Whamcloud - gitweb
LU-6361 lfsck: measure LFSCK 4 performance
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_handler.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2014, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/osd/osd_handler.c
37  *
38  * Top-level entry points into osd module
39  *
40  * Author: Nikita Danilov <nikita@clusterfs.com>
41  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
42  */
43
44 #define DEBUG_SUBSYSTEM S_OSD
45
46 #include <linux/module.h>
47 #include <linux/user_namespace.h>
48 #ifdef HAVE_UIDGID_HEADER
49 # include <linux/uidgid.h>
50 #endif
51
52 /* LUSTRE_VERSION_CODE */
53 #include <lustre_ver.h>
54 /* prerequisite for linux/xattr.h */
55 #include <linux/types.h>
56 /* prerequisite for linux/xattr.h */
57 #include <linux/fs.h>
58 /* XATTR_{REPLACE,CREATE} */
59 #include <linux/xattr.h>
60
61 #include <ldiskfs/ldiskfs.h>
62 #include <ldiskfs/xattr.h>
63 #undef ENTRY
64 /*
65  * struct OBD_{ALLOC,FREE}*()
66  * OBD_FAIL_CHECK
67  */
68 #include <obd_support.h>
69 /* struct ptlrpc_thread */
70 #include <lustre_net.h>
71 #include <lustre_fid.h>
72 /* process_config */
73 #include <lustre_param.h>
74
75 #include "osd_internal.h"
76 #include "osd_dynlocks.h"
77
78 /* llo_* api support */
79 #include <md_object.h>
80 #include <lustre_quota.h>
81
82 #include <lustre_linkea.h>
83
84 int ldiskfs_pdo = 1;
85 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
86                 "ldiskfs with parallel directory operations");
87
88 int ldiskfs_track_declares_assert;
89 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
90                 "LBUG during tracking of declares");
91
92 /* Slab to allocate dynlocks */
93 struct kmem_cache *dynlock_cachep;
94
95 /* Slab to allocate osd_it_ea */
96 struct kmem_cache *osd_itea_cachep;
97
98 static struct lu_kmem_descr ldiskfs_caches[] = {
99         {
100                 .ckd_cache = &dynlock_cachep,
101                 .ckd_name  = "dynlock_cache",
102                 .ckd_size  = sizeof(struct dynlock_handle)
103         },
104         {
105                 .ckd_cache = &osd_itea_cachep,
106                 .ckd_name  = "osd_itea_cache",
107                 .ckd_size  = sizeof(struct osd_it_ea)
108         },
109         {
110                 .ckd_cache = NULL
111         }
112 };
113
114 static const char dot[] = ".";
115 static const char dotdot[] = "..";
116 static const char remote_obj_dir[] = "REM_OBJ_DIR";
117
118 static const struct lu_object_operations      osd_lu_obj_ops;
119 static const struct dt_object_operations      osd_obj_ops;
120 static const struct dt_object_operations      osd_obj_ea_ops;
121 static const struct dt_object_operations      osd_obj_otable_it_ops;
122 static const struct dt_index_operations       osd_index_iam_ops;
123 static const struct dt_index_operations       osd_index_ea_ops;
124
125 int osd_trans_declare_op2rb[] = {
126         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
127         [OSD_OT_PUNCH]          = OSD_OT_MAX,
128         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
129         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
130         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
131         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
132         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
133         [OSD_OT_WRITE]          = OSD_OT_WRITE,
134         [OSD_OT_INSERT]         = OSD_OT_DELETE,
135         [OSD_OT_DELETE]         = OSD_OT_INSERT,
136         [OSD_OT_QUOTA]          = OSD_OT_MAX,
137 };
138
139 static int osd_has_index(const struct osd_object *obj)
140 {
141         return obj->oo_dt.do_index_ops != NULL;
142 }
143
144 static int osd_object_invariant(const struct lu_object *l)
145 {
146         return osd_invariant(osd_obj(l));
147 }
148
149 /*
150  * Concurrency: doesn't matter
151  */
152
153 /*
154  * Concurrency: doesn't matter
155  */
156 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
157 {
158         struct osd_thread_info *oti = osd_oti_get(env);
159         return oti->oti_w_locks > 0 && o->oo_owner == env;
160 }
161
162 /*
163  * Concurrency: doesn't access mutable data
164  */
165 static int osd_root_get(const struct lu_env *env,
166                         struct dt_device *dev, struct lu_fid *f)
167 {
168         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
169         return 0;
170 }
171
172 /*
173  * OSD object methods.
174  */
175
176 /*
177  * Concurrency: no concurrent access is possible that early in object
178  * life-cycle.
179  */
180 static struct lu_object *osd_object_alloc(const struct lu_env *env,
181                                           const struct lu_object_header *hdr,
182                                           struct lu_device *d)
183 {
184         struct osd_object *mo;
185
186         OBD_ALLOC_PTR(mo);
187         if (mo != NULL) {
188                 struct lu_object *l;
189
190                 l = &mo->oo_dt.do_lu;
191                 dt_object_init(&mo->oo_dt, NULL, d);
192                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
193                 l->lo_ops = &osd_lu_obj_ops;
194                 init_rwsem(&mo->oo_sem);
195                 init_rwsem(&mo->oo_ext_idx_sem);
196                 spin_lock_init(&mo->oo_guard);
197                 return l;
198         } else {
199                 return NULL;
200         }
201 }
202
203 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
204                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
205 {
206         int rc;
207
208         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
209         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
210                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
211         if (rc > 0) {
212                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
213                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
214                 rc = 0;
215                 lustre_lma_swab(lma);
216                 /* Check LMA compatibility */
217                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
218                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
219                               "for fid = "DFID", ino = %lu\n",
220                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
221                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
222                               PFID(&lma->lma_self_fid), inode->i_ino);
223                         rc = -EOPNOTSUPP;
224                 }
225         } else if (rc == 0) {
226                 rc = -ENODATA;
227         }
228
229         return rc;
230 }
231
232 /*
233  * retrieve object from backend ext fs.
234  **/
235 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
236                        struct osd_inode_id *id)
237 {
238         struct inode *inode = NULL;
239
240         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
241         if (IS_ERR(inode)) {
242                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
243                        id->oii_ino, PTR_ERR(inode));
244         } else if (id->oii_gen != OSD_OII_NOGEN &&
245                    inode->i_generation != id->oii_gen) {
246                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
247                        "i_generation = %u\n",
248                        id->oii_ino, id->oii_gen, inode->i_generation);
249                 iput(inode);
250                 inode = ERR_PTR(-ESTALE);
251         } else if (inode->i_nlink == 0) {
252                 /* due to parallel readdir and unlink,
253                 * we can have dead inode here. */
254                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
255                 iput(inode);
256                 inode = ERR_PTR(-ESTALE);
257         } else if (is_bad_inode(inode)) {
258                 CWARN("%.16s: bad inode: ino = %u\n",
259                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
260                 iput(inode);
261                 inode = ERR_PTR(-ENOENT);
262         } else {
263                 if (id->oii_gen == OSD_OII_NOGEN)
264                         osd_id_gen(id, inode->i_ino, inode->i_generation);
265
266                 /* Do not update file c/mtime in ldiskfs.
267                  * NB: we don't have any lock to protect this because we don't
268                  * have reference on osd_object now, but contention with
269                  * another lookup + attr_set can't happen in the tiny window
270                  * between if (...) and set S_NOCMTIME. */
271                 if (!(inode->i_flags & S_NOCMTIME))
272                         inode->i_flags |= S_NOCMTIME;
273         }
274         return inode;
275 }
276
277 static struct inode *
278 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
279              struct osd_inode_id *id, struct lu_fid *fid)
280 {
281         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
282         struct inode            *inode;
283         int                      rc;
284
285         inode = osd_iget(info, dev, id);
286         if (IS_ERR(inode))
287                 return inode;
288
289         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
290         if (rc == 0) {
291                 *fid = lma->lma_self_fid;
292         } else if (rc == -ENODATA) {
293                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
294                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
295                 else
296                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
297         } else {
298                 iput(inode);
299                 inode = ERR_PTR(rc);
300         }
301         return inode;
302 }
303
304 /**
305  * \retval +v: new filter_fid, does not contain self-fid
306  * \retval 0:  filter_fid_old, contains self-fid
307  * \retval -v: other failure cases
308  */
309 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
310                  struct dentry *dentry, struct lu_fid *fid)
311 {
312         struct filter_fid_old   *ff     = &info->oti_ff;
313         struct ost_id           *ostid  = &info->oti_ostid;
314         int                      rc;
315
316         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
317         if (rc == sizeof(*ff)) {
318                 rc = 0;
319                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
320                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
321                 /* XXX: should use real OST index in the future. LU-3569 */
322                 ostid_to_fid(fid, ostid, 0);
323         } else if (rc == sizeof(struct filter_fid)) {
324                 rc = 1;
325         } else if (rc >= 0) {
326                 rc = -EINVAL;
327         }
328
329         return rc;
330 }
331
332 static int osd_lma_self_repair(struct osd_thread_info *info,
333                                struct osd_device *osd, struct inode *inode,
334                                const struct lu_fid *fid, __u32 compat)
335 {
336         handle_t *jh;
337         int       rc;
338
339         LASSERT(current->journal_info == NULL);
340
341         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
342                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
343         if (IS_ERR(jh)) {
344                 rc = PTR_ERR(jh);
345                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
346                       osd_name(osd), rc);
347                 return rc;
348         }
349
350         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
351         if (rc != 0)
352                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
353                       osd_name(osd), rc);
354         ldiskfs_journal_stop(jh);
355         return rc;
356 }
357
358 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
359 {
360         struct osd_thread_info  *info   = osd_oti_get(env);
361         struct osd_device       *osd    = osd_obj2dev(obj);
362         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
363         struct inode            *inode  = obj->oo_inode;
364         struct dentry           *dentry = &info->oti_obj_dentry;
365         struct lu_fid           *fid    = NULL;
366         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
367         int                      rc;
368         ENTRY;
369
370         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
371         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
372                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
373         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
374                 fid = &lma->lma_self_fid;
375                 rc = osd_get_idif(info, inode, dentry, fid);
376                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
377                         /* For the given OST-object, if it has neither LMA nor
378                          * FID in XATTR_NAME_FID, then the given FID (which is
379                          * contained in the @obj, from client RPC for locating
380                          * the OST-object) is trusted. We use it to generate
381                          * the LMA. */
382                         osd_lma_self_repair(info, osd, inode, rfid,
383                                             LMAC_FID_ON_OST);
384                         RETURN(0);
385                 }
386         }
387
388         if (unlikely(rc == -ENODATA))
389                 RETURN(0);
390
391         if (rc < 0)
392                 RETURN(rc);
393
394         if (rc > 0) {
395                 rc = 0;
396                 lustre_lma_swab(lma);
397                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
398                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
399                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
400                               "fid = "DFID", ino = %lu\n", osd_name(osd),
401                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
402                               PFID(rfid), inode->i_ino);
403                         rc = -EOPNOTSUPP;
404                 } else {
405                         fid = &lma->lma_self_fid;
406                 }
407         }
408
409         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
410                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
411                         struct ost_id   *oi   = &info->oti_ostid;
412                         struct lu_fid   *fid1 = &info->oti_fid3;
413                         __u32            idx  = fid_idif_ost_idx(rfid);
414
415                         /* For old IDIF, the OST index is not part of the IDIF,
416                          * Means that different OSTs may have the same IDIFs.
417                          * Under such case, we need to make some compatible
418                          * check to make sure to trigger OI scrub properly. */
419                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
420                                 /* Given @rfid is new, LMA is old. */
421                                 fid_to_ostid(fid, oi);
422                                 ostid_to_fid(fid1, oi, idx);
423                                 if (lu_fid_eq(fid1, rfid)) {
424                                         if (osd->od_index_in_idif)
425                                                 osd_lma_self_repair(info, osd,
426                                                         inode, rfid,
427                                                         LMAC_FID_ON_OST);
428                                         RETURN(0);
429                                 }
430                         }
431                 }
432
433                 rc = -EREMCHG;
434         }
435
436         RETURN(rc);
437 }
438
439 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
440                           const struct lu_fid *fid,
441                           const struct lu_object_conf *conf)
442 {
443         struct osd_thread_info *info;
444         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
445         struct osd_device      *dev;
446         struct osd_idmap_cache *oic;
447         struct osd_inode_id    *id;
448         struct inode           *inode;
449         struct osd_scrub       *scrub;
450         struct scrub_file      *sf;
451         int                     result;
452         int                     saved  = 0;
453         bool                    in_oi  = false;
454         bool                    in_cache = false;
455         bool                    triggered = false;
456         ENTRY;
457
458         LINVRNT(osd_invariant(obj));
459         LASSERT(obj->oo_inode == NULL);
460         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
461
462         dev = osd_dev(ldev);
463         scrub = &dev->od_scrub;
464         sf = &scrub->os_file;
465         info = osd_oti_get(env);
466         LASSERT(info);
467         oic = &info->oti_cache;
468
469         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
470                 RETURN(-ENOENT);
471
472         /* For the object is created as locking anchor, or for the object to
473          * be created on disk. No need to osd_oi_lookup() at here because FID
474          * shouldn't never be re-used, if it's really a duplicate FID from
475          * unexpected reason, we should be able to detect it later by calling
476          * do_create->osd_oi_insert(). */
477         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
478                 GOTO(out, result = 0);
479
480         /* Search order: 1. per-thread cache. */
481         if (lu_fid_eq(fid, &oic->oic_fid) &&
482             likely(oic->oic_dev == dev)) {
483                 id = &oic->oic_lid;
484                 in_cache = true;
485                 goto iget;
486         }
487
488         id = &info->oti_id;
489         if (!list_empty(&scrub->os_inconsistent_items)) {
490                 /* Search order: 2. OI scrub pending list. */
491                 result = osd_oii_lookup(dev, fid, id);
492                 if (result == 0)
493                         goto iget;
494         }
495
496         /* Search order: 3. OI files. */
497         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
498         if (result == -ENOENT) {
499                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
500                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
501                                       sf->sf_oi_bitmap))
502                         GOTO(out, result = 0);
503
504                 goto trigger;
505         }
506
507         if (result != 0)
508                 GOTO(out, result);
509
510         in_oi = true;
511
512 iget:
513         inode = osd_iget(info, dev, id);
514         if (IS_ERR(inode)) {
515                 result = PTR_ERR(inode);
516                 if (result != -ENOENT && result != -ESTALE)
517                         GOTO(out, result);
518
519                 if (in_cache)
520                         fid_zero(&oic->oic_fid);
521
522                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
523                 if (result != 0)
524                         GOTO(out, result = (result == -ENOENT ? 0 : result));
525
526                 /* The OI mapping is there, but the inode is NOT there.
527                  * Two possible cases for that:
528                  *
529                  * 1) Backup/restore caused the OI invalid.
530                  * 2) Someone unlinked the object but NOT removed
531                  *    the OI mapping, such as mount target device
532                  *    as ldiskfs, and modify something directly.
533                  *
534                  * Generally, when the device is mounted, it will
535                  * auto check whether the system is restored from
536                  * file-level backup or not. We trust such detect
537                  * to distinguish the 1st case from the 2nd case. */
538                 if (!(scrub->os_file.sf_flags & SF_INCONSISTENT))
539                         GOTO(out, result = 0);
540
541 trigger:
542                 if (unlikely(triggered))
543                         GOTO(out, result = saved);
544
545                 triggered = true;
546                 if (thread_is_running(&scrub->os_thread)) {
547                         result = -EINPROGRESS;
548                 } else if (!dev->od_noscrub) {
549                         /* Since we do not know the right OI mapping, we have
550                          * to trigger OI scrub to scan the whole device. */
551                         result = osd_scrub_start(dev, SS_AUTO_FULL |
552                                 SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
553                         CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI "
554                                "scrub by RPC for "DFID", rc = %d [1]\n",
555                                osd_name(dev), PFID(fid), result);
556                         if (result == 0 || result == -EALREADY)
557                                 result = -EINPROGRESS;
558                         else
559                                 result = -EREMCHG;
560                 }
561
562                 /* We still have chance to get the valid inode: for the
563                  * object which is referenced by remote name entry, the
564                  * object on the local MDT will be linked under the dir
565                  * of "/REMOTE_PARENT_DIR" with its FID string as name.
566                  *
567                  * We do not know whether the object for the given FID
568                  * is referenced by some remote name entry or not, and
569                  * especially for DNE II, a multiple-linked object may
570                  * have many name entries reside on many MDTs.
571                  *
572                  * To simplify the operation, OSD will not distinguish
573                  * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
574                  * only happened for the RPC from other MDT during the
575                  * OI scrub, or for the client side RPC with FID only,
576                  * such as FID to path, or from old connected client. */
577                 saved = result;
578                 result = osd_lookup_in_remote_parent(info, dev, fid, id);
579                 if (result == 0) {
580                         in_oi = false;
581                         goto iget;
582                 }
583
584                 GOTO(out, result = saved);
585         }
586
587         obj->oo_inode = inode;
588         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
589
590         result = osd_check_lma(env, obj);
591         if (result != 0) {
592                 iput(inode);
593                 obj->oo_inode = NULL;
594
595                 if (result != -EREMCHG)
596                         GOTO(out, result);
597
598                 if (in_cache)
599                         fid_zero(&oic->oic_fid);
600
601                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
602                 if (result == 0)
603                         goto trigger;
604
605                 if (result != -ENOENT)
606                         GOTO(out, result);
607
608                 if (!in_oi && (fid_is_norm(fid) || fid_is_igif(fid)) &&
609                     ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
610                                      sf->sf_oi_bitmap))
611                         goto trigger;
612
613                 GOTO(out, result = 0);
614         }
615
616         obj->oo_compat_dot_created = 1;
617         obj->oo_compat_dotdot_created = 1;
618
619         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
620                 GOTO(out, result = 0);
621
622         LASSERT(obj->oo_hl_head == NULL);
623         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
624         if (obj->oo_hl_head == NULL) {
625                 obj->oo_inode = NULL;
626                 iput(inode);
627                 GOTO(out, result = -ENOMEM);
628         }
629         GOTO(out, result = 0);
630
631 out:
632         LINVRNT(osd_invariant(obj));
633         return result;
634 }
635
636 /*
637  * Concurrency: shouldn't matter.
638  */
639 static void osd_object_init0(struct osd_object *obj)
640 {
641         LASSERT(obj->oo_inode != NULL);
642         obj->oo_dt.do_body_ops = &osd_body_ops;
643         obj->oo_dt.do_lu.lo_header->loh_attr |=
644                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
645 }
646
647 /*
648  * Concurrency: no concurrent access is possible that early in object
649  * life-cycle.
650  */
651 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
652                            const struct lu_object_conf *conf)
653 {
654         struct osd_object *obj = osd_obj(l);
655         int result;
656
657         LINVRNT(osd_invariant(obj));
658
659         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
660                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
661                 l->lo_header->loh_attr |= LOHA_EXISTS;
662                 return 0;
663         }
664
665         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
666         obj->oo_dt.do_body_ops = &osd_body_ops_new;
667         if (result == 0 && obj->oo_inode != NULL)
668                 osd_object_init0(obj);
669
670         LINVRNT(osd_invariant(obj));
671         return result;
672 }
673
674 /*
675  * Concurrency: no concurrent access is possible that late in object
676  * life-cycle.
677  */
678 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
679 {
680         struct osd_object *obj = osd_obj(l);
681
682         LINVRNT(osd_invariant(obj));
683
684         dt_object_fini(&obj->oo_dt);
685         if (obj->oo_hl_head != NULL)
686                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
687         OBD_FREE_PTR(obj);
688 }
689
690 /*
691  * Concurrency: no concurrent access is possible that late in object
692  * life-cycle.
693  */
694 static void osd_index_fini(struct osd_object *o)
695 {
696         struct iam_container *bag;
697
698         if (o->oo_dir != NULL) {
699                 bag = &o->oo_dir->od_container;
700                 if (o->oo_inode != NULL) {
701                         if (bag->ic_object == o->oo_inode)
702                                 iam_container_fini(bag);
703                 }
704                 OBD_FREE_PTR(o->oo_dir);
705                 o->oo_dir = NULL;
706         }
707 }
708
709 /*
710  * Concurrency: no concurrent access is possible that late in object
711  * life-cycle (for all existing callers, that is. New callers have to provide
712  * their own locking.)
713  */
714 static int osd_inode_unlinked(const struct inode *inode)
715 {
716         return inode->i_nlink == 0;
717 }
718
719 enum {
720         OSD_TXN_OI_DELETE_CREDITS    = 20,
721         OSD_TXN_INODE_DELETE_CREDITS = 20
722 };
723
724 /*
725  * Journal
726  */
727
728 #if OSD_THANDLE_STATS
729 /**
730  * Set time when the handle is allocated
731  */
732 static void osd_th_alloced(struct osd_thandle *oth)
733 {
734         oth->oth_alloced = cfs_time_current();
735 }
736
737 /**
738  * Set time when the handle started
739  */
740 static void osd_th_started(struct osd_thandle *oth)
741 {
742         oth->oth_started = cfs_time_current();
743 }
744
745 /**
746  * Helper function to convert time interval to microseconds packed in
747  * long int.
748  */
749 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
750 {
751         struct timeval val;
752
753         cfs_duration_usec(cfs_time_sub(end, start), &val);
754         return val.tv_sec * 1000000 + val.tv_usec;
755 }
756
757 /**
758  * Check whether the we deal with this handle for too long.
759  */
760 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
761                                 cfs_time_t alloced, cfs_time_t started,
762                                 cfs_time_t closed)
763 {
764         cfs_time_t now = cfs_time_current();
765
766         LASSERT(dev != NULL);
767
768         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
769                             interval_to_usec(alloced, started));
770         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
771                             interval_to_usec(started, closed));
772         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
773                             interval_to_usec(closed, now));
774
775         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
776                 CWARN("transaction handle %p was open for too long: "
777                       "now "CFS_TIME_T" ,"
778                       "alloced "CFS_TIME_T" ,"
779                       "started "CFS_TIME_T" ,"
780                       "closed "CFS_TIME_T"\n",
781                       oth, now, alloced, started, closed);
782                 libcfs_debug_dumpstack(NULL);
783         }
784 }
785
786 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
787 {                                                                       \
788         cfs_time_t __closed = cfs_time_current();                       \
789         cfs_time_t __alloced = oth->oth_alloced;                        \
790         cfs_time_t __started = oth->oth_started;                        \
791                                                                         \
792         expr;                                                           \
793         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
794 }
795
796 #else /* OSD_THANDLE_STATS */
797
798 #define osd_th_alloced(h)                  do {} while(0)
799 #define osd_th_started(h)                  do {} while(0)
800 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
801
802 #endif /* OSD_THANDLE_STATS */
803
804 /*
805  * Concurrency: doesn't access mutable data.
806  */
807 static int osd_param_is_not_sane(const struct osd_device *dev,
808                                  const struct thandle *th)
809 {
810         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
811
812         return oh->ot_credits > osd_transaction_size(dev);
813 }
814
815 /*
816  * Concurrency: shouldn't matter.
817  */
818 static void osd_trans_commit_cb(struct super_block *sb,
819                                 struct ldiskfs_journal_cb_entry *jcb, int error)
820 {
821         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
822         struct thandle     *th  = &oh->ot_super;
823         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
824         struct dt_txn_commit_cb *dcb, *tmp;
825
826         LASSERT(oh->ot_handle == NULL);
827
828         if (error)
829                 CERROR("transaction @0x%p commit error: %d\n", th, error);
830
831         dt_txn_hook_commit(th);
832
833         /* call per-transaction callbacks if any */
834         list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
835                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
836                          "commit callback entry: magic=%x name='%s'\n",
837                          dcb->dcb_magic, dcb->dcb_name);
838                 list_del_init(&dcb->dcb_linkage);
839                 dcb->dcb_func(NULL, th, dcb, error);
840         }
841
842         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
843         lu_device_put(lud);
844         th->th_dev = NULL;
845
846         lu_context_exit(&th->th_ctx);
847         lu_context_fini(&th->th_ctx);
848         OBD_FREE_PTR(oh);
849 }
850
851 static struct thandle *osd_trans_create(const struct lu_env *env,
852                                         struct dt_device *d)
853 {
854         struct osd_thread_info  *oti = osd_oti_get(env);
855         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
856         struct osd_thandle      *oh;
857         struct thandle          *th;
858         ENTRY;
859
860         /* on pending IO in this thread should left from prev. request */
861         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
862
863         th = ERR_PTR(-ENOMEM);
864         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
865         if (oh != NULL) {
866                 oh->ot_quota_trans = &oti->oti_quota_trans;
867                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
868                 th = &oh->ot_super;
869                 th->th_dev = d;
870                 th->th_result = 0;
871                 th->th_tags = LCT_TX_HANDLE;
872                 oh->ot_credits = 0;
873                 INIT_LIST_HEAD(&oh->ot_dcb_list);
874                 osd_th_alloced(oh);
875
876                 memset(oti->oti_declare_ops, 0,
877                                         sizeof(oti->oti_declare_ops));
878                 memset(oti->oti_declare_ops_rb, 0,
879                                         sizeof(oti->oti_declare_ops_rb));
880                 memset(oti->oti_declare_ops_cred, 0,
881                                         sizeof(oti->oti_declare_ops_cred));
882                 oti->oti_rollback = false;
883         }
884         RETURN(th);
885 }
886
887 /*
888  * Concurrency: shouldn't matter.
889  */
890 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
891                            struct thandle *th)
892 {
893         struct osd_thread_info *oti = osd_oti_get(env);
894         struct osd_device  *dev = osd_dt_dev(d);
895         handle_t           *jh;
896         struct osd_thandle *oh;
897         int rc;
898
899         ENTRY;
900
901         LASSERT(current->journal_info == NULL);
902
903         oh = container_of0(th, struct osd_thandle, ot_super);
904         LASSERT(oh != NULL);
905         LASSERT(oh->ot_handle == NULL);
906
907         rc = dt_txn_hook_start(env, d, th);
908         if (rc != 0)
909                 GOTO(out, rc);
910
911         if (unlikely(osd_param_is_not_sane(dev, th))) {
912                 static unsigned long last_printed;
913                 static int last_credits;
914
915                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
916                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
917                       oh->ot_credits,
918                       osd_journal(dev)->j_max_transaction_buffers);
919                 CWARN("  create: %u/%u, destroy: %u/%u\n",
920                       oti->oti_declare_ops[OSD_OT_CREATE],
921                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
922                       oti->oti_declare_ops[OSD_OT_DESTROY],
923                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
924                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
925                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
926                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
927                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
928                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
929                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
930                       oti->oti_declare_ops[OSD_OT_WRITE],
931                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
932                       oti->oti_declare_ops[OSD_OT_PUNCH],
933                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
934                       oti->oti_declare_ops[OSD_OT_QUOTA],
935                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
936                 CWARN("  insert: %u/%u, delete: %u/%u\n",
937                       oti->oti_declare_ops[OSD_OT_INSERT],
938                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
939                       oti->oti_declare_ops[OSD_OT_DELETE],
940                       oti->oti_declare_ops_cred[OSD_OT_DELETE]);
941                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
942                       oti->oti_declare_ops[OSD_OT_REF_ADD],
943                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
944                       oti->oti_declare_ops[OSD_OT_REF_DEL],
945                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
946
947                 if (last_credits != oh->ot_credits &&
948                     time_after(jiffies, last_printed +
949                                msecs_to_jiffies(60 * MSEC_PER_SEC))) {
950                         libcfs_debug_dumpstack(NULL);
951                         last_credits = oh->ot_credits;
952                         last_printed = jiffies;
953                 }
954                 /* XXX Limit the credits to 'max_transaction_buffers', and
955                  *     let the underlying filesystem to catch the error if
956                  *     we really need so many credits.
957                  *
958                  *     This should be removed when we can calculate the
959                  *     credits precisely. */
960                 oh->ot_credits = osd_transaction_size(dev);
961         }
962
963         /*
964          * XXX temporary stuff. Some abstraction layer should
965          * be used.
966          */
967         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
968         osd_th_started(oh);
969         if (!IS_ERR(jh)) {
970                 oh->ot_handle = jh;
971                 LASSERT(oti->oti_txns == 0);
972                 lu_context_init(&th->th_ctx, th->th_tags);
973                 lu_context_enter(&th->th_ctx);
974
975                 lu_device_get(&d->dd_lu_dev);
976                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
977                               "osd-tx", th);
978                 oti->oti_txns++;
979                 rc = 0;
980         } else {
981                 rc = PTR_ERR(jh);
982         }
983 out:
984         RETURN(rc);
985 }
986
987 static int osd_seq_exists(const struct lu_env *env,
988                           struct osd_device *osd, u64 seq)
989 {
990         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
991         struct seq_server_site  *ss = osd_seq_site(osd);
992         int                     rc;
993         ENTRY;
994
995         LASSERT(ss != NULL);
996         LASSERT(ss->ss_server_fld != NULL);
997
998         rc = osd_fld_lookup(env, osd, seq, range);
999         if (rc != 0) {
1000                 if (rc != -ENOENT)
1001                         CERROR("%s: can't lookup FLD sequence "LPX64
1002                                ": rc = %d\n", osd_name(osd), seq, rc);
1003                 RETURN(0);
1004         }
1005
1006         RETURN(ss->ss_node_id == range->lsr_index);
1007 }
1008
1009 /*
1010  * Concurrency: shouldn't matter.
1011  */
1012 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1013                           struct thandle *th)
1014 {
1015         int                     rc = 0;
1016         struct osd_thandle     *oh;
1017         struct osd_thread_info *oti = osd_oti_get(env);
1018         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1019         struct osd_device      *osd = osd_dt_dev(th->th_dev);
1020         struct qsd_instance    *qsd = osd->od_quota_slave;
1021         struct lquota_trans    *qtrans;
1022         ENTRY;
1023
1024         oh = container_of0(th, struct osd_thandle, ot_super);
1025
1026         qtrans = oh->ot_quota_trans;
1027         oh->ot_quota_trans = NULL;
1028
1029         if (oh->ot_handle != NULL) {
1030                 handle_t *hdl = oh->ot_handle;
1031
1032                 /*
1033                  * add commit callback
1034                  * notice we don't do this in osd_trans_start()
1035                  * as underlying transaction can change during truncate
1036                  */
1037                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1038                                          &oh->ot_jcb);
1039
1040                 LASSERT(oti->oti_txns == 1);
1041                 oti->oti_txns--;
1042                 rc = dt_txn_hook_stop(env, th);
1043                 if (rc != 0)
1044                         CERROR("%s: failed in transaction hook: rc = %d\n",
1045                                osd_name(osd), rc);
1046
1047                 /* hook functions might modify th_sync */
1048                 hdl->h_sync = th->th_sync;
1049
1050                 oh->ot_handle = NULL;
1051                 OSD_CHECK_SLOW_TH(oh, osd, rc = ldiskfs_journal_stop(hdl));
1052                 if (rc != 0)
1053                         CERROR("%s: failed to stop transaction: rc = %d\n",
1054                                osd_name(osd), rc);
1055         } else {
1056                 OBD_FREE_PTR(oh);
1057         }
1058
1059         /* inform the quota slave device that the transaction is stopping */
1060         qsd_op_end(env, qsd, qtrans);
1061
1062         /* as we want IO to journal and data IO be concurrent, we don't block
1063          * awaiting data IO completion in osd_do_bio(), instead we wait here
1064          * once transaction is submitted to the journal. all reqular requests
1065          * don't do direct IO (except read/write), thus this wait_event becomes
1066          * no-op for them.
1067          *
1068          * IMPORTANT: we have to wait till any IO submited by the thread is
1069          * completed otherwise iobuf may be corrupted by different request
1070          */
1071         wait_event(iobuf->dr_wait,
1072                        atomic_read(&iobuf->dr_numreqs) == 0);
1073         osd_fini_iobuf(osd, iobuf);
1074         if (!rc)
1075                 rc = iobuf->dr_error;
1076
1077         RETURN(rc);
1078 }
1079
1080 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1081 {
1082         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1083                                                ot_super);
1084
1085         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1086         LASSERT(&dcb->dcb_func != NULL);
1087         list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1088
1089         return 0;
1090 }
1091
1092 /*
1093  * Called just before object is freed. Releases all resources except for
1094  * object itself (that is released by osd_object_free()).
1095  *
1096  * Concurrency: no concurrent access is possible that late in object
1097  * life-cycle.
1098  */
1099 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1100 {
1101         struct osd_object *obj   = osd_obj(l);
1102         struct inode      *inode = obj->oo_inode;
1103
1104         LINVRNT(osd_invariant(obj));
1105
1106         /*
1107          * If object is unlinked remove fid->ino mapping from object index.
1108          */
1109
1110         osd_index_fini(obj);
1111         if (inode != NULL) {
1112                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1113                 qid_t                    uid = i_uid_read(inode);
1114                 qid_t                    gid = i_gid_read(inode);
1115
1116                 iput(inode);
1117                 obj->oo_inode = NULL;
1118
1119                 if (qsd != NULL) {
1120                         struct osd_thread_info  *info = osd_oti_get(env);
1121                         struct lquota_id_info   *qi = &info->oti_qi;
1122
1123                         /* Release granted quota to master if necessary */
1124                         qi->lqi_id.qid_uid = uid;
1125                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1126
1127                         qi->lqi_id.qid_uid = gid;
1128                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1129                 }
1130         }
1131 }
1132
1133 /*
1134  * Concurrency: ->loo_object_release() is called under site spin-lock.
1135  */
1136 static void osd_object_release(const struct lu_env *env,
1137                                struct lu_object *l)
1138 {
1139 }
1140
1141 /*
1142  * Concurrency: shouldn't matter.
1143  */
1144 static int osd_object_print(const struct lu_env *env, void *cookie,
1145                             lu_printer_t p, const struct lu_object *l)
1146 {
1147         struct osd_object *o = osd_obj(l);
1148         struct iam_descr  *d;
1149
1150         if (o->oo_dir != NULL)
1151                 d = o->oo_dir->od_container.ic_descr;
1152         else
1153                 d = NULL;
1154         return (*p)(env, cookie,
1155                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1156                     o, o->oo_inode,
1157                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1158                     o->oo_inode ? o->oo_inode->i_generation : 0,
1159                     d ? d->id_ops->id_name : "plain");
1160 }
1161
1162 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1163
1164 /*
1165  * Concurrency: shouldn't matter.
1166  */
1167 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1168                struct obd_statfs *sfs)
1169 {
1170         struct osd_device  *osd = osd_dt_dev(d);
1171         struct super_block *sb = osd_sb(osd);
1172         struct kstatfs     *ksfs;
1173         int result = 0;
1174
1175         if (unlikely(osd->od_mnt == NULL))
1176                 return -EINPROGRESS;
1177
1178         /* osd_lproc.c call this without env, allocate ksfs for that case */
1179         if (unlikely(env == NULL)) {
1180                 OBD_ALLOC_PTR(ksfs);
1181                 if (ksfs == NULL)
1182                         return -ENOMEM;
1183         } else {
1184                 ksfs = &osd_oti_get(env)->oti_ksfs;
1185         }
1186
1187         spin_lock(&osd->od_osfs_lock);
1188         result = sb->s_op->statfs(sb->s_root, ksfs);
1189         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1190                 statfs_pack(sfs, ksfs);
1191                 if (sb->s_flags & MS_RDONLY)
1192                         sfs->os_state = OS_STATE_READONLY;
1193         }
1194
1195         spin_unlock(&osd->od_osfs_lock);
1196
1197         if (unlikely(env == NULL))
1198                 OBD_FREE_PTR(ksfs);
1199
1200         /* Reserve a small amount of space for local objects like last_rcvd,
1201          * llog, quota files, ... */
1202         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1203                 sfs->os_bavail = 0;
1204         } else {
1205                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1206                 /** Take out metadata overhead for indirect blocks */
1207                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1208         }
1209
1210         return result;
1211 }
1212
1213 /**
1214  * Estimate space needed for file creations. We assume the largest filename
1215  * which is 2^64 - 1, hence a filename of 20 chars.
1216  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1217  */
1218 #ifdef __LDISKFS_DIR_REC_LEN
1219 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1220 #else
1221 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1222 #endif
1223
1224 /*
1225  * Concurrency: doesn't access mutable data.
1226  */
1227 static void osd_conf_get(const struct lu_env *env,
1228                          const struct dt_device *dev,
1229                          struct dt_device_param *param)
1230 {
1231         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1232         int                ea_overhead;
1233
1234         /*
1235          * XXX should be taken from not-yet-existing fs abstraction layer.
1236          */
1237         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1238         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1239         param->ddp_block_shift  = sb->s_blocksize_bits;
1240         param->ddp_mount_type     = LDD_MT_LDISKFS;
1241         param->ddp_maxbytes       = sb->s_maxbytes;
1242         /* Overhead estimate should be fairly accurate, so we really take a tiny
1243          * error margin which also avoids fragmenting the filesystem too much */
1244         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1245         /* inode are statically allocated, so per-inode space consumption
1246          * is the space consumed by the directory entry */
1247         param->ddp_inodespace     = PER_OBJ_USAGE;
1248         /* per-fragment overhead to be used by the client code */
1249         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1250         param->ddp_mntopts      = 0;
1251         if (test_opt(sb, XATTR_USER))
1252                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1253         if (test_opt(sb, POSIX_ACL))
1254                 param->ddp_mntopts |= MNTOPT_ACL;
1255
1256         /* LOD might calculate the max stripe count based on max_ea_size,
1257          * so we need take account in the overhead as well,
1258          * xattr_header + magic + xattr_entry_head */
1259         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1260                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1261
1262 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1263         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1264                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1265                                                                 ea_overhead;
1266         else
1267 #endif
1268                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1269 }
1270
1271 /*
1272  * Concurrency: shouldn't matter.
1273  */
1274 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1275 {
1276         int rc;
1277
1278         CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1279
1280         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1281
1282         CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
1283                LUSTRE_OSD_LDISKFS_NAME, rc);
1284
1285         return rc;
1286 }
1287
1288 /**
1289  * Start commit for OSD device.
1290  *
1291  * An implementation of dt_commit_async method for OSD device.
1292  * Asychronously starts underlayng fs sync and thereby a transaction
1293  * commit.
1294  *
1295  * \param env environment
1296  * \param d dt device
1297  *
1298  * \see dt_device_operations
1299  */
1300 static int osd_commit_async(const struct lu_env *env,
1301                             struct dt_device *d)
1302 {
1303         struct super_block *s = osd_sb(osd_dt_dev(d));
1304         ENTRY;
1305
1306         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1307         RETURN(s->s_op->sync_fs(s, 0));
1308 }
1309
1310 /*
1311  * Concurrency: shouldn't matter.
1312  */
1313
1314 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1315 {
1316         struct super_block *sb = osd_sb(osd_dt_dev(d));
1317         struct block_device *dev = sb->s_bdev;
1318 #ifdef HAVE_DEV_SET_RDONLY
1319         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1320         int rc = 0;
1321 #else
1322         int rc = -EOPNOTSUPP;
1323 #endif
1324         ENTRY;
1325
1326 #ifdef HAVE_DEV_SET_RDONLY
1327         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1328
1329         if (jdev && (jdev != dev)) {
1330                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1331                        (long)jdev);
1332                 dev_set_rdonly(jdev);
1333         }
1334         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1335         dev_set_rdonly(dev);
1336 #else
1337         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1338                osd_dt_dev(d)->od_svname, (long)dev, rc);
1339 #endif
1340         RETURN(rc);
1341 }
1342
1343 /**
1344  * Note: we do not count into QUOTA here.
1345  * If we mount with --data_journal we may need more.
1346  */
1347 const int osd_dto_credits_noquota[DTO_NR] = {
1348         /**
1349          * Insert.
1350          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1351          * SINGLEDATA_TRANS_BLOCKS(8)
1352          * XXX Note: maybe iam need more, since iam have more level than
1353          *           EXT3 htree.
1354          */
1355         [DTO_INDEX_INSERT]  = 16,
1356         /**
1357          * Delete
1358          * just modify a single entry, probably merge few within a block
1359          */
1360         [DTO_INDEX_DELETE]  = 1,
1361         /**
1362          * Used for OI scrub
1363          */
1364         [DTO_INDEX_UPDATE]  = 16,
1365         /**
1366          * 4(inode, inode bits, groups, GDT)
1367          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1368          */
1369         [DTO_OBJECT_CREATE] = 4,
1370         /**
1371          * 4(inode, inode bits, groups, GDT)
1372          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1373          */
1374         [DTO_OBJECT_DELETE] = 4,
1375         /**
1376          * Attr set credits (inode)
1377          */
1378         [DTO_ATTR_SET_BASE] = 1,
1379         /**
1380          * Xattr set. The same as xattr of EXT3.
1381          * DATA_TRANS_BLOCKS(14)
1382          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1383          * are also counted in. Do not know why?
1384          */
1385         [DTO_XATTR_SET]     = 14,
1386         /**
1387          * credits for inode change during write.
1388          */
1389         [DTO_WRITE_BASE]    = 3,
1390         /**
1391          * credits for single block write.
1392          */
1393         [DTO_WRITE_BLOCK]   = 14,
1394         /**
1395          * Attr set credits for chown.
1396          * This is extra credits for setattr, and it is null without quota
1397          */
1398         [DTO_ATTR_SET_CHOWN] = 0
1399 };
1400
1401 static const struct dt_device_operations osd_dt_ops = {
1402         .dt_root_get       = osd_root_get,
1403         .dt_statfs         = osd_statfs,
1404         .dt_trans_create   = osd_trans_create,
1405         .dt_trans_start    = osd_trans_start,
1406         .dt_trans_stop     = osd_trans_stop,
1407         .dt_trans_cb_add   = osd_trans_cb_add,
1408         .dt_conf_get       = osd_conf_get,
1409         .dt_sync           = osd_sync,
1410         .dt_ro             = osd_ro,
1411         .dt_commit_async   = osd_commit_async,
1412 };
1413
1414 static void osd_object_read_lock(const struct lu_env *env,
1415                                  struct dt_object *dt, unsigned role)
1416 {
1417         struct osd_object *obj = osd_dt_obj(dt);
1418         struct osd_thread_info *oti = osd_oti_get(env);
1419
1420         LINVRNT(osd_invariant(obj));
1421
1422         LASSERT(obj->oo_owner != env);
1423         down_read_nested(&obj->oo_sem, role);
1424
1425         LASSERT(obj->oo_owner == NULL);
1426         oti->oti_r_locks++;
1427 }
1428
1429 static void osd_object_write_lock(const struct lu_env *env,
1430                                   struct dt_object *dt, unsigned role)
1431 {
1432         struct osd_object *obj = osd_dt_obj(dt);
1433         struct osd_thread_info *oti = osd_oti_get(env);
1434
1435         LINVRNT(osd_invariant(obj));
1436
1437         LASSERT(obj->oo_owner != env);
1438         down_write_nested(&obj->oo_sem, role);
1439
1440         LASSERT(obj->oo_owner == NULL);
1441         obj->oo_owner = env;
1442         oti->oti_w_locks++;
1443 }
1444
1445 static void osd_object_read_unlock(const struct lu_env *env,
1446                                    struct dt_object *dt)
1447 {
1448         struct osd_object *obj = osd_dt_obj(dt);
1449         struct osd_thread_info *oti = osd_oti_get(env);
1450
1451         LINVRNT(osd_invariant(obj));
1452
1453         LASSERT(oti->oti_r_locks > 0);
1454         oti->oti_r_locks--;
1455         up_read(&obj->oo_sem);
1456 }
1457
1458 static void osd_object_write_unlock(const struct lu_env *env,
1459                                     struct dt_object *dt)
1460 {
1461         struct osd_object *obj = osd_dt_obj(dt);
1462         struct osd_thread_info *oti = osd_oti_get(env);
1463
1464         LINVRNT(osd_invariant(obj));
1465
1466         LASSERT(obj->oo_owner == env);
1467         LASSERT(oti->oti_w_locks > 0);
1468         oti->oti_w_locks--;
1469         obj->oo_owner = NULL;
1470         up_write(&obj->oo_sem);
1471 }
1472
1473 static int osd_object_write_locked(const struct lu_env *env,
1474                                    struct dt_object *dt)
1475 {
1476         struct osd_object *obj = osd_dt_obj(dt);
1477
1478         LINVRNT(osd_invariant(obj));
1479
1480         return obj->oo_owner == env;
1481 }
1482
1483 static struct timespec *osd_inode_time(const struct lu_env *env,
1484                                        struct inode *inode, __u64 seconds)
1485 {
1486         struct osd_thread_info  *oti = osd_oti_get(env);
1487         struct timespec         *t   = &oti->oti_time;
1488
1489         t->tv_sec = seconds;
1490         t->tv_nsec = 0;
1491         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1492         return t;
1493 }
1494
1495
1496 static void osd_inode_getattr(const struct lu_env *env,
1497                               struct inode *inode, struct lu_attr *attr)
1498 {
1499         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1500                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1501                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1502                            LA_TYPE;
1503
1504         attr->la_atime   = LTIME_S(inode->i_atime);
1505         attr->la_mtime   = LTIME_S(inode->i_mtime);
1506         attr->la_ctime   = LTIME_S(inode->i_ctime);
1507         attr->la_mode    = inode->i_mode;
1508         attr->la_size    = i_size_read(inode);
1509         attr->la_blocks  = inode->i_blocks;
1510         attr->la_uid     = i_uid_read(inode);
1511         attr->la_gid     = i_gid_read(inode);
1512         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1513         attr->la_nlink   = inode->i_nlink;
1514         attr->la_rdev    = inode->i_rdev;
1515         attr->la_blksize = 1 << inode->i_blkbits;
1516         attr->la_blkbits = inode->i_blkbits;
1517 }
1518
1519 static int osd_attr_get(const struct lu_env *env,
1520                         struct dt_object *dt,
1521                         struct lu_attr *attr)
1522 {
1523         struct osd_object *obj = osd_dt_obj(dt);
1524
1525         if (!dt_object_exists(dt))
1526                 return -ENOENT;
1527
1528         LASSERT(!dt_object_remote(dt));
1529         LINVRNT(osd_invariant(obj));
1530
1531         spin_lock(&obj->oo_guard);
1532         osd_inode_getattr(env, obj->oo_inode, attr);
1533         spin_unlock(&obj->oo_guard);
1534         return 0;
1535 }
1536
1537 static int osd_declare_attr_set(const struct lu_env *env,
1538                                 struct dt_object *dt,
1539                                 const struct lu_attr *attr,
1540                                 struct thandle *handle)
1541 {
1542         struct osd_thandle     *oh;
1543         struct osd_object      *obj;
1544         struct osd_thread_info *info = osd_oti_get(env);
1545         struct lquota_id_info  *qi = &info->oti_qi;
1546         qid_t                   uid;
1547         qid_t                   gid;
1548         long long               bspace;
1549         int                     rc = 0;
1550         bool                    enforce;
1551         ENTRY;
1552
1553         LASSERT(dt != NULL);
1554         LASSERT(handle != NULL);
1555
1556         obj = osd_dt_obj(dt);
1557         LASSERT(osd_invariant(obj));
1558
1559         oh = container_of0(handle, struct osd_thandle, ot_super);
1560         LASSERT(oh->ot_handle == NULL);
1561
1562         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1563                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1564
1565         if (attr == NULL || obj->oo_inode == NULL)
1566                 RETURN(rc);
1567
1568         bspace   = obj->oo_inode->i_blocks;
1569         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1570         bspace   = toqb(bspace);
1571
1572         /* Changing ownership is always preformed by super user, it should not
1573          * fail with EDQUOT.
1574          *
1575          * We still need to call the osd_declare_qid() to calculate the journal
1576          * credits for updating quota accounting files and to trigger quota
1577          * space adjustment once the operation is completed.*/
1578         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1579                 /* USERQUOTA */
1580                 uid = i_uid_read(obj->oo_inode);
1581                 qi->lqi_type = USRQUOTA;
1582                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1583                 /* inode accounting */
1584                 qi->lqi_is_blk = false;
1585
1586                 /* one more inode for the new uid ... */
1587                 qi->lqi_id.qid_uid = attr->la_uid;
1588                 qi->lqi_space      = 1;
1589                 /* Reserve credits for the new uid */
1590                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1591                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1592                         rc = 0;
1593                 if (rc)
1594                         RETURN(rc);
1595
1596                 /* and one less inode for the current uid */
1597                 qi->lqi_id.qid_uid = uid;
1598                 qi->lqi_space      = -1;
1599                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1600                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1601                         rc = 0;
1602                 if (rc)
1603                         RETURN(rc);
1604
1605                 /* block accounting */
1606                 qi->lqi_is_blk = true;
1607
1608                 /* more blocks for the new uid ... */
1609                 qi->lqi_id.qid_uid = attr->la_uid;
1610                 qi->lqi_space      = bspace;
1611                 /*
1612                  * Credits for the new uid has been reserved, re-use "obj"
1613                  * to save credit reservation.
1614                  */
1615                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1616                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1617                         rc = 0;
1618                 if (rc)
1619                         RETURN(rc);
1620
1621                 /* and finally less blocks for the current uid */
1622                 qi->lqi_id.qid_uid = uid;
1623                 qi->lqi_space      = -bspace;
1624                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1625                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1626                         rc = 0;
1627                 if (rc)
1628                         RETURN(rc);
1629
1630                 /* GROUP QUOTA */
1631                 gid = i_gid_read(obj->oo_inode);
1632                 qi->lqi_type = GRPQUOTA;
1633                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1634
1635                 /* inode accounting */
1636                 qi->lqi_is_blk = false;
1637
1638                 /* one more inode for the new gid ... */
1639                 qi->lqi_id.qid_gid = attr->la_gid;
1640                 qi->lqi_space      = 1;
1641                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1642                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1643                         rc = 0;
1644                 if (rc)
1645                         RETURN(rc);
1646
1647                 /* and one less inode for the current gid */
1648                 qi->lqi_id.qid_gid = gid;
1649                 qi->lqi_space      = -1;
1650                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1651                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1652                         rc = 0;
1653                 if (rc)
1654                         RETURN(rc);
1655
1656                 /* block accounting */
1657                 qi->lqi_is_blk = true;
1658
1659                 /* more blocks for the new gid ... */
1660                 qi->lqi_id.qid_gid = attr->la_gid;
1661                 qi->lqi_space      = bspace;
1662                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1663                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1664                         rc = 0;
1665                 if (rc)
1666                         RETURN(rc);
1667
1668                 /* and finally less blocks for the current gid */
1669                 qi->lqi_id.qid_gid = gid;
1670                 qi->lqi_space      = -bspace;
1671                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1672                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1673                         rc = 0;
1674                 if (rc)
1675                         RETURN(rc);
1676         }
1677
1678         RETURN(rc);
1679 }
1680
1681 static int osd_inode_setattr(const struct lu_env *env,
1682                              struct inode *inode, const struct lu_attr *attr)
1683 {
1684         __u64 bits = attr->la_valid;
1685
1686         /* Only allow set size for regular file */
1687         if (!S_ISREG(inode->i_mode))
1688                 bits &= ~(LA_SIZE | LA_BLOCKS);
1689
1690         if (bits == 0)
1691                 return 0;
1692
1693         if (bits & LA_ATIME)
1694                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1695         if (bits & LA_CTIME)
1696                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1697         if (bits & LA_MTIME)
1698                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1699         if (bits & LA_SIZE) {
1700                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1701                 i_size_write(inode, attr->la_size);
1702         }
1703
1704 #if 0
1705         /* OSD should not change "i_blocks" which is used by quota.
1706          * "i_blocks" should be changed by ldiskfs only. */
1707         if (bits & LA_BLOCKS)
1708                 inode->i_blocks = attr->la_blocks;
1709 #endif
1710         if (bits & LA_MODE)
1711                 inode->i_mode = (inode->i_mode & S_IFMT) |
1712                                 (attr->la_mode & ~S_IFMT);
1713         if (bits & LA_UID)
1714                 i_uid_write(inode, attr->la_uid);
1715         if (bits & LA_GID)
1716                 i_gid_write(inode, attr->la_gid);
1717         if (bits & LA_NLINK)
1718                 set_nlink(inode, attr->la_nlink);
1719         if (bits & LA_RDEV)
1720                 inode->i_rdev = attr->la_rdev;
1721
1722         if (bits & LA_FLAGS) {
1723                 /* always keep S_NOCMTIME */
1724                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1725                                  S_NOCMTIME;
1726         }
1727         return 0;
1728 }
1729
1730 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1731 {
1732         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1733             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1734                 struct iattr    iattr;
1735                 int             rc;
1736
1737                 ll_vfs_dq_init(inode);
1738                 iattr.ia_valid = 0;
1739                 if (attr->la_valid & LA_UID)
1740                         iattr.ia_valid |= ATTR_UID;
1741                 if (attr->la_valid & LA_GID)
1742                         iattr.ia_valid |= ATTR_GID;
1743                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1744                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1745
1746                 rc = ll_vfs_dq_transfer(inode, &iattr);
1747                 if (rc) {
1748                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1749                                "enforcement enabled on the ldiskfs "
1750                                "filesystem?\n", inode->i_sb->s_id, rc);
1751                         return rc;
1752                 }
1753         }
1754         return 0;
1755 }
1756
1757 static int osd_attr_set(const struct lu_env *env,
1758                         struct dt_object *dt,
1759                         const struct lu_attr *attr,
1760                         struct thandle *handle)
1761 {
1762         struct osd_object *obj = osd_dt_obj(dt);
1763         struct inode      *inode;
1764         int rc;
1765
1766         if (!dt_object_exists(dt))
1767                 return -ENOENT;
1768
1769         LASSERT(handle != NULL);
1770         LASSERT(!dt_object_remote(dt));
1771         LASSERT(osd_invariant(obj));
1772
1773         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1774
1775         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1776                 struct osd_thread_info  *oti  = osd_oti_get(env);
1777                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1778                 struct lu_fid           *fid1 = &oti->oti_fid;
1779                 struct osd_inode_id     *id   = &oti->oti_id;
1780                 struct iam_path_descr   *ipd;
1781                 struct iam_container    *bag;
1782                 struct osd_thandle      *oh;
1783                 int                      rc;
1784
1785                 fid_cpu_to_be(fid1, fid0);
1786                 memset(id, 1, sizeof(*id));
1787                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1788                                   fid0)->oi_dir.od_container;
1789                 ipd = osd_idx_ipd_get(env, bag);
1790                 if (unlikely(ipd == NULL))
1791                         RETURN(-ENOMEM);
1792
1793                 oh = container_of0(handle, struct osd_thandle, ot_super);
1794                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1795                                 (const struct iam_rec *)id, ipd);
1796                 osd_ipd_put(env, bag, ipd);
1797                 return(rc > 0 ? 0 : rc);
1798         }
1799
1800         inode = obj->oo_inode;
1801
1802         rc = osd_quota_transfer(inode, attr);
1803         if (rc)
1804                 return rc;
1805
1806         spin_lock(&obj->oo_guard);
1807         rc = osd_inode_setattr(env, inode, attr);
1808         spin_unlock(&obj->oo_guard);
1809
1810         if (!rc)
1811                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1812         return rc;
1813 }
1814
1815 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
1816                                            struct osd_object *obj,
1817                                            const char *name, const int namelen)
1818 {
1819         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1820 }
1821
1822 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1823                       umode_t mode, struct dt_allocation_hint *hint,
1824                       struct thandle *th)
1825 {
1826         int result;
1827         struct osd_device  *osd = osd_obj2dev(obj);
1828         struct osd_thandle *oth;
1829         struct dt_object   *parent = NULL;
1830         struct inode       *inode;
1831
1832         LINVRNT(osd_invariant(obj));
1833         LASSERT(obj->oo_inode == NULL);
1834         LASSERT(obj->oo_hl_head == NULL);
1835
1836         if (S_ISDIR(mode) && ldiskfs_pdo) {
1837                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1838                 if (obj->oo_hl_head == NULL)
1839                         return -ENOMEM;
1840         }
1841
1842         oth = container_of(th, struct osd_thandle, ot_super);
1843         LASSERT(oth->ot_handle->h_transaction != NULL);
1844
1845         if (hint && hint->dah_parent)
1846                 parent = hint->dah_parent;
1847
1848         inode = ldiskfs_create_inode(oth->ot_handle,
1849                                      parent ? osd_dt_obj(parent)->oo_inode :
1850                                               osd_sb(osd)->s_root->d_inode,
1851                                      mode);
1852         if (!IS_ERR(inode)) {
1853                 /* Do not update file c/mtime in ldiskfs. */
1854                 inode->i_flags |= S_NOCMTIME;
1855
1856                 /* For new created object, it must be consistent,
1857                  * and it is unnecessary to scrub against it. */
1858                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
1859
1860                 obj->oo_inode = inode;
1861                 result = 0;
1862         } else {
1863                 if (obj->oo_hl_head != NULL) {
1864                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1865                         obj->oo_hl_head = NULL;
1866                 }
1867                 result = PTR_ERR(inode);
1868         }
1869         LINVRNT(osd_invariant(obj));
1870         return result;
1871 }
1872
1873 enum {
1874         OSD_NAME_LEN = 255
1875 };
1876
1877 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1878                      struct lu_attr *attr,
1879                      struct dt_allocation_hint *hint,
1880                      struct dt_object_format *dof,
1881                      struct thandle *th)
1882 {
1883         int result;
1884         struct osd_thandle *oth;
1885         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1886
1887         LASSERT(S_ISDIR(attr->la_mode));
1888
1889         oth = container_of(th, struct osd_thandle, ot_super);
1890         LASSERT(oth->ot_handle->h_transaction != NULL);
1891         result = osd_mkfile(info, obj, mode, hint, th);
1892
1893         return result;
1894 }
1895
1896 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1897                         struct lu_attr *attr,
1898                         struct dt_allocation_hint *hint,
1899                         struct dt_object_format *dof,
1900                         struct thandle *th)
1901 {
1902         int result;
1903         struct osd_thandle *oth;
1904         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1905
1906         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1907
1908         LASSERT(S_ISREG(attr->la_mode));
1909
1910         oth = container_of(th, struct osd_thandle, ot_super);
1911         LASSERT(oth->ot_handle->h_transaction != NULL);
1912
1913         result = osd_mkfile(info, obj, mode, hint, th);
1914         if (result == 0) {
1915                 LASSERT(obj->oo_inode != NULL);
1916                 if (feat->dif_flags & DT_IND_VARKEY)
1917                         result = iam_lvar_create(obj->oo_inode,
1918                                                  feat->dif_keysize_max,
1919                                                  feat->dif_ptrsize,
1920                                                  feat->dif_recsize_max,
1921                                                  oth->ot_handle);
1922                 else
1923                         result = iam_lfix_create(obj->oo_inode,
1924                                                  feat->dif_keysize_max,
1925                                                  feat->dif_ptrsize,
1926                                                  feat->dif_recsize_max,
1927                                                  oth->ot_handle);
1928
1929         }
1930         return result;
1931 }
1932
1933 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1934                      struct lu_attr *attr,
1935                      struct dt_allocation_hint *hint,
1936                      struct dt_object_format *dof,
1937                      struct thandle *th)
1938 {
1939         LASSERT(S_ISREG(attr->la_mode));
1940         return osd_mkfile(info, obj, (attr->la_mode &
1941                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1942 }
1943
1944 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1945                      struct lu_attr *attr,
1946                      struct dt_allocation_hint *hint,
1947                      struct dt_object_format *dof,
1948                      struct thandle *th)
1949 {
1950         LASSERT(S_ISLNK(attr->la_mode));
1951         return osd_mkfile(info, obj, (attr->la_mode &
1952                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1953 }
1954
1955 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1956                      struct lu_attr *attr,
1957                      struct dt_allocation_hint *hint,
1958                      struct dt_object_format *dof,
1959                      struct thandle *th)
1960 {
1961         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
1962         int result;
1963
1964         LINVRNT(osd_invariant(obj));
1965         LASSERT(obj->oo_inode == NULL);
1966         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1967                 S_ISFIFO(mode) || S_ISSOCK(mode));
1968
1969         result = osd_mkfile(info, obj, mode, hint, th);
1970         if (result == 0) {
1971                 LASSERT(obj->oo_inode != NULL);
1972                 /*
1973                  * This inode should be marked dirty for i_rdev.  Currently
1974                  * that is done in the osd_attr_init().
1975                  */
1976                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
1977                                    attr->la_rdev);
1978         }
1979         LINVRNT(osd_invariant(obj));
1980         return result;
1981 }
1982
1983 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1984                               struct lu_attr *,
1985                               struct dt_allocation_hint *hint,
1986                               struct dt_object_format *dof,
1987                               struct thandle *);
1988
1989 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1990 {
1991         osd_obj_type_f result;
1992
1993         switch (type) {
1994         case DFT_DIR:
1995                 result = osd_mkdir;
1996                 break;
1997         case DFT_REGULAR:
1998                 result = osd_mkreg;
1999                 break;
2000         case DFT_SYM:
2001                 result = osd_mksym;
2002                 break;
2003         case DFT_NODE:
2004                 result = osd_mknod;
2005                 break;
2006         case DFT_INDEX:
2007                 result = osd_mk_index;
2008                 break;
2009
2010         default:
2011                 LBUG();
2012                 break;
2013         }
2014         return result;
2015 }
2016
2017
2018 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2019                         struct dt_object *parent, struct dt_object *child,
2020                         umode_t child_mode)
2021 {
2022         LASSERT(ah);
2023
2024         ah->dah_parent = parent;
2025         ah->dah_mode = child_mode;
2026 }
2027
2028 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2029                           struct lu_attr *attr, struct dt_object_format *dof)
2030 {
2031         struct inode   *inode = obj->oo_inode;
2032         __u64           valid = attr->la_valid;
2033         int             result;
2034
2035         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2036
2037         if (dof->dof_type != DFT_NODE)
2038                 attr->la_valid &= ~LA_RDEV;
2039         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2040                 attr->la_valid &= ~LA_ATIME;
2041         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2042                 attr->la_valid &= ~LA_CTIME;
2043         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2044                 attr->la_valid &= ~LA_MTIME;
2045
2046         result = osd_quota_transfer(inode, attr);
2047         if (result)
2048                 return;
2049
2050         if (attr->la_valid != 0) {
2051                 result = osd_inode_setattr(info->oti_env, inode, attr);
2052                 /*
2053                  * The osd_inode_setattr() should always succeed here.  The
2054                  * only error that could be returned is EDQUOT when we are
2055                  * trying to change the UID or GID of the inode. However, this
2056                  * should not happen since quota enforcement is no longer
2057                  * enabled on ldiskfs (lquota takes care of it).
2058                  */
2059                 LASSERTF(result == 0, "%d\n", result);
2060                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2061         }
2062
2063         attr->la_valid = valid;
2064 }
2065
2066 /**
2067  * Helper function for osd_object_create()
2068  *
2069  * \retval 0, on success
2070  */
2071 static int __osd_object_create(struct osd_thread_info *info,
2072                                struct osd_object *obj, struct lu_attr *attr,
2073                                struct dt_allocation_hint *hint,
2074                                struct dt_object_format *dof,
2075                                struct thandle *th)
2076 {
2077         int     result;
2078         __u32   umask;
2079
2080         /* we drop umask so that permissions we pass are not affected */
2081         umask = current->fs->umask;
2082         current->fs->umask = 0;
2083
2084         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2085                                                   th);
2086         if (likely(obj->oo_inode != NULL)) {
2087                 LASSERT(obj->oo_inode->i_state & I_NEW);
2088
2089                 /* Unlock the inode before attr initialization to avoid
2090                  * unnecessary dqget operations. LU-6378 */
2091                 unlock_new_inode(obj->oo_inode);
2092         }
2093
2094         if (likely(result == 0)) {
2095                 osd_attr_init(info, obj, attr, dof);
2096                 osd_object_init0(obj);
2097         }
2098
2099         /* restore previous umask value */
2100         current->fs->umask = umask;
2101
2102         return result;
2103 }
2104
2105 /**
2106  * Helper function for osd_object_create()
2107  *
2108  * \retval 0, on success
2109  */
2110 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2111                            const struct lu_fid *fid, struct thandle *th)
2112 {
2113         struct osd_thread_info *info = osd_oti_get(env);
2114         struct osd_inode_id    *id   = &info->oti_id;
2115         struct osd_device      *osd  = osd_obj2dev(obj);
2116         struct osd_thandle     *oh;
2117
2118         LASSERT(obj->oo_inode != NULL);
2119
2120         oh = container_of0(th, struct osd_thandle, ot_super);
2121         LASSERT(oh->ot_handle);
2122
2123         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2124         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2125 }
2126
2127 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2128                    u64 seq, struct lu_seq_range *range)
2129 {
2130         struct seq_server_site  *ss = osd_seq_site(osd);
2131
2132         if (fid_seq_is_idif(seq)) {
2133                 fld_range_set_ost(range);
2134                 range->lsr_index = idif_ost_idx(seq);
2135                 return 0;
2136         }
2137
2138         if (!fid_seq_in_fldb(seq)) {
2139                 fld_range_set_mdt(range);
2140                 if (ss != NULL)
2141                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2142                          * at all */
2143                         range->lsr_index = ss->ss_node_id;
2144                 return 0;
2145         }
2146
2147         LASSERT(ss != NULL);
2148         fld_range_set_any(range);
2149         /* OSD will only do local fld lookup */
2150         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2151 }
2152
2153 /*
2154  * Concurrency: no external locking is necessary.
2155  */
2156 static int osd_declare_object_create(const struct lu_env *env,
2157                                      struct dt_object *dt,
2158                                      struct lu_attr *attr,
2159                                      struct dt_allocation_hint *hint,
2160                                      struct dt_object_format *dof,
2161                                      struct thandle *handle)
2162 {
2163         struct osd_thandle      *oh;
2164         int                      rc;
2165         ENTRY;
2166
2167         LASSERT(handle != NULL);
2168
2169         oh = container_of0(handle, struct osd_thandle, ot_super);
2170         LASSERT(oh->ot_handle == NULL);
2171
2172         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2173                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2174         /* Reuse idle OI block may cause additional one OI block
2175          * to be changed. */
2176         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2177                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2178
2179         /* If this is directory, then we expect . and .. to be inserted as
2180          * well. The one directory block always needs to be created for the
2181          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2182          * block), there is no danger of needing a tree for the first block.
2183          */
2184         if (attr && S_ISDIR(attr->la_mode)) {
2185                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2186                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2187                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2188         }
2189
2190         if (!attr)
2191                 RETURN(0);
2192
2193         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2194                                    osd_dt_obj(dt), false, NULL, false);
2195         if (rc != 0)
2196                 RETURN(rc);
2197
2198         RETURN(rc);
2199 }
2200
2201 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2202                              struct lu_attr *attr,
2203                              struct dt_allocation_hint *hint,
2204                              struct dt_object_format *dof, struct thandle *th)
2205 {
2206         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2207         struct osd_object       *obj    = osd_dt_obj(dt);
2208         struct osd_thread_info  *info   = osd_oti_get(env);
2209         int result;
2210         ENTRY;
2211
2212         if (dt_object_exists(dt))
2213                 return -EEXIST;
2214
2215         LINVRNT(osd_invariant(obj));
2216         LASSERT(!dt_object_remote(dt));
2217         LASSERT(osd_write_locked(env, obj));
2218         LASSERT(th != NULL);
2219
2220         if (unlikely(fid_is_acct(fid)))
2221                 /* Quota files can't be created from the kernel any more,
2222                  * 'tune2fs -O quota' will take care of creating them */
2223                 RETURN(-EPERM);
2224
2225         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2226         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2227
2228         result = __osd_object_create(info, obj, attr, hint, dof, th);
2229         if (result == 0)
2230                 result = __osd_oi_insert(env, obj, fid, th);
2231
2232         LASSERT(ergo(result == 0,
2233                      dt_object_exists(dt) && !dt_object_remote(dt)));
2234
2235         LASSERT(osd_invariant(obj));
2236         RETURN(result);
2237 }
2238
2239 /**
2240  * Called to destroy on-disk representation of the object
2241  *
2242  * Concurrency: must be locked
2243  */
2244 static int osd_declare_object_destroy(const struct lu_env *env,
2245                                       struct dt_object *dt,
2246                                       struct thandle *th)
2247 {
2248         struct osd_object  *obj = osd_dt_obj(dt);
2249         struct inode       *inode = obj->oo_inode;
2250         struct osd_thandle *oh;
2251         int                 rc;
2252         ENTRY;
2253
2254         oh = container_of0(th, struct osd_thandle, ot_super);
2255         LASSERT(oh->ot_handle == NULL);
2256         LASSERT(inode);
2257
2258         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2259                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2260         /* Recycle idle OI leaf may cause additional three OI blocks
2261          * to be changed. */
2262         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2263                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2264         /* one less inode */
2265         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2266                                    -1, oh, obj, false, NULL, false);
2267         if (rc)
2268                 RETURN(rc);
2269         /* data to be truncated */
2270         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2271                                    0, oh, obj, true, NULL, false);
2272         RETURN(rc);
2273 }
2274
2275 static int osd_object_destroy(const struct lu_env *env,
2276                               struct dt_object *dt,
2277                               struct thandle *th)
2278 {
2279         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2280         struct osd_object      *obj = osd_dt_obj(dt);
2281         struct inode           *inode = obj->oo_inode;
2282         struct osd_device      *osd = osd_obj2dev(obj);
2283         struct osd_thandle     *oh;
2284         int                     result;
2285         ENTRY;
2286
2287         oh = container_of0(th, struct osd_thandle, ot_super);
2288         LASSERT(oh->ot_handle);
2289         LASSERT(inode);
2290         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2291
2292         if (unlikely(fid_is_acct(fid)))
2293                 RETURN(-EPERM);
2294
2295         if (S_ISDIR(inode->i_mode)) {
2296                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2297                         inode->i_nlink == 2);
2298                 /* it will check/delete the inode from remote parent,
2299                  * how to optimize it? unlink performance impaction XXX */
2300                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2301                 if (result != 0 && result != -ENOENT) {
2302                         CERROR("%s: delete inode "DFID": rc = %d\n",
2303                                osd_name(osd), PFID(fid), result);
2304                 }
2305                 spin_lock(&obj->oo_guard);
2306                 clear_nlink(inode);
2307                 spin_unlock(&obj->oo_guard);
2308                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2309         }
2310
2311         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2312
2313         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2314                                OI_CHECK_FLD);
2315
2316         /* XXX: add to ext3 orphan list */
2317         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2318
2319         /* not needed in the cache anymore */
2320         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2321
2322         RETURN(0);
2323 }
2324
2325 /**
2326  * Put the fid into lustre_mdt_attrs, and then place the structure
2327  * inode's ea. This fid should not be altered during the life time
2328  * of the inode.
2329  *
2330  * \retval +ve, on success
2331  * \retval -ve, on error
2332  *
2333  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2334  */
2335 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2336                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2337 {
2338         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2339         int                      rc;
2340         ENTRY;
2341
2342         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2343                 RETURN(0);
2344
2345         lustre_lma_init(lma, fid, compat, incompat);
2346         lustre_lma_swab(lma);
2347
2348         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2349                              XATTR_CREATE);
2350         /* LMA may already exist, but we need to check that all the
2351          * desired compat/incompat flags have been added. */
2352         if (unlikely(rc == -EEXIST)) {
2353                 if (compat == 0 && incompat == 0)
2354                         RETURN(0);
2355
2356                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2357                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2358                                      LMA_OLD_SIZE);
2359                 if (rc <= 0)
2360                         RETURN(-EINVAL);
2361
2362                 lustre_lma_swab(lma);
2363                 if (!(~lma->lma_compat & compat) &&
2364                     !(~lma->lma_incompat & incompat))
2365                         RETURN(0);
2366
2367                 lma->lma_compat |= compat;
2368                 lma->lma_incompat |= incompat;
2369                 lustre_lma_swab(lma);
2370                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2371                                      sizeof(*lma), XATTR_REPLACE);
2372         }
2373
2374         RETURN(rc);
2375 }
2376
2377 /**
2378  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2379  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2380  * To have compatilibility with 1.8 ldiskfs driver we need to have
2381  * magic number at start of fid data.
2382  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2383  * its inmemory API.
2384  */
2385 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2386                                          const struct lu_fid *fid)
2387 {
2388         if (!fid_is_namespace_visible(fid) ||
2389             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2390                 param->edp_magic = 0;
2391                 return;
2392         }
2393
2394         param->edp_magic = LDISKFS_LUFID_MAGIC;
2395         param->edp_len =  sizeof(struct lu_fid) + 1;
2396         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2397 }
2398
2399 /**
2400  * Try to read the fid from inode ea into dt_rec.
2401  *
2402  * \param fid object fid.
2403  *
2404  * \retval 0 on success
2405  */
2406 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2407                           __u32 ino, struct lu_fid *fid,
2408                           struct osd_inode_id *id)
2409 {
2410         struct osd_thread_info *info  = osd_oti_get(env);
2411         struct inode           *inode;
2412         ENTRY;
2413
2414         osd_id_gen(id, ino, OSD_OII_NOGEN);
2415         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2416         if (IS_ERR(inode))
2417                 RETURN(PTR_ERR(inode));
2418
2419         iput(inode);
2420         RETURN(0);
2421 }
2422
2423 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2424                                         struct inode *dir,
2425                                         struct inode *parent_dir,
2426                                         const struct lu_fid *dot_fid,
2427                                         const struct lu_fid *dot_dot_fid,
2428                                         struct osd_thandle *oth)
2429 {
2430         struct ldiskfs_dentry_param *dot_ldp;
2431         struct ldiskfs_dentry_param *dot_dot_ldp;
2432
2433         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2434         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2435
2436         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2437         dot_ldp->edp_magic = 0;
2438         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2439                                         dir, dot_ldp, dot_dot_ldp);
2440 }
2441
2442 /**
2443  * Create an local agent inode for remote entry
2444  */
2445 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2446                                                   struct osd_device *osd,
2447                                                   struct osd_object *pobj,
2448                                                   const struct lu_fid *fid,
2449                                                   __u32 type,
2450                                                   struct thandle *th)
2451 {
2452         struct osd_thread_info  *info = osd_oti_get(env);
2453         struct inode            *local;
2454         struct osd_thandle      *oh;
2455         int                     rc;
2456         ENTRY;
2457
2458         LASSERT(th);
2459         oh = container_of(th, struct osd_thandle, ot_super);
2460         LASSERT(oh->ot_handle->h_transaction != NULL);
2461
2462         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2463         if (IS_ERR(local)) {
2464                 CERROR("%s: create local error %d\n", osd_name(osd),
2465                        (int)PTR_ERR(local));
2466                 RETURN(local);
2467         }
2468
2469         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
2470         unlock_new_inode(local);
2471
2472         /* Set special LMA flag for local agent inode */
2473         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2474         if (rc != 0) {
2475                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2476                        osd_name(osd), PFID(fid), rc);
2477                 RETURN(ERR_PTR(rc));
2478         }
2479
2480         if (!S_ISDIR(type))
2481                 RETURN(local);
2482
2483         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2484                                          lu_object_fid(&pobj->oo_dt.do_lu),
2485                                          fid, oh);
2486         if (rc != 0) {
2487                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2488                         osd_name(osd), PFID(fid), rc);
2489                 RETURN(ERR_PTR(rc));
2490         }
2491
2492         RETURN(local);
2493 }
2494
2495 /**
2496  * Delete local agent inode for remote entry
2497  */
2498 static int osd_delete_local_agent_inode(const struct lu_env *env,
2499                                         struct osd_device *osd,
2500                                         const struct lu_fid *fid,
2501                                         __u32 ino, struct osd_thandle *oh)
2502 {
2503         struct osd_thread_info  *oti = osd_oti_get(env);
2504         struct osd_inode_id     *id = &oti->oti_id;
2505         struct inode            *inode;
2506         ENTRY;
2507
2508         id->oii_ino = le32_to_cpu(ino);
2509         id->oii_gen = OSD_OII_NOGEN;
2510         inode = osd_iget(oti, osd, id);
2511         if (IS_ERR(inode)) {
2512                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2513                        PFID(fid), id->oii_ino, id->oii_gen);
2514                 RETURN(PTR_ERR(inode));
2515         }
2516
2517         clear_nlink(inode);
2518         mark_inode_dirty(inode);
2519         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2520                 osd_name(osd), PFID(fid), inode->i_ino);
2521         iput(inode);
2522         RETURN(0);
2523 }
2524
2525 /**
2526  * OSD layer object create function for interoperability mode (b11826).
2527  * This is mostly similar to osd_object_create(). Only difference being, fid is
2528  * inserted into inode ea here.
2529  *
2530  * \retval   0, on success
2531  * \retval -ve, on error
2532  */
2533 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2534                                 struct lu_attr *attr,
2535                                 struct dt_allocation_hint *hint,
2536                                 struct dt_object_format *dof,
2537                                 struct thandle *th)
2538 {
2539         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2540         struct osd_object       *obj    = osd_dt_obj(dt);
2541         struct osd_thread_info  *info   = osd_oti_get(env);
2542         int                      result;
2543
2544         ENTRY;
2545
2546         if (dt_object_exists(dt))
2547                 RETURN(-EEXIST);
2548
2549         LASSERT(osd_invariant(obj));
2550         LASSERT(!dt_object_remote(dt));
2551         LASSERT(osd_write_locked(env, obj));
2552         LASSERT(th != NULL);
2553
2554         if (unlikely(fid_is_acct(fid)))
2555                 /* Quota files can't be created from the kernel any more,
2556                  * 'tune2fs -O quota' will take care of creating them */
2557                 RETURN(-EPERM);
2558
2559         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2560         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2561
2562         result = __osd_object_create(info, obj, attr, hint, dof, th);
2563         if (result == 0) {
2564                 if (fid_is_idif(fid) &&
2565                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
2566                         struct lu_fid *tfid = &info->oti_fid;
2567                         struct ost_id *oi   = &info->oti_ostid;
2568
2569                         fid_to_ostid(fid, oi);
2570                         ostid_to_fid(tfid, oi, 0);
2571                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
2572                                                 LMAC_FID_ON_OST, 0);
2573                 } else {
2574                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
2575                                 fid_is_on_ost(info, osd_obj2dev(obj),
2576                                               fid, OI_CHECK_FLD) ?
2577                                 LMAC_FID_ON_OST : 0, 0);
2578                 }
2579         }
2580
2581         if (result == 0)
2582                 result = __osd_oi_insert(env, obj, fid, th);
2583
2584         LASSERT(ergo(result == 0,
2585                      dt_object_exists(dt) && !dt_object_remote(dt)));
2586         LINVRNT(osd_invariant(obj));
2587         RETURN(result);
2588 }
2589
2590 static int osd_declare_object_ref_add(const struct lu_env *env,
2591                                       struct dt_object *dt,
2592                                       struct thandle *handle)
2593 {
2594         struct osd_thandle       *oh;
2595
2596         /* it's possible that object doesn't exist yet */
2597         LASSERT(handle != NULL);
2598
2599         oh = container_of0(handle, struct osd_thandle, ot_super);
2600         LASSERT(oh->ot_handle == NULL);
2601
2602         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2603                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2604
2605         return 0;
2606 }
2607
2608 /*
2609  * Concurrency: @dt is write locked.
2610  */
2611 static int osd_object_ref_add(const struct lu_env *env,
2612                               struct dt_object *dt, struct thandle *th)
2613 {
2614         struct osd_object  *obj = osd_dt_obj(dt);
2615         struct inode       *inode = obj->oo_inode;
2616         struct osd_thandle *oh;
2617         int                 rc = 0;
2618
2619         if (!dt_object_exists(dt))
2620                 return -ENOENT;
2621
2622         LINVRNT(osd_invariant(obj));
2623         LASSERT(!dt_object_remote(dt));
2624         LASSERT(osd_write_locked(env, obj));
2625         LASSERT(th != NULL);
2626
2627         oh = container_of0(th, struct osd_thandle, ot_super);
2628         LASSERT(oh->ot_handle != NULL);
2629
2630         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2631
2632         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2633                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2634         /*
2635          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2636          * (65000) subdirectories by storing "1" in i_nlink if the link count
2637          * would otherwise overflow. Directory tranversal tools understand
2638          * that (st_nlink == 1) indicates that the filesystem dose not track
2639          * hard links count on the directory, and will not abort subdirectory
2640          * scanning early once (st_nlink - 2) subdirs have been found.
2641          *
2642          * This also has to properly handle the case of inodes with nlink == 0
2643          * in case they are being linked into the PENDING directory
2644          */
2645         spin_lock(&obj->oo_guard);
2646         if (unlikely(inode->i_nlink == 0))
2647                 /* inc_nlink from 0 may cause WARN_ON */
2648                 set_nlink(inode, 1);
2649         else {
2650                 ldiskfs_inc_count(oh->ot_handle, inode);
2651                 if (!S_ISDIR(inode->i_mode))
2652                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2653         }
2654         spin_unlock(&obj->oo_guard);
2655
2656         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2657         LINVRNT(osd_invariant(obj));
2658
2659         return rc;
2660 }
2661
2662 static int osd_declare_object_ref_del(const struct lu_env *env,
2663                                       struct dt_object *dt,
2664                                       struct thandle *handle)
2665 {
2666         struct osd_thandle *oh;
2667
2668         LASSERT(!dt_object_remote(dt));
2669         LASSERT(handle != NULL);
2670
2671         oh = container_of0(handle, struct osd_thandle, ot_super);
2672         LASSERT(oh->ot_handle == NULL);
2673
2674         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2675                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2676
2677         return 0;
2678 }
2679
2680 /*
2681  * Concurrency: @dt is write locked.
2682  */
2683 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2684                               struct thandle *th)
2685 {
2686         struct osd_object       *obj = osd_dt_obj(dt);
2687         struct inode            *inode = obj->oo_inode;
2688         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2689         struct osd_thandle      *oh;
2690
2691         if (!dt_object_exists(dt))
2692                 return -ENOENT;
2693
2694         LINVRNT(osd_invariant(obj));
2695         LASSERT(!dt_object_remote(dt));
2696         LASSERT(osd_write_locked(env, obj));
2697         LASSERT(th != NULL);
2698
2699         oh = container_of0(th, struct osd_thandle, ot_super);
2700         LASSERT(oh->ot_handle != NULL);
2701
2702         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2703
2704         spin_lock(&obj->oo_guard);
2705         /* That can be result of upgrade from old Lustre version and
2706          * applied only to local files.  Just skip this ref_del call.
2707          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2708         if (inode->i_nlink == 0) {
2709                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2710                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2711                              ", maybe an upgraded file? (LU-3915)\n",
2712                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2713                 spin_unlock(&obj->oo_guard);
2714                 return 0;
2715         }
2716
2717         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2718                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2719
2720         ldiskfs_dec_count(oh->ot_handle, inode);
2721         spin_unlock(&obj->oo_guard);
2722
2723         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2724         LINVRNT(osd_invariant(obj));
2725
2726         return 0;
2727 }
2728
2729 /*
2730  * Get the 64-bit version for an inode.
2731  */
2732 static int osd_object_version_get(const struct lu_env *env,
2733                                   struct dt_object *dt, dt_obj_version_t *ver)
2734 {
2735         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2736
2737         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2738                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2739         *ver = LDISKFS_I(inode)->i_fs_version;
2740         return 0;
2741 }
2742
2743 /*
2744  * Concurrency: @dt is read locked.
2745  */
2746 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2747                          struct lu_buf *buf, const char *name)
2748 {
2749         struct osd_object      *obj    = osd_dt_obj(dt);
2750         struct inode           *inode  = obj->oo_inode;
2751         struct osd_thread_info *info   = osd_oti_get(env);
2752         struct dentry          *dentry = &info->oti_obj_dentry;
2753
2754         /* version get is not real XATTR but uses xattr API */
2755         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2756                 /* for version we are just using xattr API but change inode
2757                  * field instead */
2758                 if (buf->lb_len == 0)
2759                         return sizeof(dt_obj_version_t);
2760
2761                 if (buf->lb_len < sizeof(dt_obj_version_t))
2762                         return -ERANGE;
2763
2764                 osd_object_version_get(env, dt, buf->lb_buf);
2765
2766                 return sizeof(dt_obj_version_t);
2767         }
2768
2769         if (!dt_object_exists(dt))
2770                 return -ENOENT;
2771
2772         LASSERT(!dt_object_remote(dt));
2773         LASSERT(inode->i_op != NULL);
2774         LASSERT(inode->i_op->getxattr != NULL);
2775
2776         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2777 }
2778
2779
2780 static int osd_declare_xattr_set(const struct lu_env *env,
2781                                  struct dt_object *dt,
2782                                  const struct lu_buf *buf, const char *name,
2783                                  int fl, struct thandle *handle)
2784 {
2785         struct osd_thandle *oh;
2786         int credits;
2787         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2788
2789         LASSERT(handle != NULL);
2790
2791         oh = container_of0(handle, struct osd_thandle, ot_super);
2792         LASSERT(oh->ot_handle == NULL);
2793
2794         /* optimistic optimization: LMA is set first and usually fit inode */
2795         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2796                 if (dt_object_exists(dt))
2797                         credits = 0;
2798                 else
2799                         credits = 1;
2800         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2801                 credits = 1;
2802         } else {
2803                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2804                 if (buf && buf->lb_len > sb->s_blocksize) {
2805                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2806                                         sb->s_blocksize_bits;
2807                 }
2808                 /*
2809                  * xattr set may involve inode quota change, reserve credits for
2810                  * dquot_initialize()
2811                  */
2812                 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
2813         }
2814
2815         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2816
2817         return 0;
2818 }
2819
2820 /*
2821  * Set the 64-bit version for object
2822  */
2823 static void osd_object_version_set(const struct lu_env *env,
2824                                    struct dt_object *dt,
2825                                    dt_obj_version_t *new_version)
2826 {
2827         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2828
2829         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2830                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2831
2832         LDISKFS_I(inode)->i_fs_version = *new_version;
2833         /** Version is set after all inode operations are finished,
2834          *  so we should mark it dirty here */
2835         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2836 }
2837
2838 /*
2839  * Concurrency: @dt is write locked.
2840  */
2841 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2842                          const struct lu_buf *buf, const char *name, int fl,
2843                          struct thandle *handle)
2844 {
2845         struct osd_object      *obj      = osd_dt_obj(dt);
2846         struct inode           *inode    = obj->oo_inode;
2847         struct osd_thread_info *info     = osd_oti_get(env);
2848         int                     fs_flags = 0;
2849         ENTRY;
2850
2851         LASSERT(handle != NULL);
2852
2853         /* version set is not real XATTR */
2854         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2855                 /* for version we are just using xattr API but change inode
2856                  * field instead */
2857                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2858                 osd_object_version_set(env, dt, buf->lb_buf);
2859                 return sizeof(dt_obj_version_t);
2860         }
2861
2862         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
2863                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
2864
2865         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2866         if (fl & LU_XATTR_REPLACE)
2867                 fs_flags |= XATTR_REPLACE;
2868
2869         if (fl & LU_XATTR_CREATE)
2870                 fs_flags |= XATTR_CREATE;
2871
2872         if (strcmp(name, XATTR_NAME_LMV) == 0) {
2873                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2874                 int                      rc;
2875
2876                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
2877                 if (rc != 0)
2878                         RETURN(rc);
2879
2880                 lma->lma_incompat |= LMAI_STRIPED;
2881                 lustre_lma_swab(lma);
2882                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2883                                      sizeof(*lma), XATTR_REPLACE);
2884                 if (rc != 0)
2885                         RETURN(rc);
2886         }
2887
2888         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
2889             strcmp(name, XATTR_NAME_LINK) == 0)
2890                 return -ENOSPC;
2891
2892         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
2893                                fs_flags);
2894 }
2895
2896 /*
2897  * Concurrency: @dt is read locked.
2898  */
2899 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2900                           const struct lu_buf *buf)
2901 {
2902         struct osd_object      *obj    = osd_dt_obj(dt);
2903         struct inode           *inode  = obj->oo_inode;
2904         struct osd_thread_info *info   = osd_oti_get(env);
2905         struct dentry          *dentry = &info->oti_obj_dentry;
2906
2907         if (!dt_object_exists(dt))
2908                 return -ENOENT;
2909
2910         LASSERT(!dt_object_remote(dt));
2911         LASSERT(inode->i_op != NULL);
2912         LASSERT(inode->i_op->listxattr != NULL);
2913
2914         dentry->d_inode = inode;
2915         dentry->d_sb = inode->i_sb;
2916         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2917 }
2918
2919 static int osd_declare_xattr_del(const struct lu_env *env,
2920                                  struct dt_object *dt, const char *name,
2921                                  struct thandle *handle)
2922 {
2923         struct osd_thandle *oh;
2924         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2925
2926         LASSERT(!dt_object_remote(dt));
2927         LASSERT(handle != NULL);
2928
2929         oh = container_of0(handle, struct osd_thandle, ot_super);
2930         LASSERT(oh->ot_handle == NULL);
2931
2932         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2933                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2934         /*
2935          * xattr del may involve inode quota change, reserve credits for
2936          * dquot_initialize()
2937          */
2938         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
2939
2940         return 0;
2941 }
2942
2943 /*
2944  * Concurrency: @dt is write locked.
2945  */
2946 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2947                          const char *name, struct thandle *handle)
2948 {
2949         struct osd_object      *obj    = osd_dt_obj(dt);
2950         struct inode           *inode  = obj->oo_inode;
2951         struct osd_thread_info *info   = osd_oti_get(env);
2952         struct dentry          *dentry = &info->oti_obj_dentry;
2953         int                     rc;
2954
2955         if (!dt_object_exists(dt))
2956                 return -ENOENT;
2957
2958         LASSERT(!dt_object_remote(dt));
2959         LASSERT(inode->i_op != NULL);
2960         LASSERT(inode->i_op->removexattr != NULL);
2961         LASSERT(handle != NULL);
2962
2963         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2964
2965         ll_vfs_dq_init(inode);
2966         dentry->d_inode = inode;
2967         dentry->d_sb = inode->i_sb;
2968         rc = inode->i_op->removexattr(dentry, name);
2969         return rc;
2970 }
2971
2972 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
2973                            __u64 start, __u64 end)
2974 {
2975         struct osd_object       *obj    = osd_dt_obj(dt);
2976         struct inode            *inode  = obj->oo_inode;
2977         struct osd_thread_info  *info   = osd_oti_get(env);
2978         struct dentry           *dentry = &info->oti_obj_dentry;
2979         struct file             *file   = &info->oti_file;
2980         int                     rc;
2981
2982         ENTRY;
2983
2984         dentry->d_inode = inode;
2985         dentry->d_sb = inode->i_sb;
2986         file->f_dentry = dentry;
2987         file->f_mapping = inode->i_mapping;
2988         file->f_op = inode->i_fop;
2989         set_file_inode(file, inode);
2990
2991         rc = ll_vfs_fsync_range(file, start, end, 0);
2992
2993         RETURN(rc);
2994 }
2995
2996 /*
2997  * Index operations.
2998  */
2999
3000 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3001                            const struct dt_index_features *feat)
3002 {
3003         struct iam_descr *descr;
3004
3005         if (osd_object_is_root(o))
3006                 return feat == &dt_directory_features;
3007
3008         LASSERT(o->oo_dir != NULL);
3009
3010         descr = o->oo_dir->od_container.ic_descr;
3011         if (feat == &dt_directory_features) {
3012                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3013                         return 1;
3014                 else
3015                         return 0;
3016         } else {
3017                 return
3018                         feat->dif_keysize_min <= descr->id_key_size &&
3019                         descr->id_key_size <= feat->dif_keysize_max &&
3020                         feat->dif_recsize_min <= descr->id_rec_size &&
3021                         descr->id_rec_size <= feat->dif_recsize_max &&
3022                         !(feat->dif_flags & (DT_IND_VARKEY |
3023                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3024                         ergo(feat->dif_flags & DT_IND_UPDATE,
3025                              1 /* XXX check that object (and file system) is
3026                                 * writable */);
3027         }
3028 }
3029
3030 static int osd_iam_container_init(const struct lu_env *env,
3031                                   struct osd_object *obj,
3032                                   struct osd_directory *dir)
3033 {
3034         struct iam_container *bag = &dir->od_container;
3035         int result;
3036
3037         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3038         if (result != 0)
3039                 return result;
3040
3041         result = iam_container_setup(bag);
3042         if (result == 0)
3043                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3044         else
3045                 iam_container_fini(bag);
3046
3047         return result;
3048 }
3049
3050
3051 /*
3052  * Concurrency: no external locking is necessary.
3053  */
3054 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3055                          const struct dt_index_features *feat)
3056 {
3057         int                      result;
3058         int                      skip_iam = 0;
3059         struct osd_object       *obj = osd_dt_obj(dt);
3060
3061         LINVRNT(osd_invariant(obj));
3062
3063         if (osd_object_is_root(obj)) {
3064                 dt->do_index_ops = &osd_index_ea_ops;
3065                 result = 0;
3066         } else if (feat == &dt_directory_features) {
3067                 dt->do_index_ops = &osd_index_ea_ops;
3068                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3069                         result = 0;
3070                 else
3071                         result = -ENOTDIR;
3072                 skip_iam = 1;
3073         } else if (unlikely(feat == &dt_otable_features)) {
3074                 dt->do_index_ops = &osd_otable_ops;
3075                 return 0;
3076         } else if (unlikely(feat == &dt_acct_features)) {
3077                 dt->do_index_ops = &osd_acct_index_ops;
3078                 result = 0;
3079                 skip_iam = 1;
3080         } else if (!osd_has_index(obj)) {
3081                 struct osd_directory *dir;
3082
3083                 OBD_ALLOC_PTR(dir);
3084                 if (dir != NULL) {
3085
3086                         spin_lock(&obj->oo_guard);
3087                         if (obj->oo_dir == NULL)
3088                                 obj->oo_dir = dir;
3089                         else
3090                                 /*
3091                                  * Concurrent thread allocated container data.
3092                                  */
3093                                 OBD_FREE_PTR(dir);
3094                         spin_unlock(&obj->oo_guard);
3095                         /*
3096                          * Now, that we have container data, serialize its
3097                          * initialization.
3098                          */
3099                         down_write(&obj->oo_ext_idx_sem);
3100                         /*
3101                          * recheck under lock.
3102                          */
3103                         if (!osd_has_index(obj))
3104                                 result = osd_iam_container_init(env, obj,
3105                                                                 obj->oo_dir);
3106                         else
3107                                 result = 0;
3108                         up_write(&obj->oo_ext_idx_sem);
3109                 } else {
3110                         result = -ENOMEM;
3111                 }
3112         } else {
3113                 result = 0;
3114         }
3115
3116         if (result == 0 && skip_iam == 0) {
3117                 if (!osd_iam_index_probe(env, obj, feat))
3118                         result = -ENOTDIR;
3119         }
3120         LINVRNT(osd_invariant(obj));
3121
3122         if (result == 0 && is_quota_glb_feat(feat) &&
3123             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3124                 result = osd_quota_migration(env, dt, feat);
3125
3126         return result;
3127 }
3128
3129 static int osd_otable_it_attr_get(const struct lu_env *env,
3130                                  struct dt_object *dt,
3131                                  struct lu_attr *attr)
3132 {
3133         attr->la_valid = 0;
3134         return 0;
3135 }
3136
3137 static const struct dt_object_operations osd_obj_ops = {
3138         .do_read_lock         = osd_object_read_lock,
3139         .do_write_lock        = osd_object_write_lock,
3140         .do_read_unlock       = osd_object_read_unlock,
3141         .do_write_unlock      = osd_object_write_unlock,
3142         .do_write_locked      = osd_object_write_locked,
3143         .do_attr_get          = osd_attr_get,
3144         .do_declare_attr_set  = osd_declare_attr_set,
3145         .do_attr_set          = osd_attr_set,
3146         .do_ah_init           = osd_ah_init,
3147         .do_declare_create    = osd_declare_object_create,
3148         .do_create            = osd_object_create,
3149         .do_declare_destroy   = osd_declare_object_destroy,
3150         .do_destroy           = osd_object_destroy,
3151         .do_index_try         = osd_index_try,
3152         .do_declare_ref_add   = osd_declare_object_ref_add,
3153         .do_ref_add           = osd_object_ref_add,
3154         .do_declare_ref_del   = osd_declare_object_ref_del,
3155         .do_ref_del           = osd_object_ref_del,
3156         .do_xattr_get         = osd_xattr_get,
3157         .do_declare_xattr_set = osd_declare_xattr_set,
3158         .do_xattr_set         = osd_xattr_set,
3159         .do_declare_xattr_del = osd_declare_xattr_del,
3160         .do_xattr_del         = osd_xattr_del,
3161         .do_xattr_list        = osd_xattr_list,
3162         .do_object_sync       = osd_object_sync,
3163 };
3164
3165 /**
3166  * dt_object_operations for interoperability mode
3167  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3168  */
3169 static const struct dt_object_operations osd_obj_ea_ops = {
3170         .do_read_lock         = osd_object_read_lock,
3171         .do_write_lock        = osd_object_write_lock,
3172         .do_read_unlock       = osd_object_read_unlock,
3173         .do_write_unlock      = osd_object_write_unlock,
3174         .do_write_locked      = osd_object_write_locked,
3175         .do_attr_get          = osd_attr_get,
3176         .do_declare_attr_set  = osd_declare_attr_set,
3177         .do_attr_set          = osd_attr_set,
3178         .do_ah_init           = osd_ah_init,