i_gid_read(obj->oo_inode);
1750                 qi->lqi_type = GRPQUOTA;
1751                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1752
1753                 /* inode accounting */
1754                 qi->lqi_is_blk = false;
1755
1756                 /* one more inode for the new gid ... */
1757                 qi->lqi_id.qid_gid = attr->la_gid;
1758                 qi->lqi_space      = 1;
1759                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1760                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1761                         rc = 0;
1762                 if (rc)
1763                         RETURN(rc);
1764
1765                 /* and one less inode for the current gid */
1766                 qi->lqi_id.qid_gid = gid;
1767                 qi->lqi_space      = -1;
1768                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1769                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1770                         rc = 0;
1771                 if (rc)
1772                         RETURN(rc);
1773
1774                 /* block accounting */
1775                 qi->lqi_is_blk = true;
1776
1777                 /* more blocks for the new gid ... */
1778                 qi->lqi_id.qid_gid = attr->la_gid;
1779                 qi->lqi_space      = bspace;
1780                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1781                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1782                         rc = 0;
1783                 if (rc)
1784                         RETURN(rc);
1785
1786                 /* and finally less blocks for the current gid */
1787                 qi->lqi_id.qid_gid = gid;
1788                 qi->lqi_space      = -bspace;
1789                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1790                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1791                         rc = 0;
1792                 if (rc)
1793                         RETURN(rc);
1794         }
1795
1796         RETURN(rc);
1797 }
1798
1799 static int osd_inode_setattr(const struct lu_env *env,
1800                              struct inode *inode, const struct lu_attr *attr)
1801 {
1802         __u64 bits = attr->la_valid;
1803
1804         /* Only allow set size for regular file */
1805         if (!S_ISREG(inode->i_mode))
1806                 bits &= ~(LA_SIZE | LA_BLOCKS);
1807
1808         if (bits == 0)
1809                 return 0;
1810
1811         if (bits & LA_ATIME)
1812                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1813         if (bits & LA_CTIME)
1814                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1815         if (bits & LA_MTIME)
1816                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1817         if (bits & LA_SIZE) {
1818                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1819                 i_size_write(inode, attr->la_size);
1820         }
1821
1822 #if 0
1823         /* OSD should not change "i_blocks" which is used by quota.
1824          * "i_blocks" should be changed by ldiskfs only. */
1825         if (bits & LA_BLOCKS)
1826                 inode->i_blocks = attr->la_blocks;
1827 #endif
1828         if (bits & LA_MODE)
1829                 inode->i_mode = (inode->i_mode & S_IFMT) |
1830                                 (attr->la_mode & ~S_IFMT);
1831         if (bits & LA_UID)
1832                 i_uid_write(inode, attr->la_uid);
1833         if (bits & LA_GID)
1834                 i_gid_write(inode, attr->la_gid);
1835         if (bits & LA_NLINK)
1836                 set_nlink(inode, attr->la_nlink);
1837         if (bits & LA_RDEV)
1838                 inode->i_rdev = attr->la_rdev;
1839
1840         if (bits & LA_FLAGS) {
1841                 /* always keep S_NOCMTIME */
1842                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1843                                  S_NOCMTIME;
1844         }
1845         return 0;
1846 }
1847
1848 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1849 {
1850         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1851             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1852                 struct iattr    iattr;
1853                 int             rc;
1854
1855                 ll_vfs_dq_init(inode);
1856                 iattr.ia_valid = 0;
1857                 if (attr->la_valid & LA_UID)
1858                         iattr.ia_valid |= ATTR_UID;
1859                 if (attr->la_valid & LA_GID)
1860                         iattr.ia_valid |= ATTR_GID;
1861                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1862                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1863
1864                 rc = ll_vfs_dq_transfer(inode, &iattr);
1865                 if (rc) {
1866                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1867                                "enforcement enabled on the ldiskfs "
1868                                "filesystem?\n", inode->i_sb->s_id, rc);
1869                         return rc;
1870                 }
1871         }
1872         return 0;
1873 }
1874
1875 static int osd_attr_set(const struct lu_env *env,
1876                         struct dt_object *dt,
1877                         const struct lu_attr *attr,
1878                         struct thandle *handle)
1879 {
1880         struct osd_object *obj = osd_dt_obj(dt);
1881         struct inode      *inode;
1882         int rc;
1883
1884         if (!dt_object_exists(dt))
1885                 return -ENOENT;
1886
1887         LASSERT(handle != NULL);
1888         LASSERT(!dt_object_remote(dt));
1889         LASSERT(osd_invariant(obj));
1890
1891         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1892
1893         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1894                 struct osd_thread_info  *oti  = osd_oti_get(env);
1895                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1896                 struct lu_fid           *fid1 = &oti->oti_fid;
1897                 struct osd_inode_id     *id   = &oti->oti_id;
1898                 struct iam_path_descr   *ipd;
1899                 struct iam_container    *bag;
1900                 struct osd_thandle      *oh;
1901                 int                      rc;
1902
1903                 fid_cpu_to_be(fid1, fid0);
1904                 memset(id, 1, sizeof(*id));
1905                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1906                                   fid0)->oi_dir.od_container;
1907                 ipd = osd_idx_ipd_get(env, bag);
1908                 if (unlikely(ipd == NULL))
1909                         RETURN(-ENOMEM);
1910
1911                 oh = container_of0(handle, struct osd_thandle, ot_super);
1912                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1913                                 (const struct iam_rec *)id, ipd);
1914                 osd_ipd_put(env, bag, ipd);
1915                 return(rc > 0 ? 0 : rc);
1916         }
1917
1918         inode = obj->oo_inode;
1919
1920         rc = osd_quota_transfer(inode, attr);
1921         if (rc)
1922                 return rc;
1923
1924         spin_lock(&obj->oo_guard);
1925         rc = osd_inode_setattr(env, inode, attr);
1926         spin_unlock(&obj->oo_guard);
1927
1928         if (!rc)
1929                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1930         return rc;
1931 }
1932
1933 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
1934                                            struct osd_object *obj,
1935                                            const char *name, const int namelen)
1936 {
1937         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1938 }
1939
1940 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1941                       umode_t mode, struct dt_allocation_hint *hint,
1942                       struct thandle *th)
1943 {
1944         int result;
1945         struct osd_device  *osd = osd_obj2dev(obj);
1946         struct osd_thandle *oth;
1947         struct dt_object   *parent = NULL;
1948         struct inode       *inode;
1949
1950         LINVRNT(osd_invariant(obj));
1951         LASSERT(obj->oo_inode == NULL);
1952         LASSERT(obj->oo_hl_head == NULL);
1953
1954         if (S_ISDIR(mode) && ldiskfs_pdo) {
1955                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1956                 if (obj->oo_hl_head == NULL)
1957                         return -ENOMEM;
1958         }
1959
1960         oth = container_of(th, struct osd_thandle, ot_super);
1961         LASSERT(oth->ot_handle->h_transaction != NULL);
1962
1963         if (hint && hint->dah_parent)
1964                 parent = hint->dah_parent;
1965
1966         inode = ldiskfs_create_inode(oth->ot_handle,
1967                                      parent ? osd_dt_obj(parent)->oo_inode :
1968                                               osd_sb(osd)->s_root->d_inode,
1969                                      mode);
1970         if (!IS_ERR(inode)) {
1971                 /* Do not update file c/mtime in ldiskfs. */
1972                 inode->i_flags |= S_NOCMTIME;
1973
1974                 /* For new created object, it must be consistent,
1975                  * and it is unnecessary to scrub against it. */
1976                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
1977
1978                 obj->oo_inode = inode;
1979                 result = 0;
1980         } else {
1981                 if (obj->oo_hl_head != NULL) {
1982                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1983                         obj->oo_hl_head = NULL;
1984                 }
1985                 result = PTR_ERR(inode);
1986         }
1987         LINVRNT(osd_invariant(obj));
1988         return result;
1989 }
1990
1991 enum {
1992         OSD_NAME_LEN = 255
1993 };
1994
1995 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1996                      struct lu_attr *attr,
1997                      struct dt_allocation_hint *hint,
1998                      struct dt_object_format *dof,
1999                      struct thandle *th)
2000 {
2001         int result;
2002         struct osd_thandle *oth;
2003         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2004
2005         LASSERT(S_ISDIR(attr->la_mode));
2006
2007         oth = container_of(th, struct osd_thandle, ot_super);
2008         LASSERT(oth->ot_handle->h_transaction != NULL);
2009         result = osd_mkfile(info, obj, mode, hint, th);
2010
2011         return result;
2012 }
2013
2014 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2015                         struct lu_attr *attr,
2016                         struct dt_allocation_hint *hint,
2017                         struct dt_object_format *dof,
2018                         struct thandle *th)
2019 {
2020         int result;
2021         struct osd_thandle *oth;
2022         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2023
2024         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2025
2026         LASSERT(S_ISREG(attr->la_mode));
2027
2028         oth = container_of(th, struct osd_thandle, ot_super);
2029         LASSERT(oth->ot_handle->h_transaction != NULL);
2030
2031         result = osd_mkfile(info, obj, mode, hint, th);
2032         if (result == 0) {
2033                 LASSERT(obj->oo_inode != NULL);
2034                 if (feat->dif_flags & DT_IND_VARKEY)
2035                         result = iam_lvar_create(obj->oo_inode,
2036                                                  feat->dif_keysize_max,
2037                                                  feat->dif_ptrsize,
2038                                                  feat->dif_recsize_max,
2039                                                  oth->ot_handle);
2040                 else
2041                         result = iam_lfix_create(obj->oo_inode,
2042                                                  feat->dif_keysize_max,
2043                                                  feat->dif_ptrsize,
2044                                                  feat->dif_recsize_max,
2045                                                  oth->ot_handle);
2046
2047         }
2048         return result;
2049 }
2050
2051 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2052                      struct lu_attr *attr,
2053                      struct dt_allocation_hint *hint,
2054                      struct dt_object_format *dof,
2055                      struct thandle *th)
2056 {
2057         LASSERT(S_ISREG(attr->la_mode));
2058         return osd_mkfile(info, obj, (attr->la_mode &
2059                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2060 }
2061
2062 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2063                      struct lu_attr *attr,
2064                      struct dt_allocation_hint *hint,
2065                      struct dt_object_format *dof,
2066                      struct thandle *th)
2067 {
2068         LASSERT(S_ISLNK(attr->la_mode));
2069         return osd_mkfile(info, obj, (attr->la_mode &
2070                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2071 }
2072
2073 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2074                      struct lu_attr *attr,
2075                      struct dt_allocation_hint *hint,
2076                      struct dt_object_format *dof,
2077                      struct thandle *th)
2078 {
2079         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2080         int result;
2081
2082         LINVRNT(osd_invariant(obj));
2083         LASSERT(obj->oo_inode == NULL);
2084         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2085                 S_ISFIFO(mode) || S_ISSOCK(mode));
2086
2087         result = osd_mkfile(info, obj, mode, hint, th);
2088         if (result == 0) {
2089                 LASSERT(obj->oo_inode != NULL);
2090                 /*
2091                  * This inode should be marked dirty for i_rdev.  Currently
2092                  * that is done in the osd_attr_init().
2093                  */
2094                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2095                                    attr->la_rdev);
2096         }
2097         LINVRNT(osd_invariant(obj));
2098         return result;
2099 }
2100
2101 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2102                               struct lu_attr *,
2103                               struct dt_allocation_hint *hint,
2104                               struct dt_object_format *dof,
2105                               struct thandle *);
2106
2107 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2108 {
2109         osd_obj_type_f result;
2110
2111         switch (type) {
2112         case DFT_DIR:
2113                 result = osd_mkdir;
2114                 break;
2115         case DFT_REGULAR:
2116                 result = osd_mkreg;
2117                 break;
2118         case DFT_SYM:
2119                 result = osd_mksym;
2120                 break;
2121         case DFT_NODE:
2122                 result = osd_mknod;
2123                 break;
2124         case DFT_INDEX:
2125                 result = osd_mk_index;
2126                 break;
2127
2128         default:
2129                 LBUG();
2130                 break;
2131         }
2132         return result;
2133 }
2134
2135
2136 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2137                         struct dt_object *parent, struct dt_object *child,
2138                         umode_t child_mode)
2139 {
2140         LASSERT(ah);
2141
2142         ah->dah_parent = parent;
2143         ah->dah_mode = child_mode;
2144 }
2145
2146 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2147                           struct lu_attr *attr, struct dt_object_format *dof)
2148 {
2149         struct inode   *inode = obj->oo_inode;
2150         __u64           valid = attr->la_valid;
2151         int             result;
2152
2153         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2154
2155         if (dof->dof_type != DFT_NODE)
2156                 attr->la_valid &= ~LA_RDEV;
2157         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2158                 attr->la_valid &= ~LA_ATIME;
2159         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2160                 attr->la_valid &= ~LA_CTIME;
2161         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2162                 attr->la_valid &= ~LA_MTIME;
2163
2164         result = osd_quota_transfer(inode, attr);
2165         if (result)
2166                 return;
2167
2168         if (attr->la_valid != 0) {
2169                 result = osd_inode_setattr(info->oti_env, inode, attr);
2170                 /*
2171                  * The osd_inode_setattr() should always succeed here.  The
2172                  * only error that could be returned is EDQUOT when we are
2173                  * trying to change the UID or GID of the inode. However, this
2174                  * should not happen since quota enforcement is no longer
2175                  * enabled on ldiskfs (lquota takes care of it).
2176                  */
2177                 LASSERTF(result == 0, "%d\n", result);
2178                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2179         }
2180
2181         attr->la_valid = valid;
2182 }
2183
2184 /**
2185  * Helper function for osd_object_create()
2186  *
2187  * \retval 0, on success
2188  */
2189 static int __osd_object_create(struct osd_thread_info *info,
2190                                struct osd_object *obj, struct lu_attr *attr,
2191                                struct dt_allocation_hint *hint,
2192                                struct dt_object_format *dof,
2193                                struct thandle *th)
2194 {
2195         int     result;
2196         __u32   umask;
2197
2198         /* we drop umask so that permissions we pass are not affected */
2199         umask = current->fs->umask;
2200         current->fs->umask = 0;
2201
2202         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2203                                                   th);
2204         if (result == 0) {
2205                 osd_attr_init(info, obj, attr, dof);
2206                 osd_object_init0(obj);
2207         }
2208
2209         if (obj->oo_inode != NULL) {
2210                 LASSERT(obj->oo_inode->i_state & I_NEW);
2211
2212                 unlock_new_inode(obj->oo_inode);
2213         }
2214
2215         /* restore previous umask value */
2216         current->fs->umask = umask;
2217
2218         return result;
2219 }
2220
2221 /**
2222  * Helper function for osd_object_create()
2223  *
2224  * \retval 0, on success
2225  */
2226 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2227                            const struct lu_fid *fid, struct thandle *th)
2228 {
2229         struct osd_thread_info *info = osd_oti_get(env);
2230         struct osd_inode_id    *id   = &info->oti_id;
2231         struct osd_device      *osd  = osd_obj2dev(obj);
2232         struct osd_thandle     *oh;
2233
2234         LASSERT(obj->oo_inode != NULL);
2235
2236         oh = container_of0(th, struct osd_thandle, ot_super);
2237         LASSERT(oh->ot_handle);
2238
2239         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2240         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2241 }
2242
2243 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2244                    u64 seq, struct lu_seq_range *range)
2245 {
2246         struct seq_server_site  *ss = osd_seq_site(osd);
2247
2248         if (fid_seq_is_idif(seq)) {
2249                 fld_range_set_ost(range);
2250                 range->lsr_index = idif_ost_idx(seq);
2251                 return 0;
2252         }
2253
2254         if (!fid_seq_in_fldb(seq)) {
2255                 fld_range_set_mdt(range);
2256                 if (ss != NULL)
2257                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2258                          * at all */
2259                         range->lsr_index = ss->ss_node_id;
2260                 return 0;
2261         }
2262
2263         LASSERT(ss != NULL);
2264         fld_range_set_any(range);
2265         /* OSD will only do local fld lookup */
2266         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2267 }
2268
2269 /*
2270  * Concurrency: no external locking is necessary.
2271  */
2272 static int osd_declare_object_create(const struct lu_env *env,
2273                                      struct dt_object *dt,
2274                                      struct lu_attr *attr,
2275                                      struct dt_allocation_hint *hint,
2276                                      struct dt_object_format *dof,
2277                                      struct thandle *handle)
2278 {
2279         struct osd_thandle      *oh;
2280         int                      rc;
2281         ENTRY;
2282
2283         LASSERT(handle != NULL);
2284
2285         oh = container_of0(handle, struct osd_thandle, ot_super);
2286         LASSERT(oh->ot_handle == NULL);
2287
2288         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2289                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2290         /* Reuse idle OI block may cause additional one OI block
2291          * to be changed. */
2292         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2293                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2294
2295         /* If this is directory, then we expect . and .. to be inserted as
2296          * well. The one directory block always needs to be created for the
2297          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2298          * block), there is no danger of needing a tree for the first block.
2299          */
2300         if (attr && S_ISDIR(attr->la_mode)) {
2301                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2302                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2303                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2304         }
2305
2306         if (!attr)
2307                 RETURN(0);
2308
2309         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2310                                    osd_dt_obj(dt), false, NULL, false);
2311         if (rc != 0)
2312                 RETURN(rc);
2313
2314         RETURN(rc);
2315 }
2316
2317 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2318                              struct lu_attr *attr,
2319                              struct dt_allocation_hint *hint,
2320                              struct dt_object_format *dof, struct thandle *th)
2321 {
2322         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2323         struct osd_object       *obj    = osd_dt_obj(dt);
2324         struct osd_thread_info  *info   = osd_oti_get(env);
2325         int result;
2326         ENTRY;
2327
2328         if (dt_object_exists(dt))
2329                 return -EEXIST;
2330
2331         LINVRNT(osd_invariant(obj));
2332         LASSERT(!dt_object_remote(dt));
2333         LASSERT(osd_write_locked(env, obj));
2334         LASSERT(th != NULL);
2335
2336         if (unlikely(fid_is_acct(fid)))
2337                 /* Quota files can't be created from the kernel any more,
2338                  * 'tune2fs -O quota' will take care of creating them */
2339                 RETURN(-EPERM);
2340
2341         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2342         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2343
2344         result = __osd_object_create(info, obj, attr, hint, dof, th);
2345         if (result == 0)
2346                 result = __osd_oi_insert(env, obj, fid, th);
2347
2348         LASSERT(ergo(result == 0,
2349                      dt_object_exists(dt) && !dt_object_remote(dt)));
2350
2351         LASSERT(osd_invariant(obj));
2352         RETURN(result);
2353 }
2354
2355 /**
2356  * Called to destroy on-disk representation of the object
2357  *
2358  * Concurrency: must be locked
2359  */
2360 static int osd_declare_object_destroy(const struct lu_env *env,
2361                                       struct dt_object *dt,
2362                                       struct thandle *th)
2363 {
2364         struct osd_object  *obj = osd_dt_obj(dt);
2365         struct inode       *inode = obj->oo_inode;
2366         struct osd_thandle *oh;
2367         int                 rc;
2368         ENTRY;
2369
2370         oh = container_of0(th, struct osd_thandle, ot_super);
2371         LASSERT(oh->ot_handle == NULL);
2372         LASSERT(inode);
2373
2374         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2375                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2376         /* Recycle idle OI leaf may cause additional three OI blocks
2377          * to be changed. */
2378         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2379                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2380         /* one less inode */
2381         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2382                                    -1, oh, obj, false, NULL, false);
2383         if (rc)
2384                 RETURN(rc);
2385         /* data to be truncated */
2386         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2387                                    0, oh, obj, true, NULL, false);
2388         RETURN(rc);
2389 }
2390
2391 static int osd_object_destroy(const struct lu_env *env,
2392                               struct dt_object *dt,
2393                               struct thandle *th)
2394 {
2395         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2396         struct osd_object      *obj = osd_dt_obj(dt);
2397         struct inode           *inode = obj->oo_inode;
2398         struct osd_device      *osd = osd_obj2dev(obj);
2399         struct osd_thandle     *oh;
2400         int                     result;
2401         ENTRY;
2402
2403         oh = container_of0(th, struct osd_thandle, ot_super);
2404         LASSERT(oh->ot_handle);
2405         LASSERT(inode);
2406         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2407
2408         if (unlikely(fid_is_acct(fid)))
2409                 RETURN(-EPERM);
2410
2411         if (S_ISDIR(inode->i_mode)) {
2412                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2413                         inode->i_nlink == 2);
2414                 /* it will check/delete the inode from remote parent,
2415                  * how to optimize it? unlink performance impaction XXX */
2416                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2417                 if (result != 0 && result != -ENOENT) {
2418                         CERROR("%s: delete inode "DFID": rc = %d\n",
2419                                osd_name(osd), PFID(fid), result);
2420                 }
2421                 spin_lock(&obj->oo_guard);
2422                 clear_nlink(inode);
2423                 spin_unlock(&obj->oo_guard);
2424                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2425         }
2426
2427         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2428
2429         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2430                                OI_CHECK_FLD);
2431
2432         /* XXX: add to ext3 orphan list */
2433         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2434
2435         /* not needed in the cache anymore */
2436         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2437
2438         RETURN(0);
2439 }
2440
2441 /**
2442  * Put the fid into lustre_mdt_attrs, and then place the structure
2443  * inode's ea. This fid should not be altered during the life time
2444  * of the inode.
2445  *
2446  * \retval +ve, on success
2447  * \retval -ve, on error
2448  *
2449  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2450  */
2451 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2452                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2453 {
2454         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2455         int                      rc;
2456         ENTRY;
2457
2458         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2459                 RETURN(0);
2460
2461         lustre_lma_init(lma, fid, compat, incompat);
2462         lustre_lma_swab(lma);
2463
2464         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2465                              XATTR_CREATE);
2466         /* LMA may already exist, but we need to check that all the
2467          * desired compat/incompat flags have been added. */
2468         if (unlikely(rc == -EEXIST)) {
2469                 if (compat == 0 && incompat == 0)
2470                         RETURN(0);
2471
2472                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2473                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2474                                      LMA_OLD_SIZE);
2475                 if (rc <= 0)
2476                         RETURN(-EINVAL);
2477
2478                 lustre_lma_swab(lma);
2479                 if (!(~lma->lma_compat & compat) &&
2480                     !(~lma->lma_incompat & incompat))
2481                         RETURN(0);
2482
2483                 lma->lma_compat |= compat;
2484                 lma->lma_incompat |= incompat;
2485                 lustre_lma_swab(lma);
2486                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2487                                      sizeof(*lma), XATTR_REPLACE);
2488         }
2489
2490         RETURN(rc);
2491 }
2492
2493 /**
2494  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2495  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2496  * To have compatilibility with 1.8 ldiskfs driver we need to have
2497  * magic number at start of fid data.
2498  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2499  * its inmemory API.
2500  */
2501 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2502                                          const struct lu_fid *fid)
2503 {
2504         if (!fid_is_namespace_visible(fid) ||
2505             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2506                 param->edp_magic = 0;
2507                 return;
2508         }
2509
2510         param->edp_magic = LDISKFS_LUFID_MAGIC;
2511         param->edp_len =  sizeof(struct lu_fid) + 1;
2512         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2513 }
2514
2515 /**
2516  * Try to read the fid from inode ea into dt_rec.
2517  *
2518  * \param fid object fid.
2519  *
2520  * \retval 0 on success
2521  */
2522 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2523                           __u32 ino, struct lu_fid *fid,
2524                           struct osd_inode_id *id)
2525 {
2526         struct osd_thread_info *info  = osd_oti_get(env);
2527         struct inode           *inode;
2528         ENTRY;
2529
2530         osd_id_gen(id, ino, OSD_OII_NOGEN);
2531         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2532         if (IS_ERR(inode))
2533                 RETURN(PTR_ERR(inode));
2534
2535         iput(inode);
2536         RETURN(0);
2537 }
2538
2539 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2540                                         struct inode *dir,
2541                                         struct inode *parent_dir,
2542                                         const struct lu_fid *dot_fid,
2543                                         const struct lu_fid *dot_dot_fid,
2544                                         struct osd_thandle *oth)
2545 {
2546         struct ldiskfs_dentry_param *dot_ldp;
2547         struct ldiskfs_dentry_param *dot_dot_ldp;
2548
2549         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2550         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2551
2552         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2553         dot_ldp->edp_magic = 0;
2554         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2555                                         dir, dot_ldp, dot_dot_ldp);
2556 }
2557
2558 /**
2559  * Create an local agent inode for remote entry
2560  */
2561 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2562                                                   struct osd_device *osd,
2563                                                   struct osd_object *pobj,
2564                                                   const struct lu_fid *fid,
2565                                                   __u32 type,
2566                                                   struct thandle *th)
2567 {
2568         struct osd_thread_info  *info = osd_oti_get(env);
2569         struct inode            *local;
2570         struct osd_thandle      *oh;
2571         int                     rc;
2572         ENTRY;
2573
2574         LASSERT(th);
2575         oh = container_of(th, struct osd_thandle, ot_super);
2576         LASSERT(oh->ot_handle->h_transaction != NULL);
2577
2578         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2579         if (IS_ERR(local)) {
2580                 CERROR("%s: create local error %d\n", osd_name(osd),
2581                        (int)PTR_ERR(local));
2582                 RETURN(local);
2583         }
2584
2585         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
2586         unlock_new_inode(local);
2587
2588         /* Set special LMA flag for local agent inode */
2589         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2590         if (rc != 0) {
2591                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2592                        osd_name(osd), PFID(fid), rc);
2593                 RETURN(ERR_PTR(rc));
2594         }
2595
2596         if (!S_ISDIR(type))
2597                 RETURN(local);
2598
2599         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2600                                          lu_object_fid(&pobj->oo_dt.do_lu),
2601                                          fid, oh);
2602         if (rc != 0) {
2603                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2604                         osd_name(osd), PFID(fid), rc);
2605                 RETURN(ERR_PTR(rc));
2606         }
2607
2608         RETURN(local);
2609 }
2610
2611 /**
2612  * Delete local agent inode for remote entry
2613  */
2614 static int osd_delete_local_agent_inode(const struct lu_env *env,
2615                                         struct osd_device *osd,
2616                                         const struct lu_fid *fid,
2617                                         __u32 ino, struct osd_thandle *oh)
2618 {
2619         struct osd_thread_info  *oti = osd_oti_get(env);
2620         struct osd_inode_id     *id = &oti->oti_id;
2621         struct inode            *inode;
2622         ENTRY;
2623
2624         id->oii_ino = le32_to_cpu(ino);
2625         id->oii_gen = OSD_OII_NOGEN;
2626         inode = osd_iget(oti, osd, id);
2627         if (IS_ERR(inode)) {
2628                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2629                        PFID(fid), id->oii_ino, id->oii_gen);
2630                 RETURN(PTR_ERR(inode));
2631         }
2632
2633         clear_nlink(inode);
2634         mark_inode_dirty(inode);
2635         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2636                 osd_name(osd), PFID(fid), inode->i_ino);
2637         iput(inode);
2638         RETURN(0);
2639 }
2640
2641 /**
2642  * OSD layer object create function for interoperability mode (b11826).
2643  * This is mostly similar to osd_object_create(). Only difference being, fid is
2644  * inserted into inode ea here.
2645  *
2646  * \retval   0, on success
2647  * \retval -ve, on error
2648  */
2649 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2650                                 struct lu_attr *attr,
2651                                 struct dt_allocation_hint *hint,
2652                                 struct dt_object_format *dof,
2653                                 struct thandle *th)
2654 {
2655         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2656         struct osd_object       *obj    = osd_dt_obj(dt);
2657         struct osd_thread_info  *info   = osd_oti_get(env);
2658         int                      result;
2659
2660         ENTRY;
2661
2662         if (dt_object_exists(dt))
2663                 RETURN(-EEXIST);
2664
2665         LASSERT(osd_invariant(obj));
2666         LASSERT(!dt_object_remote(dt));
2667         LASSERT(osd_write_locked(env, obj));
2668         LASSERT(th != NULL);
2669
2670         if (unlikely(fid_is_acct(fid)))
2671                 /* Quota files can't be created from the kernel any more,
2672                  * 'tune2fs -O quota' will take care of creating them */
2673                 RETURN(-EPERM);
2674
2675         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2676         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2677
2678         result = __osd_object_create(info, obj, attr, hint, dof, th);
2679         if (result == 0) {
2680                 if (fid_is_idif(fid) &&
2681                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
2682                         struct lu_fid *tfid = &info->oti_fid;
2683                         struct ost_id *oi   = &info->oti_ostid;
2684
2685                         fid_to_ostid(fid, oi);
2686                         ostid_to_fid(tfid, oi, 0);
2687                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
2688                                                 LMAC_FID_ON_OST, 0);
2689                 } else {
2690                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
2691                                 fid_is_on_ost(info, osd_obj2dev(obj),
2692                                               fid, OI_CHECK_FLD) ?
2693                                 LMAC_FID_ON_OST : 0, 0);
2694                 }
2695         }
2696
2697         if (result == 0)
2698                 result = __osd_oi_insert(env, obj, fid, th);
2699
2700         LASSERT(ergo(result == 0,
2701                      dt_object_exists(dt) && !dt_object_remote(dt)));
2702         LINVRNT(osd_invariant(obj));
2703         RETURN(result);
2704 }
2705
2706 static int osd_declare_object_ref_add(const struct lu_env *env,
2707                                       struct dt_object *dt,
2708                                       struct thandle *handle)
2709 {
2710         struct osd_thandle       *oh;
2711
2712         /* it's possible that object doesn't exist yet */
2713         LASSERT(handle != NULL);
2714
2715         oh = container_of0(handle, struct osd_thandle, ot_super);
2716         LASSERT(oh->ot_handle == NULL);
2717
2718         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2719                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2720
2721         return 0;
2722 }
2723
2724 /*
2725  * Concurrency: @dt is write locked.
2726  */
2727 static int osd_object_ref_add(const struct lu_env *env,
2728                               struct dt_object *dt, struct thandle *th)
2729 {
2730         struct osd_object  *obj = osd_dt_obj(dt);
2731         struct inode       *inode = obj->oo_inode;
2732         struct osd_thandle *oh;
2733         int                 rc = 0;
2734
2735         if (!dt_object_exists(dt))
2736                 return -ENOENT;
2737
2738         LINVRNT(osd_invariant(obj));
2739         LASSERT(!dt_object_remote(dt));
2740         LASSERT(osd_write_locked(env, obj));
2741         LASSERT(th != NULL);
2742
2743         oh = container_of0(th, struct osd_thandle, ot_super);
2744         LASSERT(oh->ot_handle != NULL);
2745
2746         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2747
2748         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2749                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2750         /*
2751          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2752          * (65000) subdirectories by storing "1" in i_nlink if the link count
2753          * would otherwise overflow. Directory tranversal tools understand
2754          * that (st_nlink == 1) indicates that the filesystem dose not track
2755          * hard links count on the directory, and will not abort subdirectory
2756          * scanning early once (st_nlink - 2) subdirs have been found.
2757          *
2758          * This also has to properly handle the case of inodes with nlink == 0
2759          * in case they are being linked into the PENDING directory
2760          */
2761         spin_lock(&obj->oo_guard);
2762         if (unlikely(inode->i_nlink == 0))
2763                 /* inc_nlink from 0 may cause WARN_ON */
2764                 set_nlink(inode, 1);
2765         else {
2766                 ldiskfs_inc_count(oh->ot_handle, inode);
2767                 if (!S_ISDIR(inode->i_mode))
2768                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2769         }
2770         spin_unlock(&obj->oo_guard);
2771
2772         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2773         LINVRNT(osd_invariant(obj));
2774
2775         return rc;
2776 }
2777
2778 static int osd_declare_object_ref_del(const struct lu_env *env,
2779                                       struct dt_object *dt,
2780                                       struct thandle *handle)
2781 {
2782         struct osd_thandle *oh;
2783
2784         LASSERT(!dt_object_remote(dt));
2785         LASSERT(handle != NULL);
2786
2787         oh = container_of0(handle, struct osd_thandle, ot_super);
2788         LASSERT(oh->ot_handle == NULL);
2789
2790         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2791                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2792
2793         return 0;
2794 }
2795
2796 /*
2797  * Concurrency: @dt is write locked.
2798  */
2799 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2800                               struct thandle *th)
2801 {
2802         struct osd_object       *obj = osd_dt_obj(dt);
2803         struct inode            *inode = obj->oo_inode;
2804         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2805         struct osd_thandle      *oh;
2806
2807         if (!dt_object_exists(dt))
2808                 return -ENOENT;
2809
2810         LINVRNT(osd_invariant(obj));
2811         LASSERT(!dt_object_remote(dt));
2812         LASSERT(osd_write_locked(env, obj));
2813         LASSERT(th != NULL);
2814
2815         oh = container_of0(th, struct osd_thandle, ot_super);
2816         LASSERT(oh->ot_handle != NULL);
2817
2818         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2819
2820         spin_lock(&obj->oo_guard);
2821         /* That can be result of upgrade from old Lustre version and
2822          * applied only to local files.  Just skip this ref_del call.
2823          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2824         if (inode->i_nlink == 0) {
2825                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2826                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2827                              ", maybe an upgraded file? (LU-3915)\n",
2828                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2829                 spin_unlock(&obj->oo_guard);
2830                 return 0;
2831         }
2832
2833         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2834                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2835
2836         ldiskfs_dec_count(oh->ot_handle, inode);
2837         spin_unlock(&obj->oo_guard);
2838
2839         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2840         LINVRNT(osd_invariant(obj));
2841
2842         return 0;
2843 }
2844
2845 /*
2846  * Get the 64-bit version for an inode.
2847  */
2848 static int osd_object_version_get(const struct lu_env *env,
2849                                   struct dt_object *dt, dt_obj_version_t *ver)
2850 {
2851         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2852
2853         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2854                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2855         *ver = LDISKFS_I(inode)->i_fs_version;
2856         return 0;
2857 }
2858
2859 /*
2860  * Concurrency: @dt is read locked.
2861  */
2862 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2863                          struct lu_buf *buf, const char *name)
2864 {
2865         struct osd_object      *obj    = osd_dt_obj(dt);
2866         struct inode           *inode  = obj->oo_inode;
2867         struct osd_thread_info *info   = osd_oti_get(env);
2868         struct dentry          *dentry = &info->oti_obj_dentry;
2869
2870         /* version get is not real XATTR but uses xattr API */
2871         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2872                 /* for version we are just using xattr API but change inode
2873                  * field instead */
2874                 if (buf->lb_len == 0)
2875                         return sizeof(dt_obj_version_t);
2876
2877                 if (buf->lb_len < sizeof(dt_obj_version_t))
2878                         return -ERANGE;
2879
2880                 osd_object_version_get(env, dt, buf->lb_buf);
2881
2882                 return sizeof(dt_obj_version_t);
2883         }
2884
2885         if (!dt_object_exists(dt))
2886                 return -ENOENT;
2887
2888         LASSERT(!dt_object_remote(dt));
2889         LASSERT(inode->i_op != NULL);
2890         LASSERT(inode->i_op->getxattr != NULL);
2891
2892         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2893 }
2894
2895
2896 static int osd_declare_xattr_set(const struct lu_env *env,
2897                                  struct dt_object *dt,
2898                                  const struct lu_buf *buf, const char *name,
2899                                  int fl, struct thandle *handle)
2900 {
2901         struct osd_thandle *oh;
2902         int credits;
2903         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2904
2905         LASSERT(handle != NULL);
2906
2907         oh = container_of0(handle, struct osd_thandle, ot_super);
2908         LASSERT(oh->ot_handle == NULL);
2909
2910         /* optimistic optimization: LMA is set first and usually fit inode */
2911         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2912                 if (dt_object_exists(dt))
2913                         credits = 0;
2914                 else
2915                         credits = 1;
2916         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2917                 credits = 1;
2918         } else {
2919                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2920                 if (buf && buf->lb_len > sb->s_blocksize) {
2921                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2922                                         sb->s_blocksize_bits;
2923                 }
2924                 /*
2925                  * xattr set may involve inode quota change, reserve credits for
2926                  * dquot_initialize()
2927                  */
2928                 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
2929         }
2930
2931         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2932
2933         return 0;
2934 }
2935
2936 /*
2937  * Set the 64-bit version for object
2938  */
2939 static void osd_object_version_set(const struct lu_env *env,
2940                                    struct dt_object *dt,
2941                                    dt_obj_version_t *new_version)
2942 {
2943         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2944
2945         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2946                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2947
2948         LDISKFS_I(inode)->i_fs_version = *new_version;
2949         /** Version is set after all inode operations are finished,
2950          *  so we should mark it dirty here */
2951         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2952 }
2953
2954 /*
2955  * Concurrency: @dt is write locked.
2956  */
2957 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2958                          const struct lu_buf *buf, const char *name, int fl,
2959                          struct thandle *handle)
2960 {
2961         struct osd_object      *obj      = osd_dt_obj(dt);
2962         struct inode           *inode    = obj->oo_inode;
2963         struct osd_thread_info *info     = osd_oti_get(env);
2964         int                     fs_flags = 0;
2965         ENTRY;
2966
2967         LASSERT(handle != NULL);
2968
2969         /* version set is not real XATTR */
2970         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2971                 /* for version we are just using xattr API but change inode
2972                  * field instead */
2973                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2974                 osd_object_version_set(env, dt, buf->lb_buf);
2975                 return sizeof(dt_obj_version_t);
2976         }
2977
2978         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
2979                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
2980
2981         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2982         if (fl & LU_XATTR_REPLACE)
2983                 fs_flags |= XATTR_REPLACE;
2984
2985         if (fl & LU_XATTR_CREATE)
2986                 fs_flags |= XATTR_CREATE;
2987
2988         if (strcmp(name, XATTR_NAME_LMV) == 0) {
2989                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2990                 int                      rc;
2991
2992                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
2993                 if (rc != 0)
2994                         RETURN(rc);
2995
2996                 lma->lma_incompat |= LMAI_STRIPED;
2997                 lustre_lma_swab(lma);
2998                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2999                                      sizeof(*lma), XATTR_REPLACE);
3000                 if (rc != 0)
3001                         RETURN(rc);
3002         }
3003
3004         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
3005             strcmp(name, XATTR_NAME_LINK) == 0)
3006                 return -ENOSPC;
3007
3008         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3009                                fs_flags);
3010 }
3011
3012 /*
3013  * Concurrency: @dt is read locked.
3014  */
3015 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3016                           const struct lu_buf *buf)
3017 {
3018         struct osd_object      *obj    = osd_dt_obj(dt);
3019         struct inode           *inode  = obj->oo_inode;
3020         struct osd_thread_info *info   = osd_oti_get(env);
3021         struct dentry          *dentry = &info->oti_obj_dentry;
3022
3023         if (!dt_object_exists(dt))
3024                 return -ENOENT;
3025
3026         LASSERT(!dt_object_remote(dt));
3027         LASSERT(inode->i_op != NULL);
3028         LASSERT(inode->i_op->listxattr != NULL);
3029
3030         dentry->d_inode = inode;
3031         dentry->d_sb = inode->i_sb;
3032         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3033 }
3034
3035 static int osd_declare_xattr_del(const struct lu_env *env,
3036                                  struct dt_object *dt, const char *name,
3037                                  struct thandle *handle)
3038 {
3039         struct osd_thandle *oh;
3040         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3041
3042         LASSERT(!dt_object_remote(dt));
3043         LASSERT(handle != NULL);
3044
3045         oh = container_of0(handle, struct osd_thandle, ot_super);
3046         LASSERT(oh->ot_handle == NULL);
3047
3048         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3049                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3050         /*
3051          * xattr del may involve inode quota change, reserve credits for
3052          * dquot_initialize()
3053          */
3054         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3055
3056         return 0;
3057 }
3058
3059 /*
3060  * Concurrency: @dt is write locked.
3061  */
3062 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3063                          const char *name, struct thandle *handle)
3064 {
3065         struct osd_object      *obj    = osd_dt_obj(dt);
3066         struct inode           *inode  = obj->oo_inode;
3067         struct osd_thread_info *info   = osd_oti_get(env);
3068         struct dentry          *dentry = &info->oti_obj_dentry;
3069         int                     rc;
3070
3071         if (!dt_object_exists(dt))
3072                 return -ENOENT;
3073
3074         LASSERT(!dt_object_remote(dt));
3075         LASSERT(inode->i_op != NULL);
3076         LASSERT(inode->i_op->removexattr != NULL);
3077         LASSERT(handle != NULL);
3078
3079         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3080
3081         ll_vfs_dq_init(inode);
3082         dentry->d_inode = inode;
3083         dentry->d_sb = inode->i_sb;
3084         rc = inode->i_op->removexattr(dentry, name);
3085         return rc;
3086 }
3087
3088 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3089                            __u64 start, __u64 end)
3090 {
3091         struct osd_object       *obj    = osd_dt_obj(dt);
3092         struct inode            *inode  = obj->oo_inode;
3093         struct osd_thread_info  *info   = osd_oti_get(env);
3094         struct dentry           *dentry = &info->oti_obj_dentry;
3095         struct file             *file   = &info->oti_file;
3096         int                     rc;
3097
3098         ENTRY;
3099
3100         dentry->d_inode = inode;
3101         dentry->d_sb = inode->i_sb;
3102         file->f_dentry = dentry;
3103         file->f_mapping = inode->i_mapping;
3104         file->f_op = inode->i_fop;
3105         set_file_inode(file, inode);
3106
3107         rc = ll_vfs_fsync_range(file, start, end, 0);
3108
3109         RETURN(rc);
3110 }
3111
3112 /*
3113  * Index operations.
3114  */
3115
3116 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3117                            const struct dt_index_features *feat)
3118 {
3119         struct iam_descr *descr;
3120
3121         if (osd_object_is_root(o))
3122                 return feat == &dt_directory_features;
3123
3124         LASSERT(o->oo_dir != NULL);
3125
3126         descr = o->oo_dir->od_container.ic_descr;
3127         if (feat == &dt_directory_features) {
3128                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3129                         return 1;
3130                 else
3131                         return 0;
3132         } else {
3133                 return
3134                         feat->dif_keysize_min <= descr->id_key_size &&
3135                         descr->id_key_size <= feat->dif_keysize_max &&
3136                         feat->dif_recsize_min <= descr->id_rec_size &&
3137                         descr->id_rec_size <= feat->dif_recsize_max &&
3138                         !(feat->dif_flags & (DT_IND_VARKEY |
3139                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3140                         ergo(feat->dif_flags & DT_IND_UPDATE,
3141                              1 /* XXX check that object (and file system) is
3142                                 * writable */);
3143         }
3144 }
3145
3146 static int osd_iam_container_init(const struct lu_env *env,
3147                                   struct osd_object *obj,
3148                                   struct osd_directory *dir)
3149 {
3150         struct iam_container *bag = &dir->od_container;
3151         int result;
3152
3153         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3154         if (result != 0)
3155                 return result;
3156
3157         result = iam_container_setup(bag);
3158         if (result == 0)
3159                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3160         else
3161                 iam_container_fini(bag);
3162
3163         return result;
3164 }
3165
3166
3167 /*
3168  * Concurrency: no external locking is necessary.
3169  */
3170 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3171                          const struct dt_index_features *feat)
3172 {
3173         int                      result;
3174         int                      skip_iam = 0;
3175         struct osd_object       *obj = osd_dt_obj(dt);
3176
3177         LINVRNT(osd_invariant(obj));
3178
3179         if (osd_object_is_root(obj)) {
3180                 dt->do_index_ops = &osd_index_ea_ops;
3181                 result = 0;
3182         } else if (feat == &dt_directory_features) {
3183                 dt->do_index_ops = &osd_index_ea_ops;
3184                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3185                         result = 0;
3186                 else
3187                         result = -ENOTDIR;
3188                 skip_iam = 1;
3189         } else if (unlikely(feat == &dt_otable_features)) {
3190                 dt->do_index_ops = &osd_otable_ops;
3191                 return 0;
3192         } else if (unlikely(feat == &dt_acct_features)) {
3193                 dt->do_index_ops = &osd_acct_index_ops;
3194                 result = 0;
3195                 skip_iam = 1;
3196         } else if (!osd_has_index(obj)) {
3197                 struct osd_directory *dir;
3198
3199                 OBD_ALLOC_PTR(dir);
3200                 if (dir != NULL) {
3201
3202                         spin_lock(&obj->oo_guard);
3203                         if (obj->oo_dir == NULL)
3204                                 obj->oo_dir = dir;
3205                         else
3206                                 /*
3207                                  * Concurrent thread allocated container data.
3208                                  */
3209                                 OBD_FREE_PTR(dir);
3210                         spin_unlock(&obj->oo_guard);
3211                         /*
3212                          * Now, that we have container data, serialize its
3213                          * initialization.
3214                          */
3215                         down_write(&obj->oo_ext_idx_sem);
3216                         /*
3217                          * recheck under lock.
3218                          */
3219                         if (!osd_has_index(obj))
3220                                 result = osd_iam_container_init(env, obj,
3221                                                                 obj->oo_dir);
3222                         else
3223                                 result = 0;
3224                         up_write(&obj->oo_ext_idx_sem);
3225                 } else {
3226                         result = -ENOMEM;
3227                 }
3228         } else {
3229                 result = 0;
3230         }
3231
3232         if (result == 0 && skip_iam == 0) {
3233                 if (!osd_iam_index_probe(env, obj, feat))
3234                         result = -ENOTDIR;
3235         }
3236         LINVRNT(osd_invariant(obj));
3237
3238         if (result == 0 && is_quota_glb_feat(feat) &&
3239             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3240                 result = osd_quota_migration(env, dt, feat);
3241
3242         return result;
3243 }
3244
3245 static int osd_otable_it_attr_get(const struct lu_env *env,
3246                                  struct dt_object *dt,
3247                                  struct lu_attr *attr)
3248 {
3249         attr->la_valid = 0;
3250         return 0;
3251 }
3252
3253 static const struct dt_object_operations osd_obj_ops = {
3254         .do_read_lock         = osd_object_read_lock,
3255         .do_write_lock        = osd_object_write_lock,
3256         .do_read_unlock       = osd_object_read_unlock,
3257         .do_write_unlock      = osd_object_write_unlock,
3258         .do_write_locked      = osd_object_write_locked,
3259         .do_attr_get          = osd_attr_get,
3260         .do_declare_attr_set  = osd_declare_attr_set,
3261         .do_attr_set          = osd_attr_set,
3262         .do_ah_init           = osd_ah_init,
3263         .do_declare_create    = osd_declare_object_create,
3264         .do_create            = osd_object_create,
3265         .do_declare_destroy   = osd_declare_object_destroy,
3266         .do_destroy           = osd_object_destroy,
3267         .do_index_try         = osd_index_try,
3268         .do_declare_ref_add   = osd_declare_object_ref_add,
3269         .do_ref_add           = osd_object_ref_add,
3270         .do_declare_ref_del   = osd_declare_object_ref_del,
3271         .do_ref_del           = osd_object_ref_del,
3272         .do_xattr_get         = osd_xattr_get,
3273         .do_declare_xattr_set = osd_declare_xattr_set,
3274         .do_xattr_set         = osd_xattr_set,
3275         .do_declare_xattr_del = osd_declare_xattr_del,
3276         .do_xattr_del         = osd_xattr_del,
3277         .do_xattr_list        = osd_xattr_list,
3278         .do_object_sync       = osd_object_sync,
3279 };
3280
3281 /**
3282  * dt_object_operations for interoperability mode
3283  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3284  */
3285 static const struct dt_object_operations osd_obj_ea_ops = {
3286         .do_read_lock         = osd_object_read_lock,
3287         .do_write_lock        = osd_object_write_lock,
3288         .do_read_unlock       = osd_object_read_unlock,
3289         .do_write_unlock      = osd_object_write_unlock,
3290         .do_write_locked      = osd_object_write_locked,
3291         .do_attr_get          = osd_attr_get,
3292         .do_declare_attr_set  = osd_declare_attr_set,
3293         .do_attr_set          = osd_attr_set,
3294         .do_ah_init           = osd_ah_init,
3295         .do_declare_create    = osd_declare_object_create,
3296         .do_create            = osd_object_ea_create,
3297         .do_declare_destroy   = osd_declare_object_destroy,
3298         .do_destroy           = osd_object_destroy,
3299         .do_index_try         = osd_index_try,
3300         .do_declare_ref_add   = osd_declare_object_ref_add,
3301         .do_ref_add           = osd_object_ref_add,
3302         .do_declare_ref_del   = osd_declare_object_ref_del,
3303         .do_ref_del           = osd_object_ref_del,
3304         .do_xattr_get         = osd_xattr_get,
3305         .do_declare_xattr_set = osd_declare_xattr_set,
3306         .do_xattr_set         = osd_xattr_set,
3307         .do_declare_xattr_del = osd_declare_xattr_del,
3308         .do_xattr_del         = osd_xattr_del,
3309         .do_xattr_list        = osd_xattr_list,
3310         .do_object_sync       = osd_object_sync,
3311 };
3312
3313 static const struct dt_object_operations osd_obj_otable_it_ops = {
3314         .do_attr_get    = osd_otable_it_attr_get,
3315         .do_index_try   = osd_index_try,
3316 };
3317
3318 static int osd_index_declare_iam_delete(const struct lu_env *env,
3319                                         struct dt_object *dt,
3320                                         const struct dt_key *key,
3321                                         struct thandle *handle)
3322 {
3323         struct osd_thandle    *oh;
3324
3325         oh = container_of0(handle, struct osd_thandle, ot_super);
3326         LASSERT(oh->ot_handle == NULL);
3327
3328         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3329                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3330
3331         return 0;
3332 }
3333
3334 /**
3335  *      delete a (key, value) pair from index \a dt specified by \a key
3336  *
3337  *      \param  dt      osd index object
3338  *      \param  key     key for index
3339  *      \param  rec     record reference
3340  *      \param  handle  transaction handler
3341  *
3342  *      \retval  0  success
3343  *      \retval -ve   failure
3344  */
3345 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3346                                 const struct dt_key *key,
3347                                 struct thandle *handle)
3348 {
3349         struct osd_thread_info *oti = osd_oti_get(env);
3350         struct osd_object      *obj = osd_dt_obj(dt);
3351         struct osd_thandle     *oh;
3352         struct iam_path_descr  *ipd;
3353         struct iam_container   *bag = &obj->oo_dir->od_container;
3354         int                     rc;
3355         ENTRY;
3356
3357         if (!dt_object_exists(dt))
3358                 RETURN(-ENOENT);
3359
3360         LINVRNT(osd_invariant(obj));
3361         LASSERT(!dt_object_remote(dt));
3362         LASSERT(bag->ic_object == obj->oo_inode);
3363         LASSERT(handle != NULL);
3364
3365         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3366
3367         ipd = osd_idx_ipd_get(env, bag);
3368         if (unlikely(ipd == NULL))
3369                 RETURN(-ENOMEM);
3370
3371         oh = container_of0(handle, struct osd_thandle, ot_super);
3372         LASSERT(oh->ot_handle != NULL);
3373         LASSERT(oh->ot_handle->h_transaction != NULL);
3374
3375         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3376                 /* swab quota uid/gid provided by caller */
3377                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3378                 key = (const struct dt_key *)&oti->oti_quota_id;
3379         }
3380
3381         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3382         osd_ipd_put(env, bag, ipd);
3383         LINVRNT(osd_invariant(obj));
3384         RETURN(rc);
3385 }
3386
3387 static int osd_index_declare_ea_delete(const struct lu_env *env,
3388                                        struct dt_object *dt,
3389                                        const struct dt_key *key,
3390                                        struct thandle *handle)
3391 {
3392         struct osd_thandle *oh;
3393         struct inode       *inode;
3394         int                 rc;
3395         ENTRY;
3396
3397         LASSERT(!dt_object_remote(dt));
3398         LASSERT(handle != NULL);
3399
3400         oh = container_of0(handle, struct osd_thandle, ot_super);
3401         LASSERT(oh->ot_handle == NULL);
3402
3403         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3404                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3405
3406         inode = osd_dt_obj(dt)->oo_inode;
3407         LASSERT(inode);
3408
3409         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3410                                    0, oh, osd_dt_obj(dt), true, NULL, false);
3411         RETURN(rc);
3412 }
3413
3414 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3415                                           struct dt_rec *fid)
3416 {
3417         struct osd_fid_pack *rec;
3418         int                  rc = -ENODATA;
3419
3420         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3421                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3422                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3423         }
3424         return rc;
3425 }
3426
3427 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3428                           const struct lu_fid *fid)
3429 {
3430         struct seq_server_site  *ss = osd_seq_site(osd);
3431         ENTRY;
3432
3433         /* FID seqs not in FLDB, must be local seq */
3434         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3435                 RETURN(0);
3436
3437         /* If FLD is not being initialized yet, it only happens during the
3438          * initialization, likely during mgs initialization, and we assume
3439          * this is local FID. */
3440         if (ss == NULL || ss->ss_server_fld == NULL)
3441                 RETURN(0);
3442
3443         /* Only check the local FLDB here */
3444         if (osd_seq_exists(env, osd, fid_seq(fid)))
3445                 RETURN(0);
3446
3447         RETURN(1);
3448 }
3449
3450 /**
3451  * Index delete function for interoperability mode (b11826).
3452  * It will remove the directory entry added by osd_index_ea_insert().
3453  * This entry is needed to maintain name->fid mapping.
3454  *
3455  * \param key,  key i.e. file entry to be deleted
3456  *
3457  * \retval   0, on success
3458  * \retval -ve, on error
3459  */
3460 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3461                                const struct dt_key *key, struct thandle *handle)
3462 {
3463         struct osd_object          *obj = osd_dt_obj(dt);
3464         struct inode               *dir = obj->oo_inode;
3465         struct dentry              *dentry;
3466         struct osd_thandle         *oh;
3467         struct ldiskfs_dir_entry_2 *de = NULL;
3468         struct buffer_head         *bh;
3469         struct htree_lock          *hlock = NULL;
3470         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3471         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3472         int                        rc;
3473         ENTRY;
3474
3475         if (!dt_object_exists(dt))
3476                 RETURN(-ENOENT);
3477
3478         LINVRNT(osd_invariant(obj));
3479         LASSERT(!dt_object_remote(dt));
3480         LASSERT(handle != NULL);
3481
3482         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3483
3484         oh = container_of(handle, struct osd_thandle, ot_super);
3485         LASSERT(oh->ot_handle != NULL);
3486         LASSERT(oh->ot_handle->h_transaction != NULL);
3487
3488         ll_vfs_dq_init(dir);
3489         dentry = osd_child_dentry_get(env, obj,
3490                                       (char *)key, strlen((char *)key));
3491
3492         if (obj->oo_hl_head != NULL) {
3493                 hlock = osd_oti_get(env)->oti_hlock;
3494                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3495                                    dir, LDISKFS_HLOCK_DEL);
3496         } else {
3497                 down_write(&obj->oo_ext_idx_sem);
3498         }
3499
3500         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3501         if (bh) {
3502                 __u32 ino = 0;
3503
3504                 /* If this is not the ".." entry, it might be a remote DNE
3505                  * entry and  we need to check if the FID is for a remote
3506                  * MDT.  If the FID is  not in the directory entry (e.g.
3507                  * upgraded 1.8 filesystem without dirdata enabled) then
3508                  * we need to get the FID from the LMA. For a remote directory
3509                  * there HAS to be an LMA, it cannot be an IGIF inode in this
3510                  * case.
3511                  *
3512                  * Delete the entry before the agent inode in order to
3513                  * simplify error handling.  At worst an error after deleting
3514                  * the entry first might leak the agent inode afterward. The
3515                  * reverse would need filesystem abort in case of error deleting
3516                  * the entry after the agent had been removed, or leave a
3517                  * dangling entry pointing at a random inode. */
3518                 if (strcmp((char *)key, dotdot) != 0) {
3519                         LASSERT(de != NULL);
3520                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3521                         /* If Fid is not in dentry, try to get it from LMA */
3522                         if (rc == -ENODATA) {
3523                                 struct osd_inode_id *id;
3524                                 struct inode *inode;
3525
3526                                 /* Before trying to get fid from the inode,
3527                                  * check whether the inode is valid.
3528                                  *
3529                                  * If the inode has been deleted, do not go
3530                                  * ahead to do osd_ea_fid_get, which will set
3531                                  * the inode to bad inode, which might cause
3532                                  * the inode to be deleted uncorrectly */
3533                                 inode = ldiskfs_iget(osd_sb(osd),
3534                                                      le32_to_cpu(de->inode));
3535                                 if (IS_ERR(inode)) {
3536                                         CDEBUG(D_INODE, "%s: "DFID"get inode"
3537                                                "error.\n", osd_name(osd),
3538                                                PFID(fid));
3539                                         rc = PTR_ERR(inode);
3540                                 } else {
3541                                         if (likely(inode->i_nlink != 0)) {
3542                                                 id = &osd_oti_get(env)->oti_id;
3543                                                 rc = osd_ea_fid_get(env, obj,
3544                                                         le32_to_cpu(de->inode),
3545                                                                     fid, id);
3546                                         } else {
3547                                                 CDEBUG(D_INFO, "%s: %u "DFID
3548                                                        "deleted.\n",
3549                                                        osd_name(osd),
3550                                                        le32_to_cpu(de->inode),
3551                                                        PFID(fid));
3552                                                 rc = -ESTALE;
3553                                         }
3554                                         iput(inode);
3555                                 }
3556                         }
3557                         if (rc == 0 &&
3558                             unlikely(osd_remote_fid(env, osd, fid)))
3559                                 /* Need to delete agent inode */
3560                                 ino = le32_to_cpu(de->inode);
3561                 }
3562                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3563                 brelse(bh);
3564                 if (rc == 0 && unlikely(ino != 0)) {
3565                         rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3566                                                           oh);
3567                         if (rc != 0)
3568                                 CERROR("%s: del local inode "DFID": rc = %d\n",
3569                                        osd_name(osd), PFID(fid), rc);
3570                 }
3571         } else {
3572                 rc = -ENOENT;
3573         }
3574         if (hlock != NULL)
3575                 ldiskfs_htree_unlock(hlock);
3576         else
3577                 up_write(&obj->oo_ext_idx_sem);
3578
3579         if (rc != 0)
3580                 GOTO(out, rc);
3581
3582         /* For inode on the remote MDT, .. will point to
3583          * /Agent directory, Check whether it needs to delete
3584          * from agent directory */
3585         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3586                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3587                                                    oh);
3588                 if (rc != 0 && rc != -ENOENT) {
3589                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3590                                osd_name(osd), PFID(fid), rc);
3591                 }
3592
3593                 if (rc == -ENOENT)
3594                         rc = 0;
3595
3596                 GOTO(out, rc);
3597         }
3598 out:
3599
3600         LASSERT(osd_invariant(obj));
3601         RETURN(rc);
3602 }
3603
3604 /**
3605  *      Lookup index for \a key and copy record to \a rec.
3606  *
3607  *      \param  dt      osd index object
3608  *      \param  key     key for index
3609  *      \param  rec     record reference
3610  *
3611  *      \retval  +ve  success : exact mach
3612  *      \retval  0    return record with key not greater than \a key
3613  *      \retval -ve   failure
3614  */
3615 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3616                                 struct dt_rec *rec, const struct dt_key *key)
3617 {
3618         struct osd_object      *obj = osd_dt_obj(dt);
3619         struct iam_path_descr  *ipd;
3620         struct iam_container   *bag = &obj->oo_dir->od_container;
3621         struct osd_thread_info *oti = osd_oti_get(env);
3622         struct iam_iterator    *it = &oti->oti_idx_it;
3623         struct iam_rec         *iam_rec;